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; | |
66f45005 WW |
109 | if (!dlm_lock_compatible(tmplock->ml.convert_type, |
110 | lock->ml.type)) | |
111 | return 0; | |
6714d8e8 KH |
112 | } |
113 | ||
114 | return 1; | |
115 | } | |
116 | ||
117 | /* performs lock creation at the lockres master site | |
118 | * locking: | |
119 | * caller needs: none | |
120 | * taken: takes and drops res->spinlock | |
121 | * held on exit: none | |
122 | * returns: DLM_NORMAL, DLM_NOTQUEUED | |
123 | */ | |
124 | static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm, | |
125 | struct dlm_lock_resource *res, | |
126 | struct dlm_lock *lock, int flags) | |
127 | { | |
128 | int call_ast = 0, kick_thread = 0; | |
129 | enum dlm_status status = DLM_NORMAL; | |
130 | ||
ef6b689b | 131 | mlog(0, "type=%d\n", lock->ml.type); |
6714d8e8 KH |
132 | |
133 | spin_lock(&res->spinlock); | |
134 | /* if called from dlm_create_lock_handler, need to | |
135 | * ensure it will not sleep in dlm_wait_on_lockres */ | |
136 | status = __dlm_lockres_state_to_status(res); | |
137 | if (status != DLM_NORMAL && | |
138 | lock->ml.node != dlm->node_num) { | |
139 | /* erf. state changed after lock was dropped. */ | |
140 | spin_unlock(&res->spinlock); | |
141 | dlm_error(status); | |
142 | return status; | |
143 | } | |
144 | __dlm_wait_on_lockres(res); | |
145 | __dlm_lockres_reserve_ast(res); | |
146 | ||
147 | if (dlm_can_grant_new_lock(res, lock)) { | |
148 | mlog(0, "I can grant this lock right away\n"); | |
149 | /* got it right away */ | |
150 | lock->lksb->status = DLM_NORMAL; | |
151 | status = DLM_NORMAL; | |
152 | dlm_lock_get(lock); | |
153 | list_add_tail(&lock->list, &res->granted); | |
154 | ||
155 | /* for the recovery lock, we can't allow the ast | |
156 | * to be queued since the dlmthread is already | |
157 | * frozen. but the recovery lock is always locked | |
158 | * with LKM_NOQUEUE so we do not need the ast in | |
159 | * this special case */ | |
160 | if (!dlm_is_recovery_lock(res->lockname.name, | |
161 | res->lockname.len)) { | |
162 | kick_thread = 1; | |
163 | call_ast = 1; | |
c03872f5 KH |
164 | } else { |
165 | mlog(0, "%s: returning DLM_NORMAL to " | |
166 | "node %u for reco lock\n", dlm->name, | |
167 | lock->ml.node); | |
6714d8e8 KH |
168 | } |
169 | } else { | |
170 | /* for NOQUEUE request, unless we get the | |
171 | * lock right away, return DLM_NOTQUEUED */ | |
c03872f5 | 172 | if (flags & LKM_NOQUEUE) { |
6714d8e8 | 173 | status = DLM_NOTQUEUED; |
c03872f5 KH |
174 | if (dlm_is_recovery_lock(res->lockname.name, |
175 | res->lockname.len)) { | |
176 | mlog(0, "%s: returning NOTQUEUED to " | |
177 | "node %u for reco lock\n", dlm->name, | |
178 | lock->ml.node); | |
179 | } | |
180 | } else { | |
6714d8e8 KH |
181 | dlm_lock_get(lock); |
182 | list_add_tail(&lock->list, &res->blocked); | |
183 | kick_thread = 1; | |
184 | } | |
185 | } | |
186 | ||
187 | spin_unlock(&res->spinlock); | |
188 | wake_up(&res->wq); | |
189 | ||
190 | /* either queue the ast or release it */ | |
191 | if (call_ast) | |
192 | dlm_queue_ast(dlm, lock); | |
193 | else | |
194 | dlm_lockres_release_ast(dlm, res); | |
195 | ||
196 | dlm_lockres_calc_usage(dlm, res); | |
197 | if (kick_thread) | |
198 | dlm_kick_thread(dlm, res); | |
199 | ||
200 | return status; | |
201 | } | |
202 | ||
203 | void dlm_revert_pending_lock(struct dlm_lock_resource *res, | |
204 | struct dlm_lock *lock) | |
205 | { | |
206 | /* remove from local queue if it failed */ | |
207 | list_del_init(&lock->list); | |
208 | lock->lksb->flags &= ~DLM_LKSB_GET_LVB; | |
209 | } | |
210 | ||
211 | ||
212 | /* | |
213 | * locking: | |
214 | * caller needs: none | |
215 | * taken: takes and drops res->spinlock | |
216 | * held on exit: none | |
217 | * returns: DLM_DENIED, DLM_RECOVERING, or net status | |
218 | */ | |
219 | static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm, | |
220 | struct dlm_lock_resource *res, | |
221 | struct dlm_lock *lock, int flags) | |
222 | { | |
223 | enum dlm_status status = DLM_DENIED; | |
2abaf97e | 224 | int lockres_changed = 1; |
6714d8e8 | 225 | |
ef6b689b TM |
226 | mlog(0, "type=%d, lockres %.*s, flags = 0x%x\n", |
227 | lock->ml.type, res->lockname.len, | |
6714d8e8 KH |
228 | res->lockname.name, flags); |
229 | ||
a2c0cc15 SM |
230 | /* |
231 | * Wait if resource is getting recovered, remastered, etc. | |
232 | * If the resource was remastered and new owner is self, then exit. | |
233 | */ | |
6714d8e8 | 234 | spin_lock(&res->spinlock); |
6714d8e8 | 235 | __dlm_wait_on_lockres(res); |
a2c0cc15 SM |
236 | if (res->owner == dlm->node_num) { |
237 | spin_unlock(&res->spinlock); | |
238 | return DLM_RECOVERING; | |
239 | } | |
6714d8e8 KH |
240 | res->state |= DLM_LOCK_RES_IN_PROGRESS; |
241 | ||
242 | /* add lock to local (secondary) queue */ | |
243 | dlm_lock_get(lock); | |
244 | list_add_tail(&lock->list, &res->blocked); | |
245 | lock->lock_pending = 1; | |
246 | spin_unlock(&res->spinlock); | |
247 | ||
248 | /* spec seems to say that you will get DLM_NORMAL when the lock | |
249 | * has been queued, meaning we need to wait for a reply here. */ | |
250 | status = dlm_send_remote_lock_request(dlm, res, lock, flags); | |
251 | ||
252 | spin_lock(&res->spinlock); | |
253 | res->state &= ~DLM_LOCK_RES_IN_PROGRESS; | |
254 | lock->lock_pending = 0; | |
255 | if (status != DLM_NORMAL) { | |
c8df412e KH |
256 | if (status == DLM_RECOVERING && |
257 | dlm_is_recovery_lock(res->lockname.name, | |
258 | res->lockname.len)) { | |
259 | /* recovery lock was mastered by dead node. | |
260 | * we need to have calc_usage shoot down this | |
261 | * lockres and completely remaster it. */ | |
262 | mlog(0, "%s: recovery lock was owned by " | |
263 | "dead node %u, remaster it now.\n", | |
264 | dlm->name, res->owner); | |
265 | } else if (status != DLM_NOTQUEUED) { | |
c87a9ae7 KH |
266 | /* |
267 | * DO NOT call calc_usage, as this would unhash | |
268 | * the remote lockres before we ever get to use | |
269 | * it. treat as if we never made any change to | |
270 | * the lockres. | |
271 | */ | |
272 | lockres_changed = 0; | |
6714d8e8 | 273 | dlm_error(status); |
c87a9ae7 | 274 | } |
6714d8e8 KH |
275 | dlm_revert_pending_lock(res, lock); |
276 | dlm_lock_put(lock); | |
2bd63216 | 277 | } else if (dlm_is_recovery_lock(res->lockname.name, |
558c70c5 KH |
278 | res->lockname.len)) { |
279 | /* special case for the $RECOVERY lock. | |
280 | * there will never be an AST delivered to put | |
281 | * this lock on the proper secondary queue | |
282 | * (granted), so do it manually. */ | |
283 | mlog(0, "%s: $RECOVERY lock for this node (%u) is " | |
284 | "mastered by %u; got lock, manually granting (no ast)\n", | |
285 | dlm->name, dlm->node_num, res->owner); | |
f116629d | 286 | list_move_tail(&lock->list, &res->granted); |
6714d8e8 KH |
287 | } |
288 | spin_unlock(&res->spinlock); | |
289 | ||
2abaf97e KH |
290 | if (lockres_changed) |
291 | dlm_lockres_calc_usage(dlm, res); | |
6714d8e8 KH |
292 | |
293 | wake_up(&res->wq); | |
294 | return status; | |
295 | } | |
296 | ||
297 | ||
298 | /* for remote lock creation. | |
299 | * locking: | |
300 | * caller needs: none, but need res->state & DLM_LOCK_RES_IN_PROGRESS | |
301 | * taken: none | |
302 | * held on exit: none | |
303 | * returns: DLM_NOLOCKMGR, or net status | |
304 | */ | |
305 | static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm, | |
306 | struct dlm_lock_resource *res, | |
307 | struct dlm_lock *lock, int flags) | |
308 | { | |
309 | struct dlm_create_lock create; | |
310 | int tmpret, status = 0; | |
311 | enum dlm_status ret; | |
312 | ||
6714d8e8 KH |
313 | memset(&create, 0, sizeof(create)); |
314 | create.node_idx = dlm->node_num; | |
315 | create.requested_type = lock->ml.type; | |
316 | create.cookie = lock->ml.cookie; | |
317 | create.namelen = res->lockname.len; | |
318 | create.flags = cpu_to_be32(flags); | |
319 | memcpy(create.name, res->lockname.name, create.namelen); | |
320 | ||
321 | tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create, | |
322 | sizeof(create), res->owner, &status); | |
323 | if (tmpret >= 0) { | |
8decab3c | 324 | ret = status; |
495ac96e | 325 | if (ret == DLM_REJECTED) { |
8decab3c SM |
326 | mlog(ML_ERROR, "%s: res %.*s, Stale lockres no longer " |
327 | "owned by node %u. That node is coming back up " | |
328 | "currently.\n", dlm->name, create.namelen, | |
e4eb0368 KH |
329 | create.name, res->owner); |
330 | dlm_print_one_lock_resource(res); | |
331 | BUG(); | |
332 | } | |
6714d8e8 | 333 | } else { |
8decab3c SM |
334 | mlog(ML_ERROR, "%s: res %.*s, Error %d send CREATE LOCK to " |
335 | "node %u\n", dlm->name, create.namelen, create.name, | |
336 | tmpret, res->owner); | |
337 | if (dlm_is_host_down(tmpret)) | |
6714d8e8 | 338 | ret = DLM_RECOVERING; |
8decab3c | 339 | else |
6714d8e8 | 340 | ret = dlm_err_to_dlm_status(tmpret); |
6714d8e8 KH |
341 | } |
342 | ||
343 | return ret; | |
344 | } | |
345 | ||
346 | void dlm_lock_get(struct dlm_lock *lock) | |
347 | { | |
348 | kref_get(&lock->lock_refs); | |
349 | } | |
350 | ||
351 | void dlm_lock_put(struct dlm_lock *lock) | |
352 | { | |
353 | kref_put(&lock->lock_refs, dlm_lock_release); | |
354 | } | |
355 | ||
356 | static void dlm_lock_release(struct kref *kref) | |
357 | { | |
358 | struct dlm_lock *lock; | |
359 | ||
360 | lock = container_of(kref, struct dlm_lock, lock_refs); | |
361 | ||
362 | BUG_ON(!list_empty(&lock->list)); | |
363 | BUG_ON(!list_empty(&lock->ast_list)); | |
364 | BUG_ON(!list_empty(&lock->bast_list)); | |
365 | BUG_ON(lock->ast_pending); | |
366 | BUG_ON(lock->bast_pending); | |
367 | ||
368 | dlm_lock_detach_lockres(lock); | |
369 | ||
370 | if (lock->lksb_kernel_allocated) { | |
371 | mlog(0, "freeing kernel-allocated lksb\n"); | |
372 | kfree(lock->lksb); | |
373 | } | |
724bdca9 | 374 | kmem_cache_free(dlm_lock_cache, lock); |
6714d8e8 KH |
375 | } |
376 | ||
377 | /* associate a lock with it's lockres, getting a ref on the lockres */ | |
378 | void dlm_lock_attach_lockres(struct dlm_lock *lock, | |
379 | struct dlm_lock_resource *res) | |
380 | { | |
381 | dlm_lockres_get(res); | |
382 | lock->lockres = res; | |
383 | } | |
384 | ||
385 | /* drop ref on lockres, if there is still one associated with lock */ | |
386 | static void dlm_lock_detach_lockres(struct dlm_lock *lock) | |
387 | { | |
388 | struct dlm_lock_resource *res; | |
389 | ||
390 | res = lock->lockres; | |
391 | if (res) { | |
392 | lock->lockres = NULL; | |
393 | mlog(0, "removing lock's lockres reference\n"); | |
394 | dlm_lockres_put(res); | |
395 | } | |
396 | } | |
397 | ||
398 | static void dlm_init_lock(struct dlm_lock *newlock, int type, | |
399 | u8 node, u64 cookie) | |
400 | { | |
401 | INIT_LIST_HEAD(&newlock->list); | |
402 | INIT_LIST_HEAD(&newlock->ast_list); | |
403 | INIT_LIST_HEAD(&newlock->bast_list); | |
404 | spin_lock_init(&newlock->spinlock); | |
405 | newlock->ml.type = type; | |
406 | newlock->ml.convert_type = LKM_IVMODE; | |
407 | newlock->ml.highest_blocked = LKM_IVMODE; | |
408 | newlock->ml.node = node; | |
409 | newlock->ml.pad1 = 0; | |
410 | newlock->ml.list = 0; | |
411 | newlock->ml.flags = 0; | |
412 | newlock->ast = NULL; | |
413 | newlock->bast = NULL; | |
414 | newlock->astdata = NULL; | |
415 | newlock->ml.cookie = cpu_to_be64(cookie); | |
416 | newlock->ast_pending = 0; | |
417 | newlock->bast_pending = 0; | |
418 | newlock->convert_pending = 0; | |
419 | newlock->lock_pending = 0; | |
420 | newlock->unlock_pending = 0; | |
421 | newlock->cancel_pending = 0; | |
422 | newlock->lksb_kernel_allocated = 0; | |
423 | ||
424 | kref_init(&newlock->lock_refs); | |
425 | } | |
426 | ||
427 | struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie, | |
428 | struct dlm_lockstatus *lksb) | |
429 | { | |
430 | struct dlm_lock *lock; | |
431 | int kernel_allocated = 0; | |
432 | ||
3914ed0c | 433 | lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS); |
6714d8e8 KH |
434 | if (!lock) |
435 | return NULL; | |
436 | ||
437 | if (!lksb) { | |
438 | /* zero memory only if kernel-allocated */ | |
cd861280 | 439 | lksb = kzalloc(sizeof(*lksb), GFP_NOFS); |
6714d8e8 KH |
440 | if (!lksb) { |
441 | kfree(lock); | |
442 | return NULL; | |
443 | } | |
444 | kernel_allocated = 1; | |
445 | } | |
446 | ||
447 | dlm_init_lock(lock, type, node, cookie); | |
448 | if (kernel_allocated) | |
449 | lock->lksb_kernel_allocated = 1; | |
450 | lock->lksb = lksb; | |
451 | lksb->lockid = lock; | |
452 | return lock; | |
453 | } | |
454 | ||
455 | /* handler for lock creation net message | |
456 | * locking: | |
457 | * caller needs: none | |
458 | * taken: takes and drops res->spinlock | |
459 | * held on exit: none | |
460 | * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED | |
461 | */ | |
d74c9803 KH |
462 | int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data, |
463 | void **ret_data) | |
6714d8e8 KH |
464 | { |
465 | struct dlm_ctxt *dlm = data; | |
466 | struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf; | |
467 | struct dlm_lock_resource *res = NULL; | |
468 | struct dlm_lock *newlock = NULL; | |
469 | struct dlm_lockstatus *lksb = NULL; | |
470 | enum dlm_status status = DLM_NORMAL; | |
471 | char *name; | |
472 | unsigned int namelen; | |
473 | ||
474 | BUG_ON(!dlm); | |
475 | ||
6714d8e8 KH |
476 | if (!dlm_grab(dlm)) |
477 | return DLM_REJECTED; | |
478 | ||
6714d8e8 KH |
479 | name = create->name; |
480 | namelen = create->namelen; | |
495ac96e | 481 | status = DLM_REJECTED; |
e4eb0368 KH |
482 | if (!dlm_domain_fully_joined(dlm)) { |
483 | mlog(ML_ERROR, "Domain %s not fully joined, but node %u is " | |
484 | "sending a create_lock message for lock %.*s!\n", | |
485 | dlm->name, create->node_idx, namelen, name); | |
486 | dlm_error(status); | |
487 | goto leave; | |
488 | } | |
6714d8e8 KH |
489 | |
490 | status = DLM_IVBUFLEN; | |
491 | if (namelen > DLM_LOCKID_NAME_MAX) { | |
492 | dlm_error(status); | |
493 | goto leave; | |
494 | } | |
495 | ||
496 | status = DLM_SYSERR; | |
497 | newlock = dlm_new_lock(create->requested_type, | |
498 | create->node_idx, | |
499 | be64_to_cpu(create->cookie), NULL); | |
500 | if (!newlock) { | |
501 | dlm_error(status); | |
502 | goto leave; | |
503 | } | |
504 | ||
505 | lksb = newlock->lksb; | |
506 | ||
507 | if (be32_to_cpu(create->flags) & LKM_GET_LVB) { | |
508 | lksb->flags |= DLM_LKSB_GET_LVB; | |
509 | mlog(0, "set DLM_LKSB_GET_LVB flag\n"); | |
510 | } | |
511 | ||
512 | status = DLM_IVLOCKID; | |
513 | res = dlm_lookup_lockres(dlm, name, namelen); | |
514 | if (!res) { | |
515 | dlm_error(status); | |
516 | goto leave; | |
517 | } | |
518 | ||
519 | spin_lock(&res->spinlock); | |
520 | status = __dlm_lockres_state_to_status(res); | |
521 | spin_unlock(&res->spinlock); | |
522 | ||
523 | if (status != DLM_NORMAL) { | |
524 | mlog(0, "lockres recovering/migrating/in-progress\n"); | |
525 | goto leave; | |
526 | } | |
527 | ||
528 | dlm_lock_attach_lockres(newlock, res); | |
529 | ||
530 | status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags)); | |
531 | leave: | |
532 | if (status != DLM_NORMAL) | |
533 | if (newlock) | |
534 | dlm_lock_put(newlock); | |
535 | ||
536 | if (res) | |
537 | dlm_lockres_put(res); | |
538 | ||
539 | dlm_put(dlm); | |
540 | ||
541 | return status; | |
542 | } | |
543 | ||
544 | ||
545 | /* fetch next node-local (u8 nodenum + u56 cookie) into u64 */ | |
546 | static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie) | |
547 | { | |
548 | u64 tmpnode = node_num; | |
549 | ||
550 | /* shift single byte of node num into top 8 bits */ | |
551 | tmpnode <<= 56; | |
552 | ||
553 | spin_lock(&dlm_cookie_lock); | |
554 | *cookie = (dlm_next_cookie | tmpnode); | |
555 | if (++dlm_next_cookie & 0xff00000000000000ull) { | |
556 | mlog(0, "This node's cookie will now wrap!\n"); | |
557 | dlm_next_cookie = 1; | |
558 | } | |
559 | spin_unlock(&dlm_cookie_lock); | |
560 | } | |
561 | ||
562 | enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode, | |
563 | struct dlm_lockstatus *lksb, int flags, | |
3384f3df MF |
564 | const char *name, int namelen, dlm_astlockfunc_t *ast, |
565 | void *data, dlm_bastlockfunc_t *bast) | |
6714d8e8 KH |
566 | { |
567 | enum dlm_status status; | |
568 | struct dlm_lock_resource *res = NULL; | |
569 | struct dlm_lock *lock = NULL; | |
570 | int convert = 0, recovery = 0; | |
571 | ||
572 | /* yes this function is a mess. | |
573 | * TODO: clean this up. lots of common code in the | |
574 | * lock and convert paths, especially in the retry blocks */ | |
575 | if (!lksb) { | |
576 | dlm_error(DLM_BADARGS); | |
577 | return DLM_BADARGS; | |
578 | } | |
579 | ||
580 | status = DLM_BADPARAM; | |
581 | if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) { | |
582 | dlm_error(status); | |
583 | goto error; | |
584 | } | |
585 | ||
586 | if (flags & ~LKM_VALID_FLAGS) { | |
587 | dlm_error(status); | |
588 | goto error; | |
589 | } | |
590 | ||
591 | convert = (flags & LKM_CONVERT); | |
592 | recovery = (flags & LKM_RECOVERY); | |
593 | ||
594 | if (recovery && | |
3384f3df | 595 | (!dlm_is_recovery_lock(name, namelen) || convert) ) { |
6714d8e8 KH |
596 | dlm_error(status); |
597 | goto error; | |
598 | } | |
599 | if (convert && (flags & LKM_LOCAL)) { | |
600 | mlog(ML_ERROR, "strange LOCAL convert request!\n"); | |
601 | goto error; | |
602 | } | |
603 | ||
604 | if (convert) { | |
605 | /* CONVERT request */ | |
606 | ||
607 | /* if converting, must pass in a valid dlm_lock */ | |
608 | lock = lksb->lockid; | |
609 | if (!lock) { | |
610 | mlog(ML_ERROR, "NULL lock pointer in convert " | |
611 | "request\n"); | |
612 | goto error; | |
613 | } | |
614 | ||
615 | res = lock->lockres; | |
616 | if (!res) { | |
617 | mlog(ML_ERROR, "NULL lockres pointer in convert " | |
618 | "request\n"); | |
619 | goto error; | |
620 | } | |
621 | dlm_lockres_get(res); | |
622 | ||
623 | /* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are | |
624 | * static after the original lock call. convert requests will | |
625 | * ensure that everything is the same, or return DLM_BADARGS. | |
626 | * this means that DLM_DENIED_NOASTS will never be returned. | |
627 | */ | |
628 | if (lock->lksb != lksb || lock->ast != ast || | |
629 | lock->bast != bast || lock->astdata != data) { | |
630 | status = DLM_BADARGS; | |
631 | mlog(ML_ERROR, "new args: lksb=%p, ast=%p, bast=%p, " | |
632 | "astdata=%p\n", lksb, ast, bast, data); | |
633 | mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, " | |
634 | "astdata=%p\n", lock->lksb, lock->ast, | |
635 | lock->bast, lock->astdata); | |
636 | goto error; | |
637 | } | |
638 | retry_convert: | |
639 | dlm_wait_for_recovery(dlm); | |
640 | ||
641 | if (res->owner == dlm->node_num) | |
642 | status = dlmconvert_master(dlm, res, lock, flags, mode); | |
643 | else | |
644 | status = dlmconvert_remote(dlm, res, lock, flags, mode); | |
645 | if (status == DLM_RECOVERING || status == DLM_MIGRATING || | |
646 | status == DLM_FORWARD) { | |
647 | /* for now, see how this works without sleeping | |
648 | * and just retry right away. I suspect the reco | |
649 | * or migration will complete fast enough that | |
650 | * no waiting will be necessary */ | |
651 | mlog(0, "retrying convert with migration/recovery/" | |
652 | "in-progress\n"); | |
653 | msleep(100); | |
654 | goto retry_convert; | |
655 | } | |
656 | } else { | |
657 | u64 tmpcookie; | |
658 | ||
659 | /* LOCK request */ | |
660 | status = DLM_BADARGS; | |
661 | if (!name) { | |
662 | dlm_error(status); | |
663 | goto error; | |
664 | } | |
665 | ||
666 | status = DLM_IVBUFLEN; | |
3384f3df | 667 | if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) { |
6714d8e8 KH |
668 | dlm_error(status); |
669 | goto error; | |
670 | } | |
671 | ||
672 | dlm_get_next_cookie(dlm->node_num, &tmpcookie); | |
673 | lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb); | |
674 | if (!lock) { | |
675 | dlm_error(status); | |
676 | goto error; | |
677 | } | |
678 | ||
679 | if (!recovery) | |
680 | dlm_wait_for_recovery(dlm); | |
681 | ||
682 | /* find or create the lock resource */ | |
3384f3df | 683 | res = dlm_get_lock_resource(dlm, name, namelen, flags); |
6714d8e8 KH |
684 | if (!res) { |
685 | status = DLM_IVLOCKID; | |
686 | dlm_error(status); | |
687 | goto error; | |
688 | } | |
689 | ||
690 | mlog(0, "type=%d, flags = 0x%x\n", mode, flags); | |
691 | mlog(0, "creating lock: lock=%p res=%p\n", lock, res); | |
692 | ||
693 | dlm_lock_attach_lockres(lock, res); | |
694 | lock->ast = ast; | |
695 | lock->bast = bast; | |
696 | lock->astdata = data; | |
697 | ||
698 | retry_lock: | |
699 | if (flags & LKM_VALBLK) { | |
700 | mlog(0, "LKM_VALBLK passed by caller\n"); | |
701 | ||
702 | /* LVB requests for non PR, PW or EX locks are | |
703 | * ignored. */ | |
704 | if (mode < LKM_PRMODE) | |
705 | flags &= ~LKM_VALBLK; | |
706 | else { | |
707 | flags |= LKM_GET_LVB; | |
708 | lock->lksb->flags |= DLM_LKSB_GET_LVB; | |
709 | } | |
710 | } | |
711 | ||
712 | if (res->owner == dlm->node_num) | |
713 | status = dlmlock_master(dlm, res, lock, flags); | |
714 | else | |
715 | status = dlmlock_remote(dlm, res, lock, flags); | |
716 | ||
717 | if (status == DLM_RECOVERING || status == DLM_MIGRATING || | |
718 | status == DLM_FORWARD) { | |
6714d8e8 | 719 | msleep(100); |
44465a7d | 720 | if (recovery) { |
c8df412e KH |
721 | if (status != DLM_RECOVERING) |
722 | goto retry_lock; | |
c8df412e KH |
723 | /* wait to see the node go down, then |
724 | * drop down and allow the lockres to | |
725 | * get cleaned up. need to remaster. */ | |
726 | dlm_wait_for_node_death(dlm, res->owner, | |
727 | DLM_NODE_DEATH_WAIT_MAX); | |
44465a7d KH |
728 | } else { |
729 | dlm_wait_for_recovery(dlm); | |
c8df412e | 730 | goto retry_lock; |
44465a7d | 731 | } |
6714d8e8 KH |
732 | } |
733 | ||
ff0a522e SM |
734 | /* Inflight taken in dlm_get_lock_resource() is dropped here */ |
735 | spin_lock(&res->spinlock); | |
736 | dlm_lockres_drop_inflight_ref(dlm, res); | |
737 | spin_unlock(&res->spinlock); | |
738 | ||
739 | dlm_lockres_calc_usage(dlm, res); | |
740 | dlm_kick_thread(dlm, res); | |
741 | ||
6714d8e8 KH |
742 | if (status != DLM_NORMAL) { |
743 | lock->lksb->flags &= ~DLM_LKSB_GET_LVB; | |
744 | if (status != DLM_NOTQUEUED) | |
745 | dlm_error(status); | |
746 | goto error; | |
747 | } | |
748 | } | |
749 | ||
750 | error: | |
751 | if (status != DLM_NORMAL) { | |
752 | if (lock && !convert) | |
753 | dlm_lock_put(lock); | |
754 | // this is kind of unnecessary | |
755 | lksb->status = status; | |
756 | } | |
757 | ||
758 | /* put lockres ref from the convert path | |
759 | * or from dlm_get_lock_resource */ | |
760 | if (res) | |
761 | dlm_lockres_put(res); | |
762 | ||
763 | return status; | |
764 | } | |
765 | EXPORT_SYMBOL_GPL(dlmlock); |