[go: nahoru, domu]

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) 2010, 2012, Intel Corporation.
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#include "../include/lustre_dlm.h"
44#include "../include/lustre_fid.h"
45#include "../include/obd_class.h"
46#include "ldlm_internal.h"
47
48struct kmem_cache *ldlm_resource_slab, *ldlm_lock_slab;
49
50int ldlm_srv_namespace_nr = 0;
51int ldlm_cli_namespace_nr = 0;
52
53struct mutex ldlm_srv_namespace_lock;
54LIST_HEAD(ldlm_srv_namespace_list);
55
56struct mutex ldlm_cli_namespace_lock;
57/* Client Namespaces that have active resources in them.
58 * Once all resources go away, ldlm_poold moves such namespaces to the
59 * inactive list */
60LIST_HEAD(ldlm_cli_active_namespace_list);
61/* Client namespaces that don't have any locks in them */
62LIST_HEAD(ldlm_cli_inactive_namespace_list);
63
64struct proc_dir_entry *ldlm_type_proc_dir = NULL;
65struct proc_dir_entry *ldlm_ns_proc_dir = NULL;
66struct proc_dir_entry *ldlm_svc_proc_dir = NULL;
67
68extern 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. */
72unsigned int ldlm_dump_granted_max = 256;
73
74#if defined (CONFIG_PROC_FS)
75static ssize_t lprocfs_wr_dump_ns(struct file *file, const char *buffer,
76				  size_t count, loff_t *off)
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}
82LPROC_SEQ_FOPS_WR_ONLY(ldlm, dump_ns);
83
84LPROC_SEQ_FOPS_RW_TYPE(ldlm_rw, uint);
85LPROC_SEQ_FOPS_RO_TYPE(ldlm, uint);
86
87int ldlm_proc_setup(void)
88{
89	int rc;
90	struct lprocfs_vars list[] = {
91		{ "dump_namespaces", &ldlm_dump_ns_fops, NULL, 0222 },
92		{ "dump_granted_max", &ldlm_rw_uint_fops,
93		  &ldlm_dump_granted_max },
94		{ "cancel_unused_locks_before_replay", &ldlm_rw_uint_fops,
95		  &ldlm_cancel_unused_locks_before_replay },
96		{ NULL }};
97	LASSERT(ldlm_ns_proc_dir == NULL);
98
99	ldlm_type_proc_dir = lprocfs_register(OBD_LDLM_DEVICENAME,
100					      proc_lustre_root,
101					      NULL, NULL);
102	if (IS_ERR(ldlm_type_proc_dir)) {
103		CERROR("LProcFS failed in ldlm-init\n");
104		rc = PTR_ERR(ldlm_type_proc_dir);
105		goto err;
106	}
107
108	ldlm_ns_proc_dir = lprocfs_register("namespaces",
109					    ldlm_type_proc_dir,
110					    NULL, NULL);
111	if (IS_ERR(ldlm_ns_proc_dir)) {
112		CERROR("LProcFS failed in ldlm-init\n");
113		rc = PTR_ERR(ldlm_ns_proc_dir);
114		goto err_type;
115	}
116
117	ldlm_svc_proc_dir = lprocfs_register("services",
118					    ldlm_type_proc_dir,
119					    NULL, NULL);
120	if (IS_ERR(ldlm_svc_proc_dir)) {
121		CERROR("LProcFS failed in ldlm-init\n");
122		rc = PTR_ERR(ldlm_svc_proc_dir);
123		goto err_ns;
124	}
125
126	rc = lprocfs_add_vars(ldlm_type_proc_dir, list, NULL);
127
128	return 0;
129
130err_ns:
131	lprocfs_remove(&ldlm_ns_proc_dir);
132err_type:
133	lprocfs_remove(&ldlm_type_proc_dir);
134err:
135	ldlm_svc_proc_dir = NULL;
136	ldlm_type_proc_dir = NULL;
137	ldlm_ns_proc_dir = NULL;
138	return rc;
139}
140
141void ldlm_proc_cleanup(void)
142{
143	if (ldlm_svc_proc_dir)
144		lprocfs_remove(&ldlm_svc_proc_dir);
145
146	if (ldlm_ns_proc_dir)
147		lprocfs_remove(&ldlm_ns_proc_dir);
148
149	if (ldlm_type_proc_dir)
150		lprocfs_remove(&ldlm_type_proc_dir);
151
152	ldlm_svc_proc_dir = NULL;
153	ldlm_type_proc_dir = NULL;
154	ldlm_ns_proc_dir = NULL;
155}
156
157static int lprocfs_ns_resources_seq_show(struct seq_file *m, void *v)
158{
159	struct ldlm_namespace *ns  = m->private;
160	__u64		  res = 0;
161	struct cfs_hash_bd	  bd;
162	int		    i;
163
164	/* result is not strictly consistent */
165	cfs_hash_for_each_bucket(ns->ns_rs_hash, &bd, i)
166		res += cfs_hash_bd_count_get(&bd);
167	return lprocfs_rd_u64(m, &res);
168}
169LPROC_SEQ_FOPS_RO(lprocfs_ns_resources);
170
171static int lprocfs_ns_locks_seq_show(struct seq_file *m, void *v)
172{
173	struct ldlm_namespace *ns = m->private;
174	__u64		  locks;
175
176	locks = lprocfs_stats_collector(ns->ns_stats, LDLM_NSS_LOCKS,
177					LPROCFS_FIELDS_FLAGS_SUM);
178	return lprocfs_rd_u64(m, &locks);
179}
180LPROC_SEQ_FOPS_RO(lprocfs_ns_locks);
181
182static int lprocfs_lru_size_seq_show(struct seq_file *m, void *v)
183{
184	struct ldlm_namespace *ns = m->private;
185	__u32 *nr = &ns->ns_max_unused;
186
187	if (ns_connect_lru_resize(ns))
188		nr = &ns->ns_nr_unused;
189	return lprocfs_rd_uint(m, nr);
190}
191
192static ssize_t lprocfs_lru_size_seq_write(struct file *file,
193					const char __user *buffer,
194					size_t count, loff_t *off)
195{
196	struct ldlm_namespace *ns = ((struct seq_file *)file->private_data)->private;
197	char dummy[MAX_STRING_SIZE + 1];
198	unsigned long tmp;
199	int lru_resize;
200	int err;
201
202	dummy[MAX_STRING_SIZE] = '\0';
203	if (copy_from_user(dummy, buffer, MAX_STRING_SIZE))
204		return -EFAULT;
205
206	if (strncmp(dummy, "clear", 5) == 0) {
207		CDEBUG(D_DLMTRACE,
208		       "dropping all unused locks from namespace %s\n",
209		       ldlm_ns_name(ns));
210		if (ns_connect_lru_resize(ns)) {
211			int canceled, unused  = ns->ns_nr_unused;
212
213			/* Try to cancel all @ns_nr_unused locks. */
214			canceled = ldlm_cancel_lru(ns, unused, 0,
215						   LDLM_CANCEL_PASSED);
216			if (canceled < unused) {
217				CDEBUG(D_DLMTRACE,
218				       "not all requested locks are canceled, "
219				       "requested: %d, canceled: %d\n", unused,
220				       canceled);
221				return -EINVAL;
222			}
223		} else {
224			tmp = ns->ns_max_unused;
225			ns->ns_max_unused = 0;
226			ldlm_cancel_lru(ns, 0, 0, LDLM_CANCEL_PASSED);
227			ns->ns_max_unused = tmp;
228		}
229		return count;
230	}
231
232	err = kstrtoul(dummy, 10, &tmp);
233	if (err != 0) {
234		CERROR("invalid value written\n");
235		return -EINVAL;
236	}
237	lru_resize = (tmp == 0);
238
239	if (ns_connect_lru_resize(ns)) {
240		if (!lru_resize)
241			ns->ns_max_unused = (unsigned int)tmp;
242
243		if (tmp > ns->ns_nr_unused)
244			tmp = ns->ns_nr_unused;
245		tmp = ns->ns_nr_unused - tmp;
246
247		CDEBUG(D_DLMTRACE,
248		       "changing namespace %s unused locks from %u to %u\n",
249		       ldlm_ns_name(ns), ns->ns_nr_unused,
250		       (unsigned int)tmp);
251		ldlm_cancel_lru(ns, tmp, LCF_ASYNC, LDLM_CANCEL_PASSED);
252
253		if (!lru_resize) {
254			CDEBUG(D_DLMTRACE,
255			       "disable lru_resize for namespace %s\n",
256			       ldlm_ns_name(ns));
257			ns->ns_connect_flags &= ~OBD_CONNECT_LRU_RESIZE;
258		}
259	} else {
260		CDEBUG(D_DLMTRACE,
261		       "changing namespace %s max_unused from %u to %u\n",
262		       ldlm_ns_name(ns), ns->ns_max_unused,
263		       (unsigned int)tmp);
264		ns->ns_max_unused = (unsigned int)tmp;
265		ldlm_cancel_lru(ns, 0, LCF_ASYNC, LDLM_CANCEL_PASSED);
266
267		/* Make sure that LRU resize was originally supported before
268		 * turning it on here. */
269		if (lru_resize &&
270		    (ns->ns_orig_connect_flags & OBD_CONNECT_LRU_RESIZE)) {
271			CDEBUG(D_DLMTRACE,
272			       "enable lru_resize for namespace %s\n",
273			       ldlm_ns_name(ns));
274			ns->ns_connect_flags |= OBD_CONNECT_LRU_RESIZE;
275		}
276	}
277
278	return count;
279}
280LPROC_SEQ_FOPS(lprocfs_lru_size);
281
282static int lprocfs_elc_seq_show(struct seq_file *m, void *v)
283{
284	struct ldlm_namespace *ns = m->private;
285	unsigned int supp = ns_connect_cancelset(ns);
286
287	return lprocfs_rd_uint(m, &supp);
288}
289
290static ssize_t lprocfs_elc_seq_write(struct file *file, const char *buffer,
291				 size_t count, loff_t *off)
292{
293	struct ldlm_namespace *ns = ((struct seq_file *)file->private_data)->private;
294	unsigned int supp = -1;
295	int rc;
296
297	rc = lprocfs_wr_uint(file, buffer, count, &supp);
298	if (rc < 0)
299		return rc;
300
301	if (supp == 0)
302		ns->ns_connect_flags &= ~OBD_CONNECT_CANCELSET;
303	else if (ns->ns_orig_connect_flags & OBD_CONNECT_CANCELSET)
304		ns->ns_connect_flags |= OBD_CONNECT_CANCELSET;
305	return count;
306}
307LPROC_SEQ_FOPS(lprocfs_elc);
308
309void ldlm_namespace_proc_unregister(struct ldlm_namespace *ns)
310{
311	if (ns->ns_proc_dir_entry == NULL)
312		CERROR("dlm namespace %s has no procfs dir?\n",
313		       ldlm_ns_name(ns));
314	else
315		lprocfs_remove(&ns->ns_proc_dir_entry);
316
317	if (ns->ns_stats != NULL)
318		lprocfs_free_stats(&ns->ns_stats);
319}
320
321#define LDLM_NS_ADD_VAR(name, var, ops)				\
322	do {							\
323		snprintf(lock_name, MAX_STRING_SIZE, name);	\
324		lock_vars[0].data = var;			\
325		lock_vars[0].fops = ops;			\
326		lprocfs_add_vars(ns_pde, lock_vars, NULL);	\
327	} while (0)
328
329int ldlm_namespace_proc_register(struct ldlm_namespace *ns)
330{
331	struct lprocfs_vars lock_vars[2];
332	char lock_name[MAX_STRING_SIZE + 1];
333	struct proc_dir_entry *ns_pde;
334
335	LASSERT(ns != NULL);
336	LASSERT(ns->ns_rs_hash != NULL);
337
338	if (ns->ns_proc_dir_entry != NULL) {
339		ns_pde = ns->ns_proc_dir_entry;
340	} else {
341		ns_pde = proc_mkdir(ldlm_ns_name(ns), ldlm_ns_proc_dir);
342		if (ns_pde == NULL)
343			return -ENOMEM;
344		ns->ns_proc_dir_entry = ns_pde;
345	}
346
347	ns->ns_stats = lprocfs_alloc_stats(LDLM_NSS_LAST, 0);
348	if (ns->ns_stats == NULL)
349		return -ENOMEM;
350
351	lprocfs_counter_init(ns->ns_stats, LDLM_NSS_LOCKS,
352			     LPROCFS_CNTR_AVGMINMAX, "locks", "locks");
353
354	lock_name[MAX_STRING_SIZE] = '\0';
355
356	memset(lock_vars, 0, sizeof(lock_vars));
357	lock_vars[0].name = lock_name;
358
359	LDLM_NS_ADD_VAR("resource_count", ns, &lprocfs_ns_resources_fops);
360	LDLM_NS_ADD_VAR("lock_count", ns, &lprocfs_ns_locks_fops);
361
362	if (ns_is_client(ns)) {
363		LDLM_NS_ADD_VAR("lock_unused_count", &ns->ns_nr_unused,
364				&ldlm_uint_fops);
365		LDLM_NS_ADD_VAR("lru_size", ns, &lprocfs_lru_size_fops);
366		LDLM_NS_ADD_VAR("lru_max_age", &ns->ns_max_age,
367				&ldlm_rw_uint_fops);
368		LDLM_NS_ADD_VAR("early_lock_cancel", ns, &lprocfs_elc_fops);
369	} else {
370		LDLM_NS_ADD_VAR("ctime_age_limit", &ns->ns_ctime_age_limit,
371				&ldlm_rw_uint_fops);
372		LDLM_NS_ADD_VAR("lock_timeouts", &ns->ns_timeouts,
373				&ldlm_uint_fops);
374		LDLM_NS_ADD_VAR("max_nolock_bytes", &ns->ns_max_nolock_size,
375				&ldlm_rw_uint_fops);
376		LDLM_NS_ADD_VAR("contention_seconds", &ns->ns_contention_time,
377				&ldlm_rw_uint_fops);
378		LDLM_NS_ADD_VAR("contended_locks", &ns->ns_contended_locks,
379				&ldlm_rw_uint_fops);
380		LDLM_NS_ADD_VAR("max_parallel_ast", &ns->ns_max_parallel_ast,
381				&ldlm_rw_uint_fops);
382	}
383	return 0;
384}
385#undef MAX_STRING_SIZE
386#else /* CONFIG_PROC_FS */
387
388#define ldlm_namespace_proc_unregister(ns)      ({;})
389#define ldlm_namespace_proc_register(ns)	({0;})
390
391#endif /* CONFIG_PROC_FS */
392
393static unsigned ldlm_res_hop_hash(struct cfs_hash *hs,
394				  const void *key, unsigned mask)
395{
396	const struct ldlm_res_id     *id  = key;
397	unsigned		val = 0;
398	unsigned		i;
399
400	for (i = 0; i < RES_NAME_SIZE; i++)
401		val += id->name[i];
402	return val & mask;
403}
404
405static unsigned ldlm_res_hop_fid_hash(struct cfs_hash *hs,
406				      const void *key, unsigned mask)
407{
408	const struct ldlm_res_id *id = key;
409	struct lu_fid       fid;
410	__u32	       hash;
411	__u32	       val;
412
413	fid.f_seq = id->name[LUSTRE_RES_ID_SEQ_OFF];
414	fid.f_oid = (__u32)id->name[LUSTRE_RES_ID_VER_OID_OFF];
415	fid.f_ver = (__u32)(id->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32);
416
417	hash = fid_flatten32(&fid);
418	hash += (hash >> 4) + (hash << 12); /* mixing oid and seq */
419	if (id->name[LUSTRE_RES_ID_HSH_OFF] != 0) {
420		val = id->name[LUSTRE_RES_ID_HSH_OFF];
421		hash += (val >> 5) + (val << 11);
422	} else {
423		val = fid_oid(&fid);
424	}
425	hash = hash_long(hash, hs->hs_bkt_bits);
426	/* give me another random factor */
427	hash -= hash_long((unsigned long)hs, val % 11 + 3);
428
429	hash <<= hs->hs_cur_bits - hs->hs_bkt_bits;
430	hash |= ldlm_res_hop_hash(hs, key, CFS_HASH_NBKT(hs) - 1);
431
432	return hash & mask;
433}
434
435static void *ldlm_res_hop_key(struct hlist_node *hnode)
436{
437	struct ldlm_resource   *res;
438
439	res = hlist_entry(hnode, struct ldlm_resource, lr_hash);
440	return &res->lr_name;
441}
442
443static int ldlm_res_hop_keycmp(const void *key, struct hlist_node *hnode)
444{
445	struct ldlm_resource   *res;
446
447	res = hlist_entry(hnode, struct ldlm_resource, lr_hash);
448	return ldlm_res_eq((const struct ldlm_res_id *)key,
449			   (const struct ldlm_res_id *)&res->lr_name);
450}
451
452static void *ldlm_res_hop_object(struct hlist_node *hnode)
453{
454	return hlist_entry(hnode, struct ldlm_resource, lr_hash);
455}
456
457static void ldlm_res_hop_get_locked(struct cfs_hash *hs, struct hlist_node *hnode)
458{
459	struct ldlm_resource *res;
460
461	res = hlist_entry(hnode, struct ldlm_resource, lr_hash);
462	ldlm_resource_getref(res);
463}
464
465static void ldlm_res_hop_put_locked(struct cfs_hash *hs, struct hlist_node *hnode)
466{
467	struct ldlm_resource *res;
468
469	res = hlist_entry(hnode, struct ldlm_resource, lr_hash);
470	/* cfs_hash_for_each_nolock is the only chance we call it */
471	ldlm_resource_putref_locked(res);
472}
473
474static void ldlm_res_hop_put(struct cfs_hash *hs, struct hlist_node *hnode)
475{
476	struct ldlm_resource *res;
477
478	res = hlist_entry(hnode, struct ldlm_resource, lr_hash);
479	ldlm_resource_putref(res);
480}
481
482cfs_hash_ops_t ldlm_ns_hash_ops = {
483	.hs_hash	= ldlm_res_hop_hash,
484	.hs_key	 = ldlm_res_hop_key,
485	.hs_keycmp      = ldlm_res_hop_keycmp,
486	.hs_keycpy      = NULL,
487	.hs_object      = ldlm_res_hop_object,
488	.hs_get	 = ldlm_res_hop_get_locked,
489	.hs_put_locked  = ldlm_res_hop_put_locked,
490	.hs_put	 = ldlm_res_hop_put
491};
492
493cfs_hash_ops_t ldlm_ns_fid_hash_ops = {
494	.hs_hash	= ldlm_res_hop_fid_hash,
495	.hs_key	 = ldlm_res_hop_key,
496	.hs_keycmp      = ldlm_res_hop_keycmp,
497	.hs_keycpy      = NULL,
498	.hs_object      = ldlm_res_hop_object,
499	.hs_get	 = ldlm_res_hop_get_locked,
500	.hs_put_locked  = ldlm_res_hop_put_locked,
501	.hs_put	 = ldlm_res_hop_put
502};
503
504typedef struct {
505	ldlm_ns_type_t  nsd_type;
506	/** hash bucket bits */
507	unsigned	nsd_bkt_bits;
508	/** hash bits */
509	unsigned	nsd_all_bits;
510	/** hash operations */
511	cfs_hash_ops_t *nsd_hops;
512} ldlm_ns_hash_def_t;
513
514ldlm_ns_hash_def_t ldlm_ns_hash_defs[] = {
515	{
516		.nsd_type       = LDLM_NS_TYPE_MDC,
517		.nsd_bkt_bits   = 11,
518		.nsd_all_bits   = 16,
519		.nsd_hops       = &ldlm_ns_fid_hash_ops,
520	},
521	{
522		.nsd_type       = LDLM_NS_TYPE_MDT,
523		.nsd_bkt_bits   = 14,
524		.nsd_all_bits   = 21,
525		.nsd_hops       = &ldlm_ns_fid_hash_ops,
526	},
527	{
528		.nsd_type       = LDLM_NS_TYPE_OSC,
529		.nsd_bkt_bits   = 8,
530		.nsd_all_bits   = 12,
531		.nsd_hops       = &ldlm_ns_hash_ops,
532	},
533	{
534		.nsd_type       = LDLM_NS_TYPE_OST,
535		.nsd_bkt_bits   = 11,
536		.nsd_all_bits   = 17,
537		.nsd_hops       = &ldlm_ns_hash_ops,
538	},
539	{
540		.nsd_type       = LDLM_NS_TYPE_MGC,
541		.nsd_bkt_bits   = 4,
542		.nsd_all_bits   = 4,
543		.nsd_hops       = &ldlm_ns_hash_ops,
544	},
545	{
546		.nsd_type       = LDLM_NS_TYPE_MGT,
547		.nsd_bkt_bits   = 4,
548		.nsd_all_bits   = 4,
549		.nsd_hops       = &ldlm_ns_hash_ops,
550	},
551	{
552		.nsd_type       = LDLM_NS_TYPE_UNKNOWN,
553	},
554};
555
556/**
557 * Create and initialize new empty namespace.
558 */
559struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
560					  ldlm_side_t client,
561					  ldlm_appetite_t apt,
562					  ldlm_ns_type_t ns_type)
563{
564	struct ldlm_namespace *ns = NULL;
565	struct ldlm_ns_bucket *nsb;
566	ldlm_ns_hash_def_t    *nsd;
567	struct cfs_hash_bd	  bd;
568	int		    idx;
569	int		    rc;
570
571	LASSERT(obd != NULL);
572
573	rc = ldlm_get_ref();
574	if (rc) {
575		CERROR("ldlm_get_ref failed: %d\n", rc);
576		return NULL;
577	}
578
579	for (idx = 0;;idx++) {
580		nsd = &ldlm_ns_hash_defs[idx];
581		if (nsd->nsd_type == LDLM_NS_TYPE_UNKNOWN) {
582			CERROR("Unknown type %d for ns %s\n", ns_type, name);
583			goto out_ref;
584		}
585
586		if (nsd->nsd_type == ns_type)
587			break;
588	}
589
590	OBD_ALLOC_PTR(ns);
591	if (!ns)
592		goto out_ref;
593
594	ns->ns_rs_hash = cfs_hash_create(name,
595					 nsd->nsd_all_bits, nsd->nsd_all_bits,
596					 nsd->nsd_bkt_bits, sizeof(*nsb),
597					 CFS_HASH_MIN_THETA,
598					 CFS_HASH_MAX_THETA,
599					 nsd->nsd_hops,
600					 CFS_HASH_DEPTH |
601					 CFS_HASH_BIGNAME |
602					 CFS_HASH_SPIN_BKTLOCK |
603					 CFS_HASH_NO_ITEMREF);
604	if (ns->ns_rs_hash == NULL)
605		goto out_ns;
606
607	cfs_hash_for_each_bucket(ns->ns_rs_hash, &bd, idx) {
608		nsb = cfs_hash_bd_extra_get(ns->ns_rs_hash, &bd);
609		at_init(&nsb->nsb_at_estimate, ldlm_enqueue_min, 0);
610		nsb->nsb_namespace = ns;
611	}
612
613	ns->ns_obd      = obd;
614	ns->ns_appetite = apt;
615	ns->ns_client   = client;
616
617	INIT_LIST_HEAD(&ns->ns_list_chain);
618	INIT_LIST_HEAD(&ns->ns_unused_list);
619	spin_lock_init(&ns->ns_lock);
620	atomic_set(&ns->ns_bref, 0);
621	init_waitqueue_head(&ns->ns_waitq);
622
623	ns->ns_max_nolock_size    = NS_DEFAULT_MAX_NOLOCK_BYTES;
624	ns->ns_contention_time    = NS_DEFAULT_CONTENTION_SECONDS;
625	ns->ns_contended_locks    = NS_DEFAULT_CONTENDED_LOCKS;
626
627	ns->ns_max_parallel_ast   = LDLM_DEFAULT_PARALLEL_AST_LIMIT;
628	ns->ns_nr_unused	  = 0;
629	ns->ns_max_unused	 = LDLM_DEFAULT_LRU_SIZE;
630	ns->ns_max_age	    = LDLM_DEFAULT_MAX_ALIVE;
631	ns->ns_ctime_age_limit    = LDLM_CTIME_AGE_LIMIT;
632	ns->ns_timeouts	   = 0;
633	ns->ns_orig_connect_flags = 0;
634	ns->ns_connect_flags      = 0;
635	ns->ns_stopping	   = 0;
636	rc = ldlm_namespace_proc_register(ns);
637	if (rc != 0) {
638		CERROR("Can't initialize ns proc, rc %d\n", rc);
639		goto out_hash;
640	}
641
642	idx = ldlm_namespace_nr_read(client);
643	rc = ldlm_pool_init(&ns->ns_pool, ns, idx, client);
644	if (rc) {
645		CERROR("Can't initialize lock pool, rc %d\n", rc);
646		goto out_proc;
647	}
648
649	ldlm_namespace_register(ns, client);
650	return ns;
651out_proc:
652	ldlm_namespace_proc_unregister(ns);
653	ldlm_namespace_cleanup(ns, 0);
654out_hash:
655	cfs_hash_putref(ns->ns_rs_hash);
656out_ns:
657	OBD_FREE_PTR(ns);
658out_ref:
659	ldlm_put_ref();
660	return NULL;
661}
662EXPORT_SYMBOL(ldlm_namespace_new);
663
664extern struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
665
666/**
667 * Cancel and destroy all locks on a resource.
668 *
669 * If flags contains FL_LOCAL_ONLY, don't try to tell the server, just
670 * clean up.  This is currently only used for recovery, and we make
671 * certain assumptions as a result--notably, that we shouldn't cancel
672 * locks with refs.
673 */
674static void cleanup_resource(struct ldlm_resource *res, struct list_head *q,
675			     __u64 flags)
676{
677	struct list_head *tmp;
678	int rc = 0, client = ns_is_client(ldlm_res_to_ns(res));
679	bool local_only = !!(flags & LDLM_FL_LOCAL_ONLY);
680
681	do {
682		struct ldlm_lock *lock = NULL;
683
684		/* First, we look for non-cleaned-yet lock
685		 * all cleaned locks are marked by CLEANED flag. */
686		lock_res(res);
687		list_for_each(tmp, q) {
688			lock = list_entry(tmp, struct ldlm_lock,
689					      l_res_link);
690			if (lock->l_flags & LDLM_FL_CLEANED) {
691				lock = NULL;
692				continue;
693			}
694			LDLM_LOCK_GET(lock);
695			lock->l_flags |= LDLM_FL_CLEANED;
696			break;
697		}
698
699		if (lock == NULL) {
700			unlock_res(res);
701			break;
702		}
703
704		/* Set CBPENDING so nothing in the cancellation path
705		 * can match this lock. */
706		lock->l_flags |= LDLM_FL_CBPENDING;
707		lock->l_flags |= LDLM_FL_FAILED;
708		lock->l_flags |= flags;
709
710		/* ... without sending a CANCEL message for local_only. */
711		if (local_only)
712			lock->l_flags |= LDLM_FL_LOCAL_ONLY;
713
714		if (local_only && (lock->l_readers || lock->l_writers)) {
715			/* This is a little bit gross, but much better than the
716			 * alternative: pretend that we got a blocking AST from
717			 * the server, so that when the lock is decref'd, it
718			 * will go away ... */
719			unlock_res(res);
720			LDLM_DEBUG(lock, "setting FL_LOCAL_ONLY");
721			if (lock->l_completion_ast)
722				lock->l_completion_ast(lock, 0, NULL);
723			LDLM_LOCK_RELEASE(lock);
724			continue;
725		}
726
727		if (client) {
728			struct lustre_handle lockh;
729
730			unlock_res(res);
731			ldlm_lock2handle(lock, &lockh);
732			rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
733			if (rc)
734				CERROR("ldlm_cli_cancel: %d\n", rc);
735		} else {
736			ldlm_resource_unlink_lock(lock);
737			unlock_res(res);
738			LDLM_DEBUG(lock, "Freeing a lock still held by a "
739				   "client node");
740			ldlm_lock_destroy(lock);
741		}
742		LDLM_LOCK_RELEASE(lock);
743	} while (1);
744}
745
746static int ldlm_resource_clean(struct cfs_hash *hs, struct cfs_hash_bd *bd,
747			       struct hlist_node *hnode, void *arg)
748{
749	struct ldlm_resource *res = cfs_hash_object(hs, hnode);
750	__u64 flags = *(__u64 *)arg;
751
752	cleanup_resource(res, &res->lr_granted, flags);
753	cleanup_resource(res, &res->lr_converting, flags);
754	cleanup_resource(res, &res->lr_waiting, flags);
755
756	return 0;
757}
758
759static int ldlm_resource_complain(struct cfs_hash *hs, struct cfs_hash_bd *bd,
760				  struct hlist_node *hnode, void *arg)
761{
762	struct ldlm_resource  *res = cfs_hash_object(hs, hnode);
763
764	lock_res(res);
765	CERROR("%s: namespace resource "DLDLMRES
766	       " (%p) refcount nonzero (%d) after lock cleanup; forcing cleanup.\n",
767	       ldlm_ns_name(ldlm_res_to_ns(res)), PLDLMRES(res), res,
768	       atomic_read(&res->lr_refcount) - 1);
769
770	ldlm_resource_dump(D_ERROR, res);
771	unlock_res(res);
772	return 0;
773}
774
775/**
776 * Cancel and destroy all locks in the namespace.
777 *
778 * Typically used during evictions when server notified client that it was
779 * evicted and all of its state needs to be destroyed.
780 * Also used during shutdown.
781 */
782int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags)
783{
784	if (ns == NULL) {
785		CDEBUG(D_INFO, "NULL ns, skipping cleanup\n");
786		return ELDLM_OK;
787	}
788
789	cfs_hash_for_each_nolock(ns->ns_rs_hash, ldlm_resource_clean, &flags);
790	cfs_hash_for_each_nolock(ns->ns_rs_hash, ldlm_resource_complain, NULL);
791	return ELDLM_OK;
792}
793EXPORT_SYMBOL(ldlm_namespace_cleanup);
794
795/**
796 * Attempts to free namespace.
797 *
798 * Only used when namespace goes away, like during an unmount.
799 */
800static int __ldlm_namespace_free(struct ldlm_namespace *ns, int force)
801{
802	/* At shutdown time, don't call the cancellation callback */
803	ldlm_namespace_cleanup(ns, force ? LDLM_FL_LOCAL_ONLY : 0);
804
805	if (atomic_read(&ns->ns_bref) > 0) {
806		struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
807		int rc;
808		CDEBUG(D_DLMTRACE,
809		       "dlm namespace %s free waiting on refcount %d\n",
810		       ldlm_ns_name(ns), atomic_read(&ns->ns_bref));
811force_wait:
812		if (force)
813			lwi = LWI_TIMEOUT(obd_timeout * HZ / 4, NULL, NULL);
814
815		rc = l_wait_event(ns->ns_waitq,
816				  atomic_read(&ns->ns_bref) == 0, &lwi);
817
818		/* Forced cleanups should be able to reclaim all references,
819		 * so it's safe to wait forever... we can't leak locks... */
820		if (force && rc == -ETIMEDOUT) {
821			LCONSOLE_ERROR("Forced cleanup waiting for %s "
822				       "namespace with %d resources in use, "
823				       "(rc=%d)\n", ldlm_ns_name(ns),
824				       atomic_read(&ns->ns_bref), rc);
825			goto force_wait;
826		}
827
828		if (atomic_read(&ns->ns_bref)) {
829			LCONSOLE_ERROR("Cleanup waiting for %s namespace "
830				       "with %d resources in use, (rc=%d)\n",
831				       ldlm_ns_name(ns),
832				       atomic_read(&ns->ns_bref), rc);
833			return ELDLM_NAMESPACE_EXISTS;
834		}
835		CDEBUG(D_DLMTRACE, "dlm namespace %s free done waiting\n",
836		       ldlm_ns_name(ns));
837	}
838
839	return ELDLM_OK;
840}
841
842/**
843 * Performs various cleanups for passed \a ns to make it drop refc and be
844 * ready for freeing. Waits for refc == 0.
845 *
846 * The following is done:
847 * (0) Unregister \a ns from its list to make inaccessible for potential
848 * users like pools thread and others;
849 * (1) Clear all locks in \a ns.
850 */
851void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
852			       struct obd_import *imp,
853			       int force)
854{
855	int rc;
856
857	if (!ns)
858		return;
859
860	spin_lock(&ns->ns_lock);
861	ns->ns_stopping = 1;
862	spin_unlock(&ns->ns_lock);
863
864	/*
865	 * Can fail with -EINTR when force == 0 in which case try harder.
866	 */
867	rc = __ldlm_namespace_free(ns, force);
868	if (rc != ELDLM_OK) {
869		if (imp) {
870			ptlrpc_disconnect_import(imp, 0);
871			ptlrpc_invalidate_import(imp);
872		}
873
874		/*
875		 * With all requests dropped and the import inactive
876		 * we are guaranteed all reference will be dropped.
877		 */
878		rc = __ldlm_namespace_free(ns, 1);
879		LASSERT(rc == 0);
880	}
881}
882
883/**
884 * Performs freeing memory structures related to \a ns. This is only done
885 * when ldlm_namespce_free_prior() successfully removed all resources
886 * referencing \a ns and its refc == 0.
887 */
888void ldlm_namespace_free_post(struct ldlm_namespace *ns)
889{
890	if (!ns)
891		return;
892
893	/* Make sure that nobody can find this ns in its list. */
894	ldlm_namespace_unregister(ns, ns->ns_client);
895	/* Fini pool _before_ parent proc dir is removed. This is important as
896	 * ldlm_pool_fini() removes own proc dir which is child to @dir.
897	 * Removing it after @dir may cause oops. */
898	ldlm_pool_fini(&ns->ns_pool);
899
900	ldlm_namespace_proc_unregister(ns);
901	cfs_hash_putref(ns->ns_rs_hash);
902	/* Namespace \a ns should be not on list at this time, otherwise
903	 * this will cause issues related to using freed \a ns in poold
904	 * thread. */
905	LASSERT(list_empty(&ns->ns_list_chain));
906	OBD_FREE_PTR(ns);
907	ldlm_put_ref();
908}
909
910/**
911 * Cleanup the resource, and free namespace.
912 * bug 12864:
913 * Deadlock issue:
914 * proc1: destroy import
915 *	class_disconnect_export(grab cl_sem) ->
916 *	      -> ldlm_namespace_free ->
917 *	      -> lprocfs_remove(grab _lprocfs_lock).
918 * proc2: read proc info
919 *	lprocfs_fops_read(grab _lprocfs_lock) ->
920 *	      -> osc_rd_active, etc(grab cl_sem).
921 *
922 * So that I have to split the ldlm_namespace_free into two parts - the first
923 * part ldlm_namespace_free_prior is used to cleanup the resource which is
924 * being used; the 2nd part ldlm_namespace_free_post is used to unregister the
925 * lprocfs entries, and then free memory. It will be called w/o cli->cl_sem
926 * held.
927 */
928void ldlm_namespace_free(struct ldlm_namespace *ns,
929			 struct obd_import *imp,
930			 int force)
931{
932	ldlm_namespace_free_prior(ns, imp, force);
933	ldlm_namespace_free_post(ns);
934}
935EXPORT_SYMBOL(ldlm_namespace_free);
936
937void ldlm_namespace_get(struct ldlm_namespace *ns)
938{
939	atomic_inc(&ns->ns_bref);
940}
941EXPORT_SYMBOL(ldlm_namespace_get);
942
943/* This is only for callers that care about refcount */
944int ldlm_namespace_get_return(struct ldlm_namespace *ns)
945{
946	return atomic_inc_return(&ns->ns_bref);
947}
948
949void ldlm_namespace_put(struct ldlm_namespace *ns)
950{
951	if (atomic_dec_and_lock(&ns->ns_bref, &ns->ns_lock)) {
952		wake_up(&ns->ns_waitq);
953		spin_unlock(&ns->ns_lock);
954	}
955}
956EXPORT_SYMBOL(ldlm_namespace_put);
957
958/** Register \a ns in the list of namespaces */
959void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client)
960{
961	mutex_lock(ldlm_namespace_lock(client));
962	LASSERT(list_empty(&ns->ns_list_chain));
963	list_add(&ns->ns_list_chain, ldlm_namespace_inactive_list(client));
964	ldlm_namespace_nr_inc(client);
965	mutex_unlock(ldlm_namespace_lock(client));
966}
967
968/** Unregister \a ns from the list of namespaces. */
969void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client)
970{
971	mutex_lock(ldlm_namespace_lock(client));
972	LASSERT(!list_empty(&ns->ns_list_chain));
973	/* Some asserts and possibly other parts of the code are still
974	 * using list_empty(&ns->ns_list_chain). This is why it is
975	 * important to use list_del_init() here. */
976	list_del_init(&ns->ns_list_chain);
977	ldlm_namespace_nr_dec(client);
978	mutex_unlock(ldlm_namespace_lock(client));
979}
980
981/** Should be called with ldlm_namespace_lock(client) taken. */
982void ldlm_namespace_move_to_active_locked(struct ldlm_namespace *ns,
983					  ldlm_side_t client)
984{
985	LASSERT(!list_empty(&ns->ns_list_chain));
986	LASSERT(mutex_is_locked(ldlm_namespace_lock(client)));
987	list_move_tail(&ns->ns_list_chain, ldlm_namespace_list(client));
988}
989
990/** Should be called with ldlm_namespace_lock(client) taken. */
991void ldlm_namespace_move_to_inactive_locked(struct ldlm_namespace *ns,
992					    ldlm_side_t client)
993{
994	LASSERT(!list_empty(&ns->ns_list_chain));
995	LASSERT(mutex_is_locked(ldlm_namespace_lock(client)));
996	list_move_tail(&ns->ns_list_chain,
997		       ldlm_namespace_inactive_list(client));
998}
999
1000/** Should be called with ldlm_namespace_lock(client) taken. */
1001struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client)
1002{
1003	LASSERT(mutex_is_locked(ldlm_namespace_lock(client)));
1004	LASSERT(!list_empty(ldlm_namespace_list(client)));
1005	return container_of(ldlm_namespace_list(client)->next,
1006		struct ldlm_namespace, ns_list_chain);
1007}
1008
1009/** Create and initialize new resource. */
1010static struct ldlm_resource *ldlm_resource_new(void)
1011{
1012	struct ldlm_resource *res;
1013	int idx;
1014
1015	OBD_SLAB_ALLOC_PTR_GFP(res, ldlm_resource_slab, GFP_NOFS);
1016	if (res == NULL)
1017		return NULL;
1018
1019	INIT_LIST_HEAD(&res->lr_granted);
1020	INIT_LIST_HEAD(&res->lr_converting);
1021	INIT_LIST_HEAD(&res->lr_waiting);
1022
1023	/* Initialize interval trees for each lock mode. */
1024	for (idx = 0; idx < LCK_MODE_NUM; idx++) {
1025		res->lr_itree[idx].lit_size = 0;
1026		res->lr_itree[idx].lit_mode = 1 << idx;
1027		res->lr_itree[idx].lit_root = NULL;
1028	}
1029
1030	atomic_set(&res->lr_refcount, 1);
1031	spin_lock_init(&res->lr_lock);
1032	lu_ref_init(&res->lr_reference);
1033
1034	/* The creator of the resource must unlock the mutex after LVB
1035	 * initialization. */
1036	mutex_init(&res->lr_lvb_mutex);
1037	mutex_lock(&res->lr_lvb_mutex);
1038
1039	return res;
1040}
1041
1042/**
1043 * Return a reference to resource with given name, creating it if necessary.
1044 * Args: namespace with ns_lock unlocked
1045 * Locks: takes and releases NS hash-lock and res->lr_lock
1046 * Returns: referenced, unlocked ldlm_resource or NULL
1047 */
1048struct ldlm_resource *
1049ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent,
1050		  const struct ldlm_res_id *name, ldlm_type_t type, int create)
1051{
1052	struct hlist_node     *hnode;
1053	struct ldlm_resource *res;
1054	struct cfs_hash_bd	 bd;
1055	__u64		 version;
1056	int		      ns_refcount = 0;
1057
1058	LASSERT(ns != NULL);
1059	LASSERT(parent == NULL);
1060	LASSERT(ns->ns_rs_hash != NULL);
1061	LASSERT(name->name[0] != 0);
1062
1063	cfs_hash_bd_get_and_lock(ns->ns_rs_hash, (void *)name, &bd, 0);
1064	hnode = cfs_hash_bd_lookup_locked(ns->ns_rs_hash, &bd, (void *)name);
1065	if (hnode != NULL) {
1066		cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 0);
1067		res = hlist_entry(hnode, struct ldlm_resource, lr_hash);
1068		/* Synchronize with regard to resource creation. */
1069		if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
1070			mutex_lock(&res->lr_lvb_mutex);
1071			mutex_unlock(&res->lr_lvb_mutex);
1072		}
1073
1074		if (unlikely(res->lr_lvb_len < 0)) {
1075			ldlm_resource_putref(res);
1076			res = NULL;
1077		}
1078		return res;
1079	}
1080
1081	version = cfs_hash_bd_version_get(&bd);
1082	cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 0);
1083
1084	if (create == 0)
1085		return NULL;
1086
1087	LASSERTF(type >= LDLM_MIN_TYPE && type < LDLM_MAX_TYPE,
1088		 "type: %d\n", type);
1089	res = ldlm_resource_new();
1090	if (!res)
1091		return NULL;
1092
1093	res->lr_ns_bucket  = cfs_hash_bd_extra_get(ns->ns_rs_hash, &bd);
1094	res->lr_name       = *name;
1095	res->lr_type       = type;
1096	res->lr_most_restr = LCK_NL;
1097
1098	cfs_hash_bd_lock(ns->ns_rs_hash, &bd, 1);
1099	hnode = (version == cfs_hash_bd_version_get(&bd)) ?  NULL :
1100		cfs_hash_bd_lookup_locked(ns->ns_rs_hash, &bd, (void *)name);
1101
1102	if (hnode != NULL) {
1103		/* Someone won the race and already added the resource. */
1104		cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
1105		/* Clean lu_ref for failed resource. */
1106		lu_ref_fini(&res->lr_reference);
1107		/* We have taken lr_lvb_mutex. Drop it. */
1108		mutex_unlock(&res->lr_lvb_mutex);
1109		OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof(*res));
1110
1111		res = hlist_entry(hnode, struct ldlm_resource, lr_hash);
1112		/* Synchronize with regard to resource creation. */
1113		if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
1114			mutex_lock(&res->lr_lvb_mutex);
1115			mutex_unlock(&res->lr_lvb_mutex);
1116		}
1117
1118		if (unlikely(res->lr_lvb_len < 0)) {
1119			ldlm_resource_putref(res);
1120			res = NULL;
1121		}
1122		return res;
1123	}
1124	/* We won! Let's add the resource. */
1125	cfs_hash_bd_add_locked(ns->ns_rs_hash, &bd, &res->lr_hash);
1126	if (cfs_hash_bd_count_get(&bd) == 1)
1127		ns_refcount = ldlm_namespace_get_return(ns);
1128
1129	cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
1130	if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
1131		int rc;
1132
1133		OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CREATE_RESOURCE, 2);
1134		rc = ns->ns_lvbo->lvbo_init(res);
1135		if (rc < 0) {
1136			CERROR("%s: lvbo_init failed for resource %#llx:%#llx: rc = %d\n",
1137			       ns->ns_obd->obd_name, name->name[0],
1138			       name->name[1], rc);
1139			if (res->lr_lvb_data) {
1140				OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1141				res->lr_lvb_data = NULL;
1142			}
1143			res->lr_lvb_len = rc;
1144			mutex_unlock(&res->lr_lvb_mutex);
1145			ldlm_resource_putref(res);
1146			return NULL;
1147		}
1148	}
1149
1150	/* We create resource with locked lr_lvb_mutex. */
1151	mutex_unlock(&res->lr_lvb_mutex);
1152
1153	/* Let's see if we happened to be the very first resource in this
1154	 * namespace. If so, and this is a client namespace, we need to move
1155	 * the namespace into the active namespaces list to be patrolled by
1156	 * the ldlm_poold. */
1157	if (ns_is_client(ns) && ns_refcount == 1) {
1158		mutex_lock(ldlm_namespace_lock(LDLM_NAMESPACE_CLIENT));
1159		ldlm_namespace_move_to_active_locked(ns, LDLM_NAMESPACE_CLIENT);
1160		mutex_unlock(ldlm_namespace_lock(LDLM_NAMESPACE_CLIENT));
1161	}
1162
1163	return res;
1164}
1165EXPORT_SYMBOL(ldlm_resource_get);
1166
1167struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res)
1168{
1169	LASSERT(res != NULL);
1170	LASSERT(res != LP_POISON);
1171	atomic_inc(&res->lr_refcount);
1172	CDEBUG(D_INFO, "getref res: %p count: %d\n", res,
1173	       atomic_read(&res->lr_refcount));
1174	return res;
1175}
1176
1177static void __ldlm_resource_putref_final(struct cfs_hash_bd *bd,
1178					 struct ldlm_resource *res)
1179{
1180	struct ldlm_ns_bucket *nsb = res->lr_ns_bucket;
1181
1182	if (!list_empty(&res->lr_granted)) {
1183		ldlm_resource_dump(D_ERROR, res);
1184		LBUG();
1185	}
1186
1187	if (!list_empty(&res->lr_converting)) {
1188		ldlm_resource_dump(D_ERROR, res);
1189		LBUG();
1190	}
1191
1192	if (!list_empty(&res->lr_waiting)) {
1193		ldlm_resource_dump(D_ERROR, res);
1194		LBUG();
1195	}
1196
1197	cfs_hash_bd_del_locked(nsb->nsb_namespace->ns_rs_hash,
1198			       bd, &res->lr_hash);
1199	lu_ref_fini(&res->lr_reference);
1200	if (cfs_hash_bd_count_get(bd) == 0)
1201		ldlm_namespace_put(nsb->nsb_namespace);
1202}
1203
1204/* Returns 1 if the resource was freed, 0 if it remains. */
1205int ldlm_resource_putref(struct ldlm_resource *res)
1206{
1207	struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1208	struct cfs_hash_bd   bd;
1209
1210	LASSERT_ATOMIC_GT_LT(&res->lr_refcount, 0, LI_POISON);
1211	CDEBUG(D_INFO, "putref res: %p count: %d\n",
1212	       res, atomic_read(&res->lr_refcount) - 1);
1213
1214	cfs_hash_bd_get(ns->ns_rs_hash, &res->lr_name, &bd);
1215	if (cfs_hash_bd_dec_and_lock(ns->ns_rs_hash, &bd, &res->lr_refcount)) {
1216		__ldlm_resource_putref_final(&bd, res);
1217		cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
1218		if (ns->ns_lvbo && ns->ns_lvbo->lvbo_free)
1219			ns->ns_lvbo->lvbo_free(res);
1220		OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof(*res));
1221		return 1;
1222	}
1223	return 0;
1224}
1225EXPORT_SYMBOL(ldlm_resource_putref);
1226
1227/* Returns 1 if the resource was freed, 0 if it remains. */
1228int ldlm_resource_putref_locked(struct ldlm_resource *res)
1229{
1230	struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1231
1232	LASSERT_ATOMIC_GT_LT(&res->lr_refcount, 0, LI_POISON);
1233	CDEBUG(D_INFO, "putref res: %p count: %d\n",
1234	       res, atomic_read(&res->lr_refcount) - 1);
1235
1236	if (atomic_dec_and_test(&res->lr_refcount)) {
1237		struct cfs_hash_bd bd;
1238
1239		cfs_hash_bd_get(ldlm_res_to_ns(res)->ns_rs_hash,
1240				&res->lr_name, &bd);
1241		__ldlm_resource_putref_final(&bd, res);
1242		cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
1243		/* NB: ns_rs_hash is created with CFS_HASH_NO_ITEMREF,
1244		 * so we should never be here while calling cfs_hash_del,
1245		 * cfs_hash_for_each_nolock is the only case we can get
1246		 * here, which is safe to release cfs_hash_bd_lock.
1247		 */
1248		if (ns->ns_lvbo && ns->ns_lvbo->lvbo_free)
1249			ns->ns_lvbo->lvbo_free(res);
1250		OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof(*res));
1251
1252		cfs_hash_bd_lock(ns->ns_rs_hash, &bd, 1);
1253		return 1;
1254	}
1255	return 0;
1256}
1257
1258/**
1259 * Add a lock into a given resource into specified lock list.
1260 */
1261void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
1262			    struct ldlm_lock *lock)
1263{
1264	check_res_locked(res);
1265
1266	LDLM_DEBUG(lock, "About to add this lock:\n");
1267
1268	if (lock->l_flags & LDLM_FL_DESTROYED) {
1269		CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
1270		return;
1271	}
1272
1273	LASSERT(list_empty(&lock->l_res_link));
1274
1275	list_add_tail(&lock->l_res_link, head);
1276}
1277
1278/**
1279 * Insert a lock into resource after specified lock.
1280 *
1281 * Obtain resource description from the lock we are inserting after.
1282 */
1283void ldlm_resource_insert_lock_after(struct ldlm_lock *original,
1284				     struct ldlm_lock *new)
1285{
1286	struct ldlm_resource *res = original->l_resource;
1287
1288	check_res_locked(res);
1289
1290	ldlm_resource_dump(D_INFO, res);
1291	LDLM_DEBUG(new, "About to insert this lock after %p:\n", original);
1292
1293	if (new->l_flags & LDLM_FL_DESTROYED) {
1294		CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
1295		goto out;
1296	}
1297
1298	LASSERT(list_empty(&new->l_res_link));
1299
1300	list_add(&new->l_res_link, &original->l_res_link);
1301 out:;
1302}
1303
1304void ldlm_resource_unlink_lock(struct ldlm_lock *lock)
1305{
1306	int type = lock->l_resource->lr_type;
1307
1308	check_res_locked(lock->l_resource);
1309	if (type == LDLM_IBITS || type == LDLM_PLAIN)
1310		ldlm_unlink_lock_skiplist(lock);
1311	else if (type == LDLM_EXTENT)
1312		ldlm_extent_unlink_lock(lock);
1313	list_del_init(&lock->l_res_link);
1314}
1315EXPORT_SYMBOL(ldlm_resource_unlink_lock);
1316
1317void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc)
1318{
1319	desc->lr_type = res->lr_type;
1320	desc->lr_name = res->lr_name;
1321}
1322
1323/**
1324 * Print information about all locks in all namespaces on this node to debug
1325 * log.
1326 */
1327void ldlm_dump_all_namespaces(ldlm_side_t client, int level)
1328{
1329	struct list_head *tmp;
1330
1331	if (!((libcfs_debug | D_ERROR) & level))
1332		return;
1333
1334	mutex_lock(ldlm_namespace_lock(client));
1335
1336	list_for_each(tmp, ldlm_namespace_list(client)) {
1337		struct ldlm_namespace *ns;
1338		ns = list_entry(tmp, struct ldlm_namespace, ns_list_chain);
1339		ldlm_namespace_dump(level, ns);
1340	}
1341
1342	mutex_unlock(ldlm_namespace_lock(client));
1343}
1344EXPORT_SYMBOL(ldlm_dump_all_namespaces);
1345
1346static int ldlm_res_hash_dump(struct cfs_hash *hs, struct cfs_hash_bd *bd,
1347			      struct hlist_node *hnode, void *arg)
1348{
1349	struct ldlm_resource *res = cfs_hash_object(hs, hnode);
1350	int    level = (int)(unsigned long)arg;
1351
1352	lock_res(res);
1353	ldlm_resource_dump(level, res);
1354	unlock_res(res);
1355
1356	return 0;
1357}
1358
1359/**
1360 * Print information about all locks in this namespace on this node to debug
1361 * log.
1362 */
1363void ldlm_namespace_dump(int level, struct ldlm_namespace *ns)
1364{
1365	if (!((libcfs_debug | D_ERROR) & level))
1366		return;
1367
1368	CDEBUG(level, "--- Namespace: %s (rc: %d, side: %s)\n",
1369	       ldlm_ns_name(ns), atomic_read(&ns->ns_bref),
1370	       ns_is_client(ns) ? "client" : "server");
1371
1372	if (time_before(cfs_time_current(), ns->ns_next_dump))
1373		return;
1374
1375	cfs_hash_for_each_nolock(ns->ns_rs_hash,
1376				 ldlm_res_hash_dump,
1377				 (void *)(unsigned long)level);
1378	spin_lock(&ns->ns_lock);
1379	ns->ns_next_dump = cfs_time_shift(10);
1380	spin_unlock(&ns->ns_lock);
1381}
1382EXPORT_SYMBOL(ldlm_namespace_dump);
1383
1384/**
1385 * Print information about all locks in this resource to debug log.
1386 */
1387void ldlm_resource_dump(int level, struct ldlm_resource *res)
1388{
1389	struct ldlm_lock *lock;
1390	unsigned int granted = 0;
1391
1392	CLASSERT(RES_NAME_SIZE == 4);
1393
1394	if (!((libcfs_debug | D_ERROR) & level))
1395		return;
1396
1397	CDEBUG(level, "--- Resource: "DLDLMRES" (%p) refcount = %d\n",
1398	       PLDLMRES(res), res, atomic_read(&res->lr_refcount));
1399
1400	if (!list_empty(&res->lr_granted)) {
1401		CDEBUG(level, "Granted locks (in reverse order):\n");
1402		list_for_each_entry_reverse(lock, &res->lr_granted,
1403						l_res_link) {
1404			LDLM_DEBUG_LIMIT(level, lock, "###");
1405			if (!(level & D_CANTMASK) &&
1406			    ++granted > ldlm_dump_granted_max) {
1407				CDEBUG(level, "only dump %d granted locks to "
1408				       "avoid DDOS.\n", granted);
1409				break;
1410			}
1411		}
1412	}
1413	if (!list_empty(&res->lr_converting)) {
1414		CDEBUG(level, "Converting locks:\n");
1415		list_for_each_entry(lock, &res->lr_converting, l_res_link)
1416			LDLM_DEBUG_LIMIT(level, lock, "###");
1417	}
1418	if (!list_empty(&res->lr_waiting)) {
1419		CDEBUG(level, "Waiting locks:\n");
1420		list_for_each_entry(lock, &res->lr_waiting, l_res_link)
1421			LDLM_DEBUG_LIMIT(level, lock, "###");
1422	}
1423}
1424