]>
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> |
1da177e4 | 39 | #include <linux/buffer_head.h> /* for invalidate_bdev */ |
d7603b7e | 40 | #include <linux/poll.h> |
16f17b39 | 41 | #include <linux/ctype.h> |
e7d2860b | 42 | #include <linux/string.h> |
fb4d8c76 N |
43 | #include <linux/hdreg.h> |
44 | #include <linux/proc_fs.h> | |
45 | #include <linux/random.h> | |
46 | #include <linux/reboot.h> | |
32a7627c | 47 | #include <linux/file.h> |
aa98aa31 | 48 | #include <linux/compat.h> |
25570727 | 49 | #include <linux/delay.h> |
bff61975 N |
50 | #include <linux/raid/md_p.h> |
51 | #include <linux/raid/md_u.h> | |
5a0e3ad6 | 52 | #include <linux/slab.h> |
43b2e5d8 | 53 | #include "md.h" |
ef740c37 | 54 | #include "bitmap.h" |
1da177e4 LT |
55 | |
56 | #define DEBUG 0 | |
57 | #define dprintk(x...) ((void)(DEBUG && printk(x))) | |
58 | ||
59 | ||
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 BS |
69 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
70 | ||
5e56341d AB |
71 | #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } |
72 | ||
1e50915f RB |
73 | /* |
74 | * Default number of read corrections we'll attempt on an rdev | |
75 | * before ejecting it from the array. We divide the read error | |
76 | * count by 2 for every hour elapsed between read errors. | |
77 | */ | |
78 | #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 | |
1da177e4 LT |
79 | /* |
80 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
81 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
82 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 83 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
84 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
85 | * speed limit - in case reconstruction slows down your system despite | |
86 | * idle IO detection. | |
87 | * | |
88 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 89 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
90 | */ |
91 | ||
92 | static int sysctl_speed_limit_min = 1000; | |
93 | static int sysctl_speed_limit_max = 200000; | |
88202a0c N |
94 | static inline int speed_min(mddev_t *mddev) |
95 | { | |
96 | return mddev->sync_speed_min ? | |
97 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
98 | } | |
99 | ||
100 | static inline int speed_max(mddev_t *mddev) | |
101 | { | |
102 | return mddev->sync_speed_max ? | |
103 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
104 | } | |
1da177e4 LT |
105 | |
106 | static struct ctl_table_header *raid_table_header; | |
107 | ||
108 | static ctl_table raid_table[] = { | |
109 | { | |
1da177e4 LT |
110 | .procname = "speed_limit_min", |
111 | .data = &sysctl_speed_limit_min, | |
112 | .maxlen = sizeof(int), | |
80ca3a44 | 113 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 114 | .proc_handler = proc_dointvec, |
1da177e4 LT |
115 | }, |
116 | { | |
1da177e4 LT |
117 | .procname = "speed_limit_max", |
118 | .data = &sysctl_speed_limit_max, | |
119 | .maxlen = sizeof(int), | |
80ca3a44 | 120 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 121 | .proc_handler = proc_dointvec, |
1da177e4 | 122 | }, |
894d2491 | 123 | { } |
1da177e4 LT |
124 | }; |
125 | ||
126 | static ctl_table raid_dir_table[] = { | |
127 | { | |
1da177e4 LT |
128 | .procname = "raid", |
129 | .maxlen = 0, | |
80ca3a44 | 130 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
131 | .child = raid_table, |
132 | }, | |
894d2491 | 133 | { } |
1da177e4 LT |
134 | }; |
135 | ||
136 | static ctl_table raid_root_table[] = { | |
137 | { | |
1da177e4 LT |
138 | .procname = "dev", |
139 | .maxlen = 0, | |
140 | .mode = 0555, | |
141 | .child = raid_dir_table, | |
142 | }, | |
894d2491 | 143 | { } |
1da177e4 LT |
144 | }; |
145 | ||
83d5cde4 | 146 | static const struct block_device_operations md_fops; |
1da177e4 | 147 | |
f91de92e N |
148 | static int start_readonly; |
149 | ||
d7603b7e N |
150 | /* |
151 | * We have a system wide 'event count' that is incremented | |
152 | * on any 'interesting' event, and readers of /proc/mdstat | |
153 | * can use 'poll' or 'select' to find out when the event | |
154 | * count increases. | |
155 | * | |
156 | * Events are: | |
157 | * start array, stop array, error, add device, remove device, | |
158 | * start build, activate spare | |
159 | */ | |
2989ddbd | 160 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 161 | static atomic_t md_event_count; |
29269553 | 162 | void md_new_event(mddev_t *mddev) |
d7603b7e N |
163 | { |
164 | atomic_inc(&md_event_count); | |
165 | wake_up(&md_event_waiters); | |
166 | } | |
29269553 | 167 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 168 | |
c331eb04 N |
169 | /* Alternate version that can be called from interrupts |
170 | * when calling sysfs_notify isn't needed. | |
171 | */ | |
05381954 | 172 | static void md_new_event_inintr(mddev_t *mddev) |
c331eb04 N |
173 | { |
174 | atomic_inc(&md_event_count); | |
175 | wake_up(&md_event_waiters); | |
176 | } | |
177 | ||
1da177e4 LT |
178 | /* |
179 | * Enables to iterate over all existing md arrays | |
180 | * all_mddevs_lock protects this list. | |
181 | */ | |
182 | static LIST_HEAD(all_mddevs); | |
183 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
184 | ||
185 | ||
186 | /* | |
187 | * iterates through all used mddevs in the system. | |
188 | * We take care to grab the all_mddevs_lock whenever navigating | |
189 | * the list, and to always hold a refcount when unlocked. | |
190 | * Any code which breaks out of this loop while own | |
191 | * a reference to the current mddev and must mddev_put it. | |
192 | */ | |
29ac4aa3 | 193 | #define for_each_mddev(mddev,tmp) \ |
1da177e4 LT |
194 | \ |
195 | for (({ spin_lock(&all_mddevs_lock); \ | |
196 | tmp = all_mddevs.next; \ | |
197 | mddev = NULL;}); \ | |
198 | ({ if (tmp != &all_mddevs) \ | |
199 | mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ | |
200 | spin_unlock(&all_mddevs_lock); \ | |
201 | if (mddev) mddev_put(mddev); \ | |
202 | mddev = list_entry(tmp, mddev_t, all_mddevs); \ | |
203 | tmp != &all_mddevs;}); \ | |
204 | ({ spin_lock(&all_mddevs_lock); \ | |
205 | tmp = tmp->next;}) \ | |
206 | ) | |
207 | ||
208 | ||
409c57f3 N |
209 | /* Rather than calling directly into the personality make_request function, |
210 | * IO requests come here first so that we can check if the device is | |
211 | * being suspended pending a reconfiguration. | |
212 | * We hold a refcount over the call to ->make_request. By the time that | |
213 | * call has finished, the bio has been linked into some internal structure | |
214 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
215 | */ | |
216 | static int md_make_request(struct request_queue *q, struct bio *bio) | |
1da177e4 | 217 | { |
49077326 | 218 | const int rw = bio_data_dir(bio); |
409c57f3 N |
219 | mddev_t *mddev = q->queuedata; |
220 | int rv; | |
49077326 N |
221 | int cpu; |
222 | ||
409c57f3 N |
223 | if (mddev == NULL || mddev->pers == NULL) { |
224 | bio_io_error(bio); | |
225 | return 0; | |
226 | } | |
227 | rcu_read_lock(); | |
a2826aa9 | 228 | if (mddev->suspended || mddev->barrier) { |
409c57f3 N |
229 | DEFINE_WAIT(__wait); |
230 | for (;;) { | |
231 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
232 | TASK_UNINTERRUPTIBLE); | |
a2826aa9 | 233 | if (!mddev->suspended && !mddev->barrier) |
409c57f3 N |
234 | break; |
235 | rcu_read_unlock(); | |
236 | schedule(); | |
237 | rcu_read_lock(); | |
238 | } | |
239 | finish_wait(&mddev->sb_wait, &__wait); | |
240 | } | |
241 | atomic_inc(&mddev->active_io); | |
242 | rcu_read_unlock(); | |
49077326 | 243 | |
21a52c6d | 244 | rv = mddev->pers->make_request(mddev, bio); |
49077326 N |
245 | |
246 | cpu = part_stat_lock(); | |
247 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
248 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
249 | bio_sectors(bio)); | |
250 | part_stat_unlock(); | |
251 | ||
409c57f3 N |
252 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
253 | wake_up(&mddev->sb_wait); | |
254 | ||
255 | return rv; | |
256 | } | |
257 | ||
9e35b99c N |
258 | /* mddev_suspend makes sure no new requests are submitted |
259 | * to the device, and that any requests that have been submitted | |
260 | * are completely handled. | |
261 | * Once ->stop is called and completes, the module will be completely | |
262 | * unused. | |
263 | */ | |
390ee602 | 264 | void mddev_suspend(mddev_t *mddev) |
409c57f3 N |
265 | { |
266 | BUG_ON(mddev->suspended); | |
267 | mddev->suspended = 1; | |
268 | synchronize_rcu(); | |
269 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
270 | mddev->pers->quiesce(mddev, 1); | |
409c57f3 | 271 | } |
390ee602 | 272 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 273 | |
390ee602 | 274 | void mddev_resume(mddev_t *mddev) |
409c57f3 N |
275 | { |
276 | mddev->suspended = 0; | |
277 | wake_up(&mddev->sb_wait); | |
278 | mddev->pers->quiesce(mddev, 0); | |
1da177e4 | 279 | } |
390ee602 | 280 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 281 | |
3fa841d7 N |
282 | int mddev_congested(mddev_t *mddev, int bits) |
283 | { | |
a2826aa9 N |
284 | if (mddev->barrier) |
285 | return 1; | |
3fa841d7 N |
286 | return mddev->suspended; |
287 | } | |
288 | EXPORT_SYMBOL(mddev_congested); | |
289 | ||
a2826aa9 N |
290 | /* |
291 | * Generic barrier handling for md | |
292 | */ | |
293 | ||
294 | #define POST_REQUEST_BARRIER ((void*)1) | |
295 | ||
296 | static void md_end_barrier(struct bio *bio, int err) | |
297 | { | |
298 | mdk_rdev_t *rdev = bio->bi_private; | |
299 | mddev_t *mddev = rdev->mddev; | |
300 | if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER) | |
301 | set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags); | |
302 | ||
303 | rdev_dec_pending(rdev, mddev); | |
304 | ||
305 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
306 | if (mddev->barrier == POST_REQUEST_BARRIER) { | |
307 | /* This was a post-request barrier */ | |
308 | mddev->barrier = NULL; | |
309 | wake_up(&mddev->sb_wait); | |
310 | } else | |
311 | /* The pre-request barrier has finished */ | |
312 | schedule_work(&mddev->barrier_work); | |
313 | } | |
314 | bio_put(bio); | |
315 | } | |
316 | ||
317 | static void submit_barriers(mddev_t *mddev) | |
318 | { | |
319 | mdk_rdev_t *rdev; | |
320 | ||
321 | rcu_read_lock(); | |
322 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
323 | if (rdev->raid_disk >= 0 && | |
324 | !test_bit(Faulty, &rdev->flags)) { | |
325 | /* Take two references, one is dropped | |
326 | * when request finishes, one after | |
327 | * we reclaim rcu_read_lock | |
328 | */ | |
329 | struct bio *bi; | |
330 | atomic_inc(&rdev->nr_pending); | |
331 | atomic_inc(&rdev->nr_pending); | |
332 | rcu_read_unlock(); | |
333 | bi = bio_alloc(GFP_KERNEL, 0); | |
334 | bi->bi_end_io = md_end_barrier; | |
335 | bi->bi_private = rdev; | |
336 | bi->bi_bdev = rdev->bdev; | |
337 | atomic_inc(&mddev->flush_pending); | |
338 | submit_bio(WRITE_BARRIER, bi); | |
339 | rcu_read_lock(); | |
340 | rdev_dec_pending(rdev, mddev); | |
341 | } | |
342 | rcu_read_unlock(); | |
343 | } | |
344 | ||
345 | static void md_submit_barrier(struct work_struct *ws) | |
346 | { | |
347 | mddev_t *mddev = container_of(ws, mddev_t, barrier_work); | |
348 | struct bio *bio = mddev->barrier; | |
349 | ||
350 | atomic_set(&mddev->flush_pending, 1); | |
351 | ||
352 | if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags)) | |
353 | bio_endio(bio, -EOPNOTSUPP); | |
354 | else if (bio->bi_size == 0) | |
355 | /* an empty barrier - all done */ | |
356 | bio_endio(bio, 0); | |
357 | else { | |
358 | bio->bi_rw &= ~(1<<BIO_RW_BARRIER); | |
21a52c6d | 359 | if (mddev->pers->make_request(mddev, bio)) |
a2826aa9 N |
360 | generic_make_request(bio); |
361 | mddev->barrier = POST_REQUEST_BARRIER; | |
362 | submit_barriers(mddev); | |
363 | } | |
364 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
365 | mddev->barrier = NULL; | |
366 | wake_up(&mddev->sb_wait); | |
367 | } | |
368 | } | |
369 | ||
370 | void md_barrier_request(mddev_t *mddev, struct bio *bio) | |
371 | { | |
372 | spin_lock_irq(&mddev->write_lock); | |
373 | wait_event_lock_irq(mddev->sb_wait, | |
374 | !mddev->barrier, | |
375 | mddev->write_lock, /*nothing*/); | |
376 | mddev->barrier = bio; | |
377 | spin_unlock_irq(&mddev->write_lock); | |
378 | ||
379 | atomic_set(&mddev->flush_pending, 1); | |
380 | INIT_WORK(&mddev->barrier_work, md_submit_barrier); | |
381 | ||
382 | submit_barriers(mddev); | |
383 | ||
384 | if (atomic_dec_and_test(&mddev->flush_pending)) | |
385 | schedule_work(&mddev->barrier_work); | |
386 | } | |
387 | EXPORT_SYMBOL(md_barrier_request); | |
409c57f3 | 388 | |
2ac87401 N |
389 | /* Support for plugging. |
390 | * This mirrors the plugging support in request_queue, but does not | |
391 | * require having a whole queue | |
392 | */ | |
393 | static void plugger_work(struct work_struct *work) | |
394 | { | |
395 | struct plug_handle *plug = | |
396 | container_of(work, struct plug_handle, unplug_work); | |
397 | plug->unplug_fn(plug); | |
398 | } | |
399 | static void plugger_timeout(unsigned long data) | |
400 | { | |
401 | struct plug_handle *plug = (void *)data; | |
402 | kblockd_schedule_work(NULL, &plug->unplug_work); | |
403 | } | |
404 | void plugger_init(struct plug_handle *plug, | |
405 | void (*unplug_fn)(struct plug_handle *)) | |
406 | { | |
407 | plug->unplug_flag = 0; | |
408 | plug->unplug_fn = unplug_fn; | |
409 | init_timer(&plug->unplug_timer); | |
410 | plug->unplug_timer.function = plugger_timeout; | |
411 | plug->unplug_timer.data = (unsigned long)plug; | |
412 | INIT_WORK(&plug->unplug_work, plugger_work); | |
413 | } | |
414 | EXPORT_SYMBOL_GPL(plugger_init); | |
415 | ||
416 | void plugger_set_plug(struct plug_handle *plug) | |
417 | { | |
418 | if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag)) | |
419 | mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1); | |
420 | } | |
421 | EXPORT_SYMBOL_GPL(plugger_set_plug); | |
422 | ||
423 | int plugger_remove_plug(struct plug_handle *plug) | |
424 | { | |
425 | if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) { | |
426 | del_timer(&plug->unplug_timer); | |
427 | return 1; | |
428 | } else | |
429 | return 0; | |
430 | } | |
431 | EXPORT_SYMBOL_GPL(plugger_remove_plug); | |
432 | ||
433 | ||
1da177e4 LT |
434 | static inline mddev_t *mddev_get(mddev_t *mddev) |
435 | { | |
436 | atomic_inc(&mddev->active); | |
437 | return mddev; | |
438 | } | |
439 | ||
5fd3a17e | 440 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 441 | |
1da177e4 LT |
442 | static void mddev_put(mddev_t *mddev) |
443 | { | |
444 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) | |
445 | return; | |
d3374825 | 446 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
447 | mddev->ctime == 0 && !mddev->hold_active) { |
448 | /* Array is not configured at all, and not held active, | |
449 | * so destroy it */ | |
1da177e4 | 450 | list_del(&mddev->all_mddevs); |
d3374825 N |
451 | if (mddev->gendisk) { |
452 | /* we did a probe so need to clean up. | |
453 | * Call schedule_work inside the spinlock | |
454 | * so that flush_scheduled_work() after | |
455 | * mddev_find will succeed in waiting for the | |
456 | * work to be done. | |
457 | */ | |
458 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
459 | schedule_work(&mddev->del_work); | |
460 | } else | |
461 | kfree(mddev); | |
462 | } | |
463 | spin_unlock(&all_mddevs_lock); | |
1da177e4 LT |
464 | } |
465 | ||
390ee602 | 466 | void mddev_init(mddev_t *mddev) |
fafd7fb0 N |
467 | { |
468 | mutex_init(&mddev->open_mutex); | |
469 | mutex_init(&mddev->reconfig_mutex); | |
470 | mutex_init(&mddev->bitmap_info.mutex); | |
471 | INIT_LIST_HEAD(&mddev->disks); | |
472 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
473 | init_timer(&mddev->safemode_timer); | |
474 | atomic_set(&mddev->active, 1); | |
475 | atomic_set(&mddev->openers, 0); | |
476 | atomic_set(&mddev->active_io, 0); | |
477 | spin_lock_init(&mddev->write_lock); | |
478 | atomic_set(&mddev->flush_pending, 0); | |
479 | init_waitqueue_head(&mddev->sb_wait); | |
480 | init_waitqueue_head(&mddev->recovery_wait); | |
481 | mddev->reshape_position = MaxSector; | |
482 | mddev->resync_min = 0; | |
483 | mddev->resync_max = MaxSector; | |
484 | mddev->level = LEVEL_NONE; | |
485 | } | |
390ee602 | 486 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 487 | |
1da177e4 LT |
488 | static mddev_t * mddev_find(dev_t unit) |
489 | { | |
490 | mddev_t *mddev, *new = NULL; | |
491 | ||
492 | retry: | |
493 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
494 | |
495 | if (unit) { | |
496 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
497 | if (mddev->unit == unit) { | |
498 | mddev_get(mddev); | |
499 | spin_unlock(&all_mddevs_lock); | |
500 | kfree(new); | |
501 | return mddev; | |
502 | } | |
503 | ||
504 | if (new) { | |
505 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 506 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
507 | new->hold_active = UNTIL_IOCTL; |
508 | return new; | |
1da177e4 | 509 | } |
efeb53c0 N |
510 | } else if (new) { |
511 | /* find an unused unit number */ | |
512 | static int next_minor = 512; | |
513 | int start = next_minor; | |
514 | int is_free = 0; | |
515 | int dev = 0; | |
516 | while (!is_free) { | |
517 | dev = MKDEV(MD_MAJOR, next_minor); | |
518 | next_minor++; | |
519 | if (next_minor > MINORMASK) | |
520 | next_minor = 0; | |
521 | if (next_minor == start) { | |
522 | /* Oh dear, all in use. */ | |
523 | spin_unlock(&all_mddevs_lock); | |
524 | kfree(new); | |
525 | return NULL; | |
526 | } | |
527 | ||
528 | is_free = 1; | |
529 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
530 | if (mddev->unit == dev) { | |
531 | is_free = 0; | |
532 | break; | |
533 | } | |
534 | } | |
535 | new->unit = dev; | |
536 | new->md_minor = MINOR(dev); | |
537 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
538 | list_add(&new->all_mddevs, &all_mddevs); |
539 | spin_unlock(&all_mddevs_lock); | |
540 | return new; | |
541 | } | |
542 | spin_unlock(&all_mddevs_lock); | |
543 | ||
9ffae0cf | 544 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
545 | if (!new) |
546 | return NULL; | |
547 | ||
1da177e4 LT |
548 | new->unit = unit; |
549 | if (MAJOR(unit) == MD_MAJOR) | |
550 | new->md_minor = MINOR(unit); | |
551 | else | |
552 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
553 | ||
fafd7fb0 | 554 | mddev_init(new); |
1da177e4 | 555 | |
1da177e4 LT |
556 | goto retry; |
557 | } | |
558 | ||
559 | static inline int mddev_lock(mddev_t * mddev) | |
560 | { | |
df5b89b3 | 561 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
562 | } |
563 | ||
b522adcd DW |
564 | static inline int mddev_is_locked(mddev_t *mddev) |
565 | { | |
566 | return mutex_is_locked(&mddev->reconfig_mutex); | |
567 | } | |
568 | ||
1da177e4 LT |
569 | static inline int mddev_trylock(mddev_t * mddev) |
570 | { | |
df5b89b3 | 571 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
572 | } |
573 | ||
b6eb127d N |
574 | static struct attribute_group md_redundancy_group; |
575 | ||
a64c876f | 576 | static void mddev_unlock(mddev_t * mddev) |
1da177e4 | 577 | { |
a64c876f | 578 | if (mddev->to_remove) { |
b6eb127d N |
579 | /* These cannot be removed under reconfig_mutex as |
580 | * an access to the files will try to take reconfig_mutex | |
581 | * while holding the file unremovable, which leads to | |
582 | * a deadlock. | |
583 | * So hold open_mutex instead - we are allowed to take | |
584 | * it while holding reconfig_mutex, and md_run can | |
585 | * use it to wait for the remove to complete. | |
586 | */ | |
a64c876f N |
587 | struct attribute_group *to_remove = mddev->to_remove; |
588 | mddev->to_remove = NULL; | |
b6eb127d N |
589 | mutex_lock(&mddev->open_mutex); |
590 | mutex_unlock(&mddev->reconfig_mutex); | |
591 | ||
00bcb4ac N |
592 | if (mddev->kobj.sd) { |
593 | if (to_remove != &md_redundancy_group) | |
594 | sysfs_remove_group(&mddev->kobj, to_remove); | |
595 | if (mddev->pers == NULL || | |
596 | mddev->pers->sync_request == NULL) { | |
597 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
598 | if (mddev->sysfs_action) | |
599 | sysfs_put(mddev->sysfs_action); | |
600 | mddev->sysfs_action = NULL; | |
601 | } | |
a64c876f | 602 | } |
b6eb127d N |
603 | mutex_unlock(&mddev->open_mutex); |
604 | } else | |
605 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 606 | |
005eca5e | 607 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
608 | } |
609 | ||
2989ddbd | 610 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 | 611 | { |
159ec1fc | 612 | mdk_rdev_t *rdev; |
1da177e4 | 613 | |
159ec1fc | 614 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
615 | if (rdev->desc_nr == nr) |
616 | return rdev; | |
159ec1fc | 617 | |
1da177e4 LT |
618 | return NULL; |
619 | } | |
620 | ||
621 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
622 | { | |
1da177e4 LT |
623 | mdk_rdev_t *rdev; |
624 | ||
159ec1fc | 625 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
626 | if (rdev->bdev->bd_dev == dev) |
627 | return rdev; | |
159ec1fc | 628 | |
1da177e4 LT |
629 | return NULL; |
630 | } | |
631 | ||
d9d166c2 | 632 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
633 | { |
634 | struct mdk_personality *pers; | |
d9d166c2 N |
635 | list_for_each_entry(pers, &pers_list, list) { |
636 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 637 | return pers; |
d9d166c2 N |
638 | if (strcmp(pers->name, clevel)==0) |
639 | return pers; | |
640 | } | |
2604b703 N |
641 | return NULL; |
642 | } | |
643 | ||
b73df2d3 | 644 | /* return the offset of the super block in 512byte sectors */ |
77933d72 | 645 | static inline sector_t calc_dev_sboffset(struct block_device *bdev) |
1da177e4 | 646 | { |
b73df2d3 AN |
647 | sector_t num_sectors = bdev->bd_inode->i_size / 512; |
648 | return MD_NEW_SIZE_SECTORS(num_sectors); | |
1da177e4 LT |
649 | } |
650 | ||
1da177e4 LT |
651 | static int alloc_disk_sb(mdk_rdev_t * rdev) |
652 | { | |
653 | if (rdev->sb_page) | |
654 | MD_BUG(); | |
655 | ||
656 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
657 | if (!rdev->sb_page) { | |
658 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 659 | return -ENOMEM; |
1da177e4 LT |
660 | } |
661 | ||
662 | return 0; | |
663 | } | |
664 | ||
665 | static void free_disk_sb(mdk_rdev_t * rdev) | |
666 | { | |
667 | if (rdev->sb_page) { | |
2d1f3b5d | 668 | put_page(rdev->sb_page); |
1da177e4 LT |
669 | rdev->sb_loaded = 0; |
670 | rdev->sb_page = NULL; | |
0f420358 | 671 | rdev->sb_start = 0; |
dd8ac336 | 672 | rdev->sectors = 0; |
1da177e4 LT |
673 | } |
674 | } | |
675 | ||
676 | ||
6712ecf8 | 677 | static void super_written(struct bio *bio, int error) |
7bfa19f2 N |
678 | { |
679 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 680 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 | 681 | |
3a0f5bbb N |
682 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
683 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
684 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
685 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 686 | md_error(mddev, rdev); |
3a0f5bbb | 687 | } |
7bfa19f2 | 688 | |
a9701a30 N |
689 | if (atomic_dec_and_test(&mddev->pending_writes)) |
690 | wake_up(&mddev->sb_wait); | |
f8b58edf | 691 | bio_put(bio); |
7bfa19f2 N |
692 | } |
693 | ||
6712ecf8 | 694 | static void super_written_barrier(struct bio *bio, int error) |
a9701a30 N |
695 | { |
696 | struct bio *bio2 = bio->bi_private; | |
697 | mdk_rdev_t *rdev = bio2->bi_private; | |
698 | mddev_t *mddev = rdev->mddev; | |
a9701a30 N |
699 | |
700 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
701 | error == -EOPNOTSUPP) { | |
702 | unsigned long flags; | |
703 | /* barriers don't appear to be supported :-( */ | |
704 | set_bit(BarriersNotsupp, &rdev->flags); | |
705 | mddev->barriers_work = 0; | |
706 | spin_lock_irqsave(&mddev->write_lock, flags); | |
707 | bio2->bi_next = mddev->biolist; | |
708 | mddev->biolist = bio2; | |
709 | spin_unlock_irqrestore(&mddev->write_lock, flags); | |
710 | wake_up(&mddev->sb_wait); | |
711 | bio_put(bio); | |
6712ecf8 N |
712 | } else { |
713 | bio_put(bio2); | |
714 | bio->bi_private = rdev; | |
715 | super_written(bio, error); | |
a9701a30 | 716 | } |
a9701a30 N |
717 | } |
718 | ||
7bfa19f2 N |
719 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, |
720 | sector_t sector, int size, struct page *page) | |
721 | { | |
722 | /* write first size bytes of page to sector of rdev | |
723 | * Increment mddev->pending_writes before returning | |
724 | * and decrement it on completion, waking up sb_wait | |
725 | * if zero is reached. | |
726 | * If an error occurred, call md_error | |
a9701a30 N |
727 | * |
728 | * As we might need to resubmit the request if BIO_RW_BARRIER | |
729 | * causes ENOTSUPP, we allocate a spare bio... | |
7bfa19f2 N |
730 | */ |
731 | struct bio *bio = bio_alloc(GFP_NOIO, 1); | |
93dbb393 | 732 | int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG); |
7bfa19f2 N |
733 | |
734 | bio->bi_bdev = rdev->bdev; | |
735 | bio->bi_sector = sector; | |
736 | bio_add_page(bio, page, size, 0); | |
737 | bio->bi_private = rdev; | |
738 | bio->bi_end_io = super_written; | |
a9701a30 N |
739 | bio->bi_rw = rw; |
740 | ||
7bfa19f2 | 741 | atomic_inc(&mddev->pending_writes); |
a9701a30 N |
742 | if (!test_bit(BarriersNotsupp, &rdev->flags)) { |
743 | struct bio *rbio; | |
744 | rw |= (1<<BIO_RW_BARRIER); | |
745 | rbio = bio_clone(bio, GFP_NOIO); | |
746 | rbio->bi_private = bio; | |
747 | rbio->bi_end_io = super_written_barrier; | |
748 | submit_bio(rw, rbio); | |
749 | } else | |
750 | submit_bio(rw, bio); | |
751 | } | |
752 | ||
753 | void md_super_wait(mddev_t *mddev) | |
754 | { | |
755 | /* wait for all superblock writes that were scheduled to complete. | |
756 | * if any had to be retried (due to BARRIER problems), retry them | |
757 | */ | |
758 | DEFINE_WAIT(wq); | |
759 | for(;;) { | |
760 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
761 | if (atomic_read(&mddev->pending_writes)==0) | |
762 | break; | |
763 | while (mddev->biolist) { | |
764 | struct bio *bio; | |
765 | spin_lock_irq(&mddev->write_lock); | |
766 | bio = mddev->biolist; | |
767 | mddev->biolist = bio->bi_next ; | |
768 | bio->bi_next = NULL; | |
769 | spin_unlock_irq(&mddev->write_lock); | |
770 | submit_bio(bio->bi_rw, bio); | |
771 | } | |
772 | schedule(); | |
773 | } | |
774 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
775 | } |
776 | ||
6712ecf8 | 777 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 778 | { |
1da177e4 | 779 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
780 | } |
781 | ||
a654b9d8 | 782 | int sync_page_io(struct block_device *bdev, sector_t sector, int size, |
1da177e4 LT |
783 | struct page *page, int rw) |
784 | { | |
baaa2c51 | 785 | struct bio *bio = bio_alloc(GFP_NOIO, 1); |
1da177e4 LT |
786 | struct completion event; |
787 | int ret; | |
788 | ||
93dbb393 | 789 | rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG); |
1da177e4 LT |
790 | |
791 | bio->bi_bdev = bdev; | |
792 | bio->bi_sector = sector; | |
793 | bio_add_page(bio, page, size, 0); | |
794 | init_completion(&event); | |
795 | bio->bi_private = &event; | |
796 | bio->bi_end_io = bi_complete; | |
797 | submit_bio(rw, bio); | |
798 | wait_for_completion(&event); | |
799 | ||
800 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
801 | bio_put(bio); | |
802 | return ret; | |
803 | } | |
a8745db2 | 804 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 805 | |
0002b271 | 806 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
807 | { |
808 | char b[BDEVNAME_SIZE]; | |
809 | if (!rdev->sb_page) { | |
810 | MD_BUG(); | |
811 | return -EINVAL; | |
812 | } | |
813 | if (rdev->sb_loaded) | |
814 | return 0; | |
815 | ||
816 | ||
0f420358 | 817 | if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ)) |
1da177e4 LT |
818 | goto fail; |
819 | rdev->sb_loaded = 1; | |
820 | return 0; | |
821 | ||
822 | fail: | |
823 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
824 | bdevname(rdev->bdev,b)); | |
825 | return -EINVAL; | |
826 | } | |
827 | ||
828 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
829 | { | |
05710466 AN |
830 | return sb1->set_uuid0 == sb2->set_uuid0 && |
831 | sb1->set_uuid1 == sb2->set_uuid1 && | |
832 | sb1->set_uuid2 == sb2->set_uuid2 && | |
833 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
834 | } |
835 | ||
1da177e4 LT |
836 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
837 | { | |
838 | int ret; | |
839 | mdp_super_t *tmp1, *tmp2; | |
840 | ||
841 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
842 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
843 | ||
844 | if (!tmp1 || !tmp2) { | |
845 | ret = 0; | |
35020f1a | 846 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
847 | goto abort; |
848 | } | |
849 | ||
850 | *tmp1 = *sb1; | |
851 | *tmp2 = *sb2; | |
852 | ||
853 | /* | |
854 | * nr_disks is not constant | |
855 | */ | |
856 | tmp1->nr_disks = 0; | |
857 | tmp2->nr_disks = 0; | |
858 | ||
ce0c8e05 | 859 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 860 | abort: |
990a8baf JJ |
861 | kfree(tmp1); |
862 | kfree(tmp2); | |
1da177e4 LT |
863 | return ret; |
864 | } | |
865 | ||
4d167f09 N |
866 | |
867 | static u32 md_csum_fold(u32 csum) | |
868 | { | |
869 | csum = (csum & 0xffff) + (csum >> 16); | |
870 | return (csum & 0xffff) + (csum >> 16); | |
871 | } | |
872 | ||
1da177e4 LT |
873 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
874 | { | |
4d167f09 N |
875 | u64 newcsum = 0; |
876 | u32 *sb32 = (u32*)sb; | |
877 | int i; | |
1da177e4 LT |
878 | unsigned int disk_csum, csum; |
879 | ||
880 | disk_csum = sb->sb_csum; | |
881 | sb->sb_csum = 0; | |
4d167f09 N |
882 | |
883 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
884 | newcsum += sb32[i]; | |
885 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
886 | ||
887 | ||
888 | #ifdef CONFIG_ALPHA | |
889 | /* This used to use csum_partial, which was wrong for several | |
890 | * reasons including that different results are returned on | |
891 | * different architectures. It isn't critical that we get exactly | |
892 | * the same return value as before (we always csum_fold before | |
893 | * testing, and that removes any differences). However as we | |
894 | * know that csum_partial always returned a 16bit value on | |
895 | * alphas, do a fold to maximise conformity to previous behaviour. | |
896 | */ | |
897 | sb->sb_csum = md_csum_fold(disk_csum); | |
898 | #else | |
1da177e4 | 899 | sb->sb_csum = disk_csum; |
4d167f09 | 900 | #endif |
1da177e4 LT |
901 | return csum; |
902 | } | |
903 | ||
904 | ||
905 | /* | |
906 | * Handle superblock details. | |
907 | * We want to be able to handle multiple superblock formats | |
908 | * so we have a common interface to them all, and an array of | |
909 | * different handlers. | |
910 | * We rely on user-space to write the initial superblock, and support | |
911 | * reading and updating of superblocks. | |
912 | * Interface methods are: | |
913 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
914 | * loads and validates a superblock on dev. | |
915 | * if refdev != NULL, compare superblocks on both devices | |
916 | * Return: | |
917 | * 0 - dev has a superblock that is compatible with refdev | |
918 | * 1 - dev has a superblock that is compatible and newer than refdev | |
919 | * so dev should be used as the refdev in future | |
920 | * -EINVAL superblock incompatible or invalid | |
921 | * -othererror e.g. -EIO | |
922 | * | |
923 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
924 | * Verify that dev is acceptable into mddev. | |
925 | * The first time, mddev->raid_disks will be 0, and data from | |
926 | * dev should be merged in. Subsequent calls check that dev | |
927 | * is new enough. Return 0 or -EINVAL | |
928 | * | |
929 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
930 | * Update the superblock for rdev with data in mddev | |
931 | * This does not write to disc. | |
932 | * | |
933 | */ | |
934 | ||
935 | struct super_type { | |
0cd17fec CW |
936 | char *name; |
937 | struct module *owner; | |
938 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, | |
939 | int minor_version); | |
940 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
941 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
942 | unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, | |
15f4a5fd | 943 | sector_t num_sectors); |
1da177e4 LT |
944 | }; |
945 | ||
0894cc30 AN |
946 | /* |
947 | * Check that the given mddev has no bitmap. | |
948 | * | |
949 | * This function is called from the run method of all personalities that do not | |
950 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
951 | * has a bitmap. Otherwise, it returns 0. | |
952 | * | |
953 | */ | |
954 | int md_check_no_bitmap(mddev_t *mddev) | |
955 | { | |
c3d9714e | 956 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
957 | return 0; |
958 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
959 | mdname(mddev), mddev->pers->name); | |
960 | return 1; | |
961 | } | |
962 | EXPORT_SYMBOL(md_check_no_bitmap); | |
963 | ||
1da177e4 LT |
964 | /* |
965 | * load_super for 0.90.0 | |
966 | */ | |
967 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
968 | { | |
969 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
970 | mdp_super_t *sb; | |
971 | int ret; | |
1da177e4 LT |
972 | |
973 | /* | |
0f420358 | 974 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
975 | * it's at the end of the disk. |
976 | * | |
977 | * It also happens to be a multiple of 4Kb. | |
978 | */ | |
0f420358 | 979 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 980 | |
0002b271 | 981 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
982 | if (ret) return ret; |
983 | ||
984 | ret = -EINVAL; | |
985 | ||
986 | bdevname(rdev->bdev, b); | |
987 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
988 | ||
989 | if (sb->md_magic != MD_SB_MAGIC) { | |
990 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
991 | b); | |
992 | goto abort; | |
993 | } | |
994 | ||
995 | if (sb->major_version != 0 || | |
f6705578 N |
996 | sb->minor_version < 90 || |
997 | sb->minor_version > 91) { | |
1da177e4 LT |
998 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
999 | sb->major_version, sb->minor_version, | |
1000 | b); | |
1001 | goto abort; | |
1002 | } | |
1003 | ||
1004 | if (sb->raid_disks <= 0) | |
1005 | goto abort; | |
1006 | ||
4d167f09 | 1007 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
1008 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
1009 | b); | |
1010 | goto abort; | |
1011 | } | |
1012 | ||
1013 | rdev->preferred_minor = sb->md_minor; | |
1014 | rdev->data_offset = 0; | |
0002b271 | 1015 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 LT |
1016 | |
1017 | if (sb->level == LEVEL_MULTIPATH) | |
1018 | rdev->desc_nr = -1; | |
1019 | else | |
1020 | rdev->desc_nr = sb->this_disk.number; | |
1021 | ||
9a7b2b0f | 1022 | if (!refdev) { |
1da177e4 | 1023 | ret = 1; |
9a7b2b0f | 1024 | } else { |
1da177e4 LT |
1025 | __u64 ev1, ev2; |
1026 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
1027 | if (!uuid_equal(refsb, sb)) { | |
1028 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
1029 | b, bdevname(refdev->bdev,b2)); | |
1030 | goto abort; | |
1031 | } | |
1032 | if (!sb_equal(refsb, sb)) { | |
1033 | printk(KERN_WARNING "md: %s has same UUID" | |
1034 | " but different superblock to %s\n", | |
1035 | b, bdevname(refdev->bdev, b2)); | |
1036 | goto abort; | |
1037 | } | |
1038 | ev1 = md_event(sb); | |
1039 | ev2 = md_event(refsb); | |
1040 | if (ev1 > ev2) | |
1041 | ret = 1; | |
1042 | else | |
1043 | ret = 0; | |
1044 | } | |
8190e754 | 1045 | rdev->sectors = rdev->sb_start; |
1da177e4 | 1046 | |
dd8ac336 | 1047 | if (rdev->sectors < sb->size * 2 && sb->level > 1) |
2bf071bf N |
1048 | /* "this cannot possibly happen" ... */ |
1049 | ret = -EINVAL; | |
1050 | ||
1da177e4 LT |
1051 | abort: |
1052 | return ret; | |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * validate_super for 0.90.0 | |
1057 | */ | |
1058 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1059 | { | |
1060 | mdp_disk_t *desc; | |
1061 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 1062 | __u64 ev1 = md_event(sb); |
1da177e4 | 1063 | |
41158c7e | 1064 | rdev->raid_disk = -1; |
c5d79adb N |
1065 | clear_bit(Faulty, &rdev->flags); |
1066 | clear_bit(In_sync, &rdev->flags); | |
1067 | clear_bit(WriteMostly, &rdev->flags); | |
1068 | clear_bit(BarriersNotsupp, &rdev->flags); | |
1069 | ||
1da177e4 LT |
1070 | if (mddev->raid_disks == 0) { |
1071 | mddev->major_version = 0; | |
1072 | mddev->minor_version = sb->minor_version; | |
1073 | mddev->patch_version = sb->patch_version; | |
e691063a | 1074 | mddev->external = 0; |
9d8f0363 | 1075 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1076 | mddev->ctime = sb->ctime; |
1077 | mddev->utime = sb->utime; | |
1078 | mddev->level = sb->level; | |
d9d166c2 | 1079 | mddev->clevel[0] = 0; |
1da177e4 LT |
1080 | mddev->layout = sb->layout; |
1081 | mddev->raid_disks = sb->raid_disks; | |
58c0fed4 | 1082 | mddev->dev_sectors = sb->size * 2; |
07d84d10 | 1083 | mddev->events = ev1; |
c3d9714e N |
1084 | mddev->bitmap_info.offset = 0; |
1085 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; | |
1da177e4 | 1086 | |
f6705578 N |
1087 | if (mddev->minor_version >= 91) { |
1088 | mddev->reshape_position = sb->reshape_position; | |
1089 | mddev->delta_disks = sb->delta_disks; | |
1090 | mddev->new_level = sb->new_level; | |
1091 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1092 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
f6705578 N |
1093 | } else { |
1094 | mddev->reshape_position = MaxSector; | |
1095 | mddev->delta_disks = 0; | |
1096 | mddev->new_level = mddev->level; | |
1097 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1098 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1099 | } |
1100 | ||
1da177e4 LT |
1101 | if (sb->state & (1<<MD_SB_CLEAN)) |
1102 | mddev->recovery_cp = MaxSector; | |
1103 | else { | |
1104 | if (sb->events_hi == sb->cp_events_hi && | |
1105 | sb->events_lo == sb->cp_events_lo) { | |
1106 | mddev->recovery_cp = sb->recovery_cp; | |
1107 | } else | |
1108 | mddev->recovery_cp = 0; | |
1109 | } | |
1110 | ||
1111 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1112 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1113 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1114 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1115 | ||
1116 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1117 | |
1118 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
c3d9714e N |
1119 | mddev->bitmap_info.file == NULL) |
1120 | mddev->bitmap_info.offset = | |
1121 | mddev->bitmap_info.default_offset; | |
a654b9d8 | 1122 | |
41158c7e | 1123 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1124 | /* Insist on good event counter while assembling, except |
1125 | * for spares (which don't need an event count) */ | |
1da177e4 | 1126 | ++ev1; |
be6800a7 N |
1127 | if (sb->disks[rdev->desc_nr].state & ( |
1128 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
1129 | if (ev1 < mddev->events) | |
1130 | return -EINVAL; | |
41158c7e N |
1131 | } else if (mddev->bitmap) { |
1132 | /* if adding to array with a bitmap, then we can accept an | |
1133 | * older device ... but not too old. | |
1134 | */ | |
41158c7e N |
1135 | if (ev1 < mddev->bitmap->events_cleared) |
1136 | return 0; | |
07d84d10 N |
1137 | } else { |
1138 | if (ev1 < mddev->events) | |
1139 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1140 | return 0; | |
1141 | } | |
41158c7e | 1142 | |
1da177e4 | 1143 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1144 | desc = sb->disks + rdev->desc_nr; |
1145 | ||
1146 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1147 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1148 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1149 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1150 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1151 | rdev->raid_disk = desc->raid_disk; |
0261cd9f N |
1152 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1153 | /* active but not in sync implies recovery up to | |
1154 | * reshape position. We don't know exactly where | |
1155 | * that is, so set to zero for now */ | |
1156 | if (mddev->minor_version >= 91) { | |
1157 | rdev->recovery_offset = 0; | |
1158 | rdev->raid_disk = desc->raid_disk; | |
1159 | } | |
1da177e4 | 1160 | } |
8ddf9efe N |
1161 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1162 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1163 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1164 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1165 | return 0; |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * sync_super for 0.90.0 | |
1170 | */ | |
1171 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1172 | { | |
1173 | mdp_super_t *sb; | |
1da177e4 LT |
1174 | mdk_rdev_t *rdev2; |
1175 | int next_spare = mddev->raid_disks; | |
19133a42 | 1176 | |
1da177e4 LT |
1177 | |
1178 | /* make rdev->sb match mddev data.. | |
1179 | * | |
1180 | * 1/ zero out disks | |
1181 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1182 | * 3/ any empty disks < next_spare become removed | |
1183 | * | |
1184 | * disks[0] gets initialised to REMOVED because | |
1185 | * we cannot be sure from other fields if it has | |
1186 | * been initialised or not. | |
1187 | */ | |
1188 | int i; | |
1189 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1190 | ||
61181565 N |
1191 | rdev->sb_size = MD_SB_BYTES; |
1192 | ||
1da177e4 LT |
1193 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
1194 | ||
1195 | memset(sb, 0, sizeof(*sb)); | |
1196 | ||
1197 | sb->md_magic = MD_SB_MAGIC; | |
1198 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1199 | sb->patch_version = mddev->patch_version; |
1200 | sb->gvalid_words = 0; /* ignored */ | |
1201 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1202 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1203 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1204 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1205 | ||
1206 | sb->ctime = mddev->ctime; | |
1207 | sb->level = mddev->level; | |
58c0fed4 | 1208 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1209 | sb->raid_disks = mddev->raid_disks; |
1210 | sb->md_minor = mddev->md_minor; | |
e691063a | 1211 | sb->not_persistent = 0; |
1da177e4 LT |
1212 | sb->utime = mddev->utime; |
1213 | sb->state = 0; | |
1214 | sb->events_hi = (mddev->events>>32); | |
1215 | sb->events_lo = (u32)mddev->events; | |
1216 | ||
f6705578 N |
1217 | if (mddev->reshape_position == MaxSector) |
1218 | sb->minor_version = 90; | |
1219 | else { | |
1220 | sb->minor_version = 91; | |
1221 | sb->reshape_position = mddev->reshape_position; | |
1222 | sb->new_level = mddev->new_level; | |
1223 | sb->delta_disks = mddev->delta_disks; | |
1224 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1225 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1226 | } |
1227 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1228 | if (mddev->in_sync) |
1229 | { | |
1230 | sb->recovery_cp = mddev->recovery_cp; | |
1231 | sb->cp_events_hi = (mddev->events>>32); | |
1232 | sb->cp_events_lo = (u32)mddev->events; | |
1233 | if (mddev->recovery_cp == MaxSector) | |
1234 | sb->state = (1<< MD_SB_CLEAN); | |
1235 | } else | |
1236 | sb->recovery_cp = 0; | |
1237 | ||
1238 | sb->layout = mddev->layout; | |
9d8f0363 | 1239 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1240 | |
c3d9714e | 1241 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1242 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1243 | ||
1da177e4 | 1244 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
159ec1fc | 1245 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1246 | mdp_disk_t *d; |
86e6ffdd | 1247 | int desc_nr; |
0261cd9f N |
1248 | int is_active = test_bit(In_sync, &rdev2->flags); |
1249 | ||
1250 | if (rdev2->raid_disk >= 0 && | |
1251 | sb->minor_version >= 91) | |
1252 | /* we have nowhere to store the recovery_offset, | |
1253 | * but if it is not below the reshape_position, | |
1254 | * we can piggy-back on that. | |
1255 | */ | |
1256 | is_active = 1; | |
1257 | if (rdev2->raid_disk < 0 || | |
1258 | test_bit(Faulty, &rdev2->flags)) | |
1259 | is_active = 0; | |
1260 | if (is_active) | |
86e6ffdd | 1261 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1262 | else |
86e6ffdd | 1263 | desc_nr = next_spare++; |
19133a42 | 1264 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1265 | d = &sb->disks[rdev2->desc_nr]; |
1266 | nr_disks++; | |
1267 | d->number = rdev2->desc_nr; | |
1268 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1269 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1270 | if (is_active) |
1da177e4 LT |
1271 | d->raid_disk = rdev2->raid_disk; |
1272 | else | |
1273 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1274 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1275 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1276 | else if (is_active) { |
1da177e4 | 1277 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1278 | if (test_bit(In_sync, &rdev2->flags)) |
1279 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1280 | active++; |
1281 | working++; | |
1282 | } else { | |
1283 | d->state = 0; | |
1284 | spare++; | |
1285 | working++; | |
1286 | } | |
8ddf9efe N |
1287 | if (test_bit(WriteMostly, &rdev2->flags)) |
1288 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1289 | } |
1da177e4 LT |
1290 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1291 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1292 | mdp_disk_t *d = &sb->disks[i]; | |
1293 | if (d->state == 0 && d->number == 0) { | |
1294 | d->number = i; | |
1295 | d->raid_disk = i; | |
1296 | d->state = (1<<MD_DISK_REMOVED); | |
1297 | d->state |= (1<<MD_DISK_FAULTY); | |
1298 | failed++; | |
1299 | } | |
1300 | } | |
1301 | sb->nr_disks = nr_disks; | |
1302 | sb->active_disks = active; | |
1303 | sb->working_disks = working; | |
1304 | sb->failed_disks = failed; | |
1305 | sb->spare_disks = spare; | |
1306 | ||
1307 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1308 | sb->sb_csum = calc_sb_csum(sb); | |
1309 | } | |
1310 | ||
0cd17fec CW |
1311 | /* |
1312 | * rdev_size_change for 0.90.0 | |
1313 | */ | |
1314 | static unsigned long long | |
15f4a5fd | 1315 | super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec | 1316 | { |
58c0fed4 | 1317 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1318 | return 0; /* component must fit device */ |
c3d9714e | 1319 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1320 | return 0; /* can't move bitmap */ |
0f420358 | 1321 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
15f4a5fd AN |
1322 | if (!num_sectors || num_sectors > rdev->sb_start) |
1323 | num_sectors = rdev->sb_start; | |
0f420358 | 1324 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1325 | rdev->sb_page); |
1326 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1327 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec CW |
1328 | } |
1329 | ||
1330 | ||
1da177e4 LT |
1331 | /* |
1332 | * version 1 superblock | |
1333 | */ | |
1334 | ||
1c05b4bc | 1335 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1336 | { |
1c05b4bc N |
1337 | __le32 disk_csum; |
1338 | u32 csum; | |
1da177e4 LT |
1339 | unsigned long long newcsum; |
1340 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1341 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1342 | int i; |
1343 | ||
1344 | disk_csum = sb->sb_csum; | |
1345 | sb->sb_csum = 0; | |
1346 | newcsum = 0; | |
1347 | for (i=0; size>=4; size -= 4 ) | |
1348 | newcsum += le32_to_cpu(*isuper++); | |
1349 | ||
1350 | if (size == 2) | |
1c05b4bc | 1351 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1352 | |
1353 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1354 | sb->sb_csum = disk_csum; | |
1355 | return cpu_to_le32(csum); | |
1356 | } | |
1357 | ||
1358 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1359 | { | |
1360 | struct mdp_superblock_1 *sb; | |
1361 | int ret; | |
0f420358 | 1362 | sector_t sb_start; |
1da177e4 | 1363 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1364 | int bmask; |
1da177e4 LT |
1365 | |
1366 | /* | |
0f420358 | 1367 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1368 | * It is always aligned to a 4K boundary and |
1369 | * depeding on minor_version, it can be: | |
1370 | * 0: At least 8K, but less than 12K, from end of device | |
1371 | * 1: At start of device | |
1372 | * 2: 4K from start of device. | |
1373 | */ | |
1374 | switch(minor_version) { | |
1375 | case 0: | |
0f420358 AN |
1376 | sb_start = rdev->bdev->bd_inode->i_size >> 9; |
1377 | sb_start -= 8*2; | |
1378 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1379 | break; |
1380 | case 1: | |
0f420358 | 1381 | sb_start = 0; |
1da177e4 LT |
1382 | break; |
1383 | case 2: | |
0f420358 | 1384 | sb_start = 8; |
1da177e4 LT |
1385 | break; |
1386 | default: | |
1387 | return -EINVAL; | |
1388 | } | |
0f420358 | 1389 | rdev->sb_start = sb_start; |
1da177e4 | 1390 | |
0002b271 N |
1391 | /* superblock is rarely larger than 1K, but it can be larger, |
1392 | * and it is safe to read 4k, so we do that | |
1393 | */ | |
1394 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1395 | if (ret) return ret; |
1396 | ||
1397 | ||
1398 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1399 | ||
1400 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1401 | sb->major_version != cpu_to_le32(1) || | |
1402 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1403 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1404 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1405 | return -EINVAL; |
1406 | ||
1407 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1408 | printk("md: invalid superblock checksum on %s\n", | |
1409 | bdevname(rdev->bdev,b)); | |
1410 | return -EINVAL; | |
1411 | } | |
1412 | if (le64_to_cpu(sb->data_size) < 10) { | |
1413 | printk("md: data_size too small on %s\n", | |
1414 | bdevname(rdev->bdev,b)); | |
1415 | return -EINVAL; | |
1416 | } | |
e11e93fa | 1417 | |
1da177e4 LT |
1418 | rdev->preferred_minor = 0xffff; |
1419 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1420 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1421 | |
0002b271 | 1422 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1423 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1424 | if (rdev->sb_size & bmask) |
a1801f85 N |
1425 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1426 | ||
1427 | if (minor_version | |
0f420358 | 1428 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1429 | return -EINVAL; |
0002b271 | 1430 | |
31b65a0d N |
1431 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1432 | rdev->desc_nr = -1; | |
1433 | else | |
1434 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1435 | ||
9a7b2b0f | 1436 | if (!refdev) { |
8ed75463 | 1437 | ret = 1; |
9a7b2b0f | 1438 | } else { |
1da177e4 LT |
1439 | __u64 ev1, ev2; |
1440 | struct mdp_superblock_1 *refsb = | |
1441 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1442 | ||
1443 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1444 | sb->level != refsb->level || | |
1445 | sb->layout != refsb->layout || | |
1446 | sb->chunksize != refsb->chunksize) { | |
1447 | printk(KERN_WARNING "md: %s has strangely different" | |
1448 | " superblock to %s\n", | |
1449 | bdevname(rdev->bdev,b), | |
1450 | bdevname(refdev->bdev,b2)); | |
1451 | return -EINVAL; | |
1452 | } | |
1453 | ev1 = le64_to_cpu(sb->events); | |
1454 | ev2 = le64_to_cpu(refsb->events); | |
1455 | ||
1456 | if (ev1 > ev2) | |
8ed75463 N |
1457 | ret = 1; |
1458 | else | |
1459 | ret = 0; | |
1da177e4 | 1460 | } |
a1801f85 | 1461 | if (minor_version) |
dd8ac336 AN |
1462 | rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) - |
1463 | le64_to_cpu(sb->data_offset); | |
1da177e4 | 1464 | else |
dd8ac336 AN |
1465 | rdev->sectors = rdev->sb_start; |
1466 | if (rdev->sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1467 | return -EINVAL; |
dd8ac336 | 1468 | rdev->sectors = le64_to_cpu(sb->data_size); |
dd8ac336 | 1469 | if (le64_to_cpu(sb->size) > rdev->sectors) |
2bf071bf | 1470 | return -EINVAL; |
8ed75463 | 1471 | return ret; |
1da177e4 LT |
1472 | } |
1473 | ||
1474 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1475 | { | |
1476 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1477 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1478 | |
41158c7e | 1479 | rdev->raid_disk = -1; |
c5d79adb N |
1480 | clear_bit(Faulty, &rdev->flags); |
1481 | clear_bit(In_sync, &rdev->flags); | |
1482 | clear_bit(WriteMostly, &rdev->flags); | |
1483 | clear_bit(BarriersNotsupp, &rdev->flags); | |
1484 | ||
1da177e4 LT |
1485 | if (mddev->raid_disks == 0) { |
1486 | mddev->major_version = 1; | |
1487 | mddev->patch_version = 0; | |
e691063a | 1488 | mddev->external = 0; |
9d8f0363 | 1489 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1490 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1491 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1492 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1493 | mddev->clevel[0] = 0; |
1da177e4 LT |
1494 | mddev->layout = le32_to_cpu(sb->layout); |
1495 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1496 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1497 | mddev->events = ev1; |
c3d9714e N |
1498 | mddev->bitmap_info.offset = 0; |
1499 | mddev->bitmap_info.default_offset = 1024 >> 9; | |
1da177e4 LT |
1500 | |
1501 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1502 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1503 | ||
1504 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1505 | |
71c0805c | 1506 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
c3d9714e N |
1507 | mddev->bitmap_info.file == NULL ) |
1508 | mddev->bitmap_info.offset = | |
1509 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
e11e93fa | 1510 | |
f6705578 N |
1511 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1512 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1513 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1514 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1515 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1516 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
f6705578 N |
1517 | } else { |
1518 | mddev->reshape_position = MaxSector; | |
1519 | mddev->delta_disks = 0; | |
1520 | mddev->new_level = mddev->level; | |
1521 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1522 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1523 | } |
1524 | ||
41158c7e | 1525 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1526 | /* Insist of good event counter while assembling, except for |
1527 | * spares (which don't need an event count) */ | |
1da177e4 | 1528 | ++ev1; |
be6800a7 N |
1529 | if (rdev->desc_nr >= 0 && |
1530 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
1531 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) | |
1532 | if (ev1 < mddev->events) | |
1533 | return -EINVAL; | |
41158c7e N |
1534 | } else if (mddev->bitmap) { |
1535 | /* If adding to array with a bitmap, then we can accept an | |
1536 | * older device, but not too old. | |
1537 | */ | |
41158c7e N |
1538 | if (ev1 < mddev->bitmap->events_cleared) |
1539 | return 0; | |
07d84d10 N |
1540 | } else { |
1541 | if (ev1 < mddev->events) | |
1542 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1543 | return 0; | |
1544 | } | |
1da177e4 LT |
1545 | if (mddev->level != LEVEL_MULTIPATH) { |
1546 | int role; | |
3673f305 N |
1547 | if (rdev->desc_nr < 0 || |
1548 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1549 | role = 0xffff; | |
1550 | rdev->desc_nr = -1; | |
1551 | } else | |
1552 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1553 | switch(role) { |
1554 | case 0xffff: /* spare */ | |
1da177e4 LT |
1555 | break; |
1556 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1557 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1558 | break; |
1559 | default: | |
5fd6c1dc N |
1560 | if ((le32_to_cpu(sb->feature_map) & |
1561 | MD_FEATURE_RECOVERY_OFFSET)) | |
1562 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1563 | else | |
1564 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1565 | rdev->raid_disk = role; |
1566 | break; | |
1567 | } | |
8ddf9efe N |
1568 | if (sb->devflags & WriteMostly1) |
1569 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1570 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1571 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1572 | |
1da177e4 LT |
1573 | return 0; |
1574 | } | |
1575 | ||
1576 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1577 | { | |
1578 | struct mdp_superblock_1 *sb; | |
1da177e4 LT |
1579 | mdk_rdev_t *rdev2; |
1580 | int max_dev, i; | |
1581 | /* make rdev->sb match mddev and rdev data. */ | |
1582 | ||
1583 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1584 | ||
1585 | sb->feature_map = 0; | |
1586 | sb->pad0 = 0; | |
5fd6c1dc | 1587 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1588 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1589 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1590 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1591 | ||
1592 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1593 | sb->events = cpu_to_le64(mddev->events); | |
1594 | if (mddev->in_sync) | |
1595 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1596 | else | |
1597 | sb->resync_offset = cpu_to_le64(0); | |
1598 | ||
1c05b4bc | 1599 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1600 | |
f0ca340c | 1601 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1602 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1603 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1604 | sb->level = cpu_to_le32(mddev->level); |
1605 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1606 | |
c3d9714e N |
1607 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1608 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1609 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1610 | } |
5fd6c1dc N |
1611 | |
1612 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1613 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1614 | sb->feature_map |= |
1615 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1616 | sb->recovery_offset = | |
1617 | cpu_to_le64(rdev->recovery_offset); | |
5fd6c1dc N |
1618 | } |
1619 | ||
f6705578 N |
1620 | if (mddev->reshape_position != MaxSector) { |
1621 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1622 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1623 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1624 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1625 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1626 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
f6705578 | 1627 | } |
a654b9d8 | 1628 | |
1da177e4 | 1629 | max_dev = 0; |
159ec1fc | 1630 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
1da177e4 LT |
1631 | if (rdev2->desc_nr+1 > max_dev) |
1632 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1633 | |
70471daf N |
1634 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1635 | int bmask; | |
a778b73f | 1636 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1637 | rdev->sb_size = max_dev * 2 + 256; |
1638 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1639 | if (rdev->sb_size & bmask) | |
1640 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
1641 | } | |
1da177e4 LT |
1642 | for (i=0; i<max_dev;i++) |
1643 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1644 | ||
159ec1fc | 1645 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1646 | i = rdev2->desc_nr; |
b2d444d7 | 1647 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1648 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1649 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1650 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1651 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1652 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1653 | else |
1654 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1655 | } | |
1656 | ||
1da177e4 LT |
1657 | sb->sb_csum = calc_sb_1_csum(sb); |
1658 | } | |
1659 | ||
0cd17fec | 1660 | static unsigned long long |
15f4a5fd | 1661 | super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec CW |
1662 | { |
1663 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1664 | sector_t max_sectors; |
58c0fed4 | 1665 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1666 | return 0; /* component must fit device */ |
0f420358 | 1667 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1668 | /* minor versions 1 and 2; superblock before data */ |
15f4a5fd AN |
1669 | max_sectors = rdev->bdev->bd_inode->i_size >> 9; |
1670 | max_sectors -= rdev->data_offset; | |
1671 | if (!num_sectors || num_sectors > max_sectors) | |
1672 | num_sectors = max_sectors; | |
c3d9714e | 1673 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1674 | /* minor version 0 with bitmap we can't move */ |
1675 | return 0; | |
1676 | } else { | |
1677 | /* minor version 0; superblock after data */ | |
0f420358 AN |
1678 | sector_t sb_start; |
1679 | sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2; | |
1680 | sb_start &= ~(sector_t)(4*2 - 1); | |
dd8ac336 | 1681 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1682 | if (!num_sectors || num_sectors > max_sectors) |
1683 | num_sectors = max_sectors; | |
0f420358 | 1684 | rdev->sb_start = sb_start; |
0cd17fec CW |
1685 | } |
1686 | sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); | |
15f4a5fd | 1687 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1688 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1689 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1690 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1691 | rdev->sb_page); |
1692 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1693 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec | 1694 | } |
1da177e4 | 1695 | |
75c96f85 | 1696 | static struct super_type super_types[] = { |
1da177e4 LT |
1697 | [0] = { |
1698 | .name = "0.90.0", | |
1699 | .owner = THIS_MODULE, | |
0cd17fec CW |
1700 | .load_super = super_90_load, |
1701 | .validate_super = super_90_validate, | |
1702 | .sync_super = super_90_sync, | |
1703 | .rdev_size_change = super_90_rdev_size_change, | |
1da177e4 LT |
1704 | }, |
1705 | [1] = { | |
1706 | .name = "md-1", | |
1707 | .owner = THIS_MODULE, | |
0cd17fec CW |
1708 | .load_super = super_1_load, |
1709 | .validate_super = super_1_validate, | |
1710 | .sync_super = super_1_sync, | |
1711 | .rdev_size_change = super_1_rdev_size_change, | |
1da177e4 LT |
1712 | }, |
1713 | }; | |
1da177e4 LT |
1714 | |
1715 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1716 | { | |
7dd5e7c3 | 1717 | mdk_rdev_t *rdev, *rdev2; |
1da177e4 | 1718 | |
4b80991c N |
1719 | rcu_read_lock(); |
1720 | rdev_for_each_rcu(rdev, mddev1) | |
1721 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1722 | if (rdev->bdev->bd_contains == |
4b80991c N |
1723 | rdev2->bdev->bd_contains) { |
1724 | rcu_read_unlock(); | |
7dd5e7c3 | 1725 | return 1; |
4b80991c N |
1726 | } |
1727 | rcu_read_unlock(); | |
1da177e4 LT |
1728 | return 0; |
1729 | } | |
1730 | ||
1731 | static LIST_HEAD(pending_raid_disks); | |
1732 | ||
ac5e7113 AN |
1733 | /* |
1734 | * Try to register data integrity profile for an mddev | |
1735 | * | |
1736 | * This is called when an array is started and after a disk has been kicked | |
1737 | * from the array. It only succeeds if all working and active component devices | |
1738 | * are integrity capable with matching profiles. | |
1739 | */ | |
1740 | int md_integrity_register(mddev_t *mddev) | |
1741 | { | |
1742 | mdk_rdev_t *rdev, *reference = NULL; | |
1743 | ||
1744 | if (list_empty(&mddev->disks)) | |
1745 | return 0; /* nothing to do */ | |
1746 | if (blk_get_integrity(mddev->gendisk)) | |
1747 | return 0; /* already registered */ | |
1748 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
1749 | /* skip spares and non-functional disks */ | |
1750 | if (test_bit(Faulty, &rdev->flags)) | |
1751 | continue; | |
1752 | if (rdev->raid_disk < 0) | |
1753 | continue; | |
1754 | /* | |
1755 | * If at least one rdev is not integrity capable, we can not | |
1756 | * enable data integrity for the md device. | |
1757 | */ | |
1758 | if (!bdev_get_integrity(rdev->bdev)) | |
1759 | return -EINVAL; | |
1760 | if (!reference) { | |
1761 | /* Use the first rdev as the reference */ | |
1762 | reference = rdev; | |
1763 | continue; | |
1764 | } | |
1765 | /* does this rdev's profile match the reference profile? */ | |
1766 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1767 | rdev->bdev->bd_disk) < 0) | |
1768 | return -EINVAL; | |
1769 | } | |
1770 | /* | |
1771 | * All component devices are integrity capable and have matching | |
1772 | * profiles, register the common profile for the md device. | |
1773 | */ | |
1774 | if (blk_integrity_register(mddev->gendisk, | |
1775 | bdev_get_integrity(reference->bdev)) != 0) { | |
1776 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1777 | mdname(mddev)); | |
1778 | return -EINVAL; | |
1779 | } | |
1780 | printk(KERN_NOTICE "md: data integrity on %s enabled\n", | |
1781 | mdname(mddev)); | |
1782 | return 0; | |
1783 | } | |
1784 | EXPORT_SYMBOL(md_integrity_register); | |
1785 | ||
1786 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
1787 | void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev) | |
3f9d99c1 | 1788 | { |
3f9d99c1 | 1789 | struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); |
ac5e7113 | 1790 | struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); |
3f9d99c1 | 1791 | |
ac5e7113 | 1792 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 1793 | return; |
ac5e7113 | 1794 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 1795 | return; |
ac5e7113 AN |
1796 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
1797 | rdev->bdev->bd_disk) >= 0) | |
1798 | return; | |
1799 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
1800 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 1801 | } |
ac5e7113 | 1802 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 1803 | |
1da177e4 LT |
1804 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) |
1805 | { | |
7dd5e7c3 | 1806 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 1807 | struct kobject *ko; |
1edf80d3 | 1808 | char *s; |
5e55e2f5 | 1809 | int err; |
1da177e4 LT |
1810 | |
1811 | if (rdev->mddev) { | |
1812 | MD_BUG(); | |
1813 | return -EINVAL; | |
1814 | } | |
11e2ede0 DW |
1815 | |
1816 | /* prevent duplicates */ | |
1817 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
1818 | return -EEXIST; | |
1819 | ||
dd8ac336 AN |
1820 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
1821 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
1822 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
1823 | if (mddev->pers) { |
1824 | /* Cannot change size, so fail | |
1825 | * If mddev->level <= 0, then we don't care | |
1826 | * about aligning sizes (e.g. linear) | |
1827 | */ | |
1828 | if (mddev->level > 0) | |
1829 | return -ENOSPC; | |
1830 | } else | |
dd8ac336 | 1831 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 1832 | } |
1da177e4 LT |
1833 | |
1834 | /* Verify rdev->desc_nr is unique. | |
1835 | * If it is -1, assign a free number, else | |
1836 | * check number is not in use | |
1837 | */ | |
1838 | if (rdev->desc_nr < 0) { | |
1839 | int choice = 0; | |
1840 | if (mddev->pers) choice = mddev->raid_disks; | |
1841 | while (find_rdev_nr(mddev, choice)) | |
1842 | choice++; | |
1843 | rdev->desc_nr = choice; | |
1844 | } else { | |
1845 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1846 | return -EBUSY; | |
1847 | } | |
de01dfad N |
1848 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
1849 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
1850 | mdname(mddev), mddev->max_disks); | |
1851 | return -EBUSY; | |
1852 | } | |
19133a42 | 1853 | bdevname(rdev->bdev,b); |
649316b2 | 1854 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 1855 | *s = '!'; |
649316b2 | 1856 | |
1da177e4 | 1857 | rdev->mddev = mddev; |
19133a42 | 1858 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1859 | |
b2d6db58 | 1860 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 1861 | goto fail; |
86e6ffdd | 1862 | |
0762b8bd | 1863 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
1864 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
1865 | /* failure here is OK */; | |
1866 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 1867 | |
4b80991c | 1868 | list_add_rcu(&rdev->same_set, &mddev->disks); |
c5d79adb | 1869 | bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk); |
4044ba58 N |
1870 | |
1871 | /* May as well allow recovery to be retried once */ | |
1872 | mddev->recovery_disabled = 0; | |
3f9d99c1 | 1873 | |
1da177e4 | 1874 | return 0; |
5e55e2f5 N |
1875 | |
1876 | fail: | |
1877 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
1878 | b, mdname(mddev)); | |
1879 | return err; | |
1da177e4 LT |
1880 | } |
1881 | ||
177a99b2 | 1882 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 N |
1883 | { |
1884 | mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); | |
1885 | kobject_del(&rdev->kobj); | |
177a99b2 | 1886 | kobject_put(&rdev->kobj); |
5792a285 N |
1887 | } |
1888 | ||
1da177e4 LT |
1889 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) |
1890 | { | |
1891 | char b[BDEVNAME_SIZE]; | |
1892 | if (!rdev->mddev) { | |
1893 | MD_BUG(); | |
1894 | return; | |
1895 | } | |
5463c790 | 1896 | bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 1897 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
1898 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1899 | rdev->mddev = NULL; | |
86e6ffdd | 1900 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
1901 | sysfs_put(rdev->sysfs_state); |
1902 | rdev->sysfs_state = NULL; | |
5792a285 | 1903 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
1904 | * writing to 'remove' to "dev/state". We also need |
1905 | * to delay it due to rcu usage. | |
5792a285 | 1906 | */ |
4b80991c | 1907 | synchronize_rcu(); |
177a99b2 N |
1908 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
1909 | kobject_get(&rdev->kobj); | |
5792a285 | 1910 | schedule_work(&rdev->del_work); |
1da177e4 LT |
1911 | } |
1912 | ||
1913 | /* | |
1914 | * prevent the device from being mounted, repartitioned or | |
1915 | * otherwise reused by a RAID array (or any other kernel | |
1916 | * subsystem), by bd_claiming the device. | |
1917 | */ | |
c5d79adb | 1918 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) |
1da177e4 LT |
1919 | { |
1920 | int err = 0; | |
1921 | struct block_device *bdev; | |
1922 | char b[BDEVNAME_SIZE]; | |
1923 | ||
2e7b651d | 1924 | bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1925 | if (IS_ERR(bdev)) { |
1926 | printk(KERN_ERR "md: could not open %s.\n", | |
1927 | __bdevname(dev, b)); | |
1928 | return PTR_ERR(bdev); | |
1929 | } | |
c5d79adb | 1930 | err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev); |
1da177e4 LT |
1931 | if (err) { |
1932 | printk(KERN_ERR "md: could not bd_claim %s.\n", | |
1933 | bdevname(bdev, b)); | |
9a1c3542 | 1934 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1935 | return err; |
1936 | } | |
c5d79adb N |
1937 | if (!shared) |
1938 | set_bit(AllReserved, &rdev->flags); | |
1da177e4 LT |
1939 | rdev->bdev = bdev; |
1940 | return err; | |
1941 | } | |
1942 | ||
1943 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1944 | { | |
1945 | struct block_device *bdev = rdev->bdev; | |
1946 | rdev->bdev = NULL; | |
1947 | if (!bdev) | |
1948 | MD_BUG(); | |
1949 | bd_release(bdev); | |
9a1c3542 | 1950 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1951 | } |
1952 | ||
1953 | void md_autodetect_dev(dev_t dev); | |
1954 | ||
1955 | static void export_rdev(mdk_rdev_t * rdev) | |
1956 | { | |
1957 | char b[BDEVNAME_SIZE]; | |
1958 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1959 | bdevname(rdev->bdev,b)); | |
1960 | if (rdev->mddev) | |
1961 | MD_BUG(); | |
1962 | free_disk_sb(rdev); | |
1da177e4 | 1963 | #ifndef MODULE |
d0fae18f N |
1964 | if (test_bit(AutoDetected, &rdev->flags)) |
1965 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
1966 | #endif |
1967 | unlock_rdev(rdev); | |
86e6ffdd | 1968 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1969 | } |
1970 | ||
1971 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1972 | { | |
1973 | unbind_rdev_from_array(rdev); | |
1974 | export_rdev(rdev); | |
1975 | } | |
1976 | ||
1977 | static void export_array(mddev_t *mddev) | |
1978 | { | |
159ec1fc | 1979 | mdk_rdev_t *rdev, *tmp; |
1da177e4 | 1980 | |
d089c6af | 1981 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1982 | if (!rdev->mddev) { |
1983 | MD_BUG(); | |
1984 | continue; | |
1985 | } | |
1986 | kick_rdev_from_array(rdev); | |
1987 | } | |
1988 | if (!list_empty(&mddev->disks)) | |
1989 | MD_BUG(); | |
1990 | mddev->raid_disks = 0; | |
1991 | mddev->major_version = 0; | |
1992 | } | |
1993 | ||
1994 | static void print_desc(mdp_disk_t *desc) | |
1995 | { | |
1996 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
1997 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
1998 | } | |
1999 | ||
cd2ac932 | 2000 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
2001 | { |
2002 | int i; | |
2003 | ||
2004 | printk(KERN_INFO | |
2005 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
2006 | sb->major_version, sb->minor_version, sb->patch_version, | |
2007 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
2008 | sb->ctime); | |
2009 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
2010 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
2011 | sb->md_minor, sb->layout, sb->chunk_size); | |
2012 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
2013 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
2014 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
2015 | sb->failed_disks, sb->spare_disks, | |
2016 | sb->sb_csum, (unsigned long)sb->events_lo); | |
2017 | ||
2018 | printk(KERN_INFO); | |
2019 | for (i = 0; i < MD_SB_DISKS; i++) { | |
2020 | mdp_disk_t *desc; | |
2021 | ||
2022 | desc = sb->disks + i; | |
2023 | if (desc->number || desc->major || desc->minor || | |
2024 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
2025 | printk(" D %2d: ", i); | |
2026 | print_desc(desc); | |
2027 | } | |
2028 | } | |
2029 | printk(KERN_INFO "md: THIS: "); | |
2030 | print_desc(&sb->this_disk); | |
cd2ac932 | 2031 | } |
1da177e4 | 2032 | |
cd2ac932 CR |
2033 | static void print_sb_1(struct mdp_superblock_1 *sb) |
2034 | { | |
2035 | __u8 *uuid; | |
2036 | ||
2037 | uuid = sb->set_uuid; | |
ad361c98 | 2038 | printk(KERN_INFO |
7b75c2f8 | 2039 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" |
ad361c98 | 2040 | "md: Name: \"%s\" CT:%llu\n", |
cd2ac932 CR |
2041 | le32_to_cpu(sb->major_version), |
2042 | le32_to_cpu(sb->feature_map), | |
7b75c2f8 | 2043 | uuid, |
cd2ac932 CR |
2044 | sb->set_name, |
2045 | (unsigned long long)le64_to_cpu(sb->ctime) | |
2046 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
2047 | ||
2048 | uuid = sb->device_uuid; | |
ad361c98 JP |
2049 | printk(KERN_INFO |
2050 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 2051 | " RO:%llu\n" |
7b75c2f8 | 2052 | "md: Dev:%08x UUID: %pU\n" |
ad361c98 JP |
2053 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" |
2054 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
2055 | le32_to_cpu(sb->level), |
2056 | (unsigned long long)le64_to_cpu(sb->size), | |
2057 | le32_to_cpu(sb->raid_disks), | |
2058 | le32_to_cpu(sb->layout), | |
2059 | le32_to_cpu(sb->chunksize), | |
2060 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
2061 | (unsigned long long)le64_to_cpu(sb->data_size), | |
2062 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
2063 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
2064 | le32_to_cpu(sb->dev_number), | |
7b75c2f8 | 2065 | uuid, |
cd2ac932 CR |
2066 | sb->devflags, |
2067 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
2068 | (unsigned long long)le64_to_cpu(sb->events), | |
2069 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
2070 | le32_to_cpu(sb->sb_csum), | |
2071 | le32_to_cpu(sb->max_dev) | |
2072 | ); | |
1da177e4 LT |
2073 | } |
2074 | ||
cd2ac932 | 2075 | static void print_rdev(mdk_rdev_t *rdev, int major_version) |
1da177e4 LT |
2076 | { |
2077 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
2078 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
2079 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
2080 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
2081 | rdev->desc_nr); | |
1da177e4 | 2082 | if (rdev->sb_loaded) { |
cd2ac932 CR |
2083 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
2084 | switch (major_version) { | |
2085 | case 0: | |
2086 | print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); | |
2087 | break; | |
2088 | case 1: | |
2089 | print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); | |
2090 | break; | |
2091 | } | |
1da177e4 LT |
2092 | } else |
2093 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
2094 | } | |
2095 | ||
5e56341d | 2096 | static void md_print_devices(void) |
1da177e4 | 2097 | { |
159ec1fc | 2098 | struct list_head *tmp; |
1da177e4 LT |
2099 | mdk_rdev_t *rdev; |
2100 | mddev_t *mddev; | |
2101 | char b[BDEVNAME_SIZE]; | |
2102 | ||
2103 | printk("\n"); | |
2104 | printk("md: **********************************\n"); | |
2105 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
2106 | printk("md: **********************************\n"); | |
29ac4aa3 | 2107 | for_each_mddev(mddev, tmp) { |
1da177e4 | 2108 | |
32a7627c N |
2109 | if (mddev->bitmap) |
2110 | bitmap_print_sb(mddev->bitmap); | |
2111 | else | |
2112 | printk("%s: ", mdname(mddev)); | |
159ec1fc | 2113 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
2114 | printk("<%s>", bdevname(rdev->bdev,b)); |
2115 | printk("\n"); | |
2116 | ||
159ec1fc | 2117 | list_for_each_entry(rdev, &mddev->disks, same_set) |
cd2ac932 | 2118 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
2119 | } |
2120 | printk("md: **********************************\n"); | |
2121 | printk("\n"); | |
2122 | } | |
2123 | ||
2124 | ||
42543769 | 2125 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 2126 | { |
42543769 N |
2127 | /* Update each superblock (in-memory image), but |
2128 | * if we are allowed to, skip spares which already | |
2129 | * have the right event counter, or have one earlier | |
2130 | * (which would mean they aren't being marked as dirty | |
2131 | * with the rest of the array) | |
2132 | */ | |
1da177e4 | 2133 | mdk_rdev_t *rdev; |
1da177e4 | 2134 | |
5e865106 N |
2135 | /* First make sure individual recovery_offsets are correct */ |
2136 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
2137 | if (rdev->raid_disk >= 0 && | |
70fffd0b | 2138 | mddev->delta_disks >= 0 && |
5e865106 N |
2139 | !test_bit(In_sync, &rdev->flags) && |
2140 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2141 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2142 | ||
2143 | } | |
159ec1fc | 2144 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
42543769 N |
2145 | if (rdev->sb_events == mddev->events || |
2146 | (nospares && | |
2147 | rdev->raid_disk < 0 && | |
42543769 N |
2148 | rdev->sb_events+1 == mddev->events)) { |
2149 | /* Don't update this superblock */ | |
2150 | rdev->sb_loaded = 2; | |
2151 | } else { | |
2152 | super_types[mddev->major_version]. | |
2153 | sync_super(mddev, rdev); | |
2154 | rdev->sb_loaded = 1; | |
2155 | } | |
1da177e4 LT |
2156 | } |
2157 | } | |
2158 | ||
850b2b42 | 2159 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 2160 | { |
1da177e4 | 2161 | mdk_rdev_t *rdev; |
06d91a5f | 2162 | int sync_req; |
42543769 | 2163 | int nospares = 0; |
1da177e4 | 2164 | |
1b57f132 | 2165 | mddev->utime = get_seconds(); |
8377bc80 N |
2166 | if (mddev->external) |
2167 | return; | |
1da177e4 | 2168 | repeat: |
a9701a30 | 2169 | spin_lock_irq(&mddev->write_lock); |
84692195 | 2170 | |
850b2b42 N |
2171 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2172 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) | |
2173 | force_change = 1; | |
2174 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2175 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2176 | * though if events isn't the right even/odd, we will have to do | |
2177 | * spares after all | |
2178 | */ | |
2179 | nospares = 1; | |
2180 | if (force_change) | |
2181 | nospares = 0; | |
2182 | if (mddev->degraded) | |
84692195 N |
2183 | /* If the array is degraded, then skipping spares is both |
2184 | * dangerous and fairly pointless. | |
2185 | * Dangerous because a device that was removed from the array | |
2186 | * might have a event_count that still looks up-to-date, | |
2187 | * so it can be re-added without a resync. | |
2188 | * Pointless because if there are any spares to skip, | |
2189 | * then a recovery will happen and soon that array won't | |
2190 | * be degraded any more and the spare can go back to sleep then. | |
2191 | */ | |
850b2b42 | 2192 | nospares = 0; |
84692195 | 2193 | |
06d91a5f | 2194 | sync_req = mddev->in_sync; |
42543769 N |
2195 | |
2196 | /* If this is just a dirty<->clean transition, and the array is clean | |
2197 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2198 | if (nospares |
42543769 | 2199 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2200 | && mddev->can_decrease_events |
2201 | && mddev->events != 1) { | |
42543769 | 2202 | mddev->events--; |
a8707c08 N |
2203 | mddev->can_decrease_events = 0; |
2204 | } else { | |
42543769 N |
2205 | /* otherwise we have to go forward and ... */ |
2206 | mddev->events ++; | |
a8707c08 | 2207 | mddev->can_decrease_events = nospares; |
42543769 | 2208 | } |
1da177e4 LT |
2209 | |
2210 | if (!mddev->events) { | |
2211 | /* | |
2212 | * oops, this 64-bit counter should never wrap. | |
2213 | * Either we are in around ~1 trillion A.C., assuming | |
2214 | * 1 reboot per second, or we have a bug: | |
2215 | */ | |
2216 | MD_BUG(); | |
2217 | mddev->events --; | |
2218 | } | |
1da177e4 LT |
2219 | |
2220 | /* | |
2221 | * do not write anything to disk if using | |
2222 | * nonpersistent superblocks | |
2223 | */ | |
06d91a5f | 2224 | if (!mddev->persistent) { |
e691063a N |
2225 | if (!mddev->external) |
2226 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
2227 | ||
a9701a30 | 2228 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2229 | wake_up(&mddev->sb_wait); |
1da177e4 | 2230 | return; |
06d91a5f | 2231 | } |
e691063a | 2232 | sync_sbs(mddev, nospares); |
a9701a30 | 2233 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
2234 | |
2235 | dprintk(KERN_INFO | |
2236 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
2237 | mdname(mddev),mddev->in_sync); | |
2238 | ||
4ad13663 | 2239 | bitmap_update_sb(mddev->bitmap); |
159ec1fc | 2240 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
2241 | char b[BDEVNAME_SIZE]; |
2242 | dprintk(KERN_INFO "md: "); | |
42543769 N |
2243 | if (rdev->sb_loaded != 1) |
2244 | continue; /* no noise on spare devices */ | |
b2d444d7 | 2245 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2246 | dprintk("(skipping faulty "); |
2247 | ||
2248 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 2249 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2250 | md_super_write(mddev,rdev, |
0f420358 | 2251 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 N |
2252 | rdev->sb_page); |
2253 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
2254 | bdevname(rdev->bdev,b), | |
0f420358 | 2255 | (unsigned long long)rdev->sb_start); |
42543769 | 2256 | rdev->sb_events = mddev->events; |
7bfa19f2 | 2257 | |
1da177e4 LT |
2258 | } else |
2259 | dprintk(")\n"); | |
7bfa19f2 | 2260 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2261 | /* only need to write one superblock... */ |
2262 | break; | |
2263 | } | |
a9701a30 | 2264 | md_super_wait(mddev); |
850b2b42 | 2265 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2266 | |
a9701a30 | 2267 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2268 | if (mddev->in_sync != sync_req || |
2269 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2270 | /* have to write it out again */ |
a9701a30 | 2271 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2272 | goto repeat; |
2273 | } | |
850b2b42 | 2274 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2275 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2276 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2277 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2278 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2279 | |
1da177e4 LT |
2280 | } |
2281 | ||
7f6ce769 | 2282 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2283 | * We want to accept with case. For this we use cmd_match. |
2284 | */ | |
2285 | static int cmd_match(const char *cmd, const char *str) | |
2286 | { | |
2287 | /* See if cmd, written into a sysfs file, matches | |
2288 | * str. They must either be the same, or cmd can | |
2289 | * have a trailing newline | |
2290 | */ | |
2291 | while (*cmd && *str && *cmd == *str) { | |
2292 | cmd++; | |
2293 | str++; | |
2294 | } | |
2295 | if (*cmd == '\n') | |
2296 | cmd++; | |
2297 | if (*str || *cmd) | |
2298 | return 0; | |
2299 | return 1; | |
2300 | } | |
2301 | ||
86e6ffdd N |
2302 | struct rdev_sysfs_entry { |
2303 | struct attribute attr; | |
2304 | ssize_t (*show)(mdk_rdev_t *, char *); | |
2305 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
2306 | }; | |
2307 | ||
2308 | static ssize_t | |
96de1e66 | 2309 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
2310 | { |
2311 | char *sep = ""; | |
20a49ff6 | 2312 | size_t len = 0; |
86e6ffdd | 2313 | |
b2d444d7 | 2314 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
2315 | len+= sprintf(page+len, "%sfaulty",sep); |
2316 | sep = ","; | |
2317 | } | |
b2d444d7 | 2318 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2319 | len += sprintf(page+len, "%sin_sync",sep); |
2320 | sep = ","; | |
2321 | } | |
f655675b N |
2322 | if (test_bit(WriteMostly, &rdev->flags)) { |
2323 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2324 | sep = ","; | |
2325 | } | |
6bfe0b49 DW |
2326 | if (test_bit(Blocked, &rdev->flags)) { |
2327 | len += sprintf(page+len, "%sblocked", sep); | |
2328 | sep = ","; | |
2329 | } | |
b2d444d7 N |
2330 | if (!test_bit(Faulty, &rdev->flags) && |
2331 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2332 | len += sprintf(page+len, "%sspare", sep); |
2333 | sep = ","; | |
2334 | } | |
2335 | return len+sprintf(page+len, "\n"); | |
2336 | } | |
2337 | ||
45dc2de1 N |
2338 | static ssize_t |
2339 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2340 | { | |
2341 | /* can write | |
2342 | * faulty - simulates and error | |
2343 | * remove - disconnects the device | |
f655675b N |
2344 | * writemostly - sets write_mostly |
2345 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
2346 | * blocked - sets the Blocked flag |
2347 | * -blocked - clears the Blocked flag | |
6d56e278 | 2348 | * insync - sets Insync providing device isn't active |
45dc2de1 N |
2349 | */ |
2350 | int err = -EINVAL; | |
2351 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2352 | md_error(rdev->mddev, rdev); | |
2353 | err = 0; | |
2354 | } else if (cmd_match(buf, "remove")) { | |
2355 | if (rdev->raid_disk >= 0) | |
2356 | err = -EBUSY; | |
2357 | else { | |
2358 | mddev_t *mddev = rdev->mddev; | |
2359 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
2360 | if (mddev->pers) |
2361 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2362 | md_new_event(mddev); |
2363 | err = 0; | |
2364 | } | |
f655675b N |
2365 | } else if (cmd_match(buf, "writemostly")) { |
2366 | set_bit(WriteMostly, &rdev->flags); | |
2367 | err = 0; | |
2368 | } else if (cmd_match(buf, "-writemostly")) { | |
2369 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2370 | err = 0; |
2371 | } else if (cmd_match(buf, "blocked")) { | |
2372 | set_bit(Blocked, &rdev->flags); | |
2373 | err = 0; | |
2374 | } else if (cmd_match(buf, "-blocked")) { | |
2375 | clear_bit(Blocked, &rdev->flags); | |
2376 | wake_up(&rdev->blocked_wait); | |
2377 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2378 | md_wakeup_thread(rdev->mddev->thread); | |
2379 | ||
6d56e278 N |
2380 | err = 0; |
2381 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2382 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2383 | err = 0; |
45dc2de1 | 2384 | } |
00bcb4ac N |
2385 | if (!err) |
2386 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2387 | return err ? err : len; |
2388 | } | |
80ca3a44 N |
2389 | static struct rdev_sysfs_entry rdev_state = |
2390 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2391 | |
4dbcdc75 N |
2392 | static ssize_t |
2393 | errors_show(mdk_rdev_t *rdev, char *page) | |
2394 | { | |
2395 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2396 | } | |
2397 | ||
2398 | static ssize_t | |
2399 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2400 | { | |
2401 | char *e; | |
2402 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2403 | if (*buf && (*e == 0 || *e == '\n')) { | |
2404 | atomic_set(&rdev->corrected_errors, n); | |
2405 | return len; | |
2406 | } | |
2407 | return -EINVAL; | |
2408 | } | |
2409 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2410 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2411 | |
014236d2 N |
2412 | static ssize_t |
2413 | slot_show(mdk_rdev_t *rdev, char *page) | |
2414 | { | |
2415 | if (rdev->raid_disk < 0) | |
2416 | return sprintf(page, "none\n"); | |
2417 | else | |
2418 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2419 | } | |
2420 | ||
2421 | static ssize_t | |
2422 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2423 | { | |
2424 | char *e; | |
c303da6d N |
2425 | int err; |
2426 | char nm[20]; | |
014236d2 N |
2427 | int slot = simple_strtoul(buf, &e, 10); |
2428 | if (strncmp(buf, "none", 4)==0) | |
2429 | slot = -1; | |
2430 | else if (e==buf || (*e && *e!= '\n')) | |
2431 | return -EINVAL; | |
6c2fce2e | 2432 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2433 | /* Setting 'slot' on an active array requires also |
2434 | * updating the 'rd%d' link, and communicating | |
2435 | * with the personality with ->hot_*_disk. | |
2436 | * For now we only support removing | |
2437 | * failed/spare devices. This normally happens automatically, | |
2438 | * but not when the metadata is externally managed. | |
2439 | */ | |
c303da6d N |
2440 | if (rdev->raid_disk == -1) |
2441 | return -EEXIST; | |
2442 | /* personality does all needed checks */ | |
2443 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2444 | return -EINVAL; | |
2445 | err = rdev->mddev->pers-> | |
2446 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2447 | if (err) | |
2448 | return err; | |
2449 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2450 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
b7103107 | 2451 | rdev->raid_disk = -1; |
c303da6d N |
2452 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2453 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2454 | } else if (rdev->mddev->pers) { |
2455 | mdk_rdev_t *rdev2; | |
6c2fce2e | 2456 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2457 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2458 | */ |
2459 | ||
2460 | if (rdev->raid_disk != -1) | |
2461 | return -EBUSY; | |
2462 | ||
2463 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2464 | return -EINVAL; | |
2465 | ||
159ec1fc | 2466 | list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) |
6c2fce2e NB |
2467 | if (rdev2->raid_disk == slot) |
2468 | return -EEXIST; | |
2469 | ||
2470 | rdev->raid_disk = slot; | |
2471 | if (test_bit(In_sync, &rdev->flags)) | |
2472 | rdev->saved_raid_disk = slot; | |
2473 | else | |
2474 | rdev->saved_raid_disk = -1; | |
2475 | err = rdev->mddev->pers-> | |
2476 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2477 | if (err) { |
6c2fce2e | 2478 | rdev->raid_disk = -1; |
6c2fce2e | 2479 | return err; |
52664732 | 2480 | } else |
00bcb4ac | 2481 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6c2fce2e NB |
2482 | sprintf(nm, "rd%d", rdev->raid_disk); |
2483 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
00bcb4ac | 2484 | /* failure here is OK */; |
6c2fce2e | 2485 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d N |
2486 | } else { |
2487 | if (slot >= rdev->mddev->raid_disks) | |
2488 | return -ENOSPC; | |
2489 | rdev->raid_disk = slot; | |
2490 | /* assume it is working */ | |
c5d79adb N |
2491 | clear_bit(Faulty, &rdev->flags); |
2492 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2493 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2494 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2495 | } |
014236d2 N |
2496 | return len; |
2497 | } | |
2498 | ||
2499 | ||
2500 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2501 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2502 | |
93c8cad0 N |
2503 | static ssize_t |
2504 | offset_show(mdk_rdev_t *rdev, char *page) | |
2505 | { | |
6961ece4 | 2506 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2507 | } |
2508 | ||
2509 | static ssize_t | |
2510 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2511 | { | |
2512 | char *e; | |
2513 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2514 | if (e==buf || (*e && *e != '\n')) | |
2515 | return -EINVAL; | |
8ed0a521 | 2516 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2517 | return -EBUSY; |
dd8ac336 | 2518 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2519 | /* Must set offset before size, so overlap checks |
2520 | * can be sane */ | |
2521 | return -EBUSY; | |
93c8cad0 N |
2522 | rdev->data_offset = offset; |
2523 | return len; | |
2524 | } | |
2525 | ||
2526 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2527 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2528 | |
83303b61 N |
2529 | static ssize_t |
2530 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2531 | { | |
dd8ac336 | 2532 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2533 | } |
2534 | ||
c5d79adb N |
2535 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2536 | { | |
2537 | /* check if two start/length pairs overlap */ | |
2538 | if (s1+l1 <= s2) | |
2539 | return 0; | |
2540 | if (s2+l2 <= s1) | |
2541 | return 0; | |
2542 | return 1; | |
2543 | } | |
2544 | ||
b522adcd DW |
2545 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2546 | { | |
2547 | unsigned long long blocks; | |
2548 | sector_t new; | |
2549 | ||
2550 | if (strict_strtoull(buf, 10, &blocks) < 0) | |
2551 | return -EINVAL; | |
2552 | ||
2553 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2554 | return -EINVAL; /* sector conversion overflow */ | |
2555 | ||
2556 | new = blocks * 2; | |
2557 | if (new != blocks * 2) | |
2558 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2559 | ||
2560 | *sectors = new; | |
2561 | return 0; | |
2562 | } | |
2563 | ||
83303b61 N |
2564 | static ssize_t |
2565 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2566 | { | |
27c529bb | 2567 | mddev_t *my_mddev = rdev->mddev; |
dd8ac336 | 2568 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2569 | sector_t sectors; |
27c529bb | 2570 | |
b522adcd | 2571 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2572 | return -EINVAL; |
0cd17fec | 2573 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2574 | if (my_mddev->persistent) { |
dd8ac336 AN |
2575 | sectors = super_types[my_mddev->major_version]. |
2576 | rdev_size_change(rdev, sectors); | |
2577 | if (!sectors) | |
0cd17fec | 2578 | return -EBUSY; |
dd8ac336 AN |
2579 | } else if (!sectors) |
2580 | sectors = (rdev->bdev->bd_inode->i_size >> 9) - | |
2581 | rdev->data_offset; | |
0cd17fec | 2582 | } |
dd8ac336 | 2583 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2584 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2585 | |
dd8ac336 AN |
2586 | rdev->sectors = sectors; |
2587 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2588 | /* need to check that all other rdevs with the same ->bdev |
2589 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2590 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2591 | * we have to change it back, we will have the lock again. |
2592 | */ | |
2593 | mddev_t *mddev; | |
2594 | int overlap = 0; | |
159ec1fc | 2595 | struct list_head *tmp; |
c5d79adb | 2596 | |
27c529bb | 2597 | mddev_unlock(my_mddev); |
29ac4aa3 | 2598 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2599 | mdk_rdev_t *rdev2; |
2600 | ||
2601 | mddev_lock(mddev); | |
159ec1fc | 2602 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
c5d79adb N |
2603 | if (test_bit(AllReserved, &rdev2->flags) || |
2604 | (rdev->bdev == rdev2->bdev && | |
2605 | rdev != rdev2 && | |
dd8ac336 | 2606 | overlaps(rdev->data_offset, rdev->sectors, |
d07bd3bc | 2607 | rdev2->data_offset, |
dd8ac336 | 2608 | rdev2->sectors))) { |
c5d79adb N |
2609 | overlap = 1; |
2610 | break; | |
2611 | } | |
2612 | mddev_unlock(mddev); | |
2613 | if (overlap) { | |
2614 | mddev_put(mddev); | |
2615 | break; | |
2616 | } | |
2617 | } | |
27c529bb | 2618 | mddev_lock(my_mddev); |
c5d79adb N |
2619 | if (overlap) { |
2620 | /* Someone else could have slipped in a size | |
2621 | * change here, but doing so is just silly. | |
dd8ac336 | 2622 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2623 | * safe, and trust userspace not to race with |
2624 | * itself | |
2625 | */ | |
dd8ac336 | 2626 | rdev->sectors = oldsectors; |
c5d79adb N |
2627 | return -EBUSY; |
2628 | } | |
2629 | } | |
83303b61 N |
2630 | return len; |
2631 | } | |
2632 | ||
2633 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2634 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2635 | |
06e3c817 DW |
2636 | |
2637 | static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page) | |
2638 | { | |
2639 | unsigned long long recovery_start = rdev->recovery_offset; | |
2640 | ||
2641 | if (test_bit(In_sync, &rdev->flags) || | |
2642 | recovery_start == MaxSector) | |
2643 | return sprintf(page, "none\n"); | |
2644 | ||
2645 | return sprintf(page, "%llu\n", recovery_start); | |
2646 | } | |
2647 | ||
2648 | static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2649 | { | |
2650 | unsigned long long recovery_start; | |
2651 | ||
2652 | if (cmd_match(buf, "none")) | |
2653 | recovery_start = MaxSector; | |
2654 | else if (strict_strtoull(buf, 10, &recovery_start)) | |
2655 | return -EINVAL; | |
2656 | ||
2657 | if (rdev->mddev->pers && | |
2658 | rdev->raid_disk >= 0) | |
2659 | return -EBUSY; | |
2660 | ||
2661 | rdev->recovery_offset = recovery_start; | |
2662 | if (recovery_start == MaxSector) | |
2663 | set_bit(In_sync, &rdev->flags); | |
2664 | else | |
2665 | clear_bit(In_sync, &rdev->flags); | |
2666 | return len; | |
2667 | } | |
2668 | ||
2669 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2670 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2671 | ||
86e6ffdd N |
2672 | static struct attribute *rdev_default_attrs[] = { |
2673 | &rdev_state.attr, | |
4dbcdc75 | 2674 | &rdev_errors.attr, |
014236d2 | 2675 | &rdev_slot.attr, |
93c8cad0 | 2676 | &rdev_offset.attr, |
83303b61 | 2677 | &rdev_size.attr, |
06e3c817 | 2678 | &rdev_recovery_start.attr, |
86e6ffdd N |
2679 | NULL, |
2680 | }; | |
2681 | static ssize_t | |
2682 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2683 | { | |
2684 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2685 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2686 | mddev_t *mddev = rdev->mddev; |
2687 | ssize_t rv; | |
86e6ffdd N |
2688 | |
2689 | if (!entry->show) | |
2690 | return -EIO; | |
27c529bb N |
2691 | |
2692 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2693 | if (!rv) { | |
2694 | if (rdev->mddev == NULL) | |
2695 | rv = -EBUSY; | |
2696 | else | |
2697 | rv = entry->show(rdev, page); | |
2698 | mddev_unlock(mddev); | |
2699 | } | |
2700 | return rv; | |
86e6ffdd N |
2701 | } |
2702 | ||
2703 | static ssize_t | |
2704 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2705 | const char *page, size_t length) | |
2706 | { | |
2707 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2708 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2709 | ssize_t rv; |
2710 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2711 | |
2712 | if (!entry->store) | |
2713 | return -EIO; | |
67463acb N |
2714 | if (!capable(CAP_SYS_ADMIN)) |
2715 | return -EACCES; | |
27c529bb | 2716 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2717 | if (!rv) { |
27c529bb N |
2718 | if (rdev->mddev == NULL) |
2719 | rv = -EBUSY; | |
2720 | else | |
2721 | rv = entry->store(rdev, page, length); | |
6a51830e | 2722 | mddev_unlock(mddev); |
ca388059 N |
2723 | } |
2724 | return rv; | |
86e6ffdd N |
2725 | } |
2726 | ||
2727 | static void rdev_free(struct kobject *ko) | |
2728 | { | |
2729 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2730 | kfree(rdev); | |
2731 | } | |
52cf25d0 | 2732 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
2733 | .show = rdev_attr_show, |
2734 | .store = rdev_attr_store, | |
2735 | }; | |
2736 | static struct kobj_type rdev_ktype = { | |
2737 | .release = rdev_free, | |
2738 | .sysfs_ops = &rdev_sysfs_ops, | |
2739 | .default_attrs = rdev_default_attrs, | |
2740 | }; | |
2741 | ||
e8bb9a83 N |
2742 | void md_rdev_init(mdk_rdev_t *rdev) |
2743 | { | |
2744 | rdev->desc_nr = -1; | |
2745 | rdev->saved_raid_disk = -1; | |
2746 | rdev->raid_disk = -1; | |
2747 | rdev->flags = 0; | |
2748 | rdev->data_offset = 0; | |
2749 | rdev->sb_events = 0; | |
2750 | rdev->last_read_error.tv_sec = 0; | |
2751 | rdev->last_read_error.tv_nsec = 0; | |
2752 | atomic_set(&rdev->nr_pending, 0); | |
2753 | atomic_set(&rdev->read_errors, 0); | |
2754 | atomic_set(&rdev->corrected_errors, 0); | |
2755 | ||
2756 | INIT_LIST_HEAD(&rdev->same_set); | |
2757 | init_waitqueue_head(&rdev->blocked_wait); | |
2758 | } | |
2759 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
2760 | /* |
2761 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2762 | * | |
2763 | * mark the device faulty if: | |
2764 | * | |
2765 | * - the device is nonexistent (zero size) | |
2766 | * - the device has no valid superblock | |
2767 | * | |
2768 | * a faulty rdev _never_ has rdev->sb set. | |
2769 | */ | |
2770 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2771 | { | |
2772 | char b[BDEVNAME_SIZE]; | |
2773 | int err; | |
2774 | mdk_rdev_t *rdev; | |
2775 | sector_t size; | |
2776 | ||
9ffae0cf | 2777 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2778 | if (!rdev) { |
2779 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2780 | return ERR_PTR(-ENOMEM); | |
2781 | } | |
1da177e4 | 2782 | |
e8bb9a83 | 2783 | md_rdev_init(rdev); |
1da177e4 LT |
2784 | if ((err = alloc_disk_sb(rdev))) |
2785 | goto abort_free; | |
2786 | ||
c5d79adb | 2787 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2788 | if (err) |
2789 | goto abort_free; | |
2790 | ||
f9cb074b | 2791 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2792 | |
1da177e4 LT |
2793 | size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; |
2794 | if (!size) { | |
2795 | printk(KERN_WARNING | |
2796 | "md: %s has zero or unknown size, marking faulty!\n", | |
2797 | bdevname(rdev->bdev,b)); | |
2798 | err = -EINVAL; | |
2799 | goto abort_free; | |
2800 | } | |
2801 | ||
2802 | if (super_format >= 0) { | |
2803 | err = super_types[super_format]. | |
2804 | load_super(rdev, NULL, super_minor); | |
2805 | if (err == -EINVAL) { | |
df968c4e N |
2806 | printk(KERN_WARNING |
2807 | "md: %s does not have a valid v%d.%d " | |
2808 | "superblock, not importing!\n", | |
2809 | bdevname(rdev->bdev,b), | |
2810 | super_format, super_minor); | |
1da177e4 LT |
2811 | goto abort_free; |
2812 | } | |
2813 | if (err < 0) { | |
2814 | printk(KERN_WARNING | |
2815 | "md: could not read %s's sb, not importing!\n", | |
2816 | bdevname(rdev->bdev,b)); | |
2817 | goto abort_free; | |
2818 | } | |
2819 | } | |
6bfe0b49 | 2820 | |
1da177e4 LT |
2821 | return rdev; |
2822 | ||
2823 | abort_free: | |
2824 | if (rdev->sb_page) { | |
2825 | if (rdev->bdev) | |
2826 | unlock_rdev(rdev); | |
2827 | free_disk_sb(rdev); | |
2828 | } | |
2829 | kfree(rdev); | |
2830 | return ERR_PTR(err); | |
2831 | } | |
2832 | ||
2833 | /* | |
2834 | * Check a full RAID array for plausibility | |
2835 | */ | |
2836 | ||
2837 | ||
a757e64c | 2838 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2839 | { |
2840 | int i; | |
159ec1fc | 2841 | mdk_rdev_t *rdev, *freshest, *tmp; |
1da177e4 LT |
2842 | char b[BDEVNAME_SIZE]; |
2843 | ||
2844 | freshest = NULL; | |
d089c6af | 2845 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2846 | switch (super_types[mddev->major_version]. |
2847 | load_super(rdev, freshest, mddev->minor_version)) { | |
2848 | case 1: | |
2849 | freshest = rdev; | |
2850 | break; | |
2851 | case 0: | |
2852 | break; | |
2853 | default: | |
2854 | printk( KERN_ERR \ | |
2855 | "md: fatal superblock inconsistency in %s" | |
2856 | " -- removing from array\n", | |
2857 | bdevname(rdev->bdev,b)); | |
2858 | kick_rdev_from_array(rdev); | |
2859 | } | |
2860 | ||
2861 | ||
2862 | super_types[mddev->major_version]. | |
2863 | validate_super(mddev, freshest); | |
2864 | ||
2865 | i = 0; | |
d089c6af | 2866 | rdev_for_each(rdev, tmp, mddev) { |
233fca36 N |
2867 | if (mddev->max_disks && |
2868 | (rdev->desc_nr >= mddev->max_disks || | |
2869 | i > mddev->max_disks)) { | |
de01dfad N |
2870 | printk(KERN_WARNING |
2871 | "md: %s: %s: only %d devices permitted\n", | |
2872 | mdname(mddev), bdevname(rdev->bdev, b), | |
2873 | mddev->max_disks); | |
2874 | kick_rdev_from_array(rdev); | |
2875 | continue; | |
2876 | } | |
1da177e4 LT |
2877 | if (rdev != freshest) |
2878 | if (super_types[mddev->major_version]. | |
2879 | validate_super(mddev, rdev)) { | |
2880 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2881 | " from array!\n", | |
2882 | bdevname(rdev->bdev,b)); | |
2883 | kick_rdev_from_array(rdev); | |
2884 | continue; | |
2885 | } | |
2886 | if (mddev->level == LEVEL_MULTIPATH) { | |
2887 | rdev->desc_nr = i++; | |
2888 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2889 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 2890 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
2891 | rdev->raid_disk = -1; |
2892 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2893 | } |
2894 | } | |
1da177e4 LT |
2895 | } |
2896 | ||
72e02075 N |
2897 | /* Read a fixed-point number. |
2898 | * Numbers in sysfs attributes should be in "standard" units where | |
2899 | * possible, so time should be in seconds. | |
2900 | * However we internally use a a much smaller unit such as | |
2901 | * milliseconds or jiffies. | |
2902 | * This function takes a decimal number with a possible fractional | |
2903 | * component, and produces an integer which is the result of | |
2904 | * multiplying that number by 10^'scale'. | |
2905 | * all without any floating-point arithmetic. | |
2906 | */ | |
2907 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
2908 | { | |
2909 | unsigned long result = 0; | |
2910 | long decimals = -1; | |
2911 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
2912 | if (*cp == '.') | |
2913 | decimals = 0; | |
2914 | else if (decimals < scale) { | |
2915 | unsigned int value; | |
2916 | value = *cp - '0'; | |
2917 | result = result * 10 + value; | |
2918 | if (decimals >= 0) | |
2919 | decimals++; | |
2920 | } | |
2921 | cp++; | |
2922 | } | |
2923 | if (*cp == '\n') | |
2924 | cp++; | |
2925 | if (*cp) | |
2926 | return -EINVAL; | |
2927 | if (decimals < 0) | |
2928 | decimals = 0; | |
2929 | while (decimals < scale) { | |
2930 | result *= 10; | |
2931 | decimals ++; | |
2932 | } | |
2933 | *res = result; | |
2934 | return 0; | |
2935 | } | |
2936 | ||
2937 | ||
19052c0e N |
2938 | static void md_safemode_timeout(unsigned long data); |
2939 | ||
16f17b39 N |
2940 | static ssize_t |
2941 | safe_delay_show(mddev_t *mddev, char *page) | |
2942 | { | |
2943 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2944 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2945 | } | |
2946 | static ssize_t | |
2947 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2948 | { | |
16f17b39 | 2949 | unsigned long msec; |
97ce0a7f | 2950 | |
72e02075 | 2951 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 2952 | return -EINVAL; |
16f17b39 N |
2953 | if (msec == 0) |
2954 | mddev->safemode_delay = 0; | |
2955 | else { | |
19052c0e | 2956 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
2957 | mddev->safemode_delay = (msec*HZ)/1000; |
2958 | if (mddev->safemode_delay == 0) | |
2959 | mddev->safemode_delay = 1; | |
19052c0e N |
2960 | if (mddev->safemode_delay < old_delay) |
2961 | md_safemode_timeout((unsigned long)mddev); | |
16f17b39 N |
2962 | } |
2963 | return len; | |
2964 | } | |
2965 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 2966 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 2967 | |
eae1701f | 2968 | static ssize_t |
96de1e66 | 2969 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2970 | { |
2604b703 | 2971 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2972 | if (p) |
eae1701f | 2973 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2974 | else if (mddev->clevel[0]) |
2975 | return sprintf(page, "%s\n", mddev->clevel); | |
2976 | else if (mddev->level != LEVEL_NONE) | |
2977 | return sprintf(page, "%d\n", mddev->level); | |
2978 | else | |
2979 | return 0; | |
eae1701f N |
2980 | } |
2981 | ||
d9d166c2 N |
2982 | static ssize_t |
2983 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2984 | { | |
f2859af6 | 2985 | char clevel[16]; |
20a49ff6 | 2986 | ssize_t rv = len; |
245f46c2 | 2987 | struct mdk_personality *pers; |
f2859af6 | 2988 | long level; |
245f46c2 | 2989 | void *priv; |
3a981b03 | 2990 | mdk_rdev_t *rdev; |
245f46c2 N |
2991 | |
2992 | if (mddev->pers == NULL) { | |
2993 | if (len == 0) | |
2994 | return 0; | |
2995 | if (len >= sizeof(mddev->clevel)) | |
2996 | return -ENOSPC; | |
2997 | strncpy(mddev->clevel, buf, len); | |
2998 | if (mddev->clevel[len-1] == '\n') | |
2999 | len--; | |
3000 | mddev->clevel[len] = 0; | |
3001 | mddev->level = LEVEL_NONE; | |
3002 | return rv; | |
3003 | } | |
3004 | ||
3005 | /* request to change the personality. Need to ensure: | |
3006 | * - array is not engaged in resync/recovery/reshape | |
3007 | * - old personality can be suspended | |
3008 | * - new personality will access other array. | |
3009 | */ | |
3010 | ||
3011 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) | |
d9d166c2 | 3012 | return -EBUSY; |
245f46c2 N |
3013 | |
3014 | if (!mddev->pers->quiesce) { | |
3015 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3016 | mdname(mddev), mddev->pers->name); | |
3017 | return -EINVAL; | |
3018 | } | |
3019 | ||
3020 | /* Now find the new personality */ | |
f2859af6 | 3021 | if (len == 0 || len >= sizeof(clevel)) |
245f46c2 | 3022 | return -EINVAL; |
f2859af6 DW |
3023 | strncpy(clevel, buf, len); |
3024 | if (clevel[len-1] == '\n') | |
d9d166c2 | 3025 | len--; |
f2859af6 DW |
3026 | clevel[len] = 0; |
3027 | if (strict_strtol(clevel, 10, &level)) | |
3028 | level = LEVEL_NONE; | |
245f46c2 | 3029 | |
f2859af6 DW |
3030 | if (request_module("md-%s", clevel) != 0) |
3031 | request_module("md-level-%s", clevel); | |
245f46c2 | 3032 | spin_lock(&pers_lock); |
f2859af6 | 3033 | pers = find_pers(level, clevel); |
245f46c2 N |
3034 | if (!pers || !try_module_get(pers->owner)) { |
3035 | spin_unlock(&pers_lock); | |
f2859af6 | 3036 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
245f46c2 N |
3037 | return -EINVAL; |
3038 | } | |
3039 | spin_unlock(&pers_lock); | |
3040 | ||
3041 | if (pers == mddev->pers) { | |
3042 | /* Nothing to do! */ | |
3043 | module_put(pers->owner); | |
3044 | return rv; | |
3045 | } | |
3046 | if (!pers->takeover) { | |
3047 | module_put(pers->owner); | |
3048 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3049 | mdname(mddev), clevel); |
245f46c2 N |
3050 | return -EINVAL; |
3051 | } | |
3052 | ||
e93f68a1 N |
3053 | list_for_each_entry(rdev, &mddev->disks, same_set) |
3054 | rdev->new_raid_disk = rdev->raid_disk; | |
3055 | ||
245f46c2 N |
3056 | /* ->takeover must set new_* and/or delta_disks |
3057 | * if it succeeds, and may set them when it fails. | |
3058 | */ | |
3059 | priv = pers->takeover(mddev); | |
3060 | if (IS_ERR(priv)) { | |
3061 | mddev->new_level = mddev->level; | |
3062 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3063 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3064 | mddev->raid_disks -= mddev->delta_disks; |
3065 | mddev->delta_disks = 0; | |
3066 | module_put(pers->owner); | |
3067 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3068 | mdname(mddev), clevel); |
245f46c2 N |
3069 | return PTR_ERR(priv); |
3070 | } | |
3071 | ||
3072 | /* Looks like we have a winner */ | |
3073 | mddev_suspend(mddev); | |
3074 | mddev->pers->stop(mddev); | |
a64c876f N |
3075 | |
3076 | if (mddev->pers->sync_request == NULL && | |
3077 | pers->sync_request != NULL) { | |
3078 | /* need to add the md_redundancy_group */ | |
3079 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3080 | printk(KERN_WARNING | |
3081 | "md: cannot register extra attributes for %s\n", | |
3082 | mdname(mddev)); | |
19fdb9ee | 3083 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action"); |
a64c876f N |
3084 | } |
3085 | if (mddev->pers->sync_request != NULL && | |
3086 | pers->sync_request == NULL) { | |
3087 | /* need to remove the md_redundancy_group */ | |
3088 | if (mddev->to_remove == NULL) | |
3089 | mddev->to_remove = &md_redundancy_group; | |
3090 | } | |
3091 | ||
54071b38 TM |
3092 | if (mddev->pers->sync_request == NULL && |
3093 | mddev->external) { | |
3094 | /* We are converting from a no-redundancy array | |
3095 | * to a redundancy array and metadata is managed | |
3096 | * externally so we need to be sure that writes | |
3097 | * won't block due to a need to transition | |
3098 | * clean->dirty | |
3099 | * until external management is started. | |
3100 | */ | |
3101 | mddev->in_sync = 0; | |
3102 | mddev->safemode_delay = 0; | |
3103 | mddev->safemode = 0; | |
3104 | } | |
3105 | ||
e93f68a1 N |
3106 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
3107 | char nm[20]; | |
3108 | if (rdev->raid_disk < 0) | |
3109 | continue; | |
3110 | if (rdev->new_raid_disk > mddev->raid_disks) | |
3111 | rdev->new_raid_disk = -1; | |
3112 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3113 | continue; | |
3114 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3115 | sysfs_remove_link(&mddev->kobj, nm); | |
3116 | } | |
3117 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
3118 | if (rdev->raid_disk < 0) | |
3119 | continue; | |
3120 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3121 | continue; | |
3122 | rdev->raid_disk = rdev->new_raid_disk; | |
3123 | if (rdev->raid_disk < 0) | |
3a981b03 | 3124 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 N |
3125 | else { |
3126 | char nm[20]; | |
3127 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3128 | if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) | |
3129 | printk("md: cannot register %s for %s after level change\n", | |
3130 | nm, mdname(mddev)); | |
3a981b03 | 3131 | } |
e93f68a1 N |
3132 | } |
3133 | ||
3134 | module_put(mddev->pers->owner); | |
245f46c2 N |
3135 | mddev->pers = pers; |
3136 | mddev->private = priv; | |
3137 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3138 | mddev->level = mddev->new_level; | |
3139 | mddev->layout = mddev->new_layout; | |
664e7c41 | 3140 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 | 3141 | mddev->delta_disks = 0; |
9af204cf TM |
3142 | if (mddev->pers->sync_request == NULL) { |
3143 | /* this is now an array without redundancy, so | |
3144 | * it must always be in_sync | |
3145 | */ | |
3146 | mddev->in_sync = 1; | |
3147 | del_timer_sync(&mddev->safemode_timer); | |
3148 | } | |
245f46c2 N |
3149 | pers->run(mddev); |
3150 | mddev_resume(mddev); | |
3151 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3152 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3153 | md_wakeup_thread(mddev->thread); | |
5cac7861 | 3154 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3155 | md_new_event(mddev); |
d9d166c2 N |
3156 | return rv; |
3157 | } | |
3158 | ||
3159 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3160 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3161 | |
d4dbd025 N |
3162 | |
3163 | static ssize_t | |
3164 | layout_show(mddev_t *mddev, char *page) | |
3165 | { | |
3166 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3167 | if (mddev->reshape_position != MaxSector && |
3168 | mddev->layout != mddev->new_layout) | |
3169 | return sprintf(page, "%d (%d)\n", | |
3170 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3171 | return sprintf(page, "%d\n", mddev->layout); |
3172 | } | |
3173 | ||
3174 | static ssize_t | |
3175 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
3176 | { | |
3177 | char *e; | |
3178 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
3179 | |
3180 | if (!*buf || (*e && *e != '\n')) | |
3181 | return -EINVAL; | |
3182 | ||
b3546035 N |
3183 | if (mddev->pers) { |
3184 | int err; | |
50ac168a | 3185 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3186 | return -EBUSY; |
597a711b | 3187 | mddev->new_layout = n; |
50ac168a | 3188 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3189 | if (err) { |
3190 | mddev->new_layout = mddev->layout; | |
b3546035 | 3191 | return err; |
597a711b | 3192 | } |
b3546035 | 3193 | } else { |
08a02ecd | 3194 | mddev->new_layout = n; |
b3546035 N |
3195 | if (mddev->reshape_position == MaxSector) |
3196 | mddev->layout = n; | |
3197 | } | |
d4dbd025 N |
3198 | return len; |
3199 | } | |
3200 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3201 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
3202 | |
3203 | ||
eae1701f | 3204 | static ssize_t |
96de1e66 | 3205 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 3206 | { |
bb636547 N |
3207 | if (mddev->raid_disks == 0) |
3208 | return 0; | |
08a02ecd N |
3209 | if (mddev->reshape_position != MaxSector && |
3210 | mddev->delta_disks != 0) | |
3211 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3212 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3213 | return sprintf(page, "%d\n", mddev->raid_disks); |
3214 | } | |
3215 | ||
da943b99 N |
3216 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
3217 | ||
3218 | static ssize_t | |
3219 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
3220 | { | |
da943b99 N |
3221 | char *e; |
3222 | int rv = 0; | |
3223 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3224 | ||
3225 | if (!*buf || (*e && *e != '\n')) | |
3226 | return -EINVAL; | |
3227 | ||
3228 | if (mddev->pers) | |
3229 | rv = update_raid_disks(mddev, n); | |
08a02ecd N |
3230 | else if (mddev->reshape_position != MaxSector) { |
3231 | int olddisks = mddev->raid_disks - mddev->delta_disks; | |
3232 | mddev->delta_disks = n - olddisks; | |
3233 | mddev->raid_disks = n; | |
3234 | } else | |
da943b99 N |
3235 | mddev->raid_disks = n; |
3236 | return rv ? rv : len; | |
3237 | } | |
3238 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3239 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3240 | |
3b34380a N |
3241 | static ssize_t |
3242 | chunk_size_show(mddev_t *mddev, char *page) | |
3243 | { | |
08a02ecd | 3244 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3245 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3246 | return sprintf(page, "%d (%d)\n", | |
3247 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3248 | mddev->chunk_sectors << 9); |
3249 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3250 | } |
3251 | ||
3252 | static ssize_t | |
3253 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
3254 | { | |
3b34380a N |
3255 | char *e; |
3256 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3257 | ||
3b34380a N |
3258 | if (!*buf || (*e && *e != '\n')) |
3259 | return -EINVAL; | |
3260 | ||
b3546035 N |
3261 | if (mddev->pers) { |
3262 | int err; | |
50ac168a | 3263 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3264 | return -EBUSY; |
597a711b | 3265 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 3266 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3267 | if (err) { |
3268 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 3269 | return err; |
597a711b | 3270 | } |
b3546035 | 3271 | } else { |
664e7c41 | 3272 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3273 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3274 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3275 | } |
3b34380a N |
3276 | return len; |
3277 | } | |
3278 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3279 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3280 | |
a94213b1 N |
3281 | static ssize_t |
3282 | resync_start_show(mddev_t *mddev, char *page) | |
3283 | { | |
d1a7c503 N |
3284 | if (mddev->recovery_cp == MaxSector) |
3285 | return sprintf(page, "none\n"); | |
a94213b1 N |
3286 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3287 | } | |
3288 | ||
3289 | static ssize_t | |
3290 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
3291 | { | |
a94213b1 N |
3292 | char *e; |
3293 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
3294 | ||
3295 | if (mddev->pers) | |
3296 | return -EBUSY; | |
06e3c817 DW |
3297 | if (cmd_match(buf, "none")) |
3298 | n = MaxSector; | |
3299 | else if (!*buf || (*e && *e != '\n')) | |
a94213b1 N |
3300 | return -EINVAL; |
3301 | ||
3302 | mddev->recovery_cp = n; | |
3303 | return len; | |
3304 | } | |
3305 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 3306 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 3307 | |
9e653b63 N |
3308 | /* |
3309 | * The array state can be: | |
3310 | * | |
3311 | * clear | |
3312 | * No devices, no size, no level | |
3313 | * Equivalent to STOP_ARRAY ioctl | |
3314 | * inactive | |
3315 | * May have some settings, but array is not active | |
3316 | * all IO results in error | |
3317 | * When written, doesn't tear down array, but just stops it | |
3318 | * suspended (not supported yet) | |
3319 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3320 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3321 | * readonly |
3322 | * no resync can happen. no superblocks get written. | |
3323 | * write requests fail | |
3324 | * read-auto | |
3325 | * like readonly, but behaves like 'clean' on a write request. | |
3326 | * | |
3327 | * clean - no pending writes, but otherwise active. | |
3328 | * When written to inactive array, starts without resync | |
3329 | * If a write request arrives then | |
3330 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3331 | * if not known, block and switch to write-pending | |
3332 | * If written to an active array that has pending writes, then fails. | |
3333 | * active | |
3334 | * fully active: IO and resync can be happening. | |
3335 | * When written to inactive array, starts with resync | |
3336 | * | |
3337 | * write-pending | |
3338 | * clean, but writes are blocked waiting for 'active' to be written. | |
3339 | * | |
3340 | * active-idle | |
3341 | * like active, but no writes have been seen for a while (100msec). | |
3342 | * | |
3343 | */ | |
3344 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3345 | write_pending, active_idle, bad_word}; | |
05381954 | 3346 | static char *array_states[] = { |
9e653b63 N |
3347 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3348 | "write-pending", "active-idle", NULL }; | |
3349 | ||
3350 | static int match_word(const char *word, char **list) | |
3351 | { | |
3352 | int n; | |
3353 | for (n=0; list[n]; n++) | |
3354 | if (cmd_match(word, list[n])) | |
3355 | break; | |
3356 | return n; | |
3357 | } | |
3358 | ||
3359 | static ssize_t | |
3360 | array_state_show(mddev_t *mddev, char *page) | |
3361 | { | |
3362 | enum array_state st = inactive; | |
3363 | ||
3364 | if (mddev->pers) | |
3365 | switch(mddev->ro) { | |
3366 | case 1: | |
3367 | st = readonly; | |
3368 | break; | |
3369 | case 2: | |
3370 | st = read_auto; | |
3371 | break; | |
3372 | case 0: | |
3373 | if (mddev->in_sync) | |
3374 | st = clean; | |
e691063a N |
3375 | else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) |
3376 | st = write_pending; | |
9e653b63 N |
3377 | else if (mddev->safemode) |
3378 | st = active_idle; | |
3379 | else | |
3380 | st = active; | |
3381 | } | |
3382 | else { | |
3383 | if (list_empty(&mddev->disks) && | |
3384 | mddev->raid_disks == 0 && | |
58c0fed4 | 3385 | mddev->dev_sectors == 0) |
9e653b63 N |
3386 | st = clear; |
3387 | else | |
3388 | st = inactive; | |
3389 | } | |
3390 | return sprintf(page, "%s\n", array_states[st]); | |
3391 | } | |
3392 | ||
df5b20cf | 3393 | static int do_md_stop(mddev_t * mddev, int ro, int is_open); |
a4bd82d0 | 3394 | static int md_set_readonly(mddev_t * mddev, int is_open); |
9e653b63 N |
3395 | static int do_md_run(mddev_t * mddev); |
3396 | static int restart_array(mddev_t *mddev); | |
3397 | ||
3398 | static ssize_t | |
3399 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
3400 | { | |
3401 | int err = -EINVAL; | |
3402 | enum array_state st = match_word(buf, array_states); | |
3403 | switch(st) { | |
3404 | case bad_word: | |
3405 | break; | |
3406 | case clear: | |
3407 | /* stopping an active array */ | |
f2ea68cf | 3408 | if (atomic_read(&mddev->openers) > 0) |
e691063a | 3409 | return -EBUSY; |
df5b20cf | 3410 | err = do_md_stop(mddev, 0, 0); |
9e653b63 N |
3411 | break; |
3412 | case inactive: | |
3413 | /* stopping an active array */ | |
3414 | if (mddev->pers) { | |
f2ea68cf | 3415 | if (atomic_read(&mddev->openers) > 0) |
9e653b63 | 3416 | return -EBUSY; |
df5b20cf | 3417 | err = do_md_stop(mddev, 2, 0); |
e691063a N |
3418 | } else |
3419 | err = 0; /* already inactive */ | |
9e653b63 N |
3420 | break; |
3421 | case suspended: | |
3422 | break; /* not supported yet */ | |
3423 | case readonly: | |
3424 | if (mddev->pers) | |
a4bd82d0 | 3425 | err = md_set_readonly(mddev, 0); |
9e653b63 N |
3426 | else { |
3427 | mddev->ro = 1; | |
648b629e | 3428 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3429 | err = do_md_run(mddev); |
3430 | } | |
3431 | break; | |
3432 | case read_auto: | |
9e653b63 | 3433 | if (mddev->pers) { |
80268ee9 | 3434 | if (mddev->ro == 0) |
a4bd82d0 | 3435 | err = md_set_readonly(mddev, 0); |
80268ee9 | 3436 | else if (mddev->ro == 1) |
648b629e N |
3437 | err = restart_array(mddev); |
3438 | if (err == 0) { | |
3439 | mddev->ro = 2; | |
3440 | set_disk_ro(mddev->gendisk, 0); | |
3441 | } | |
9e653b63 N |
3442 | } else { |
3443 | mddev->ro = 2; | |
3444 | err = do_md_run(mddev); | |
3445 | } | |
3446 | break; | |
3447 | case clean: | |
3448 | if (mddev->pers) { | |
3449 | restart_array(mddev); | |
3450 | spin_lock_irq(&mddev->write_lock); | |
3451 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3452 | if (mddev->in_sync == 0) { |
3453 | mddev->in_sync = 1; | |
31a59e34 N |
3454 | if (mddev->safemode == 1) |
3455 | mddev->safemode = 0; | |
e691063a N |
3456 | if (mddev->persistent) |
3457 | set_bit(MD_CHANGE_CLEAN, | |
3458 | &mddev->flags); | |
3459 | } | |
3460 | err = 0; | |
3461 | } else | |
3462 | err = -EBUSY; | |
9e653b63 | 3463 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3464 | } else |
3465 | err = -EINVAL; | |
9e653b63 N |
3466 | break; |
3467 | case active: | |
3468 | if (mddev->pers) { | |
3469 | restart_array(mddev); | |
e691063a N |
3470 | if (mddev->external) |
3471 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
9e653b63 N |
3472 | wake_up(&mddev->sb_wait); |
3473 | err = 0; | |
3474 | } else { | |
3475 | mddev->ro = 0; | |
648b629e | 3476 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3477 | err = do_md_run(mddev); |
3478 | } | |
3479 | break; | |
3480 | case write_pending: | |
3481 | case active_idle: | |
3482 | /* these cannot be set */ | |
3483 | break; | |
3484 | } | |
3485 | if (err) | |
3486 | return err; | |
0fd62b86 | 3487 | else { |
00bcb4ac | 3488 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
9e653b63 | 3489 | return len; |
0fd62b86 | 3490 | } |
9e653b63 | 3491 | } |
80ca3a44 N |
3492 | static struct md_sysfs_entry md_array_state = |
3493 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3494 | |
1e50915f RB |
3495 | static ssize_t |
3496 | max_corrected_read_errors_show(mddev_t *mddev, char *page) { | |
3497 | return sprintf(page, "%d\n", | |
3498 | atomic_read(&mddev->max_corr_read_errors)); | |
3499 | } | |
3500 | ||
3501 | static ssize_t | |
3502 | max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len) | |
3503 | { | |
3504 | char *e; | |
3505 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3506 | ||
3507 | if (*buf && (*e == 0 || *e == '\n')) { | |
3508 | atomic_set(&mddev->max_corr_read_errors, n); | |
3509 | return len; | |
3510 | } | |
3511 | return -EINVAL; | |
3512 | } | |
3513 | ||
3514 | static struct md_sysfs_entry max_corr_read_errors = | |
3515 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3516 | max_corrected_read_errors_store); | |
3517 | ||
6d7ff738 N |
3518 | static ssize_t |
3519 | null_show(mddev_t *mddev, char *page) | |
3520 | { | |
3521 | return -EINVAL; | |
3522 | } | |
3523 | ||
3524 | static ssize_t | |
3525 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
3526 | { | |
3527 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3528 | /* The new device is added to the array. | |
3529 | * If the array has a persistent superblock, we read the | |
3530 | * superblock to initialise info and check validity. | |
3531 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3532 | * which mainly checks size. | |
3533 | */ | |
3534 | char *e; | |
3535 | int major = simple_strtoul(buf, &e, 10); | |
3536 | int minor; | |
3537 | dev_t dev; | |
3538 | mdk_rdev_t *rdev; | |
3539 | int err; | |
3540 | ||
3541 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3542 | return -EINVAL; | |
3543 | minor = simple_strtoul(e+1, &e, 10); | |
3544 | if (*e && *e != '\n') | |
3545 | return -EINVAL; | |
3546 | dev = MKDEV(major, minor); | |
3547 | if (major != MAJOR(dev) || | |
3548 | minor != MINOR(dev)) | |
3549 | return -EOVERFLOW; | |
3550 | ||
3551 | ||
3552 | if (mddev->persistent) { | |
3553 | rdev = md_import_device(dev, mddev->major_version, | |
3554 | mddev->minor_version); | |
3555 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3556 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3557 | mdk_rdev_t, same_set); | |
3558 | err = super_types[mddev->major_version] | |
3559 | .load_super(rdev, rdev0, mddev->minor_version); | |
3560 | if (err < 0) | |
3561 | goto out; | |
3562 | } | |
c5d79adb N |
3563 | } else if (mddev->external) |
3564 | rdev = md_import_device(dev, -2, -1); | |
3565 | else | |
6d7ff738 N |
3566 | rdev = md_import_device(dev, -1, -1); |
3567 | ||
3568 | if (IS_ERR(rdev)) | |
3569 | return PTR_ERR(rdev); | |
3570 | err = bind_rdev_to_array(rdev, mddev); | |
3571 | out: | |
3572 | if (err) | |
3573 | export_rdev(rdev); | |
3574 | return err ? err : len; | |
3575 | } | |
3576 | ||
3577 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3578 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3579 | |
9b1d1dac PC |
3580 | static ssize_t |
3581 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
3582 | { | |
3583 | char *end; | |
3584 | unsigned long chunk, end_chunk; | |
3585 | ||
3586 | if (!mddev->bitmap) | |
3587 | goto out; | |
3588 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3589 | while (*buf) { | |
3590 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3591 | if (buf == end) break; | |
3592 | if (*end == '-') { /* range */ | |
3593 | buf = end + 1; | |
3594 | end_chunk = simple_strtoul(buf, &end, 0); | |
3595 | if (buf == end) break; | |
3596 | } | |
3597 | if (*end && !isspace(*end)) break; | |
3598 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 3599 | buf = skip_spaces(end); |
9b1d1dac PC |
3600 | } |
3601 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3602 | out: | |
3603 | return len; | |
3604 | } | |
3605 | ||
3606 | static struct md_sysfs_entry md_bitmap = | |
3607 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3608 | ||
a35b0d69 N |
3609 | static ssize_t |
3610 | size_show(mddev_t *mddev, char *page) | |
3611 | { | |
58c0fed4 AN |
3612 | return sprintf(page, "%llu\n", |
3613 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3614 | } |
3615 | ||
d71f9f88 | 3616 | static int update_size(mddev_t *mddev, sector_t num_sectors); |
a35b0d69 N |
3617 | |
3618 | static ssize_t | |
3619 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
3620 | { | |
3621 | /* If array is inactive, we can reduce the component size, but | |
3622 | * not increase it (except from 0). | |
3623 | * If array is active, we can try an on-line resize | |
3624 | */ | |
b522adcd DW |
3625 | sector_t sectors; |
3626 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3627 | |
58c0fed4 AN |
3628 | if (err < 0) |
3629 | return err; | |
a35b0d69 | 3630 | if (mddev->pers) { |
58c0fed4 | 3631 | err = update_size(mddev, sectors); |
850b2b42 | 3632 | md_update_sb(mddev, 1); |
a35b0d69 | 3633 | } else { |
58c0fed4 AN |
3634 | if (mddev->dev_sectors == 0 || |
3635 | mddev->dev_sectors > sectors) | |
3636 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
3637 | else |
3638 | err = -ENOSPC; | |
3639 | } | |
3640 | return err ? err : len; | |
3641 | } | |
3642 | ||
3643 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 3644 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 3645 | |
8bb93aac N |
3646 | |
3647 | /* Metdata version. | |
e691063a N |
3648 | * This is one of |
3649 | * 'none' for arrays with no metadata (good luck...) | |
3650 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
3651 | * or N.M for internally known formats |
3652 | */ | |
3653 | static ssize_t | |
3654 | metadata_show(mddev_t *mddev, char *page) | |
3655 | { | |
3656 | if (mddev->persistent) | |
3657 | return sprintf(page, "%d.%d\n", | |
3658 | mddev->major_version, mddev->minor_version); | |
e691063a N |
3659 | else if (mddev->external) |
3660 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
3661 | else |
3662 | return sprintf(page, "none\n"); | |
3663 | } | |
3664 | ||
3665 | static ssize_t | |
3666 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
3667 | { | |
3668 | int major, minor; | |
3669 | char *e; | |
ea43ddd8 N |
3670 | /* Changing the details of 'external' metadata is |
3671 | * always permitted. Otherwise there must be | |
3672 | * no devices attached to the array. | |
3673 | */ | |
3674 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
3675 | ; | |
3676 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
3677 | return -EBUSY; |
3678 | ||
3679 | if (cmd_match(buf, "none")) { | |
3680 | mddev->persistent = 0; | |
e691063a N |
3681 | mddev->external = 0; |
3682 | mddev->major_version = 0; | |
3683 | mddev->minor_version = 90; | |
3684 | return len; | |
3685 | } | |
3686 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 3687 | size_t namelen = len-9; |
e691063a N |
3688 | if (namelen >= sizeof(mddev->metadata_type)) |
3689 | namelen = sizeof(mddev->metadata_type)-1; | |
3690 | strncpy(mddev->metadata_type, buf+9, namelen); | |
3691 | mddev->metadata_type[namelen] = 0; | |
3692 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
3693 | mddev->metadata_type[--namelen] = 0; | |
3694 | mddev->persistent = 0; | |
3695 | mddev->external = 1; | |
8bb93aac N |
3696 | mddev->major_version = 0; |
3697 | mddev->minor_version = 90; | |
3698 | return len; | |
3699 | } | |
3700 | major = simple_strtoul(buf, &e, 10); | |
3701 | if (e==buf || *e != '.') | |
3702 | return -EINVAL; | |
3703 | buf = e+1; | |
3704 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 3705 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 3706 | return -EINVAL; |
50511da3 | 3707 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
3708 | return -ENOENT; |
3709 | mddev->major_version = major; | |
3710 | mddev->minor_version = minor; | |
3711 | mddev->persistent = 1; | |
e691063a | 3712 | mddev->external = 0; |
8bb93aac N |
3713 | return len; |
3714 | } | |
3715 | ||
3716 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 3717 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 3718 | |
24dd469d | 3719 | static ssize_t |
7eec314d | 3720 | action_show(mddev_t *mddev, char *page) |
24dd469d | 3721 | { |
7eec314d | 3722 | char *type = "idle"; |
b6a9ce68 N |
3723 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
3724 | type = "frozen"; | |
3725 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 3726 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
3727 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3728 | type = "reshape"; | |
3729 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3730 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3731 | type = "resync"; | |
3732 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3733 | type = "check"; | |
3734 | else | |
3735 | type = "repair"; | |
72a23c21 | 3736 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3737 | type = "recover"; |
3738 | } | |
3739 | return sprintf(page, "%s\n", type); | |
3740 | } | |
3741 | ||
3742 | static ssize_t | |
7eec314d | 3743 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3744 | { |
7eec314d N |
3745 | if (!mddev->pers || !mddev->pers->sync_request) |
3746 | return -EINVAL; | |
3747 | ||
b6a9ce68 N |
3748 | if (cmd_match(page, "frozen")) |
3749 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3750 | else | |
3751 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3752 | ||
3753 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
3754 | if (mddev->sync_thread) { |
3755 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
3756 | md_unregister_thread(mddev->sync_thread); | |
3757 | mddev->sync_thread = NULL; | |
3758 | mddev->recovery = 0; | |
3759 | } | |
03c902e1 N |
3760 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3761 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3762 | return -EBUSY; |
72a23c21 NB |
3763 | else if (cmd_match(page, "resync")) |
3764 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3765 | else if (cmd_match(page, "recover")) { | |
3766 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3767 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3768 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3769 | int err; |
3770 | if (mddev->pers->start_reshape == NULL) | |
3771 | return -EINVAL; | |
3772 | err = mddev->pers->start_reshape(mddev); | |
3773 | if (err) | |
3774 | return err; | |
a99ac971 | 3775 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3776 | } else { |
bce74dac | 3777 | if (cmd_match(page, "check")) |
7eec314d | 3778 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3779 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3780 | return -EINVAL; |
3781 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3782 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3783 | } |
03c902e1 | 3784 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3785 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 3786 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
3787 | return len; |
3788 | } | |
3789 | ||
9d88883e | 3790 | static ssize_t |
96de1e66 | 3791 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3792 | { |
3793 | return sprintf(page, "%llu\n", | |
3794 | (unsigned long long) mddev->resync_mismatches); | |
3795 | } | |
3796 | ||
80ca3a44 N |
3797 | static struct md_sysfs_entry md_scan_mode = |
3798 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3799 | |
96de1e66 | 3800 | |
80ca3a44 | 3801 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3802 | |
88202a0c N |
3803 | static ssize_t |
3804 | sync_min_show(mddev_t *mddev, char *page) | |
3805 | { | |
3806 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3807 | mddev->sync_speed_min ? "local": "system"); | |
3808 | } | |
3809 | ||
3810 | static ssize_t | |
3811 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3812 | { | |
3813 | int min; | |
3814 | char *e; | |
3815 | if (strncmp(buf, "system", 6)==0) { | |
3816 | mddev->sync_speed_min = 0; | |
3817 | return len; | |
3818 | } | |
3819 | min = simple_strtoul(buf, &e, 10); | |
3820 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3821 | return -EINVAL; | |
3822 | mddev->sync_speed_min = min; | |
3823 | return len; | |
3824 | } | |
3825 | ||
3826 | static struct md_sysfs_entry md_sync_min = | |
3827 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3828 | ||
3829 | static ssize_t | |
3830 | sync_max_show(mddev_t *mddev, char *page) | |
3831 | { | |
3832 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3833 | mddev->sync_speed_max ? "local": "system"); | |
3834 | } | |
3835 | ||
3836 | static ssize_t | |
3837 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3838 | { | |
3839 | int max; | |
3840 | char *e; | |
3841 | if (strncmp(buf, "system", 6)==0) { | |
3842 | mddev->sync_speed_max = 0; | |
3843 | return len; | |
3844 | } | |
3845 | max = simple_strtoul(buf, &e, 10); | |
3846 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3847 | return -EINVAL; | |
3848 | mddev->sync_speed_max = max; | |
3849 | return len; | |
3850 | } | |
3851 | ||
3852 | static struct md_sysfs_entry md_sync_max = | |
3853 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3854 | ||
d7f3d291 IP |
3855 | static ssize_t |
3856 | degraded_show(mddev_t *mddev, char *page) | |
3857 | { | |
3858 | return sprintf(page, "%d\n", mddev->degraded); | |
3859 | } | |
3860 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3861 | |
90b08710 BS |
3862 | static ssize_t |
3863 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3864 | { | |
3865 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3866 | } | |
3867 | ||
3868 | static ssize_t | |
3869 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3870 | { | |
3871 | long n; | |
3872 | ||
3873 | if (strict_strtol(buf, 10, &n)) | |
3874 | return -EINVAL; | |
3875 | ||
3876 | if (n != 0 && n != 1) | |
3877 | return -EINVAL; | |
3878 | ||
3879 | mddev->parallel_resync = n; | |
3880 | ||
3881 | if (mddev->sync_thread) | |
3882 | wake_up(&resync_wait); | |
3883 | ||
3884 | return len; | |
3885 | } | |
3886 | ||
3887 | /* force parallel resync, even with shared block devices */ | |
3888 | static struct md_sysfs_entry md_sync_force_parallel = | |
3889 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3890 | sync_force_parallel_show, sync_force_parallel_store); | |
3891 | ||
88202a0c N |
3892 | static ssize_t |
3893 | sync_speed_show(mddev_t *mddev, char *page) | |
3894 | { | |
3895 | unsigned long resync, dt, db; | |
d1a7c503 N |
3896 | if (mddev->curr_resync == 0) |
3897 | return sprintf(page, "none\n"); | |
9687a60c AN |
3898 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3899 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3900 | if (!dt) dt++; |
9687a60c AN |
3901 | db = resync - mddev->resync_mark_cnt; |
3902 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3903 | } |
3904 | ||
80ca3a44 | 3905 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3906 | |
3907 | static ssize_t | |
3908 | sync_completed_show(mddev_t *mddev, char *page) | |
3909 | { | |
58c0fed4 | 3910 | unsigned long max_sectors, resync; |
88202a0c | 3911 | |
acb180b0 N |
3912 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3913 | return sprintf(page, "none\n"); | |
3914 | ||
88202a0c | 3915 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
58c0fed4 | 3916 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 3917 | else |
58c0fed4 | 3918 | max_sectors = mddev->dev_sectors; |
88202a0c | 3919 | |
acb180b0 | 3920 | resync = mddev->curr_resync_completed; |
58c0fed4 | 3921 | return sprintf(page, "%lu / %lu\n", resync, max_sectors); |
88202a0c N |
3922 | } |
3923 | ||
80ca3a44 | 3924 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3925 | |
5e96ee65 NB |
3926 | static ssize_t |
3927 | min_sync_show(mddev_t *mddev, char *page) | |
3928 | { | |
3929 | return sprintf(page, "%llu\n", | |
3930 | (unsigned long long)mddev->resync_min); | |
3931 | } | |
3932 | static ssize_t | |
3933 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3934 | { | |
3935 | unsigned long long min; | |
3936 | if (strict_strtoull(buf, 10, &min)) | |
3937 | return -EINVAL; | |
3938 | if (min > mddev->resync_max) | |
3939 | return -EINVAL; | |
3940 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3941 | return -EBUSY; | |
3942 | ||
3943 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3944 | if (mddev->chunk_sectors) { |
2ac06c33 | 3945 | sector_t temp = min; |
9d8f0363 | 3946 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
3947 | return -EINVAL; |
3948 | } | |
3949 | mddev->resync_min = min; | |
3950 | ||
3951 | return len; | |
3952 | } | |
3953 | ||
3954 | static struct md_sysfs_entry md_min_sync = | |
3955 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3956 | ||
c6207277 N |
3957 | static ssize_t |
3958 | max_sync_show(mddev_t *mddev, char *page) | |
3959 | { | |
3960 | if (mddev->resync_max == MaxSector) | |
3961 | return sprintf(page, "max\n"); | |
3962 | else | |
3963 | return sprintf(page, "%llu\n", | |
3964 | (unsigned long long)mddev->resync_max); | |
3965 | } | |
3966 | static ssize_t | |
3967 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3968 | { | |
3969 | if (strncmp(buf, "max", 3) == 0) | |
3970 | mddev->resync_max = MaxSector; | |
3971 | else { | |
5e96ee65 NB |
3972 | unsigned long long max; |
3973 | if (strict_strtoull(buf, 10, &max)) | |
3974 | return -EINVAL; | |
3975 | if (max < mddev->resync_min) | |
c6207277 N |
3976 | return -EINVAL; |
3977 | if (max < mddev->resync_max && | |
4d484a4a | 3978 | mddev->ro == 0 && |
c6207277 N |
3979 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3980 | return -EBUSY; | |
3981 | ||
3982 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3983 | if (mddev->chunk_sectors) { |
2ac06c33 | 3984 | sector_t temp = max; |
9d8f0363 | 3985 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
3986 | return -EINVAL; |
3987 | } | |
3988 | mddev->resync_max = max; | |
3989 | } | |
3990 | wake_up(&mddev->recovery_wait); | |
3991 | return len; | |
3992 | } | |
3993 | ||
3994 | static struct md_sysfs_entry md_max_sync = | |
3995 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
3996 | ||
e464eafd N |
3997 | static ssize_t |
3998 | suspend_lo_show(mddev_t *mddev, char *page) | |
3999 | { | |
4000 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4001 | } | |
4002 | ||
4003 | static ssize_t | |
4004 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
4005 | { | |
4006 | char *e; | |
4007 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
4008 | ||
b8d966ef N |
4009 | if (mddev->pers == NULL || |
4010 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4011 | return -EINVAL; |
4012 | if (buf == e || (*e && *e != '\n')) | |
4013 | return -EINVAL; | |
4014 | if (new >= mddev->suspend_hi || | |
4015 | (new > mddev->suspend_lo && new < mddev->suspend_hi)) { | |
4016 | mddev->suspend_lo = new; | |
4017 | mddev->pers->quiesce(mddev, 2); | |
4018 | return len; | |
4019 | } else | |
4020 | return -EINVAL; | |
4021 | } | |
4022 | static struct md_sysfs_entry md_suspend_lo = | |
4023 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4024 | ||
4025 | ||
4026 | static ssize_t | |
4027 | suspend_hi_show(mddev_t *mddev, char *page) | |
4028 | { | |
4029 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4030 | } | |
4031 | ||
4032 | static ssize_t | |
4033 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
4034 | { | |
4035 | char *e; | |
4036 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
4037 | ||
b8d966ef N |
4038 | if (mddev->pers == NULL || |
4039 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4040 | return -EINVAL; |
4041 | if (buf == e || (*e && *e != '\n')) | |
4042 | return -EINVAL; | |
4043 | if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || | |
4044 | (new > mddev->suspend_lo && new > mddev->suspend_hi)) { | |
4045 | mddev->suspend_hi = new; | |
4046 | mddev->pers->quiesce(mddev, 1); | |
4047 | mddev->pers->quiesce(mddev, 0); | |
4048 | return len; | |
4049 | } else | |
4050 | return -EINVAL; | |
4051 | } | |
4052 | static struct md_sysfs_entry md_suspend_hi = | |
4053 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4054 | ||
08a02ecd N |
4055 | static ssize_t |
4056 | reshape_position_show(mddev_t *mddev, char *page) | |
4057 | { | |
4058 | if (mddev->reshape_position != MaxSector) | |
4059 | return sprintf(page, "%llu\n", | |
4060 | (unsigned long long)mddev->reshape_position); | |
4061 | strcpy(page, "none\n"); | |
4062 | return 5; | |
4063 | } | |
4064 | ||
4065 | static ssize_t | |
4066 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
4067 | { | |
4068 | char *e; | |
4069 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
4070 | if (mddev->pers) | |
4071 | return -EBUSY; | |
4072 | if (buf == e || (*e && *e != '\n')) | |
4073 | return -EINVAL; | |
4074 | mddev->reshape_position = new; | |
4075 | mddev->delta_disks = 0; | |
4076 | mddev->new_level = mddev->level; | |
4077 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4078 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
08a02ecd N |
4079 | return len; |
4080 | } | |
4081 | ||
4082 | static struct md_sysfs_entry md_reshape_position = | |
4083 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4084 | reshape_position_store); | |
4085 | ||
b522adcd DW |
4086 | static ssize_t |
4087 | array_size_show(mddev_t *mddev, char *page) | |
4088 | { | |
4089 | if (mddev->external_size) | |
4090 | return sprintf(page, "%llu\n", | |
4091 | (unsigned long long)mddev->array_sectors/2); | |
4092 | else | |
4093 | return sprintf(page, "default\n"); | |
4094 | } | |
4095 | ||
4096 | static ssize_t | |
4097 | array_size_store(mddev_t *mddev, const char *buf, size_t len) | |
4098 | { | |
4099 | sector_t sectors; | |
4100 | ||
4101 | if (strncmp(buf, "default", 7) == 0) { | |
4102 | if (mddev->pers) | |
4103 | sectors = mddev->pers->size(mddev, 0, 0); | |
4104 | else | |
4105 | sectors = mddev->array_sectors; | |
4106 | ||
4107 | mddev->external_size = 0; | |
4108 | } else { | |
4109 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
4110 | return -EINVAL; | |
4111 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 4112 | return -E2BIG; |
b522adcd DW |
4113 | |
4114 | mddev->external_size = 1; | |
4115 | } | |
4116 | ||
4117 | mddev->array_sectors = sectors; | |
4118 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e N |
4119 | if (mddev->pers) |
4120 | revalidate_disk(mddev->gendisk); | |
b522adcd DW |
4121 | |
4122 | return len; | |
4123 | } | |
4124 | ||
4125 | static struct md_sysfs_entry md_array_size = | |
4126 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4127 | array_size_store); | |
e464eafd | 4128 | |
eae1701f N |
4129 | static struct attribute *md_default_attrs[] = { |
4130 | &md_level.attr, | |
d4dbd025 | 4131 | &md_layout.attr, |
eae1701f | 4132 | &md_raid_disks.attr, |
3b34380a | 4133 | &md_chunk_size.attr, |
a35b0d69 | 4134 | &md_size.attr, |
a94213b1 | 4135 | &md_resync_start.attr, |
8bb93aac | 4136 | &md_metadata.attr, |
6d7ff738 | 4137 | &md_new_device.attr, |
16f17b39 | 4138 | &md_safe_delay.attr, |
9e653b63 | 4139 | &md_array_state.attr, |
08a02ecd | 4140 | &md_reshape_position.attr, |
b522adcd | 4141 | &md_array_size.attr, |
1e50915f | 4142 | &max_corr_read_errors.attr, |
411036fa N |
4143 | NULL, |
4144 | }; | |
4145 | ||
4146 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4147 | &md_scan_mode.attr, |
9d88883e | 4148 | &md_mismatches.attr, |
88202a0c N |
4149 | &md_sync_min.attr, |
4150 | &md_sync_max.attr, | |
4151 | &md_sync_speed.attr, | |
90b08710 | 4152 | &md_sync_force_parallel.attr, |
88202a0c | 4153 | &md_sync_completed.attr, |
5e96ee65 | 4154 | &md_min_sync.attr, |
c6207277 | 4155 | &md_max_sync.attr, |
e464eafd N |
4156 | &md_suspend_lo.attr, |
4157 | &md_suspend_hi.attr, | |
9b1d1dac | 4158 | &md_bitmap.attr, |
d7f3d291 | 4159 | &md_degraded.attr, |
eae1701f N |
4160 | NULL, |
4161 | }; | |
411036fa N |
4162 | static struct attribute_group md_redundancy_group = { |
4163 | .name = NULL, | |
4164 | .attrs = md_redundancy_attrs, | |
4165 | }; | |
4166 | ||
eae1701f N |
4167 | |
4168 | static ssize_t | |
4169 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4170 | { | |
4171 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4172 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4173 | ssize_t rv; |
eae1701f N |
4174 | |
4175 | if (!entry->show) | |
4176 | return -EIO; | |
5dc5cf7d IM |
4177 | rv = mddev_lock(mddev); |
4178 | if (!rv) { | |
4179 | rv = entry->show(mddev, page); | |
4180 | mddev_unlock(mddev); | |
4181 | } | |
96de1e66 | 4182 | return rv; |
eae1701f N |
4183 | } |
4184 | ||
4185 | static ssize_t | |
4186 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4187 | const char *page, size_t length) | |
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->store) | |
4194 | return -EIO; | |
67463acb N |
4195 | if (!capable(CAP_SYS_ADMIN)) |
4196 | return -EACCES; | |
5dc5cf7d | 4197 | rv = mddev_lock(mddev); |
d3374825 N |
4198 | if (mddev->hold_active == UNTIL_IOCTL) |
4199 | mddev->hold_active = 0; | |
5dc5cf7d IM |
4200 | if (!rv) { |
4201 | rv = entry->store(mddev, page, length); | |
4202 | mddev_unlock(mddev); | |
4203 | } | |
96de1e66 | 4204 | return rv; |
eae1701f N |
4205 | } |
4206 | ||
4207 | static void md_free(struct kobject *ko) | |
4208 | { | |
4209 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
a21d1504 N |
4210 | |
4211 | if (mddev->sysfs_state) | |
4212 | sysfs_put(mddev->sysfs_state); | |
4213 | ||
4214 | if (mddev->gendisk) { | |
4215 | del_gendisk(mddev->gendisk); | |
4216 | put_disk(mddev->gendisk); | |
4217 | } | |
4218 | if (mddev->queue) | |
4219 | blk_cleanup_queue(mddev->queue); | |
4220 | ||
eae1701f N |
4221 | kfree(mddev); |
4222 | } | |
4223 | ||
52cf25d0 | 4224 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4225 | .show = md_attr_show, |
4226 | .store = md_attr_store, | |
4227 | }; | |
4228 | static struct kobj_type md_ktype = { | |
4229 | .release = md_free, | |
4230 | .sysfs_ops = &md_sysfs_ops, | |
4231 | .default_attrs = md_default_attrs, | |
4232 | }; | |
4233 | ||
1da177e4 LT |
4234 | int mdp_major = 0; |
4235 | ||
5fd3a17e DW |
4236 | static void mddev_delayed_delete(struct work_struct *ws) |
4237 | { | |
4238 | mddev_t *mddev = container_of(ws, mddev_t, del_work); | |
4239 | ||
43a70507 | 4240 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4241 | kobject_del(&mddev->kobj); |
4242 | kobject_put(&mddev->kobj); | |
4243 | } | |
4244 | ||
efeb53c0 | 4245 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4246 | { |
48c9c27b | 4247 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
4248 | mddev_t *mddev = mddev_find(dev); |
4249 | struct gendisk *disk; | |
efeb53c0 N |
4250 | int partitioned; |
4251 | int shift; | |
4252 | int unit; | |
3830c62f | 4253 | int error; |
1da177e4 LT |
4254 | |
4255 | if (!mddev) | |
efeb53c0 N |
4256 | return -ENODEV; |
4257 | ||
4258 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4259 | shift = partitioned ? MdpMinorShift : 0; | |
4260 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4261 | |
d3374825 N |
4262 | /* wait for any previous instance if this device |
4263 | * to be completed removed (mddev_delayed_delete). | |
4264 | */ | |
4265 | flush_scheduled_work(); | |
4266 | ||
48c9c27b | 4267 | mutex_lock(&disks_mutex); |
0909dc44 N |
4268 | error = -EEXIST; |
4269 | if (mddev->gendisk) | |
4270 | goto abort; | |
efeb53c0 N |
4271 | |
4272 | if (name) { | |
4273 | /* Need to ensure that 'name' is not a duplicate. | |
4274 | */ | |
4275 | mddev_t *mddev2; | |
4276 | spin_lock(&all_mddevs_lock); | |
4277 | ||
4278 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4279 | if (mddev2->gendisk && | |
4280 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4281 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4282 | goto abort; |
efeb53c0 N |
4283 | } |
4284 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4285 | } |
8b765398 | 4286 | |
0909dc44 | 4287 | error = -ENOMEM; |
8b765398 | 4288 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4289 | if (!mddev->queue) |
4290 | goto abort; | |
409c57f3 N |
4291 | mddev->queue->queuedata = mddev; |
4292 | ||
8b765398 N |
4293 | /* Can be unlocked because the queue is new: no concurrency */ |
4294 | queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue); | |
4295 | ||
409c57f3 | 4296 | blk_queue_make_request(mddev->queue, md_make_request); |
8b765398 | 4297 | |
1da177e4 LT |
4298 | disk = alloc_disk(1 << shift); |
4299 | if (!disk) { | |
8b765398 N |
4300 | blk_cleanup_queue(mddev->queue); |
4301 | mddev->queue = NULL; | |
0909dc44 | 4302 | goto abort; |
1da177e4 | 4303 | } |
efeb53c0 | 4304 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4305 | disk->first_minor = unit << shift; |
efeb53c0 N |
4306 | if (name) |
4307 | strcpy(disk->disk_name, name); | |
4308 | else if (partitioned) | |
1da177e4 | 4309 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4310 | else |
1da177e4 | 4311 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4312 | disk->fops = &md_fops; |
4313 | disk->private_data = mddev; | |
4314 | disk->queue = mddev->queue; | |
92850bbd | 4315 | /* Allow extended partitions. This makes the |
d3374825 | 4316 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4317 | * remove it now. |
4318 | */ | |
4319 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 LT |
4320 | add_disk(disk); |
4321 | mddev->gendisk = disk; | |
ed9e1982 TH |
4322 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4323 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4324 | if (error) { |
4325 | /* This isn't possible, but as kobject_init_and_add is marked | |
4326 | * __must_check, we must do something with the result | |
4327 | */ | |
5e55e2f5 N |
4328 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4329 | disk->disk_name); | |
0909dc44 N |
4330 | error = 0; |
4331 | } | |
00bcb4ac N |
4332 | if (mddev->kobj.sd && |
4333 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4334 | printk(KERN_DEBUG "pointless warning\n"); |
0909dc44 N |
4335 | abort: |
4336 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4337 | if (!error && mddev->kobj.sd) { |
3830c62f | 4338 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4339 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4340 | } |
d3374825 | 4341 | mddev_put(mddev); |
0909dc44 | 4342 | return error; |
efeb53c0 N |
4343 | } |
4344 | ||
4345 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4346 | { | |
4347 | md_alloc(dev, NULL); | |
1da177e4 LT |
4348 | return NULL; |
4349 | } | |
4350 | ||
efeb53c0 N |
4351 | static int add_named_array(const char *val, struct kernel_param *kp) |
4352 | { | |
4353 | /* val must be "md_*" where * is not all digits. | |
4354 | * We allocate an array with a large free minor number, and | |
4355 | * set the name to val. val must not already be an active name. | |
4356 | */ | |
4357 | int len = strlen(val); | |
4358 | char buf[DISK_NAME_LEN]; | |
4359 | ||
4360 | while (len && val[len-1] == '\n') | |
4361 | len--; | |
4362 | if (len >= DISK_NAME_LEN) | |
4363 | return -E2BIG; | |
4364 | strlcpy(buf, val, len+1); | |
4365 | if (strncmp(buf, "md_", 3) != 0) | |
4366 | return -EINVAL; | |
4367 | return md_alloc(0, buf); | |
4368 | } | |
4369 | ||
1da177e4 LT |
4370 | static void md_safemode_timeout(unsigned long data) |
4371 | { | |
4372 | mddev_t *mddev = (mddev_t *) data; | |
4373 | ||
0fd62b86 NB |
4374 | if (!atomic_read(&mddev->writes_pending)) { |
4375 | mddev->safemode = 1; | |
4376 | if (mddev->external) | |
00bcb4ac | 4377 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4378 | } |
1da177e4 LT |
4379 | md_wakeup_thread(mddev->thread); |
4380 | } | |
4381 | ||
6ff8d8ec | 4382 | static int start_dirty_degraded; |
1da177e4 | 4383 | |
390ee602 | 4384 | int md_run(mddev_t *mddev) |
1da177e4 | 4385 | { |
2604b703 | 4386 | int err; |
1da177e4 | 4387 | mdk_rdev_t *rdev; |
2604b703 | 4388 | struct mdk_personality *pers; |
1da177e4 | 4389 | |
a757e64c N |
4390 | if (list_empty(&mddev->disks)) |
4391 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4392 | return -EINVAL; |
1da177e4 LT |
4393 | |
4394 | if (mddev->pers) | |
4395 | return -EBUSY; | |
4396 | ||
b6eb127d N |
4397 | /* These two calls synchronise us with the |
4398 | * sysfs_remove_group calls in mddev_unlock, | |
4399 | * so they must have completed. | |
4400 | */ | |
4401 | mutex_lock(&mddev->open_mutex); | |
4402 | mutex_unlock(&mddev->open_mutex); | |
4403 | ||
1da177e4 LT |
4404 | /* |
4405 | * Analyze all RAID superblock(s) | |
4406 | */ | |
1ec4a939 N |
4407 | if (!mddev->raid_disks) { |
4408 | if (!mddev->persistent) | |
4409 | return -EINVAL; | |
a757e64c | 4410 | analyze_sbs(mddev); |
1ec4a939 | 4411 | } |
1da177e4 | 4412 | |
d9d166c2 N |
4413 | if (mddev->level != LEVEL_NONE) |
4414 | request_module("md-level-%d", mddev->level); | |
4415 | else if (mddev->clevel[0]) | |
4416 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4417 | |
4418 | /* | |
4419 | * Drop all container device buffers, from now on | |
4420 | * the only valid external interface is through the md | |
4421 | * device. | |
1da177e4 | 4422 | */ |
159ec1fc | 4423 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 | 4424 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4425 | continue; |
4426 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4427 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4428 | |
4429 | /* perform some consistency tests on the device. | |
4430 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4431 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4432 | */ |
0f420358 | 4433 | if (rdev->data_offset < rdev->sb_start) { |
58c0fed4 AN |
4434 | if (mddev->dev_sectors && |
4435 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4436 | > rdev->sb_start) { |
f0d76d70 N |
4437 | printk("md: %s: data overlaps metadata\n", |
4438 | mdname(mddev)); | |
4439 | return -EINVAL; | |
4440 | } | |
4441 | } else { | |
0f420358 | 4442 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4443 | > rdev->data_offset) { |
4444 | printk("md: %s: metadata overlaps data\n", | |
4445 | mdname(mddev)); | |
4446 | return -EINVAL; | |
4447 | } | |
4448 | } | |
00bcb4ac | 4449 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
4450 | } |
4451 | ||
1da177e4 | 4452 | spin_lock(&pers_lock); |
d9d166c2 | 4453 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4454 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4455 | spin_unlock(&pers_lock); |
d9d166c2 N |
4456 | if (mddev->level != LEVEL_NONE) |
4457 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
4458 | mddev->level); | |
4459 | else | |
4460 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
4461 | mddev->clevel); | |
1da177e4 LT |
4462 | return -EINVAL; |
4463 | } | |
2604b703 | 4464 | mddev->pers = pers; |
1da177e4 | 4465 | spin_unlock(&pers_lock); |
34817e8c N |
4466 | if (mddev->level != pers->level) { |
4467 | mddev->level = pers->level; | |
4468 | mddev->new_level = pers->level; | |
4469 | } | |
d9d166c2 | 4470 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 4471 | |
f6705578 | 4472 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 4473 | pers->start_reshape == NULL) { |
f6705578 N |
4474 | /* This personality cannot handle reshaping... */ |
4475 | mddev->pers = NULL; | |
4476 | module_put(pers->owner); | |
4477 | return -EINVAL; | |
4478 | } | |
4479 | ||
7dd5e7c3 N |
4480 | if (pers->sync_request) { |
4481 | /* Warn if this is a potentially silly | |
4482 | * configuration. | |
4483 | */ | |
4484 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4485 | mdk_rdev_t *rdev2; | |
7dd5e7c3 | 4486 | int warned = 0; |
159ec1fc CR |
4487 | |
4488 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4489 | list_for_each_entry(rdev2, &mddev->disks, same_set) { | |
7dd5e7c3 N |
4490 | if (rdev < rdev2 && |
4491 | rdev->bdev->bd_contains == | |
4492 | rdev2->bdev->bd_contains) { | |
4493 | printk(KERN_WARNING | |
4494 | "%s: WARNING: %s appears to be" | |
4495 | " on the same physical disk as" | |
4496 | " %s.\n", | |
4497 | mdname(mddev), | |
4498 | bdevname(rdev->bdev,b), | |
4499 | bdevname(rdev2->bdev,b2)); | |
4500 | warned = 1; | |
4501 | } | |
4502 | } | |
159ec1fc | 4503 | |
7dd5e7c3 N |
4504 | if (warned) |
4505 | printk(KERN_WARNING | |
4506 | "True protection against single-disk" | |
4507 | " failure might be compromised.\n"); | |
4508 | } | |
4509 | ||
657390d2 | 4510 | mddev->recovery = 0; |
58c0fed4 AN |
4511 | /* may be over-ridden by personality */ |
4512 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4513 | ||
a9701a30 | 4514 | mddev->barriers_work = 1; |
6ff8d8ec | 4515 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 4516 | |
0f9552b5 | 4517 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
4518 | mddev->ro = 2; /* read-only, but switch on first write */ |
4519 | ||
b15c2e57 | 4520 | err = mddev->pers->run(mddev); |
13e53df3 AN |
4521 | if (err) |
4522 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
4523 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
4524 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
4525 | " but 'external_size' not in effect?\n", __func__); | |
4526 | printk(KERN_ERR | |
4527 | "md: invalid array_size %llu > default size %llu\n", | |
4528 | (unsigned long long)mddev->array_sectors / 2, | |
4529 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
4530 | err = -EINVAL; | |
4531 | mddev->pers->stop(mddev); | |
4532 | } | |
4533 | if (err == 0 && mddev->pers->sync_request) { | |
b15c2e57 N |
4534 | err = bitmap_create(mddev); |
4535 | if (err) { | |
4536 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
4537 | mdname(mddev), err); | |
4538 | mddev->pers->stop(mddev); | |
4539 | } | |
4540 | } | |
1da177e4 | 4541 | if (err) { |
1da177e4 LT |
4542 | module_put(mddev->pers->owner); |
4543 | mddev->pers = NULL; | |
32a7627c N |
4544 | bitmap_destroy(mddev); |
4545 | return err; | |
1da177e4 | 4546 | } |
5e55e2f5 | 4547 | if (mddev->pers->sync_request) { |
00bcb4ac N |
4548 | if (mddev->kobj.sd && |
4549 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
4550 | printk(KERN_WARNING |
4551 | "md: cannot register extra attributes for %s\n", | |
4552 | mdname(mddev)); | |
00bcb4ac | 4553 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 4554 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
4555 | mddev->ro = 0; |
4556 | ||
1da177e4 | 4557 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
4558 | atomic_set(&mddev->max_corr_read_errors, |
4559 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
4560 | mddev->safemode = 0; |
4561 | mddev->safemode_timer.function = md_safemode_timeout; | |
4562 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 4563 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 4564 | mddev->in_sync = 1; |
86e6ffdd | 4565 | |
159ec1fc | 4566 | list_for_each_entry(rdev, &mddev->disks, same_set) |
86e6ffdd N |
4567 | if (rdev->raid_disk >= 0) { |
4568 | char nm[20]; | |
4569 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 | 4570 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
00bcb4ac | 4571 | /* failure here is OK */; |
86e6ffdd | 4572 | } |
1da177e4 LT |
4573 | |
4574 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4575 | ||
850b2b42 N |
4576 | if (mddev->flags) |
4577 | md_update_sb(mddev, 0); | |
1da177e4 | 4578 | |
0b8c9de0 N |
4579 | md_wakeup_thread(mddev->thread); |
4580 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 4581 | |
d7603b7e | 4582 | md_new_event(mddev); |
00bcb4ac N |
4583 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
4584 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 4585 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
4586 | return 0; |
4587 | } | |
390ee602 | 4588 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 4589 | |
fe60b014 N |
4590 | static int do_md_run(mddev_t *mddev) |
4591 | { | |
4592 | int err; | |
4593 | ||
4594 | err = md_run(mddev); | |
4595 | if (err) | |
4596 | goto out; | |
4597 | ||
4598 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4599 | revalidate_disk(mddev->gendisk); | |
4600 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); | |
4601 | out: | |
4602 | return err; | |
4603 | } | |
4604 | ||
1da177e4 LT |
4605 | static int restart_array(mddev_t *mddev) |
4606 | { | |
4607 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 4608 | |
80fab1d7 | 4609 | /* Complain if it has no devices */ |
1da177e4 | 4610 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
4611 | return -ENXIO; |
4612 | if (!mddev->pers) | |
4613 | return -EINVAL; | |
4614 | if (!mddev->ro) | |
4615 | return -EBUSY; | |
4616 | mddev->safemode = 0; | |
4617 | mddev->ro = 0; | |
4618 | set_disk_ro(disk, 0); | |
4619 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
4620 | mdname(mddev)); | |
4621 | /* Kick recovery or resync if necessary */ | |
4622 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4623 | md_wakeup_thread(mddev->thread); | |
4624 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 4625 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 4626 | return 0; |
1da177e4 LT |
4627 | } |
4628 | ||
acc55e22 N |
4629 | /* similar to deny_write_access, but accounts for our holding a reference |
4630 | * to the file ourselves */ | |
4631 | static int deny_bitmap_write_access(struct file * file) | |
4632 | { | |
4633 | struct inode *inode = file->f_mapping->host; | |
4634 | ||
4635 | spin_lock(&inode->i_lock); | |
4636 | if (atomic_read(&inode->i_writecount) > 1) { | |
4637 | spin_unlock(&inode->i_lock); | |
4638 | return -ETXTBSY; | |
4639 | } | |
4640 | atomic_set(&inode->i_writecount, -1); | |
4641 | spin_unlock(&inode->i_lock); | |
4642 | ||
4643 | return 0; | |
4644 | } | |
4645 | ||
43a70507 | 4646 | void restore_bitmap_write_access(struct file *file) |
acc55e22 N |
4647 | { |
4648 | struct inode *inode = file->f_mapping->host; | |
4649 | ||
4650 | spin_lock(&inode->i_lock); | |
4651 | atomic_set(&inode->i_writecount, 1); | |
4652 | spin_unlock(&inode->i_lock); | |
4653 | } | |
4654 | ||
6177b472 N |
4655 | static void md_clean(mddev_t *mddev) |
4656 | { | |
4657 | mddev->array_sectors = 0; | |
4658 | mddev->external_size = 0; | |
4659 | mddev->dev_sectors = 0; | |
4660 | mddev->raid_disks = 0; | |
4661 | mddev->recovery_cp = 0; | |
4662 | mddev->resync_min = 0; | |
4663 | mddev->resync_max = MaxSector; | |
4664 | mddev->reshape_position = MaxSector; | |
4665 | mddev->external = 0; | |
4666 | mddev->persistent = 0; | |
4667 | mddev->level = LEVEL_NONE; | |
4668 | mddev->clevel[0] = 0; | |
4669 | mddev->flags = 0; | |
4670 | mddev->ro = 0; | |
4671 | mddev->metadata_type[0] = 0; | |
4672 | mddev->chunk_sectors = 0; | |
4673 | mddev->ctime = mddev->utime = 0; | |
4674 | mddev->layout = 0; | |
4675 | mddev->max_disks = 0; | |
4676 | mddev->events = 0; | |
a8707c08 | 4677 | mddev->can_decrease_events = 0; |
6177b472 N |
4678 | mddev->delta_disks = 0; |
4679 | mddev->new_level = LEVEL_NONE; | |
4680 | mddev->new_layout = 0; | |
4681 | mddev->new_chunk_sectors = 0; | |
4682 | mddev->curr_resync = 0; | |
4683 | mddev->resync_mismatches = 0; | |
4684 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
4685 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
4686 | mddev->recovery = 0; | |
4687 | mddev->in_sync = 0; | |
4688 | mddev->degraded = 0; | |
4689 | mddev->barriers_work = 0; | |
4690 | mddev->safemode = 0; | |
4691 | mddev->bitmap_info.offset = 0; | |
4692 | mddev->bitmap_info.default_offset = 0; | |
4693 | mddev->bitmap_info.chunksize = 0; | |
4694 | mddev->bitmap_info.daemon_sleep = 0; | |
4695 | mddev->bitmap_info.max_write_behind = 0; | |
252ac522 | 4696 | mddev->plug = NULL; |
6177b472 N |
4697 | } |
4698 | ||
390ee602 | 4699 | void md_stop_writes(mddev_t *mddev) |
a047e125 N |
4700 | { |
4701 | if (mddev->sync_thread) { | |
4702 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4703 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
4704 | md_unregister_thread(mddev->sync_thread); | |
4705 | mddev->sync_thread = NULL; | |
4706 | } | |
4707 | ||
4708 | del_timer_sync(&mddev->safemode_timer); | |
4709 | ||
4710 | bitmap_flush(mddev); | |
4711 | md_super_wait(mddev); | |
4712 | ||
4713 | if (!mddev->in_sync || mddev->flags) { | |
4714 | /* mark array as shutdown cleanly */ | |
4715 | mddev->in_sync = 1; | |
4716 | md_update_sb(mddev, 1); | |
4717 | } | |
4718 | } | |
390ee602 | 4719 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 4720 | |
390ee602 | 4721 | void md_stop(mddev_t *mddev) |
6177b472 N |
4722 | { |
4723 | mddev->pers->stop(mddev); | |
4724 | if (mddev->pers->sync_request && mddev->to_remove == NULL) | |
4725 | mddev->to_remove = &md_redundancy_group; | |
4726 | module_put(mddev->pers->owner); | |
4727 | mddev->pers = NULL; | |
cca9cf90 | 4728 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 4729 | } |
390ee602 | 4730 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 4731 | |
a4bd82d0 N |
4732 | static int md_set_readonly(mddev_t *mddev, int is_open) |
4733 | { | |
4734 | int err = 0; | |
4735 | mutex_lock(&mddev->open_mutex); | |
4736 | if (atomic_read(&mddev->openers) > is_open) { | |
4737 | printk("md: %s still in use.\n",mdname(mddev)); | |
4738 | err = -EBUSY; | |
4739 | goto out; | |
4740 | } | |
4741 | if (mddev->pers) { | |
4742 | md_stop_writes(mddev); | |
4743 | ||
4744 | err = -ENXIO; | |
4745 | if (mddev->ro==1) | |
4746 | goto out; | |
4747 | mddev->ro = 1; | |
4748 | set_disk_ro(mddev->gendisk, 1); | |
4749 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
00bcb4ac | 4750 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
a4bd82d0 N |
4751 | err = 0; |
4752 | } | |
4753 | out: | |
4754 | mutex_unlock(&mddev->open_mutex); | |
4755 | return err; | |
4756 | } | |
4757 | ||
9e653b63 N |
4758 | /* mode: |
4759 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
4760 | * 2 - stop but do not disassemble array |
4761 | */ | |
df5b20cf | 4762 | static int do_md_stop(mddev_t * mddev, int mode, int is_open) |
1da177e4 LT |
4763 | { |
4764 | int err = 0; | |
4765 | struct gendisk *disk = mddev->gendisk; | |
c4647292 | 4766 | mdk_rdev_t *rdev; |
1da177e4 | 4767 | |
c8c00a69 | 4768 | mutex_lock(&mddev->open_mutex); |
f2ea68cf | 4769 | if (atomic_read(&mddev->openers) > is_open) { |
df5b20cf | 4770 | printk("md: %s still in use.\n",mdname(mddev)); |
c8c00a69 N |
4771 | err = -EBUSY; |
4772 | } else if (mddev->pers) { | |
1da177e4 | 4773 | |
a4bd82d0 N |
4774 | if (mddev->ro) |
4775 | set_disk_ro(disk, 0); | |
409c57f3 | 4776 | |
390ee602 | 4777 | md_stop_writes(mddev); |
a4bd82d0 N |
4778 | md_stop(mddev); |
4779 | mddev->queue->merge_bvec_fn = NULL; | |
4780 | mddev->queue->unplug_fn = NULL; | |
4781 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
6177b472 | 4782 | |
a4bd82d0 | 4783 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 4784 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 4785 | |
a4bd82d0 N |
4786 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4787 | if (rdev->raid_disk >= 0) { | |
4788 | char nm[20]; | |
4789 | sprintf(nm, "rd%d", rdev->raid_disk); | |
4790 | sysfs_remove_link(&mddev->kobj, nm); | |
4791 | } | |
c4647292 | 4792 | |
a4bd82d0 N |
4793 | set_capacity(disk, 0); |
4794 | revalidate_disk(disk); | |
0d4ca600 | 4795 | |
a4bd82d0 N |
4796 | if (mddev->ro) |
4797 | mddev->ro = 0; | |
4798 | ||
80ffb3cc | 4799 | err = 0; |
1da177e4 | 4800 | } |
c8c00a69 N |
4801 | mutex_unlock(&mddev->open_mutex); |
4802 | if (err) | |
4803 | return err; | |
1da177e4 LT |
4804 | /* |
4805 | * Free resources if final stop | |
4806 | */ | |
9e653b63 | 4807 | if (mode == 0) { |
0d4ca600 | 4808 | |
1da177e4 LT |
4809 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
4810 | ||
978f946b | 4811 | bitmap_destroy(mddev); |
c3d9714e N |
4812 | if (mddev->bitmap_info.file) { |
4813 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
4814 | fput(mddev->bitmap_info.file); | |
4815 | mddev->bitmap_info.file = NULL; | |
978f946b | 4816 | } |
c3d9714e | 4817 | mddev->bitmap_info.offset = 0; |
978f946b | 4818 | |
1da177e4 LT |
4819 | export_array(mddev); |
4820 | ||
6177b472 | 4821 | md_clean(mddev); |
934d9c23 | 4822 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
4823 | if (mddev->hold_active == UNTIL_STOP) |
4824 | mddev->hold_active = 0; | |
9e653b63 | 4825 | |
a4bd82d0 | 4826 | } |
1da177e4 | 4827 | err = 0; |
3f9d99c1 | 4828 | blk_integrity_unregister(disk); |
d7603b7e | 4829 | md_new_event(mddev); |
00bcb4ac | 4830 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
1da177e4 LT |
4831 | return err; |
4832 | } | |
4833 | ||
fdee8ae4 | 4834 | #ifndef MODULE |
1da177e4 LT |
4835 | static void autorun_array(mddev_t *mddev) |
4836 | { | |
4837 | mdk_rdev_t *rdev; | |
1da177e4 LT |
4838 | int err; |
4839 | ||
a757e64c | 4840 | if (list_empty(&mddev->disks)) |
1da177e4 | 4841 | return; |
1da177e4 LT |
4842 | |
4843 | printk(KERN_INFO "md: running: "); | |
4844 | ||
159ec1fc | 4845 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
4846 | char b[BDEVNAME_SIZE]; |
4847 | printk("<%s>", bdevname(rdev->bdev,b)); | |
4848 | } | |
4849 | printk("\n"); | |
4850 | ||
d710e138 | 4851 | err = do_md_run(mddev); |
1da177e4 LT |
4852 | if (err) { |
4853 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
d710e138 | 4854 | do_md_stop(mddev, 0, 0); |
1da177e4 LT |
4855 | } |
4856 | } | |
4857 | ||
4858 | /* | |
4859 | * lets try to run arrays based on all disks that have arrived | |
4860 | * until now. (those are in pending_raid_disks) | |
4861 | * | |
4862 | * the method: pick the first pending disk, collect all disks with | |
4863 | * the same UUID, remove all from the pending list and put them into | |
4864 | * the 'same_array' list. Then order this list based on superblock | |
4865 | * update time (freshest comes first), kick out 'old' disks and | |
4866 | * compare superblocks. If everything's fine then run it. | |
4867 | * | |
4868 | * If "unit" is allocated, then bump its reference count | |
4869 | */ | |
4870 | static void autorun_devices(int part) | |
4871 | { | |
159ec1fc | 4872 | mdk_rdev_t *rdev0, *rdev, *tmp; |
1da177e4 LT |
4873 | mddev_t *mddev; |
4874 | char b[BDEVNAME_SIZE]; | |
4875 | ||
4876 | printk(KERN_INFO "md: autorun ...\n"); | |
4877 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4878 | int unit; |
1da177e4 | 4879 | dev_t dev; |
ad01c9e3 | 4880 | LIST_HEAD(candidates); |
1da177e4 LT |
4881 | rdev0 = list_entry(pending_raid_disks.next, |
4882 | mdk_rdev_t, same_set); | |
4883 | ||
4884 | printk(KERN_INFO "md: considering %s ...\n", | |
4885 | bdevname(rdev0->bdev,b)); | |
4886 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 4887 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
4888 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4889 | printk(KERN_INFO "md: adding %s ...\n", | |
4890 | bdevname(rdev->bdev,b)); | |
4891 | list_move(&rdev->same_set, &candidates); | |
4892 | } | |
4893 | /* | |
4894 | * now we have a set of devices, with all of them having | |
4895 | * mostly sane superblocks. It's time to allocate the | |
4896 | * mddev. | |
4897 | */ | |
e8703fe1 N |
4898 | if (part) { |
4899 | dev = MKDEV(mdp_major, | |
4900 | rdev0->preferred_minor << MdpMinorShift); | |
4901 | unit = MINOR(dev) >> MdpMinorShift; | |
4902 | } else { | |
4903 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4904 | unit = MINOR(dev); | |
4905 | } | |
4906 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4907 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4908 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4909 | break; | |
4910 | } | |
1da177e4 LT |
4911 | |
4912 | md_probe(dev, NULL, NULL); | |
4913 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4914 | if (!mddev || !mddev->gendisk) { |
4915 | if (mddev) | |
4916 | mddev_put(mddev); | |
4917 | printk(KERN_ERR | |
1da177e4 LT |
4918 | "md: cannot allocate memory for md drive.\n"); |
4919 | break; | |
4920 | } | |
4921 | if (mddev_lock(mddev)) | |
4922 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4923 | mdname(mddev)); | |
4924 | else if (mddev->raid_disks || mddev->major_version | |
4925 | || !list_empty(&mddev->disks)) { | |
4926 | printk(KERN_WARNING | |
4927 | "md: %s already running, cannot run %s\n", | |
4928 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4929 | mddev_unlock(mddev); | |
4930 | } else { | |
4931 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4932 | mddev->persistent = 1; |
159ec1fc | 4933 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
4934 | list_del_init(&rdev->same_set); |
4935 | if (bind_rdev_to_array(rdev, mddev)) | |
4936 | export_rdev(rdev); | |
4937 | } | |
4938 | autorun_array(mddev); | |
4939 | mddev_unlock(mddev); | |
4940 | } | |
4941 | /* on success, candidates will be empty, on error | |
4942 | * it won't... | |
4943 | */ | |
159ec1fc | 4944 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 4945 | list_del_init(&rdev->same_set); |
1da177e4 | 4946 | export_rdev(rdev); |
4b80991c | 4947 | } |
1da177e4 LT |
4948 | mddev_put(mddev); |
4949 | } | |
4950 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
4951 | } | |
fdee8ae4 | 4952 | #endif /* !MODULE */ |
1da177e4 | 4953 | |
1da177e4 LT |
4954 | static int get_version(void __user * arg) |
4955 | { | |
4956 | mdu_version_t ver; | |
4957 | ||
4958 | ver.major = MD_MAJOR_VERSION; | |
4959 | ver.minor = MD_MINOR_VERSION; | |
4960 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
4961 | ||
4962 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
4963 | return -EFAULT; | |
4964 | ||
4965 | return 0; | |
4966 | } | |
4967 | ||
4968 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
4969 | { | |
4970 | mdu_array_info_t info; | |
a9f326eb | 4971 | int nr,working,insync,failed,spare; |
1da177e4 | 4972 | mdk_rdev_t *rdev; |
1da177e4 | 4973 | |
a9f326eb | 4974 | nr=working=insync=failed=spare=0; |
159ec1fc | 4975 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4976 | nr++; |
b2d444d7 | 4977 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4978 | failed++; |
4979 | else { | |
4980 | working++; | |
b2d444d7 | 4981 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 4982 | insync++; |
1da177e4 LT |
4983 | else |
4984 | spare++; | |
4985 | } | |
4986 | } | |
4987 | ||
4988 | info.major_version = mddev->major_version; | |
4989 | info.minor_version = mddev->minor_version; | |
4990 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
4991 | info.ctime = mddev->ctime; | |
4992 | info.level = mddev->level; | |
58c0fed4 AN |
4993 | info.size = mddev->dev_sectors / 2; |
4994 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 4995 | info.size = -1; |
1da177e4 LT |
4996 | info.nr_disks = nr; |
4997 | info.raid_disks = mddev->raid_disks; | |
4998 | info.md_minor = mddev->md_minor; | |
4999 | info.not_persistent= !mddev->persistent; | |
5000 | ||
5001 | info.utime = mddev->utime; | |
5002 | info.state = 0; | |
5003 | if (mddev->in_sync) | |
5004 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5005 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5006 | info.state = (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 5007 | info.active_disks = insync; |
1da177e4 LT |
5008 | info.working_disks = working; |
5009 | info.failed_disks = failed; | |
5010 | info.spare_disks = spare; | |
5011 | ||
5012 | info.layout = mddev->layout; | |
9d8f0363 | 5013 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5014 | |
5015 | if (copy_to_user(arg, &info, sizeof(info))) | |
5016 | return -EFAULT; | |
5017 | ||
5018 | return 0; | |
5019 | } | |
5020 | ||
87162a28 | 5021 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
5022 | { |
5023 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
5024 | char *ptr, *buf = NULL; | |
5025 | int err = -ENOMEM; | |
5026 | ||
b5470dc5 DW |
5027 | if (md_allow_write(mddev)) |
5028 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
5029 | else | |
5030 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 5031 | |
32a7627c N |
5032 | if (!file) |
5033 | goto out; | |
5034 | ||
5035 | /* bitmap disabled, zero the first byte and copy out */ | |
5036 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
5037 | file->pathname[0] = '\0'; | |
5038 | goto copy_out; | |
5039 | } | |
5040 | ||
5041 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
5042 | if (!buf) | |
5043 | goto out; | |
5044 | ||
6bcfd601 CH |
5045 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
5046 | if (IS_ERR(ptr)) | |
32a7627c N |
5047 | goto out; |
5048 | ||
5049 | strcpy(file->pathname, ptr); | |
5050 | ||
5051 | copy_out: | |
5052 | err = 0; | |
5053 | if (copy_to_user(arg, file, sizeof(*file))) | |
5054 | err = -EFAULT; | |
5055 | out: | |
5056 | kfree(buf); | |
5057 | kfree(file); | |
5058 | return err; | |
5059 | } | |
5060 | ||
1da177e4 LT |
5061 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
5062 | { | |
5063 | mdu_disk_info_t info; | |
1da177e4 LT |
5064 | mdk_rdev_t *rdev; |
5065 | ||
5066 | if (copy_from_user(&info, arg, sizeof(info))) | |
5067 | return -EFAULT; | |
5068 | ||
26ef379f | 5069 | rdev = find_rdev_nr(mddev, info.number); |
1da177e4 LT |
5070 | if (rdev) { |
5071 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5072 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5073 | info.raid_disk = rdev->raid_disk; | |
5074 | info.state = 0; | |
b2d444d7 | 5075 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5076 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5077 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5078 | info.state |= (1<<MD_DISK_ACTIVE); |
5079 | info.state |= (1<<MD_DISK_SYNC); | |
5080 | } | |
8ddf9efe N |
5081 | if (test_bit(WriteMostly, &rdev->flags)) |
5082 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5083 | } else { |
5084 | info.major = info.minor = 0; | |
5085 | info.raid_disk = -1; | |
5086 | info.state = (1<<MD_DISK_REMOVED); | |
5087 | } | |
5088 | ||
5089 | if (copy_to_user(arg, &info, sizeof(info))) | |
5090 | return -EFAULT; | |
5091 | ||
5092 | return 0; | |
5093 | } | |
5094 | ||
5095 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
5096 | { | |
5097 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
5098 | mdk_rdev_t *rdev; | |
5099 | dev_t dev = MKDEV(info->major,info->minor); | |
5100 | ||
5101 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
5102 | return -EOVERFLOW; | |
5103 | ||
5104 | if (!mddev->raid_disks) { | |
5105 | int err; | |
5106 | /* expecting a device which has a superblock */ | |
5107 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5108 | if (IS_ERR(rdev)) { | |
5109 | printk(KERN_WARNING | |
5110 | "md: md_import_device returned %ld\n", | |
5111 | PTR_ERR(rdev)); | |
5112 | return PTR_ERR(rdev); | |
5113 | } | |
5114 | if (!list_empty(&mddev->disks)) { | |
5115 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
5116 | mdk_rdev_t, same_set); | |
a9f326eb | 5117 | err = super_types[mddev->major_version] |
1da177e4 LT |
5118 | .load_super(rdev, rdev0, mddev->minor_version); |
5119 | if (err < 0) { | |
5120 | printk(KERN_WARNING | |
5121 | "md: %s has different UUID to %s\n", | |
5122 | bdevname(rdev->bdev,b), | |
5123 | bdevname(rdev0->bdev,b2)); | |
5124 | export_rdev(rdev); | |
5125 | return -EINVAL; | |
5126 | } | |
5127 | } | |
5128 | err = bind_rdev_to_array(rdev, mddev); | |
5129 | if (err) | |
5130 | export_rdev(rdev); | |
5131 | return err; | |
5132 | } | |
5133 | ||
5134 | /* | |
5135 | * add_new_disk can be used once the array is assembled | |
5136 | * to add "hot spares". They must already have a superblock | |
5137 | * written | |
5138 | */ | |
5139 | if (mddev->pers) { | |
5140 | int err; | |
5141 | if (!mddev->pers->hot_add_disk) { | |
5142 | printk(KERN_WARNING | |
5143 | "%s: personality does not support diskops!\n", | |
5144 | mdname(mddev)); | |
5145 | return -EINVAL; | |
5146 | } | |
7b1e35f6 N |
5147 | if (mddev->persistent) |
5148 | rdev = md_import_device(dev, mddev->major_version, | |
5149 | mddev->minor_version); | |
5150 | else | |
5151 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
5152 | if (IS_ERR(rdev)) { |
5153 | printk(KERN_WARNING | |
5154 | "md: md_import_device returned %ld\n", | |
5155 | PTR_ERR(rdev)); | |
5156 | return PTR_ERR(rdev); | |
5157 | } | |
41158c7e N |
5158 | /* set save_raid_disk if appropriate */ |
5159 | if (!mddev->persistent) { | |
5160 | if (info->state & (1<<MD_DISK_SYNC) && | |
5161 | info->raid_disk < mddev->raid_disks) | |
5162 | rdev->raid_disk = info->raid_disk; | |
5163 | else | |
5164 | rdev->raid_disk = -1; | |
5165 | } else | |
5166 | super_types[mddev->major_version]. | |
5167 | validate_super(mddev, rdev); | |
5168 | rdev->saved_raid_disk = rdev->raid_disk; | |
5169 | ||
b2d444d7 | 5170 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5171 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5172 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5173 | else |
5174 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5175 | |
1da177e4 LT |
5176 | rdev->raid_disk = -1; |
5177 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5178 | if (!err && !mddev->pers->hot_remove_disk) { |
5179 | /* If there is hot_add_disk but no hot_remove_disk | |
5180 | * then added disks for geometry changes, | |
5181 | * and should be added immediately. | |
5182 | */ | |
5183 | super_types[mddev->major_version]. | |
5184 | validate_super(mddev, rdev); | |
5185 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5186 | if (err) | |
5187 | unbind_rdev_from_array(rdev); | |
5188 | } | |
1da177e4 LT |
5189 | if (err) |
5190 | export_rdev(rdev); | |
52664732 | 5191 | else |
00bcb4ac | 5192 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c361777f | 5193 | |
17571284 | 5194 | md_update_sb(mddev, 1); |
72a23c21 NB |
5195 | if (mddev->degraded) |
5196 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5197 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
005eca5e | 5198 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5199 | return err; |
5200 | } | |
5201 | ||
5202 | /* otherwise, add_new_disk is only allowed | |
5203 | * for major_version==0 superblocks | |
5204 | */ | |
5205 | if (mddev->major_version != 0) { | |
5206 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5207 | mdname(mddev)); | |
5208 | return -EINVAL; | |
5209 | } | |
5210 | ||
5211 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5212 | int err; | |
d710e138 | 5213 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5214 | if (IS_ERR(rdev)) { |
5215 | printk(KERN_WARNING | |
5216 | "md: error, md_import_device() returned %ld\n", | |
5217 | PTR_ERR(rdev)); | |
5218 | return PTR_ERR(rdev); | |
5219 | } | |
5220 | rdev->desc_nr = info->number; | |
5221 | if (info->raid_disk < mddev->raid_disks) | |
5222 | rdev->raid_disk = info->raid_disk; | |
5223 | else | |
5224 | rdev->raid_disk = -1; | |
5225 | ||
1da177e4 | 5226 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5227 | if (info->state & (1<<MD_DISK_SYNC)) |
5228 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5229 | |
8ddf9efe N |
5230 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5231 | set_bit(WriteMostly, &rdev->flags); | |
5232 | ||
1da177e4 LT |
5233 | if (!mddev->persistent) { |
5234 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
0f420358 | 5235 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 5236 | } else |
0f420358 | 5237 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
8190e754 | 5238 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5239 | |
2bf071bf N |
5240 | err = bind_rdev_to_array(rdev, mddev); |
5241 | if (err) { | |
5242 | export_rdev(rdev); | |
5243 | return err; | |
5244 | } | |
1da177e4 LT |
5245 | } |
5246 | ||
5247 | return 0; | |
5248 | } | |
5249 | ||
5250 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
5251 | { | |
5252 | char b[BDEVNAME_SIZE]; | |
5253 | mdk_rdev_t *rdev; | |
5254 | ||
1da177e4 LT |
5255 | rdev = find_rdev(mddev, dev); |
5256 | if (!rdev) | |
5257 | return -ENXIO; | |
5258 | ||
5259 | if (rdev->raid_disk >= 0) | |
5260 | goto busy; | |
5261 | ||
5262 | kick_rdev_from_array(rdev); | |
850b2b42 | 5263 | md_update_sb(mddev, 1); |
d7603b7e | 5264 | md_new_event(mddev); |
1da177e4 LT |
5265 | |
5266 | return 0; | |
5267 | busy: | |
fdefa4d8 | 5268 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5269 | bdevname(rdev->bdev,b), mdname(mddev)); |
5270 | return -EBUSY; | |
5271 | } | |
5272 | ||
5273 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
5274 | { | |
5275 | char b[BDEVNAME_SIZE]; | |
5276 | int err; | |
1da177e4 LT |
5277 | mdk_rdev_t *rdev; |
5278 | ||
5279 | if (!mddev->pers) | |
5280 | return -ENODEV; | |
5281 | ||
5282 | if (mddev->major_version != 0) { | |
5283 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5284 | " version-0 superblocks.\n", | |
5285 | mdname(mddev)); | |
5286 | return -EINVAL; | |
5287 | } | |
5288 | if (!mddev->pers->hot_add_disk) { | |
5289 | printk(KERN_WARNING | |
5290 | "%s: personality does not support diskops!\n", | |
5291 | mdname(mddev)); | |
5292 | return -EINVAL; | |
5293 | } | |
5294 | ||
d710e138 | 5295 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5296 | if (IS_ERR(rdev)) { |
5297 | printk(KERN_WARNING | |
5298 | "md: error, md_import_device() returned %ld\n", | |
5299 | PTR_ERR(rdev)); | |
5300 | return -EINVAL; | |
5301 | } | |
5302 | ||
5303 | if (mddev->persistent) | |
0f420358 | 5304 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 5305 | else |
0f420358 | 5306 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 5307 | |
8190e754 | 5308 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5309 | |
b2d444d7 | 5310 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5311 | printk(KERN_WARNING |
5312 | "md: can not hot-add faulty %s disk to %s!\n", | |
5313 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5314 | err = -EINVAL; | |
5315 | goto abort_export; | |
5316 | } | |
b2d444d7 | 5317 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5318 | rdev->desc_nr = -1; |
5842730d | 5319 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5320 | err = bind_rdev_to_array(rdev, mddev); |
5321 | if (err) | |
5322 | goto abort_export; | |
1da177e4 LT |
5323 | |
5324 | /* | |
5325 | * The rest should better be atomic, we can have disk failures | |
5326 | * noticed in interrupt contexts ... | |
5327 | */ | |
5328 | ||
1da177e4 LT |
5329 | rdev->raid_disk = -1; |
5330 | ||
850b2b42 | 5331 | md_update_sb(mddev, 1); |
1da177e4 LT |
5332 | |
5333 | /* | |
5334 | * Kick recovery, maybe this spare has to be added to the | |
5335 | * array immediately. | |
5336 | */ | |
5337 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5338 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5339 | md_new_event(mddev); |
1da177e4 LT |
5340 | return 0; |
5341 | ||
1da177e4 LT |
5342 | abort_export: |
5343 | export_rdev(rdev); | |
5344 | return err; | |
5345 | } | |
5346 | ||
32a7627c N |
5347 | static int set_bitmap_file(mddev_t *mddev, int fd) |
5348 | { | |
5349 | int err; | |
5350 | ||
36fa3063 N |
5351 | if (mddev->pers) { |
5352 | if (!mddev->pers->quiesce) | |
5353 | return -EBUSY; | |
5354 | if (mddev->recovery || mddev->sync_thread) | |
5355 | return -EBUSY; | |
5356 | /* we should be able to change the bitmap.. */ | |
5357 | } | |
32a7627c | 5358 | |
32a7627c | 5359 | |
36fa3063 N |
5360 | if (fd >= 0) { |
5361 | if (mddev->bitmap) | |
5362 | return -EEXIST; /* cannot add when bitmap is present */ | |
c3d9714e | 5363 | mddev->bitmap_info.file = fget(fd); |
32a7627c | 5364 | |
c3d9714e | 5365 | if (mddev->bitmap_info.file == NULL) { |
36fa3063 N |
5366 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
5367 | mdname(mddev)); | |
5368 | return -EBADF; | |
5369 | } | |
5370 | ||
c3d9714e | 5371 | err = deny_bitmap_write_access(mddev->bitmap_info.file); |
36fa3063 N |
5372 | if (err) { |
5373 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
5374 | mdname(mddev)); | |
c3d9714e N |
5375 | fput(mddev->bitmap_info.file); |
5376 | mddev->bitmap_info.file = NULL; | |
36fa3063 N |
5377 | return err; |
5378 | } | |
c3d9714e | 5379 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
5380 | } else if (mddev->bitmap == NULL) |
5381 | return -ENOENT; /* cannot remove what isn't there */ | |
5382 | err = 0; | |
5383 | if (mddev->pers) { | |
5384 | mddev->pers->quiesce(mddev, 1); | |
5385 | if (fd >= 0) | |
5386 | err = bitmap_create(mddev); | |
d7375ab3 | 5387 | if (fd < 0 || err) { |
36fa3063 | 5388 | bitmap_destroy(mddev); |
d7375ab3 N |
5389 | fd = -1; /* make sure to put the file */ |
5390 | } | |
36fa3063 | 5391 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
5392 | } |
5393 | if (fd < 0) { | |
c3d9714e N |
5394 | if (mddev->bitmap_info.file) { |
5395 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
5396 | fput(mddev->bitmap_info.file); | |
acc55e22 | 5397 | } |
c3d9714e | 5398 | mddev->bitmap_info.file = NULL; |
36fa3063 N |
5399 | } |
5400 | ||
32a7627c N |
5401 | return err; |
5402 | } | |
5403 | ||
1da177e4 LT |
5404 | /* |
5405 | * set_array_info is used two different ways | |
5406 | * The original usage is when creating a new array. | |
5407 | * In this usage, raid_disks is > 0 and it together with | |
5408 | * level, size, not_persistent,layout,chunksize determine the | |
5409 | * shape of the array. | |
5410 | * This will always create an array with a type-0.90.0 superblock. | |
5411 | * The newer usage is when assembling an array. | |
5412 | * In this case raid_disks will be 0, and the major_version field is | |
5413 | * use to determine which style super-blocks are to be found on the devices. | |
5414 | * The minor and patch _version numbers are also kept incase the | |
5415 | * super_block handler wishes to interpret them. | |
5416 | */ | |
5417 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
5418 | { | |
5419 | ||
5420 | if (info->raid_disks == 0) { | |
5421 | /* just setting version number for superblock loading */ | |
5422 | if (info->major_version < 0 || | |
50511da3 | 5423 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
5424 | super_types[info->major_version].name == NULL) { |
5425 | /* maybe try to auto-load a module? */ | |
5426 | printk(KERN_INFO | |
5427 | "md: superblock version %d not known\n", | |
5428 | info->major_version); | |
5429 | return -EINVAL; | |
5430 | } | |
5431 | mddev->major_version = info->major_version; | |
5432 | mddev->minor_version = info->minor_version; | |
5433 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 5434 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
5435 | /* ensure mddev_put doesn't delete this now that there |
5436 | * is some minimal configuration. | |
5437 | */ | |
5438 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
5439 | return 0; |
5440 | } | |
5441 | mddev->major_version = MD_MAJOR_VERSION; | |
5442 | mddev->minor_version = MD_MINOR_VERSION; | |
5443 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
5444 | mddev->ctime = get_seconds(); | |
5445 | ||
5446 | mddev->level = info->level; | |
17115e03 | 5447 | mddev->clevel[0] = 0; |
58c0fed4 | 5448 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
5449 | mddev->raid_disks = info->raid_disks; |
5450 | /* don't set md_minor, it is determined by which /dev/md* was | |
5451 | * openned | |
5452 | */ | |
5453 | if (info->state & (1<<MD_SB_CLEAN)) | |
5454 | mddev->recovery_cp = MaxSector; | |
5455 | else | |
5456 | mddev->recovery_cp = 0; | |
5457 | mddev->persistent = ! info->not_persistent; | |
e691063a | 5458 | mddev->external = 0; |
1da177e4 LT |
5459 | |
5460 | mddev->layout = info->layout; | |
9d8f0363 | 5461 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
5462 | |
5463 | mddev->max_disks = MD_SB_DISKS; | |
5464 | ||
e691063a N |
5465 | if (mddev->persistent) |
5466 | mddev->flags = 0; | |
850b2b42 | 5467 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5468 | |
c3d9714e N |
5469 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
5470 | mddev->bitmap_info.offset = 0; | |
b2a2703c | 5471 | |
f6705578 N |
5472 | mddev->reshape_position = MaxSector; |
5473 | ||
1da177e4 LT |
5474 | /* |
5475 | * Generate a 128 bit UUID | |
5476 | */ | |
5477 | get_random_bytes(mddev->uuid, 16); | |
5478 | ||
f6705578 | 5479 | mddev->new_level = mddev->level; |
664e7c41 | 5480 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
5481 | mddev->new_layout = mddev->layout; |
5482 | mddev->delta_disks = 0; | |
5483 | ||
1da177e4 LT |
5484 | return 0; |
5485 | } | |
5486 | ||
1f403624 DW |
5487 | void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) |
5488 | { | |
b522adcd DW |
5489 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
5490 | ||
5491 | if (mddev->external_size) | |
5492 | return; | |
5493 | ||
1f403624 DW |
5494 | mddev->array_sectors = array_sectors; |
5495 | } | |
5496 | EXPORT_SYMBOL(md_set_array_sectors); | |
5497 | ||
d71f9f88 | 5498 | static int update_size(mddev_t *mddev, sector_t num_sectors) |
a35b0d69 | 5499 | { |
159ec1fc | 5500 | mdk_rdev_t *rdev; |
a35b0d69 | 5501 | int rv; |
d71f9f88 | 5502 | int fit = (num_sectors == 0); |
a35b0d69 N |
5503 | |
5504 | if (mddev->pers->resize == NULL) | |
5505 | return -EINVAL; | |
d71f9f88 AN |
5506 | /* The "num_sectors" is the number of sectors of each device that |
5507 | * is used. This can only make sense for arrays with redundancy. | |
5508 | * linear and raid0 always use whatever space is available. We can only | |
5509 | * consider changing this number if no resync or reconstruction is | |
5510 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 5511 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
5512 | * of each device. If num_sectors is zero, we find the largest size |
5513 | * that fits. | |
5514 | ||
a35b0d69 N |
5515 | */ |
5516 | if (mddev->sync_thread) | |
5517 | return -EBUSY; | |
dba034ee N |
5518 | if (mddev->bitmap) |
5519 | /* Sorry, cannot grow a bitmap yet, just remove it, | |
5520 | * grow, and re-add. | |
5521 | */ | |
5522 | return -EBUSY; | |
159ec1fc | 5523 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dd8ac336 | 5524 | sector_t avail = rdev->sectors; |
01ab5662 | 5525 | |
d71f9f88 AN |
5526 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
5527 | num_sectors = avail; | |
5528 | if (avail < num_sectors) | |
a35b0d69 N |
5529 | return -ENOSPC; |
5530 | } | |
d71f9f88 | 5531 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
5532 | if (!rv) |
5533 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
5534 | return rv; |
5535 | } | |
5536 | ||
da943b99 N |
5537 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
5538 | { | |
5539 | int rv; | |
5540 | /* change the number of raid disks */ | |
63c70c4f | 5541 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
5542 | return -EINVAL; |
5543 | if (raid_disks <= 0 || | |
233fca36 | 5544 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 5545 | return -EINVAL; |
63c70c4f | 5546 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 5547 | return -EBUSY; |
63c70c4f N |
5548 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
5549 | ||
5550 | rv = mddev->pers->check_reshape(mddev); | |
da943b99 N |
5551 | return rv; |
5552 | } | |
5553 | ||
5554 | ||
1da177e4 LT |
5555 | /* |
5556 | * update_array_info is used to change the configuration of an | |
5557 | * on-line array. | |
5558 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
5559 | * fields in the info are checked against the array. | |
5560 | * Any differences that cannot be handled will cause an error. | |
5561 | * Normally, only one change can be managed at a time. | |
5562 | */ | |
5563 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
5564 | { | |
5565 | int rv = 0; | |
5566 | int cnt = 0; | |
36fa3063 N |
5567 | int state = 0; |
5568 | ||
5569 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 5570 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5571 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
5572 | |
5573 | if (mddev->major_version != info->major_version || | |
5574 | mddev->minor_version != info->minor_version || | |
5575 | /* mddev->patch_version != info->patch_version || */ | |
5576 | mddev->ctime != info->ctime || | |
5577 | mddev->level != info->level || | |
5578 | /* mddev->layout != info->layout || */ | |
5579 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 5580 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
5581 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
5582 | ((state^info->state) & 0xfffffe00) | |
5583 | ) | |
1da177e4 LT |
5584 | return -EINVAL; |
5585 | /* Check there is only one change */ | |
58c0fed4 AN |
5586 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
5587 | cnt++; | |
5588 | if (mddev->raid_disks != info->raid_disks) | |
5589 | cnt++; | |
5590 | if (mddev->layout != info->layout) | |
5591 | cnt++; | |
5592 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
5593 | cnt++; | |
5594 | if (cnt == 0) | |
5595 | return 0; | |
5596 | if (cnt > 1) | |
5597 | return -EINVAL; | |
1da177e4 LT |
5598 | |
5599 | if (mddev->layout != info->layout) { | |
5600 | /* Change layout | |
5601 | * we don't need to do anything at the md level, the | |
5602 | * personality will take care of it all. | |
5603 | */ | |
50ac168a | 5604 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 5605 | return -EINVAL; |
597a711b N |
5606 | else { |
5607 | mddev->new_layout = info->layout; | |
50ac168a | 5608 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
5609 | if (rv) |
5610 | mddev->new_layout = mddev->layout; | |
5611 | return rv; | |
5612 | } | |
1da177e4 | 5613 | } |
58c0fed4 | 5614 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 5615 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 5616 | |
da943b99 N |
5617 | if (mddev->raid_disks != info->raid_disks) |
5618 | rv = update_raid_disks(mddev, info->raid_disks); | |
5619 | ||
36fa3063 N |
5620 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
5621 | if (mddev->pers->quiesce == NULL) | |
5622 | return -EINVAL; | |
5623 | if (mddev->recovery || mddev->sync_thread) | |
5624 | return -EBUSY; | |
5625 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
5626 | /* add the bitmap */ | |
5627 | if (mddev->bitmap) | |
5628 | return -EEXIST; | |
c3d9714e | 5629 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 5630 | return -EINVAL; |
c3d9714e N |
5631 | mddev->bitmap_info.offset = |
5632 | mddev->bitmap_info.default_offset; | |
36fa3063 N |
5633 | mddev->pers->quiesce(mddev, 1); |
5634 | rv = bitmap_create(mddev); | |
5635 | if (rv) | |
5636 | bitmap_destroy(mddev); | |
5637 | mddev->pers->quiesce(mddev, 0); | |
5638 | } else { | |
5639 | /* remove the bitmap */ | |
5640 | if (!mddev->bitmap) | |
5641 | return -ENOENT; | |
5642 | if (mddev->bitmap->file) | |
5643 | return -EINVAL; | |
5644 | mddev->pers->quiesce(mddev, 1); | |
5645 | bitmap_destroy(mddev); | |
5646 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 5647 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
5648 | } |
5649 | } | |
850b2b42 | 5650 | md_update_sb(mddev, 1); |
1da177e4 LT |
5651 | return rv; |
5652 | } | |
5653 | ||
5654 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
5655 | { | |
5656 | mdk_rdev_t *rdev; | |
5657 | ||
5658 | if (mddev->pers == NULL) | |
5659 | return -ENODEV; | |
5660 | ||
5661 | rdev = find_rdev(mddev, dev); | |
5662 | if (!rdev) | |
5663 | return -ENODEV; | |
5664 | ||
5665 | md_error(mddev, rdev); | |
5666 | return 0; | |
5667 | } | |
5668 | ||
2f9618ce AN |
5669 | /* |
5670 | * We have a problem here : there is no easy way to give a CHS | |
5671 | * virtual geometry. We currently pretend that we have a 2 heads | |
5672 | * 4 sectors (with a BIG number of cylinders...). This drives | |
5673 | * dosfs just mad... ;-) | |
5674 | */ | |
a885c8c4 CH |
5675 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
5676 | { | |
5677 | mddev_t *mddev = bdev->bd_disk->private_data; | |
5678 | ||
5679 | geo->heads = 2; | |
5680 | geo->sectors = 4; | |
49ce6cea | 5681 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
5682 | return 0; |
5683 | } | |
5684 | ||
a39907fa | 5685 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
5686 | unsigned int cmd, unsigned long arg) |
5687 | { | |
5688 | int err = 0; | |
5689 | void __user *argp = (void __user *)arg; | |
1da177e4 | 5690 | mddev_t *mddev = NULL; |
e2218350 | 5691 | int ro; |
1da177e4 LT |
5692 | |
5693 | if (!capable(CAP_SYS_ADMIN)) | |
5694 | return -EACCES; | |
5695 | ||
5696 | /* | |
5697 | * Commands dealing with the RAID driver but not any | |
5698 | * particular array: | |
5699 | */ | |
5700 | switch (cmd) | |
5701 | { | |
5702 | case RAID_VERSION: | |
5703 | err = get_version(argp); | |
5704 | goto done; | |
5705 | ||
5706 | case PRINT_RAID_DEBUG: | |
5707 | err = 0; | |
5708 | md_print_devices(); | |
5709 | goto done; | |
5710 | ||
5711 | #ifndef MODULE | |
5712 | case RAID_AUTORUN: | |
5713 | err = 0; | |
5714 | autostart_arrays(arg); | |
5715 | goto done; | |
5716 | #endif | |
5717 | default:; | |
5718 | } | |
5719 | ||
5720 | /* | |
5721 | * Commands creating/starting a new array: | |
5722 | */ | |
5723 | ||
a39907fa | 5724 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
5725 | |
5726 | if (!mddev) { | |
5727 | BUG(); | |
5728 | goto abort; | |
5729 | } | |
5730 | ||
1da177e4 LT |
5731 | err = mddev_lock(mddev); |
5732 | if (err) { | |
5733 | printk(KERN_INFO | |
5734 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
5735 | err, cmd); | |
5736 | goto abort; | |
5737 | } | |
5738 | ||
5739 | switch (cmd) | |
5740 | { | |
5741 | case SET_ARRAY_INFO: | |
5742 | { | |
5743 | mdu_array_info_t info; | |
5744 | if (!arg) | |
5745 | memset(&info, 0, sizeof(info)); | |
5746 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
5747 | err = -EFAULT; | |
5748 | goto abort_unlock; | |
5749 | } | |
5750 | if (mddev->pers) { | |
5751 | err = update_array_info(mddev, &info); | |
5752 | if (err) { | |
5753 | printk(KERN_WARNING "md: couldn't update" | |
5754 | " array info. %d\n", err); | |
5755 | goto abort_unlock; | |
5756 | } | |
5757 | goto done_unlock; | |
5758 | } | |
5759 | if (!list_empty(&mddev->disks)) { | |
5760 | printk(KERN_WARNING | |
5761 | "md: array %s already has disks!\n", | |
5762 | mdname(mddev)); | |
5763 | err = -EBUSY; | |
5764 | goto abort_unlock; | |
5765 | } | |
5766 | if (mddev->raid_disks) { | |
5767 | printk(KERN_WARNING | |
5768 | "md: array %s already initialised!\n", | |
5769 | mdname(mddev)); | |
5770 | err = -EBUSY; | |
5771 | goto abort_unlock; | |
5772 | } | |
5773 | err = set_array_info(mddev, &info); | |
5774 | if (err) { | |
5775 | printk(KERN_WARNING "md: couldn't set" | |
5776 | " array info. %d\n", err); | |
5777 | goto abort_unlock; | |
5778 | } | |
5779 | } | |
5780 | goto done_unlock; | |
5781 | ||
5782 | default:; | |
5783 | } | |
5784 | ||
5785 | /* | |
5786 | * Commands querying/configuring an existing array: | |
5787 | */ | |
32a7627c | 5788 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 5789 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
5790 | if ((!mddev->raid_disks && !mddev->external) |
5791 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
5792 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
5793 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
5794 | err = -ENODEV; |
5795 | goto abort_unlock; | |
5796 | } | |
5797 | ||
5798 | /* | |
5799 | * Commands even a read-only array can execute: | |
5800 | */ | |
5801 | switch (cmd) | |
5802 | { | |
5803 | case GET_ARRAY_INFO: | |
5804 | err = get_array_info(mddev, argp); | |
5805 | goto done_unlock; | |
5806 | ||
32a7627c | 5807 | case GET_BITMAP_FILE: |
87162a28 | 5808 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
5809 | goto done_unlock; |
5810 | ||
1da177e4 LT |
5811 | case GET_DISK_INFO: |
5812 | err = get_disk_info(mddev, argp); | |
5813 | goto done_unlock; | |
5814 | ||
5815 | case RESTART_ARRAY_RW: | |
5816 | err = restart_array(mddev); | |
5817 | goto done_unlock; | |
5818 | ||
5819 | case STOP_ARRAY: | |
d710e138 | 5820 | err = do_md_stop(mddev, 0, 1); |
1da177e4 LT |
5821 | goto done_unlock; |
5822 | ||
5823 | case STOP_ARRAY_RO: | |
a4bd82d0 | 5824 | err = md_set_readonly(mddev, 1); |
1da177e4 LT |
5825 | goto done_unlock; |
5826 | ||
e2218350 DW |
5827 | case BLKROSET: |
5828 | if (get_user(ro, (int __user *)(arg))) { | |
5829 | err = -EFAULT; | |
5830 | goto done_unlock; | |
5831 | } | |
5832 | err = -EINVAL; | |
5833 | ||
5834 | /* if the bdev is going readonly the value of mddev->ro | |
5835 | * does not matter, no writes are coming | |
5836 | */ | |
5837 | if (ro) | |
5838 | goto done_unlock; | |
5839 | ||
5840 | /* are we are already prepared for writes? */ | |
5841 | if (mddev->ro != 1) | |
5842 | goto done_unlock; | |
5843 | ||
5844 | /* transitioning to readauto need only happen for | |
5845 | * arrays that call md_write_start | |
5846 | */ | |
5847 | if (mddev->pers) { | |
5848 | err = restart_array(mddev); | |
5849 | if (err == 0) { | |
5850 | mddev->ro = 2; | |
5851 | set_disk_ro(mddev->gendisk, 0); | |
5852 | } | |
5853 | } | |
5854 | goto done_unlock; | |
1da177e4 LT |
5855 | } |
5856 | ||
5857 | /* | |
5858 | * The remaining ioctls are changing the state of the | |
f91de92e N |
5859 | * superblock, so we do not allow them on read-only arrays. |
5860 | * However non-MD ioctls (e.g. get-size) will still come through | |
5861 | * here and hit the 'default' below, so only disallow | |
5862 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 5863 | */ |
bb57fc64 | 5864 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
5865 | if (mddev->ro == 2) { |
5866 | mddev->ro = 0; | |
00bcb4ac | 5867 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 NB |
5868 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5869 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
5870 | } else { |
5871 | err = -EROFS; | |
5872 | goto abort_unlock; | |
5873 | } | |
1da177e4 LT |
5874 | } |
5875 | ||
5876 | switch (cmd) | |
5877 | { | |
5878 | case ADD_NEW_DISK: | |
5879 | { | |
5880 | mdu_disk_info_t info; | |
5881 | if (copy_from_user(&info, argp, sizeof(info))) | |
5882 | err = -EFAULT; | |
5883 | else | |
5884 | err = add_new_disk(mddev, &info); | |
5885 | goto done_unlock; | |
5886 | } | |
5887 | ||
5888 | case HOT_REMOVE_DISK: | |
5889 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
5890 | goto done_unlock; | |
5891 | ||
5892 | case HOT_ADD_DISK: | |
5893 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
5894 | goto done_unlock; | |
5895 | ||
5896 | case SET_DISK_FAULTY: | |
5897 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5898 | goto done_unlock; | |
5899 | ||
5900 | case RUN_ARRAY: | |
d710e138 | 5901 | err = do_md_run(mddev); |
1da177e4 LT |
5902 | goto done_unlock; |
5903 | ||
32a7627c N |
5904 | case SET_BITMAP_FILE: |
5905 | err = set_bitmap_file(mddev, (int)arg); | |
5906 | goto done_unlock; | |
5907 | ||
1da177e4 | 5908 | default: |
1da177e4 LT |
5909 | err = -EINVAL; |
5910 | goto abort_unlock; | |
5911 | } | |
5912 | ||
5913 | done_unlock: | |
5914 | abort_unlock: | |
d3374825 N |
5915 | if (mddev->hold_active == UNTIL_IOCTL && |
5916 | err != -EINVAL) | |
5917 | mddev->hold_active = 0; | |
1da177e4 LT |
5918 | mddev_unlock(mddev); |
5919 | ||
5920 | return err; | |
5921 | done: | |
5922 | if (err) | |
5923 | MD_BUG(); | |
5924 | abort: | |
5925 | return err; | |
5926 | } | |
aa98aa31 AB |
5927 | #ifdef CONFIG_COMPAT |
5928 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
5929 | unsigned int cmd, unsigned long arg) | |
5930 | { | |
5931 | switch (cmd) { | |
5932 | case HOT_REMOVE_DISK: | |
5933 | case HOT_ADD_DISK: | |
5934 | case SET_DISK_FAULTY: | |
5935 | case SET_BITMAP_FILE: | |
5936 | /* These take in integer arg, do not convert */ | |
5937 | break; | |
5938 | default: | |
5939 | arg = (unsigned long)compat_ptr(arg); | |
5940 | break; | |
5941 | } | |
5942 | ||
5943 | return md_ioctl(bdev, mode, cmd, arg); | |
5944 | } | |
5945 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 5946 | |
a39907fa | 5947 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
5948 | { |
5949 | /* | |
5950 | * Succeed if we can lock the mddev, which confirms that | |
5951 | * it isn't being stopped right now. | |
5952 | */ | |
d3374825 | 5953 | mddev_t *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
5954 | int err; |
5955 | ||
d3374825 N |
5956 | if (mddev->gendisk != bdev->bd_disk) { |
5957 | /* we are racing with mddev_put which is discarding this | |
5958 | * bd_disk. | |
5959 | */ | |
5960 | mddev_put(mddev); | |
5961 | /* Wait until bdev->bd_disk is definitely gone */ | |
5962 | flush_scheduled_work(); | |
5963 | /* Then retry the open from the top */ | |
5964 | return -ERESTARTSYS; | |
5965 | } | |
5966 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
5967 | ||
c8c00a69 | 5968 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
5969 | goto out; |
5970 | ||
5971 | err = 0; | |
f2ea68cf | 5972 | atomic_inc(&mddev->openers); |
c8c00a69 | 5973 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 5974 | |
f3b99be1 | 5975 | check_disk_size_change(mddev->gendisk, bdev); |
1da177e4 LT |
5976 | out: |
5977 | return err; | |
5978 | } | |
5979 | ||
a39907fa | 5980 | static int md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 5981 | { |
a39907fa | 5982 | mddev_t *mddev = disk->private_data; |
1da177e4 | 5983 | |
52e5f9d1 | 5984 | BUG_ON(!mddev); |
f2ea68cf | 5985 | atomic_dec(&mddev->openers); |
1da177e4 LT |
5986 | mddev_put(mddev); |
5987 | ||
5988 | return 0; | |
5989 | } | |
83d5cde4 | 5990 | static const struct block_device_operations md_fops = |
1da177e4 LT |
5991 | { |
5992 | .owner = THIS_MODULE, | |
a39907fa AV |
5993 | .open = md_open, |
5994 | .release = md_release, | |
b492b852 | 5995 | .ioctl = md_ioctl, |
aa98aa31 AB |
5996 | #ifdef CONFIG_COMPAT |
5997 | .compat_ioctl = md_compat_ioctl, | |
5998 | #endif | |
a885c8c4 | 5999 | .getgeo = md_getgeo, |
1da177e4 LT |
6000 | }; |
6001 | ||
75c96f85 | 6002 | static int md_thread(void * arg) |
1da177e4 LT |
6003 | { |
6004 | mdk_thread_t *thread = arg; | |
6005 | ||
1da177e4 LT |
6006 | /* |
6007 | * md_thread is a 'system-thread', it's priority should be very | |
6008 | * high. We avoid resource deadlocks individually in each | |
6009 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6010 | * the very same RT priority as kswapd, thus we will never get | |
6011 | * into a priority inversion deadlock. | |
6012 | * | |
6013 | * we definitely have to have equal or higher priority than | |
6014 | * bdflush, otherwise bdflush will deadlock if there are too | |
6015 | * many dirty RAID5 blocks. | |
6016 | */ | |
1da177e4 | 6017 | |
6985c43f | 6018 | allow_signal(SIGKILL); |
a6fb0934 | 6019 | while (!kthread_should_stop()) { |
1da177e4 | 6020 | |
93588e22 N |
6021 | /* We need to wait INTERRUPTIBLE so that |
6022 | * we don't add to the load-average. | |
6023 | * That means we need to be sure no signals are | |
6024 | * pending | |
6025 | */ | |
6026 | if (signal_pending(current)) | |
6027 | flush_signals(current); | |
6028 | ||
6029 | wait_event_interruptible_timeout | |
6030 | (thread->wqueue, | |
6031 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6032 | || kthread_should_stop(), | |
6033 | thread->timeout); | |
1da177e4 LT |
6034 | |
6035 | clear_bit(THREAD_WAKEUP, &thread->flags); | |
6036 | ||
787453c2 | 6037 | thread->run(thread->mddev); |
1da177e4 | 6038 | } |
a6fb0934 | 6039 | |
1da177e4 LT |
6040 | return 0; |
6041 | } | |
6042 | ||
6043 | void md_wakeup_thread(mdk_thread_t *thread) | |
6044 | { | |
6045 | if (thread) { | |
6046 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
6047 | set_bit(THREAD_WAKEUP, &thread->flags); | |
6048 | wake_up(&thread->wqueue); | |
6049 | } | |
6050 | } | |
6051 | ||
6052 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
6053 | const char *name) | |
6054 | { | |
6055 | mdk_thread_t *thread; | |
1da177e4 | 6056 | |
9ffae0cf | 6057 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
6058 | if (!thread) |
6059 | return NULL; | |
6060 | ||
1da177e4 LT |
6061 | init_waitqueue_head(&thread->wqueue); |
6062 | ||
1da177e4 LT |
6063 | thread->run = run; |
6064 | thread->mddev = mddev; | |
32a7627c | 6065 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
6066 | thread->tsk = kthread_run(md_thread, thread, |
6067 | "%s_%s", | |
6068 | mdname(thread->mddev), | |
6069 | name ?: mddev->pers->name); | |
a6fb0934 | 6070 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
6071 | kfree(thread); |
6072 | return NULL; | |
6073 | } | |
1da177e4 LT |
6074 | return thread; |
6075 | } | |
6076 | ||
1da177e4 LT |
6077 | void md_unregister_thread(mdk_thread_t *thread) |
6078 | { | |
e0cf8f04 N |
6079 | if (!thread) |
6080 | return; | |
ba25f9dc | 6081 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
6082 | |
6083 | kthread_stop(thread->tsk); | |
1da177e4 LT |
6084 | kfree(thread); |
6085 | } | |
6086 | ||
6087 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
6088 | { | |
6089 | if (!mddev) { | |
6090 | MD_BUG(); | |
6091 | return; | |
6092 | } | |
6093 | ||
b2d444d7 | 6094 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6095 | return; |
6bfe0b49 DW |
6096 | |
6097 | if (mddev->external) | |
6098 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 6099 | /* |
1da177e4 LT |
6100 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
6101 | mdname(mddev), | |
6102 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
6103 | __builtin_return_address(0),__builtin_return_address(1), | |
6104 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 6105 | */ |
d0a0a5ee AM |
6106 | if (!mddev->pers) |
6107 | return; | |
1da177e4 LT |
6108 | if (!mddev->pers->error_handler) |
6109 | return; | |
6110 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6111 | if (mddev->degraded) |
6112 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 6113 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
6114 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6115 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6116 | md_wakeup_thread(mddev->thread); | |
768a418d N |
6117 | if (mddev->event_work.func) |
6118 | schedule_work(&mddev->event_work); | |
c331eb04 | 6119 | md_new_event_inintr(mddev); |
1da177e4 LT |
6120 | } |
6121 | ||
6122 | /* seq_file implementation /proc/mdstat */ | |
6123 | ||
6124 | static void status_unused(struct seq_file *seq) | |
6125 | { | |
6126 | int i = 0; | |
6127 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6128 | |
6129 | seq_printf(seq, "unused devices: "); | |
6130 | ||
159ec1fc | 6131 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6132 | char b[BDEVNAME_SIZE]; |
6133 | i++; | |
6134 | seq_printf(seq, "%s ", | |
6135 | bdevname(rdev->bdev,b)); | |
6136 | } | |
6137 | if (!i) | |
6138 | seq_printf(seq, "<none>"); | |
6139 | ||
6140 | seq_printf(seq, "\n"); | |
6141 | } | |
6142 | ||
6143 | ||
6144 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
6145 | { | |
dd71cf6b N |
6146 | sector_t max_sectors, resync, res; |
6147 | unsigned long dt, db; | |
6148 | sector_t rt; | |
4588b42e N |
6149 | int scale; |
6150 | unsigned int per_milli; | |
1da177e4 | 6151 | |
dd71cf6b | 6152 | resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6153 | |
6154 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
dd71cf6b | 6155 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6156 | else |
dd71cf6b | 6157 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
6158 | |
6159 | /* | |
6160 | * Should not happen. | |
6161 | */ | |
dd71cf6b | 6162 | if (!max_sectors) { |
1da177e4 LT |
6163 | MD_BUG(); |
6164 | return; | |
6165 | } | |
4588b42e | 6166 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6167 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6168 | * u32, as those are the requirements for sector_div. |
6169 | * Thus 'scale' must be at least 10 | |
6170 | */ | |
6171 | scale = 10; | |
6172 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6173 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6174 | scale++; |
6175 | } | |
6176 | res = (resync>>scale)*1000; | |
dd71cf6b | 6177 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6178 | |
6179 | per_milli = res; | |
1da177e4 | 6180 | { |
4588b42e | 6181 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6182 | seq_printf(seq, "["); |
6183 | for (i = 0; i < x; i++) | |
6184 | seq_printf(seq, "="); | |
6185 | seq_printf(seq, ">"); | |
6186 | for (i = 0; i < y; i++) | |
6187 | seq_printf(seq, "."); | |
6188 | seq_printf(seq, "] "); | |
6189 | } | |
4588b42e | 6190 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6191 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6192 | "reshape" : | |
61df9d91 N |
6193 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6194 | "check" : | |
6195 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6196 | "resync" : "recovery"))), | |
6197 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6198 | (unsigned long long) resync/2, |
6199 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6200 | |
6201 | /* | |
1da177e4 LT |
6202 | * dt: time from mark until now |
6203 | * db: blocks written from mark until now | |
6204 | * rt: remaining time | |
dd71cf6b N |
6205 | * |
6206 | * rt is a sector_t, so could be 32bit or 64bit. | |
6207 | * So we divide before multiply in case it is 32bit and close | |
6208 | * to the limit. | |
6209 | * We scale the divisor (db) by 32 to avoid loosing precision | |
6210 | * near the end of resync when the number of remaining sectors | |
6211 | * is close to 'db'. | |
6212 | * We then divide rt by 32 after multiplying by db to compensate. | |
6213 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6214 | */ |
6215 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6216 | if (!dt) dt++; | |
ff4e8d9a N |
6217 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6218 | - mddev->resync_mark_cnt; | |
1da177e4 | 6219 | |
dd71cf6b N |
6220 | rt = max_sectors - resync; /* number of remaining sectors */ |
6221 | sector_div(rt, db/32+1); | |
6222 | rt *= dt; | |
6223 | rt >>= 5; | |
6224 | ||
6225 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6226 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6227 | |
ff4e8d9a | 6228 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
6229 | } |
6230 | ||
6231 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
6232 | { | |
6233 | struct list_head *tmp; | |
6234 | loff_t l = *pos; | |
6235 | mddev_t *mddev; | |
6236 | ||
6237 | if (l >= 0x10000) | |
6238 | return NULL; | |
6239 | if (!l--) | |
6240 | /* header */ | |
6241 | return (void*)1; | |
6242 | ||
6243 | spin_lock(&all_mddevs_lock); | |
6244 | list_for_each(tmp,&all_mddevs) | |
6245 | if (!l--) { | |
6246 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
6247 | mddev_get(mddev); | |
6248 | spin_unlock(&all_mddevs_lock); | |
6249 | return mddev; | |
6250 | } | |
6251 | spin_unlock(&all_mddevs_lock); | |
6252 | if (!l--) | |
6253 | return (void*)2;/* tail */ | |
6254 | return NULL; | |
6255 | } | |
6256 | ||
6257 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
6258 | { | |
6259 | struct list_head *tmp; | |
6260 | mddev_t *next_mddev, *mddev = v; | |
6261 | ||
6262 | ++*pos; | |
6263 | if (v == (void*)2) | |
6264 | return NULL; | |
6265 | ||
6266 | spin_lock(&all_mddevs_lock); | |
6267 | if (v == (void*)1) | |
6268 | tmp = all_mddevs.next; | |
6269 | else | |
6270 | tmp = mddev->all_mddevs.next; | |
6271 | if (tmp != &all_mddevs) | |
6272 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
6273 | else { | |
6274 | next_mddev = (void*)2; | |
6275 | *pos = 0x10000; | |
6276 | } | |
6277 | spin_unlock(&all_mddevs_lock); | |
6278 | ||
6279 | if (v != (void*)1) | |
6280 | mddev_put(mddev); | |
6281 | return next_mddev; | |
6282 | ||
6283 | } | |
6284 | ||
6285 | static void md_seq_stop(struct seq_file *seq, void *v) | |
6286 | { | |
6287 | mddev_t *mddev = v; | |
6288 | ||
6289 | if (mddev && v != (void*)1 && v != (void*)2) | |
6290 | mddev_put(mddev); | |
6291 | } | |
6292 | ||
d7603b7e N |
6293 | struct mdstat_info { |
6294 | int event; | |
6295 | }; | |
6296 | ||
1da177e4 LT |
6297 | static int md_seq_show(struct seq_file *seq, void *v) |
6298 | { | |
6299 | mddev_t *mddev = v; | |
dd8ac336 | 6300 | sector_t sectors; |
1da177e4 | 6301 | mdk_rdev_t *rdev; |
d7603b7e | 6302 | struct mdstat_info *mi = seq->private; |
32a7627c | 6303 | struct bitmap *bitmap; |
1da177e4 LT |
6304 | |
6305 | if (v == (void*)1) { | |
2604b703 | 6306 | struct mdk_personality *pers; |
1da177e4 LT |
6307 | seq_printf(seq, "Personalities : "); |
6308 | spin_lock(&pers_lock); | |
2604b703 N |
6309 | list_for_each_entry(pers, &pers_list, list) |
6310 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
6311 | |
6312 | spin_unlock(&pers_lock); | |
6313 | seq_printf(seq, "\n"); | |
d7603b7e | 6314 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
6315 | return 0; |
6316 | } | |
6317 | if (v == (void*)2) { | |
6318 | status_unused(seq); | |
6319 | return 0; | |
6320 | } | |
6321 | ||
5dc5cf7d | 6322 | if (mddev_lock(mddev) < 0) |
1da177e4 | 6323 | return -EINTR; |
5dc5cf7d | 6324 | |
1da177e4 LT |
6325 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
6326 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
6327 | mddev->pers ? "" : "in"); | |
6328 | if (mddev->pers) { | |
f91de92e | 6329 | if (mddev->ro==1) |
1da177e4 | 6330 | seq_printf(seq, " (read-only)"); |
f91de92e | 6331 | if (mddev->ro==2) |
52720ae7 | 6332 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
6333 | seq_printf(seq, " %s", mddev->pers->name); |
6334 | } | |
6335 | ||
dd8ac336 | 6336 | sectors = 0; |
159ec1fc | 6337 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
6338 | char b[BDEVNAME_SIZE]; |
6339 | seq_printf(seq, " %s[%d]", | |
6340 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
6341 | if (test_bit(WriteMostly, &rdev->flags)) |
6342 | seq_printf(seq, "(W)"); | |
b2d444d7 | 6343 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6344 | seq_printf(seq, "(F)"); |
6345 | continue; | |
b325a32e N |
6346 | } else if (rdev->raid_disk < 0) |
6347 | seq_printf(seq, "(S)"); /* spare */ | |
dd8ac336 | 6348 | sectors += rdev->sectors; |
1da177e4 LT |
6349 | } |
6350 | ||
6351 | if (!list_empty(&mddev->disks)) { | |
6352 | if (mddev->pers) | |
6353 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
6354 | (unsigned long long) |
6355 | mddev->array_sectors / 2); | |
1da177e4 LT |
6356 | else |
6357 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 6358 | (unsigned long long)sectors / 2); |
1da177e4 | 6359 | } |
1cd6bf19 N |
6360 | if (mddev->persistent) { |
6361 | if (mddev->major_version != 0 || | |
6362 | mddev->minor_version != 90) { | |
6363 | seq_printf(seq," super %d.%d", | |
6364 | mddev->major_version, | |
6365 | mddev->minor_version); | |
6366 | } | |
e691063a N |
6367 | } else if (mddev->external) |
6368 | seq_printf(seq, " super external:%s", | |
6369 | mddev->metadata_type); | |
6370 | else | |
1cd6bf19 | 6371 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
6372 | |
6373 | if (mddev->pers) { | |
d710e138 | 6374 | mddev->pers->status(seq, mddev); |
1da177e4 | 6375 | seq_printf(seq, "\n "); |
8e1b39d6 N |
6376 | if (mddev->pers->sync_request) { |
6377 | if (mddev->curr_resync > 2) { | |
d710e138 | 6378 | status_resync(seq, mddev); |
8e1b39d6 N |
6379 | seq_printf(seq, "\n "); |
6380 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
6381 | seq_printf(seq, "\tresync=DELAYED\n "); | |
6382 | else if (mddev->recovery_cp < MaxSector) | |
6383 | seq_printf(seq, "\tresync=PENDING\n "); | |
6384 | } | |
32a7627c N |
6385 | } else |
6386 | seq_printf(seq, "\n "); | |
6387 | ||
6388 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
6389 | unsigned long chunk_kb; |
6390 | unsigned long flags; | |
32a7627c | 6391 | spin_lock_irqsave(&bitmap->lock, flags); |
42a04b50 | 6392 | chunk_kb = mddev->bitmap_info.chunksize >> 10; |
32a7627c N |
6393 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " |
6394 | "%lu%s chunk", | |
6395 | bitmap->pages - bitmap->missing_pages, | |
6396 | bitmap->pages, | |
6397 | (bitmap->pages - bitmap->missing_pages) | |
6398 | << (PAGE_SHIFT - 10), | |
42a04b50 | 6399 | chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize, |
32a7627c | 6400 | chunk_kb ? "KB" : "B"); |
78d742d8 N |
6401 | if (bitmap->file) { |
6402 | seq_printf(seq, ", file: "); | |
c32c2f63 | 6403 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 6404 | } |
78d742d8 | 6405 | |
32a7627c N |
6406 | seq_printf(seq, "\n"); |
6407 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
6408 | } |
6409 | ||
6410 | seq_printf(seq, "\n"); | |
6411 | } | |
6412 | mddev_unlock(mddev); | |
6413 | ||
6414 | return 0; | |
6415 | } | |
6416 | ||
110518bc | 6417 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
6418 | .start = md_seq_start, |
6419 | .next = md_seq_next, | |
6420 | .stop = md_seq_stop, | |
6421 | .show = md_seq_show, | |
6422 | }; | |
6423 | ||
6424 | static int md_seq_open(struct inode *inode, struct file *file) | |
6425 | { | |
6426 | int error; | |
d7603b7e N |
6427 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
6428 | if (mi == NULL) | |
6429 | return -ENOMEM; | |
1da177e4 LT |
6430 | |
6431 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
6432 | if (error) |
6433 | kfree(mi); | |
6434 | else { | |
6435 | struct seq_file *p = file->private_data; | |
6436 | p->private = mi; | |
6437 | mi->event = atomic_read(&md_event_count); | |
6438 | } | |
1da177e4 LT |
6439 | return error; |
6440 | } | |
6441 | ||
d7603b7e N |
6442 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
6443 | { | |
6444 | struct seq_file *m = filp->private_data; | |
6445 | struct mdstat_info *mi = m->private; | |
6446 | int mask; | |
6447 | ||
6448 | poll_wait(filp, &md_event_waiters, wait); | |
6449 | ||
6450 | /* always allow read */ | |
6451 | mask = POLLIN | POLLRDNORM; | |
6452 | ||
6453 | if (mi->event != atomic_read(&md_event_count)) | |
6454 | mask |= POLLERR | POLLPRI; | |
6455 | return mask; | |
6456 | } | |
6457 | ||
fa027c2a | 6458 | static const struct file_operations md_seq_fops = { |
e24650c2 | 6459 | .owner = THIS_MODULE, |
1da177e4 LT |
6460 | .open = md_seq_open, |
6461 | .read = seq_read, | |
6462 | .llseek = seq_lseek, | |
c3f94b40 | 6463 | .release = seq_release_private, |
d7603b7e | 6464 | .poll = mdstat_poll, |
1da177e4 LT |
6465 | }; |
6466 | ||
2604b703 | 6467 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 6468 | { |
1da177e4 | 6469 | spin_lock(&pers_lock); |
2604b703 N |
6470 | list_add_tail(&p->list, &pers_list); |
6471 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
6472 | spin_unlock(&pers_lock); |
6473 | return 0; | |
6474 | } | |
6475 | ||
2604b703 | 6476 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 6477 | { |
2604b703 | 6478 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 6479 | spin_lock(&pers_lock); |
2604b703 | 6480 | list_del_init(&p->list); |
1da177e4 LT |
6481 | spin_unlock(&pers_lock); |
6482 | return 0; | |
6483 | } | |
6484 | ||
eea1bf38 | 6485 | static int is_mddev_idle(mddev_t *mddev, int init) |
1da177e4 LT |
6486 | { |
6487 | mdk_rdev_t * rdev; | |
1da177e4 | 6488 | int idle; |
eea1bf38 | 6489 | int curr_events; |
1da177e4 LT |
6490 | |
6491 | idle = 1; | |
4b80991c N |
6492 | rcu_read_lock(); |
6493 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6494 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
6495 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
6496 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
6497 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
6498 | /* sync IO will cause sync_io to increase before the disk_stats |
6499 | * as sync_io is counted when a request starts, and | |
6500 | * disk_stats is counted when it completes. | |
6501 | * So resync activity will cause curr_events to be smaller than | |
6502 | * when there was no such activity. | |
6503 | * non-sync IO will cause disk_stat to increase without | |
6504 | * increasing sync_io so curr_events will (eventually) | |
6505 | * be larger than it was before. Once it becomes | |
6506 | * substantially larger, the test below will cause | |
6507 | * the array to appear non-idle, and resync will slow | |
6508 | * down. | |
6509 | * If there is a lot of outstanding resync activity when | |
6510 | * we set last_event to curr_events, then all that activity | |
6511 | * completing might cause the array to appear non-idle | |
6512 | * and resync will be slowed down even though there might | |
6513 | * not have been non-resync activity. This will only | |
6514 | * happen once though. 'last_events' will soon reflect | |
6515 | * the state where there is little or no outstanding | |
6516 | * resync requests, and further resync activity will | |
6517 | * always make curr_events less than last_events. | |
c0e48521 | 6518 | * |
1da177e4 | 6519 | */ |
eea1bf38 | 6520 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
6521 | rdev->last_events = curr_events; |
6522 | idle = 0; | |
6523 | } | |
6524 | } | |
4b80991c | 6525 | rcu_read_unlock(); |
1da177e4 LT |
6526 | return idle; |
6527 | } | |
6528 | ||
6529 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
6530 | { | |
6531 | /* another "blocks" (512byte) blocks have been synced */ | |
6532 | atomic_sub(blocks, &mddev->recovery_active); | |
6533 | wake_up(&mddev->recovery_wait); | |
6534 | if (!ok) { | |
dfc70645 | 6535 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6536 | md_wakeup_thread(mddev->thread); |
6537 | // stop recovery, signal do_sync .... | |
6538 | } | |
6539 | } | |
6540 | ||
6541 | ||
06d91a5f N |
6542 | /* md_write_start(mddev, bi) |
6543 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
6544 | * in superblock) before writing, schedule a superblock update |
6545 | * and wait for it to complete. | |
06d91a5f | 6546 | */ |
3d310eb7 | 6547 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 6548 | { |
0fd62b86 | 6549 | int did_change = 0; |
06d91a5f | 6550 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 6551 | return; |
06d91a5f | 6552 | |
f91de92e N |
6553 | BUG_ON(mddev->ro == 1); |
6554 | if (mddev->ro == 2) { | |
6555 | /* need to switch to read/write */ | |
6556 | mddev->ro = 0; | |
6557 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6558 | md_wakeup_thread(mddev->thread); | |
25156198 | 6559 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 6560 | did_change = 1; |
f91de92e | 6561 | } |
06d91a5f | 6562 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
6563 | if (mddev->safemode == 1) |
6564 | mddev->safemode = 0; | |
06d91a5f | 6565 | if (mddev->in_sync) { |
a9701a30 | 6566 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
6567 | if (mddev->in_sync) { |
6568 | mddev->in_sync = 0; | |
850b2b42 | 6569 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3d310eb7 | 6570 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 6571 | did_change = 1; |
3d310eb7 | 6572 | } |
a9701a30 | 6573 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 6574 | } |
0fd62b86 | 6575 | if (did_change) |
00bcb4ac | 6576 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 N |
6577 | wait_event(mddev->sb_wait, |
6578 | !test_bit(MD_CHANGE_CLEAN, &mddev->flags) && | |
6579 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
1da177e4 LT |
6580 | } |
6581 | ||
6582 | void md_write_end(mddev_t *mddev) | |
6583 | { | |
6584 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
6585 | if (mddev->safemode == 2) | |
6586 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 6587 | else if (mddev->safemode_delay) |
1da177e4 LT |
6588 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
6589 | } | |
6590 | } | |
6591 | ||
2a2275d6 N |
6592 | /* md_allow_write(mddev) |
6593 | * Calling this ensures that the array is marked 'active' so that writes | |
6594 | * may proceed without blocking. It is important to call this before | |
6595 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
6596 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
6597 | * |
6598 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
6599 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 6600 | */ |
b5470dc5 | 6601 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
6602 | { |
6603 | if (!mddev->pers) | |
b5470dc5 | 6604 | return 0; |
2a2275d6 | 6605 | if (mddev->ro) |
b5470dc5 | 6606 | return 0; |
1a0fd497 | 6607 | if (!mddev->pers->sync_request) |
b5470dc5 | 6608 | return 0; |
2a2275d6 N |
6609 | |
6610 | spin_lock_irq(&mddev->write_lock); | |
6611 | if (mddev->in_sync) { | |
6612 | mddev->in_sync = 0; | |
6613 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6614 | if (mddev->safemode_delay && | |
6615 | mddev->safemode == 0) | |
6616 | mddev->safemode = 1; | |
6617 | spin_unlock_irq(&mddev->write_lock); | |
6618 | md_update_sb(mddev, 0); | |
00bcb4ac | 6619 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 N |
6620 | } else |
6621 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 DW |
6622 | |
6623 | if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
6624 | return -EAGAIN; | |
6625 | else | |
6626 | return 0; | |
2a2275d6 N |
6627 | } |
6628 | EXPORT_SYMBOL_GPL(md_allow_write); | |
6629 | ||
252ac522 N |
6630 | static void md_unplug(mddev_t *mddev) |
6631 | { | |
6632 | if (mddev->queue) | |
6633 | blk_unplug(mddev->queue); | |
6634 | if (mddev->plug) | |
6635 | mddev->plug->unplug_fn(mddev->plug); | |
6636 | } | |
6637 | ||
1da177e4 LT |
6638 | #define SYNC_MARKS 10 |
6639 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 6640 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
6641 | { |
6642 | mddev_t *mddev2; | |
6643 | unsigned int currspeed = 0, | |
6644 | window; | |
57afd89f | 6645 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
6646 | unsigned long mark[SYNC_MARKS]; |
6647 | sector_t mark_cnt[SYNC_MARKS]; | |
6648 | int last_mark,m; | |
6649 | struct list_head *tmp; | |
6650 | sector_t last_check; | |
57afd89f | 6651 | int skipped = 0; |
5fd6c1dc | 6652 | mdk_rdev_t *rdev; |
61df9d91 | 6653 | char *desc; |
1da177e4 LT |
6654 | |
6655 | /* just incase thread restarts... */ | |
6656 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
6657 | return; | |
5fd6c1dc N |
6658 | if (mddev->ro) /* never try to sync a read-only array */ |
6659 | return; | |
1da177e4 | 6660 | |
61df9d91 N |
6661 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
6662 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
6663 | desc = "data-check"; | |
6664 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6665 | desc = "requested-resync"; | |
6666 | else | |
6667 | desc = "resync"; | |
6668 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
6669 | desc = "reshape"; | |
6670 | else | |
6671 | desc = "recovery"; | |
6672 | ||
1da177e4 LT |
6673 | /* we overload curr_resync somewhat here. |
6674 | * 0 == not engaged in resync at all | |
6675 | * 2 == checking that there is no conflict with another sync | |
6676 | * 1 == like 2, but have yielded to allow conflicting resync to | |
6677 | * commense | |
6678 | * other == active in resync - this many blocks | |
6679 | * | |
6680 | * Before starting a resync we must have set curr_resync to | |
6681 | * 2, and then checked that every "conflicting" array has curr_resync | |
6682 | * less than ours. When we find one that is the same or higher | |
6683 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
6684 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
6685 | * This will mean we have to start checking from the beginning again. | |
6686 | * | |
6687 | */ | |
6688 | ||
6689 | do { | |
6690 | mddev->curr_resync = 2; | |
6691 | ||
6692 | try_again: | |
404e4b43 | 6693 | if (kthread_should_stop()) |
6985c43f | 6694 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
404e4b43 N |
6695 | |
6696 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
1da177e4 | 6697 | goto skip; |
29ac4aa3 | 6698 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
6699 | if (mddev2 == mddev) |
6700 | continue; | |
90b08710 BS |
6701 | if (!mddev->parallel_resync |
6702 | && mddev2->curr_resync | |
6703 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
6704 | DEFINE_WAIT(wq); |
6705 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
6706 | /* arbitrarily yield */ | |
6707 | mddev->curr_resync = 1; | |
6708 | wake_up(&resync_wait); | |
6709 | } | |
6710 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
6711 | /* no need to wait here, we can wait the next | |
6712 | * time 'round when curr_resync == 2 | |
6713 | */ | |
6714 | continue; | |
9744197c N |
6715 | /* We need to wait 'interruptible' so as not to |
6716 | * contribute to the load average, and not to | |
6717 | * be caught by 'softlockup' | |
6718 | */ | |
6719 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
787453c2 | 6720 | if (!kthread_should_stop() && |
8712e553 | 6721 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
6722 | printk(KERN_INFO "md: delaying %s of %s" |
6723 | " until %s has finished (they" | |
1da177e4 | 6724 | " share one or more physical units)\n", |
61df9d91 | 6725 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 6726 | mddev_put(mddev2); |
9744197c N |
6727 | if (signal_pending(current)) |
6728 | flush_signals(current); | |
1da177e4 LT |
6729 | schedule(); |
6730 | finish_wait(&resync_wait, &wq); | |
6731 | goto try_again; | |
6732 | } | |
6733 | finish_wait(&resync_wait, &wq); | |
6734 | } | |
6735 | } | |
6736 | } while (mddev->curr_resync < 2); | |
6737 | ||
5fd6c1dc | 6738 | j = 0; |
9d88883e | 6739 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 6740 | /* resync follows the size requested by the personality, |
57afd89f | 6741 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
6742 | */ |
6743 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 6744 | mddev->resync_mismatches = 0; |
5fd6c1dc | 6745 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
6746 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
6747 | j = mddev->resync_min; | |
6748 | else if (!mddev->bitmap) | |
5fd6c1dc | 6749 | j = mddev->recovery_cp; |
5e96ee65 | 6750 | |
ccfcc3c1 | 6751 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
58c0fed4 | 6752 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6753 | else { |
1da177e4 | 6754 | /* recovery follows the physical size of devices */ |
58c0fed4 | 6755 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6756 | j = MaxSector; |
4e59ca7d DW |
6757 | rcu_read_lock(); |
6758 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc N |
6759 | if (rdev->raid_disk >= 0 && |
6760 | !test_bit(Faulty, &rdev->flags) && | |
6761 | !test_bit(In_sync, &rdev->flags) && | |
6762 | rdev->recovery_offset < j) | |
6763 | j = rdev->recovery_offset; | |
4e59ca7d | 6764 | rcu_read_unlock(); |
5fd6c1dc | 6765 | } |
1da177e4 | 6766 | |
61df9d91 N |
6767 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
6768 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
6769 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 6770 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
6771 | "(but not more than %d KB/sec) for %s.\n", |
6772 | speed_max(mddev), desc); | |
1da177e4 | 6773 | |
eea1bf38 | 6774 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 6775 | |
57afd89f | 6776 | io_sectors = 0; |
1da177e4 LT |
6777 | for (m = 0; m < SYNC_MARKS; m++) { |
6778 | mark[m] = jiffies; | |
57afd89f | 6779 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
6780 | } |
6781 | last_mark = 0; | |
6782 | mddev->resync_mark = mark[last_mark]; | |
6783 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
6784 | ||
6785 | /* | |
6786 | * Tune reconstruction: | |
6787 | */ | |
6788 | window = 32*(PAGE_SIZE/512); | |
6789 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
6790 | window/2,(unsigned long long) max_sectors/2); | |
6791 | ||
6792 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
6793 | last_check = 0; |
6794 | ||
6795 | if (j>2) { | |
6796 | printk(KERN_INFO | |
61df9d91 N |
6797 | "md: resuming %s of %s from checkpoint.\n", |
6798 | desc, mdname(mddev)); | |
1da177e4 LT |
6799 | mddev->curr_resync = j; |
6800 | } | |
efa59339 | 6801 | mddev->curr_resync_completed = mddev->curr_resync; |
1da177e4 LT |
6802 | |
6803 | while (j < max_sectors) { | |
57afd89f | 6804 | sector_t sectors; |
1da177e4 | 6805 | |
57afd89f | 6806 | skipped = 0; |
97e4f42d | 6807 | |
7a91ee1f N |
6808 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6809 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
6810 | (mddev->curr_resync - mddev->curr_resync_completed) | |
6811 | > (max_sectors >> 4)) || | |
6812 | (j - mddev->curr_resync_completed)*2 | |
6813 | >= mddev->resync_max - mddev->curr_resync_completed | |
6814 | )) { | |
97e4f42d | 6815 | /* time to update curr_resync_completed */ |
252ac522 | 6816 | md_unplug(mddev); |
97e4f42d N |
6817 | wait_event(mddev->recovery_wait, |
6818 | atomic_read(&mddev->recovery_active) == 0); | |
6819 | mddev->curr_resync_completed = | |
6820 | mddev->curr_resync; | |
676e42d8 N |
6821 | if (mddev->persistent) |
6822 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
acb180b0 | 6823 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 6824 | } |
acb180b0 | 6825 | |
e62e58a5 N |
6826 | while (j >= mddev->resync_max && !kthread_should_stop()) { |
6827 | /* As this condition is controlled by user-space, | |
6828 | * we can block indefinitely, so use '_interruptible' | |
6829 | * to avoid triggering warnings. | |
6830 | */ | |
6831 | flush_signals(current); /* just in case */ | |
6832 | wait_event_interruptible(mddev->recovery_wait, | |
6833 | mddev->resync_max > j | |
6834 | || kthread_should_stop()); | |
6835 | } | |
acb180b0 N |
6836 | |
6837 | if (kthread_should_stop()) | |
6838 | goto interrupted; | |
6839 | ||
57afd89f | 6840 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 6841 | currspeed < speed_min(mddev)); |
57afd89f | 6842 | if (sectors == 0) { |
dfc70645 | 6843 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6844 | goto out; |
6845 | } | |
57afd89f N |
6846 | |
6847 | if (!skipped) { /* actual IO requested */ | |
6848 | io_sectors += sectors; | |
6849 | atomic_add(sectors, &mddev->recovery_active); | |
6850 | } | |
6851 | ||
1da177e4 LT |
6852 | j += sectors; |
6853 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 6854 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
6855 | if (last_check == 0) |
6856 | /* this is the earliers that rebuilt will be | |
6857 | * visible in /proc/mdstat | |
6858 | */ | |
6859 | md_new_event(mddev); | |
57afd89f N |
6860 | |
6861 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
6862 | continue; |
6863 | ||
57afd89f | 6864 | last_check = io_sectors; |
1da177e4 | 6865 | |
dfc70645 | 6866 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
6867 | break; |
6868 | ||
6869 | repeat: | |
6870 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
6871 | /* step marks */ | |
6872 | int next = (last_mark+1) % SYNC_MARKS; | |
6873 | ||
6874 | mddev->resync_mark = mark[next]; | |
6875 | mddev->resync_mark_cnt = mark_cnt[next]; | |
6876 | mark[next] = jiffies; | |
57afd89f | 6877 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6878 | last_mark = next; |
6879 | } | |
6880 | ||
6881 | ||
c6207277 N |
6882 | if (kthread_should_stop()) |
6883 | goto interrupted; | |
6884 | ||
1da177e4 LT |
6885 | |
6886 | /* | |
6887 | * this loop exits only if either when we are slower than | |
6888 | * the 'hard' speed limit, or the system was IO-idle for | |
6889 | * a jiffy. | |
6890 | * the system might be non-idle CPU-wise, but we only care | |
6891 | * about not overloading the IO subsystem. (things like an | |
6892 | * e2fsck being done on the RAID array should execute fast) | |
6893 | */ | |
252ac522 | 6894 | md_unplug(mddev); |
1da177e4 LT |
6895 | cond_resched(); |
6896 | ||
57afd89f N |
6897 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
6898 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 6899 | |
88202a0c N |
6900 | if (currspeed > speed_min(mddev)) { |
6901 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 6902 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 6903 | msleep(500); |
1da177e4 LT |
6904 | goto repeat; |
6905 | } | |
6906 | } | |
6907 | } | |
61df9d91 | 6908 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
6909 | /* |
6910 | * this also signals 'finished resyncing' to md_stop | |
6911 | */ | |
6912 | out: | |
252ac522 | 6913 | md_unplug(mddev); |
1da177e4 LT |
6914 | |
6915 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
6916 | ||
6917 | /* tell personality that we are finished */ | |
57afd89f | 6918 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 6919 | |
dfc70645 | 6920 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
6921 | mddev->curr_resync > 2) { |
6922 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
6923 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
6924 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
6925 | printk(KERN_INFO | |
61df9d91 N |
6926 | "md: checkpointing %s of %s.\n", |
6927 | desc, mdname(mddev)); | |
5fd6c1dc N |
6928 | mddev->recovery_cp = mddev->curr_resync; |
6929 | } | |
6930 | } else | |
6931 | mddev->recovery_cp = MaxSector; | |
6932 | } else { | |
6933 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6934 | mddev->curr_resync = MaxSector; | |
4e59ca7d DW |
6935 | rcu_read_lock(); |
6936 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc | 6937 | if (rdev->raid_disk >= 0 && |
70fffd0b | 6938 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
6939 | !test_bit(Faulty, &rdev->flags) && |
6940 | !test_bit(In_sync, &rdev->flags) && | |
6941 | rdev->recovery_offset < mddev->curr_resync) | |
6942 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 6943 | rcu_read_unlock(); |
5fd6c1dc | 6944 | } |
1da177e4 | 6945 | } |
17571284 | 6946 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6947 | |
1da177e4 | 6948 | skip: |
c07b70ad N |
6949 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
6950 | /* We completed so min/max setting can be forgotten if used. */ | |
6951 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6952 | mddev->resync_min = 0; | |
6953 | mddev->resync_max = MaxSector; | |
6954 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6955 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 | 6956 | mddev->curr_resync = 0; |
efa59339 N |
6957 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
6958 | mddev->curr_resync_completed = 0; | |
c6207277 | 6959 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
1da177e4 LT |
6960 | wake_up(&resync_wait); |
6961 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
6962 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
6963 | return; |
6964 | ||
6965 | interrupted: | |
6966 | /* | |
6967 | * got a signal, exit. | |
6968 | */ | |
6969 | printk(KERN_INFO | |
6970 | "md: md_do_sync() got signal ... exiting\n"); | |
6971 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6972 | goto out; | |
6973 | ||
1da177e4 | 6974 | } |
29269553 | 6975 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
6976 | |
6977 | ||
b4c4c7b8 N |
6978 | static int remove_and_add_spares(mddev_t *mddev) |
6979 | { | |
6980 | mdk_rdev_t *rdev; | |
b4c4c7b8 N |
6981 | int spares = 0; |
6982 | ||
97e4f42d N |
6983 | mddev->curr_resync_completed = 0; |
6984 | ||
159ec1fc | 6985 | list_for_each_entry(rdev, &mddev->disks, same_set) |
b4c4c7b8 | 6986 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 6987 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
6988 | (test_bit(Faulty, &rdev->flags) || |
6989 | ! test_bit(In_sync, &rdev->flags)) && | |
6990 | atomic_read(&rdev->nr_pending)==0) { | |
6991 | if (mddev->pers->hot_remove_disk( | |
6992 | mddev, rdev->raid_disk)==0) { | |
6993 | char nm[20]; | |
6994 | sprintf(nm,"rd%d", rdev->raid_disk); | |
6995 | sysfs_remove_link(&mddev->kobj, nm); | |
6996 | rdev->raid_disk = -1; | |
6997 | } | |
6998 | } | |
6999 | ||
4044ba58 | 7000 | if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) { |
159ec1fc | 7001 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dfc70645 | 7002 | if (rdev->raid_disk >= 0 && |
e5427135 DW |
7003 | !test_bit(In_sync, &rdev->flags) && |
7004 | !test_bit(Blocked, &rdev->flags)) | |
dfc70645 | 7005 | spares++; |
b4c4c7b8 N |
7006 | if (rdev->raid_disk < 0 |
7007 | && !test_bit(Faulty, &rdev->flags)) { | |
7008 | rdev->recovery_offset = 0; | |
199050ea NB |
7009 | if (mddev->pers-> |
7010 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
7011 | char nm[20]; |
7012 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
7013 | if (sysfs_create_link(&mddev->kobj, |
7014 | &rdev->kobj, nm)) | |
00bcb4ac | 7015 | /* failure here is OK */; |
b4c4c7b8 N |
7016 | spares++; |
7017 | md_new_event(mddev); | |
93be75ff | 7018 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
b4c4c7b8 N |
7019 | } else |
7020 | break; | |
7021 | } | |
dfc70645 | 7022 | } |
b4c4c7b8 N |
7023 | } |
7024 | return spares; | |
7025 | } | |
1da177e4 LT |
7026 | /* |
7027 | * This routine is regularly called by all per-raid-array threads to | |
7028 | * deal with generic issues like resync and super-block update. | |
7029 | * Raid personalities that don't have a thread (linear/raid0) do not | |
7030 | * need this as they never do any recovery or update the superblock. | |
7031 | * | |
7032 | * It does not do any resync itself, but rather "forks" off other threads | |
7033 | * to do that as needed. | |
7034 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
7035 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 7036 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
7037 | * and wakeups up this thread which will reap the thread and finish up. |
7038 | * This thread also removes any faulty devices (with nr_pending == 0). | |
7039 | * | |
7040 | * The overall approach is: | |
7041 | * 1/ if the superblock needs updating, update it. | |
7042 | * 2/ If a recovery thread is running, don't do anything else. | |
7043 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
7044 | * 4/ If there are any faulty devices, remove them. | |
7045 | * 5/ If array is degraded, try to add spares devices | |
7046 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
7047 | */ | |
7048 | void md_check_recovery(mddev_t *mddev) | |
7049 | { | |
7050 | mdk_rdev_t *rdev; | |
1da177e4 LT |
7051 | |
7052 | ||
5f40402d | 7053 | if (mddev->bitmap) |
aa5cbd10 | 7054 | bitmap_daemon_work(mddev); |
1da177e4 LT |
7055 | |
7056 | if (mddev->ro) | |
7057 | return; | |
fca4d848 N |
7058 | |
7059 | if (signal_pending(current)) { | |
31a59e34 | 7060 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
7061 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
7062 | mdname(mddev)); | |
7063 | mddev->safemode = 2; | |
7064 | } | |
7065 | flush_signals(current); | |
7066 | } | |
7067 | ||
c89a8eee N |
7068 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
7069 | return; | |
1da177e4 | 7070 | if ( ! ( |
e691063a | 7071 | (mddev->flags && !mddev->external) || |
1da177e4 | 7072 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 7073 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 7074 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
7075 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
7076 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
7077 | )) |
7078 | return; | |
fca4d848 | 7079 | |
df5b89b3 | 7080 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 7081 | int spares = 0; |
fca4d848 | 7082 | |
c89a8eee N |
7083 | if (mddev->ro) { |
7084 | /* Only thing we do on a ro array is remove | |
7085 | * failed devices. | |
7086 | */ | |
7087 | remove_and_add_spares(mddev); | |
7088 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7089 | goto unlock; | |
7090 | } | |
7091 | ||
31a59e34 | 7092 | if (!mddev->external) { |
0fd62b86 | 7093 | int did_change = 0; |
31a59e34 N |
7094 | spin_lock_irq(&mddev->write_lock); |
7095 | if (mddev->safemode && | |
7096 | !atomic_read(&mddev->writes_pending) && | |
7097 | !mddev->in_sync && | |
7098 | mddev->recovery_cp == MaxSector) { | |
7099 | mddev->in_sync = 1; | |
0fd62b86 | 7100 | did_change = 1; |
31a59e34 N |
7101 | if (mddev->persistent) |
7102 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
7103 | } | |
7104 | if (mddev->safemode == 1) | |
7105 | mddev->safemode = 0; | |
7106 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 7107 | if (did_change) |
00bcb4ac | 7108 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 7109 | } |
fca4d848 | 7110 | |
850b2b42 N |
7111 | if (mddev->flags) |
7112 | md_update_sb(mddev, 0); | |
06d91a5f | 7113 | |
1da177e4 LT |
7114 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7115 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7116 | /* resync/recovery still happening */ | |
7117 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7118 | goto unlock; | |
7119 | } | |
7120 | if (mddev->sync_thread) { | |
7121 | /* resync has finished, collect result */ | |
7122 | md_unregister_thread(mddev->sync_thread); | |
7123 | mddev->sync_thread = NULL; | |
56ac36d7 DW |
7124 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
7125 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
1da177e4 LT |
7126 | /* success...*/ |
7127 | /* activate any spares */ | |
a99ac971 NB |
7128 | if (mddev->pers->spare_active(mddev)) |
7129 | sysfs_notify(&mddev->kobj, NULL, | |
7130 | "degraded"); | |
1da177e4 | 7131 | } |
cea9c228 N |
7132 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7133 | mddev->pers->finish_reshape) | |
7134 | mddev->pers->finish_reshape(mddev); | |
850b2b42 | 7135 | md_update_sb(mddev, 1); |
41158c7e N |
7136 | |
7137 | /* if array is no-longer degraded, then any saved_raid_disk | |
7138 | * information must be scrapped | |
7139 | */ | |
7140 | if (!mddev->degraded) | |
159ec1fc | 7141 | list_for_each_entry(rdev, &mddev->disks, same_set) |
41158c7e N |
7142 | rdev->saved_raid_disk = -1; |
7143 | ||
1da177e4 LT |
7144 | mddev->recovery = 0; |
7145 | /* flag recovery needed just to double check */ | |
7146 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
00bcb4ac | 7147 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
d7603b7e | 7148 | md_new_event(mddev); |
1da177e4 LT |
7149 | goto unlock; |
7150 | } | |
72a23c21 NB |
7151 | /* Set RUNNING before clearing NEEDED to avoid |
7152 | * any transients in the value of "sync_action". | |
7153 | */ | |
7154 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7155 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
7156 | /* Clear some bits that don't mean anything, but |
7157 | * might be left set | |
7158 | */ | |
24dd469d N |
7159 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7160 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7161 | |
5fd6c1dc N |
7162 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
7163 | goto unlock; | |
1da177e4 LT |
7164 | /* no recovery is running. |
7165 | * remove any failed drives, then | |
7166 | * add spares if possible. | |
7167 | * Spare are also removed and re-added, to allow | |
7168 | * the personality to fail the re-add. | |
7169 | */ | |
1da177e4 | 7170 | |
b4c4c7b8 | 7171 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7172 | if (mddev->pers->check_reshape == NULL || |
7173 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
7174 | /* Cannot proceed */ |
7175 | goto unlock; | |
7176 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 7177 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 7178 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
7179 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7180 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7181 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7182 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7183 | } else if (mddev->recovery_cp < MaxSector) { |
7184 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7185 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7186 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7187 | /* nothing to be done ... */ | |
1da177e4 | 7188 | goto unlock; |
24dd469d | 7189 | |
1da177e4 | 7190 | if (mddev->pers->sync_request) { |
a654b9d8 N |
7191 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
7192 | /* We are adding a device or devices to an array | |
7193 | * which has the bitmap stored on all devices. | |
7194 | * So make sure all bitmap pages get written | |
7195 | */ | |
7196 | bitmap_write_all(mddev->bitmap); | |
7197 | } | |
1da177e4 LT |
7198 | mddev->sync_thread = md_register_thread(md_do_sync, |
7199 | mddev, | |
0da3c619 | 7200 | "resync"); |
1da177e4 LT |
7201 | if (!mddev->sync_thread) { |
7202 | printk(KERN_ERR "%s: could not start resync" | |
7203 | " thread...\n", | |
7204 | mdname(mddev)); | |
7205 | /* leave the spares where they are, it shouldn't hurt */ | |
7206 | mddev->recovery = 0; | |
d7603b7e | 7207 | } else |
1da177e4 | 7208 | md_wakeup_thread(mddev->sync_thread); |
00bcb4ac | 7209 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
d7603b7e | 7210 | md_new_event(mddev); |
1da177e4 LT |
7211 | } |
7212 | unlock: | |
72a23c21 NB |
7213 | if (!mddev->sync_thread) { |
7214 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7215 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7216 | &mddev->recovery)) | |
0c3573f1 | 7217 | if (mddev->sysfs_action) |
00bcb4ac | 7218 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 7219 | } |
1da177e4 LT |
7220 | mddev_unlock(mddev); |
7221 | } | |
7222 | } | |
7223 | ||
6bfe0b49 DW |
7224 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
7225 | { | |
00bcb4ac | 7226 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 DW |
7227 | wait_event_timeout(rdev->blocked_wait, |
7228 | !test_bit(Blocked, &rdev->flags), | |
7229 | msecs_to_jiffies(5000)); | |
7230 | rdev_dec_pending(rdev, mddev); | |
7231 | } | |
7232 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
7233 | ||
75c96f85 AB |
7234 | static int md_notify_reboot(struct notifier_block *this, |
7235 | unsigned long code, void *x) | |
1da177e4 LT |
7236 | { |
7237 | struct list_head *tmp; | |
7238 | mddev_t *mddev; | |
7239 | ||
7240 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
7241 | ||
7242 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
7243 | ||
29ac4aa3 | 7244 | for_each_mddev(mddev, tmp) |
c71d4887 | 7245 | if (mddev_trylock(mddev)) { |
2b25000b N |
7246 | /* Force a switch to readonly even array |
7247 | * appears to still be in use. Hence | |
7248 | * the '100'. | |
7249 | */ | |
a4bd82d0 | 7250 | md_set_readonly(mddev, 100); |
c71d4887 NB |
7251 | mddev_unlock(mddev); |
7252 | } | |
1da177e4 LT |
7253 | /* |
7254 | * certain more exotic SCSI devices are known to be | |
7255 | * volatile wrt too early system reboots. While the | |
7256 | * right place to handle this issue is the given | |
7257 | * driver, we do want to have a safe RAID driver ... | |
7258 | */ | |
7259 | mdelay(1000*1); | |
7260 | } | |
7261 | return NOTIFY_DONE; | |
7262 | } | |
7263 | ||
75c96f85 | 7264 | static struct notifier_block md_notifier = { |
1da177e4 LT |
7265 | .notifier_call = md_notify_reboot, |
7266 | .next = NULL, | |
7267 | .priority = INT_MAX, /* before any real devices */ | |
7268 | }; | |
7269 | ||
7270 | static void md_geninit(void) | |
7271 | { | |
1da177e4 LT |
7272 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
7273 | ||
c7705f34 | 7274 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
7275 | } |
7276 | ||
75c96f85 | 7277 | static int __init md_init(void) |
1da177e4 | 7278 | { |
3dbd8c2e | 7279 | if (register_blkdev(MD_MAJOR, "md")) |
1da177e4 LT |
7280 | return -1; |
7281 | if ((mdp_major=register_blkdev(0, "mdp"))<=0) { | |
3dbd8c2e | 7282 | unregister_blkdev(MD_MAJOR, "md"); |
1da177e4 LT |
7283 | return -1; |
7284 | } | |
3dbd8c2e | 7285 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
7286 | md_probe, NULL, NULL); |
7287 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
7288 | md_probe, NULL, NULL); |
7289 | ||
1da177e4 | 7290 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 7291 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
7292 | |
7293 | md_geninit(); | |
d710e138 | 7294 | return 0; |
1da177e4 LT |
7295 | } |
7296 | ||
7297 | ||
7298 | #ifndef MODULE | |
7299 | ||
7300 | /* | |
7301 | * Searches all registered partitions for autorun RAID arrays | |
7302 | * at boot time. | |
7303 | */ | |
4d936ec1 ME |
7304 | |
7305 | static LIST_HEAD(all_detected_devices); | |
7306 | struct detected_devices_node { | |
7307 | struct list_head list; | |
7308 | dev_t dev; | |
7309 | }; | |
1da177e4 LT |
7310 | |
7311 | void md_autodetect_dev(dev_t dev) | |
7312 | { | |
4d936ec1 ME |
7313 | struct detected_devices_node *node_detected_dev; |
7314 | ||
7315 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
7316 | if (node_detected_dev) { | |
7317 | node_detected_dev->dev = dev; | |
7318 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
7319 | } else { | |
7320 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
7321 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
7322 | } | |
1da177e4 LT |
7323 | } |
7324 | ||
7325 | ||
7326 | static void autostart_arrays(int part) | |
7327 | { | |
7328 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
7329 | struct detected_devices_node *node_detected_dev; |
7330 | dev_t dev; | |
7331 | int i_scanned, i_passed; | |
1da177e4 | 7332 | |
4d936ec1 ME |
7333 | i_scanned = 0; |
7334 | i_passed = 0; | |
1da177e4 | 7335 | |
4d936ec1 | 7336 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 7337 | |
4d936ec1 ME |
7338 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
7339 | i_scanned++; | |
7340 | node_detected_dev = list_entry(all_detected_devices.next, | |
7341 | struct detected_devices_node, list); | |
7342 | list_del(&node_detected_dev->list); | |
7343 | dev = node_detected_dev->dev; | |
7344 | kfree(node_detected_dev); | |
df968c4e | 7345 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
7346 | if (IS_ERR(rdev)) |
7347 | continue; | |
7348 | ||
b2d444d7 | 7349 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7350 | MD_BUG(); |
7351 | continue; | |
7352 | } | |
d0fae18f | 7353 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 7354 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 7355 | i_passed++; |
1da177e4 | 7356 | } |
4d936ec1 ME |
7357 | |
7358 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
7359 | i_scanned, i_passed); | |
1da177e4 LT |
7360 | |
7361 | autorun_devices(part); | |
7362 | } | |
7363 | ||
fdee8ae4 | 7364 | #endif /* !MODULE */ |
1da177e4 LT |
7365 | |
7366 | static __exit void md_exit(void) | |
7367 | { | |
7368 | mddev_t *mddev; | |
7369 | struct list_head *tmp; | |
8ab5e4c1 | 7370 | |
3dbd8c2e | 7371 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 7372 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 7373 | |
3dbd8c2e | 7374 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
7375 | unregister_blkdev(mdp_major, "mdp"); |
7376 | unregister_reboot_notifier(&md_notifier); | |
7377 | unregister_sysctl_table(raid_table_header); | |
7378 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 7379 | for_each_mddev(mddev, tmp) { |
1da177e4 | 7380 | export_array(mddev); |
d3374825 | 7381 | mddev->hold_active = 0; |
1da177e4 LT |
7382 | } |
7383 | } | |
7384 | ||
685784aa | 7385 | subsys_initcall(md_init); |
1da177e4 LT |
7386 | module_exit(md_exit) |
7387 | ||
f91de92e N |
7388 | static int get_ro(char *buffer, struct kernel_param *kp) |
7389 | { | |
7390 | return sprintf(buffer, "%d", start_readonly); | |
7391 | } | |
7392 | static int set_ro(const char *val, struct kernel_param *kp) | |
7393 | { | |
7394 | char *e; | |
7395 | int num = simple_strtoul(val, &e, 10); | |
7396 | if (*val && (*e == '\0' || *e == '\n')) { | |
7397 | start_readonly = num; | |
4dbcdc75 | 7398 | return 0; |
f91de92e N |
7399 | } |
7400 | return -EINVAL; | |
7401 | } | |
7402 | ||
80ca3a44 N |
7403 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
7404 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 7405 | |
efeb53c0 | 7406 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 7407 | |
1da177e4 LT |
7408 | EXPORT_SYMBOL(register_md_personality); |
7409 | EXPORT_SYMBOL(unregister_md_personality); | |
7410 | EXPORT_SYMBOL(md_error); | |
7411 | EXPORT_SYMBOL(md_done_sync); | |
7412 | EXPORT_SYMBOL(md_write_start); | |
7413 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
7414 | EXPORT_SYMBOL(md_register_thread); |
7415 | EXPORT_SYMBOL(md_unregister_thread); | |
7416 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
7417 | EXPORT_SYMBOL(md_check_recovery); |
7418 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7419 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 7420 | MODULE_ALIAS("md"); |
72008652 | 7421 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |