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