Commit | Line | Data |
---|---|---|
2522fe45 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
e7fd4179 DT |
2 | /****************************************************************************** |
3 | ******************************************************************************* | |
4 | ** | |
5 | ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
60f98d18 | 6 | ** Copyright (C) 2004-2011 Red Hat, Inc. All rights reserved. |
e7fd4179 | 7 | ** |
e7fd4179 DT |
8 | ** |
9 | ******************************************************************************* | |
10 | ******************************************************************************/ | |
11 | ||
7963b8a5 PG |
12 | #include <linux/module.h> |
13 | ||
e7fd4179 DT |
14 | #include "dlm_internal.h" |
15 | #include "lockspace.h" | |
16 | #include "member.h" | |
17 | #include "recoverd.h" | |
e7fd4179 | 18 | #include "dir.h" |
a070a91c | 19 | #include "midcomms.h" |
e7fd4179 DT |
20 | #include "lowcomms.h" |
21 | #include "config.h" | |
22 | #include "memory.h" | |
23 | #include "lock.h" | |
c56b39cd | 24 | #include "recover.h" |
2896ee37 | 25 | #include "requestqueue.h" |
0f8e0d9a | 26 | #include "user.h" |
23e8e1aa | 27 | #include "ast.h" |
e7fd4179 | 28 | |
e7fd4179 | 29 | static int ls_count; |
90135925 | 30 | static struct mutex ls_lock; |
e7fd4179 DT |
31 | static struct list_head lslist; |
32 | static spinlock_t lslist_lock; | |
33 | static struct task_struct * scand_task; | |
34 | ||
35 | ||
36 | static ssize_t dlm_control_store(struct dlm_ls *ls, const char *buf, size_t len) | |
37 | { | |
38 | ssize_t ret = len; | |
6edb5687 FF |
39 | int n; |
40 | int rc = kstrtoint(buf, 0, &n); | |
e7fd4179 | 41 | |
6edb5687 FF |
42 | if (rc) |
43 | return rc; | |
e2de7f56 PC |
44 | ls = dlm_find_lockspace_local(ls->ls_local_handle); |
45 | if (!ls) | |
46 | return -EINVAL; | |
47 | ||
e7fd4179 DT |
48 | switch (n) { |
49 | case 0: | |
50 | dlm_ls_stop(ls); | |
51 | break; | |
52 | case 1: | |
53 | dlm_ls_start(ls); | |
54 | break; | |
55 | default: | |
56 | ret = -EINVAL; | |
57 | } | |
e2de7f56 | 58 | dlm_put_lockspace(ls); |
e7fd4179 DT |
59 | return ret; |
60 | } | |
61 | ||
62 | static ssize_t dlm_event_store(struct dlm_ls *ls, const char *buf, size_t len) | |
63 | { | |
6edb5687 FF |
64 | int rc = kstrtoint(buf, 0, &ls->ls_uevent_result); |
65 | ||
66 | if (rc) | |
67 | return rc; | |
e7fd4179 DT |
68 | set_bit(LSFL_UEVENT_WAIT, &ls->ls_flags); |
69 | wake_up(&ls->ls_uevent_wait); | |
70 | return len; | |
71 | } | |
72 | ||
73 | static ssize_t dlm_id_show(struct dlm_ls *ls, char *buf) | |
74 | { | |
a1d144c7 | 75 | return snprintf(buf, PAGE_SIZE, "%u\n", ls->ls_global_id); |
e7fd4179 DT |
76 | } |
77 | ||
78 | static ssize_t dlm_id_store(struct dlm_ls *ls, const char *buf, size_t len) | |
79 | { | |
6edb5687 FF |
80 | int rc = kstrtouint(buf, 0, &ls->ls_global_id); |
81 | ||
82 | if (rc) | |
83 | return rc; | |
e7fd4179 DT |
84 | return len; |
85 | } | |
86 | ||
4875647a DT |
87 | static ssize_t dlm_nodir_show(struct dlm_ls *ls, char *buf) |
88 | { | |
89 | return snprintf(buf, PAGE_SIZE, "%u\n", dlm_no_directory(ls)); | |
90 | } | |
91 | ||
92 | static ssize_t dlm_nodir_store(struct dlm_ls *ls, const char *buf, size_t len) | |
93 | { | |
6edb5687 FF |
94 | int val; |
95 | int rc = kstrtoint(buf, 0, &val); | |
96 | ||
97 | if (rc) | |
98 | return rc; | |
4875647a DT |
99 | if (val == 1) |
100 | set_bit(LSFL_NODIR, &ls->ls_flags); | |
101 | return len; | |
102 | } | |
103 | ||
c56b39cd DT |
104 | static ssize_t dlm_recover_status_show(struct dlm_ls *ls, char *buf) |
105 | { | |
106 | uint32_t status = dlm_recover_status(ls); | |
a1d144c7 | 107 | return snprintf(buf, PAGE_SIZE, "%x\n", status); |
c56b39cd DT |
108 | } |
109 | ||
faa0f267 DT |
110 | static ssize_t dlm_recover_nodeid_show(struct dlm_ls *ls, char *buf) |
111 | { | |
a1d144c7 | 112 | return snprintf(buf, PAGE_SIZE, "%d\n", ls->ls_recover_nodeid); |
faa0f267 DT |
113 | } |
114 | ||
e7fd4179 DT |
115 | struct dlm_attr { |
116 | struct attribute attr; | |
117 | ssize_t (*show)(struct dlm_ls *, char *); | |
118 | ssize_t (*store)(struct dlm_ls *, const char *, size_t); | |
119 | }; | |
120 | ||
121 | static struct dlm_attr dlm_attr_control = { | |
122 | .attr = {.name = "control", .mode = S_IWUSR}, | |
123 | .store = dlm_control_store | |
124 | }; | |
125 | ||
126 | static struct dlm_attr dlm_attr_event = { | |
127 | .attr = {.name = "event_done", .mode = S_IWUSR}, | |
128 | .store = dlm_event_store | |
129 | }; | |
130 | ||
131 | static struct dlm_attr dlm_attr_id = { | |
132 | .attr = {.name = "id", .mode = S_IRUGO | S_IWUSR}, | |
133 | .show = dlm_id_show, | |
134 | .store = dlm_id_store | |
135 | }; | |
136 | ||
4875647a DT |
137 | static struct dlm_attr dlm_attr_nodir = { |
138 | .attr = {.name = "nodir", .mode = S_IRUGO | S_IWUSR}, | |
139 | .show = dlm_nodir_show, | |
140 | .store = dlm_nodir_store | |
141 | }; | |
142 | ||
c56b39cd DT |
143 | static struct dlm_attr dlm_attr_recover_status = { |
144 | .attr = {.name = "recover_status", .mode = S_IRUGO}, | |
145 | .show = dlm_recover_status_show | |
146 | }; | |
147 | ||
faa0f267 DT |
148 | static struct dlm_attr dlm_attr_recover_nodeid = { |
149 | .attr = {.name = "recover_nodeid", .mode = S_IRUGO}, | |
150 | .show = dlm_recover_nodeid_show | |
151 | }; | |
152 | ||
e7fd4179 DT |
153 | static struct attribute *dlm_attrs[] = { |
154 | &dlm_attr_control.attr, | |
155 | &dlm_attr_event.attr, | |
156 | &dlm_attr_id.attr, | |
4875647a | 157 | &dlm_attr_nodir.attr, |
c56b39cd | 158 | &dlm_attr_recover_status.attr, |
faa0f267 | 159 | &dlm_attr_recover_nodeid.attr, |
e7fd4179 DT |
160 | NULL, |
161 | }; | |
c9c5b5e1 | 162 | ATTRIBUTE_GROUPS(dlm); |
e7fd4179 DT |
163 | |
164 | static ssize_t dlm_attr_show(struct kobject *kobj, struct attribute *attr, | |
165 | char *buf) | |
166 | { | |
167 | struct dlm_ls *ls = container_of(kobj, struct dlm_ls, ls_kobj); | |
168 | struct dlm_attr *a = container_of(attr, struct dlm_attr, attr); | |
169 | return a->show ? a->show(ls, buf) : 0; | |
170 | } | |
171 | ||
172 | static ssize_t dlm_attr_store(struct kobject *kobj, struct attribute *attr, | |
173 | const char *buf, size_t len) | |
174 | { | |
175 | struct dlm_ls *ls = container_of(kobj, struct dlm_ls, ls_kobj); | |
176 | struct dlm_attr *a = container_of(attr, struct dlm_attr, attr); | |
177 | return a->store ? a->store(ls, buf, len) : len; | |
178 | } | |
179 | ||
ba542e3b PC |
180 | static void lockspace_kobj_release(struct kobject *k) |
181 | { | |
182 | struct dlm_ls *ls = container_of(k, struct dlm_ls, ls_kobj); | |
183 | kfree(ls); | |
184 | } | |
185 | ||
52cf25d0 | 186 | static const struct sysfs_ops dlm_attr_ops = { |
e7fd4179 DT |
187 | .show = dlm_attr_show, |
188 | .store = dlm_attr_store, | |
189 | }; | |
190 | ||
191 | static struct kobj_type dlm_ktype = { | |
c9c5b5e1 | 192 | .default_groups = dlm_groups, |
e7fd4179 | 193 | .sysfs_ops = &dlm_attr_ops, |
ba542e3b | 194 | .release = lockspace_kobj_release, |
e7fd4179 DT |
195 | }; |
196 | ||
d405936b | 197 | static struct kset *dlm_kset; |
e7fd4179 | 198 | |
e7fd4179 DT |
199 | static int do_uevent(struct dlm_ls *ls, int in) |
200 | { | |
e7fd4179 DT |
201 | if (in) |
202 | kobject_uevent(&ls->ls_kobj, KOBJ_ONLINE); | |
203 | else | |
204 | kobject_uevent(&ls->ls_kobj, KOBJ_OFFLINE); | |
205 | ||
075f0177 | 206 | log_rinfo(ls, "%s the lockspace group...", in ? "joining" : "leaving"); |
8b0e7b2c DT |
207 | |
208 | /* dlm_controld will see the uevent, do the necessary group management | |
209 | and then write to sysfs to wake us */ | |
210 | ||
f084a4f4 RL |
211 | wait_event(ls->ls_uevent_wait, |
212 | test_and_clear_bit(LSFL_UEVENT_WAIT, &ls->ls_flags)); | |
8b0e7b2c | 213 | |
f084a4f4 | 214 | log_rinfo(ls, "group event done %d", ls->ls_uevent_result); |
e7fd4179 | 215 | |
f084a4f4 | 216 | return ls->ls_uevent_result; |
e7fd4179 DT |
217 | } |
218 | ||
cf6299b6 | 219 | static int dlm_uevent(struct kobject *kobj, struct kobj_uevent_env *env) |
b4a5d4bc SW |
220 | { |
221 | struct dlm_ls *ls = container_of(kobj, struct dlm_ls, ls_kobj); | |
222 | ||
223 | add_uevent_var(env, "LOCKSPACE=%s", ls->ls_name); | |
224 | return 0; | |
225 | } | |
226 | ||
417f7c59 | 227 | static const struct kset_uevent_ops dlm_uevent_ops = { |
b4a5d4bc SW |
228 | .uevent = dlm_uevent, |
229 | }; | |
e7fd4179 | 230 | |
30727174 | 231 | int __init dlm_lockspace_init(void) |
e7fd4179 | 232 | { |
e7fd4179 | 233 | ls_count = 0; |
90135925 | 234 | mutex_init(&ls_lock); |
e7fd4179 DT |
235 | INIT_LIST_HEAD(&lslist); |
236 | spin_lock_init(&lslist_lock); | |
237 | ||
b4a5d4bc | 238 | dlm_kset = kset_create_and_add("dlm", &dlm_uevent_ops, kernel_kobj); |
d405936b | 239 | if (!dlm_kset) { |
8e24eea7 | 240 | printk(KERN_WARNING "%s: can not create kset\n", __func__); |
d405936b GKH |
241 | return -ENOMEM; |
242 | } | |
243 | return 0; | |
e7fd4179 DT |
244 | } |
245 | ||
246 | void dlm_lockspace_exit(void) | |
247 | { | |
d405936b | 248 | kset_unregister(dlm_kset); |
e7fd4179 DT |
249 | } |
250 | ||
c1dcf65f DT |
251 | static struct dlm_ls *find_ls_to_scan(void) |
252 | { | |
253 | struct dlm_ls *ls; | |
254 | ||
255 | spin_lock(&lslist_lock); | |
256 | list_for_each_entry(ls, &lslist, ls_list) { | |
257 | if (time_after_eq(jiffies, ls->ls_scan_time + | |
258 | dlm_config.ci_scan_secs * HZ)) { | |
259 | spin_unlock(&lslist_lock); | |
260 | return ls; | |
261 | } | |
262 | } | |
263 | spin_unlock(&lslist_lock); | |
264 | return NULL; | |
265 | } | |
266 | ||
e7fd4179 DT |
267 | static int dlm_scand(void *data) |
268 | { | |
269 | struct dlm_ls *ls; | |
270 | ||
271 | while (!kthread_should_stop()) { | |
c1dcf65f DT |
272 | ls = find_ls_to_scan(); |
273 | if (ls) { | |
85e86edf | 274 | if (dlm_lock_recovery_try(ls)) { |
c1dcf65f | 275 | ls->ls_scan_time = jiffies; |
85e86edf | 276 | dlm_scan_rsbs(ls); |
3ae1acf9 | 277 | dlm_scan_timeout(ls); |
85e86edf | 278 | dlm_unlock_recovery(ls); |
c1dcf65f DT |
279 | } else { |
280 | ls->ls_scan_time += HZ; | |
85e86edf | 281 | } |
c6ff669b | 282 | continue; |
85e86edf | 283 | } |
c6ff669b | 284 | schedule_timeout_interruptible(dlm_config.ci_scan_secs * HZ); |
e7fd4179 DT |
285 | } |
286 | return 0; | |
287 | } | |
288 | ||
289 | static int dlm_scand_start(void) | |
290 | { | |
291 | struct task_struct *p; | |
292 | int error = 0; | |
293 | ||
294 | p = kthread_run(dlm_scand, NULL, "dlm_scand"); | |
295 | if (IS_ERR(p)) | |
296 | error = PTR_ERR(p); | |
297 | else | |
298 | scand_task = p; | |
299 | return error; | |
300 | } | |
301 | ||
302 | static void dlm_scand_stop(void) | |
303 | { | |
304 | kthread_stop(scand_task); | |
305 | } | |
306 | ||
e7fd4179 DT |
307 | struct dlm_ls *dlm_find_lockspace_global(uint32_t id) |
308 | { | |
309 | struct dlm_ls *ls; | |
310 | ||
311 | spin_lock(&lslist_lock); | |
312 | ||
313 | list_for_each_entry(ls, &lslist, ls_list) { | |
314 | if (ls->ls_global_id == id) { | |
3cb5977c | 315 | atomic_inc(&ls->ls_count); |
e7fd4179 DT |
316 | goto out; |
317 | } | |
318 | } | |
319 | ls = NULL; | |
320 | out: | |
321 | spin_unlock(&lslist_lock); | |
322 | return ls; | |
323 | } | |
324 | ||
597d0cae | 325 | struct dlm_ls *dlm_find_lockspace_local(dlm_lockspace_t *lockspace) |
e7fd4179 | 326 | { |
597d0cae | 327 | struct dlm_ls *ls; |
e7fd4179 DT |
328 | |
329 | spin_lock(&lslist_lock); | |
597d0cae DT |
330 | list_for_each_entry(ls, &lslist, ls_list) { |
331 | if (ls->ls_local_handle == lockspace) { | |
3cb5977c | 332 | atomic_inc(&ls->ls_count); |
597d0cae DT |
333 | goto out; |
334 | } | |
335 | } | |
336 | ls = NULL; | |
337 | out: | |
338 | spin_unlock(&lslist_lock); | |
339 | return ls; | |
340 | } | |
341 | ||
342 | struct dlm_ls *dlm_find_lockspace_device(int minor) | |
343 | { | |
344 | struct dlm_ls *ls; | |
345 | ||
346 | spin_lock(&lslist_lock); | |
347 | list_for_each_entry(ls, &lslist, ls_list) { | |
348 | if (ls->ls_device.minor == minor) { | |
3cb5977c | 349 | atomic_inc(&ls->ls_count); |
597d0cae DT |
350 | goto out; |
351 | } | |
352 | } | |
353 | ls = NULL; | |
354 | out: | |
e7fd4179 DT |
355 | spin_unlock(&lslist_lock); |
356 | return ls; | |
357 | } | |
358 | ||
359 | void dlm_put_lockspace(struct dlm_ls *ls) | |
360 | { | |
3cb5977c AA |
361 | if (atomic_dec_and_test(&ls->ls_count)) |
362 | wake_up(&ls->ls_count_wait); | |
e7fd4179 DT |
363 | } |
364 | ||
365 | static void remove_lockspace(struct dlm_ls *ls) | |
366 | { | |
3cb5977c AA |
367 | retry: |
368 | wait_event(ls->ls_count_wait, atomic_read(&ls->ls_count) == 0); | |
369 | ||
370 | spin_lock(&lslist_lock); | |
371 | if (atomic_read(&ls->ls_count) != 0) { | |
e7fd4179 | 372 | spin_unlock(&lslist_lock); |
3cb5977c | 373 | goto retry; |
e7fd4179 | 374 | } |
3cb5977c AA |
375 | |
376 | WARN_ON(ls->ls_create_count != 0); | |
377 | list_del(&ls->ls_list); | |
378 | spin_unlock(&lslist_lock); | |
e7fd4179 DT |
379 | } |
380 | ||
381 | static int threads_start(void) | |
382 | { | |
383 | int error; | |
384 | ||
e7fd4179 DT |
385 | error = dlm_scand_start(); |
386 | if (error) { | |
387 | log_print("cannot start dlm_scand thread %d", error); | |
23e8e1aa | 388 | goto fail; |
e7fd4179 DT |
389 | } |
390 | ||
391 | /* Thread for sending/receiving messages for all lockspace's */ | |
a070a91c | 392 | error = dlm_midcomms_start(); |
e7fd4179 DT |
393 | if (error) { |
394 | log_print("cannot start dlm lowcomms %d", error); | |
395 | goto scand_fail; | |
396 | } | |
397 | ||
398 | return 0; | |
399 | ||
400 | scand_fail: | |
401 | dlm_scand_stop(); | |
e7fd4179 DT |
402 | fail: |
403 | return error; | |
404 | } | |
405 | ||
60f98d18 DT |
406 | static int new_lockspace(const char *name, const char *cluster, |
407 | uint32_t flags, int lvblen, | |
408 | const struct dlm_lockspace_ops *ops, void *ops_arg, | |
409 | int *ops_result, dlm_lockspace_t **lockspace) | |
e7fd4179 DT |
410 | { |
411 | struct dlm_ls *ls; | |
0f8e0d9a | 412 | int i, size, error; |
79d72b54 | 413 | int do_unreg = 0; |
60f98d18 | 414 | int namelen = strlen(name); |
e7fd4179 | 415 | |
3f0806d2 | 416 | if (namelen > DLM_LOCKSPACE_LEN || namelen == 0) |
e7fd4179 DT |
417 | return -EINVAL; |
418 | ||
b5c9d37c | 419 | if (lvblen % 8) |
e7fd4179 DT |
420 | return -EINVAL; |
421 | ||
422 | if (!try_module_get(THIS_MODULE)) | |
423 | return -EINVAL; | |
424 | ||
dc68c7ed | 425 | if (!dlm_user_daemon_available()) { |
60f98d18 DT |
426 | log_print("dlm user daemon not available"); |
427 | error = -EUNATCH; | |
428 | goto out; | |
429 | } | |
430 | ||
431 | if (ops && ops_result) { | |
432 | if (!dlm_config.ci_recover_callbacks) | |
433 | *ops_result = -EOPNOTSUPP; | |
434 | else | |
435 | *ops_result = 0; | |
436 | } | |
437 | ||
3b0e761b ZL |
438 | if (!cluster) |
439 | log_print("dlm cluster name '%s' is being used without an application provided cluster name", | |
440 | dlm_config.ci_cluster_name); | |
441 | ||
60f98d18 DT |
442 | if (dlm_config.ci_recover_callbacks && cluster && |
443 | strncmp(cluster, dlm_config.ci_cluster_name, DLM_LOCKSPACE_LEN)) { | |
8e174374 GH |
444 | log_print("dlm cluster name '%s' does not match " |
445 | "the application cluster name '%s'", | |
60f98d18 DT |
446 | dlm_config.ci_cluster_name, cluster); |
447 | error = -EBADR; | |
448 | goto out; | |
dc68c7ed DT |
449 | } |
450 | ||
0f8e0d9a DT |
451 | error = 0; |
452 | ||
453 | spin_lock(&lslist_lock); | |
454 | list_for_each_entry(ls, &lslist, ls_list) { | |
455 | WARN_ON(ls->ls_create_count <= 0); | |
456 | if (ls->ls_namelen != namelen) | |
457 | continue; | |
458 | if (memcmp(ls->ls_name, name, namelen)) | |
459 | continue; | |
460 | if (flags & DLM_LSFL_NEWEXCL) { | |
461 | error = -EEXIST; | |
462 | break; | |
463 | } | |
464 | ls->ls_create_count++; | |
8511a272 DT |
465 | *lockspace = ls; |
466 | error = 1; | |
0f8e0d9a | 467 | break; |
e7fd4179 | 468 | } |
0f8e0d9a DT |
469 | spin_unlock(&lslist_lock); |
470 | ||
0f8e0d9a | 471 | if (error) |
8511a272 | 472 | goto out; |
0f8e0d9a DT |
473 | |
474 | error = -ENOMEM; | |
e7fd4179 | 475 | |
d96d0f96 | 476 | ls = kzalloc(sizeof(*ls), GFP_NOFS); |
e7fd4179 DT |
477 | if (!ls) |
478 | goto out; | |
e7fd4179 DT |
479 | memcpy(ls->ls_name, name, namelen); |
480 | ls->ls_namelen = namelen; | |
e7fd4179 | 481 | ls->ls_lvblen = lvblen; |
3cb5977c AA |
482 | atomic_set(&ls->ls_count, 0); |
483 | init_waitqueue_head(&ls->ls_count_wait); | |
e7fd4179 | 484 | ls->ls_flags = 0; |
c1dcf65f | 485 | ls->ls_scan_time = jiffies; |
e7fd4179 | 486 | |
60f98d18 DT |
487 | if (ops && dlm_config.ci_recover_callbacks) { |
488 | ls->ls_ops = ops; | |
489 | ls->ls_ops_arg = ops_arg; | |
490 | } | |
491 | ||
81eeb82f | 492 | #ifdef CONFIG_DLM_DEPRECATED_API |
6b0afc0c | 493 | if (flags & DLM_LSFL_TIMEWARN) { |
81eeb82f AA |
494 | pr_warn_once("===============================================================\n" |
495 | "WARNING: the dlm DLM_LSFL_TIMEWARN flag is being deprecated and\n" | |
496 | " will be removed in v6.2!\n" | |
497 | " Inclusive DLM_LSFL_TIMEWARN define in UAPI header!\n" | |
498 | "===============================================================\n"); | |
81eeb82f | 499 | |
3ae1acf9 | 500 | set_bit(LSFL_TIMEWARN, &ls->ls_flags); |
81eeb82f | 501 | } |
3ae1acf9 | 502 | |
fad59c13 | 503 | /* ls_exflags are forced to match among nodes, and we don't |
6b0afc0c AA |
504 | * need to require all nodes to have some flags set |
505 | */ | |
0f8e0d9a DT |
506 | ls->ls_exflags = (flags & ~(DLM_LSFL_TIMEWARN | DLM_LSFL_FS | |
507 | DLM_LSFL_NEWEXCL)); | |
6b0afc0c AA |
508 | #else |
509 | /* ls_exflags are forced to match among nodes, and we don't | |
510 | * need to require all nodes to have some flags set | |
511 | */ | |
512 | ls->ls_exflags = (flags & ~(DLM_LSFL_FS | DLM_LSFL_NEWEXCL)); | |
513 | #endif | |
fad59c13 | 514 | |
d921a23f | 515 | size = READ_ONCE(dlm_config.ci_rsbtbl_size); |
e7fd4179 DT |
516 | ls->ls_rsbtbl_size = size; |
517 | ||
42bc47b3 | 518 | ls->ls_rsbtbl = vmalloc(array_size(size, sizeof(struct dlm_rsbtable))); |
e7fd4179 DT |
519 | if (!ls->ls_rsbtbl) |
520 | goto out_lsfree; | |
521 | for (i = 0; i < size; i++) { | |
9beb3bf5 BP |
522 | ls->ls_rsbtbl[i].keep.rb_node = NULL; |
523 | ls->ls_rsbtbl[i].toss.rb_node = NULL; | |
c7be761a | 524 | spin_lock_init(&ls->ls_rsbtbl[i].lock); |
e7fd4179 DT |
525 | } |
526 | ||
05c32f47 | 527 | spin_lock_init(&ls->ls_remove_spin); |
21d9ac1a | 528 | init_waitqueue_head(&ls->ls_remove_wait); |
05c32f47 DT |
529 | |
530 | for (i = 0; i < DLM_REMOVE_NAMES_MAX; i++) { | |
531 | ls->ls_remove_names[i] = kzalloc(DLM_RESNAME_MAXLEN+1, | |
532 | GFP_KERNEL); | |
533 | if (!ls->ls_remove_names[i]) | |
534 | goto out_rsbtbl; | |
535 | } | |
536 | ||
3d6aa675 DT |
537 | idr_init(&ls->ls_lkbidr); |
538 | spin_lock_init(&ls->ls_lkbidr_spin); | |
e7fd4179 | 539 | |
e7fd4179 | 540 | INIT_LIST_HEAD(&ls->ls_waiters); |
90135925 | 541 | mutex_init(&ls->ls_waiters_mutex); |
ef0c2bb0 DT |
542 | INIT_LIST_HEAD(&ls->ls_orphans); |
543 | mutex_init(&ls->ls_orphans_mutex); | |
6b0afc0c | 544 | #ifdef CONFIG_DLM_DEPRECATED_API |
3ae1acf9 DT |
545 | INIT_LIST_HEAD(&ls->ls_timeout); |
546 | mutex_init(&ls->ls_timeout_mutex); | |
6b0afc0c | 547 | #endif |
e7fd4179 | 548 | |
3881ac04 DT |
549 | INIT_LIST_HEAD(&ls->ls_new_rsb); |
550 | spin_lock_init(&ls->ls_new_rsb_spin); | |
551 | ||
e7fd4179 DT |
552 | INIT_LIST_HEAD(&ls->ls_nodes); |
553 | INIT_LIST_HEAD(&ls->ls_nodes_gone); | |
554 | ls->ls_num_nodes = 0; | |
555 | ls->ls_low_nodeid = 0; | |
556 | ls->ls_total_weight = 0; | |
557 | ls->ls_node_array = NULL; | |
558 | ||
559 | memset(&ls->ls_stub_rsb, 0, sizeof(struct dlm_rsb)); | |
560 | ls->ls_stub_rsb.res_ls = ls; | |
561 | ||
5de6319b DT |
562 | ls->ls_debug_rsb_dentry = NULL; |
563 | ls->ls_debug_waiters_dentry = NULL; | |
e7fd4179 DT |
564 | |
565 | init_waitqueue_head(&ls->ls_uevent_wait); | |
566 | ls->ls_uevent_result = 0; | |
682bb91b AA |
567 | init_completion(&ls->ls_recovery_done); |
568 | ls->ls_recovery_result = -1; | |
e7fd4179 | 569 | |
23e8e1aa DT |
570 | mutex_init(&ls->ls_cb_mutex); |
571 | INIT_LIST_HEAD(&ls->ls_cb_delay); | |
572 | ||
e7fd4179 | 573 | ls->ls_recoverd_task = NULL; |
90135925 | 574 | mutex_init(&ls->ls_recoverd_active); |
e7fd4179 | 575 | spin_lock_init(&ls->ls_recover_lock); |
98f176fb DT |
576 | spin_lock_init(&ls->ls_rcom_spin); |
577 | get_random_bytes(&ls->ls_rcom_seq, sizeof(uint64_t)); | |
e7fd4179 DT |
578 | ls->ls_recover_status = 0; |
579 | ls->ls_recover_seq = 0; | |
580 | ls->ls_recover_args = NULL; | |
581 | init_rwsem(&ls->ls_in_recovery); | |
c36258b5 | 582 | init_rwsem(&ls->ls_recv_active); |
e7fd4179 | 583 | INIT_LIST_HEAD(&ls->ls_requestqueue); |
164d88ab AA |
584 | atomic_set(&ls->ls_requestqueue_cnt, 0); |
585 | init_waitqueue_head(&ls->ls_requestqueue_wait); | |
90135925 | 586 | mutex_init(&ls->ls_requestqueue_mutex); |
296d9d1e | 587 | spin_lock_init(&ls->ls_clear_proc_locks); |
e7fd4179 | 588 | |
489d8e55 AA |
589 | /* Due backwards compatibility with 3.1 we need to use maximum |
590 | * possible dlm message size to be sure the message will fit and | |
591 | * not having out of bounds issues. However on sending side 3.2 | |
592 | * might send less. | |
593 | */ | |
d10a0b88 | 594 | ls->ls_recover_buf = kmalloc(DLM_MAX_SOCKET_BUFSIZE, GFP_NOFS); |
e7fd4179 | 595 | if (!ls->ls_recover_buf) |
05c32f47 | 596 | goto out_lkbidr; |
e7fd4179 | 597 | |
757a4271 DT |
598 | ls->ls_slot = 0; |
599 | ls->ls_num_slots = 0; | |
600 | ls->ls_slots_size = 0; | |
601 | ls->ls_slots = NULL; | |
602 | ||
e7fd4179 DT |
603 | INIT_LIST_HEAD(&ls->ls_recover_list); |
604 | spin_lock_init(&ls->ls_recover_list_lock); | |
1d7c484e DT |
605 | idr_init(&ls->ls_recover_idr); |
606 | spin_lock_init(&ls->ls_recover_idr_lock); | |
e7fd4179 | 607 | ls->ls_recover_list_count = 0; |
597d0cae | 608 | ls->ls_local_handle = ls; |
e7fd4179 DT |
609 | init_waitqueue_head(&ls->ls_wait_general); |
610 | INIT_LIST_HEAD(&ls->ls_root_list); | |
611 | init_rwsem(&ls->ls_root_sem); | |
612 | ||
5f88f1ea | 613 | spin_lock(&lslist_lock); |
0f8e0d9a | 614 | ls->ls_create_count = 1; |
5f88f1ea DT |
615 | list_add(&ls->ls_list, &lslist); |
616 | spin_unlock(&lslist_lock); | |
617 | ||
23e8e1aa DT |
618 | if (flags & DLM_LSFL_FS) { |
619 | error = dlm_callback_start(ls); | |
620 | if (error) { | |
621 | log_error(ls, "can't start dlm_callback %d", error); | |
622 | goto out_delist; | |
623 | } | |
624 | } | |
625 | ||
475f230c DT |
626 | init_waitqueue_head(&ls->ls_recover_lock_wait); |
627 | ||
628 | /* | |
629 | * Once started, dlm_recoverd first looks for ls in lslist, then | |
630 | * initializes ls_in_recovery as locked in "down" mode. We need | |
631 | * to wait for the wakeup from dlm_recoverd because in_recovery | |
632 | * has to start out in down mode. | |
633 | */ | |
634 | ||
e7fd4179 DT |
635 | error = dlm_recoverd_start(ls); |
636 | if (error) { | |
637 | log_error(ls, "can't start dlm_recoverd %d", error); | |
23e8e1aa | 638 | goto out_callback; |
e7fd4179 DT |
639 | } |
640 | ||
475f230c DT |
641 | wait_event(ls->ls_recover_lock_wait, |
642 | test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags)); | |
643 | ||
0ffddafc WH |
644 | /* let kobject handle freeing of ls if there's an error */ |
645 | do_unreg = 1; | |
646 | ||
901195ed GKH |
647 | ls->ls_kobj.kset = dlm_kset; |
648 | error = kobject_init_and_add(&ls->ls_kobj, &dlm_ktype, NULL, | |
649 | "%s", ls->ls_name); | |
e7fd4179 | 650 | if (error) |
23e8e1aa | 651 | goto out_recoverd; |
901195ed | 652 | kobject_uevent(&ls->ls_kobj, KOBJ_ADD); |
79d72b54 | 653 | |
8b0e7b2c DT |
654 | /* This uevent triggers dlm_controld in userspace to add us to the |
655 | group of nodes that are members of this lockspace (managed by the | |
656 | cluster infrastructure.) Once it's done that, it tells us who the | |
657 | current lockspace members are (via configfs) and then tells the | |
658 | lockspace to start running (via sysfs) in dlm_ls_start(). */ | |
659 | ||
e7fd4179 DT |
660 | error = do_uevent(ls, 1); |
661 | if (error) | |
23e8e1aa | 662 | goto out_recoverd; |
79d72b54 | 663 | |
682bb91b AA |
664 | /* wait until recovery is successful or failed */ |
665 | wait_for_completion(&ls->ls_recovery_done); | |
666 | error = ls->ls_recovery_result; | |
8b0e7b2c DT |
667 | if (error) |
668 | goto out_members; | |
669 | ||
79d72b54 DT |
670 | dlm_create_debug_file(ls); |
671 | ||
075f0177 | 672 | log_rinfo(ls, "join complete"); |
e7fd4179 DT |
673 | *lockspace = ls; |
674 | return 0; | |
675 | ||
8b0e7b2c DT |
676 | out_members: |
677 | do_uevent(ls, 0); | |
678 | dlm_clear_members(ls); | |
679 | kfree(ls->ls_node_array); | |
23e8e1aa | 680 | out_recoverd: |
5f88f1ea | 681 | dlm_recoverd_stop(ls); |
23e8e1aa DT |
682 | out_callback: |
683 | dlm_callback_stop(ls); | |
79d72b54 | 684 | out_delist: |
e7fd4179 DT |
685 | spin_lock(&lslist_lock); |
686 | list_del(&ls->ls_list); | |
687 | spin_unlock(&lslist_lock); | |
1d7c484e | 688 | idr_destroy(&ls->ls_recover_idr); |
e7fd4179 | 689 | kfree(ls->ls_recover_buf); |
05c32f47 | 690 | out_lkbidr: |
3d6aa675 | 691 | idr_destroy(&ls->ls_lkbidr); |
b982896c | 692 | out_rsbtbl: |
3456880f TM |
693 | for (i = 0; i < DLM_REMOVE_NAMES_MAX; i++) |
694 | kfree(ls->ls_remove_names[i]); | |
c282af49 | 695 | vfree(ls->ls_rsbtbl); |
e7fd4179 | 696 | out_lsfree: |
79d72b54 | 697 | if (do_unreg) |
197b12d6 | 698 | kobject_put(&ls->ls_kobj); |
79d72b54 DT |
699 | else |
700 | kfree(ls); | |
e7fd4179 DT |
701 | out: |
702 | module_put(THIS_MODULE); | |
703 | return error; | |
704 | } | |
705 | ||
12cda13c AA |
706 | static int __dlm_new_lockspace(const char *name, const char *cluster, |
707 | uint32_t flags, int lvblen, | |
708 | const struct dlm_lockspace_ops *ops, | |
709 | void *ops_arg, int *ops_result, | |
710 | dlm_lockspace_t **lockspace) | |
e7fd4179 DT |
711 | { |
712 | int error = 0; | |
713 | ||
90135925 | 714 | mutex_lock(&ls_lock); |
e7fd4179 DT |
715 | if (!ls_count) |
716 | error = threads_start(); | |
717 | if (error) | |
718 | goto out; | |
719 | ||
60f98d18 DT |
720 | error = new_lockspace(name, cluster, flags, lvblen, ops, ops_arg, |
721 | ops_result, lockspace); | |
e7fd4179 DT |
722 | if (!error) |
723 | ls_count++; | |
8511a272 DT |
724 | if (error > 0) |
725 | error = 0; | |
9d232469 AA |
726 | if (!ls_count) { |
727 | dlm_scand_stop(); | |
a070a91c | 728 | dlm_midcomms_shutdown(); |
9d232469 AA |
729 | dlm_lowcomms_stop(); |
730 | } | |
e7fd4179 | 731 | out: |
90135925 | 732 | mutex_unlock(&ls_lock); |
e7fd4179 DT |
733 | return error; |
734 | } | |
735 | ||
12cda13c AA |
736 | int dlm_new_lockspace(const char *name, const char *cluster, uint32_t flags, |
737 | int lvblen, const struct dlm_lockspace_ops *ops, | |
738 | void *ops_arg, int *ops_result, | |
739 | dlm_lockspace_t **lockspace) | |
740 | { | |
741 | return __dlm_new_lockspace(name, cluster, flags | DLM_LSFL_FS, lvblen, | |
742 | ops, ops_arg, ops_result, lockspace); | |
743 | } | |
744 | ||
745 | int dlm_new_user_lockspace(const char *name, const char *cluster, | |
746 | uint32_t flags, int lvblen, | |
747 | const struct dlm_lockspace_ops *ops, | |
748 | void *ops_arg, int *ops_result, | |
749 | dlm_lockspace_t **lockspace) | |
750 | { | |
751 | return __dlm_new_lockspace(name, cluster, flags, lvblen, ops, | |
752 | ops_arg, ops_result, lockspace); | |
753 | } | |
754 | ||
3d6aa675 | 755 | static int lkb_idr_is_local(int id, void *p, void *data) |
e7fd4179 | 756 | { |
3d6aa675 DT |
757 | struct dlm_lkb *lkb = p; |
758 | ||
a97f4a66 | 759 | return lkb->lkb_nodeid == 0 && lkb->lkb_grmode != DLM_LOCK_IV; |
3d6aa675 DT |
760 | } |
761 | ||
762 | static int lkb_idr_is_any(int id, void *p, void *data) | |
763 | { | |
764 | return 1; | |
765 | } | |
766 | ||
767 | static int lkb_idr_free(int id, void *p, void *data) | |
768 | { | |
769 | struct dlm_lkb *lkb = p; | |
770 | ||
3d6aa675 DT |
771 | if (lkb->lkb_lvbptr && lkb->lkb_flags & DLM_IFL_MSTCPY) |
772 | dlm_free_lvb(lkb->lkb_lvbptr); | |
773 | ||
774 | dlm_free_lkb(lkb); | |
775 | return 0; | |
776 | } | |
777 | ||
778 | /* NOTE: We check the lkbidr here rather than the resource table. | |
779 | This is because there may be LKBs queued as ASTs that have been unlinked | |
780 | from their RSBs and are pending deletion once the AST has been delivered */ | |
781 | ||
782 | static int lockspace_busy(struct dlm_ls *ls, int force) | |
783 | { | |
784 | int rv; | |
785 | ||
786 | spin_lock(&ls->ls_lkbidr_spin); | |
787 | if (force == 0) { | |
788 | rv = idr_for_each(&ls->ls_lkbidr, lkb_idr_is_any, ls); | |
789 | } else if (force == 1) { | |
790 | rv = idr_for_each(&ls->ls_lkbidr, lkb_idr_is_local, ls); | |
791 | } else { | |
792 | rv = 0; | |
e7fd4179 | 793 | } |
3d6aa675 DT |
794 | spin_unlock(&ls->ls_lkbidr_spin); |
795 | return rv; | |
e7fd4179 DT |
796 | } |
797 | ||
798 | static int release_lockspace(struct dlm_ls *ls, int force) | |
799 | { | |
e7fd4179 | 800 | struct dlm_rsb *rsb; |
9beb3bf5 | 801 | struct rb_node *n; |
0f8e0d9a DT |
802 | int i, busy, rv; |
803 | ||
3d6aa675 | 804 | busy = lockspace_busy(ls, force); |
0f8e0d9a DT |
805 | |
806 | spin_lock(&lslist_lock); | |
807 | if (ls->ls_create_count == 1) { | |
3d6aa675 | 808 | if (busy) { |
0f8e0d9a | 809 | rv = -EBUSY; |
3d6aa675 | 810 | } else { |
0f8e0d9a DT |
811 | /* remove_lockspace takes ls off lslist */ |
812 | ls->ls_create_count = 0; | |
813 | rv = 0; | |
814 | } | |
815 | } else if (ls->ls_create_count > 1) { | |
816 | rv = --ls->ls_create_count; | |
817 | } else { | |
818 | rv = -EINVAL; | |
819 | } | |
820 | spin_unlock(&lslist_lock); | |
821 | ||
822 | if (rv) { | |
823 | log_debug(ls, "release_lockspace no remove %d", rv); | |
824 | return rv; | |
825 | } | |
e7fd4179 | 826 | |
0f8e0d9a | 827 | dlm_device_deregister(ls); |
e7fd4179 | 828 | |
dc68c7ed | 829 | if (force < 3 && dlm_user_daemon_available()) |
e7fd4179 DT |
830 | do_uevent(ls, 0); |
831 | ||
832 | dlm_recoverd_stop(ls); | |
833 | ||
9d232469 AA |
834 | if (ls_count == 1) { |
835 | dlm_scand_stop(); | |
ecd95673 | 836 | dlm_clear_members(ls); |
a070a91c | 837 | dlm_midcomms_shutdown(); |
9d232469 AA |
838 | } |
839 | ||
23e8e1aa DT |
840 | dlm_callback_stop(ls); |
841 | ||
e7fd4179 DT |
842 | remove_lockspace(ls); |
843 | ||
844 | dlm_delete_debug_file(ls); | |
845 | ||
8fc6ed9a | 846 | idr_destroy(&ls->ls_recover_idr); |
e7fd4179 DT |
847 | kfree(ls->ls_recover_buf); |
848 | ||
e7fd4179 | 849 | /* |
3d6aa675 | 850 | * Free all lkb's in idr |
e7fd4179 DT |
851 | */ |
852 | ||
3d6aa675 | 853 | idr_for_each(&ls->ls_lkbidr, lkb_idr_free, ls); |
3d6aa675 | 854 | idr_destroy(&ls->ls_lkbidr); |
e7fd4179 | 855 | |
e7fd4179 DT |
856 | /* |
857 | * Free all rsb's on rsbtbl[] lists | |
858 | */ | |
859 | ||
860 | for (i = 0; i < ls->ls_rsbtbl_size; i++) { | |
9beb3bf5 BP |
861 | while ((n = rb_first(&ls->ls_rsbtbl[i].keep))) { |
862 | rsb = rb_entry(n, struct dlm_rsb, res_hashnode); | |
863 | rb_erase(n, &ls->ls_rsbtbl[i].keep); | |
52bda2b5 | 864 | dlm_free_rsb(rsb); |
e7fd4179 DT |
865 | } |
866 | ||
9beb3bf5 BP |
867 | while ((n = rb_first(&ls->ls_rsbtbl[i].toss))) { |
868 | rsb = rb_entry(n, struct dlm_rsb, res_hashnode); | |
869 | rb_erase(n, &ls->ls_rsbtbl[i].toss); | |
52bda2b5 | 870 | dlm_free_rsb(rsb); |
e7fd4179 DT |
871 | } |
872 | } | |
873 | ||
c282af49 | 874 | vfree(ls->ls_rsbtbl); |
e7fd4179 | 875 | |
05c32f47 DT |
876 | for (i = 0; i < DLM_REMOVE_NAMES_MAX; i++) |
877 | kfree(ls->ls_remove_names[i]); | |
878 | ||
3881ac04 DT |
879 | while (!list_empty(&ls->ls_new_rsb)) { |
880 | rsb = list_first_entry(&ls->ls_new_rsb, struct dlm_rsb, | |
881 | res_hashchain); | |
882 | list_del(&rsb->res_hashchain); | |
883 | dlm_free_rsb(rsb); | |
884 | } | |
885 | ||
e7fd4179 DT |
886 | /* |
887 | * Free structures on any other lists | |
888 | */ | |
889 | ||
2896ee37 | 890 | dlm_purge_requestqueue(ls); |
e7fd4179 | 891 | kfree(ls->ls_recover_args); |
e7fd4179 DT |
892 | dlm_clear_members(ls); |
893 | dlm_clear_members_gone(ls); | |
894 | kfree(ls->ls_node_array); | |
075f0177 | 895 | log_rinfo(ls, "release_lockspace final free"); |
197b12d6 | 896 | kobject_put(&ls->ls_kobj); |
79d72b54 | 897 | /* The ls structure will be freed when the kobject is done with */ |
e7fd4179 | 898 | |
e7fd4179 DT |
899 | module_put(THIS_MODULE); |
900 | return 0; | |
901 | } | |
902 | ||
903 | /* | |
904 | * Called when a system has released all its locks and is not going to use the | |
905 | * lockspace any longer. We free everything we're managing for this lockspace. | |
906 | * Remaining nodes will go through the recovery process as if we'd died. The | |
907 | * lockspace must continue to function as usual, participating in recoveries, | |
908 | * until this returns. | |
909 | * | |
910 | * Force has 4 possible values: | |
bb6866a5 | 911 | * 0 - don't destroy lockspace if it has any LKBs |
e7fd4179 DT |
912 | * 1 - destroy lockspace if it has remote LKBs but not if it has local LKBs |
913 | * 2 - destroy lockspace regardless of LKBs | |
914 | * 3 - destroy lockspace as part of a forced shutdown | |
915 | */ | |
916 | ||
917 | int dlm_release_lockspace(void *lockspace, int force) | |
918 | { | |
919 | struct dlm_ls *ls; | |
0f8e0d9a | 920 | int error; |
e7fd4179 DT |
921 | |
922 | ls = dlm_find_lockspace_local(lockspace); | |
923 | if (!ls) | |
924 | return -EINVAL; | |
925 | dlm_put_lockspace(ls); | |
0f8e0d9a DT |
926 | |
927 | mutex_lock(&ls_lock); | |
928 | error = release_lockspace(ls, force); | |
929 | if (!error) | |
930 | ls_count--; | |
278afcbf | 931 | if (!ls_count) |
9d232469 | 932 | dlm_lowcomms_stop(); |
0f8e0d9a DT |
933 | mutex_unlock(&ls_lock); |
934 | ||
935 | return error; | |
e7fd4179 DT |
936 | } |
937 | ||
dc68c7ed DT |
938 | void dlm_stop_lockspaces(void) |
939 | { | |
940 | struct dlm_ls *ls; | |
696b3d84 | 941 | int count; |
dc68c7ed DT |
942 | |
943 | restart: | |
696b3d84 | 944 | count = 0; |
dc68c7ed DT |
945 | spin_lock(&lslist_lock); |
946 | list_for_each_entry(ls, &lslist, ls_list) { | |
696b3d84 DT |
947 | if (!test_bit(LSFL_RUNNING, &ls->ls_flags)) { |
948 | count++; | |
dc68c7ed | 949 | continue; |
696b3d84 | 950 | } |
dc68c7ed DT |
951 | spin_unlock(&lslist_lock); |
952 | log_error(ls, "no userland control daemon, stopping lockspace"); | |
953 | dlm_ls_stop(ls); | |
954 | goto restart; | |
955 | } | |
956 | spin_unlock(&lslist_lock); | |
696b3d84 DT |
957 | |
958 | if (count) | |
959 | log_print("dlm user daemon left %d lockspaces", count); | |
dc68c7ed DT |
960 | } |
961 | ||
2c3fa6ae AA |
962 | void dlm_stop_lockspaces_check(void) |
963 | { | |
964 | struct dlm_ls *ls; | |
965 | ||
966 | spin_lock(&lslist_lock); | |
967 | list_for_each_entry(ls, &lslist, ls_list) { | |
968 | if (WARN_ON(!rwsem_is_locked(&ls->ls_in_recovery) || | |
969 | !dlm_locking_stopped(ls))) | |
970 | break; | |
971 | } | |
972 | spin_unlock(&lslist_lock); | |
973 | } |