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.sun.com/software/products/lustre/docs/GPLv2.pdf
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
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * Copyright (c) 2011, 2014, Intel Corporation.
34 * This file is part of Lustre, http://www.lustre.org/
35 * Lustre is a trademark of Sun Microsystems, Inc.
37 * lustre/include/lustre_mdc.h
39 * MDS data structures.
40 * See also lustre_idl.h for wire formats of requests.
52 #include <linux/dcache.h>
53 #ifdef CONFIG_FS_POSIX_ACL
54 # include <linux/posix_acl_xattr.h>
55 #endif /* CONFIG_FS_POSIX_ACL */
56 #include <lustre_handles.h>
57 #include <lustre_intent.h>
58 #include <libcfs/libcfs.h>
59 #include <obd_class.h>
60 #include <lustre/lustre_idl.h>
61 #include <lustre_lib.h>
62 #include <lustre_dlm.h>
63 #include <lustre_export.h>
67 struct ptlrpc_request;
71 * Serializes in-flight MDT-modifying RPC requests to preserve idempotency.
73 * This mutex is used to implement execute-once semantics on the MDT.
74 * The MDT stores the last transaction ID and result for every client in
75 * its last_rcvd file. If the client doesn't get a reply, it can safely
76 * resend the request and the MDT will reconstruct the reply being aware
77 * that the request has already been executed. Without this lock,
78 * execution status of concurrent in-flight requests would be
81 * This design limits the extent to which we can keep a full pipeline of
82 * in-flight requests from a single client. This limitation could be
83 * overcome by allowing multiple slots per client in the last_rcvd file.
86 /** Lock protecting in-flight RPC concurrency. */
87 struct mutex rpcl_mutex;
88 /** Intent associated with currently executing request. */
89 struct lookup_intent *rpcl_it;
90 /** Used for MDS/RPC load testing purposes. */
94 #define MDC_FAKE_RPCL_IT ((void *)0x2c0012bfUL)
96 static inline void mdc_init_rpc_lock(struct mdc_rpc_lock *lck)
98 mutex_init(&lck->rpcl_mutex);
102 static inline void mdc_get_rpc_lock(struct mdc_rpc_lock *lck,
103 struct lookup_intent *it)
107 if (it != NULL && (it->it_op == IT_GETATTR || it->it_op == IT_LOOKUP ||
108 it->it_op == IT_LAYOUT || it->it_op == IT_READDIR))
111 /* This would normally block until the existing request finishes.
112 * If fail_loc is set it will block until the regular request is
113 * done, then set rpcl_it to MDC_FAKE_RPCL_IT. Once that is set
114 * it will only be cleared when all fake requests are finished.
115 * Only when all fake requests are finished can normal requests
116 * be sent, to ensure they are recoverable again. */
118 mutex_lock(&lck->rpcl_mutex);
120 if (CFS_FAIL_CHECK_QUIET(OBD_FAIL_MDC_RPCS_SEM)) {
121 lck->rpcl_it = MDC_FAKE_RPCL_IT;
123 mutex_unlock(&lck->rpcl_mutex);
127 /* This will only happen when the CFS_FAIL_CHECK() was
128 * just turned off but there are still requests in progress.
129 * Wait until they finish. It doesn't need to be efficient
130 * in this extremely rare case, just have low overhead in
131 * the common case when it isn't true. */
132 while (unlikely(lck->rpcl_it == MDC_FAKE_RPCL_IT)) {
133 mutex_unlock(&lck->rpcl_mutex);
134 schedule_timeout(cfs_time_seconds(1) / 4);
138 LASSERT(lck->rpcl_it == NULL);
142 static inline void mdc_put_rpc_lock(struct mdc_rpc_lock *lck,
143 struct lookup_intent *it)
145 if (it != NULL && (it->it_op == IT_GETATTR || it->it_op == IT_LOOKUP ||
146 it->it_op == IT_LAYOUT || it->it_op == IT_READDIR))
149 if (lck->rpcl_it == MDC_FAKE_RPCL_IT) { /* OBD_FAIL_MDC_RPCS_SEM */
150 mutex_lock(&lck->rpcl_mutex);
152 LASSERTF(lck->rpcl_fakes > 0, "%d\n", lck->rpcl_fakes);
155 if (lck->rpcl_fakes == 0)
159 LASSERTF(it == lck->rpcl_it, "%p != %p\n", it, lck->rpcl_it);
163 mutex_unlock(&lck->rpcl_mutex);
168 static inline void mdc_get_mod_rpc_slot(struct ptlrpc_request *req,
169 struct lookup_intent *it)
171 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
175 opc = lustre_msg_get_opc(req->rq_reqmsg);
176 tag = obd_get_mod_rpc_slot(cli, opc, it);
177 lustre_msg_set_tag(req->rq_reqmsg, tag);
180 static inline void mdc_put_mod_rpc_slot(struct ptlrpc_request *req,
181 struct lookup_intent *it)
183 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
187 opc = lustre_msg_get_opc(req->rq_reqmsg);
188 tag = lustre_msg_get_tag(req->rq_reqmsg);
189 obd_put_mod_rpc_slot(cli, opc, it, tag);
194 * Update the maximum possible easize.
196 * This value is learned from ptlrpc replies sent by the MDT. The
197 * default easize is initialized to the minimum value but allowed to
198 * grow up to a single page in size if required to handle the common
201 * \see client_obd::cl_default_mds_easize
203 * \param[in] exp export for MDC device
204 * \param[in] body body of ptlrpc reply from MDT
207 static inline void mdc_update_max_ea_from_body(struct obd_export *exp,
208 struct mdt_body *body)
210 if (body->mbo_valid & OBD_MD_FLMODEASIZE) {
211 struct client_obd *cli = &exp->exp_obd->u.cli;
214 if (cli->cl_max_mds_easize < body->mbo_max_mdsize)
215 cli->cl_max_mds_easize = body->mbo_max_mdsize;
217 def_easize = min_t(__u32, body->mbo_max_mdsize,
218 OBD_MAX_DEFAULT_EA_SIZE);
219 cli->cl_default_mds_easize = def_easize;
224 /* mdc/mdc_locks.c */
225 int it_open_error(int phase, struct lookup_intent *it);
227 static inline bool cl_is_lov_delay_create(unsigned int flags)
229 return (flags & O_LOV_DELAY_CREATE_1_8) != 0 ||
230 (flags & O_LOV_DELAY_CREATE_MASK) == O_LOV_DELAY_CREATE_MASK;
233 static inline void cl_lov_delay_create_clear(unsigned int *flags)
235 if ((*flags & O_LOV_DELAY_CREATE_1_8) != 0)
236 *flags &= ~O_LOV_DELAY_CREATE_1_8;
237 if ((*flags & O_LOV_DELAY_CREATE_MASK) == O_LOV_DELAY_CREATE_MASK)
238 *flags &= ~O_LOV_DELAY_CREATE_MASK;