]>
Commit | Line | Data |
---|---|---|
6714d8e8 KH |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * dlmlock.c | |
5 | * | |
6 | * underlying calls for lock creation | |
7 | * | |
8 | * Copyright (C) 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | * | |
25 | */ | |
26 | ||
27 | ||
28 | #include <linux/module.h> | |
29 | #include <linux/fs.h> | |
30 | #include <linux/types.h> | |
31 | #include <linux/slab.h> | |
32 | #include <linux/highmem.h> | |
6714d8e8 KH |
33 | #include <linux/init.h> |
34 | #include <linux/sysctl.h> | |
35 | #include <linux/random.h> | |
36 | #include <linux/blkdev.h> | |
37 | #include <linux/socket.h> | |
38 | #include <linux/inet.h> | |
39 | #include <linux/spinlock.h> | |
40 | #include <linux/delay.h> | |
41 | ||
42 | ||
43 | #include "cluster/heartbeat.h" | |
44 | #include "cluster/nodemanager.h" | |
45 | #include "cluster/tcp.h" | |
46 | ||
47 | #include "dlmapi.h" | |
48 | #include "dlmcommon.h" | |
49 | ||
50 | #include "dlmconvert.h" | |
51 | ||
52 | #define MLOG_MASK_PREFIX ML_DLM | |
53 | #include "cluster/masklog.h" | |
54 | ||
724bdca9 SM |
55 | static struct kmem_cache *dlm_lock_cache = NULL; |
56 | ||
34af946a | 57 | static DEFINE_SPINLOCK(dlm_cookie_lock); |
6714d8e8 KH |
58 | static u64 dlm_next_cookie = 1; |
59 | ||
60 | static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm, | |
61 | struct dlm_lock_resource *res, | |
62 | struct dlm_lock *lock, int flags); | |
63 | static void dlm_init_lock(struct dlm_lock *newlock, int type, | |
64 | u8 node, u64 cookie); | |
65 | static void dlm_lock_release(struct kref *kref); | |
66 | static void dlm_lock_detach_lockres(struct dlm_lock *lock); | |
67 | ||
724bdca9 SM |
68 | int dlm_init_lock_cache(void) |
69 | { | |
70 | dlm_lock_cache = kmem_cache_create("o2dlm_lock", | |
71 | sizeof(struct dlm_lock), | |
72 | 0, SLAB_HWCACHE_ALIGN, NULL); | |
73 | if (dlm_lock_cache == NULL) | |
74 | return -ENOMEM; | |
75 | return 0; | |
76 | } | |
77 | ||
78 | void dlm_destroy_lock_cache(void) | |
79 | { | |
80 | if (dlm_lock_cache) | |
81 | kmem_cache_destroy(dlm_lock_cache); | |
82 | } | |
83 | ||
6714d8e8 KH |
84 | /* Tell us whether we can grant a new lock request. |
85 | * locking: | |
86 | * caller needs: res->spinlock | |
87 | * taken: none | |
88 | * held on exit: none | |
89 | * returns: 1 if the lock can be granted, 0 otherwise. | |
90 | */ | |
91 | static int dlm_can_grant_new_lock(struct dlm_lock_resource *res, | |
92 | struct dlm_lock *lock) | |
93 | { | |
94 | struct list_head *iter; | |
95 | struct dlm_lock *tmplock; | |
96 | ||
97 | list_for_each(iter, &res->granted) { | |
98 | tmplock = list_entry(iter, struct dlm_lock, list); | |
99 | ||
100 | if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type)) | |
101 | return 0; | |
102 | } | |
103 | ||
104 | list_for_each(iter, &res->converting) { | |
105 | tmplock = list_entry(iter, struct dlm_lock, list); | |
106 | ||
107 | if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type)) | |
108 | return 0; | |
109 | } | |
110 | ||
111 | return 1; | |
112 | } | |
113 | ||
114 | /* performs lock creation at the lockres master site | |
115 | * locking: | |
116 | * caller needs: none | |
117 | * taken: takes and drops res->spinlock | |
118 | * held on exit: none | |
119 | * returns: DLM_NORMAL, DLM_NOTQUEUED | |
120 | */ | |
121 | static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm, | |
122 | struct dlm_lock_resource *res, | |
123 | struct dlm_lock *lock, int flags) | |
124 | { | |
125 | int call_ast = 0, kick_thread = 0; | |
126 | enum dlm_status status = DLM_NORMAL; | |
127 | ||
128 | mlog_entry("type=%d\n", lock->ml.type); | |
129 | ||
130 | spin_lock(&res->spinlock); | |
131 | /* if called from dlm_create_lock_handler, need to | |
132 | * ensure it will not sleep in dlm_wait_on_lockres */ | |
133 | status = __dlm_lockres_state_to_status(res); | |
134 | if (status != DLM_NORMAL && | |
135 | lock->ml.node != dlm->node_num) { | |
136 | /* erf. state changed after lock was dropped. */ | |
137 | spin_unlock(&res->spinlock); | |
138 | dlm_error(status); | |
139 | return status; | |
140 | } | |
141 | __dlm_wait_on_lockres(res); | |
142 | __dlm_lockres_reserve_ast(res); | |
143 | ||
144 | if (dlm_can_grant_new_lock(res, lock)) { | |
145 | mlog(0, "I can grant this lock right away\n"); | |
146 | /* got it right away */ | |
147 | lock->lksb->status = DLM_NORMAL; | |
148 | status = DLM_NORMAL; | |
149 | dlm_lock_get(lock); | |
150 | list_add_tail(&lock->list, &res->granted); | |
151 | ||
152 | /* for the recovery lock, we can't allow the ast | |
153 | * to be queued since the dlmthread is already | |
154 | * frozen. but the recovery lock is always locked | |
155 | * with LKM_NOQUEUE so we do not need the ast in | |
156 | * this special case */ | |
157 | if (!dlm_is_recovery_lock(res->lockname.name, | |
158 | res->lockname.len)) { | |
159 | kick_thread = 1; | |
160 | call_ast = 1; | |
c03872f5 KH |
161 | } else { |
162 | mlog(0, "%s: returning DLM_NORMAL to " | |
163 | "node %u for reco lock\n", dlm->name, | |
164 | lock->ml.node); | |
6714d8e8 KH |
165 | } |
166 | } else { | |
167 | /* for NOQUEUE request, unless we get the | |
168 | * lock right away, return DLM_NOTQUEUED */ | |
c03872f5 | 169 | if (flags & LKM_NOQUEUE) { |
6714d8e8 | 170 | status = DLM_NOTQUEUED; |
c03872f5 KH |
171 | if (dlm_is_recovery_lock(res->lockname.name, |
172 | res->lockname.len)) { | |
173 | mlog(0, "%s: returning NOTQUEUED to " | |
174 | "node %u for reco lock\n", dlm->name, | |
175 | lock->ml.node); | |
176 | } | |
177 | } else { | |
6714d8e8 KH |
178 | dlm_lock_get(lock); |
179 | list_add_tail(&lock->list, &res->blocked); | |
180 | kick_thread = 1; | |
181 | } | |
182 | } | |
ba2bf218 KH |
183 | /* reduce the inflight count, this may result in the lockres |
184 | * being purged below during calc_usage */ | |
185 | if (lock->ml.node == dlm->node_num) | |
186 | dlm_lockres_drop_inflight_ref(dlm, res); | |
6714d8e8 KH |
187 | |
188 | spin_unlock(&res->spinlock); | |
189 | wake_up(&res->wq); | |
190 | ||
191 | /* either queue the ast or release it */ | |
192 | if (call_ast) | |
193 | dlm_queue_ast(dlm, lock); | |
194 | else | |
195 | dlm_lockres_release_ast(dlm, res); | |
196 | ||
197 | dlm_lockres_calc_usage(dlm, res); | |
198 | if (kick_thread) | |
199 | dlm_kick_thread(dlm, res); | |
200 | ||
201 | return status; | |
202 | } | |
203 | ||
204 | void dlm_revert_pending_lock(struct dlm_lock_resource *res, | |
205 | struct dlm_lock *lock) | |
206 | { | |
207 | /* remove from local queue if it failed */ | |
208 | list_del_init(&lock->list); | |
209 | lock->lksb->flags &= ~DLM_LKSB_GET_LVB; | |
210 | } | |
211 | ||
212 | ||
213 | /* | |
214 | * locking: | |
215 | * caller needs: none | |
216 | * taken: takes and drops res->spinlock | |
217 | * held on exit: none | |
218 | * returns: DLM_DENIED, DLM_RECOVERING, or net status | |
219 | */ | |
220 | static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm, | |
221 | struct dlm_lock_resource *res, | |
222 | struct dlm_lock *lock, int flags) | |
223 | { | |
224 | enum dlm_status status = DLM_DENIED; | |
2abaf97e | 225 | int lockres_changed = 1; |
6714d8e8 KH |
226 | |
227 | mlog_entry("type=%d\n", lock->ml.type); | |
228 | mlog(0, "lockres %.*s, flags = 0x%x\n", res->lockname.len, | |
229 | res->lockname.name, flags); | |
230 | ||
231 | spin_lock(&res->spinlock); | |
232 | ||
233 | /* will exit this call with spinlock held */ | |
234 | __dlm_wait_on_lockres(res); | |
235 | res->state |= DLM_LOCK_RES_IN_PROGRESS; | |
236 | ||
237 | /* add lock to local (secondary) queue */ | |
238 | dlm_lock_get(lock); | |
239 | list_add_tail(&lock->list, &res->blocked); | |
240 | lock->lock_pending = 1; | |
241 | spin_unlock(&res->spinlock); | |
242 | ||
243 | /* spec seems to say that you will get DLM_NORMAL when the lock | |
244 | * has been queued, meaning we need to wait for a reply here. */ | |
245 | status = dlm_send_remote_lock_request(dlm, res, lock, flags); | |
246 | ||
247 | spin_lock(&res->spinlock); | |
248 | res->state &= ~DLM_LOCK_RES_IN_PROGRESS; | |
249 | lock->lock_pending = 0; | |
250 | if (status != DLM_NORMAL) { | |
c8df412e KH |
251 | if (status == DLM_RECOVERING && |
252 | dlm_is_recovery_lock(res->lockname.name, | |
253 | res->lockname.len)) { | |
254 | /* recovery lock was mastered by dead node. | |
255 | * we need to have calc_usage shoot down this | |
256 | * lockres and completely remaster it. */ | |
257 | mlog(0, "%s: recovery lock was owned by " | |
258 | "dead node %u, remaster it now.\n", | |
259 | dlm->name, res->owner); | |
260 | } else if (status != DLM_NOTQUEUED) { | |
c87a9ae7 KH |
261 | /* |
262 | * DO NOT call calc_usage, as this would unhash | |
263 | * the remote lockres before we ever get to use | |
264 | * it. treat as if we never made any change to | |
265 | * the lockres. | |
266 | */ | |
267 | lockres_changed = 0; | |
6714d8e8 | 268 | dlm_error(status); |
c87a9ae7 | 269 | } |
6714d8e8 KH |
270 | dlm_revert_pending_lock(res, lock); |
271 | dlm_lock_put(lock); | |
2bd63216 | 272 | } else if (dlm_is_recovery_lock(res->lockname.name, |
558c70c5 KH |
273 | res->lockname.len)) { |
274 | /* special case for the $RECOVERY lock. | |
275 | * there will never be an AST delivered to put | |
276 | * this lock on the proper secondary queue | |
277 | * (granted), so do it manually. */ | |
278 | mlog(0, "%s: $RECOVERY lock for this node (%u) is " | |
279 | "mastered by %u; got lock, manually granting (no ast)\n", | |
280 | dlm->name, dlm->node_num, res->owner); | |
f116629d | 281 | list_move_tail(&lock->list, &res->granted); |
6714d8e8 KH |
282 | } |
283 | spin_unlock(&res->spinlock); | |
284 | ||
2abaf97e KH |
285 | if (lockres_changed) |
286 | dlm_lockres_calc_usage(dlm, res); | |
6714d8e8 KH |
287 | |
288 | wake_up(&res->wq); | |
289 | return status; | |
290 | } | |
291 | ||
292 | ||
293 | /* for remote lock creation. | |
294 | * locking: | |
295 | * caller needs: none, but need res->state & DLM_LOCK_RES_IN_PROGRESS | |
296 | * taken: none | |
297 | * held on exit: none | |
298 | * returns: DLM_NOLOCKMGR, or net status | |
299 | */ | |
300 | static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm, | |
301 | struct dlm_lock_resource *res, | |
302 | struct dlm_lock *lock, int flags) | |
303 | { | |
304 | struct dlm_create_lock create; | |
305 | int tmpret, status = 0; | |
306 | enum dlm_status ret; | |
307 | ||
308 | mlog_entry_void(); | |
309 | ||
310 | memset(&create, 0, sizeof(create)); | |
311 | create.node_idx = dlm->node_num; | |
312 | create.requested_type = lock->ml.type; | |
313 | create.cookie = lock->ml.cookie; | |
314 | create.namelen = res->lockname.len; | |
315 | create.flags = cpu_to_be32(flags); | |
316 | memcpy(create.name, res->lockname.name, create.namelen); | |
317 | ||
318 | tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create, | |
319 | sizeof(create), res->owner, &status); | |
320 | if (tmpret >= 0) { | |
321 | // successfully sent and received | |
322 | ret = status; // this is already a dlm_status | |
495ac96e | 323 | if (ret == DLM_REJECTED) { |
e4eb0368 KH |
324 | mlog(ML_ERROR, "%s:%.*s: BUG. this is a stale lockres " |
325 | "no longer owned by %u. that node is coming back " | |
326 | "up currently.\n", dlm->name, create.namelen, | |
327 | create.name, res->owner); | |
328 | dlm_print_one_lock_resource(res); | |
329 | BUG(); | |
330 | } | |
6714d8e8 | 331 | } else { |
a5196ec5 WW |
332 | mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to " |
333 | "node %u\n", tmpret, DLM_CREATE_LOCK_MSG, dlm->key, | |
334 | res->owner); | |
6714d8e8 KH |
335 | if (dlm_is_host_down(tmpret)) { |
336 | ret = DLM_RECOVERING; | |
337 | mlog(0, "node %u died so returning DLM_RECOVERING " | |
338 | "from lock message!\n", res->owner); | |
339 | } else { | |
340 | ret = dlm_err_to_dlm_status(tmpret); | |
341 | } | |
342 | } | |
343 | ||
344 | return ret; | |
345 | } | |
346 | ||
347 | void dlm_lock_get(struct dlm_lock *lock) | |
348 | { | |
349 | kref_get(&lock->lock_refs); | |
350 | } | |
351 | ||
352 | void dlm_lock_put(struct dlm_lock *lock) | |
353 | { | |
354 | kref_put(&lock->lock_refs, dlm_lock_release); | |
355 | } | |
356 | ||
357 | static void dlm_lock_release(struct kref *kref) | |
358 | { | |
359 | struct dlm_lock *lock; | |
360 | ||
361 | lock = container_of(kref, struct dlm_lock, lock_refs); | |
362 | ||
363 | BUG_ON(!list_empty(&lock->list)); | |
364 | BUG_ON(!list_empty(&lock->ast_list)); | |
365 | BUG_ON(!list_empty(&lock->bast_list)); | |
366 | BUG_ON(lock->ast_pending); | |
367 | BUG_ON(lock->bast_pending); | |
368 | ||
369 | dlm_lock_detach_lockres(lock); | |
370 | ||
371 | if (lock->lksb_kernel_allocated) { | |
372 | mlog(0, "freeing kernel-allocated lksb\n"); | |
373 | kfree(lock->lksb); | |
374 | } | |
724bdca9 | 375 | kmem_cache_free(dlm_lock_cache, lock); |
6714d8e8 KH |
376 | } |
377 | ||
378 | /* associate a lock with it's lockres, getting a ref on the lockres */ | |
379 | void dlm_lock_attach_lockres(struct dlm_lock *lock, | |
380 | struct dlm_lock_resource *res) | |
381 | { | |
382 | dlm_lockres_get(res); | |
383 | lock->lockres = res; | |
384 | } | |
385 | ||
386 | /* drop ref on lockres, if there is still one associated with lock */ | |
387 | static void dlm_lock_detach_lockres(struct dlm_lock *lock) | |
388 | { | |
389 | struct dlm_lock_resource *res; | |
390 | ||
391 | res = lock->lockres; | |
392 | if (res) { | |
393 | lock->lockres = NULL; | |
394 | mlog(0, "removing lock's lockres reference\n"); | |
395 | dlm_lockres_put(res); | |
396 | } | |
397 | } | |
398 | ||
399 | static void dlm_init_lock(struct dlm_lock *newlock, int type, | |
400 | u8 node, u64 cookie) | |
401 | { | |
402 | INIT_LIST_HEAD(&newlock->list); | |
403 | INIT_LIST_HEAD(&newlock->ast_list); | |
404 | INIT_LIST_HEAD(&newlock->bast_list); | |
405 | spin_lock_init(&newlock->spinlock); | |
406 | newlock->ml.type = type; | |
407 | newlock->ml.convert_type = LKM_IVMODE; | |
408 | newlock->ml.highest_blocked = LKM_IVMODE; | |
409 | newlock->ml.node = node; | |
410 | newlock->ml.pad1 = 0; | |
411 | newlock->ml.list = 0; | |
412 | newlock->ml.flags = 0; | |
413 | newlock->ast = NULL; | |
414 | newlock->bast = NULL; | |
415 | newlock->astdata = NULL; | |
416 | newlock->ml.cookie = cpu_to_be64(cookie); | |
417 | newlock->ast_pending = 0; | |
418 | newlock->bast_pending = 0; | |
419 | newlock->convert_pending = 0; | |
420 | newlock->lock_pending = 0; | |
421 | newlock->unlock_pending = 0; | |
422 | newlock->cancel_pending = 0; | |
423 | newlock->lksb_kernel_allocated = 0; | |
424 | ||
425 | kref_init(&newlock->lock_refs); | |
426 | } | |
427 | ||
428 | struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie, | |
429 | struct dlm_lockstatus *lksb) | |
430 | { | |
431 | struct dlm_lock *lock; | |
432 | int kernel_allocated = 0; | |
433 | ||
3914ed0c | 434 | lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS); |
6714d8e8 KH |
435 | if (!lock) |
436 | return NULL; | |
437 | ||
438 | if (!lksb) { | |
439 | /* zero memory only if kernel-allocated */ | |
cd861280 | 440 | lksb = kzalloc(sizeof(*lksb), GFP_NOFS); |
6714d8e8 KH |
441 | if (!lksb) { |
442 | kfree(lock); | |
443 | return NULL; | |
444 | } | |
445 | kernel_allocated = 1; | |
446 | } | |
447 | ||
448 | dlm_init_lock(lock, type, node, cookie); | |
449 | if (kernel_allocated) | |
450 | lock->lksb_kernel_allocated = 1; | |
451 | lock->lksb = lksb; | |
452 | lksb->lockid = lock; | |
453 | return lock; | |
454 | } | |
455 | ||
456 | /* handler for lock creation net message | |
457 | * locking: | |
458 | * caller needs: none | |
459 | * taken: takes and drops res->spinlock | |
460 | * held on exit: none | |
461 | * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED | |
462 | */ | |
d74c9803 KH |
463 | int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data, |
464 | void **ret_data) | |
6714d8e8 KH |
465 | { |
466 | struct dlm_ctxt *dlm = data; | |
467 | struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf; | |
468 | struct dlm_lock_resource *res = NULL; | |
469 | struct dlm_lock *newlock = NULL; | |
470 | struct dlm_lockstatus *lksb = NULL; | |
471 | enum dlm_status status = DLM_NORMAL; | |
472 | char *name; | |
473 | unsigned int namelen; | |
474 | ||
475 | BUG_ON(!dlm); | |
476 | ||
477 | mlog_entry_void(); | |
478 | ||
479 | if (!dlm_grab(dlm)) | |
480 | return DLM_REJECTED; | |
481 | ||
6714d8e8 KH |
482 | name = create->name; |
483 | namelen = create->namelen; | |
495ac96e | 484 | status = DLM_REJECTED; |
e4eb0368 KH |
485 | if (!dlm_domain_fully_joined(dlm)) { |
486 | mlog(ML_ERROR, "Domain %s not fully joined, but node %u is " | |
487 | "sending a create_lock message for lock %.*s!\n", | |
488 | dlm->name, create->node_idx, namelen, name); | |
489 | dlm_error(status); | |
490 | goto leave; | |
491 | } | |
6714d8e8 KH |
492 | |
493 | status = DLM_IVBUFLEN; | |
494 | if (namelen > DLM_LOCKID_NAME_MAX) { | |
495 | dlm_error(status); | |
496 | goto leave; | |
497 | } | |
498 | ||
499 | status = DLM_SYSERR; | |
500 | newlock = dlm_new_lock(create->requested_type, | |
501 | create->node_idx, | |
502 | be64_to_cpu(create->cookie), NULL); | |
503 | if (!newlock) { | |
504 | dlm_error(status); | |
505 | goto leave; | |
506 | } | |
507 | ||
508 | lksb = newlock->lksb; | |
509 | ||
510 | if (be32_to_cpu(create->flags) & LKM_GET_LVB) { | |
511 | lksb->flags |= DLM_LKSB_GET_LVB; | |
512 | mlog(0, "set DLM_LKSB_GET_LVB flag\n"); | |
513 | } | |
514 | ||
515 | status = DLM_IVLOCKID; | |
516 | res = dlm_lookup_lockres(dlm, name, namelen); | |
517 | if (!res) { | |
518 | dlm_error(status); | |
519 | goto leave; | |
520 | } | |
521 | ||
522 | spin_lock(&res->spinlock); | |
523 | status = __dlm_lockres_state_to_status(res); | |
524 | spin_unlock(&res->spinlock); | |
525 | ||
526 | if (status != DLM_NORMAL) { | |
527 | mlog(0, "lockres recovering/migrating/in-progress\n"); | |
528 | goto leave; | |
529 | } | |
530 | ||
531 | dlm_lock_attach_lockres(newlock, res); | |
532 | ||
533 | status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags)); | |
534 | leave: | |
535 | if (status != DLM_NORMAL) | |
536 | if (newlock) | |
537 | dlm_lock_put(newlock); | |
538 | ||
539 | if (res) | |
540 | dlm_lockres_put(res); | |
541 | ||
542 | dlm_put(dlm); | |
543 | ||
544 | return status; | |
545 | } | |
546 | ||
547 | ||
548 | /* fetch next node-local (u8 nodenum + u56 cookie) into u64 */ | |
549 | static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie) | |
550 | { | |
551 | u64 tmpnode = node_num; | |
552 | ||
553 | /* shift single byte of node num into top 8 bits */ | |
554 | tmpnode <<= 56; | |
555 | ||
556 | spin_lock(&dlm_cookie_lock); | |
557 | *cookie = (dlm_next_cookie | tmpnode); | |
558 | if (++dlm_next_cookie & 0xff00000000000000ull) { | |
559 | mlog(0, "This node's cookie will now wrap!\n"); | |
560 | dlm_next_cookie = 1; | |
561 | } | |
562 | spin_unlock(&dlm_cookie_lock); | |
563 | } | |
564 | ||
565 | enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode, | |
566 | struct dlm_lockstatus *lksb, int flags, | |
3384f3df MF |
567 | const char *name, int namelen, dlm_astlockfunc_t *ast, |
568 | void *data, dlm_bastlockfunc_t *bast) | |
6714d8e8 KH |
569 | { |
570 | enum dlm_status status; | |
571 | struct dlm_lock_resource *res = NULL; | |
572 | struct dlm_lock *lock = NULL; | |
573 | int convert = 0, recovery = 0; | |
574 | ||
575 | /* yes this function is a mess. | |
576 | * TODO: clean this up. lots of common code in the | |
577 | * lock and convert paths, especially in the retry blocks */ | |
578 | if (!lksb) { | |
579 | dlm_error(DLM_BADARGS); | |
580 | return DLM_BADARGS; | |
581 | } | |
582 | ||
583 | status = DLM_BADPARAM; | |
584 | if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) { | |
585 | dlm_error(status); | |
586 | goto error; | |
587 | } | |
588 | ||
589 | if (flags & ~LKM_VALID_FLAGS) { | |
590 | dlm_error(status); | |
591 | goto error; | |
592 | } | |
593 | ||
594 | convert = (flags & LKM_CONVERT); | |
595 | recovery = (flags & LKM_RECOVERY); | |
596 | ||
597 | if (recovery && | |
3384f3df | 598 | (!dlm_is_recovery_lock(name, namelen) || convert) ) { |
6714d8e8 KH |
599 | dlm_error(status); |
600 | goto error; | |
601 | } | |
602 | if (convert && (flags & LKM_LOCAL)) { | |
603 | mlog(ML_ERROR, "strange LOCAL convert request!\n"); | |
604 | goto error; | |
605 | } | |
606 | ||
607 | if (convert) { | |
608 | /* CONVERT request */ | |
609 | ||
610 | /* if converting, must pass in a valid dlm_lock */ | |
611 | lock = lksb->lockid; | |
612 | if (!lock) { | |
613 | mlog(ML_ERROR, "NULL lock pointer in convert " | |
614 | "request\n"); | |
615 | goto error; | |
616 | } | |
617 | ||
618 | res = lock->lockres; | |
619 | if (!res) { | |
620 | mlog(ML_ERROR, "NULL lockres pointer in convert " | |
621 | "request\n"); | |
622 | goto error; | |
623 | } | |
624 | dlm_lockres_get(res); | |
625 | ||
626 | /* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are | |
627 | * static after the original lock call. convert requests will | |
628 | * ensure that everything is the same, or return DLM_BADARGS. | |
629 | * this means that DLM_DENIED_NOASTS will never be returned. | |
630 | */ | |
631 | if (lock->lksb != lksb || lock->ast != ast || | |
632 | lock->bast != bast || lock->astdata != data) { | |
633 | status = DLM_BADARGS; | |
634 | mlog(ML_ERROR, "new args: lksb=%p, ast=%p, bast=%p, " | |
635 | "astdata=%p\n", lksb, ast, bast, data); | |
636 | mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, " | |
637 | "astdata=%p\n", lock->lksb, lock->ast, | |
638 | lock->bast, lock->astdata); | |
639 | goto error; | |
640 | } | |
641 | retry_convert: | |
642 | dlm_wait_for_recovery(dlm); | |
643 | ||
644 | if (res->owner == dlm->node_num) | |
645 | status = dlmconvert_master(dlm, res, lock, flags, mode); | |
646 | else | |
647 | status = dlmconvert_remote(dlm, res, lock, flags, mode); | |
648 | if (status == DLM_RECOVERING || status == DLM_MIGRATING || | |
649 | status == DLM_FORWARD) { | |
650 | /* for now, see how this works without sleeping | |
651 | * and just retry right away. I suspect the reco | |
652 | * or migration will complete fast enough that | |
653 | * no waiting will be necessary */ | |
654 | mlog(0, "retrying convert with migration/recovery/" | |
655 | "in-progress\n"); | |
656 | msleep(100); | |
657 | goto retry_convert; | |
658 | } | |
659 | } else { | |
660 | u64 tmpcookie; | |
661 | ||
662 | /* LOCK request */ | |
663 | status = DLM_BADARGS; | |
664 | if (!name) { | |
665 | dlm_error(status); | |
666 | goto error; | |
667 | } | |
668 | ||
669 | status = DLM_IVBUFLEN; | |
3384f3df | 670 | if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) { |
6714d8e8 KH |
671 | dlm_error(status); |
672 | goto error; | |
673 | } | |
674 | ||
675 | dlm_get_next_cookie(dlm->node_num, &tmpcookie); | |
676 | lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb); | |
677 | if (!lock) { | |
678 | dlm_error(status); | |
679 | goto error; | |
680 | } | |
681 | ||
682 | if (!recovery) | |
683 | dlm_wait_for_recovery(dlm); | |
684 | ||
685 | /* find or create the lock resource */ | |
3384f3df | 686 | res = dlm_get_lock_resource(dlm, name, namelen, flags); |
6714d8e8 KH |
687 | if (!res) { |
688 | status = DLM_IVLOCKID; | |
689 | dlm_error(status); | |
690 | goto error; | |
691 | } | |
692 | ||
693 | mlog(0, "type=%d, flags = 0x%x\n", mode, flags); | |
694 | mlog(0, "creating lock: lock=%p res=%p\n", lock, res); | |
695 | ||
696 | dlm_lock_attach_lockres(lock, res); | |
697 | lock->ast = ast; | |
698 | lock->bast = bast; | |
699 | lock->astdata = data; | |
700 | ||
701 | retry_lock: | |
702 | if (flags & LKM_VALBLK) { | |
703 | mlog(0, "LKM_VALBLK passed by caller\n"); | |
704 | ||
705 | /* LVB requests for non PR, PW or EX locks are | |
706 | * ignored. */ | |
707 | if (mode < LKM_PRMODE) | |
708 | flags &= ~LKM_VALBLK; | |
709 | else { | |
710 | flags |= LKM_GET_LVB; | |
711 | lock->lksb->flags |= DLM_LKSB_GET_LVB; | |
712 | } | |
713 | } | |
714 | ||
715 | if (res->owner == dlm->node_num) | |
716 | status = dlmlock_master(dlm, res, lock, flags); | |
717 | else | |
718 | status = dlmlock_remote(dlm, res, lock, flags); | |
719 | ||
720 | if (status == DLM_RECOVERING || status == DLM_MIGRATING || | |
721 | status == DLM_FORWARD) { | |
722 | mlog(0, "retrying lock with migration/" | |
723 | "recovery/in progress\n"); | |
724 | msleep(100); | |
44465a7d KH |
725 | /* no waiting for dlm_reco_thread */ |
726 | if (recovery) { | |
c8df412e KH |
727 | if (status != DLM_RECOVERING) |
728 | goto retry_lock; | |
729 | ||
730 | mlog(0, "%s: got RECOVERING " | |
731 | "for $RECOVERY lock, master " | |
732 | "was %u\n", dlm->name, | |
733 | res->owner); | |
734 | /* wait to see the node go down, then | |
735 | * drop down and allow the lockres to | |
736 | * get cleaned up. need to remaster. */ | |
737 | dlm_wait_for_node_death(dlm, res->owner, | |
738 | DLM_NODE_DEATH_WAIT_MAX); | |
44465a7d KH |
739 | } else { |
740 | dlm_wait_for_recovery(dlm); | |
c8df412e | 741 | goto retry_lock; |
44465a7d | 742 | } |
6714d8e8 KH |
743 | } |
744 | ||
745 | if (status != DLM_NORMAL) { | |
746 | lock->lksb->flags &= ~DLM_LKSB_GET_LVB; | |
747 | if (status != DLM_NOTQUEUED) | |
748 | dlm_error(status); | |
749 | goto error; | |
750 | } | |
751 | } | |
752 | ||
753 | error: | |
754 | if (status != DLM_NORMAL) { | |
755 | if (lock && !convert) | |
756 | dlm_lock_put(lock); | |
757 | // this is kind of unnecessary | |
758 | lksb->status = status; | |
759 | } | |
760 | ||
761 | /* put lockres ref from the convert path | |
762 | * or from dlm_get_lock_resource */ | |
763 | if (res) | |
764 | dlm_lockres_put(res); | |
765 | ||
766 | return status; | |
767 | } | |
768 | EXPORT_SYMBOL_GPL(dlmlock); |