xref: /openbmc/linux/fs/ocfs2/dlm/dlmlock.c (revision 4800cd83)
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 = NULL;
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 	if (dlm_lock_cache)
81 		kmem_cache_destroy(dlm_lock_cache);
82 }
83 
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 		if (!dlm_lock_compatible(tmplock->ml.convert_type,
110 					 lock->ml.type))
111 			return 0;
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 
131 	mlog_entry("type=%d\n", lock->ml.type);
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;
164 		} else {
165 			mlog(0, "%s: returning DLM_NORMAL to "
166 			     "node %u for reco lock\n", dlm->name,
167 			     lock->ml.node);
168 		}
169 	} else {
170 		/* for NOQUEUE request, unless we get the
171 		 * lock right away, return DLM_NOTQUEUED */
172 		if (flags & LKM_NOQUEUE) {
173 			status = DLM_NOTQUEUED;
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 {
181 			dlm_lock_get(lock);
182 			list_add_tail(&lock->list, &res->blocked);
183 			kick_thread = 1;
184 		}
185 	}
186 	/* reduce the inflight count, this may result in the lockres
187 	 * being purged below during calc_usage */
188 	if (lock->ml.node == dlm->node_num)
189 		dlm_lockres_drop_inflight_ref(dlm, res);
190 
191 	spin_unlock(&res->spinlock);
192 	wake_up(&res->wq);
193 
194 	/* either queue the ast or release it */
195 	if (call_ast)
196 		dlm_queue_ast(dlm, lock);
197 	else
198 		dlm_lockres_release_ast(dlm, res);
199 
200 	dlm_lockres_calc_usage(dlm, res);
201 	if (kick_thread)
202 		dlm_kick_thread(dlm, res);
203 
204 	return status;
205 }
206 
207 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
208 			     struct dlm_lock *lock)
209 {
210 	/* remove from local queue if it failed */
211 	list_del_init(&lock->list);
212 	lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
213 }
214 
215 
216 /*
217  * locking:
218  *   caller needs:  none
219  *   taken:         takes and drops res->spinlock
220  *   held on exit:  none
221  * returns: DLM_DENIED, DLM_RECOVERING, or net status
222  */
223 static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
224 				      struct dlm_lock_resource *res,
225 				      struct dlm_lock *lock, int flags)
226 {
227 	enum dlm_status status = DLM_DENIED;
228 	int lockres_changed = 1;
229 
230 	mlog_entry("type=%d\n", lock->ml.type);
231 	mlog(0, "lockres %.*s, flags = 0x%x\n", res->lockname.len,
232 	     res->lockname.name, flags);
233 
234 	spin_lock(&res->spinlock);
235 
236 	/* will exit this call with spinlock held */
237 	__dlm_wait_on_lockres(res);
238 	res->state |= DLM_LOCK_RES_IN_PROGRESS;
239 
240 	/* add lock to local (secondary) queue */
241 	dlm_lock_get(lock);
242 	list_add_tail(&lock->list, &res->blocked);
243 	lock->lock_pending = 1;
244 	spin_unlock(&res->spinlock);
245 
246 	/* spec seems to say that you will get DLM_NORMAL when the lock
247 	 * has been queued, meaning we need to wait for a reply here. */
248 	status = dlm_send_remote_lock_request(dlm, res, lock, flags);
249 
250 	spin_lock(&res->spinlock);
251 	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
252 	lock->lock_pending = 0;
253 	if (status != DLM_NORMAL) {
254 		if (status == DLM_RECOVERING &&
255 		    dlm_is_recovery_lock(res->lockname.name,
256 					 res->lockname.len)) {
257 			/* recovery lock was mastered by dead node.
258 			 * we need to have calc_usage shoot down this
259 			 * lockres and completely remaster it. */
260 			mlog(0, "%s: recovery lock was owned by "
261 			     "dead node %u, remaster it now.\n",
262 			     dlm->name, res->owner);
263 		} else if (status != DLM_NOTQUEUED) {
264 			/*
265 			 * DO NOT call calc_usage, as this would unhash
266 			 * the remote lockres before we ever get to use
267 			 * it.  treat as if we never made any change to
268 			 * the lockres.
269 			 */
270 			lockres_changed = 0;
271 			dlm_error(status);
272 		}
273 		dlm_revert_pending_lock(res, lock);
274 		dlm_lock_put(lock);
275 	} else if (dlm_is_recovery_lock(res->lockname.name,
276 					res->lockname.len)) {
277 		/* special case for the $RECOVERY lock.
278 		 * there will never be an AST delivered to put
279 		 * this lock on the proper secondary queue
280 		 * (granted), so do it manually. */
281 		mlog(0, "%s: $RECOVERY lock for this node (%u) is "
282 		     "mastered by %u; got lock, manually granting (no ast)\n",
283 		     dlm->name, dlm->node_num, res->owner);
284 		list_move_tail(&lock->list, &res->granted);
285 	}
286 	spin_unlock(&res->spinlock);
287 
288 	if (lockres_changed)
289 		dlm_lockres_calc_usage(dlm, res);
290 
291 	wake_up(&res->wq);
292 	return status;
293 }
294 
295 
296 /* for remote lock creation.
297  * locking:
298  *   caller needs:  none, but need res->state & DLM_LOCK_RES_IN_PROGRESS
299  *   taken:         none
300  *   held on exit:  none
301  * returns: DLM_NOLOCKMGR, or net status
302  */
303 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
304 					       struct dlm_lock_resource *res,
305 					       struct dlm_lock *lock, int flags)
306 {
307 	struct dlm_create_lock create;
308 	int tmpret, status = 0;
309 	enum dlm_status ret;
310 
311 	mlog_entry_void();
312 
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) {
324 		// successfully sent and received
325 		ret = status;  // this is already a dlm_status
326 		if (ret == DLM_REJECTED) {
327 			mlog(ML_ERROR, "%s:%.*s: BUG.  this is a stale lockres "
328 			     "no longer owned by %u.  that node is coming back "
329 			     "up currently.\n", dlm->name, create.namelen,
330 			     create.name, res->owner);
331 			dlm_print_one_lock_resource(res);
332 			BUG();
333 		}
334 	} else {
335 		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
336 		     "node %u\n", tmpret, DLM_CREATE_LOCK_MSG, dlm->key,
337 		     res->owner);
338 		if (dlm_is_host_down(tmpret)) {
339 			ret = DLM_RECOVERING;
340 			mlog(0, "node %u died so returning DLM_RECOVERING "
341 			     "from lock message!\n", res->owner);
342 		} else {
343 			ret = dlm_err_to_dlm_status(tmpret);
344 		}
345 	}
346 
347 	return ret;
348 }
349 
350 void dlm_lock_get(struct dlm_lock *lock)
351 {
352 	kref_get(&lock->lock_refs);
353 }
354 
355 void dlm_lock_put(struct dlm_lock *lock)
356 {
357 	kref_put(&lock->lock_refs, dlm_lock_release);
358 }
359 
360 static void dlm_lock_release(struct kref *kref)
361 {
362 	struct dlm_lock *lock;
363 
364 	lock = container_of(kref, struct dlm_lock, lock_refs);
365 
366 	BUG_ON(!list_empty(&lock->list));
367 	BUG_ON(!list_empty(&lock->ast_list));
368 	BUG_ON(!list_empty(&lock->bast_list));
369 	BUG_ON(lock->ast_pending);
370 	BUG_ON(lock->bast_pending);
371 
372 	dlm_lock_detach_lockres(lock);
373 
374 	if (lock->lksb_kernel_allocated) {
375 		mlog(0, "freeing kernel-allocated lksb\n");
376 		kfree(lock->lksb);
377 	}
378 	kmem_cache_free(dlm_lock_cache, lock);
379 }
380 
381 /* associate a lock with it's lockres, getting a ref on the lockres */
382 void dlm_lock_attach_lockres(struct dlm_lock *lock,
383 			     struct dlm_lock_resource *res)
384 {
385 	dlm_lockres_get(res);
386 	lock->lockres = res;
387 }
388 
389 /* drop ref on lockres, if there is still one associated with lock */
390 static void dlm_lock_detach_lockres(struct dlm_lock *lock)
391 {
392 	struct dlm_lock_resource *res;
393 
394 	res = lock->lockres;
395 	if (res) {
396 		lock->lockres = NULL;
397 		mlog(0, "removing lock's lockres reference\n");
398 		dlm_lockres_put(res);
399 	}
400 }
401 
402 static void dlm_init_lock(struct dlm_lock *newlock, int type,
403 			  u8 node, u64 cookie)
404 {
405 	INIT_LIST_HEAD(&newlock->list);
406 	INIT_LIST_HEAD(&newlock->ast_list);
407 	INIT_LIST_HEAD(&newlock->bast_list);
408 	spin_lock_init(&newlock->spinlock);
409 	newlock->ml.type = type;
410 	newlock->ml.convert_type = LKM_IVMODE;
411 	newlock->ml.highest_blocked = LKM_IVMODE;
412 	newlock->ml.node = node;
413 	newlock->ml.pad1 = 0;
414 	newlock->ml.list = 0;
415 	newlock->ml.flags = 0;
416 	newlock->ast = NULL;
417 	newlock->bast = NULL;
418 	newlock->astdata = NULL;
419 	newlock->ml.cookie = cpu_to_be64(cookie);
420 	newlock->ast_pending = 0;
421 	newlock->bast_pending = 0;
422 	newlock->convert_pending = 0;
423 	newlock->lock_pending = 0;
424 	newlock->unlock_pending = 0;
425 	newlock->cancel_pending = 0;
426 	newlock->lksb_kernel_allocated = 0;
427 
428 	kref_init(&newlock->lock_refs);
429 }
430 
431 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
432 			       struct dlm_lockstatus *lksb)
433 {
434 	struct dlm_lock *lock;
435 	int kernel_allocated = 0;
436 
437 	lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS);
438 	if (!lock)
439 		return NULL;
440 
441 	if (!lksb) {
442 		/* zero memory only if kernel-allocated */
443 		lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
444 		if (!lksb) {
445 			kfree(lock);
446 			return NULL;
447 		}
448 		kernel_allocated = 1;
449 	}
450 
451 	dlm_init_lock(lock, type, node, cookie);
452 	if (kernel_allocated)
453 		lock->lksb_kernel_allocated = 1;
454 	lock->lksb = lksb;
455 	lksb->lockid = lock;
456 	return lock;
457 }
458 
459 /* handler for lock creation net message
460  * locking:
461  *   caller needs:  none
462  *   taken:         takes and drops res->spinlock
463  *   held on exit:  none
464  * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED
465  */
466 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
467 			    void **ret_data)
468 {
469 	struct dlm_ctxt *dlm = data;
470 	struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
471 	struct dlm_lock_resource *res = NULL;
472 	struct dlm_lock *newlock = NULL;
473 	struct dlm_lockstatus *lksb = NULL;
474 	enum dlm_status status = DLM_NORMAL;
475 	char *name;
476 	unsigned int namelen;
477 
478 	BUG_ON(!dlm);
479 
480 	mlog_entry_void();
481 
482 	if (!dlm_grab(dlm))
483 		return DLM_REJECTED;
484 
485 	name = create->name;
486 	namelen = create->namelen;
487 	status = DLM_REJECTED;
488 	if (!dlm_domain_fully_joined(dlm)) {
489 		mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
490 		     "sending a create_lock message for lock %.*s!\n",
491 		     dlm->name, create->node_idx, namelen, name);
492 		dlm_error(status);
493 		goto leave;
494 	}
495 
496 	status = DLM_IVBUFLEN;
497 	if (namelen > DLM_LOCKID_NAME_MAX) {
498 		dlm_error(status);
499 		goto leave;
500 	}
501 
502 	status = DLM_SYSERR;
503 	newlock = dlm_new_lock(create->requested_type,
504 			       create->node_idx,
505 			       be64_to_cpu(create->cookie), NULL);
506 	if (!newlock) {
507 		dlm_error(status);
508 		goto leave;
509 	}
510 
511 	lksb = newlock->lksb;
512 
513 	if (be32_to_cpu(create->flags) & LKM_GET_LVB) {
514 		lksb->flags |= DLM_LKSB_GET_LVB;
515 		mlog(0, "set DLM_LKSB_GET_LVB flag\n");
516 	}
517 
518 	status = DLM_IVLOCKID;
519 	res = dlm_lookup_lockres(dlm, name, namelen);
520 	if (!res) {
521 		dlm_error(status);
522 		goto leave;
523 	}
524 
525 	spin_lock(&res->spinlock);
526 	status = __dlm_lockres_state_to_status(res);
527 	spin_unlock(&res->spinlock);
528 
529 	if (status != DLM_NORMAL) {
530 		mlog(0, "lockres recovering/migrating/in-progress\n");
531 		goto leave;
532 	}
533 
534 	dlm_lock_attach_lockres(newlock, res);
535 
536 	status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags));
537 leave:
538 	if (status != DLM_NORMAL)
539 		if (newlock)
540 			dlm_lock_put(newlock);
541 
542 	if (res)
543 		dlm_lockres_put(res);
544 
545 	dlm_put(dlm);
546 
547 	return status;
548 }
549 
550 
551 /* fetch next node-local (u8 nodenum + u56 cookie) into u64 */
552 static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
553 {
554 	u64 tmpnode = node_num;
555 
556 	/* shift single byte of node num into top 8 bits */
557 	tmpnode <<= 56;
558 
559 	spin_lock(&dlm_cookie_lock);
560 	*cookie = (dlm_next_cookie | tmpnode);
561 	if (++dlm_next_cookie & 0xff00000000000000ull) {
562 		mlog(0, "This node's cookie will now wrap!\n");
563 		dlm_next_cookie = 1;
564 	}
565 	spin_unlock(&dlm_cookie_lock);
566 }
567 
568 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
569 			struct dlm_lockstatus *lksb, int flags,
570 			const char *name, int namelen, dlm_astlockfunc_t *ast,
571 			void *data, dlm_bastlockfunc_t *bast)
572 {
573 	enum dlm_status status;
574 	struct dlm_lock_resource *res = NULL;
575 	struct dlm_lock *lock = NULL;
576 	int convert = 0, recovery = 0;
577 
578 	/* yes this function is a mess.
579 	 * TODO: clean this up.  lots of common code in the
580 	 *       lock and convert paths, especially in the retry blocks */
581 	if (!lksb) {
582 		dlm_error(DLM_BADARGS);
583 		return DLM_BADARGS;
584 	}
585 
586 	status = DLM_BADPARAM;
587 	if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) {
588 		dlm_error(status);
589 		goto error;
590 	}
591 
592 	if (flags & ~LKM_VALID_FLAGS) {
593 		dlm_error(status);
594 		goto error;
595 	}
596 
597 	convert = (flags & LKM_CONVERT);
598 	recovery = (flags & LKM_RECOVERY);
599 
600 	if (recovery &&
601 	    (!dlm_is_recovery_lock(name, namelen) || convert) ) {
602 		dlm_error(status);
603 		goto error;
604 	}
605 	if (convert && (flags & LKM_LOCAL)) {
606 		mlog(ML_ERROR, "strange LOCAL convert request!\n");
607 		goto error;
608 	}
609 
610 	if (convert) {
611 		/* CONVERT request */
612 
613 		/* if converting, must pass in a valid dlm_lock */
614 		lock = lksb->lockid;
615 		if (!lock) {
616 			mlog(ML_ERROR, "NULL lock pointer in convert "
617 			     "request\n");
618 			goto error;
619 		}
620 
621 		res = lock->lockres;
622 		if (!res) {
623 			mlog(ML_ERROR, "NULL lockres pointer in convert "
624 			     "request\n");
625 			goto error;
626 		}
627 		dlm_lockres_get(res);
628 
629 		/* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are
630 	 	 * static after the original lock call.  convert requests will
631 		 * ensure that everything is the same, or return DLM_BADARGS.
632 	 	 * this means that DLM_DENIED_NOASTS will never be returned.
633 	 	 */
634 		if (lock->lksb != lksb || lock->ast != ast ||
635 		    lock->bast != bast || lock->astdata != data) {
636 			status = DLM_BADARGS;
637 			mlog(ML_ERROR, "new args:  lksb=%p, ast=%p, bast=%p, "
638 			     "astdata=%p\n", lksb, ast, bast, data);
639 			mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, "
640 			     "astdata=%p\n", lock->lksb, lock->ast,
641 			     lock->bast, lock->astdata);
642 			goto error;
643 		}
644 retry_convert:
645 		dlm_wait_for_recovery(dlm);
646 
647 		if (res->owner == dlm->node_num)
648 			status = dlmconvert_master(dlm, res, lock, flags, mode);
649 		else
650 			status = dlmconvert_remote(dlm, res, lock, flags, mode);
651 		if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
652 		    status == DLM_FORWARD) {
653 			/* for now, see how this works without sleeping
654 			 * and just retry right away.  I suspect the reco
655 			 * or migration will complete fast enough that
656 			 * no waiting will be necessary */
657 			mlog(0, "retrying convert with migration/recovery/"
658 			     "in-progress\n");
659 			msleep(100);
660 			goto retry_convert;
661 		}
662 	} else {
663 		u64 tmpcookie;
664 
665 		/* LOCK request */
666 		status = DLM_BADARGS;
667 		if (!name) {
668 			dlm_error(status);
669 			goto error;
670 		}
671 
672 		status = DLM_IVBUFLEN;
673 		if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
674 			dlm_error(status);
675 			goto error;
676 		}
677 
678 		dlm_get_next_cookie(dlm->node_num, &tmpcookie);
679 		lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb);
680 		if (!lock) {
681 			dlm_error(status);
682 			goto error;
683 		}
684 
685 		if (!recovery)
686 			dlm_wait_for_recovery(dlm);
687 
688 		/* find or create the lock resource */
689 		res = dlm_get_lock_resource(dlm, name, namelen, flags);
690 		if (!res) {
691 			status = DLM_IVLOCKID;
692 			dlm_error(status);
693 			goto error;
694 		}
695 
696 		mlog(0, "type=%d, flags = 0x%x\n", mode, flags);
697 		mlog(0, "creating lock: lock=%p res=%p\n", lock, res);
698 
699 		dlm_lock_attach_lockres(lock, res);
700 		lock->ast = ast;
701 		lock->bast = bast;
702 		lock->astdata = data;
703 
704 retry_lock:
705 		if (flags & LKM_VALBLK) {
706 			mlog(0, "LKM_VALBLK passed by caller\n");
707 
708 			/* LVB requests for non PR, PW or EX locks are
709 			 * ignored. */
710 			if (mode < LKM_PRMODE)
711 				flags &= ~LKM_VALBLK;
712 			else {
713 				flags |= LKM_GET_LVB;
714 				lock->lksb->flags |= DLM_LKSB_GET_LVB;
715 			}
716 		}
717 
718 		if (res->owner == dlm->node_num)
719 			status = dlmlock_master(dlm, res, lock, flags);
720 		else
721 			status = dlmlock_remote(dlm, res, lock, flags);
722 
723 		if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
724 		    status == DLM_FORWARD) {
725 			mlog(0, "retrying lock with migration/"
726 			     "recovery/in progress\n");
727 			msleep(100);
728 			/* no waiting for dlm_reco_thread */
729 			if (recovery) {
730 				if (status != DLM_RECOVERING)
731 					goto retry_lock;
732 
733 				mlog(0, "%s: got RECOVERING "
734 				     "for $RECOVERY lock, master "
735 				     "was %u\n", dlm->name,
736 				     res->owner);
737 				/* wait to see the node go down, then
738 				 * drop down and allow the lockres to
739 				 * get cleaned up.  need to remaster. */
740 				dlm_wait_for_node_death(dlm, res->owner,
741 						DLM_NODE_DEATH_WAIT_MAX);
742 			} else {
743 				dlm_wait_for_recovery(dlm);
744 				goto retry_lock;
745 			}
746 		}
747 
748 		if (status != DLM_NORMAL) {
749 			lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
750 			if (status != DLM_NOTQUEUED)
751 				dlm_error(status);
752 			goto error;
753 		}
754 	}
755 
756 error:
757 	if (status != DLM_NORMAL) {
758 		if (lock && !convert)
759 			dlm_lock_put(lock);
760 		// this is kind of unnecessary
761 		lksb->status = status;
762 	}
763 
764 	/* put lockres ref from the convert path
765 	 * or from dlm_get_lock_resource */
766 	if (res)
767 		dlm_lockres_put(res);
768 
769 	return status;
770 }
771 EXPORT_SYMBOL_GPL(dlmlock);
772