xref: /openbmc/linux/fs/ocfs2/dlm/dlmlock.c (revision ae213c44)
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/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 
55 static struct kmem_cache *dlm_lock_cache;
56 
57 static DEFINE_SPINLOCK(dlm_cookie_lock);
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 
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 	kmem_cache_destroy(dlm_lock_cache);
81 }
82 
83 /* Tell us whether we can grant a new lock request.
84  * locking:
85  *   caller needs:  res->spinlock
86  *   taken:         none
87  *   held on exit:  none
88  * returns: 1 if the lock can be granted, 0 otherwise.
89  */
90 static int dlm_can_grant_new_lock(struct dlm_lock_resource *res,
91 				  struct dlm_lock *lock)
92 {
93 	struct dlm_lock *tmplock;
94 
95 	list_for_each_entry(tmplock, &res->granted, list) {
96 		if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
97 			return 0;
98 	}
99 
100 	list_for_each_entry(tmplock, &res->converting, list) {
101 		if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
102 			return 0;
103 		if (!dlm_lock_compatible(tmplock->ml.convert_type,
104 					 lock->ml.type))
105 			return 0;
106 	}
107 
108 	return 1;
109 }
110 
111 /* performs lock creation at the lockres master site
112  * locking:
113  *   caller needs:  none
114  *   taken:         takes and drops res->spinlock
115  *   held on exit:  none
116  * returns: DLM_NORMAL, DLM_NOTQUEUED
117  */
118 static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
119 				      struct dlm_lock_resource *res,
120 				      struct dlm_lock *lock, int flags)
121 {
122 	int call_ast = 0, kick_thread = 0;
123 	enum dlm_status status = DLM_NORMAL;
124 
125 	mlog(0, "type=%d\n", lock->ml.type);
126 
127 	spin_lock(&res->spinlock);
128 	/* if called from dlm_create_lock_handler, need to
129 	 * ensure it will not sleep in dlm_wait_on_lockres */
130 	status = __dlm_lockres_state_to_status(res);
131 	if (status != DLM_NORMAL &&
132 	    lock->ml.node != dlm->node_num) {
133 		/* erf.  state changed after lock was dropped. */
134 		spin_unlock(&res->spinlock);
135 		dlm_error(status);
136 		return status;
137 	}
138 	__dlm_wait_on_lockres(res);
139 	__dlm_lockres_reserve_ast(res);
140 
141 	if (dlm_can_grant_new_lock(res, lock)) {
142 		mlog(0, "I can grant this lock right away\n");
143 		/* got it right away */
144 		lock->lksb->status = DLM_NORMAL;
145 		status = DLM_NORMAL;
146 		dlm_lock_get(lock);
147 		list_add_tail(&lock->list, &res->granted);
148 
149 		/* for the recovery lock, we can't allow the ast
150 		 * to be queued since the dlmthread is already
151 		 * frozen.  but the recovery lock is always locked
152 		 * with LKM_NOQUEUE so we do not need the ast in
153 		 * this special case */
154 		if (!dlm_is_recovery_lock(res->lockname.name,
155 					  res->lockname.len)) {
156 			kick_thread = 1;
157 			call_ast = 1;
158 		} else {
159 			mlog(0, "%s: returning DLM_NORMAL to "
160 			     "node %u for reco lock\n", dlm->name,
161 			     lock->ml.node);
162 		}
163 	} else {
164 		/* for NOQUEUE request, unless we get the
165 		 * lock right away, return DLM_NOTQUEUED */
166 		if (flags & LKM_NOQUEUE) {
167 			status = DLM_NOTQUEUED;
168 			if (dlm_is_recovery_lock(res->lockname.name,
169 						 res->lockname.len)) {
170 				mlog(0, "%s: returning NOTQUEUED to "
171 				     "node %u for reco lock\n", dlm->name,
172 				     lock->ml.node);
173 			}
174 		} else {
175 			status = DLM_NORMAL;
176 			dlm_lock_get(lock);
177 			list_add_tail(&lock->list, &res->blocked);
178 			kick_thread = 1;
179 		}
180 	}
181 
182 	spin_unlock(&res->spinlock);
183 	wake_up(&res->wq);
184 
185 	/* either queue the ast or release it */
186 	if (call_ast)
187 		dlm_queue_ast(dlm, lock);
188 	else
189 		dlm_lockres_release_ast(dlm, res);
190 
191 	dlm_lockres_calc_usage(dlm, res);
192 	if (kick_thread)
193 		dlm_kick_thread(dlm, res);
194 
195 	return status;
196 }
197 
198 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
199 			     struct dlm_lock *lock)
200 {
201 	/* remove from local queue if it failed */
202 	list_del_init(&lock->list);
203 	lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
204 }
205 
206 
207 /*
208  * locking:
209  *   caller needs:  none
210  *   taken:         takes and drops res->spinlock
211  *   held on exit:  none
212  * returns: DLM_DENIED, DLM_RECOVERING, or net status
213  */
214 static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
215 				      struct dlm_lock_resource *res,
216 				      struct dlm_lock *lock, int flags)
217 {
218 	enum dlm_status status = DLM_DENIED;
219 	int lockres_changed = 1;
220 
221 	mlog(0, "type=%d, lockres %.*s, flags = 0x%x\n",
222 	     lock->ml.type, res->lockname.len,
223 	     res->lockname.name, flags);
224 
225 	/*
226 	 * Wait if resource is getting recovered, remastered, etc.
227 	 * If the resource was remastered and new owner is self, then exit.
228 	 */
229 	spin_lock(&res->spinlock);
230 	__dlm_wait_on_lockres(res);
231 	if (res->owner == dlm->node_num) {
232 		spin_unlock(&res->spinlock);
233 		return DLM_RECOVERING;
234 	}
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) {
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) {
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;
268 			dlm_error(status);
269 		}
270 		dlm_revert_pending_lock(res, lock);
271 		dlm_lock_put(lock);
272 	} else if (dlm_is_recovery_lock(res->lockname.name,
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);
281 		list_move_tail(&lock->list, &res->granted);
282 	}
283 	spin_unlock(&res->spinlock);
284 
285 	if (lockres_changed)
286 		dlm_lockres_calc_usage(dlm, res);
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 	memset(&create, 0, sizeof(create));
309 	create.node_idx = dlm->node_num;
310 	create.requested_type = lock->ml.type;
311 	create.cookie = lock->ml.cookie;
312 	create.namelen = res->lockname.len;
313 	create.flags = cpu_to_be32(flags);
314 	memcpy(create.name, res->lockname.name, create.namelen);
315 
316 	tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create,
317 				    sizeof(create), res->owner, &status);
318 	if (tmpret >= 0) {
319 		ret = status;
320 		if (ret == DLM_REJECTED) {
321 			mlog(ML_ERROR, "%s: res %.*s, Stale lockres no longer "
322 			     "owned by node %u. That node is coming back up "
323 			     "currently.\n", dlm->name, create.namelen,
324 			     create.name, res->owner);
325 			dlm_print_one_lock_resource(res);
326 			BUG();
327 		}
328 	} else {
329 		mlog(ML_ERROR, "%s: res %.*s, Error %d send CREATE LOCK to "
330 		     "node %u\n", dlm->name, create.namelen, create.name,
331 		     tmpret, res->owner);
332 		if (dlm_is_host_down(tmpret))
333 			ret = DLM_RECOVERING;
334 		else
335 			ret = dlm_err_to_dlm_status(tmpret);
336 	}
337 
338 	return ret;
339 }
340 
341 void dlm_lock_get(struct dlm_lock *lock)
342 {
343 	kref_get(&lock->lock_refs);
344 }
345 
346 void dlm_lock_put(struct dlm_lock *lock)
347 {
348 	kref_put(&lock->lock_refs, dlm_lock_release);
349 }
350 
351 static void dlm_lock_release(struct kref *kref)
352 {
353 	struct dlm_lock *lock;
354 
355 	lock = container_of(kref, struct dlm_lock, lock_refs);
356 
357 	BUG_ON(!list_empty(&lock->list));
358 	BUG_ON(!list_empty(&lock->ast_list));
359 	BUG_ON(!list_empty(&lock->bast_list));
360 	BUG_ON(lock->ast_pending);
361 	BUG_ON(lock->bast_pending);
362 
363 	dlm_lock_detach_lockres(lock);
364 
365 	if (lock->lksb_kernel_allocated) {
366 		mlog(0, "freeing kernel-allocated lksb\n");
367 		kfree(lock->lksb);
368 	}
369 	kmem_cache_free(dlm_lock_cache, lock);
370 }
371 
372 /* associate a lock with it's lockres, getting a ref on the lockres */
373 void dlm_lock_attach_lockres(struct dlm_lock *lock,
374 			     struct dlm_lock_resource *res)
375 {
376 	dlm_lockres_get(res);
377 	lock->lockres = res;
378 }
379 
380 /* drop ref on lockres, if there is still one associated with lock */
381 static void dlm_lock_detach_lockres(struct dlm_lock *lock)
382 {
383 	struct dlm_lock_resource *res;
384 
385 	res = lock->lockres;
386 	if (res) {
387 		lock->lockres = NULL;
388 		mlog(0, "removing lock's lockres reference\n");
389 		dlm_lockres_put(res);
390 	}
391 }
392 
393 static void dlm_init_lock(struct dlm_lock *newlock, int type,
394 			  u8 node, u64 cookie)
395 {
396 	INIT_LIST_HEAD(&newlock->list);
397 	INIT_LIST_HEAD(&newlock->ast_list);
398 	INIT_LIST_HEAD(&newlock->bast_list);
399 	spin_lock_init(&newlock->spinlock);
400 	newlock->ml.type = type;
401 	newlock->ml.convert_type = LKM_IVMODE;
402 	newlock->ml.highest_blocked = LKM_IVMODE;
403 	newlock->ml.node = node;
404 	newlock->ml.pad1 = 0;
405 	newlock->ml.list = 0;
406 	newlock->ml.flags = 0;
407 	newlock->ast = NULL;
408 	newlock->bast = NULL;
409 	newlock->astdata = NULL;
410 	newlock->ml.cookie = cpu_to_be64(cookie);
411 	newlock->ast_pending = 0;
412 	newlock->bast_pending = 0;
413 	newlock->convert_pending = 0;
414 	newlock->lock_pending = 0;
415 	newlock->unlock_pending = 0;
416 	newlock->cancel_pending = 0;
417 	newlock->lksb_kernel_allocated = 0;
418 
419 	kref_init(&newlock->lock_refs);
420 }
421 
422 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
423 			       struct dlm_lockstatus *lksb)
424 {
425 	struct dlm_lock *lock;
426 	int kernel_allocated = 0;
427 
428 	lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS);
429 	if (!lock)
430 		return NULL;
431 
432 	if (!lksb) {
433 		/* zero memory only if kernel-allocated */
434 		lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
435 		if (!lksb) {
436 			kmem_cache_free(dlm_lock_cache, lock);
437 			return NULL;
438 		}
439 		kernel_allocated = 1;
440 	}
441 
442 	dlm_init_lock(lock, type, node, cookie);
443 	if (kernel_allocated)
444 		lock->lksb_kernel_allocated = 1;
445 	lock->lksb = lksb;
446 	lksb->lockid = lock;
447 	return lock;
448 }
449 
450 /* handler for lock creation net message
451  * locking:
452  *   caller needs:  none
453  *   taken:         takes and drops res->spinlock
454  *   held on exit:  none
455  * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED
456  */
457 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
458 			    void **ret_data)
459 {
460 	struct dlm_ctxt *dlm = data;
461 	struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
462 	struct dlm_lock_resource *res = NULL;
463 	struct dlm_lock *newlock = NULL;
464 	struct dlm_lockstatus *lksb = NULL;
465 	enum dlm_status status = DLM_NORMAL;
466 	char *name;
467 	unsigned int namelen;
468 
469 	BUG_ON(!dlm);
470 
471 	if (!dlm_grab(dlm))
472 		return DLM_REJECTED;
473 
474 	name = create->name;
475 	namelen = create->namelen;
476 	status = DLM_REJECTED;
477 	if (!dlm_domain_fully_joined(dlm)) {
478 		mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
479 		     "sending a create_lock message for lock %.*s!\n",
480 		     dlm->name, create->node_idx, namelen, name);
481 		dlm_error(status);
482 		goto leave;
483 	}
484 
485 	status = DLM_IVBUFLEN;
486 	if (namelen > DLM_LOCKID_NAME_MAX) {
487 		dlm_error(status);
488 		goto leave;
489 	}
490 
491 	status = DLM_SYSERR;
492 	newlock = dlm_new_lock(create->requested_type,
493 			       create->node_idx,
494 			       be64_to_cpu(create->cookie), NULL);
495 	if (!newlock) {
496 		dlm_error(status);
497 		goto leave;
498 	}
499 
500 	lksb = newlock->lksb;
501 
502 	if (be32_to_cpu(create->flags) & LKM_GET_LVB) {
503 		lksb->flags |= DLM_LKSB_GET_LVB;
504 		mlog(0, "set DLM_LKSB_GET_LVB flag\n");
505 	}
506 
507 	status = DLM_IVLOCKID;
508 	res = dlm_lookup_lockres(dlm, name, namelen);
509 	if (!res) {
510 		dlm_error(status);
511 		goto leave;
512 	}
513 
514 	spin_lock(&res->spinlock);
515 	status = __dlm_lockres_state_to_status(res);
516 	spin_unlock(&res->spinlock);
517 
518 	if (status != DLM_NORMAL) {
519 		mlog(0, "lockres recovering/migrating/in-progress\n");
520 		goto leave;
521 	}
522 
523 	dlm_lock_attach_lockres(newlock, res);
524 
525 	status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags));
526 leave:
527 	if (status != DLM_NORMAL)
528 		if (newlock)
529 			dlm_lock_put(newlock);
530 
531 	if (res)
532 		dlm_lockres_put(res);
533 
534 	dlm_put(dlm);
535 
536 	return status;
537 }
538 
539 
540 /* fetch next node-local (u8 nodenum + u56 cookie) into u64 */
541 static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
542 {
543 	u64 tmpnode = node_num;
544 
545 	/* shift single byte of node num into top 8 bits */
546 	tmpnode <<= 56;
547 
548 	spin_lock(&dlm_cookie_lock);
549 	*cookie = (dlm_next_cookie | tmpnode);
550 	if (++dlm_next_cookie & 0xff00000000000000ull) {
551 		mlog(0, "This node's cookie will now wrap!\n");
552 		dlm_next_cookie = 1;
553 	}
554 	spin_unlock(&dlm_cookie_lock);
555 }
556 
557 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
558 			struct dlm_lockstatus *lksb, int flags,
559 			const char *name, int namelen, dlm_astlockfunc_t *ast,
560 			void *data, dlm_bastlockfunc_t *bast)
561 {
562 	enum dlm_status status;
563 	struct dlm_lock_resource *res = NULL;
564 	struct dlm_lock *lock = NULL;
565 	int convert = 0, recovery = 0;
566 
567 	/* yes this function is a mess.
568 	 * TODO: clean this up.  lots of common code in the
569 	 *       lock and convert paths, especially in the retry blocks */
570 	if (!lksb) {
571 		dlm_error(DLM_BADARGS);
572 		return DLM_BADARGS;
573 	}
574 
575 	status = DLM_BADPARAM;
576 	if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) {
577 		dlm_error(status);
578 		goto error;
579 	}
580 
581 	if (flags & ~LKM_VALID_FLAGS) {
582 		dlm_error(status);
583 		goto error;
584 	}
585 
586 	convert = (flags & LKM_CONVERT);
587 	recovery = (flags & LKM_RECOVERY);
588 
589 	if (recovery &&
590 	    (!dlm_is_recovery_lock(name, namelen) || convert) ) {
591 		dlm_error(status);
592 		goto error;
593 	}
594 	if (convert && (flags & LKM_LOCAL)) {
595 		mlog(ML_ERROR, "strange LOCAL convert request!\n");
596 		goto error;
597 	}
598 
599 	if (convert) {
600 		/* CONVERT request */
601 
602 		/* if converting, must pass in a valid dlm_lock */
603 		lock = lksb->lockid;
604 		if (!lock) {
605 			mlog(ML_ERROR, "NULL lock pointer in convert "
606 			     "request\n");
607 			goto error;
608 		}
609 
610 		res = lock->lockres;
611 		if (!res) {
612 			mlog(ML_ERROR, "NULL lockres pointer in convert "
613 			     "request\n");
614 			goto error;
615 		}
616 		dlm_lockres_get(res);
617 
618 		/* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are
619 	 	 * static after the original lock call.  convert requests will
620 		 * ensure that everything is the same, or return DLM_BADARGS.
621 	 	 * this means that DLM_DENIED_NOASTS will never be returned.
622 	 	 */
623 		if (lock->lksb != lksb || lock->ast != ast ||
624 		    lock->bast != bast || lock->astdata != data) {
625 			status = DLM_BADARGS;
626 			mlog(ML_ERROR, "new args:  lksb=%p, ast=%p, bast=%p, "
627 			     "astdata=%p\n", lksb, ast, bast, data);
628 			mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, "
629 			     "astdata=%p\n", lock->lksb, lock->ast,
630 			     lock->bast, lock->astdata);
631 			goto error;
632 		}
633 retry_convert:
634 		dlm_wait_for_recovery(dlm);
635 
636 		if (res->owner == dlm->node_num)
637 			status = dlmconvert_master(dlm, res, lock, flags, mode);
638 		else
639 			status = dlmconvert_remote(dlm, res, lock, flags, mode);
640 		if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
641 		    status == DLM_FORWARD) {
642 			/* for now, see how this works without sleeping
643 			 * and just retry right away.  I suspect the reco
644 			 * or migration will complete fast enough that
645 			 * no waiting will be necessary */
646 			mlog(0, "retrying convert with migration/recovery/"
647 			     "in-progress\n");
648 			msleep(100);
649 			goto retry_convert;
650 		}
651 	} else {
652 		u64 tmpcookie;
653 
654 		/* LOCK request */
655 		status = DLM_BADARGS;
656 		if (!name) {
657 			dlm_error(status);
658 			goto error;
659 		}
660 
661 		status = DLM_IVBUFLEN;
662 		if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
663 			dlm_error(status);
664 			goto error;
665 		}
666 
667 		dlm_get_next_cookie(dlm->node_num, &tmpcookie);
668 		lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb);
669 		if (!lock) {
670 			dlm_error(status);
671 			goto error;
672 		}
673 
674 		if (!recovery)
675 			dlm_wait_for_recovery(dlm);
676 
677 		/* find or create the lock resource */
678 		res = dlm_get_lock_resource(dlm, name, namelen, flags);
679 		if (!res) {
680 			status = DLM_IVLOCKID;
681 			dlm_error(status);
682 			goto error;
683 		}
684 
685 		mlog(0, "type=%d, flags = 0x%x\n", mode, flags);
686 		mlog(0, "creating lock: lock=%p res=%p\n", lock, res);
687 
688 		dlm_lock_attach_lockres(lock, res);
689 		lock->ast = ast;
690 		lock->bast = bast;
691 		lock->astdata = data;
692 
693 retry_lock:
694 		if (flags & LKM_VALBLK) {
695 			mlog(0, "LKM_VALBLK passed by caller\n");
696 
697 			/* LVB requests for non PR, PW or EX locks are
698 			 * ignored. */
699 			if (mode < LKM_PRMODE)
700 				flags &= ~LKM_VALBLK;
701 			else {
702 				flags |= LKM_GET_LVB;
703 				lock->lksb->flags |= DLM_LKSB_GET_LVB;
704 			}
705 		}
706 
707 		if (res->owner == dlm->node_num)
708 			status = dlmlock_master(dlm, res, lock, flags);
709 		else
710 			status = dlmlock_remote(dlm, res, lock, flags);
711 
712 		if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
713 		    status == DLM_FORWARD) {
714 			msleep(100);
715 			if (recovery) {
716 				if (status != DLM_RECOVERING)
717 					goto retry_lock;
718 				/* wait to see the node go down, then
719 				 * drop down and allow the lockres to
720 				 * get cleaned up.  need to remaster. */
721 				dlm_wait_for_node_death(dlm, res->owner,
722 						DLM_NODE_DEATH_WAIT_MAX);
723 			} else {
724 				dlm_wait_for_recovery(dlm);
725 				goto retry_lock;
726 			}
727 		}
728 
729 		/* Inflight taken in dlm_get_lock_resource() is dropped here */
730 		spin_lock(&res->spinlock);
731 		dlm_lockres_drop_inflight_ref(dlm, res);
732 		spin_unlock(&res->spinlock);
733 
734 		dlm_lockres_calc_usage(dlm, res);
735 		dlm_kick_thread(dlm, res);
736 
737 		if (status != DLM_NORMAL) {
738 			lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
739 			if (status != DLM_NOTQUEUED)
740 				dlm_error(status);
741 			goto error;
742 		}
743 	}
744 
745 error:
746 	if (status != DLM_NORMAL) {
747 		if (lock && !convert)
748 			dlm_lock_put(lock);
749 		// this is kind of unnecessary
750 		lksb->status = status;
751 	}
752 
753 	/* put lockres ref from the convert path
754 	 * or from dlm_get_lock_resource */
755 	if (res)
756 		dlm_lockres_put(res);
757 
758 	return status;
759 }
760 EXPORT_SYMBOL_GPL(dlmlock);
761