Whamcloud - gitweb
28bc7041e6abc3552b6b1fecf75bd79554c94c5c
[fs/lustre-release.git] / lustre / ldlm / ldlm_lock.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * GPL HEADER START
5  *
6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 only,
10  * as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope that it will be useful, but
13  * WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * General Public License version 2 for more details (a copy is included
16  * in the LICENSE file that accompanied this code).
17  *
18  * You should have received a copy of the GNU General Public License
19  * version 2 along with this program; If not, see
20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
21  *
22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23  * CA 95054 USA or visit www.sun.com if you need additional information or
24  * have any questions.
25  *
26  * GPL HEADER END
27  */
28 /*
29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
30  * Use is subject to license terms.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/ldlm/ldlm_lock.c
37  *
38  * Author: Peter Braam <braam@clusterfs.com>
39  * Author: Phil Schwan <phil@clusterfs.com>
40  */
41
42 #define DEBUG_SUBSYSTEM S_LDLM
43
44 #ifdef __KERNEL__
45 # include <libcfs/libcfs.h>
46 # ifndef HAVE_VFS_INTENT_PATCHES
47 # include <linux/lustre_intent.h>
48 # endif
49 #else
50 # include <liblustre.h>
51 #endif
52
53 #include <obd_class.h>
54 #include "ldlm_internal.h"
55
56 /* lock types */
57 char *ldlm_lockname[] = {
58         [0] "--",
59         [LCK_EX] "EX",
60         [LCK_PW] "PW",
61         [LCK_PR] "PR",
62         [LCK_CW] "CW",
63         [LCK_CR] "CR",
64         [LCK_NL] "NL",
65         [LCK_GROUP] "GROUP",
66         [LCK_COS] "COS"
67 };
68
69 char *ldlm_typename[] = {
70         [LDLM_PLAIN] "PLN",
71         [LDLM_EXTENT] "EXT",
72         [LDLM_FLOCK] "FLK",
73         [LDLM_IBITS] "IBT",
74 };
75
76 char *ldlm_it2str(int it)
77 {
78         switch (it) {
79         case IT_OPEN:
80                 return "open";
81         case IT_CREAT:
82                 return "creat";
83         case (IT_OPEN | IT_CREAT):
84                 return "open|creat";
85         case IT_READDIR:
86                 return "readdir";
87         case IT_GETATTR:
88                 return "getattr";
89         case IT_LOOKUP:
90                 return "lookup";
91         case IT_UNLINK:
92                 return "unlink";
93         case IT_GETXATTR:
94                 return "getxattr";
95         default:
96                 CERROR("Unknown intent %d\n", it);
97                 return "UNKNOWN";
98         }
99 }
100
101 extern cfs_mem_cache_t *ldlm_lock_slab;
102
103 static ldlm_processing_policy ldlm_processing_policy_table[] = {
104         [LDLM_PLAIN] ldlm_process_plain_lock,
105         [LDLM_EXTENT] ldlm_process_extent_lock,
106 #ifdef __KERNEL__
107         [LDLM_FLOCK] ldlm_process_flock_lock,
108 #endif
109         [LDLM_IBITS] ldlm_process_inodebits_lock,
110 };
111
112 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
113 {
114         return ldlm_processing_policy_table[res->lr_type];
115 }
116
117 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
118 {
119         ns->ns_policy = arg;
120 }
121
122 /*
123  * REFCOUNTED LOCK OBJECTS
124  */
125
126
127 /*
128  * Lock refcounts, during creation:
129  *   - one special one for allocation, dec'd only once in destroy
130  *   - one for being a lock that's in-use
131  *   - one for the addref associated with a new lock
132  */
133 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
134 {
135         atomic_inc(&lock->l_refc);
136         return lock;
137 }
138
139 static void ldlm_lock_free(struct ldlm_lock *lock, size_t size)
140 {
141         LASSERT(size == sizeof(*lock));
142         OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
143 }
144
145 void ldlm_lock_put(struct ldlm_lock *lock)
146 {
147         ENTRY;
148
149         LASSERT(lock->l_resource != LP_POISON);
150         LASSERT(atomic_read(&lock->l_refc) > 0);
151         if (atomic_dec_and_test(&lock->l_refc)) {
152                 struct ldlm_resource *res;
153
154                 LDLM_DEBUG(lock,
155                            "final lock_put on destroyed lock, freeing it.");
156
157                 res = lock->l_resource;
158                 LASSERT(lock->l_destroyed);
159                 LASSERT(list_empty(&lock->l_res_link));
160                 LASSERT(list_empty(&lock->l_pending_chain));
161
162                 atomic_dec(&res->lr_namespace->ns_locks);
163                 lu_ref_del(&res->lr_reference, "lock", lock);
164                 ldlm_resource_putref(res);
165                 lock->l_resource = NULL;
166                 if (lock->l_export) {
167                         class_export_lock_put(lock->l_export, lock);
168                         lock->l_export = NULL;
169                 }
170
171                 if (lock->l_lvb_data != NULL)
172                         OBD_FREE(lock->l_lvb_data, lock->l_lvb_len);
173
174                 ldlm_interval_free(ldlm_interval_detach(lock));
175                 lu_ref_fini(&lock->l_reference);
176                 OBD_FREE_RCU_CB(lock, sizeof(*lock), &lock->l_handle,
177                                 ldlm_lock_free);
178         }
179
180         EXIT;
181 }
182
183 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
184 {
185         int rc = 0;
186         if (!list_empty(&lock->l_lru)) {
187                 struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
188                 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
189                 list_del_init(&lock->l_lru);
190                 LASSERT(ns->ns_nr_unused > 0);
191                 ns->ns_nr_unused--;
192                 rc = 1;
193         }
194         return rc;
195 }
196
197 int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
198 {
199         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
200         int rc;
201         ENTRY;
202         spin_lock(&ns->ns_unused_lock);
203         rc = ldlm_lock_remove_from_lru_nolock(lock);
204         spin_unlock(&ns->ns_unused_lock);
205         EXIT;
206         return rc;
207 }
208
209 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
210 {
211         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
212         lock->l_last_used = cfs_time_current();
213         LASSERT(list_empty(&lock->l_lru));
214         LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
215         list_add_tail(&lock->l_lru, &ns->ns_unused_list);
216         LASSERT(ns->ns_nr_unused >= 0);
217         ns->ns_nr_unused++;
218 }
219
220 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
221 {
222         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
223         ENTRY;
224         spin_lock(&ns->ns_unused_lock);
225         ldlm_lock_add_to_lru_nolock(lock);
226         spin_unlock(&ns->ns_unused_lock);
227         EXIT;
228 }
229
230 void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
231 {
232         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
233         ENTRY;
234         spin_lock(&ns->ns_unused_lock);
235         if (!list_empty(&lock->l_lru)) {
236                 ldlm_lock_remove_from_lru_nolock(lock);
237                 ldlm_lock_add_to_lru_nolock(lock);
238         }
239         spin_unlock(&ns->ns_unused_lock);
240         EXIT;
241 }
242
243 /* This used to have a 'strict' flag, which recovery would use to mark an
244  * in-use lock as needing-to-die.  Lest I am ever tempted to put it back, I
245  * shall explain why it's gone: with the new hash table scheme, once you call
246  * ldlm_lock_destroy, you can never drop your final references on this lock.
247  * Because it's not in the hash table anymore.  -phil */
248 int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
249 {
250         ENTRY;
251
252         if (lock->l_readers || lock->l_writers) {
253                 LDLM_ERROR(lock, "lock still has references");
254                 ldlm_lock_dump(D_ERROR, lock, 0);
255                 LBUG();
256         }
257
258         if (!list_empty(&lock->l_res_link)) {
259                 LDLM_ERROR(lock, "lock still on resource");
260                 ldlm_lock_dump(D_ERROR, lock, 0);
261                 LBUG();
262         }
263
264         if (lock->l_destroyed) {
265                 LASSERT(list_empty(&lock->l_lru));
266                 EXIT;
267                 return 0;
268         }
269         lock->l_destroyed = 1;
270
271         if (lock->l_export && lock->l_export->exp_lock_hash &&
272             !hlist_unhashed(&lock->l_exp_hash))
273                 cfs_hash_del(lock->l_export->exp_lock_hash,
274                              &lock->l_remote_handle, &lock->l_exp_hash);
275
276         ldlm_lock_remove_from_lru(lock);
277         class_handle_unhash(&lock->l_handle);
278
279 #if 0
280         /* Wake anyone waiting for this lock */
281         /* FIXME: I should probably add yet another flag, instead of using
282          * l_export to only call this on clients */
283         if (lock->l_export)
284                 class_export_put(lock->l_export);
285         lock->l_export = NULL;
286         if (lock->l_export && lock->l_completion_ast)
287                 lock->l_completion_ast(lock, 0);
288 #endif
289         EXIT;
290         return 1;
291 }
292
293 void ldlm_lock_destroy(struct ldlm_lock *lock)
294 {
295         int first;
296         ENTRY;
297         lock_res_and_lock(lock);
298         first = ldlm_lock_destroy_internal(lock);
299         unlock_res_and_lock(lock);
300
301         /* drop reference from hashtable only for first destroy */
302         if (first) {
303                 lu_ref_del(&lock->l_reference, "hash", lock);
304                 LDLM_LOCK_RELEASE(lock);
305         }
306         EXIT;
307 }
308
309 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
310 {
311         int first;
312         ENTRY;
313         first = ldlm_lock_destroy_internal(lock);
314         /* drop reference from hashtable only for first destroy */
315         if (first) {
316                 lu_ref_del(&lock->l_reference, "hash", lock);
317                 LDLM_LOCK_RELEASE(lock);
318         }
319         EXIT;
320 }
321
322 /* this is called by portals_handle2object with the handle lock taken */
323 static void lock_handle_addref(void *lock)
324 {
325         LDLM_LOCK_GET((struct ldlm_lock *)lock);
326 }
327
328 /*
329  * usage: pass in a resource on which you have done ldlm_resource_get
330  *        pass in a parent lock on which you have done a ldlm_lock_get
331  *        after return, ldlm_*_put the resource and parent
332  * returns: lock with refcount 2 - one for current caller and one for remote
333  */
334 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
335 {
336         struct ldlm_lock *lock;
337         ENTRY;
338
339         if (resource == NULL)
340                 LBUG();
341
342         OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, CFS_ALLOC_IO);
343         if (lock == NULL)
344                 RETURN(NULL);
345
346         spin_lock_init(&lock->l_lock);
347         lock->l_resource = ldlm_resource_getref(resource);
348         lu_ref_add(&resource->lr_reference, "lock", lock);
349
350         atomic_set(&lock->l_refc, 2);
351         CFS_INIT_LIST_HEAD(&lock->l_res_link);
352         CFS_INIT_LIST_HEAD(&lock->l_lru);
353         CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
354         CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
355         CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
356         CFS_INIT_LIST_HEAD(&lock->l_rk_ast);
357         cfs_waitq_init(&lock->l_waitq);
358         lock->l_blocking_lock = NULL;
359         CFS_INIT_LIST_HEAD(&lock->l_sl_mode);
360         CFS_INIT_LIST_HEAD(&lock->l_sl_policy);
361         CFS_INIT_HLIST_NODE(&lock->l_exp_hash);
362
363         atomic_inc(&resource->lr_namespace->ns_locks);
364         CFS_INIT_LIST_HEAD(&lock->l_handle.h_link);
365         class_handle_hash(&lock->l_handle, lock_handle_addref);
366
367         CFS_INIT_LIST_HEAD(&lock->l_extents_list);
368         spin_lock_init(&lock->l_extents_list_lock);
369         CFS_INIT_LIST_HEAD(&lock->l_cache_locks_list);
370         lu_ref_init(&lock->l_reference);
371         lu_ref_add(&lock->l_reference, "hash", lock);
372         lock->l_callback_timeout = 0;
373
374 #if LUSTRE_TRACKS_LOCK_EXP_REFS
375         CFS_INIT_LIST_HEAD(&lock->l_exp_refs_link);
376         lock->l_exp_refs_nr = 0;
377         lock->l_exp_refs_target = NULL;
378 #endif
379
380         RETURN(lock);
381 }
382
383 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
384                               const struct ldlm_res_id *new_resid)
385 {
386         struct ldlm_resource *oldres = lock->l_resource;
387         struct ldlm_resource *newres;
388         int type;
389         ENTRY;
390
391         LASSERT(ns_is_client(ns));
392
393         lock_res_and_lock(lock);
394         if (memcmp(new_resid, &lock->l_resource->lr_name,
395                    sizeof(lock->l_resource->lr_name)) == 0) {
396                 /* Nothing to do */
397                 unlock_res_and_lock(lock);
398                 RETURN(0);
399         }
400
401         LASSERT(new_resid->name[0] != 0);
402
403         /* This function assumes that the lock isn't on any lists */
404         LASSERT(list_empty(&lock->l_res_link));
405
406         type = oldres->lr_type;
407         unlock_res_and_lock(lock);
408
409         newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
410         lu_ref_add(&newres->lr_reference, "lock", lock);
411         if (newres == NULL)
412                 RETURN(-ENOMEM);
413         /*
414          * To flip the lock from the old to the new resource, lock, oldres and
415          * newres have to be locked. Resource spin-locks are nested within
416          * lock->l_lock, and are taken in the memory address order to avoid
417          * dead-locks.
418          */
419         spin_lock(&lock->l_lock);
420         oldres = lock->l_resource;
421         if (oldres < newres) {
422                 lock_res(oldres);
423                 lock_res_nested(newres, LRT_NEW);
424         } else {
425                 lock_res(newres);
426                 lock_res_nested(oldres, LRT_NEW);
427         }
428         LASSERT(memcmp(new_resid, &oldres->lr_name,
429                        sizeof oldres->lr_name) != 0);
430         lock->l_resource = newres;
431         unlock_res(oldres);
432         unlock_res_and_lock(lock);
433
434         /* ...and the flowers are still standing! */
435         lu_ref_del(&oldres->lr_reference, "lock", lock);
436         ldlm_resource_putref(oldres);
437
438         RETURN(0);
439 }
440
441 /*
442  *  HANDLES
443  */
444
445 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
446 {
447         lockh->cookie = lock->l_handle.h_cookie;
448 }
449
450 /* if flags: atomically get the lock and set the flags.
451  *           Return NULL if flag already set
452  */
453
454 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
455                                      int flags)
456 {
457         struct ldlm_namespace *ns;
458         struct ldlm_lock *lock, *retval = NULL;
459         ENTRY;
460
461         LASSERT(handle);
462
463         lock = class_handle2object(handle->cookie);
464         if (lock == NULL)
465                 RETURN(NULL);
466
467         LASSERT(lock->l_resource != NULL);
468         ns = lock->l_resource->lr_namespace;
469         LASSERT(ns != NULL);
470
471         lu_ref_add_atomic(&lock->l_reference, "handle", cfs_current());
472         lock_res_and_lock(lock);
473
474         /* It's unlikely but possible that someone marked the lock as
475          * destroyed after we did handle2object on it */
476         if (lock->l_destroyed) {
477                 unlock_res_and_lock(lock);
478                 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
479                 LDLM_LOCK_PUT(lock);
480                 GOTO(out, retval);
481         }
482
483         if (flags && (lock->l_flags & flags)) {
484                 unlock_res_and_lock(lock);
485                 LDLM_LOCK_PUT(lock);
486                 GOTO(out, retval);
487         }
488
489         if (flags)
490                 lock->l_flags |= flags;
491
492         unlock_res_and_lock(lock);
493         retval = lock;
494         EXIT;
495  out:
496         return retval;
497 }
498
499 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
500 {
501         struct obd_export *exp = lock->l_export?:lock->l_conn_export;
502         /* INODEBITS_INTEROP: If the other side does not support
503          * inodebits, reply with a plain lock descriptor.
504          */
505         if ((lock->l_resource->lr_type == LDLM_IBITS) &&
506             (exp && !(exp->exp_connect_flags & OBD_CONNECT_IBITS))) {
507                 /* Make sure all the right bits are set in this lock we
508                    are going to pass to client */
509                 LASSERTF(lock->l_policy_data.l_inodebits.bits ==
510                          (MDS_INODELOCK_LOOKUP|MDS_INODELOCK_UPDATE),
511                          "Inappropriate inode lock bits during "
512                          "conversion " LPU64 "\n",
513                          lock->l_policy_data.l_inodebits.bits);
514
515                 ldlm_res2desc(lock->l_resource, &desc->l_resource);
516                 desc->l_resource.lr_type = LDLM_PLAIN;
517
518                 /* Convert "new" lock mode to something old client can
519                    understand */
520                 if ((lock->l_req_mode == LCK_CR) ||
521                     (lock->l_req_mode == LCK_CW))
522                         desc->l_req_mode = LCK_PR;
523                 else
524                         desc->l_req_mode = lock->l_req_mode;
525                 if ((lock->l_granted_mode == LCK_CR) ||
526                     (lock->l_granted_mode == LCK_CW)) {
527                         desc->l_granted_mode = LCK_PR;
528                 } else {
529                         /* We never grant PW/EX locks to clients */
530                         LASSERT((lock->l_granted_mode != LCK_PW) &&
531                                 (lock->l_granted_mode != LCK_EX));
532                         desc->l_granted_mode = lock->l_granted_mode;
533                 }
534
535                 /* We do not copy policy here, because there is no
536                    policy for plain locks */
537         } else {
538                 ldlm_res2desc(lock->l_resource, &desc->l_resource);
539                 desc->l_req_mode = lock->l_req_mode;
540                 desc->l_granted_mode = lock->l_granted_mode;
541                 desc->l_policy_data = lock->l_policy_data;
542         }
543 }
544
545 void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
546                            struct list_head *work_list)
547 {
548         if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) {
549                 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
550                 lock->l_flags |= LDLM_FL_AST_SENT;
551                 /* If the enqueuing client said so, tell the AST recipient to
552                  * discard dirty data, rather than writing back. */
553                 if (new->l_flags & LDLM_AST_DISCARD_DATA)
554                         lock->l_flags |= LDLM_FL_DISCARD_DATA;
555                 LASSERT(list_empty(&lock->l_bl_ast));
556                 list_add(&lock->l_bl_ast, work_list);
557                 LDLM_LOCK_GET(lock);
558                 LASSERT(lock->l_blocking_lock == NULL);
559                 lock->l_blocking_lock = LDLM_LOCK_GET(new);
560         }
561 }
562
563 void ldlm_add_cp_work_item(struct ldlm_lock *lock, struct list_head *work_list)
564 {
565         if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) {
566                 lock->l_flags |= LDLM_FL_CP_REQD;
567                 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
568                 LASSERT(list_empty(&lock->l_cp_ast));
569                 list_add(&lock->l_cp_ast, work_list);
570                 LDLM_LOCK_GET(lock);
571         }
572 }
573
574 /* must be called with lr_lock held */
575 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
576                                 struct list_head *work_list)
577 {
578         ENTRY;
579         check_res_locked(lock->l_resource);
580         if (new)
581                 ldlm_add_bl_work_item(lock, new, work_list);
582         else
583                 ldlm_add_cp_work_item(lock, work_list);
584         EXIT;
585 }
586
587 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
588 {
589         struct ldlm_lock *lock;
590
591         lock = ldlm_handle2lock(lockh);
592         LASSERT(lock != NULL);
593         ldlm_lock_addref_internal(lock, mode);
594         LDLM_LOCK_PUT(lock);
595 }
596
597 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
598 {
599         ldlm_lock_remove_from_lru(lock);
600         if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
601                 lock->l_readers++;
602                 lu_ref_add_atomic(&lock->l_reference, "reader", lock);
603         }
604         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
605                 lock->l_writers++;
606                 lu_ref_add_atomic(&lock->l_reference, "writer", lock);
607         }
608         LDLM_LOCK_GET(lock);
609         lu_ref_add_atomic(&lock->l_reference, "user", lock);
610         LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
611 }
612
613 /**
614  * Attempts to addref a lock, and fails if lock is already LDLM_FL_CBPENDING
615  * or destroyed.
616  *
617  * \retval 0 success, lock was addref-ed
618  *
619  * \retval -EAGAIN lock is being canceled.
620  */
621 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode)
622 {
623         struct ldlm_lock *lock;
624         int               result;
625
626         result = -EAGAIN;
627         lock = ldlm_handle2lock(lockh);
628         if (lock != NULL) {
629                 lock_res_and_lock(lock);
630                 if (lock->l_readers != 0 || lock->l_writers != 0 ||
631                     !(lock->l_flags & LDLM_FL_CBPENDING)) {
632                         ldlm_lock_addref_internal_nolock(lock, mode);
633                         result = 0;
634                 }
635                 unlock_res_and_lock(lock);
636                 LDLM_LOCK_PUT(lock);
637         }
638         return result;
639 }
640
641 /* only called for local locks */
642 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
643 {
644         lock_res_and_lock(lock);
645         ldlm_lock_addref_internal_nolock(lock, mode);
646         unlock_res_and_lock(lock);
647 }
648
649 /* only called in ldlm_flock_destroy and for local locks.
650  *  * for LDLM_FLOCK type locks, l_blocking_ast is null, and
651  *   * ldlm_lock_remove_from_lru() does nothing, it is safe
652  *    * for ldlm_flock_destroy usage by dropping some code */
653 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
654 {
655         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
656         if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
657                 LASSERT(lock->l_readers > 0);
658                 lu_ref_del(&lock->l_reference, "reader", lock);
659                 lock->l_readers--;
660         }
661         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
662                 LASSERT(lock->l_writers > 0);
663                 lu_ref_del(&lock->l_reference, "writer", lock);
664                 lock->l_writers--;
665         }
666
667         lu_ref_del(&lock->l_reference, "user", lock);
668         LDLM_LOCK_RELEASE(lock);    /* matches the LDLM_LOCK_GET() in addref */
669 }
670
671 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
672 {
673         struct ldlm_namespace *ns;
674         ENTRY;
675
676         lock_res_and_lock(lock);
677
678         ns = lock->l_resource->lr_namespace;
679
680         ldlm_lock_decref_internal_nolock(lock, mode);
681
682         if (lock->l_flags & LDLM_FL_LOCAL &&
683             !lock->l_readers && !lock->l_writers) {
684                 /* If this is a local lock on a server namespace and this was
685                  * the last reference, cancel the lock. */
686                 CDEBUG(D_INFO, "forcing cancel of local lock\n");
687                 lock->l_flags |= LDLM_FL_CBPENDING;
688         }
689
690         if (!lock->l_readers && !lock->l_writers &&
691             (lock->l_flags & LDLM_FL_CBPENDING)) {
692                 /* If we received a blocked AST and this was the last reference,
693                  * run the callback. */
694                 if (ns_is_server(ns) && lock->l_export)
695                         CERROR("FL_CBPENDING set on non-local lock--just a "
696                                "warning\n");
697
698                 LDLM_DEBUG(lock, "final decref done on cbpending lock");
699
700                 LDLM_LOCK_GET(lock); /* dropped by bl thread */
701                 ldlm_lock_remove_from_lru(lock);
702                 unlock_res_and_lock(lock);
703
704                 if (lock->l_flags & LDLM_FL_FAIL_LOC)
705                         OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
706
707                 if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
708                     ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
709                         ldlm_handle_bl_callback(ns, NULL, lock);
710         } else if (ns_is_client(ns) &&
711                    !lock->l_readers && !lock->l_writers &&
712                    !(lock->l_flags & LDLM_FL_BL_AST)) {
713                 /* If this is a client-side namespace and this was the last
714                  * reference, put it on the LRU. */
715                 ldlm_lock_add_to_lru(lock);
716                 unlock_res_and_lock(lock);
717
718                 if (lock->l_flags & LDLM_FL_FAIL_LOC)
719                         OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
720
721                 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
722                  * are not supported by the server, otherwise, it is done on
723                  * enqueue. */
724                 if (!exp_connect_cancelset(lock->l_conn_export) &&
725                     !ns_connect_lru_resize(ns))
726                         ldlm_cancel_lru(ns, 0, LDLM_ASYNC, 0);
727         } else {
728                 unlock_res_and_lock(lock);
729         }
730
731         EXIT;
732 }
733
734 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
735 {
736         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
737         LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
738         ldlm_lock_decref_internal(lock, mode);
739         LDLM_LOCK_PUT(lock);
740 }
741
742 /* This will drop a lock reference and mark it for destruction, but will not
743  * necessarily cancel the lock before returning. */
744 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
745 {
746         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
747         ENTRY;
748
749         LASSERT(lock != NULL);
750
751         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
752         lock_res_and_lock(lock);
753         lock->l_flags |= LDLM_FL_CBPENDING;
754         unlock_res_and_lock(lock);
755         ldlm_lock_decref_internal(lock, mode);
756         LDLM_LOCK_PUT(lock);
757 }
758
759 struct sl_insert_point {
760         struct list_head *res_link;
761         struct list_head *mode_link;
762         struct list_head *policy_link;
763 };
764
765 /*
766  * search_granted_lock
767  *
768  * Description:
769  *      Finds a position to insert the new lock.
770  * Parameters:
771  *      queue [input]:  the granted list where search acts on;
772  *      req [input]:    the lock whose position to be located;
773  *      prev [output]:  positions within 3 lists to insert @req to
774  * Return Value:
775  *      filled @prev
776  * NOTE: called by
777  *  - ldlm_grant_lock_with_skiplist
778  */
779 static void search_granted_lock(struct list_head *queue,
780                                 struct ldlm_lock *req,
781                                 struct sl_insert_point *prev)
782 {
783         struct list_head *tmp;
784         struct ldlm_lock *lock, *mode_end, *policy_end;
785         ENTRY;
786
787         list_for_each(tmp, queue) {
788                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
789
790                 mode_end = list_entry(lock->l_sl_mode.prev, struct ldlm_lock,
791                                       l_sl_mode);
792
793                 if (lock->l_req_mode != req->l_req_mode) {
794                         /* jump to last lock of mode group */
795                         tmp = &mode_end->l_res_link;
796                         continue;
797                 }
798
799                 /* suitable mode group is found */
800                 if (lock->l_resource->lr_type == LDLM_PLAIN) {
801                         /* insert point is last lock of the mode group */
802                         prev->res_link = &mode_end->l_res_link;
803                         prev->mode_link = &mode_end->l_sl_mode;
804                         prev->policy_link = &req->l_sl_policy;
805                         EXIT;
806                         return;
807                 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
808                         for (;;) {
809                                 policy_end = list_entry(lock->l_sl_policy.prev,
810                                                         struct ldlm_lock,
811                                                         l_sl_policy);
812
813                                 if (lock->l_policy_data.l_inodebits.bits ==
814                                     req->l_policy_data.l_inodebits.bits) {
815                                         /* insert point is last lock of
816                                          * the policy group */
817                                         prev->res_link =
818                                                 &policy_end->l_res_link;
819                                         prev->mode_link =
820                                                 &policy_end->l_sl_mode;
821                                         prev->policy_link =
822                                                 &policy_end->l_sl_policy;
823                                         EXIT;
824                                         return;
825                                 }
826
827                                 if (policy_end == mode_end)
828                                         /* done with mode group */
829                                         break;
830
831                                 /* go to next policy group within mode group */
832                                 tmp = policy_end->l_res_link.next;
833                                 lock = list_entry(tmp, struct ldlm_lock,
834                                                   l_res_link);
835                         }  /* loop over policy groups within the mode group */
836
837                         /* insert point is last lock of the mode group,
838                          * new policy group is started */
839                         prev->res_link = &mode_end->l_res_link;
840                         prev->mode_link = &mode_end->l_sl_mode;
841                         prev->policy_link = &req->l_sl_policy;
842                         EXIT;
843                         return;
844                 } else {
845                         LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock");
846                         LBUG();
847                 }
848         }
849
850         /* insert point is last lock on the queue,
851          * new mode group and new policy group are started */
852         prev->res_link = queue->prev;
853         prev->mode_link = &req->l_sl_mode;
854         prev->policy_link = &req->l_sl_policy;
855         EXIT;
856         return;
857 }
858
859 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
860                                        struct sl_insert_point *prev)
861 {
862         struct ldlm_resource *res = lock->l_resource;
863         ENTRY;
864
865         check_res_locked(res);
866
867         ldlm_resource_dump(D_INFO, res);
868         CDEBUG(D_OTHER, "About to add this lock:\n");
869         ldlm_lock_dump(D_OTHER, lock, 0);
870
871         if (lock->l_destroyed) {
872                 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
873                 return;
874         }
875
876         LASSERT(list_empty(&lock->l_res_link));
877         LASSERT(list_empty(&lock->l_sl_mode));
878         LASSERT(list_empty(&lock->l_sl_policy));
879
880         list_add(&lock->l_res_link, prev->res_link);
881         list_add(&lock->l_sl_mode, prev->mode_link);
882         list_add(&lock->l_sl_policy, prev->policy_link);
883
884         EXIT;
885 }
886
887 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
888 {
889         struct sl_insert_point prev;
890         ENTRY;
891
892         LASSERT(lock->l_req_mode == lock->l_granted_mode);
893
894         search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
895         ldlm_granted_list_add_lock(lock, &prev);
896         EXIT;
897 }
898
899 /* NOTE: called by
900  *  - ldlm_lock_enqueue
901  *  - ldlm_reprocess_queue
902  *  - ldlm_lock_convert
903  *
904  * must be called with lr_lock held
905  */
906 void ldlm_grant_lock(struct ldlm_lock *lock, struct list_head *work_list)
907 {
908         struct ldlm_resource *res = lock->l_resource;
909         ENTRY;
910
911         check_res_locked(res);
912
913         lock->l_granted_mode = lock->l_req_mode;
914         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
915                 ldlm_grant_lock_with_skiplist(lock);
916         else if (res->lr_type == LDLM_EXTENT)
917                 ldlm_extent_add_lock(res, lock);
918         else
919                 ldlm_resource_add_lock(res, &res->lr_granted, lock);
920
921         if (lock->l_granted_mode < res->lr_most_restr)
922                 res->lr_most_restr = lock->l_granted_mode;
923
924         if (work_list && lock->l_completion_ast != NULL)
925                 ldlm_add_ast_work_item(lock, NULL, work_list);
926
927         ldlm_pool_add(&res->lr_namespace->ns_pool, lock);
928         EXIT;
929 }
930
931 /* returns a referenced lock or NULL.  See the flag descriptions below, in the
932  * comment above ldlm_lock_match */
933 static struct ldlm_lock *search_queue(struct list_head *queue,
934                                       ldlm_mode_t *mode,
935                                       ldlm_policy_data_t *policy,
936                                       struct ldlm_lock *old_lock,
937                                       int flags, int unref)
938 {
939         struct ldlm_lock *lock;
940         struct list_head *tmp;
941
942         list_for_each(tmp, queue) {
943                 ldlm_mode_t match;
944
945                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
946
947                 if (lock == old_lock)
948                         break;
949
950                 /* llite sometimes wants to match locks that will be
951                  * canceled when their users drop, but we allow it to match
952                  * if it passes in CBPENDING and the lock still has users.
953                  * this is generally only going to be used by children
954                  * whose parents already hold a lock so forward progress
955                  * can still happen. */
956                 if (lock->l_flags & LDLM_FL_CBPENDING &&
957                     !(flags & LDLM_FL_CBPENDING))
958                         continue;
959                 if (!unref && lock->l_flags & LDLM_FL_CBPENDING &&
960                     lock->l_readers == 0 && lock->l_writers == 0)
961                         continue;
962
963                 if (!(lock->l_req_mode & *mode))
964                         continue;
965                 match = lock->l_req_mode;
966
967                 if (lock->l_resource->lr_type == LDLM_EXTENT &&
968                     (lock->l_policy_data.l_extent.start >
969                      policy->l_extent.start ||
970                      lock->l_policy_data.l_extent.end < policy->l_extent.end))
971                         continue;
972
973                 if (unlikely(match == LCK_GROUP) &&
974                     lock->l_resource->lr_type == LDLM_EXTENT &&
975                     lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
976                         continue;
977
978                 /* We match if we have existing lock with same or wider set
979                    of bits. */
980                 if (lock->l_resource->lr_type == LDLM_IBITS &&
981                      ((lock->l_policy_data.l_inodebits.bits &
982                       policy->l_inodebits.bits) !=
983                       policy->l_inodebits.bits))
984                         continue;
985
986                 if (!unref &&
987                     (lock->l_destroyed || (lock->l_flags & LDLM_FL_FAILED)))
988                         continue;
989
990                 if ((flags & LDLM_FL_LOCAL_ONLY) &&
991                     !(lock->l_flags & LDLM_FL_LOCAL))
992                         continue;
993
994                 if (flags & LDLM_FL_TEST_LOCK) {
995                         LDLM_LOCK_GET(lock);
996                         ldlm_lock_touch_in_lru(lock);
997                 } else {
998                         ldlm_lock_addref_internal_nolock(lock, match);
999                 }
1000                 *mode = match;
1001                 return lock;
1002         }
1003
1004         return NULL;
1005 }
1006
1007 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
1008 {
1009         lock->l_flags |= LDLM_FL_LVB_READY;
1010         cfs_waitq_signal(&lock->l_waitq);
1011 }
1012
1013 void ldlm_lock_allow_match(struct ldlm_lock *lock)
1014 {
1015         lock_res_and_lock(lock);
1016         ldlm_lock_allow_match_locked(lock);
1017         unlock_res_and_lock(lock);
1018 }
1019
1020 /* Can be called in two ways:
1021  *
1022  * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1023  * for a duplicate of.
1024  *
1025  * Otherwise, all of the fields must be filled in, to match against.
1026  *
1027  * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1028  *     server (ie, connh is NULL)
1029  * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1030  *     list will be considered
1031  * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1032  *     to be canceled can still be matched as long as they still have reader
1033  *     or writer refernces
1034  * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1035  *     just tell us if we would have matched.
1036  *
1037  * Returns 1 if it finds an already-existing lock that is compatible; in this
1038  * case, lockh is filled in with a addref()ed lock
1039  *
1040  * we also check security context, if that failed we simply return 0 (to keep
1041  * caller code unchanged), the context failure will be discovered by caller
1042  * sometime later.
1043  */
1044 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
1045                             const struct ldlm_res_id *res_id, ldlm_type_t type,
1046                             ldlm_policy_data_t *policy, ldlm_mode_t mode,
1047                             struct lustre_handle *lockh, int unref)
1048 {
1049         struct ldlm_resource *res;
1050         struct ldlm_lock *lock, *old_lock = NULL;
1051         int rc = 0;
1052         ENTRY;
1053
1054         if (ns == NULL) {
1055                 old_lock = ldlm_handle2lock(lockh);
1056                 LASSERT(old_lock);
1057
1058                 ns = old_lock->l_resource->lr_namespace;
1059                 res_id = &old_lock->l_resource->lr_name;
1060                 type = old_lock->l_resource->lr_type;
1061                 mode = old_lock->l_req_mode;
1062         }
1063
1064         res = ldlm_resource_get(ns, NULL, res_id, type, 0);
1065         if (res == NULL) {
1066                 LASSERT(old_lock == NULL);
1067                 RETURN(0);
1068         }
1069
1070         LDLM_RESOURCE_ADDREF(res);
1071         lock_res(res);
1072
1073         lock = search_queue(&res->lr_granted, &mode, policy, old_lock,
1074                             flags, unref);
1075         if (lock != NULL)
1076                 GOTO(out, rc = 1);
1077         if (flags & LDLM_FL_BLOCK_GRANTED)
1078                 GOTO(out, rc = 0);
1079         lock = search_queue(&res->lr_converting, &mode, policy, old_lock,
1080                             flags, unref);
1081         if (lock != NULL)
1082                 GOTO(out, rc = 1);
1083         lock = search_queue(&res->lr_waiting, &mode, policy, old_lock,
1084                             flags, unref);
1085         if (lock != NULL)
1086                 GOTO(out, rc = 1);
1087
1088         EXIT;
1089  out:
1090         unlock_res(res);
1091         LDLM_RESOURCE_DELREF(res);
1092         ldlm_resource_putref(res);
1093
1094         if (lock) {
1095                 ldlm_lock2handle(lock, lockh);
1096                 if ((flags & LDLM_FL_LVB_READY) &&
1097                     (!(lock->l_flags & LDLM_FL_LVB_READY))) {
1098                         struct l_wait_info lwi;
1099                         if (lock->l_completion_ast) {
1100                                 int err = lock->l_completion_ast(lock,
1101                                                           LDLM_FL_WAIT_NOREPROC,
1102                                                                  NULL);
1103                                 if (err) {
1104                                         if (flags & LDLM_FL_TEST_LOCK)
1105                                                 LDLM_LOCK_RELEASE(lock);
1106                                         else
1107                                                 ldlm_lock_decref_internal(lock,
1108                                                                           mode);
1109                                         rc = 0;
1110                                         goto out2;
1111                                 }
1112                         }
1113
1114                         lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
1115                                                NULL, LWI_ON_SIGNAL_NOOP, NULL);
1116
1117                         /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1118                         l_wait_event(lock->l_waitq,
1119                                      (lock->l_flags & LDLM_FL_LVB_READY), &lwi);
1120                 }
1121         }
1122  out2:
1123         if (rc) {
1124                 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1125                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1126                                 res_id->name[2] : policy->l_extent.start,
1127                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1128                                 res_id->name[3] : policy->l_extent.end);
1129
1130                 /* check user's security context */
1131                 if (lock->l_conn_export &&
1132                     sptlrpc_import_check_ctx(
1133                                 class_exp2cliimp(lock->l_conn_export))) {
1134                         if (!(flags & LDLM_FL_TEST_LOCK))
1135                                 ldlm_lock_decref_internal(lock, mode);
1136                         rc = 0;
1137                 }
1138
1139                 if (flags & LDLM_FL_TEST_LOCK)
1140                         LDLM_LOCK_RELEASE(lock);
1141
1142         } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1143                 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1144                                   LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1145                                   type, mode, res_id->name[0], res_id->name[1],
1146                                   (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1147                                         res_id->name[2] :policy->l_extent.start,
1148                                   (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1149                                         res_id->name[3] : policy->l_extent.end);
1150         }
1151         if (old_lock)
1152                 LDLM_LOCK_PUT(old_lock);
1153
1154         return rc ? mode : 0;
1155 }
1156
1157 /* Returns a referenced lock */
1158 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1159                                    const struct ldlm_res_id *res_id,
1160                                    ldlm_type_t type,
1161                                    ldlm_mode_t mode,
1162                                    const struct ldlm_callback_suite *cbs,
1163                                    void *data, __u32 lvb_len)
1164 {
1165         struct ldlm_lock *lock;
1166         struct ldlm_resource *res;
1167         ENTRY;
1168
1169         res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1170         if (res == NULL)
1171                 RETURN(NULL);
1172
1173         lock = ldlm_lock_new(res);
1174         ldlm_resource_putref(res);
1175
1176         if (lock == NULL)
1177                 RETURN(NULL);
1178
1179         lock->l_req_mode = mode;
1180         lock->l_ast_data = data;
1181         lock->l_pid = cfs_curproc_pid();
1182         if (cbs) {
1183                 lock->l_blocking_ast = cbs->lcs_blocking;
1184                 lock->l_completion_ast = cbs->lcs_completion;
1185                 lock->l_glimpse_ast = cbs->lcs_glimpse;
1186                 lock->l_weigh_ast = cbs->lcs_weigh;
1187         }
1188
1189         lock->l_tree_node = NULL;
1190         /* if this is the extent lock, allocate the interval tree node */
1191         if (type == LDLM_EXTENT) {
1192                 if (ldlm_interval_alloc(lock) == NULL)
1193                         GOTO(out, 0);
1194         }
1195
1196         if (lvb_len) {
1197                 lock->l_lvb_len = lvb_len;
1198                 OBD_ALLOC(lock->l_lvb_data, lvb_len);
1199                 if (lock->l_lvb_data == NULL)
1200                         GOTO(out, 0);
1201         }
1202
1203         RETURN(lock);
1204
1205 out:
1206         if (lock->l_lvb_data)
1207                 OBD_FREE(lock->l_lvb_data, lvb_len);
1208         ldlm_interval_free(ldlm_interval_detach(lock));
1209         OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
1210         return NULL;
1211 }
1212
1213 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1214                                struct ldlm_lock **lockp,
1215                                void *cookie, int *flags)
1216 {
1217         struct ldlm_lock *lock = *lockp;
1218         struct ldlm_resource *res = lock->l_resource;
1219         int local = ns_is_client(res->lr_namespace);
1220         ldlm_processing_policy policy;
1221         ldlm_error_t rc = ELDLM_OK;
1222         struct ldlm_interval *node = NULL;
1223         ENTRY;
1224
1225         lock->l_last_activity = cfs_time_current_sec();
1226         /* policies are not executed on the client or during replay */
1227         if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1228             && !local && ns->ns_policy) {
1229                 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1230                                    NULL);
1231                 if (rc == ELDLM_LOCK_REPLACED) {
1232                         /* The lock that was returned has already been granted,
1233                          * and placed into lockp.  If it's not the same as the
1234                          * one we passed in, then destroy the old one and our
1235                          * work here is done. */
1236                         if (lock != *lockp) {
1237                                 ldlm_lock_destroy(lock);
1238                                 LDLM_LOCK_RELEASE(lock);
1239                         }
1240                         *flags |= LDLM_FL_LOCK_CHANGED;
1241                         RETURN(0);
1242                 } else if (rc != ELDLM_OK ||
1243                            (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1244                         ldlm_lock_destroy(lock);
1245                         RETURN(rc);
1246                 }
1247         }
1248
1249         /* For a replaying lock, it might be already in granted list. So
1250          * unlinking the lock will cause the interval node to be freed, we
1251          * have to allocate the interval node early otherwise we can't regrant
1252          * this lock in the future. - jay */
1253         if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1254                 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO);
1255
1256         lock_res_and_lock(lock);
1257         if (local && lock->l_req_mode == lock->l_granted_mode) {
1258                 /* The server returned a blocked lock, but it was granted
1259                  * before we got a chance to actually enqueue it.  We don't
1260                  * need to do anything else. */
1261                 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
1262                             LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
1263                 GOTO(out, ELDLM_OK);
1264         }
1265
1266         ldlm_resource_unlink_lock(lock);
1267         if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1268                 if (node == NULL) {
1269                         ldlm_lock_destroy_nolock(lock);
1270                         GOTO(out, rc = -ENOMEM);
1271                 }
1272
1273                 CFS_INIT_LIST_HEAD(&node->li_group);
1274                 ldlm_interval_attach(node, lock);
1275                 node = NULL;
1276         }
1277
1278         /* Some flags from the enqueue want to make it into the AST, via the
1279          * lock's l_flags. */
1280         lock->l_flags |= *flags & LDLM_AST_DISCARD_DATA;
1281
1282         /* This distinction between local lock trees is very important; a client
1283          * namespace only has information about locks taken by that client, and
1284          * thus doesn't have enough information to decide for itself if it can
1285          * be granted (below).  In this case, we do exactly what the server
1286          * tells us to do, as dictated by the 'flags'.
1287          *
1288          * We do exactly the same thing during recovery, when the server is
1289          * more or less trusting the clients not to lie.
1290          *
1291          * FIXME (bug 268): Detect obvious lies by checking compatibility in
1292          * granted/converting queues. */
1293         if (local) {
1294                 if (*flags & LDLM_FL_BLOCK_CONV)
1295                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1296                 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1297                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1298                 else
1299                         ldlm_grant_lock(lock, NULL);
1300                 GOTO(out, ELDLM_OK);
1301         } else if (*flags & LDLM_FL_REPLAY) {
1302                 if (*flags & LDLM_FL_BLOCK_CONV) {
1303                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1304                         GOTO(out, ELDLM_OK);
1305                 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1306                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1307                         GOTO(out, ELDLM_OK);
1308                 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1309                         ldlm_grant_lock(lock, NULL);
1310                         GOTO(out, ELDLM_OK);
1311                 }
1312                 /* If no flags, fall through to normal enqueue path. */
1313         }
1314
1315         policy = ldlm_processing_policy_table[res->lr_type];
1316         policy(lock, flags, 1, &rc, NULL);
1317         GOTO(out, rc);
1318 out:
1319         unlock_res_and_lock(lock);
1320         if (node)
1321                 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1322         return rc;
1323 }
1324
1325 /* Must be called with namespace taken: queue is waiting or converting. */
1326 int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
1327                          struct list_head *work_list)
1328 {
1329         struct list_head *tmp, *pos;
1330         ldlm_processing_policy policy;
1331         int flags;
1332         int rc = LDLM_ITER_CONTINUE;
1333         ldlm_error_t err;
1334         ENTRY;
1335
1336         check_res_locked(res);
1337
1338         policy = ldlm_processing_policy_table[res->lr_type];
1339         LASSERT(policy);
1340
1341         list_for_each_safe(tmp, pos, queue) {
1342                 struct ldlm_lock *pending;
1343                 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
1344
1345                 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1346
1347                 flags = 0;
1348                 rc = policy(pending, &flags, 0, &err, work_list);
1349                 if (rc != LDLM_ITER_CONTINUE)
1350                         break;
1351         }
1352
1353         RETURN(rc);
1354 }
1355
1356 /* Helper function for ldlm_run_ast_work().
1357  *
1358  * Send an existing rpc set specified by @arg->set and then
1359  * destroy it. Create new one if @do_create flag is set. */
1360 static void
1361 ldlm_send_and_maybe_create_set(struct ldlm_cb_set_arg *arg, int do_create)
1362 {
1363         ENTRY;
1364
1365         ptlrpc_set_wait(arg->set);
1366         if (arg->type == LDLM_BL_CALLBACK)
1367                 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_GLIMPSE, 2);
1368         ptlrpc_set_destroy(arg->set);
1369
1370         if (do_create)
1371                 arg->set = ptlrpc_prep_set();
1372
1373         EXIT;
1374 }
1375
1376 static int
1377 ldlm_work_bl_ast_lock(struct list_head *tmp, struct ldlm_cb_set_arg *arg)
1378 {
1379         struct ldlm_lock_desc d;
1380         struct ldlm_lock *lock = list_entry(tmp, struct ldlm_lock, l_bl_ast);
1381         ENTRY;
1382
1383         /* nobody should touch l_bl_ast */
1384         lock_res_and_lock(lock);
1385         list_del_init(&lock->l_bl_ast);
1386
1387         LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1388         LASSERT(lock->l_bl_ast_run == 0);
1389         LASSERT(lock->l_blocking_lock);
1390         lock->l_bl_ast_run++;
1391         unlock_res_and_lock(lock);
1392
1393         ldlm_lock2desc(lock->l_blocking_lock, &d);
1394
1395         lock->l_blocking_ast(lock, &d, (void *)arg,
1396                              LDLM_CB_BLOCKING);
1397         LDLM_LOCK_RELEASE(lock->l_blocking_lock);
1398         lock->l_blocking_lock = NULL;
1399         LDLM_LOCK_RELEASE(lock);
1400
1401         RETURN(1);
1402 }
1403
1404 static int
1405 ldlm_work_cp_ast_lock(struct list_head *tmp, struct ldlm_cb_set_arg *arg)
1406 {
1407         struct ldlm_lock *lock = list_entry(tmp, struct ldlm_lock, l_cp_ast);
1408         ldlm_completion_callback completion_callback;
1409         int rc = 0;
1410         ENTRY;
1411
1412         /* It's possible to receive a completion AST before we've set
1413          * the l_completion_ast pointer: either because the AST arrived
1414          * before the reply, or simply because there's a small race
1415          * window between receiving the reply and finishing the local
1416          * enqueue. (bug 842)
1417          *
1418          * This can't happen with the blocking_ast, however, because we
1419          * will never call the local blocking_ast until we drop our
1420          * reader/writer reference, which we won't do until we get the
1421          * reply and finish enqueueing. */
1422
1423         /* nobody should touch l_cp_ast */
1424         lock_res_and_lock(lock);
1425         list_del_init(&lock->l_cp_ast);
1426         LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1427         /* save l_completion_ast since it can be changed by
1428          * mds_intent_policy(), see bug 14225 */
1429         completion_callback = lock->l_completion_ast;
1430         lock->l_flags &= ~LDLM_FL_CP_REQD;
1431         unlock_res_and_lock(lock);
1432
1433         if (completion_callback != NULL) {
1434                 completion_callback(lock, 0, (void *)arg);
1435                 rc = 1;
1436         }
1437         LDLM_LOCK_RELEASE(lock);
1438
1439         RETURN(rc);
1440 }
1441
1442 static int
1443 ldlm_work_revoke_ast_lock(struct list_head *tmp, struct ldlm_cb_set_arg *arg)
1444 {
1445         struct ldlm_lock_desc desc;
1446         struct ldlm_lock *lock = list_entry(tmp, struct ldlm_lock, l_rk_ast);
1447         ENTRY;
1448
1449         list_del_init(&lock->l_rk_ast);
1450
1451         /* the desc just pretend to exclusive */
1452         ldlm_lock2desc(lock, &desc);
1453         desc.l_req_mode = LCK_EX;
1454         desc.l_granted_mode = 0;
1455
1456         lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
1457         LDLM_LOCK_RELEASE(lock);
1458
1459         RETURN(1);
1460 }
1461
1462 int ldlm_run_ast_work(struct list_head *rpc_list, ldlm_desc_ast_t ast_type)
1463 {
1464         struct ldlm_cb_set_arg arg;
1465         struct list_head *tmp, *pos;
1466         int (*work_ast_lock)(struct list_head *tmp,struct ldlm_cb_set_arg *arg);
1467         int ast_count;
1468         ENTRY;
1469
1470         if (list_empty(rpc_list))
1471                 RETURN(0);
1472
1473         arg.set = ptlrpc_prep_set();
1474         if (NULL == arg.set)
1475                 RETURN(-ERESTART);
1476         atomic_set(&arg.restart, 0);
1477         switch (ast_type) {
1478         case LDLM_WORK_BL_AST:
1479                 arg.type = LDLM_BL_CALLBACK;
1480                 work_ast_lock = ldlm_work_bl_ast_lock;
1481                 break;
1482         case LDLM_WORK_CP_AST:
1483                 arg.type = LDLM_CP_CALLBACK;
1484                 work_ast_lock = ldlm_work_cp_ast_lock;
1485                 break;
1486         case LDLM_WORK_REVOKE_AST:
1487                 arg.type = LDLM_BL_CALLBACK;
1488                 work_ast_lock = ldlm_work_revoke_ast_lock;
1489                 break;
1490         default:
1491                 LBUG();
1492         }
1493
1494         ast_count = 0;
1495         list_for_each_safe(tmp, pos, rpc_list) {
1496                 ast_count += work_ast_lock(tmp, &arg);
1497
1498                 /* Send the request set if it exceeds the PARALLEL_AST_LIMIT,
1499                  * and create a new set for requests that remained in
1500                  * @rpc_list */
1501                 if (unlikely(ast_count == PARALLEL_AST_LIMIT)) {
1502                         ldlm_send_and_maybe_create_set(&arg, 1);
1503                         ast_count = 0;
1504                 }
1505         }
1506
1507         if (ast_count > 0)
1508                 ldlm_send_and_maybe_create_set(&arg, 0);
1509         else
1510                 /* In case when number of ASTs is multiply of
1511                  * PARALLEL_AST_LIMIT or @rpc_list was initially empty,
1512                  * @arg.set must be destroyed here, otherwise we get
1513                  * write memory leaking. */
1514                 ptlrpc_set_destroy(arg.set);
1515
1516         RETURN(atomic_read(&arg.restart) ? -ERESTART : 0);
1517 }
1518
1519 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
1520 {
1521         ldlm_reprocess_all(res);
1522         return LDLM_ITER_CONTINUE;
1523 }
1524
1525 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
1526 {
1527         struct list_head *tmp;
1528         int i, rc;
1529
1530         if (ns == NULL)
1531                 return;
1532
1533         ENTRY;
1534         spin_lock(&ns->ns_hash_lock);
1535         for (i = 0; i < RES_HASH_SIZE; i++) {
1536                 tmp = ns->ns_hash[i].next;
1537                 while (tmp != &(ns->ns_hash[i])) {
1538                         struct ldlm_resource *res =
1539                                 list_entry(tmp, struct ldlm_resource, lr_hash);
1540
1541                         ldlm_resource_getref(res);
1542                         spin_unlock(&ns->ns_hash_lock);
1543                         LDLM_RESOURCE_ADDREF(res);
1544
1545                         rc = reprocess_one_queue(res, NULL);
1546
1547                         LDLM_RESOURCE_DELREF(res);
1548                         spin_lock(&ns->ns_hash_lock);
1549                         tmp = tmp->next;
1550                         ldlm_resource_putref_locked(res);
1551
1552                         if (rc == LDLM_ITER_STOP)
1553                                 GOTO(out, rc);
1554                 }
1555         }
1556  out:
1557         spin_unlock(&ns->ns_hash_lock);
1558         EXIT;
1559 }
1560
1561 void ldlm_reprocess_all(struct ldlm_resource *res)
1562 {
1563         CFS_LIST_HEAD(rpc_list);
1564         int rc;
1565         ENTRY;
1566
1567         /* Local lock trees don't get reprocessed. */
1568         if (ns_is_client(res->lr_namespace)) {
1569                 EXIT;
1570                 return;
1571         }
1572
1573  restart:
1574         lock_res(res);
1575         rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
1576         if (rc == LDLM_ITER_CONTINUE)
1577                 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
1578         unlock_res(res);
1579
1580         rc = ldlm_run_ast_work(&rpc_list, LDLM_WORK_CP_AST);
1581         if (rc == -ERESTART) {
1582                 LASSERT(list_empty(&rpc_list));
1583                 goto restart;
1584         }
1585         EXIT;
1586 }
1587
1588 void ldlm_cancel_callback(struct ldlm_lock *lock)
1589 {
1590         check_res_locked(lock->l_resource);
1591         if (!(lock->l_flags & LDLM_FL_CANCEL)) {
1592                 lock->l_flags |= LDLM_FL_CANCEL;
1593                 if (lock->l_blocking_ast) {
1594                         // l_check_no_ns_lock(ns);
1595                         unlock_res_and_lock(lock);
1596                         lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
1597                                              LDLM_CB_CANCELING);
1598                         lock_res_and_lock(lock);
1599                 } else {
1600                         LDLM_DEBUG(lock, "no blocking ast");
1601                 }
1602         }
1603         lock->l_flags |= LDLM_FL_BL_DONE;
1604 }
1605
1606 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
1607 {
1608         if (req->l_resource->lr_type != LDLM_PLAIN &&
1609             req->l_resource->lr_type != LDLM_IBITS)
1610                 return;
1611
1612         list_del_init(&req->l_sl_policy);
1613         list_del_init(&req->l_sl_mode);
1614 }
1615
1616 void ldlm_lock_cancel(struct ldlm_lock *lock)
1617 {
1618         struct ldlm_resource *res;
1619         struct ldlm_namespace *ns;
1620         ENTRY;
1621
1622         lock_res_and_lock(lock);
1623
1624         res = lock->l_resource;
1625         ns = res->lr_namespace;
1626
1627         /* Please do not, no matter how tempting, remove this LBUG without
1628          * talking to me first. -phik */
1629         if (lock->l_readers || lock->l_writers) {
1630                 LDLM_ERROR(lock, "lock still has references");
1631                 LBUG();
1632         }
1633
1634         ldlm_del_waiting_lock(lock);
1635
1636         /* Releases cancel callback. */
1637         ldlm_cancel_callback(lock);
1638
1639         /* Yes, second time, just in case it was added again while we were
1640            running with no res lock in ldlm_cancel_callback */
1641         ldlm_del_waiting_lock(lock);
1642         ldlm_resource_unlink_lock(lock);
1643         ldlm_lock_destroy_nolock(lock);
1644
1645         if (lock->l_granted_mode == lock->l_req_mode)
1646                 ldlm_pool_del(&ns->ns_pool, lock);
1647
1648         /* Make sure we will not be called again for same lock what is possible
1649          * if not to zero out lock->l_granted_mode */
1650         lock->l_granted_mode = LCK_MINMODE;
1651         unlock_res_and_lock(lock);
1652
1653         EXIT;
1654 }
1655
1656 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
1657 {
1658         struct ldlm_lock *lock = ldlm_handle2lock(lockh);
1659         ENTRY;
1660
1661         if (lock == NULL)
1662                 RETURN(-EINVAL);
1663
1664         lock->l_ast_data = data;
1665         LDLM_LOCK_PUT(lock);
1666         RETURN(0);
1667 }
1668
1669 void ldlm_cancel_locks_for_export_cb(void *obj, void *data)
1670 {
1671         struct obd_export    *exp = data;
1672         struct ldlm_lock     *lock = obj;
1673         struct ldlm_resource *res;
1674
1675         res = ldlm_resource_getref(lock->l_resource);
1676         LDLM_LOCK_GET(lock);
1677
1678         LDLM_DEBUG(lock, "export %p", exp);
1679         ldlm_res_lvbo_update(res, NULL, 1);
1680         ldlm_lock_cancel(lock);
1681         ldlm_reprocess_all(res);
1682         ldlm_resource_putref(res);
1683         LDLM_LOCK_RELEASE(lock);
1684 }
1685
1686 void ldlm_cancel_locks_for_export(struct obd_export *exp)
1687 {
1688         cfs_hash_for_each_empty(exp->exp_lock_hash,
1689                                 ldlm_cancel_locks_for_export_cb, exp);
1690 }
1691
1692 /**
1693  * Downgrade an exclusive lock.
1694  *
1695  * A fast variant of ldlm_lock_convert for convertion of exclusive
1696  * locks. The convertion is always successful.
1697  *
1698  * \param lock A lock to convert
1699  * \param new_mode new lock mode
1700  */
1701 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode)
1702 {
1703         struct ldlm_namespace *ns;
1704         ENTRY;
1705
1706         LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
1707         LASSERT(new_mode == LCK_COS);
1708
1709         lock_res_and_lock(lock);
1710         ldlm_resource_unlink_lock(lock);
1711         /*
1712          * Remove the lock from pool as it will be added again in
1713          * ldlm_grant_lock() called below.
1714          */
1715         ns = lock->l_resource->lr_namespace;
1716         ldlm_pool_del(&ns->ns_pool, lock);
1717
1718         lock->l_req_mode = new_mode;
1719         ldlm_grant_lock(lock, NULL);
1720         unlock_res_and_lock(lock);
1721         ldlm_reprocess_all(lock->l_resource);
1722
1723         EXIT;
1724 }
1725
1726 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1727                                         __u32 *flags)
1728 {
1729         CFS_LIST_HEAD(rpc_list);
1730         struct ldlm_resource *res;
1731         struct ldlm_namespace *ns;
1732         int granted = 0;
1733         int old_mode, rc;
1734         struct sl_insert_point prev;
1735         ldlm_error_t err;
1736         struct ldlm_interval *node;
1737         ENTRY;
1738
1739         if (new_mode == lock->l_granted_mode) { // No changes? Just return.
1740                 *flags |= LDLM_FL_BLOCK_GRANTED;
1741                 RETURN(lock->l_resource);
1742         }
1743
1744         /* I can't check the type of lock here because the bitlock of lock
1745          * is not held here, so do the allocation blindly. -jay */
1746         OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO);
1747         if (node == NULL)  /* Actually, this causes EDEADLOCK to be returned */
1748                 RETURN(NULL);
1749
1750         LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
1751                  "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
1752
1753         lock_res_and_lock(lock);
1754
1755         res = lock->l_resource;
1756         ns = res->lr_namespace;
1757
1758         old_mode = lock->l_req_mode;
1759         lock->l_req_mode = new_mode;
1760         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
1761                 /* remember the lock position where the lock might be
1762                  * added back to the granted list later and also
1763                  * remember the join mode for skiplist fixing. */
1764                 prev.res_link = lock->l_res_link.prev;
1765                 prev.mode_link = lock->l_sl_mode.prev;
1766                 prev.policy_link = lock->l_sl_policy.prev;
1767                 ldlm_resource_unlink_lock(lock);
1768         } else {
1769                 ldlm_resource_unlink_lock(lock);
1770                 if (res->lr_type == LDLM_EXTENT) {
1771                         /* FIXME: ugly code, I have to attach the lock to a
1772                          * interval node again since perhaps it will be granted
1773                          * soon */
1774                         CFS_INIT_LIST_HEAD(&node->li_group);
1775                         ldlm_interval_attach(node, lock);
1776                         node = NULL;
1777                 }
1778         }
1779
1780         /*
1781          * Remove old lock from the pool before adding the lock with new
1782          * mode below in ->policy()
1783          */
1784         ldlm_pool_del(&ns->ns_pool, lock);
1785
1786         /* If this is a local resource, put it on the appropriate list. */
1787         if (ns_is_client(res->lr_namespace)) {
1788                 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
1789                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1790                 } else {
1791                         /* This should never happen, because of the way the
1792                          * server handles conversions. */
1793                         LDLM_ERROR(lock, "Erroneous flags %d on local lock\n",
1794                                    *flags);
1795                         LBUG();
1796
1797                         ldlm_grant_lock(lock, &rpc_list);
1798                         granted = 1;
1799                         /* FIXME: completion handling not with ns_lock held ! */
1800                         if (lock->l_completion_ast)
1801                                 lock->l_completion_ast(lock, 0, NULL);
1802                 }
1803         } else {
1804                 int pflags = 0;
1805                 ldlm_processing_policy policy;
1806                 policy = ldlm_processing_policy_table[res->lr_type];
1807                 rc = policy(lock, &pflags, 0, &err, &rpc_list);
1808                 if (rc == LDLM_ITER_STOP) {
1809                         lock->l_req_mode = old_mode;
1810                         if (res->lr_type == LDLM_EXTENT)
1811                                 ldlm_extent_add_lock(res, lock);
1812                         else
1813                                 ldlm_granted_list_add_lock(lock, &prev);
1814
1815                         res = NULL;
1816                 } else {
1817                         *flags |= LDLM_FL_BLOCK_GRANTED;
1818                         granted = 1;
1819                 }
1820         }
1821         unlock_res_and_lock(lock);
1822
1823         if (granted)
1824                 ldlm_run_ast_work(&rpc_list, LDLM_WORK_CP_AST);
1825         if (node)
1826                 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1827         RETURN(res);
1828 }
1829
1830 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos)
1831 {
1832         struct obd_device *obd = NULL;
1833
1834         if (!((libcfs_debug | D_ERROR) & level))
1835                 return;
1836
1837         if (!lock) {
1838                 CDEBUG(level, "  NULL LDLM lock\n");
1839                 return;
1840         }
1841
1842         CDEBUG(level," -- Lock dump: %p/"LPX64" (rc: %d) (pos: %d) (pid: %d)\n",
1843                lock, lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1844                pos, lock->l_pid);
1845         if (lock->l_conn_export != NULL)
1846                 obd = lock->l_conn_export->exp_obd;
1847         if (lock->l_export && lock->l_export->exp_connection) {
1848                 CDEBUG(level, "  Node: NID %s (rhandle: "LPX64")\n",
1849                      libcfs_nid2str(lock->l_export->exp_connection->c_peer.nid),
1850                      lock->l_remote_handle.cookie);
1851         } else if (obd == NULL) {
1852                 CDEBUG(level, "  Node: local\n");
1853         } else {
1854                 struct obd_import *imp = obd->u.cli.cl_import;
1855                 CDEBUG(level, "  Node: NID %s (rhandle: "LPX64")\n",
1856                        libcfs_nid2str(imp->imp_connection->c_peer.nid),
1857                        lock->l_remote_handle.cookie);
1858         }
1859         CDEBUG(level, "  Resource: %p ("LPU64"/"LPU64"/"LPU64")\n",
1860                   lock->l_resource,
1861                   lock->l_resource->lr_name.name[0],
1862                   lock->l_resource->lr_name.name[1],
1863                   lock->l_resource->lr_name.name[2]);
1864         CDEBUG(level, "  Req mode: %s, grant mode: %s, rc: %u, read: %d, "
1865                "write: %d flags: "LPX64"\n", ldlm_lockname[lock->l_req_mode],
1866                ldlm_lockname[lock->l_granted_mode],
1867                atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers,
1868                lock->l_flags);
1869         if (lock->l_resource->lr_type == LDLM_EXTENT)
1870                 CDEBUG(level, "  Extent: "LPU64" -> "LPU64
1871                        " (req "LPU64"-"LPU64")\n",
1872                        lock->l_policy_data.l_extent.start,
1873                        lock->l_policy_data.l_extent.end,
1874                        lock->l_req_extent.start, lock->l_req_extent.end);
1875         else if (lock->l_resource->lr_type == LDLM_FLOCK)
1876                 CDEBUG(level, "  Pid: %d Extent: "LPU64" -> "LPU64"\n",
1877                        lock->l_policy_data.l_flock.pid,
1878                        lock->l_policy_data.l_flock.start,
1879                        lock->l_policy_data.l_flock.end);
1880        else if (lock->l_resource->lr_type == LDLM_IBITS)
1881                 CDEBUG(level, "  Bits: "LPX64"\n",
1882                        lock->l_policy_data.l_inodebits.bits);
1883 }
1884
1885 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
1886 {
1887         struct ldlm_lock *lock;
1888
1889         if (!((libcfs_debug | D_ERROR) & level))
1890                 return;
1891
1892         lock = ldlm_handle2lock(lockh);
1893         if (lock == NULL)
1894                 return;
1895
1896         ldlm_lock_dump(D_OTHER, lock, 0);
1897
1898         LDLM_LOCK_PUT(lock);
1899 }
1900
1901 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 level,
1902                       struct libcfs_debug_msg_data *data, const char *fmt,
1903                       ...)
1904 {
1905         va_list args;
1906         cfs_debug_limit_state_t *cdls = data->msg_cdls;
1907
1908         va_start(args, fmt);
1909
1910         if (lock->l_resource == NULL) {
1911                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file,
1912                                    data->msg_fn, data->msg_line, fmt, args,
1913                        " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1914                        "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" remote: "
1915                        LPX64" expref: %d pid: %u timeout: %lu\n", lock,
1916                        lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1917                        lock->l_readers, lock->l_writers,
1918                        ldlm_lockname[lock->l_granted_mode],
1919                        ldlm_lockname[lock->l_req_mode],
1920                        lock->l_flags, lock->l_remote_handle.cookie,
1921                        lock->l_export ?
1922                        atomic_read(&lock->l_export->exp_refcount) : -99,
1923                        lock->l_pid, lock->l_callback_timeout);
1924                 va_end(args);
1925                 return;
1926         }
1927
1928         switch (lock->l_resource->lr_type) {
1929         case LDLM_EXTENT:
1930                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file,
1931                                    data->msg_fn, data->msg_line, fmt, args,
1932                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1933                        "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64
1934                        "] (req "LPU64"->"LPU64") flags: "LPX64" remote: "LPX64
1935                        " expref: %d pid: %u timeout %lu\n",
1936                        lock->l_resource->lr_namespace->ns_name, lock,
1937                        lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1938                        lock->l_readers, lock->l_writers,
1939                        ldlm_lockname[lock->l_granted_mode],
1940                        ldlm_lockname[lock->l_req_mode],
1941                        lock->l_resource->lr_name.name[0],
1942                        lock->l_resource->lr_name.name[1],
1943                        atomic_read(&lock->l_resource->lr_refcount),
1944                        ldlm_typename[lock->l_resource->lr_type],
1945                        lock->l_policy_data.l_extent.start,
1946                        lock->l_policy_data.l_extent.end,
1947                        lock->l_req_extent.start, lock->l_req_extent.end,
1948                        lock->l_flags, lock->l_remote_handle.cookie,
1949                        lock->l_export ?
1950                        atomic_read(&lock->l_export->exp_refcount) : -99,
1951                        lock->l_pid, lock->l_callback_timeout);
1952                 break;
1953
1954         case LDLM_FLOCK:
1955                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file,
1956                                    data->msg_fn, data->msg_line, fmt, args,
1957                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1958                        "res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d "
1959                        "["LPU64"->"LPU64"] flags: "LPX64" remote: "LPX64
1960                        " expref: %d pid: %u timeout: %lu\n",
1961                        lock->l_resource->lr_namespace->ns_name, lock,
1962                        lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1963                        lock->l_readers, lock->l_writers,
1964                        ldlm_lockname[lock->l_granted_mode],
1965                        ldlm_lockname[lock->l_req_mode],
1966                        lock->l_resource->lr_name.name[0],
1967                        lock->l_resource->lr_name.name[1],
1968                        atomic_read(&lock->l_resource->lr_refcount),
1969                        ldlm_typename[lock->l_resource->lr_type],
1970                        lock->l_policy_data.l_flock.pid,
1971                        lock->l_policy_data.l_flock.start,
1972                        lock->l_policy_data.l_flock.end,
1973                        lock->l_flags, lock->l_remote_handle.cookie,
1974                        lock->l_export ?
1975                        atomic_read(&lock->l_export->exp_refcount) : -99,
1976                        lock->l_pid, lock->l_callback_timeout);
1977                 break;
1978
1979         case LDLM_IBITS:
1980                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file,
1981                                    data->msg_fn, data->msg_line, fmt, args,
1982                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1983                        "res: "LPU64"/"LPU64" bits "LPX64" rrc: %d type: %s "
1984                        "flags: "LPX64" remote: "LPX64" expref: %d "
1985                        "pid: %u timeout: %lu\n",
1986                        lock->l_resource->lr_namespace->ns_name,
1987                        lock, lock->l_handle.h_cookie,
1988                        atomic_read (&lock->l_refc),
1989                        lock->l_readers, lock->l_writers,
1990                        ldlm_lockname[lock->l_granted_mode],
1991                        ldlm_lockname[lock->l_req_mode],
1992                        lock->l_resource->lr_name.name[0],
1993                        lock->l_resource->lr_name.name[1],
1994                        lock->l_policy_data.l_inodebits.bits,
1995                        atomic_read(&lock->l_resource->lr_refcount),
1996                        ldlm_typename[lock->l_resource->lr_type],
1997                        lock->l_flags, lock->l_remote_handle.cookie,
1998                        lock->l_export ?
1999                        atomic_read(&lock->l_export->exp_refcount) : -99,
2000                        lock->l_pid, lock->l_callback_timeout);
2001                 break;
2002
2003         default:
2004                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file,
2005                                    data->msg_fn, data->msg_line, fmt, args,
2006                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2007                        "res: "LPU64"/"LPU64" rrc: %d type: %s flags: "LPX64" "
2008                        "remote: "LPX64" expref: %d pid: %u timeout %lu\n",
2009                        lock->l_resource->lr_namespace->ns_name,
2010                        lock, lock->l_handle.h_cookie,
2011                        atomic_read (&lock->l_refc),
2012                        lock->l_readers, lock->l_writers,
2013                        ldlm_lockname[lock->l_granted_mode],
2014                        ldlm_lockname[lock->l_req_mode],
2015                        lock->l_resource->lr_name.name[0],
2016                        lock->l_resource->lr_name.name[1],
2017                        atomic_read(&lock->l_resource->lr_refcount),
2018                        ldlm_typename[lock->l_resource->lr_type],
2019                        lock->l_flags, lock->l_remote_handle.cookie,
2020                        lock->l_export ?
2021                        atomic_read(&lock->l_export->exp_refcount) : -99,
2022                        lock->l_pid, lock->l_callback_timeout);
2023                 break;
2024         }
2025         va_end(args);
2026 }
2027 EXPORT_SYMBOL(_ldlm_lock_debug);