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
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
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>
56 static struct ptlrpcd_ctl ptlrpcd_pc;
57 static struct ptlrpcd_ctl ptlrpcd_recovery_pc;
59 struct semaphore ptlrpcd_sem;
60 static int ptlrpcd_users = 0;
62 void ptlrpcd_wake(struct ptlrpc_request *req)
64 struct ptlrpc_request_set *rq_set = req->rq_set;
66 LASSERT(rq_set != NULL);
68 cfs_waitq_signal(&rq_set->set_waitq);
72 * Requests that are added to the ptlrpcd queue are sent via
73 * ptlrpcd_check->ptlrpc_check_set().
75 void ptlrpcd_add_req(struct ptlrpc_request *req)
77 struct ptlrpcd_ctl *pc;
80 if (req->rq_send_state == LUSTRE_IMP_FULL)
83 pc = &ptlrpcd_recovery_pc;
85 rc = ptlrpc_set_add_new_req(pc, req);
87 ptlrpc_interpterer_t interpreter;
89 interpreter = req->rq_interpret_reply;
92 * Thread is probably in stop now so we need to
93 * kill this rpc as it was not added. Let's call
94 * interpret for it to let know we're killing it
95 * so that higher levels might free assosiated
98 req->rq_status = -EBADR;
99 interpreter(NULL, req, &req->rq_async_args,
102 ptlrpc_req_finished(req);
106 static int ptlrpcd_check(const struct lu_env *env, struct ptlrpcd_ctl *pc)
108 struct list_head *tmp, *pos;
109 struct ptlrpc_request *req;
113 if (test_bit(LIOD_STOP, &pc->pc_flags))
116 spin_lock(&pc->pc_set->set_new_req_lock);
117 list_for_each_safe(pos, tmp, &pc->pc_set->set_new_requests) {
118 req = list_entry(pos, struct ptlrpc_request, rq_set_chain);
119 list_del_init(&req->rq_set_chain);
120 ptlrpc_set_add_req(pc->pc_set, req);
122 * Need to calculate its timeout.
126 spin_unlock(&pc->pc_set->set_new_req_lock);
128 if (pc->pc_set->set_remaining) {
129 rc = rc | ptlrpc_check_set(env, pc->pc_set);
132 * XXX: our set never completes, so we prune the completed
133 * reqs after each iteration. boy could this be smarter.
135 list_for_each_safe(pos, tmp, &pc->pc_set->set_requests) {
136 req = list_entry(pos, struct ptlrpc_request,
138 if (req->rq_phase != RQ_PHASE_COMPLETE)
141 list_del_init(&req->rq_set_chain);
143 ptlrpc_req_finished (req);
149 * If new requests have been added, make sure to wake up.
151 spin_lock(&pc->pc_set->set_new_req_lock);
152 rc = !list_empty(&pc->pc_set->set_new_requests);
153 spin_unlock(&pc->pc_set->set_new_req_lock);
161 * ptlrpc's code paths like to execute in process context, so we have this
162 * thread which spins on a set which contains the io rpcs. llite specifies
163 * ptlrpcd's set when it pushes pages down into the oscs.
165 static int ptlrpcd(void *arg)
167 struct ptlrpcd_ctl *pc = arg;
171 if ((rc = cfs_daemonize_ctxt(pc->pc_name))) {
172 complete(&pc->pc_starting);
176 complete(&pc->pc_starting);
179 * This mainloop strongly resembles ptlrpc_set_wait() except that our
180 * set never completes. ptlrpcd_check() calls ptlrpc_check_set() when
181 * there are requests in the set. New requests come in on the set's
182 * new_req_list and ptlrpcd_check() moves them into the set.
185 struct l_wait_info lwi;
186 cfs_duration_t timeout;
188 timeout = cfs_time_seconds(ptlrpc_set_next_timeout(pc->pc_set));
189 lwi = LWI_TIMEOUT(timeout, ptlrpc_expired_set, pc->pc_set);
191 lu_context_enter(&pc->pc_env.le_ctx);
192 l_wait_event(pc->pc_set->set_waitq,
193 ptlrpcd_check(&pc->pc_env, pc), &lwi);
194 lu_context_exit(&pc->pc_env.le_ctx);
197 * Abort inflight rpcs for forced stop case.
199 if (test_bit(LIOD_STOP_FORCE, &pc->pc_flags))
200 ptlrpc_abort_set(pc->pc_set);
202 if (test_bit(LIOD_STOP, &pc->pc_flags))
207 * Wait for inflight requests to drain.
209 if (!list_empty(&pc->pc_set->set_requests))
210 ptlrpc_set_wait(pc->pc_set);
212 complete(&pc->pc_finishing);
214 clear_bit(LIOD_START, &pc->pc_flags);
215 clear_bit(LIOD_STOP, &pc->pc_flags);
219 #else /* !__KERNEL__ */
221 int ptlrpcd_check_async_rpcs(void *arg)
223 struct ptlrpcd_ctl *pc = arg;
231 if (pc->pc_recurred == 1) {
232 lu_context_enter(&pc->pc_env.le_ctx);
233 rc = ptlrpcd_check(&pc->pc_env, pc);
234 lu_context_exit(&pc->pc_env.le_ctx);
236 ptlrpc_expired_set(pc->pc_set);
238 * XXX: send replay requests.
240 if (pc == &ptlrpcd_recovery_pc)
241 rc = ptlrpcd_check(&pc->pc_env, pc);
248 int ptlrpcd_idle(void *arg)
250 struct ptlrpcd_ctl *pc = arg;
252 return (list_empty(&pc->pc_set->set_new_requests) &&
253 pc->pc_set->set_remaining == 0);
258 int ptlrpcd_start(char *name, struct ptlrpcd_ctl *pc)
264 * Do not allow start second thread for one pc.
266 if (test_and_set_bit(LIOD_START, &pc->pc_flags)) {
267 CERROR("Starting second thread (%s) for same pc %p\n",
272 init_completion(&pc->pc_starting);
273 init_completion(&pc->pc_finishing);
274 spin_lock_init(&pc->pc_lock);
275 snprintf (pc->pc_name, sizeof (pc->pc_name), name);
276 pc->pc_set = ptlrpc_prep_set();
277 if (pc->pc_set == NULL)
278 GOTO(out, rc = -ENOMEM);
280 * So far only "client" ptlrpcd uses an environment. In the future,
281 * ptlrpcd thread (or a thread-set) has to be given an argument,
282 * describing its "scope".
284 rc = lu_context_init(&pc->pc_env.le_ctx, LCT_CL_THREAD|LCT_REMEMBER);
286 ptlrpc_set_destroy(pc->pc_set);
291 rc = cfs_kernel_thread(ptlrpcd, pc, 0);
293 lu_context_fini(&pc->pc_env.le_ctx);
294 ptlrpc_set_destroy(pc->pc_set);
298 wait_for_completion(&pc->pc_starting);
300 pc->pc_wait_callback =
301 liblustre_register_wait_callback("ptlrpcd_check_async_rpcs",
302 &ptlrpcd_check_async_rpcs, pc);
303 pc->pc_idle_callback =
304 liblustre_register_idle_callback("ptlrpcd_check_idle_rpcs",
309 clear_bit(LIOD_START, &pc->pc_flags);
313 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force)
315 if (!test_bit(LIOD_START, &pc->pc_flags)) {
316 CERROR("Thread for pc %p was not started\n", pc);
320 set_bit(LIOD_STOP, &pc->pc_flags);
322 set_bit(LIOD_STOP_FORCE, &pc->pc_flags);
323 cfs_waitq_signal(&pc->pc_set->set_waitq);
325 wait_for_completion(&pc->pc_finishing);
327 liblustre_deregister_wait_callback(pc->pc_wait_callback);
328 liblustre_deregister_idle_callback(pc->pc_idle_callback);
330 lu_context_fini(&pc->pc_env.le_ctx);
331 ptlrpc_set_destroy(pc->pc_set);
334 int ptlrpcd_addref(void)
339 mutex_down(&ptlrpcd_sem);
340 if (++ptlrpcd_users != 1)
343 rc = ptlrpcd_start("ptlrpcd", &ptlrpcd_pc);
349 rc = ptlrpcd_start("ptlrpcd-recov", &ptlrpcd_recovery_pc);
351 ptlrpcd_stop(&ptlrpcd_pc, 0);
356 mutex_up(&ptlrpcd_sem);
360 void ptlrpcd_decref(void)
362 mutex_down(&ptlrpcd_sem);
363 if (--ptlrpcd_users == 0) {
364 ptlrpcd_stop(&ptlrpcd_pc, 0);
365 ptlrpcd_stop(&ptlrpcd_recovery_pc, 0);
367 mutex_up(&ptlrpcd_sem);