4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
27 * Copyright (c) 2011, 2014, Intel Corporation.
30 * This file is part of Lustre, http://www.lustre.org/
31 * Lustre is a trademark of Sun Microsystems, Inc.
33 * lustre/include/lustre_mdc.h
35 * MDS data structures.
36 * See also lustre_idl.h for wire formats of requests.
48 #include <linux/dcache.h>
49 #ifdef CONFIG_FS_POSIX_ACL
50 # include <linux/posix_acl_xattr.h>
51 #endif /* CONFIG_FS_POSIX_ACL */
52 #include <lustre_handles.h>
53 #include <lustre_intent.h>
54 #include <libcfs/libcfs.h>
55 #include <obd_class.h>
56 #include <lustre/lustre_idl.h>
57 #include <lustre_lib.h>
58 #include <lustre_dlm.h>
59 #include <lustre_export.h>
63 struct ptlrpc_request;
67 * Serializes in-flight MDT-modifying RPC requests to preserve idempotency.
69 * This mutex is used to implement execute-once semantics on the MDT.
70 * The MDT stores the last transaction ID and result for every client in
71 * its last_rcvd file. If the client doesn't get a reply, it can safely
72 * resend the request and the MDT will reconstruct the reply being aware
73 * that the request has already been executed. Without this lock,
74 * execution status of concurrent in-flight requests would be
77 * This design limits the extent to which we can keep a full pipeline of
78 * in-flight requests from a single client. This limitation could be
79 * overcome by allowing multiple slots per client in the last_rcvd file.
82 /** Lock protecting in-flight RPC concurrency. */
83 struct mutex rpcl_mutex;
84 /** Intent associated with currently executing request. */
85 struct lookup_intent *rpcl_it;
86 /** Used for MDS/RPC load testing purposes. */
90 #define MDC_FAKE_RPCL_IT ((void *)0x2c0012bfUL)
92 static inline void mdc_init_rpc_lock(struct mdc_rpc_lock *lck)
94 mutex_init(&lck->rpcl_mutex);
98 static inline void mdc_get_rpc_lock(struct mdc_rpc_lock *lck,
99 struct lookup_intent *it)
103 if (it != NULL && (it->it_op == IT_GETATTR || it->it_op == IT_LOOKUP ||
104 it->it_op == IT_LAYOUT || it->it_op == IT_READDIR))
107 /* This would normally block until the existing request finishes.
108 * If fail_loc is set it will block until the regular request is
109 * done, then set rpcl_it to MDC_FAKE_RPCL_IT. Once that is set
110 * it will only be cleared when all fake requests are finished.
111 * Only when all fake requests are finished can normal requests
112 * be sent, to ensure they are recoverable again. */
114 mutex_lock(&lck->rpcl_mutex);
116 if (CFS_FAIL_CHECK_QUIET(OBD_FAIL_MDC_RPCS_SEM)) {
117 lck->rpcl_it = MDC_FAKE_RPCL_IT;
119 mutex_unlock(&lck->rpcl_mutex);
123 /* This will only happen when the CFS_FAIL_CHECK() was
124 * just turned off but there are still requests in progress.
125 * Wait until they finish. It doesn't need to be efficient
126 * in this extremely rare case, just have low overhead in
127 * the common case when it isn't true. */
128 while (unlikely(lck->rpcl_it == MDC_FAKE_RPCL_IT)) {
129 mutex_unlock(&lck->rpcl_mutex);
130 schedule_timeout(cfs_time_seconds(1) / 4);
134 LASSERT(lck->rpcl_it == NULL);
138 static inline void mdc_put_rpc_lock(struct mdc_rpc_lock *lck,
139 struct lookup_intent *it)
141 if (it != NULL && (it->it_op == IT_GETATTR || it->it_op == IT_LOOKUP ||
142 it->it_op == IT_LAYOUT || it->it_op == IT_READDIR))
145 if (lck->rpcl_it == MDC_FAKE_RPCL_IT) { /* OBD_FAIL_MDC_RPCS_SEM */
146 mutex_lock(&lck->rpcl_mutex);
148 LASSERTF(lck->rpcl_fakes > 0, "%d\n", lck->rpcl_fakes);
151 if (lck->rpcl_fakes == 0)
155 LASSERTF(it == lck->rpcl_it, "%p != %p\n", it, lck->rpcl_it);
159 mutex_unlock(&lck->rpcl_mutex);
164 static inline void mdc_get_mod_rpc_slot(struct ptlrpc_request *req,
165 struct lookup_intent *it)
167 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
171 opc = lustre_msg_get_opc(req->rq_reqmsg);
172 tag = obd_get_mod_rpc_slot(cli, opc, it);
173 lustre_msg_set_tag(req->rq_reqmsg, tag);
176 static inline void mdc_put_mod_rpc_slot(struct ptlrpc_request *req,
177 struct lookup_intent *it)
179 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
183 opc = lustre_msg_get_opc(req->rq_reqmsg);
184 tag = lustre_msg_get_tag(req->rq_reqmsg);
185 obd_put_mod_rpc_slot(cli, opc, it, tag);
190 * Update the maximum possible easize.
192 * This value is learned from ptlrpc replies sent by the MDT. The
193 * default easize is initialized to the minimum value but allowed to
194 * grow up to a single page in size if required to handle the common
197 * \see client_obd::cl_default_mds_easize
199 * \param[in] exp export for MDC device
200 * \param[in] body body of ptlrpc reply from MDT
203 static inline void mdc_update_max_ea_from_body(struct obd_export *exp,
204 struct mdt_body *body)
206 if (body->mbo_valid & OBD_MD_FLMODEASIZE) {
207 struct client_obd *cli = &exp->exp_obd->u.cli;
210 if (cli->cl_max_mds_easize < body->mbo_max_mdsize)
211 cli->cl_max_mds_easize = body->mbo_max_mdsize;
213 def_easize = min_t(__u32, body->mbo_max_mdsize,
214 OBD_MAX_DEFAULT_EA_SIZE);
215 cli->cl_default_mds_easize = def_easize;
220 /* mdc/mdc_locks.c */
221 int it_open_error(int phase, struct lookup_intent *it);
223 static inline bool cl_is_lov_delay_create(unsigned int flags)
225 return (flags & O_LOV_DELAY_CREATE_1_8) != 0 ||
226 (flags & O_LOV_DELAY_CREATE_MASK) == O_LOV_DELAY_CREATE_MASK;
229 static inline void cl_lov_delay_create_clear(unsigned int *flags)
231 if ((*flags & O_LOV_DELAY_CREATE_1_8) != 0)
232 *flags &= ~O_LOV_DELAY_CREATE_1_8;
233 if ((*flags & O_LOV_DELAY_CREATE_MASK) == O_LOV_DELAY_CREATE_MASK)
234 *flags &= ~O_LOV_DELAY_CREATE_MASK;