Whamcloud - gitweb
LU-1346 libcfs: replace libcfs wrappers with kernel API
[fs/lustre-release.git] / lustre / ldlm / ldlm_resource.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
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_resource.c
37  *
38  * Author: Phil Schwan <phil@clusterfs.com>
39  * Author: Peter Braam <braam@clusterfs.com>
40  */
41
42 #define DEBUG_SUBSYSTEM S_LDLM
43 #ifdef __KERNEL__
44 # include <lustre_dlm.h>
45 #else
46 # include <liblustre.h>
47 #endif
48
49 #include <lustre_fid.h>
50 #include <obd_class.h>
51 #include "ldlm_internal.h"
52
53 cfs_mem_cache_t *ldlm_resource_slab, *ldlm_lock_slab;
54
55 cfs_atomic_t ldlm_srv_namespace_nr = CFS_ATOMIC_INIT(0);
56 cfs_atomic_t ldlm_cli_namespace_nr = CFS_ATOMIC_INIT(0);
57
58 struct mutex ldlm_srv_namespace_lock;
59 CFS_LIST_HEAD(ldlm_srv_namespace_list);
60
61 struct mutex ldlm_cli_namespace_lock;
62 CFS_LIST_HEAD(ldlm_cli_namespace_list);
63
64 cfs_proc_dir_entry_t *ldlm_type_proc_dir = NULL;
65 cfs_proc_dir_entry_t *ldlm_ns_proc_dir = NULL;
66 cfs_proc_dir_entry_t *ldlm_svc_proc_dir = NULL;
67
68 extern unsigned int ldlm_cancel_unused_locks_before_replay;
69
70 /* during debug dump certain amount of granted locks for one resource to avoid
71  * DDOS. */
72 unsigned int ldlm_dump_granted_max = 256;
73
74 #ifdef LPROCFS
75 static int ldlm_proc_dump_ns(struct file *file, const char *buffer,
76                              unsigned long count, void *data)
77 {
78         ldlm_dump_all_namespaces(LDLM_NAMESPACE_SERVER, D_DLMTRACE);
79         ldlm_dump_all_namespaces(LDLM_NAMESPACE_CLIENT, D_DLMTRACE);
80         RETURN(count);
81 }
82
83 int ldlm_proc_setup(void)
84 {
85         int rc;
86         struct lprocfs_vars list[] = {
87                 { "dump_namespaces", NULL, ldlm_proc_dump_ns, NULL },
88                 { "dump_granted_max",
89                   lprocfs_rd_uint, lprocfs_wr_uint,
90                   &ldlm_dump_granted_max, NULL },
91                 { "cancel_unused_locks_before_replay",
92                   lprocfs_rd_uint, lprocfs_wr_uint,
93                   &ldlm_cancel_unused_locks_before_replay, NULL },
94                 { NULL }};
95         ENTRY;
96         LASSERT(ldlm_ns_proc_dir == NULL);
97
98         ldlm_type_proc_dir = lprocfs_register(OBD_LDLM_DEVICENAME,
99                                               proc_lustre_root,
100                                               NULL, NULL);
101         if (IS_ERR(ldlm_type_proc_dir)) {
102                 CERROR("LProcFS failed in ldlm-init\n");
103                 rc = PTR_ERR(ldlm_type_proc_dir);
104                 GOTO(err, rc);
105         }
106
107         ldlm_ns_proc_dir = lprocfs_register("namespaces",
108                                             ldlm_type_proc_dir,
109                                             NULL, NULL);
110         if (IS_ERR(ldlm_ns_proc_dir)) {
111                 CERROR("LProcFS failed in ldlm-init\n");
112                 rc = PTR_ERR(ldlm_ns_proc_dir);
113                 GOTO(err_type, rc);
114         }
115
116         ldlm_svc_proc_dir = lprocfs_register("services",
117                                             ldlm_type_proc_dir,
118                                             NULL, NULL);
119         if (IS_ERR(ldlm_svc_proc_dir)) {
120                 CERROR("LProcFS failed in ldlm-init\n");
121                 rc = PTR_ERR(ldlm_svc_proc_dir);
122                 GOTO(err_ns, rc);
123         }
124
125         rc = lprocfs_add_vars(ldlm_type_proc_dir, list, NULL);
126
127         RETURN(0);
128
129 err_ns:
130         lprocfs_remove(&ldlm_ns_proc_dir);
131 err_type:
132         lprocfs_remove(&ldlm_type_proc_dir);
133 err:
134         ldlm_svc_proc_dir = NULL;
135         RETURN(rc);
136 }
137
138 void ldlm_proc_cleanup(void)
139 {
140         if (ldlm_svc_proc_dir)
141                 lprocfs_remove(&ldlm_svc_proc_dir);
142
143         if (ldlm_ns_proc_dir)
144                 lprocfs_remove(&ldlm_ns_proc_dir);
145
146         if (ldlm_type_proc_dir)
147                 lprocfs_remove(&ldlm_type_proc_dir);
148 }
149
150 static int lprocfs_rd_ns_resources(char *page, char **start, off_t off,
151                                    int count, int *eof, void *data)
152 {
153         struct ldlm_namespace *ns  = data;
154         __u64                  res = 0;
155         cfs_hash_bd_t          bd;
156         int                    i;
157
158         /* result is not strictly consistant */
159         cfs_hash_for_each_bucket(ns->ns_rs_hash, &bd, i)
160                 res += cfs_hash_bd_count_get(&bd);
161         return lprocfs_rd_u64(page, start, off, count, eof, &res);
162 }
163
164 static int lprocfs_rd_ns_locks(char *page, char **start, off_t off,
165                                int count, int *eof, void *data)
166 {
167         struct ldlm_namespace *ns = data;
168         __u64                  locks;
169
170         locks = lprocfs_stats_collector(ns->ns_stats, LDLM_NSS_LOCKS,
171                                         LPROCFS_FIELDS_FLAGS_SUM);
172         return lprocfs_rd_u64(page, start, off, count, eof, &locks);
173 }
174
175 static int lprocfs_rd_lru_size(char *page, char **start, off_t off,
176                                int count, int *eof, void *data)
177 {
178         struct ldlm_namespace *ns = data;
179         __u32 *nr = &ns->ns_max_unused;
180
181         if (ns_connect_lru_resize(ns))
182                 nr = &ns->ns_nr_unused;
183         return lprocfs_rd_uint(page, start, off, count, eof, nr);
184 }
185
186 static int lprocfs_wr_lru_size(struct file *file, const char *buffer,
187                                unsigned long count, void *data)
188 {
189         struct ldlm_namespace *ns = data;
190         char dummy[MAX_STRING_SIZE + 1], *end;
191         unsigned long tmp;
192         int lru_resize;
193
194         dummy[MAX_STRING_SIZE] = '\0';
195         if (cfs_copy_from_user(dummy, buffer, MAX_STRING_SIZE))
196                 return -EFAULT;
197
198         if (strncmp(dummy, "clear", 5) == 0) {
199                 CDEBUG(D_DLMTRACE,
200                        "dropping all unused locks from namespace %s\n",
201                        ldlm_ns_name(ns));
202                 if (ns_connect_lru_resize(ns)) {
203                         int canceled, unused  = ns->ns_nr_unused;
204
205                         /* Try to cancel all @ns_nr_unused locks. */
206                         canceled = ldlm_cancel_lru(ns, unused, LDLM_SYNC,
207                                                    LDLM_CANCEL_PASSED);
208                         if (canceled < unused) {
209                                 CDEBUG(D_DLMTRACE,
210                                        "not all requested locks are canceled, "
211                                        "requested: %d, canceled: %d\n", unused,
212                                        canceled);
213                                 return -EINVAL;
214                         }
215                 } else {
216                         tmp = ns->ns_max_unused;
217                         ns->ns_max_unused = 0;
218                         ldlm_cancel_lru(ns, 0, LDLM_SYNC, LDLM_CANCEL_PASSED);
219                         ns->ns_max_unused = tmp;
220                 }
221                 return count;
222         }
223
224         tmp = simple_strtoul(dummy, &end, 0);
225         if (dummy == end) {
226                 CERROR("invalid value written\n");
227                 return -EINVAL;
228         }
229         lru_resize = (tmp == 0);
230
231         if (ns_connect_lru_resize(ns)) {
232                 if (!lru_resize)
233                         ns->ns_max_unused = (unsigned int)tmp;
234
235                 if (tmp > ns->ns_nr_unused)
236                         tmp = ns->ns_nr_unused;
237                 tmp = ns->ns_nr_unused - tmp;
238
239                 CDEBUG(D_DLMTRACE,
240                        "changing namespace %s unused locks from %u to %u\n",
241                        ldlm_ns_name(ns), ns->ns_nr_unused,
242                        (unsigned int)tmp);
243                 ldlm_cancel_lru(ns, tmp, LDLM_ASYNC, LDLM_CANCEL_PASSED);
244
245                 if (!lru_resize) {
246                         CDEBUG(D_DLMTRACE,
247                                "disable lru_resize for namespace %s\n",
248                                ldlm_ns_name(ns));
249                         ns->ns_connect_flags &= ~OBD_CONNECT_LRU_RESIZE;
250                 }
251         } else {
252                 CDEBUG(D_DLMTRACE,
253                        "changing namespace %s max_unused from %u to %u\n",
254                        ldlm_ns_name(ns), ns->ns_max_unused,
255                        (unsigned int)tmp);
256                 ns->ns_max_unused = (unsigned int)tmp;
257                 ldlm_cancel_lru(ns, 0, LDLM_ASYNC, LDLM_CANCEL_PASSED);
258
259                 /* Make sure that originally lru resize was supported before
260                  * turning it on here. */
261                 if (lru_resize &&
262                     (ns->ns_orig_connect_flags & OBD_CONNECT_LRU_RESIZE)) {
263                         CDEBUG(D_DLMTRACE,
264                                "enable lru_resize for namespace %s\n",
265                                ldlm_ns_name(ns));
266                         ns->ns_connect_flags |= OBD_CONNECT_LRU_RESIZE;
267                 }
268         }
269
270         return count;
271 }
272
273 static int lprocfs_rd_elc(char *page, char **start, off_t off,
274                           int count, int *eof, void *data)
275 {
276         struct ldlm_namespace *ns = data;
277         unsigned int supp = ns_connect_cancelset(ns);
278
279         return lprocfs_rd_uint(page, start, off, count, eof, &supp);
280 }
281
282 static int lprocfs_wr_elc(struct file *file, const char *buffer,
283                                unsigned long count, void *data)
284 {
285         struct ldlm_namespace *ns = data;
286         unsigned int supp = -1;
287         int rc;
288
289         rc = lprocfs_wr_uint(file, buffer, count, &supp);
290         if (rc < 0)
291                 return rc;
292
293         if (supp == 0)
294                 ns->ns_connect_flags &= ~OBD_CONNECT_CANCELSET;
295         else if (ns->ns_orig_connect_flags & OBD_CONNECT_CANCELSET)
296                 ns->ns_connect_flags |= OBD_CONNECT_CANCELSET;
297         return count;
298 }
299
300 void ldlm_namespace_proc_unregister(struct ldlm_namespace *ns)
301 {
302         struct proc_dir_entry *dir;
303
304         dir = lprocfs_srch(ldlm_ns_proc_dir, ldlm_ns_name(ns));
305         if (dir == NULL) {
306                 CERROR("dlm namespace %s has no procfs dir?\n",
307                        ldlm_ns_name(ns));
308         } else {
309                 lprocfs_remove(&dir);
310         }
311
312         if (ns->ns_stats != NULL)
313                 lprocfs_free_stats(&ns->ns_stats);
314 }
315
316 int ldlm_namespace_proc_register(struct ldlm_namespace *ns)
317 {
318         struct lprocfs_vars lock_vars[2];
319         char lock_name[MAX_STRING_SIZE + 1];
320
321         LASSERT(ns != NULL);
322         LASSERT(ns->ns_rs_hash != NULL);
323
324         ns->ns_stats = lprocfs_alloc_stats(LDLM_NSS_LAST, 0);
325         if (ns->ns_stats == NULL)
326                 return -ENOMEM;
327
328         lprocfs_counter_init(ns->ns_stats, LDLM_NSS_LOCKS,
329                              LPROCFS_CNTR_AVGMINMAX, "locks", "locks");
330
331         lock_name[MAX_STRING_SIZE] = '\0';
332
333         memset(lock_vars, 0, sizeof(lock_vars));
334         lock_vars[0].name = lock_name;
335
336         snprintf(lock_name, MAX_STRING_SIZE, "%s/resource_count",
337                  ldlm_ns_name(ns));
338         lock_vars[0].data = ns;
339         lock_vars[0].read_fptr = lprocfs_rd_ns_resources;
340         lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
341
342         snprintf(lock_name, MAX_STRING_SIZE, "%s/lock_count",
343                  ldlm_ns_name(ns));
344         lock_vars[0].data = ns;
345         lock_vars[0].read_fptr = lprocfs_rd_ns_locks;
346         lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
347
348         if (ns_is_client(ns)) {
349                 snprintf(lock_name, MAX_STRING_SIZE, "%s/lock_unused_count",
350                          ldlm_ns_name(ns));
351                 lock_vars[0].data = &ns->ns_nr_unused;
352                 lock_vars[0].read_fptr = lprocfs_rd_uint;
353                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
354
355                 snprintf(lock_name, MAX_STRING_SIZE, "%s/lru_size",
356                          ldlm_ns_name(ns));
357                 lock_vars[0].data = ns;
358                 lock_vars[0].read_fptr = lprocfs_rd_lru_size;
359                 lock_vars[0].write_fptr = lprocfs_wr_lru_size;
360                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
361
362                 snprintf(lock_name, MAX_STRING_SIZE, "%s/lru_max_age",
363                          ldlm_ns_name(ns));
364                 lock_vars[0].data = &ns->ns_max_age;
365                 lock_vars[0].read_fptr = lprocfs_rd_uint;
366                 lock_vars[0].write_fptr = lprocfs_wr_uint;
367                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
368
369                 snprintf(lock_name, MAX_STRING_SIZE, "%s/early_lock_cancel",
370                          ldlm_ns_name(ns));
371                 lock_vars[0].data = ns;
372                 lock_vars[0].read_fptr = lprocfs_rd_elc;
373                 lock_vars[0].write_fptr = lprocfs_wr_elc;
374                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
375         } else {
376                 snprintf(lock_name, MAX_STRING_SIZE, "%s/ctime_age_limit",
377                          ldlm_ns_name(ns));
378                 lock_vars[0].data = &ns->ns_ctime_age_limit;
379                 lock_vars[0].read_fptr = lprocfs_rd_uint;
380                 lock_vars[0].write_fptr = lprocfs_wr_uint;
381                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
382
383                 snprintf(lock_name, MAX_STRING_SIZE, "%s/lock_timeouts",
384                          ldlm_ns_name(ns));
385                 lock_vars[0].data = &ns->ns_timeouts;
386                 lock_vars[0].read_fptr = lprocfs_rd_uint;
387                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
388
389                 snprintf(lock_name, MAX_STRING_SIZE, "%s/max_nolock_bytes",
390                          ldlm_ns_name(ns));
391                 lock_vars[0].data = &ns->ns_max_nolock_size;
392                 lock_vars[0].read_fptr = lprocfs_rd_uint;
393                 lock_vars[0].write_fptr = lprocfs_wr_uint;
394                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
395
396                 snprintf(lock_name, MAX_STRING_SIZE, "%s/contention_seconds",
397                          ldlm_ns_name(ns));
398                 lock_vars[0].data = &ns->ns_contention_time;
399                 lock_vars[0].read_fptr = lprocfs_rd_uint;
400                 lock_vars[0].write_fptr = lprocfs_wr_uint;
401                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
402
403                 snprintf(lock_name, MAX_STRING_SIZE, "%s/contended_locks",
404                          ldlm_ns_name(ns));
405                 lock_vars[0].data = &ns->ns_contended_locks;
406                 lock_vars[0].read_fptr = lprocfs_rd_uint;
407                 lock_vars[0].write_fptr = lprocfs_wr_uint;
408                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
409
410                 snprintf(lock_name, MAX_STRING_SIZE, "%s/max_parallel_ast",
411                          ldlm_ns_name(ns));
412                 lock_vars[0].data = &ns->ns_max_parallel_ast;
413                 lock_vars[0].read_fptr = lprocfs_rd_uint;
414                 lock_vars[0].write_fptr = lprocfs_wr_uint;
415                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
416         }
417         return 0;
418 }
419 #undef MAX_STRING_SIZE
420 #else /* LPROCFS */
421
422 #define ldlm_namespace_proc_unregister(ns)      ({;})
423 #define ldlm_namespace_proc_register(ns)        ({0;})
424
425 #endif /* LPROCFS */
426
427 static unsigned ldlm_res_hop_hash(cfs_hash_t *hs,
428                                   const void *key, unsigned mask)
429 {
430         const struct ldlm_res_id     *id  = key;
431         unsigned                val = 0;
432         unsigned                i;
433
434         for (i = 0; i < RES_NAME_SIZE; i++)
435                 val += id->name[i];
436         return val & mask;
437 }
438
439 static unsigned ldlm_res_hop_fid_hash(cfs_hash_t *hs,
440                                       const void *key, unsigned mask)
441 {
442         const struct ldlm_res_id *id = key;
443         struct lu_fid       fid;
444         __u32               hash;
445         __u32               val;
446
447         fid.f_seq = id->name[LUSTRE_RES_ID_SEQ_OFF];
448         fid.f_oid = (__u32)id->name[LUSTRE_RES_ID_VER_OID_OFF];
449         fid.f_ver = (__u32)(id->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32);
450
451         hash = fid_flatten32(&fid);
452         hash += (hash >> 4) + (hash << 12); /* mixing oid and seq */
453         if (id->name[LUSTRE_RES_ID_HSH_OFF] != 0) {
454                 val = id->name[LUSTRE_RES_ID_HSH_OFF];
455                 hash += (val >> 5) + (val << 11);
456         } else {
457                 val = fid_oid(&fid);
458         }
459         hash = cfs_hash_long(hash, hs->hs_bkt_bits);
460         /* give me another random factor */
461         hash -= cfs_hash_long((unsigned long)hs, val % 11 + 3);
462
463         hash <<= hs->hs_cur_bits - hs->hs_bkt_bits;
464         hash |= ldlm_res_hop_hash(hs, key, CFS_HASH_NBKT(hs) - 1);
465
466         return hash & mask;
467 }
468
469 static void *ldlm_res_hop_key(cfs_hlist_node_t *hnode)
470 {
471         struct ldlm_resource   *res;
472
473         res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
474         return &res->lr_name;
475 }
476
477 static int ldlm_res_hop_keycmp(const void *key, cfs_hlist_node_t *hnode)
478 {
479         struct ldlm_resource   *res;
480
481         res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
482         return ldlm_res_eq((const struct ldlm_res_id *)key,
483                            (const struct ldlm_res_id *)&res->lr_name);
484 }
485
486 static void *ldlm_res_hop_object(cfs_hlist_node_t *hnode)
487 {
488         return cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
489 }
490
491 static void ldlm_res_hop_get_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
492 {
493         struct ldlm_resource *res;
494
495         res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
496         ldlm_resource_getref(res);
497 }
498
499 static void ldlm_res_hop_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
500 {
501         struct ldlm_resource *res;
502
503         res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
504         /* cfs_hash_for_each_nolock is the only chance we call it */
505         ldlm_resource_putref_locked(res);
506 }
507
508 static void ldlm_res_hop_put(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
509 {
510         struct ldlm_resource *res;
511
512         res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
513         ldlm_resource_putref(res);
514 }
515
516 cfs_hash_ops_t ldlm_ns_hash_ops = {
517         .hs_hash        = ldlm_res_hop_hash,
518         .hs_key         = ldlm_res_hop_key,
519         .hs_keycmp      = ldlm_res_hop_keycmp,
520         .hs_keycpy      = NULL,
521         .hs_object      = ldlm_res_hop_object,
522         .hs_get         = ldlm_res_hop_get_locked,
523         .hs_put_locked  = ldlm_res_hop_put_locked,
524         .hs_put         = ldlm_res_hop_put
525 };
526
527 cfs_hash_ops_t ldlm_ns_fid_hash_ops = {
528         .hs_hash        = ldlm_res_hop_fid_hash,
529         .hs_key         = ldlm_res_hop_key,
530         .hs_keycmp      = ldlm_res_hop_keycmp,
531         .hs_keycpy      = NULL,
532         .hs_object      = ldlm_res_hop_object,
533         .hs_get         = ldlm_res_hop_get_locked,
534         .hs_put_locked  = ldlm_res_hop_put_locked,
535         .hs_put         = ldlm_res_hop_put
536 };
537
538 typedef struct {
539         ldlm_ns_type_t  nsd_type;
540         /** hash bucket bits */
541         unsigned        nsd_bkt_bits;
542         /** hash bits */
543         unsigned        nsd_all_bits;
544         /** hash operations */
545         cfs_hash_ops_t *nsd_hops;
546 } ldlm_ns_hash_def_t;
547
548 ldlm_ns_hash_def_t ldlm_ns_hash_defs[] =
549 {
550         {
551                 .nsd_type       = LDLM_NS_TYPE_MDC,
552                 .nsd_bkt_bits   = 11,
553                 .nsd_all_bits   = 16,
554                 .nsd_hops       = &ldlm_ns_fid_hash_ops,
555         },
556         {
557                 .nsd_type       = LDLM_NS_TYPE_MDT,
558                 .nsd_bkt_bits   = 14,
559                 .nsd_all_bits   = 21,
560                 .nsd_hops       = &ldlm_ns_fid_hash_ops,
561         },
562         {
563                 .nsd_type       = LDLM_NS_TYPE_OSC,
564                 .nsd_bkt_bits   = 8,
565                 .nsd_all_bits   = 12,
566                 .nsd_hops       = &ldlm_ns_hash_ops,
567         },
568         {
569                 .nsd_type       = LDLM_NS_TYPE_OST,
570                 .nsd_bkt_bits   = 11,
571                 .nsd_all_bits   = 17,
572                 .nsd_hops       = &ldlm_ns_hash_ops,
573         },
574         {
575                 .nsd_type       = LDLM_NS_TYPE_MGC,
576                 .nsd_bkt_bits   = 4,
577                 .nsd_all_bits   = 4,
578                 .nsd_hops       = &ldlm_ns_hash_ops,
579         },
580         {
581                 .nsd_type       = LDLM_NS_TYPE_MGT,
582                 .nsd_bkt_bits   = 4,
583                 .nsd_all_bits   = 4,
584                 .nsd_hops       = &ldlm_ns_hash_ops,
585         },
586         {
587                 .nsd_type       = LDLM_NS_TYPE_UNKNOWN,
588         },
589 };
590
591 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
592                                           ldlm_side_t client,
593                                           ldlm_appetite_t apt,
594                                           ldlm_ns_type_t ns_type)
595 {
596         struct ldlm_namespace *ns = NULL;
597         struct ldlm_ns_bucket *nsb;
598         ldlm_ns_hash_def_t    *nsd;
599         cfs_hash_bd_t          bd;
600         int                    idx;
601         int                    rc;
602         ENTRY;
603
604         LASSERT(obd != NULL);
605
606         rc = ldlm_get_ref();
607         if (rc) {
608                 CERROR("ldlm_get_ref failed: %d\n", rc);
609                 RETURN(NULL);
610         }
611
612         for (idx = 0;;idx++) {
613                 nsd = &ldlm_ns_hash_defs[idx];
614                 if (nsd->nsd_type == LDLM_NS_TYPE_UNKNOWN) {
615                         CERROR("Unknown type %d for ns %s\n", ns_type, name);
616                         GOTO(out_ref, NULL);
617                 }
618
619                 if (nsd->nsd_type == ns_type)
620                         break;
621         }
622
623         OBD_ALLOC_PTR(ns);
624         if (!ns)
625                 GOTO(out_ref, NULL);
626
627         ns->ns_rs_hash = cfs_hash_create(name,
628                                          nsd->nsd_all_bits, nsd->nsd_all_bits,
629                                          nsd->nsd_bkt_bits, sizeof(*nsb),
630                                          CFS_HASH_MIN_THETA,
631                                          CFS_HASH_MAX_THETA,
632                                          nsd->nsd_hops,
633                                          CFS_HASH_DEPTH |
634                                          CFS_HASH_BIGNAME |
635                                          CFS_HASH_SPIN_BKTLOCK |
636                                          CFS_HASH_NO_ITEMREF);
637         if (ns->ns_rs_hash == NULL)
638                 GOTO(out_ns, NULL);
639
640         cfs_hash_for_each_bucket(ns->ns_rs_hash, &bd, idx) {
641                 nsb = cfs_hash_bd_extra_get(ns->ns_rs_hash, &bd);
642                 at_init(&nsb->nsb_at_estimate, ldlm_enqueue_min, 0);
643                 nsb->nsb_namespace = ns;
644         }
645
646         ns->ns_obd      = obd;
647         ns->ns_appetite = apt;
648         ns->ns_client   = client;
649
650         CFS_INIT_LIST_HEAD(&ns->ns_list_chain);
651         CFS_INIT_LIST_HEAD(&ns->ns_unused_list);
652         spin_lock_init(&ns->ns_lock);
653         cfs_atomic_set(&ns->ns_bref, 0);
654         cfs_waitq_init(&ns->ns_waitq);
655
656         ns->ns_max_nolock_size    = NS_DEFAULT_MAX_NOLOCK_BYTES;
657         ns->ns_contention_time    = NS_DEFAULT_CONTENTION_SECONDS;
658         ns->ns_contended_locks    = NS_DEFAULT_CONTENDED_LOCKS;
659
660         ns->ns_max_parallel_ast   = LDLM_DEFAULT_PARALLEL_AST_LIMIT;
661         ns->ns_nr_unused          = 0;
662         ns->ns_max_unused         = LDLM_DEFAULT_LRU_SIZE;
663         ns->ns_max_age            = LDLM_DEFAULT_MAX_ALIVE;
664         ns->ns_ctime_age_limit    = LDLM_CTIME_AGE_LIMIT;
665         ns->ns_timeouts           = 0;
666         ns->ns_orig_connect_flags = 0;
667         ns->ns_connect_flags      = 0;
668         ns->ns_stopping           = 0;
669         rc = ldlm_namespace_proc_register(ns);
670         if (rc != 0) {
671                 CERROR("Can't initialize ns proc, rc %d\n", rc);
672                 GOTO(out_hash, rc);
673         }
674
675         idx = cfs_atomic_read(ldlm_namespace_nr(client));
676         rc = ldlm_pool_init(&ns->ns_pool, ns, idx, client);
677         if (rc) {
678                 CERROR("Can't initialize lock pool, rc %d\n", rc);
679                 GOTO(out_proc, rc);
680         }
681
682         ldlm_namespace_register(ns, client);
683         RETURN(ns);
684 out_proc:
685         ldlm_namespace_proc_unregister(ns);
686         ldlm_namespace_cleanup(ns, 0);
687 out_hash:
688         cfs_hash_putref(ns->ns_rs_hash);
689 out_ns:
690         OBD_FREE_PTR(ns);
691 out_ref:
692         ldlm_put_ref();
693         RETURN(NULL);
694 }
695 EXPORT_SYMBOL(ldlm_namespace_new);
696
697 extern struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
698
699 /* If flags contains FL_LOCAL_ONLY, don't try to tell the server, just cleanup.
700  * This is currently only used for recovery, and we make certain assumptions
701  * as a result--notably, that we shouldn't cancel locks with refs. -phil */
702 static void cleanup_resource(struct ldlm_resource *res, cfs_list_t *q,
703                              __u64 flags)
704 {
705         cfs_list_t *tmp;
706         int rc = 0, client = ns_is_client(ldlm_res_to_ns(res));
707         bool local_only = !!(flags & LDLM_FL_LOCAL_ONLY);
708
709         do {
710                 struct ldlm_lock *lock = NULL;
711
712                 /* first, we look for non-cleaned-yet lock
713                  * all cleaned locks are marked by CLEANED flag */
714                 lock_res(res);
715                 cfs_list_for_each(tmp, q) {
716                         lock = cfs_list_entry(tmp, struct ldlm_lock,
717                                               l_res_link);
718                         if (lock->l_flags & LDLM_FL_CLEANED) {
719                                 lock = NULL;
720                                 continue;
721                         }
722                         LDLM_LOCK_GET(lock);
723                         lock->l_flags |= LDLM_FL_CLEANED;
724                         break;
725                 }
726
727                 if (lock == NULL) {
728                         unlock_res(res);
729                         break;
730                 }
731
732                 /* Set CBPENDING so nothing in the cancellation path
733                  * can match this lock */
734                 lock->l_flags |= LDLM_FL_CBPENDING;
735                 lock->l_flags |= LDLM_FL_FAILED;
736                 lock->l_flags |= flags;
737
738                 /* ... without sending a CANCEL message for local_only. */
739                 if (local_only)
740                         lock->l_flags |= LDLM_FL_LOCAL_ONLY;
741
742                 if (local_only && (lock->l_readers || lock->l_writers)) {
743                         /* This is a little bit gross, but much better than the
744                          * alternative: pretend that we got a blocking AST from
745                          * the server, so that when the lock is decref'd, it
746                          * will go away ... */
747                         unlock_res(res);
748                         LDLM_DEBUG(lock, "setting FL_LOCAL_ONLY");
749                         if (lock->l_completion_ast)
750                                 lock->l_completion_ast(lock, 0, NULL);
751                         LDLM_LOCK_RELEASE(lock);
752                         continue;
753                 }
754
755                 if (client) {
756                         struct lustre_handle lockh;
757
758                         unlock_res(res);
759                         ldlm_lock2handle(lock, &lockh);
760                         rc = ldlm_cli_cancel(&lockh);
761                         if (rc)
762                                 CERROR("ldlm_cli_cancel: %d\n", rc);
763                 } else {
764                         ldlm_resource_unlink_lock(lock);
765                         unlock_res(res);
766                         LDLM_DEBUG(lock, "Freeing a lock still held by a "
767                                    "client node");
768                         ldlm_lock_destroy(lock);
769                 }
770                 LDLM_LOCK_RELEASE(lock);
771         } while (1);
772 }
773
774 static int ldlm_resource_clean(cfs_hash_t *hs, cfs_hash_bd_t *bd,
775                                cfs_hlist_node_t *hnode, void *arg)
776 {
777         struct ldlm_resource *res = cfs_hash_object(hs, hnode);
778         __u64 flags = *(__u64 *)arg;
779
780         cleanup_resource(res, &res->lr_granted, flags);
781         cleanup_resource(res, &res->lr_converting, flags);
782         cleanup_resource(res, &res->lr_waiting, flags);
783
784         return 0;
785 }
786
787 static int ldlm_resource_complain(cfs_hash_t *hs, cfs_hash_bd_t *bd,
788                                   cfs_hlist_node_t *hnode, void *arg)
789 {
790         struct ldlm_resource  *res = cfs_hash_object(hs, hnode);
791
792         CERROR("Namespace %s resource refcount nonzero "
793                "(%d) after lock cleanup; forcing "
794                "cleanup.\n",
795                ldlm_ns_name(ldlm_res_to_ns(res)),
796                cfs_atomic_read(&res->lr_refcount) - 1);
797
798         CERROR("Resource: %p ("LPU64"/"LPU64"/"LPU64"/"
799                LPU64") (rc: %d)\n", res,
800                res->lr_name.name[0], res->lr_name.name[1],
801                res->lr_name.name[2], res->lr_name.name[3],
802                cfs_atomic_read(&res->lr_refcount) - 1);
803         return 0;
804 }
805
806 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags)
807 {
808         if (ns == NULL) {
809                 CDEBUG(D_INFO, "NULL ns, skipping cleanup\n");
810                 return ELDLM_OK;
811         }
812
813         cfs_hash_for_each_nolock(ns->ns_rs_hash, ldlm_resource_clean, &flags);
814         cfs_hash_for_each_nolock(ns->ns_rs_hash, ldlm_resource_complain, NULL);
815         return ELDLM_OK;
816 }
817 EXPORT_SYMBOL(ldlm_namespace_cleanup);
818
819 static int __ldlm_namespace_free(struct ldlm_namespace *ns, int force)
820 {
821         ENTRY;
822
823         /* At shutdown time, don't call the cancellation callback */
824         ldlm_namespace_cleanup(ns, force ? LDLM_FL_LOCAL_ONLY : 0);
825
826         if (cfs_atomic_read(&ns->ns_bref) > 0) {
827                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
828                 int rc;
829                 CDEBUG(D_DLMTRACE,
830                        "dlm namespace %s free waiting on refcount %d\n",
831                        ldlm_ns_name(ns), cfs_atomic_read(&ns->ns_bref));
832 force_wait:
833                 if (force)
834                         lwi = LWI_TIMEOUT(obd_timeout * CFS_HZ / 4, NULL, NULL);
835
836                 rc = l_wait_event(ns->ns_waitq,
837                                   cfs_atomic_read(&ns->ns_bref) == 0, &lwi);
838
839                 /* Forced cleanups should be able to reclaim all references,
840                  * so it's safe to wait forever... we can't leak locks... */
841                 if (force && rc == -ETIMEDOUT) {
842                         LCONSOLE_ERROR("Forced cleanup waiting for %s "
843                                        "namespace with %d resources in use, "
844                                        "(rc=%d)\n", ldlm_ns_name(ns),
845                                        cfs_atomic_read(&ns->ns_bref), rc);
846                         GOTO(force_wait, rc);
847                 }
848
849                 if (cfs_atomic_read(&ns->ns_bref)) {
850                         LCONSOLE_ERROR("Cleanup waiting for %s namespace "
851                                        "with %d resources in use, (rc=%d)\n",
852                                        ldlm_ns_name(ns),
853                                        cfs_atomic_read(&ns->ns_bref), rc);
854                         RETURN(ELDLM_NAMESPACE_EXISTS);
855                 }
856                 CDEBUG(D_DLMTRACE, "dlm namespace %s free done waiting\n",
857                        ldlm_ns_name(ns));
858         }
859
860         RETURN(ELDLM_OK);
861 }
862
863 /**
864  * Performs various cleanups for passed \a ns to make it drop refc and be ready
865  * for freeing. Waits for refc == 0.
866  *
867  * The following is done:
868  * (0) Unregister \a ns from its list to make inaccessible for potential users
869  * like pools thread and others;
870  * (1) Clear all locks in \a ns.
871  */
872 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
873                                struct obd_import *imp,
874                                int force)
875 {
876         int rc;
877         ENTRY;
878         if (!ns) {
879                 EXIT;
880                 return;
881         }
882
883         spin_lock(&ns->ns_lock);
884         ns->ns_stopping = 1;
885         spin_unlock(&ns->ns_lock);
886
887         /*
888          * Can fail with -EINTR when force == 0 in which case try harder.
889          */
890         rc = __ldlm_namespace_free(ns, force);
891         if (rc != ELDLM_OK) {
892                 if (imp) {
893                         ptlrpc_disconnect_import(imp, 0);
894                         ptlrpc_invalidate_import(imp);
895                 }
896
897                 /*
898                  * With all requests dropped and the import inactive
899                  * we are gaurenteed all reference will be dropped.
900                  */
901                 rc = __ldlm_namespace_free(ns, 1);
902                 LASSERT(rc == 0);
903         }
904         EXIT;
905 }
906
907 /**
908  * Performs freeing memory structures related to \a ns. This is only done when
909  * ldlm_namespce_free_prior() successfully removed all resources referencing
910  * \a ns and its refc == 0.
911  */
912 void ldlm_namespace_free_post(struct ldlm_namespace *ns)
913 {
914         ENTRY;
915         if (!ns) {
916                 EXIT;
917                 return;
918         }
919
920
921         /*
922          * Make sure that nobody can find this ns in its list.
923          */
924         ldlm_namespace_unregister(ns, ns->ns_client);
925         /*
926          * Fini pool _before_ parent proc dir is removed. This is important as
927          * ldlm_pool_fini() removes own proc dir which is child to @dir. Removing
928          * it after @dir may cause oops.
929          */
930         ldlm_pool_fini(&ns->ns_pool);
931
932         ldlm_namespace_proc_unregister(ns);
933         cfs_hash_putref(ns->ns_rs_hash);
934         /*
935          * Namespace \a ns should be not on list in this time, otherwise this
936          * will cause issues realted to using freed \a ns in pools thread.
937          */
938         LASSERT(cfs_list_empty(&ns->ns_list_chain));
939         OBD_FREE_PTR(ns);
940         ldlm_put_ref();
941         EXIT;
942 }
943
944
945 /* Cleanup the resource, and free namespace.
946  * bug 12864:
947  * Deadlock issue:
948  * proc1: destroy import
949  *        class_disconnect_export(grab cl_sem) ->
950  *              -> ldlm_namespace_free ->
951  *              -> lprocfs_remove(grab _lprocfs_lock).
952  * proc2: read proc info
953  *        lprocfs_fops_read(grab _lprocfs_lock) ->
954  *              -> osc_rd_active, etc(grab cl_sem).
955  *
956  * So that I have to split the ldlm_namespace_free into two parts - the first
957  * part ldlm_namespace_free_prior is used to cleanup the resource which is
958  * being used; the 2nd part ldlm_namespace_free_post is used to unregister the
959  * lprocfs entries, and then free memory. It will be called w/o cli->cl_sem
960  * held.
961  */
962 void ldlm_namespace_free(struct ldlm_namespace *ns,
963                          struct obd_import *imp,
964                          int force)
965 {
966         ldlm_namespace_free_prior(ns, imp, force);
967         ldlm_namespace_free_post(ns);
968 }
969 EXPORT_SYMBOL(ldlm_namespace_free);
970
971 void ldlm_namespace_get(struct ldlm_namespace *ns)
972 {
973         cfs_atomic_inc(&ns->ns_bref);
974 }
975 EXPORT_SYMBOL(ldlm_namespace_get);
976
977 void ldlm_namespace_put(struct ldlm_namespace *ns)
978 {
979         if (cfs_atomic_dec_and_lock(&ns->ns_bref, &ns->ns_lock)) {
980                 cfs_waitq_signal(&ns->ns_waitq);
981                 spin_unlock(&ns->ns_lock);
982         }
983 }
984 EXPORT_SYMBOL(ldlm_namespace_put);
985
986 /* Register @ns in the list of namespaces */
987 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client)
988 {
989         mutex_lock(ldlm_namespace_lock(client));
990         LASSERT(cfs_list_empty(&ns->ns_list_chain));
991         cfs_list_add(&ns->ns_list_chain, ldlm_namespace_list(client));
992         cfs_atomic_inc(ldlm_namespace_nr(client));
993         mutex_unlock(ldlm_namespace_lock(client));
994 }
995
996 /* Unregister @ns from the list of namespaces */
997 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client)
998 {
999         mutex_lock(ldlm_namespace_lock(client));
1000         LASSERT(!cfs_list_empty(&ns->ns_list_chain));
1001         /*
1002          * Some asserts and possibly other parts of code still using
1003          * list_empty(&ns->ns_list_chain). This is why it is important
1004          * to use list_del_init() here.
1005          */
1006         cfs_list_del_init(&ns->ns_list_chain);
1007         cfs_atomic_dec(ldlm_namespace_nr(client));
1008         mutex_unlock(ldlm_namespace_lock(client));
1009 }
1010
1011 /* Should be called under ldlm_namespace_lock(client) taken */
1012 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client)
1013 {
1014         LASSERT(!cfs_list_empty(&ns->ns_list_chain));
1015         LASSERT_MUTEX_LOCKED(ldlm_namespace_lock(client));
1016         cfs_list_move_tail(&ns->ns_list_chain, ldlm_namespace_list(client));
1017 }
1018
1019 /* Should be called under ldlm_namespace_lock(client) taken */
1020 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client)
1021 {
1022         LASSERT_MUTEX_LOCKED(ldlm_namespace_lock(client));
1023         LASSERT(!cfs_list_empty(ldlm_namespace_list(client)));
1024         return container_of(ldlm_namespace_list(client)->next,
1025                 struct ldlm_namespace, ns_list_chain);
1026 }
1027
1028 static struct ldlm_resource *ldlm_resource_new(void)
1029 {
1030         struct ldlm_resource *res;
1031         int idx;
1032
1033         OBD_SLAB_ALLOC_PTR_GFP(res, ldlm_resource_slab, CFS_ALLOC_IO);
1034         if (res == NULL)
1035                 return NULL;
1036
1037         CFS_INIT_LIST_HEAD(&res->lr_granted);
1038         CFS_INIT_LIST_HEAD(&res->lr_converting);
1039         CFS_INIT_LIST_HEAD(&res->lr_waiting);
1040
1041         /* initialize interval trees for each lock mode*/
1042         for (idx = 0; idx < LCK_MODE_NUM; idx++) {
1043                 res->lr_itree[idx].lit_size = 0;
1044                 res->lr_itree[idx].lit_mode = 1 << idx;
1045                 res->lr_itree[idx].lit_root = NULL;
1046         }
1047
1048         cfs_atomic_set(&res->lr_refcount, 1);
1049         spin_lock_init(&res->lr_lock);
1050         lu_ref_init(&res->lr_reference);
1051
1052         /* one who creates the resource must unlock
1053          * the mutex after lvb initialization */
1054         mutex_init(&res->lr_lvb_mutex);
1055         mutex_lock(&res->lr_lvb_mutex);
1056
1057         return res;
1058 }
1059
1060 /* Args: unlocked namespace
1061  *  * Locks: takes and releases NS hash-lock and res->lr_lock
1062  *   * Returns: referenced, unlocked ldlm_resource or NULL */
1063 struct ldlm_resource *
1064 ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent,
1065                   const struct ldlm_res_id *name, ldlm_type_t type, int create)
1066 {
1067         cfs_hlist_node_t     *hnode;
1068         struct ldlm_resource *res;
1069         cfs_hash_bd_t         bd;
1070         __u64                 version;
1071
1072         LASSERT(ns != NULL);
1073         LASSERT(parent == NULL);
1074         LASSERT(ns->ns_rs_hash != NULL);
1075         LASSERT(name->name[0] != 0);
1076
1077         cfs_hash_bd_get_and_lock(ns->ns_rs_hash, (void *)name, &bd, 0);
1078         hnode = cfs_hash_bd_lookup_locked(ns->ns_rs_hash, &bd, (void *)name);
1079         if (hnode != NULL) {
1080                 cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 0);
1081                 res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
1082                 /* synchronize WRT resource creation */
1083                 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
1084                         mutex_lock(&res->lr_lvb_mutex);
1085                         mutex_unlock(&res->lr_lvb_mutex);
1086                 }
1087                 return res;
1088         }
1089
1090         version = cfs_hash_bd_version_get(&bd);
1091         cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 0);
1092
1093         if (create == 0)
1094                 return NULL;
1095
1096         LASSERTF(type >= LDLM_MIN_TYPE && type < LDLM_MAX_TYPE,
1097                  "type: %d\n", type);
1098         res = ldlm_resource_new();
1099         if (!res)
1100                 return NULL;
1101
1102         res->lr_ns_bucket  = cfs_hash_bd_extra_get(ns->ns_rs_hash, &bd);
1103         res->lr_name       = *name;
1104         res->lr_type       = type;
1105         res->lr_most_restr = LCK_NL;
1106
1107         cfs_hash_bd_lock(ns->ns_rs_hash, &bd, 1);
1108         hnode = (version == cfs_hash_bd_version_get(&bd)) ?  NULL :
1109                 cfs_hash_bd_lookup_locked(ns->ns_rs_hash, &bd, (void *)name);
1110
1111         if (hnode != NULL) {
1112                 /* someone won the race and added the resource before */
1113                 cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
1114                 /* clean lu_ref for failed resource */
1115                 lu_ref_fini(&res->lr_reference);
1116                 /* We have taken lr_lvb_mutex. Drop it. */
1117                 mutex_unlock(&res->lr_lvb_mutex);
1118                 OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
1119
1120                 res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
1121                 /* synchronize WRT resource creation */
1122                 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
1123                         mutex_lock(&res->lr_lvb_mutex);
1124                         mutex_unlock(&res->lr_lvb_mutex);
1125                 }
1126                 return res;
1127         }
1128         /* we won! let's add the resource */
1129         cfs_hash_bd_add_locked(ns->ns_rs_hash, &bd, &res->lr_hash);
1130         if (cfs_hash_bd_count_get(&bd) == 1)
1131                 ldlm_namespace_get(ns);
1132
1133         cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
1134         if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
1135                 int rc;
1136
1137                 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CREATE_RESOURCE, 2);
1138                 rc = ns->ns_lvbo->lvbo_init(res);
1139                 if (rc)
1140                         CERROR("lvbo_init failed for resource "
1141                                LPU64": rc %d\n", name->name[0], rc);
1142         }
1143
1144         /* we create resource with locked lr_lvb_mutex */
1145         mutex_unlock(&res->lr_lvb_mutex);
1146
1147         return res;
1148 }
1149 EXPORT_SYMBOL(ldlm_resource_get);
1150
1151 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res)
1152 {
1153         LASSERT(res != NULL);
1154         LASSERT(res != LP_POISON);
1155         cfs_atomic_inc(&res->lr_refcount);
1156         CDEBUG(D_INFO, "getref res: %p count: %d\n", res,
1157                cfs_atomic_read(&res->lr_refcount));
1158         return res;
1159 }
1160
1161 static void __ldlm_resource_putref_final(cfs_hash_bd_t *bd,
1162                                          struct ldlm_resource *res)
1163 {
1164         struct ldlm_ns_bucket *nsb = res->lr_ns_bucket;
1165
1166         if (!cfs_list_empty(&res->lr_granted)) {
1167                 ldlm_resource_dump(D_ERROR, res);
1168                 LBUG();
1169         }
1170
1171         if (!cfs_list_empty(&res->lr_converting)) {
1172                 ldlm_resource_dump(D_ERROR, res);
1173                 LBUG();
1174         }
1175
1176         if (!cfs_list_empty(&res->lr_waiting)) {
1177                 ldlm_resource_dump(D_ERROR, res);
1178                 LBUG();
1179         }
1180
1181         cfs_hash_bd_del_locked(nsb->nsb_namespace->ns_rs_hash,
1182                                bd, &res->lr_hash);
1183         lu_ref_fini(&res->lr_reference);
1184         if (cfs_hash_bd_count_get(bd) == 0)
1185                 ldlm_namespace_put(nsb->nsb_namespace);
1186 }
1187
1188 /* Returns 1 if the resource was freed, 0 if it remains. */
1189 int ldlm_resource_putref(struct ldlm_resource *res)
1190 {
1191         struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1192         cfs_hash_bd_t   bd;
1193
1194         LASSERT_ATOMIC_GT_LT(&res->lr_refcount, 0, LI_POISON);
1195         CDEBUG(D_INFO, "putref res: %p count: %d\n",
1196                res, cfs_atomic_read(&res->lr_refcount) - 1);
1197
1198         cfs_hash_bd_get(ns->ns_rs_hash, &res->lr_name, &bd);
1199         if (cfs_hash_bd_dec_and_lock(ns->ns_rs_hash, &bd, &res->lr_refcount)) {
1200                 __ldlm_resource_putref_final(&bd, res);
1201                 cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
1202                 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_free)
1203                         ns->ns_lvbo->lvbo_free(res);
1204                 OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
1205                 return 1;
1206         }
1207         return 0;
1208 }
1209 EXPORT_SYMBOL(ldlm_resource_putref);
1210
1211 /* Returns 1 if the resource was freed, 0 if it remains. */
1212 int ldlm_resource_putref_locked(struct ldlm_resource *res)
1213 {
1214         struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1215
1216         LASSERT_ATOMIC_GT_LT(&res->lr_refcount, 0, LI_POISON);
1217         CDEBUG(D_INFO, "putref res: %p count: %d\n",
1218                res, cfs_atomic_read(&res->lr_refcount) - 1);
1219
1220         if (cfs_atomic_dec_and_test(&res->lr_refcount)) {
1221                 cfs_hash_bd_t bd;
1222
1223                 cfs_hash_bd_get(ldlm_res_to_ns(res)->ns_rs_hash,
1224                                 &res->lr_name, &bd);
1225                 __ldlm_resource_putref_final(&bd, res);
1226                 cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
1227                 /* NB: ns_rs_hash is created with CFS_HASH_NO_ITEMREF,
1228                  * so we should never be here while calling cfs_hash_del,
1229                  * cfs_hash_for_each_nolock is the only case we can get
1230                  * here, which is safe to release cfs_hash_bd_lock.
1231                  */
1232                 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_free)
1233                         ns->ns_lvbo->lvbo_free(res);
1234                 OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
1235
1236                 cfs_hash_bd_lock(ns->ns_rs_hash, &bd, 1);
1237                 return 1;
1238         }
1239         return 0;
1240 }
1241
1242 void ldlm_resource_add_lock(struct ldlm_resource *res, cfs_list_t *head,
1243                             struct ldlm_lock *lock)
1244 {
1245         check_res_locked(res);
1246
1247         LDLM_DEBUG(lock, "About to add this lock:\n");
1248
1249         if (lock->l_destroyed) {
1250                 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
1251                 return;
1252         }
1253
1254         LASSERT(cfs_list_empty(&lock->l_res_link));
1255
1256         cfs_list_add_tail(&lock->l_res_link, head);
1257 }
1258
1259 void ldlm_resource_insert_lock_after(struct ldlm_lock *original,
1260                                      struct ldlm_lock *new)
1261 {
1262         struct ldlm_resource *res = original->l_resource;
1263
1264         check_res_locked(res);
1265
1266         ldlm_resource_dump(D_INFO, res);
1267         LDLM_DEBUG(new, "About to insert this lock after %p:\n", original);
1268
1269         if (new->l_destroyed) {
1270                 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
1271                 goto out;
1272         }
1273
1274         LASSERT(cfs_list_empty(&new->l_res_link));
1275
1276         cfs_list_add(&new->l_res_link, &original->l_res_link);
1277  out:;
1278 }
1279
1280 void ldlm_resource_unlink_lock(struct ldlm_lock *lock)
1281 {
1282         int type = lock->l_resource->lr_type;
1283
1284         check_res_locked(lock->l_resource);
1285         if (type == LDLM_IBITS || type == LDLM_PLAIN)
1286                 ldlm_unlink_lock_skiplist(lock);
1287         else if (type == LDLM_EXTENT)
1288                 ldlm_extent_unlink_lock(lock);
1289         cfs_list_del_init(&lock->l_res_link);
1290 }
1291 EXPORT_SYMBOL(ldlm_resource_unlink_lock);
1292
1293 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc)
1294 {
1295         desc->lr_type = res->lr_type;
1296         desc->lr_name = res->lr_name;
1297 }
1298
1299 void ldlm_dump_all_namespaces(ldlm_side_t client, int level)
1300 {
1301         cfs_list_t *tmp;
1302
1303         if (!((libcfs_debug | D_ERROR) & level))
1304                 return;
1305
1306         mutex_lock(ldlm_namespace_lock(client));
1307
1308         cfs_list_for_each(tmp, ldlm_namespace_list(client)) {
1309                 struct ldlm_namespace *ns;
1310                 ns = cfs_list_entry(tmp, struct ldlm_namespace, ns_list_chain);
1311                 ldlm_namespace_dump(level, ns);
1312         }
1313
1314         mutex_unlock(ldlm_namespace_lock(client));
1315 }
1316 EXPORT_SYMBOL(ldlm_dump_all_namespaces);
1317
1318 static int ldlm_res_hash_dump(cfs_hash_t *hs, cfs_hash_bd_t *bd,
1319                               cfs_hlist_node_t *hnode, void *arg)
1320 {
1321         struct ldlm_resource *res = cfs_hash_object(hs, hnode);
1322         int    level = (int)(unsigned long)arg;
1323
1324         lock_res(res);
1325         ldlm_resource_dump(level, res);
1326         unlock_res(res);
1327
1328         return 0;
1329 }
1330
1331 void ldlm_namespace_dump(int level, struct ldlm_namespace *ns)
1332 {
1333         if (!((libcfs_debug | D_ERROR) & level))
1334                 return;
1335
1336         CDEBUG(level, "--- Namespace: %s (rc: %d, side: %s)\n",
1337                ldlm_ns_name(ns), cfs_atomic_read(&ns->ns_bref),
1338                ns_is_client(ns) ? "client" : "server");
1339
1340         if (cfs_time_before(cfs_time_current(), ns->ns_next_dump))
1341                 return;
1342
1343         cfs_hash_for_each_nolock(ns->ns_rs_hash,
1344                                  ldlm_res_hash_dump,
1345                                  (void *)(unsigned long)level);
1346         spin_lock(&ns->ns_lock);
1347         ns->ns_next_dump = cfs_time_shift(10);
1348         spin_unlock(&ns->ns_lock);
1349 }
1350 EXPORT_SYMBOL(ldlm_namespace_dump);
1351
1352 void ldlm_resource_dump(int level, struct ldlm_resource *res)
1353 {
1354         struct ldlm_lock *lock;
1355         unsigned int granted = 0;
1356
1357         CLASSERT(RES_NAME_SIZE == 4);
1358
1359         if (!((libcfs_debug | D_ERROR) & level))
1360                 return;
1361
1362         CDEBUG(level, "--- Resource: %p ("LPU64"/"LPU64"/"LPU64"/"LPU64
1363                ") (rc: %d)\n", res, res->lr_name.name[0], res->lr_name.name[1],
1364                res->lr_name.name[2], res->lr_name.name[3],
1365                cfs_atomic_read(&res->lr_refcount));
1366
1367         if (!cfs_list_empty(&res->lr_granted)) {
1368                 CDEBUG(level, "Granted locks (in reverse order):\n");
1369                 cfs_list_for_each_entry_reverse(lock, &res->lr_granted,
1370                                                 l_res_link) {
1371                         LDLM_DEBUG_LIMIT(level, lock, "###");
1372                         if (!(level & D_CANTMASK) &&
1373                             ++granted > ldlm_dump_granted_max) {
1374                                 CDEBUG(level, "only dump %d granted locks to "
1375                                        "avoid DDOS.\n", granted);
1376                                 break;
1377                         }
1378                 }
1379         }
1380         if (!cfs_list_empty(&res->lr_converting)) {
1381                 CDEBUG(level, "Converting locks:\n");
1382                 cfs_list_for_each_entry(lock, &res->lr_converting, l_res_link)
1383                         LDLM_DEBUG_LIMIT(level, lock, "###");
1384         }
1385         if (!cfs_list_empty(&res->lr_waiting)) {
1386                 CDEBUG(level, "Waiting locks:\n");
1387                 cfs_list_for_each_entry(lock, &res->lr_waiting, l_res_link)
1388                         LDLM_DEBUG_LIMIT(level, lock, "###");
1389         }
1390 }