1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see [sun.com URL with a
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/ptlrpc/ptlrpcd.c
39 #define DEBUG_SUBSYSTEM S_RPC
42 # include <libcfs/libcfs.h>
43 #else /* __KERNEL__ */
44 # include <liblustre.h>
48 #include <lustre_net.h>
49 # include <lustre_lib.h>
51 #include <lustre_ha.h>
52 #include <obd_class.h> /* for obd_zombie */
53 #include <obd_support.h> /* for OBD_FAIL_CHECK */
54 #include <lprocfs_status.h>
58 unsigned long pc_flags;
60 struct completion pc_starting;
61 struct completion pc_finishing;
62 struct ptlrpc_request_set *pc_set;
67 void *pc_wait_callback;
68 void *pc_idle_callback;
72 static struct ptlrpcd_ctl ptlrpcd_pc;
73 static struct ptlrpcd_ctl ptlrpcd_recovery_pc;
75 struct semaphore ptlrpcd_sem;
76 static int ptlrpcd_users = 0;
78 void ptlrpcd_wake(struct ptlrpc_request *req)
80 struct ptlrpc_request_set *rq_set = req->rq_set;
82 LASSERT(rq_set != NULL);
84 cfs_waitq_signal(&rq_set->set_waitq);
87 /* requests that are added to the ptlrpcd queue are sent via
88 * ptlrpcd_check->ptlrpc_check_set() */
89 void ptlrpcd_add_req(struct ptlrpc_request *req)
91 struct ptlrpcd_ctl *pc;
93 if (req->rq_send_state == LUSTRE_IMP_FULL)
96 pc = &ptlrpcd_recovery_pc;
98 ptlrpc_set_add_new_req(pc->pc_set, req);
99 cfs_waitq_signal(&pc->pc_set->set_waitq);
102 static int ptlrpcd_check(struct ptlrpcd_ctl *pc)
104 struct list_head *tmp, *pos;
105 struct ptlrpc_request *req;
109 if (test_bit(LIOD_STOP, &pc->pc_flags))
112 spin_lock(&pc->pc_set->set_new_req_lock);
113 list_for_each_safe(pos, tmp, &pc->pc_set->set_new_requests) {
114 req = list_entry(pos, struct ptlrpc_request, rq_set_chain);
115 list_del_init(&req->rq_set_chain);
116 ptlrpc_set_add_req(pc->pc_set, req);
117 rc = 1; /* need to calculate its timeout */
119 spin_unlock(&pc->pc_set->set_new_req_lock);
121 if (pc->pc_set->set_remaining) {
122 rc = rc | ptlrpc_check_set(pc->pc_set);
124 /* XXX our set never completes, so we prune the completed
125 * reqs after each iteration. boy could this be smarter. */
126 list_for_each_safe(pos, tmp, &pc->pc_set->set_requests) {
127 req = list_entry(pos, struct ptlrpc_request,
129 if (req->rq_phase != RQ_PHASE_COMPLETE)
132 list_del_init(&req->rq_set_chain);
134 ptlrpc_req_finished (req);
139 /* If new requests have been added, make sure to wake up */
140 spin_lock(&pc->pc_set->set_new_req_lock);
141 rc = !list_empty(&pc->pc_set->set_new_requests);
142 spin_unlock(&pc->pc_set->set_new_req_lock);
149 /* ptlrpc's code paths like to execute in process context, so we have this
150 * thread which spins on a set which contains the io rpcs. llite specifies
151 * ptlrpcd's set when it pushes pages down into the oscs */
152 static int ptlrpcd(void *arg)
154 struct ptlrpcd_ctl *pc = arg;
158 if ((rc = cfs_daemonize_ctxt(pc->pc_name))) {
159 complete(&pc->pc_starting);
163 complete(&pc->pc_starting);
165 /* this mainloop strongly resembles ptlrpc_set_wait except
166 * that our set never completes. ptlrpcd_check calls ptlrpc_check_set
167 * when there are requests in the set. new requests come in
168 * on the set's new_req_list and ptlrpcd_check moves them into
171 struct l_wait_info lwi;
172 cfs_duration_t timeout;
174 timeout = cfs_time_seconds(ptlrpc_set_next_timeout(pc->pc_set));
175 lwi = LWI_TIMEOUT(timeout, ptlrpc_expired_set, pc->pc_set);
177 l_wait_event(pc->pc_set->set_waitq, ptlrpcd_check(pc), &lwi);
179 if (test_bit(LIOD_STOP, &pc->pc_flags))
182 /* wait for inflight requests to drain */
183 if (!list_empty(&pc->pc_set->set_requests))
184 ptlrpc_set_wait(pc->pc_set);
185 complete(&pc->pc_finishing);
191 int ptlrpcd_check_async_rpcs(void *arg)
193 struct ptlrpcd_ctl *pc = arg;
196 /* single threaded!! */
199 if (pc->pc_recurred == 1) {
200 rc = ptlrpcd_check(pc);
202 ptlrpc_expired_set(pc->pc_set);
203 /*XXX send replay requests */
204 if (pc == &ptlrpcd_recovery_pc)
205 rc = ptlrpcd_check(pc);
212 int ptlrpcd_idle(void *arg)
214 struct ptlrpcd_ctl *pc = arg;
216 return (list_empty(&pc->pc_set->set_new_requests) &&
217 pc->pc_set->set_remaining == 0);
222 static int ptlrpcd_start(char *name, struct ptlrpcd_ctl *pc)
227 memset(pc, 0, sizeof(*pc));
228 init_completion(&pc->pc_starting);
229 init_completion(&pc->pc_finishing);
231 spin_lock_init(&pc->pc_lock);
232 snprintf (pc->pc_name, sizeof (pc->pc_name), name);
234 pc->pc_set = ptlrpc_prep_set();
235 if (pc->pc_set == NULL)
239 rc = cfs_kernel_thread(ptlrpcd, pc, 0);
241 ptlrpc_set_destroy(pc->pc_set);
245 wait_for_completion(&pc->pc_starting);
247 pc->pc_wait_callback =
248 liblustre_register_wait_callback("ptlrpcd_check_async_rpcs",
249 &ptlrpcd_check_async_rpcs, pc);
250 pc->pc_idle_callback =
251 liblustre_register_idle_callback("ptlrpcd_check_idle_rpcs",
258 static void ptlrpcd_stop(struct ptlrpcd_ctl *pc)
260 set_bit(LIOD_STOP, &pc->pc_flags);
261 cfs_waitq_signal(&pc->pc_set->set_waitq);
263 wait_for_completion(&pc->pc_finishing);
265 liblustre_deregister_wait_callback(pc->pc_wait_callback);
266 liblustre_deregister_idle_callback(pc->pc_idle_callback);
268 ptlrpc_set_destroy(pc->pc_set);
271 int ptlrpcd_addref(void)
276 mutex_down(&ptlrpcd_sem);
277 if (++ptlrpcd_users != 1)
280 rc = ptlrpcd_start("ptlrpcd", &ptlrpcd_pc);
286 rc = ptlrpcd_start("ptlrpcd-recov", &ptlrpcd_recovery_pc);
288 ptlrpcd_stop(&ptlrpcd_pc);
293 mutex_up(&ptlrpcd_sem);
297 void ptlrpcd_decref(void)
299 mutex_down(&ptlrpcd_sem);
300 if (--ptlrpcd_users == 0) {
301 ptlrpcd_stop(&ptlrpcd_pc);
302 ptlrpcd_stop(&ptlrpcd_recovery_pc);
304 mutex_up(&ptlrpcd_sem);