Commit | Line | Data |
---|---|---|
e7fd4179 DT |
1 | /****************************************************************************** |
2 | ******************************************************************************* | |
3 | ** | |
4 | ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
60f98d18 | 5 | ** Copyright (C) 2004-2011 Red Hat, Inc. All rights reserved. |
e7fd4179 DT |
6 | ** |
7 | ** This copyrighted material is made available to anyone wishing to use, | |
8 | ** modify, copy, or redistribute it subject to the terms and conditions | |
9 | ** of the GNU General Public License v.2. | |
10 | ** | |
11 | ******************************************************************************* | |
12 | ******************************************************************************/ | |
13 | ||
14 | #include "dlm_internal.h" | |
15 | #include "lockspace.h" | |
16 | #include "member.h" | |
17 | #include "recoverd.h" | |
e7fd4179 DT |
18 | #include "dir.h" |
19 | #include "lowcomms.h" | |
20 | #include "config.h" | |
21 | #include "memory.h" | |
22 | #include "lock.h" | |
c56b39cd | 23 | #include "recover.h" |
2896ee37 | 24 | #include "requestqueue.h" |
0f8e0d9a | 25 | #include "user.h" |
23e8e1aa | 26 | #include "ast.h" |
e7fd4179 | 27 | |
e7fd4179 | 28 | static int ls_count; |
90135925 | 29 | static struct mutex ls_lock; |
e7fd4179 DT |
30 | static struct list_head lslist; |
31 | static spinlock_t lslist_lock; | |
32 | static struct task_struct * scand_task; | |
33 | ||
34 | ||
35 | static ssize_t dlm_control_store(struct dlm_ls *ls, const char *buf, size_t len) | |
36 | { | |
37 | ssize_t ret = len; | |
38 | int n = simple_strtol(buf, NULL, 0); | |
39 | ||
e2de7f56 PC |
40 | ls = dlm_find_lockspace_local(ls->ls_local_handle); |
41 | if (!ls) | |
42 | return -EINVAL; | |
43 | ||
e7fd4179 DT |
44 | switch (n) { |
45 | case 0: | |
46 | dlm_ls_stop(ls); | |
47 | break; | |
48 | case 1: | |
49 | dlm_ls_start(ls); | |
50 | break; | |
51 | default: | |
52 | ret = -EINVAL; | |
53 | } | |
e2de7f56 | 54 | dlm_put_lockspace(ls); |
e7fd4179 DT |
55 | return ret; |
56 | } | |
57 | ||
58 | static ssize_t dlm_event_store(struct dlm_ls *ls, const char *buf, size_t len) | |
59 | { | |
60 | ls->ls_uevent_result = simple_strtol(buf, NULL, 0); | |
61 | set_bit(LSFL_UEVENT_WAIT, &ls->ls_flags); | |
62 | wake_up(&ls->ls_uevent_wait); | |
63 | return len; | |
64 | } | |
65 | ||
66 | static ssize_t dlm_id_show(struct dlm_ls *ls, char *buf) | |
67 | { | |
a1d144c7 | 68 | return snprintf(buf, PAGE_SIZE, "%u\n", ls->ls_global_id); |
e7fd4179 DT |
69 | } |
70 | ||
71 | static ssize_t dlm_id_store(struct dlm_ls *ls, const char *buf, size_t len) | |
72 | { | |
73 | ls->ls_global_id = simple_strtoul(buf, NULL, 0); | |
74 | return len; | |
75 | } | |
76 | ||
c56b39cd DT |
77 | static ssize_t dlm_recover_status_show(struct dlm_ls *ls, char *buf) |
78 | { | |
79 | uint32_t status = dlm_recover_status(ls); | |
a1d144c7 | 80 | return snprintf(buf, PAGE_SIZE, "%x\n", status); |
c56b39cd DT |
81 | } |
82 | ||
faa0f267 DT |
83 | static ssize_t dlm_recover_nodeid_show(struct dlm_ls *ls, char *buf) |
84 | { | |
a1d144c7 | 85 | return snprintf(buf, PAGE_SIZE, "%d\n", ls->ls_recover_nodeid); |
faa0f267 DT |
86 | } |
87 | ||
e7fd4179 DT |
88 | struct dlm_attr { |
89 | struct attribute attr; | |
90 | ssize_t (*show)(struct dlm_ls *, char *); | |
91 | ssize_t (*store)(struct dlm_ls *, const char *, size_t); | |
92 | }; | |
93 | ||
94 | static struct dlm_attr dlm_attr_control = { | |
95 | .attr = {.name = "control", .mode = S_IWUSR}, | |
96 | .store = dlm_control_store | |
97 | }; | |
98 | ||
99 | static struct dlm_attr dlm_attr_event = { | |
100 | .attr = {.name = "event_done", .mode = S_IWUSR}, | |
101 | .store = dlm_event_store | |
102 | }; | |
103 | ||
104 | static struct dlm_attr dlm_attr_id = { | |
105 | .attr = {.name = "id", .mode = S_IRUGO | S_IWUSR}, | |
106 | .show = dlm_id_show, | |
107 | .store = dlm_id_store | |
108 | }; | |
109 | ||
c56b39cd DT |
110 | static struct dlm_attr dlm_attr_recover_status = { |
111 | .attr = {.name = "recover_status", .mode = S_IRUGO}, | |
112 | .show = dlm_recover_status_show | |
113 | }; | |
114 | ||
faa0f267 DT |
115 | static struct dlm_attr dlm_attr_recover_nodeid = { |
116 | .attr = {.name = "recover_nodeid", .mode = S_IRUGO}, | |
117 | .show = dlm_recover_nodeid_show | |
118 | }; | |
119 | ||
e7fd4179 DT |
120 | static struct attribute *dlm_attrs[] = { |
121 | &dlm_attr_control.attr, | |
122 | &dlm_attr_event.attr, | |
123 | &dlm_attr_id.attr, | |
c56b39cd | 124 | &dlm_attr_recover_status.attr, |
faa0f267 | 125 | &dlm_attr_recover_nodeid.attr, |
e7fd4179 DT |
126 | NULL, |
127 | }; | |
128 | ||
129 | static ssize_t dlm_attr_show(struct kobject *kobj, struct attribute *attr, | |
130 | char *buf) | |
131 | { | |
132 | struct dlm_ls *ls = container_of(kobj, struct dlm_ls, ls_kobj); | |
133 | struct dlm_attr *a = container_of(attr, struct dlm_attr, attr); | |
134 | return a->show ? a->show(ls, buf) : 0; | |
135 | } | |
136 | ||
137 | static ssize_t dlm_attr_store(struct kobject *kobj, struct attribute *attr, | |
138 | const char *buf, size_t len) | |
139 | { | |
140 | struct dlm_ls *ls = container_of(kobj, struct dlm_ls, ls_kobj); | |
141 | struct dlm_attr *a = container_of(attr, struct dlm_attr, attr); | |
142 | return a->store ? a->store(ls, buf, len) : len; | |
143 | } | |
144 | ||
ba542e3b PC |
145 | static void lockspace_kobj_release(struct kobject *k) |
146 | { | |
147 | struct dlm_ls *ls = container_of(k, struct dlm_ls, ls_kobj); | |
148 | kfree(ls); | |
149 | } | |
150 | ||
52cf25d0 | 151 | static const struct sysfs_ops dlm_attr_ops = { |
e7fd4179 DT |
152 | .show = dlm_attr_show, |
153 | .store = dlm_attr_store, | |
154 | }; | |
155 | ||
156 | static struct kobj_type dlm_ktype = { | |
157 | .default_attrs = dlm_attrs, | |
158 | .sysfs_ops = &dlm_attr_ops, | |
ba542e3b | 159 | .release = lockspace_kobj_release, |
e7fd4179 DT |
160 | }; |
161 | ||
d405936b | 162 | static struct kset *dlm_kset; |
e7fd4179 | 163 | |
e7fd4179 DT |
164 | static int do_uevent(struct dlm_ls *ls, int in) |
165 | { | |
166 | int error; | |
167 | ||
168 | if (in) | |
169 | kobject_uevent(&ls->ls_kobj, KOBJ_ONLINE); | |
170 | else | |
171 | kobject_uevent(&ls->ls_kobj, KOBJ_OFFLINE); | |
172 | ||
8b0e7b2c DT |
173 | log_debug(ls, "%s the lockspace group...", in ? "joining" : "leaving"); |
174 | ||
175 | /* dlm_controld will see the uevent, do the necessary group management | |
176 | and then write to sysfs to wake us */ | |
177 | ||
e7fd4179 DT |
178 | error = wait_event_interruptible(ls->ls_uevent_wait, |
179 | test_and_clear_bit(LSFL_UEVENT_WAIT, &ls->ls_flags)); | |
8b0e7b2c DT |
180 | |
181 | log_debug(ls, "group event done %d %d", error, ls->ls_uevent_result); | |
182 | ||
e7fd4179 DT |
183 | if (error) |
184 | goto out; | |
185 | ||
186 | error = ls->ls_uevent_result; | |
187 | out: | |
8b0e7b2c DT |
188 | if (error) |
189 | log_error(ls, "group %s failed %d %d", in ? "join" : "leave", | |
190 | error, ls->ls_uevent_result); | |
e7fd4179 DT |
191 | return error; |
192 | } | |
193 | ||
b4a5d4bc SW |
194 | static int dlm_uevent(struct kset *kset, struct kobject *kobj, |
195 | struct kobj_uevent_env *env) | |
196 | { | |
197 | struct dlm_ls *ls = container_of(kobj, struct dlm_ls, ls_kobj); | |
198 | ||
199 | add_uevent_var(env, "LOCKSPACE=%s", ls->ls_name); | |
200 | return 0; | |
201 | } | |
202 | ||
203 | static struct kset_uevent_ops dlm_uevent_ops = { | |
204 | .uevent = dlm_uevent, | |
205 | }; | |
e7fd4179 | 206 | |
30727174 | 207 | int __init dlm_lockspace_init(void) |
e7fd4179 | 208 | { |
e7fd4179 | 209 | ls_count = 0; |
90135925 | 210 | mutex_init(&ls_lock); |
e7fd4179 DT |
211 | INIT_LIST_HEAD(&lslist); |
212 | spin_lock_init(&lslist_lock); | |
213 | ||
b4a5d4bc | 214 | dlm_kset = kset_create_and_add("dlm", &dlm_uevent_ops, kernel_kobj); |
d405936b | 215 | if (!dlm_kset) { |
8e24eea7 | 216 | printk(KERN_WARNING "%s: can not create kset\n", __func__); |
d405936b GKH |
217 | return -ENOMEM; |
218 | } | |
219 | return 0; | |
e7fd4179 DT |
220 | } |
221 | ||
222 | void dlm_lockspace_exit(void) | |
223 | { | |
d405936b | 224 | kset_unregister(dlm_kset); |
e7fd4179 DT |
225 | } |
226 | ||
c1dcf65f DT |
227 | static struct dlm_ls *find_ls_to_scan(void) |
228 | { | |
229 | struct dlm_ls *ls; | |
230 | ||
231 | spin_lock(&lslist_lock); | |
232 | list_for_each_entry(ls, &lslist, ls_list) { | |
233 | if (time_after_eq(jiffies, ls->ls_scan_time + | |
234 | dlm_config.ci_scan_secs * HZ)) { | |
235 | spin_unlock(&lslist_lock); | |
236 | return ls; | |
237 | } | |
238 | } | |
239 | spin_unlock(&lslist_lock); | |
240 | return NULL; | |
241 | } | |
242 | ||
e7fd4179 DT |
243 | static int dlm_scand(void *data) |
244 | { | |
245 | struct dlm_ls *ls; | |
246 | ||
247 | while (!kthread_should_stop()) { | |
c1dcf65f DT |
248 | ls = find_ls_to_scan(); |
249 | if (ls) { | |
85e86edf | 250 | if (dlm_lock_recovery_try(ls)) { |
c1dcf65f | 251 | ls->ls_scan_time = jiffies; |
85e86edf | 252 | dlm_scan_rsbs(ls); |
3ae1acf9 | 253 | dlm_scan_timeout(ls); |
c6ff669b | 254 | dlm_scan_waiters(ls); |
85e86edf | 255 | dlm_unlock_recovery(ls); |
c1dcf65f DT |
256 | } else { |
257 | ls->ls_scan_time += HZ; | |
85e86edf | 258 | } |
c6ff669b | 259 | continue; |
85e86edf | 260 | } |
c6ff669b | 261 | schedule_timeout_interruptible(dlm_config.ci_scan_secs * HZ); |
e7fd4179 DT |
262 | } |
263 | return 0; | |
264 | } | |
265 | ||
266 | static int dlm_scand_start(void) | |
267 | { | |
268 | struct task_struct *p; | |
269 | int error = 0; | |
270 | ||
271 | p = kthread_run(dlm_scand, NULL, "dlm_scand"); | |
272 | if (IS_ERR(p)) | |
273 | error = PTR_ERR(p); | |
274 | else | |
275 | scand_task = p; | |
276 | return error; | |
277 | } | |
278 | ||
279 | static void dlm_scand_stop(void) | |
280 | { | |
281 | kthread_stop(scand_task); | |
282 | } | |
283 | ||
e7fd4179 DT |
284 | struct dlm_ls *dlm_find_lockspace_global(uint32_t id) |
285 | { | |
286 | struct dlm_ls *ls; | |
287 | ||
288 | spin_lock(&lslist_lock); | |
289 | ||
290 | list_for_each_entry(ls, &lslist, ls_list) { | |
291 | if (ls->ls_global_id == id) { | |
292 | ls->ls_count++; | |
293 | goto out; | |
294 | } | |
295 | } | |
296 | ls = NULL; | |
297 | out: | |
298 | spin_unlock(&lslist_lock); | |
299 | return ls; | |
300 | } | |
301 | ||
597d0cae | 302 | struct dlm_ls *dlm_find_lockspace_local(dlm_lockspace_t *lockspace) |
e7fd4179 | 303 | { |
597d0cae | 304 | struct dlm_ls *ls; |
e7fd4179 DT |
305 | |
306 | spin_lock(&lslist_lock); | |
597d0cae DT |
307 | list_for_each_entry(ls, &lslist, ls_list) { |
308 | if (ls->ls_local_handle == lockspace) { | |
309 | ls->ls_count++; | |
310 | goto out; | |
311 | } | |
312 | } | |
313 | ls = NULL; | |
314 | out: | |
315 | spin_unlock(&lslist_lock); | |
316 | return ls; | |
317 | } | |
318 | ||
319 | struct dlm_ls *dlm_find_lockspace_device(int minor) | |
320 | { | |
321 | struct dlm_ls *ls; | |
322 | ||
323 | spin_lock(&lslist_lock); | |
324 | list_for_each_entry(ls, &lslist, ls_list) { | |
325 | if (ls->ls_device.minor == minor) { | |
326 | ls->ls_count++; | |
327 | goto out; | |
328 | } | |
329 | } | |
330 | ls = NULL; | |
331 | out: | |
e7fd4179 DT |
332 | spin_unlock(&lslist_lock); |
333 | return ls; | |
334 | } | |
335 | ||
336 | void dlm_put_lockspace(struct dlm_ls *ls) | |
337 | { | |
338 | spin_lock(&lslist_lock); | |
339 | ls->ls_count--; | |
340 | spin_unlock(&lslist_lock); | |
341 | } | |
342 | ||
343 | static void remove_lockspace(struct dlm_ls *ls) | |
344 | { | |
345 | for (;;) { | |
346 | spin_lock(&lslist_lock); | |
347 | if (ls->ls_count == 0) { | |
0f8e0d9a | 348 | WARN_ON(ls->ls_create_count != 0); |
e7fd4179 DT |
349 | list_del(&ls->ls_list); |
350 | spin_unlock(&lslist_lock); | |
351 | return; | |
352 | } | |
353 | spin_unlock(&lslist_lock); | |
354 | ssleep(1); | |
355 | } | |
356 | } | |
357 | ||
358 | static int threads_start(void) | |
359 | { | |
360 | int error; | |
361 | ||
e7fd4179 DT |
362 | error = dlm_scand_start(); |
363 | if (error) { | |
364 | log_print("cannot start dlm_scand thread %d", error); | |
23e8e1aa | 365 | goto fail; |
e7fd4179 DT |
366 | } |
367 | ||
368 | /* Thread for sending/receiving messages for all lockspace's */ | |
369 | error = dlm_lowcomms_start(); | |
370 | if (error) { | |
371 | log_print("cannot start dlm lowcomms %d", error); | |
372 | goto scand_fail; | |
373 | } | |
374 | ||
375 | return 0; | |
376 | ||
377 | scand_fail: | |
378 | dlm_scand_stop(); | |
e7fd4179 DT |
379 | fail: |
380 | return error; | |
381 | } | |
382 | ||
383 | static void threads_stop(void) | |
384 | { | |
385 | dlm_scand_stop(); | |
386 | dlm_lowcomms_stop(); | |
e7fd4179 DT |
387 | } |
388 | ||
60f98d18 DT |
389 | static int new_lockspace(const char *name, const char *cluster, |
390 | uint32_t flags, int lvblen, | |
391 | const struct dlm_lockspace_ops *ops, void *ops_arg, | |
392 | int *ops_result, dlm_lockspace_t **lockspace) | |
e7fd4179 DT |
393 | { |
394 | struct dlm_ls *ls; | |
0f8e0d9a | 395 | int i, size, error; |
79d72b54 | 396 | int do_unreg = 0; |
60f98d18 | 397 | int namelen = strlen(name); |
e7fd4179 DT |
398 | |
399 | if (namelen > DLM_LOCKSPACE_LEN) | |
400 | return -EINVAL; | |
401 | ||
402 | if (!lvblen || (lvblen % 8)) | |
403 | return -EINVAL; | |
404 | ||
405 | if (!try_module_get(THIS_MODULE)) | |
406 | return -EINVAL; | |
407 | ||
dc68c7ed | 408 | if (!dlm_user_daemon_available()) { |
60f98d18 DT |
409 | log_print("dlm user daemon not available"); |
410 | error = -EUNATCH; | |
411 | goto out; | |
412 | } | |
413 | ||
414 | if (ops && ops_result) { | |
415 | if (!dlm_config.ci_recover_callbacks) | |
416 | *ops_result = -EOPNOTSUPP; | |
417 | else | |
418 | *ops_result = 0; | |
419 | } | |
420 | ||
421 | if (dlm_config.ci_recover_callbacks && cluster && | |
422 | strncmp(cluster, dlm_config.ci_cluster_name, DLM_LOCKSPACE_LEN)) { | |
423 | log_print("dlm cluster name %s mismatch %s", | |
424 | dlm_config.ci_cluster_name, cluster); | |
425 | error = -EBADR; | |
426 | goto out; | |
dc68c7ed DT |
427 | } |
428 | ||
0f8e0d9a DT |
429 | error = 0; |
430 | ||
431 | spin_lock(&lslist_lock); | |
432 | list_for_each_entry(ls, &lslist, ls_list) { | |
433 | WARN_ON(ls->ls_create_count <= 0); | |
434 | if (ls->ls_namelen != namelen) | |
435 | continue; | |
436 | if (memcmp(ls->ls_name, name, namelen)) | |
437 | continue; | |
438 | if (flags & DLM_LSFL_NEWEXCL) { | |
439 | error = -EEXIST; | |
440 | break; | |
441 | } | |
442 | ls->ls_create_count++; | |
8511a272 DT |
443 | *lockspace = ls; |
444 | error = 1; | |
0f8e0d9a | 445 | break; |
e7fd4179 | 446 | } |
0f8e0d9a DT |
447 | spin_unlock(&lslist_lock); |
448 | ||
0f8e0d9a | 449 | if (error) |
8511a272 | 450 | goto out; |
0f8e0d9a DT |
451 | |
452 | error = -ENOMEM; | |
e7fd4179 | 453 | |
573c24c4 | 454 | ls = kzalloc(sizeof(struct dlm_ls) + namelen, GFP_NOFS); |
e7fd4179 DT |
455 | if (!ls) |
456 | goto out; | |
e7fd4179 DT |
457 | memcpy(ls->ls_name, name, namelen); |
458 | ls->ls_namelen = namelen; | |
e7fd4179 DT |
459 | ls->ls_lvblen = lvblen; |
460 | ls->ls_count = 0; | |
461 | ls->ls_flags = 0; | |
c1dcf65f | 462 | ls->ls_scan_time = jiffies; |
e7fd4179 | 463 | |
60f98d18 DT |
464 | if (ops && dlm_config.ci_recover_callbacks) { |
465 | ls->ls_ops = ops; | |
466 | ls->ls_ops_arg = ops_arg; | |
467 | } | |
468 | ||
3ae1acf9 DT |
469 | if (flags & DLM_LSFL_TIMEWARN) |
470 | set_bit(LSFL_TIMEWARN, &ls->ls_flags); | |
3ae1acf9 | 471 | |
fad59c13 | 472 | /* ls_exflags are forced to match among nodes, and we don't |
0f8e0d9a DT |
473 | need to require all nodes to have some flags set */ |
474 | ls->ls_exflags = (flags & ~(DLM_LSFL_TIMEWARN | DLM_LSFL_FS | | |
475 | DLM_LSFL_NEWEXCL)); | |
fad59c13 | 476 | |
68c817a1 | 477 | size = dlm_config.ci_rsbtbl_size; |
e7fd4179 DT |
478 | ls->ls_rsbtbl_size = size; |
479 | ||
c282af49 | 480 | ls->ls_rsbtbl = vmalloc(sizeof(struct dlm_rsbtable) * size); |
e7fd4179 DT |
481 | if (!ls->ls_rsbtbl) |
482 | goto out_lsfree; | |
483 | for (i = 0; i < size; i++) { | |
9beb3bf5 BP |
484 | ls->ls_rsbtbl[i].keep.rb_node = NULL; |
485 | ls->ls_rsbtbl[i].toss.rb_node = NULL; | |
c7be761a | 486 | spin_lock_init(&ls->ls_rsbtbl[i].lock); |
e7fd4179 DT |
487 | } |
488 | ||
3d6aa675 DT |
489 | idr_init(&ls->ls_lkbidr); |
490 | spin_lock_init(&ls->ls_lkbidr_spin); | |
e7fd4179 | 491 | |
68c817a1 | 492 | size = dlm_config.ci_dirtbl_size; |
e7fd4179 DT |
493 | ls->ls_dirtbl_size = size; |
494 | ||
c282af49 | 495 | ls->ls_dirtbl = vmalloc(sizeof(struct dlm_dirtable) * size); |
e7fd4179 DT |
496 | if (!ls->ls_dirtbl) |
497 | goto out_lkbfree; | |
498 | for (i = 0; i < size; i++) { | |
499 | INIT_LIST_HEAD(&ls->ls_dirtbl[i].list); | |
305a47b1 | 500 | spin_lock_init(&ls->ls_dirtbl[i].lock); |
e7fd4179 DT |
501 | } |
502 | ||
503 | INIT_LIST_HEAD(&ls->ls_waiters); | |
90135925 | 504 | mutex_init(&ls->ls_waiters_mutex); |
ef0c2bb0 DT |
505 | INIT_LIST_HEAD(&ls->ls_orphans); |
506 | mutex_init(&ls->ls_orphans_mutex); | |
3ae1acf9 DT |
507 | INIT_LIST_HEAD(&ls->ls_timeout); |
508 | mutex_init(&ls->ls_timeout_mutex); | |
e7fd4179 | 509 | |
3881ac04 DT |
510 | INIT_LIST_HEAD(&ls->ls_new_rsb); |
511 | spin_lock_init(&ls->ls_new_rsb_spin); | |
512 | ||
e7fd4179 DT |
513 | INIT_LIST_HEAD(&ls->ls_nodes); |
514 | INIT_LIST_HEAD(&ls->ls_nodes_gone); | |
515 | ls->ls_num_nodes = 0; | |
516 | ls->ls_low_nodeid = 0; | |
517 | ls->ls_total_weight = 0; | |
518 | ls->ls_node_array = NULL; | |
519 | ||
520 | memset(&ls->ls_stub_rsb, 0, sizeof(struct dlm_rsb)); | |
521 | ls->ls_stub_rsb.res_ls = ls; | |
522 | ||
5de6319b DT |
523 | ls->ls_debug_rsb_dentry = NULL; |
524 | ls->ls_debug_waiters_dentry = NULL; | |
e7fd4179 DT |
525 | |
526 | init_waitqueue_head(&ls->ls_uevent_wait); | |
527 | ls->ls_uevent_result = 0; | |
8b0e7b2c DT |
528 | init_completion(&ls->ls_members_done); |
529 | ls->ls_members_result = -1; | |
e7fd4179 | 530 | |
23e8e1aa DT |
531 | mutex_init(&ls->ls_cb_mutex); |
532 | INIT_LIST_HEAD(&ls->ls_cb_delay); | |
533 | ||
e7fd4179 | 534 | ls->ls_recoverd_task = NULL; |
90135925 | 535 | mutex_init(&ls->ls_recoverd_active); |
e7fd4179 | 536 | spin_lock_init(&ls->ls_recover_lock); |
98f176fb DT |
537 | spin_lock_init(&ls->ls_rcom_spin); |
538 | get_random_bytes(&ls->ls_rcom_seq, sizeof(uint64_t)); | |
e7fd4179 DT |
539 | ls->ls_recover_status = 0; |
540 | ls->ls_recover_seq = 0; | |
541 | ls->ls_recover_args = NULL; | |
542 | init_rwsem(&ls->ls_in_recovery); | |
c36258b5 | 543 | init_rwsem(&ls->ls_recv_active); |
e7fd4179 | 544 | INIT_LIST_HEAD(&ls->ls_requestqueue); |
90135925 | 545 | mutex_init(&ls->ls_requestqueue_mutex); |
597d0cae | 546 | mutex_init(&ls->ls_clear_proc_locks); |
e7fd4179 | 547 | |
573c24c4 | 548 | ls->ls_recover_buf = kmalloc(dlm_config.ci_buffer_size, GFP_NOFS); |
e7fd4179 DT |
549 | if (!ls->ls_recover_buf) |
550 | goto out_dirfree; | |
551 | ||
757a4271 DT |
552 | ls->ls_slot = 0; |
553 | ls->ls_num_slots = 0; | |
554 | ls->ls_slots_size = 0; | |
555 | ls->ls_slots = NULL; | |
556 | ||
e7fd4179 DT |
557 | INIT_LIST_HEAD(&ls->ls_recover_list); |
558 | spin_lock_init(&ls->ls_recover_list_lock); | |
559 | ls->ls_recover_list_count = 0; | |
597d0cae | 560 | ls->ls_local_handle = ls; |
e7fd4179 DT |
561 | init_waitqueue_head(&ls->ls_wait_general); |
562 | INIT_LIST_HEAD(&ls->ls_root_list); | |
563 | init_rwsem(&ls->ls_root_sem); | |
564 | ||
565 | down_write(&ls->ls_in_recovery); | |
566 | ||
5f88f1ea | 567 | spin_lock(&lslist_lock); |
0f8e0d9a | 568 | ls->ls_create_count = 1; |
5f88f1ea DT |
569 | list_add(&ls->ls_list, &lslist); |
570 | spin_unlock(&lslist_lock); | |
571 | ||
23e8e1aa DT |
572 | if (flags & DLM_LSFL_FS) { |
573 | error = dlm_callback_start(ls); | |
574 | if (error) { | |
575 | log_error(ls, "can't start dlm_callback %d", error); | |
576 | goto out_delist; | |
577 | } | |
578 | } | |
579 | ||
5f88f1ea | 580 | /* needs to find ls in lslist */ |
e7fd4179 DT |
581 | error = dlm_recoverd_start(ls); |
582 | if (error) { | |
583 | log_error(ls, "can't start dlm_recoverd %d", error); | |
23e8e1aa | 584 | goto out_callback; |
e7fd4179 DT |
585 | } |
586 | ||
901195ed GKH |
587 | ls->ls_kobj.kset = dlm_kset; |
588 | error = kobject_init_and_add(&ls->ls_kobj, &dlm_ktype, NULL, | |
589 | "%s", ls->ls_name); | |
e7fd4179 | 590 | if (error) |
23e8e1aa | 591 | goto out_recoverd; |
901195ed | 592 | kobject_uevent(&ls->ls_kobj, KOBJ_ADD); |
79d72b54 DT |
593 | |
594 | /* let kobject handle freeing of ls if there's an error */ | |
595 | do_unreg = 1; | |
e7fd4179 | 596 | |
8b0e7b2c DT |
597 | /* This uevent triggers dlm_controld in userspace to add us to the |
598 | group of nodes that are members of this lockspace (managed by the | |
599 | cluster infrastructure.) Once it's done that, it tells us who the | |
600 | current lockspace members are (via configfs) and then tells the | |
601 | lockspace to start running (via sysfs) in dlm_ls_start(). */ | |
602 | ||
e7fd4179 DT |
603 | error = do_uevent(ls, 1); |
604 | if (error) | |
23e8e1aa | 605 | goto out_recoverd; |
79d72b54 | 606 | |
8b0e7b2c DT |
607 | wait_for_completion(&ls->ls_members_done); |
608 | error = ls->ls_members_result; | |
609 | if (error) | |
610 | goto out_members; | |
611 | ||
79d72b54 DT |
612 | dlm_create_debug_file(ls); |
613 | ||
614 | log_debug(ls, "join complete"); | |
e7fd4179 DT |
615 | *lockspace = ls; |
616 | return 0; | |
617 | ||
8b0e7b2c DT |
618 | out_members: |
619 | do_uevent(ls, 0); | |
620 | dlm_clear_members(ls); | |
621 | kfree(ls->ls_node_array); | |
23e8e1aa | 622 | out_recoverd: |
5f88f1ea | 623 | dlm_recoverd_stop(ls); |
23e8e1aa DT |
624 | out_callback: |
625 | dlm_callback_stop(ls); | |
79d72b54 | 626 | out_delist: |
e7fd4179 DT |
627 | spin_lock(&lslist_lock); |
628 | list_del(&ls->ls_list); | |
629 | spin_unlock(&lslist_lock); | |
e7fd4179 DT |
630 | kfree(ls->ls_recover_buf); |
631 | out_dirfree: | |
c282af49 | 632 | vfree(ls->ls_dirtbl); |
e7fd4179 | 633 | out_lkbfree: |
3d6aa675 | 634 | idr_destroy(&ls->ls_lkbidr); |
c282af49 | 635 | vfree(ls->ls_rsbtbl); |
e7fd4179 | 636 | out_lsfree: |
79d72b54 | 637 | if (do_unreg) |
197b12d6 | 638 | kobject_put(&ls->ls_kobj); |
79d72b54 DT |
639 | else |
640 | kfree(ls); | |
e7fd4179 DT |
641 | out: |
642 | module_put(THIS_MODULE); | |
643 | return error; | |
644 | } | |
645 | ||
60f98d18 DT |
646 | int dlm_new_lockspace(const char *name, const char *cluster, |
647 | uint32_t flags, int lvblen, | |
648 | const struct dlm_lockspace_ops *ops, void *ops_arg, | |
649 | int *ops_result, dlm_lockspace_t **lockspace) | |
e7fd4179 DT |
650 | { |
651 | int error = 0; | |
652 | ||
90135925 | 653 | mutex_lock(&ls_lock); |
e7fd4179 DT |
654 | if (!ls_count) |
655 | error = threads_start(); | |
656 | if (error) | |
657 | goto out; | |
658 | ||
60f98d18 DT |
659 | error = new_lockspace(name, cluster, flags, lvblen, ops, ops_arg, |
660 | ops_result, lockspace); | |
e7fd4179 DT |
661 | if (!error) |
662 | ls_count++; | |
8511a272 DT |
663 | if (error > 0) |
664 | error = 0; | |
665 | if (!ls_count) | |
8b0e7b2c | 666 | threads_stop(); |
e7fd4179 | 667 | out: |
90135925 | 668 | mutex_unlock(&ls_lock); |
e7fd4179 DT |
669 | return error; |
670 | } | |
671 | ||
3d6aa675 | 672 | static int lkb_idr_is_local(int id, void *p, void *data) |
e7fd4179 | 673 | { |
3d6aa675 DT |
674 | struct dlm_lkb *lkb = p; |
675 | ||
676 | if (!lkb->lkb_nodeid) | |
677 | return 1; | |
678 | return 0; | |
679 | } | |
680 | ||
681 | static int lkb_idr_is_any(int id, void *p, void *data) | |
682 | { | |
683 | return 1; | |
684 | } | |
685 | ||
686 | static int lkb_idr_free(int id, void *p, void *data) | |
687 | { | |
688 | struct dlm_lkb *lkb = p; | |
689 | ||
3d6aa675 DT |
690 | if (lkb->lkb_lvbptr && lkb->lkb_flags & DLM_IFL_MSTCPY) |
691 | dlm_free_lvb(lkb->lkb_lvbptr); | |
692 | ||
693 | dlm_free_lkb(lkb); | |
694 | return 0; | |
695 | } | |
696 | ||
697 | /* NOTE: We check the lkbidr here rather than the resource table. | |
698 | This is because there may be LKBs queued as ASTs that have been unlinked | |
699 | from their RSBs and are pending deletion once the AST has been delivered */ | |
700 | ||
701 | static int lockspace_busy(struct dlm_ls *ls, int force) | |
702 | { | |
703 | int rv; | |
704 | ||
705 | spin_lock(&ls->ls_lkbidr_spin); | |
706 | if (force == 0) { | |
707 | rv = idr_for_each(&ls->ls_lkbidr, lkb_idr_is_any, ls); | |
708 | } else if (force == 1) { | |
709 | rv = idr_for_each(&ls->ls_lkbidr, lkb_idr_is_local, ls); | |
710 | } else { | |
711 | rv = 0; | |
e7fd4179 | 712 | } |
3d6aa675 DT |
713 | spin_unlock(&ls->ls_lkbidr_spin); |
714 | return rv; | |
e7fd4179 DT |
715 | } |
716 | ||
717 | static int release_lockspace(struct dlm_ls *ls, int force) | |
718 | { | |
e7fd4179 | 719 | struct dlm_rsb *rsb; |
9beb3bf5 | 720 | struct rb_node *n; |
0f8e0d9a DT |
721 | int i, busy, rv; |
722 | ||
3d6aa675 | 723 | busy = lockspace_busy(ls, force); |
0f8e0d9a DT |
724 | |
725 | spin_lock(&lslist_lock); | |
726 | if (ls->ls_create_count == 1) { | |
3d6aa675 | 727 | if (busy) { |
0f8e0d9a | 728 | rv = -EBUSY; |
3d6aa675 | 729 | } else { |
0f8e0d9a DT |
730 | /* remove_lockspace takes ls off lslist */ |
731 | ls->ls_create_count = 0; | |
732 | rv = 0; | |
733 | } | |
734 | } else if (ls->ls_create_count > 1) { | |
735 | rv = --ls->ls_create_count; | |
736 | } else { | |
737 | rv = -EINVAL; | |
738 | } | |
739 | spin_unlock(&lslist_lock); | |
740 | ||
741 | if (rv) { | |
742 | log_debug(ls, "release_lockspace no remove %d", rv); | |
743 | return rv; | |
744 | } | |
e7fd4179 | 745 | |
0f8e0d9a | 746 | dlm_device_deregister(ls); |
e7fd4179 | 747 | |
dc68c7ed | 748 | if (force < 3 && dlm_user_daemon_available()) |
e7fd4179 DT |
749 | do_uevent(ls, 0); |
750 | ||
751 | dlm_recoverd_stop(ls); | |
752 | ||
23e8e1aa DT |
753 | dlm_callback_stop(ls); |
754 | ||
e7fd4179 DT |
755 | remove_lockspace(ls); |
756 | ||
757 | dlm_delete_debug_file(ls); | |
758 | ||
e7fd4179 DT |
759 | kfree(ls->ls_recover_buf); |
760 | ||
761 | /* | |
762 | * Free direntry structs. | |
763 | */ | |
764 | ||
765 | dlm_dir_clear(ls); | |
c282af49 | 766 | vfree(ls->ls_dirtbl); |
e7fd4179 DT |
767 | |
768 | /* | |
3d6aa675 | 769 | * Free all lkb's in idr |
e7fd4179 DT |
770 | */ |
771 | ||
3d6aa675 DT |
772 | idr_for_each(&ls->ls_lkbidr, lkb_idr_free, ls); |
773 | idr_remove_all(&ls->ls_lkbidr); | |
774 | idr_destroy(&ls->ls_lkbidr); | |
e7fd4179 | 775 | |
e7fd4179 DT |
776 | /* |
777 | * Free all rsb's on rsbtbl[] lists | |
778 | */ | |
779 | ||
780 | for (i = 0; i < ls->ls_rsbtbl_size; i++) { | |
9beb3bf5 BP |
781 | while ((n = rb_first(&ls->ls_rsbtbl[i].keep))) { |
782 | rsb = rb_entry(n, struct dlm_rsb, res_hashnode); | |
783 | rb_erase(n, &ls->ls_rsbtbl[i].keep); | |
52bda2b5 | 784 | dlm_free_rsb(rsb); |
e7fd4179 DT |
785 | } |
786 | ||
9beb3bf5 BP |
787 | while ((n = rb_first(&ls->ls_rsbtbl[i].toss))) { |
788 | rsb = rb_entry(n, struct dlm_rsb, res_hashnode); | |
789 | rb_erase(n, &ls->ls_rsbtbl[i].toss); | |
52bda2b5 | 790 | dlm_free_rsb(rsb); |
e7fd4179 DT |
791 | } |
792 | } | |
793 | ||
c282af49 | 794 | vfree(ls->ls_rsbtbl); |
e7fd4179 | 795 | |
3881ac04 DT |
796 | while (!list_empty(&ls->ls_new_rsb)) { |
797 | rsb = list_first_entry(&ls->ls_new_rsb, struct dlm_rsb, | |
798 | res_hashchain); | |
799 | list_del(&rsb->res_hashchain); | |
800 | dlm_free_rsb(rsb); | |
801 | } | |
802 | ||
e7fd4179 DT |
803 | /* |
804 | * Free structures on any other lists | |
805 | */ | |
806 | ||
2896ee37 | 807 | dlm_purge_requestqueue(ls); |
e7fd4179 DT |
808 | kfree(ls->ls_recover_args); |
809 | dlm_clear_free_entries(ls); | |
810 | dlm_clear_members(ls); | |
811 | dlm_clear_members_gone(ls); | |
812 | kfree(ls->ls_node_array); | |
0f8e0d9a | 813 | log_debug(ls, "release_lockspace final free"); |
197b12d6 | 814 | kobject_put(&ls->ls_kobj); |
79d72b54 | 815 | /* The ls structure will be freed when the kobject is done with */ |
e7fd4179 | 816 | |
e7fd4179 DT |
817 | module_put(THIS_MODULE); |
818 | return 0; | |
819 | } | |
820 | ||
821 | /* | |
822 | * Called when a system has released all its locks and is not going to use the | |
823 | * lockspace any longer. We free everything we're managing for this lockspace. | |
824 | * Remaining nodes will go through the recovery process as if we'd died. The | |
825 | * lockspace must continue to function as usual, participating in recoveries, | |
826 | * until this returns. | |
827 | * | |
828 | * Force has 4 possible values: | |
829 | * 0 - don't destroy locksapce if it has any LKBs | |
830 | * 1 - destroy lockspace if it has remote LKBs but not if it has local LKBs | |
831 | * 2 - destroy lockspace regardless of LKBs | |
832 | * 3 - destroy lockspace as part of a forced shutdown | |
833 | */ | |
834 | ||
835 | int dlm_release_lockspace(void *lockspace, int force) | |
836 | { | |
837 | struct dlm_ls *ls; | |
0f8e0d9a | 838 | int error; |
e7fd4179 DT |
839 | |
840 | ls = dlm_find_lockspace_local(lockspace); | |
841 | if (!ls) | |
842 | return -EINVAL; | |
843 | dlm_put_lockspace(ls); | |
0f8e0d9a DT |
844 | |
845 | mutex_lock(&ls_lock); | |
846 | error = release_lockspace(ls, force); | |
847 | if (!error) | |
848 | ls_count--; | |
278afcbf | 849 | if (!ls_count) |
0f8e0d9a DT |
850 | threads_stop(); |
851 | mutex_unlock(&ls_lock); | |
852 | ||
853 | return error; | |
e7fd4179 DT |
854 | } |
855 | ||
dc68c7ed DT |
856 | void dlm_stop_lockspaces(void) |
857 | { | |
858 | struct dlm_ls *ls; | |
859 | ||
860 | restart: | |
861 | spin_lock(&lslist_lock); | |
862 | list_for_each_entry(ls, &lslist, ls_list) { | |
863 | if (!test_bit(LSFL_RUNNING, &ls->ls_flags)) | |
864 | continue; | |
865 | spin_unlock(&lslist_lock); | |
866 | log_error(ls, "no userland control daemon, stopping lockspace"); | |
867 | dlm_ls_stop(ls); | |
868 | goto restart; | |
869 | } | |
870 | spin_unlock(&lslist_lock); | |
871 | } | |
872 |