[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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 */
30/*
31 * Copyright (c) 2011, 2012, Intel Corporation.
32 */
33/*
34 * This file is part of Lustre, http://www.lustre.org/
35 * Lustre is a trademark of Sun Microsystems, Inc.
36 *
37 * lustre/include/lustre_mdc.h
38 *
39 * MDS data structures.
40 * See also lustre_idl.h for wire formats of requests.
41 */
42
43#ifndef _LUSTRE_MDC_H
44#define _LUSTRE_MDC_H
45
46/** \defgroup mdc mdc
47 *
48 * @{
49 */
50
51#include <linux/fs.h>
52#include <linux/dcache.h>
53#include "lustre_intent.h"
54#include "lustre_handles.h"
55#include "../../include/linux/libcfs/libcfs.h"
56#include "obd_class.h"
57#include "lustre/lustre_idl.h"
58#include "lustre_lib.h"
59#include "lustre_dlm.h"
60#include "lustre_export.h"
61
62struct ptlrpc_client;
63struct obd_export;
64struct ptlrpc_request;
65struct obd_device;
66
67struct mdc_rpc_lock {
68	struct mutex		rpcl_mutex;
69	struct lookup_intent	*rpcl_it;
70	int			rpcl_fakes;
71};
72
73#define MDC_FAKE_RPCL_IT ((void *)0x2c0012bfUL)
74
75static inline void mdc_init_rpc_lock(struct mdc_rpc_lock *lck)
76{
77	mutex_init(&lck->rpcl_mutex);
78	lck->rpcl_it = NULL;
79}
80
81static inline void mdc_get_rpc_lock(struct mdc_rpc_lock *lck,
82				    struct lookup_intent *it)
83{
84	if (it != NULL && (it->it_op == IT_GETATTR || it->it_op == IT_LOOKUP ||
85			   it->it_op == IT_LAYOUT))
86		return;
87
88	/* This would normally block until the existing request finishes.
89	 * If fail_loc is set it will block until the regular request is
90	 * done, then set rpcl_it to MDC_FAKE_RPCL_IT.  Once that is set
91	 * it will only be cleared when all fake requests are finished.
92	 * Only when all fake requests are finished can normal requests
93	 * be sent, to ensure they are recoverable again. */
94 again:
95	mutex_lock(&lck->rpcl_mutex);
96
97	if (CFS_FAIL_CHECK_QUIET(OBD_FAIL_MDC_RPCS_SEM)) {
98		lck->rpcl_it = MDC_FAKE_RPCL_IT;
99		lck->rpcl_fakes++;
100		mutex_unlock(&lck->rpcl_mutex);
101		return;
102	}
103
104	/* This will only happen when the CFS_FAIL_CHECK() was
105	 * just turned off but there are still requests in progress.
106	 * Wait until they finish.  It doesn't need to be efficient
107	 * in this extremely rare case, just have low overhead in
108	 * the common case when it isn't true. */
109	while (unlikely(lck->rpcl_it == MDC_FAKE_RPCL_IT)) {
110		mutex_unlock(&lck->rpcl_mutex);
111		schedule_timeout(cfs_time_seconds(1) / 4);
112		goto again;
113	}
114
115	LASSERT(lck->rpcl_it == NULL);
116	lck->rpcl_it = it;
117}
118
119static inline void mdc_put_rpc_lock(struct mdc_rpc_lock *lck,
120				    struct lookup_intent *it)
121{
122	if (it != NULL && (it->it_op == IT_GETATTR || it->it_op == IT_LOOKUP ||
123			   it->it_op == IT_LAYOUT))
124		return;
125
126	if (lck->rpcl_it == MDC_FAKE_RPCL_IT) { /* OBD_FAIL_MDC_RPCS_SEM */
127		mutex_lock(&lck->rpcl_mutex);
128
129		LASSERTF(lck->rpcl_fakes > 0, "%d\n", lck->rpcl_fakes);
130		lck->rpcl_fakes--;
131
132		if (lck->rpcl_fakes == 0)
133			lck->rpcl_it = NULL;
134
135	} else {
136		LASSERTF(it == lck->rpcl_it, "%p != %p\n", it, lck->rpcl_it);
137		lck->rpcl_it = NULL;
138	}
139
140	mutex_unlock(&lck->rpcl_mutex);
141}
142
143/* Update the maximum observed easize and cookiesize.  The default easize
144 * and cookiesize is initialized to the minimum value but allowed to grow
145 * up to a single page in size if required to handle the common case.
146 */
147static inline void mdc_update_max_ea_from_body(struct obd_export *exp,
148					       struct mdt_body *body)
149{
150	if (body->valid & OBD_MD_FLMODEASIZE) {
151		struct client_obd *cli = &exp->exp_obd->u.cli;
152
153		if (cli->cl_max_mds_easize < body->max_mdsize) {
154			cli->cl_max_mds_easize = body->max_mdsize;
155			cli->cl_default_mds_easize =
156			    min_t(__u32, body->max_mdsize, PAGE_CACHE_SIZE);
157		}
158		if (cli->cl_max_mds_cookiesize < body->max_cookiesize) {
159			cli->cl_max_mds_cookiesize = body->max_cookiesize;
160			cli->cl_default_mds_cookiesize =
161			    min_t(__u32, body->max_cookiesize, PAGE_CACHE_SIZE);
162		}
163	}
164}
165
166
167struct mdc_cache_waiter {
168	struct list_head	      mcw_entry;
169	wait_queue_head_t	     mcw_waitq;
170};
171
172/* mdc/mdc_locks.c */
173int it_disposition(struct lookup_intent *it, int flag);
174void it_clear_disposition(struct lookup_intent *it, int flag);
175void it_set_disposition(struct lookup_intent *it, int flag);
176int it_open_error(int phase, struct lookup_intent *it);
177
178static inline bool cl_is_lov_delay_create(unsigned int flags)
179{
180	return (flags & O_LOV_DELAY_CREATE) == O_LOV_DELAY_CREATE;
181}
182
183static inline void cl_lov_delay_create_clear(unsigned int *flags)
184{
185	if ((*flags & O_LOV_DELAY_CREATE) == O_LOV_DELAY_CREATE)
186		*flags &= ~O_LOV_DELAY_CREATE;
187}
188
189/** @} mdc */
190
191#endif
192