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 <libcfs/kp30.h>
49 #include <lustre_net.h>
50 # include <lustre_lib.h>
52 #include <lustre_ha.h>
53 #include <obd_class.h> /* for obd_zombie */
54 #include <obd_support.h> /* for OBD_FAIL_CHECK */
55 #include <lprocfs_status.h>
57 static struct ptlrpcd_ctl ptlrpcd_pc;
58 static struct ptlrpcd_ctl ptlrpcd_recovery_pc;
60 struct semaphore ptlrpcd_sem;
61 static int ptlrpcd_users = 0;
63 void ptlrpcd_wake(struct ptlrpc_request *req)
65 struct ptlrpc_request_set *rq_set = req->rq_set;
67 LASSERT(rq_set != NULL);
69 cfs_waitq_signal(&rq_set->set_waitq);
73 * Requests that are added to the ptlrpcd queue are sent via
74 * ptlrpcd_check->ptlrpc_check_set().
76 void ptlrpcd_add_req(struct ptlrpc_request *req)
78 struct ptlrpcd_ctl *pc;
81 if (req->rq_send_state == LUSTRE_IMP_FULL)
84 pc = &ptlrpcd_recovery_pc;
85 rc = ptlrpc_set_add_new_req(pc, req);
87 int (*interpreter)(struct ptlrpc_request *,
90 interpreter = req->rq_interpret_reply;
93 * Thread is probably in stop now so we need to
94 * kill this rpc as it was not added. Let's call
95 * interpret for it to let know we're killing it
96 * so that higher levels might free assosiated
99 req->rq_status = -EBADR;
100 interpreter(req, &req->rq_async_args,
103 ptlrpc_req_finished(req);
107 static int ptlrpcd_check(struct ptlrpcd_ctl *pc)
109 struct list_head *tmp, *pos;
110 struct ptlrpc_request *req;
114 spin_lock(&pc->pc_set->set_new_req_lock);
115 list_for_each_safe(pos, tmp, &pc->pc_set->set_new_requests) {
116 req = list_entry(pos, struct ptlrpc_request, rq_set_chain);
117 list_del_init(&req->rq_set_chain);
118 ptlrpc_set_add_req(pc->pc_set, req);
120 * Need to calculate its timeout.
124 spin_unlock(&pc->pc_set->set_new_req_lock);
126 if (pc->pc_set->set_remaining) {
127 rc = rc | ptlrpc_check_set(pc->pc_set);
130 * XXX: our set never completes, so we prune the completed
131 * reqs after each iteration. boy could this be smarter.
133 list_for_each_safe(pos, tmp, &pc->pc_set->set_requests) {
134 req = list_entry(pos, struct ptlrpc_request,
136 if (req->rq_phase != RQ_PHASE_COMPLETE)
139 list_del_init(&req->rq_set_chain);
141 ptlrpc_req_finished (req);
147 * If new requests have been added, make sure to wake up.
149 spin_lock(&pc->pc_set->set_new_req_lock);
150 rc = !list_empty(&pc->pc_set->set_new_requests);
151 spin_unlock(&pc->pc_set->set_new_req_lock);
159 * ptlrpc's code paths like to execute in process context, so we have this
160 * thread which spins on a set which contains the io rpcs. llite specifies
161 * ptlrpcd's set when it pushes pages down into the oscs.
163 static int ptlrpcd(void *arg)
165 struct ptlrpcd_ctl *pc = arg;
169 if ((rc = cfs_daemonize_ctxt(pc->pc_name))) {
170 complete(&pc->pc_starting);
174 complete(&pc->pc_starting);
177 * This mainloop strongly resembles ptlrpc_set_wait() except that our
178 * set never completes. ptlrpcd_check() calls ptlrpc_check_set() when
179 * there are requests in the set. New requests come in on the set's
180 * new_req_list and ptlrpcd_check() moves them into the set.
183 struct l_wait_info lwi;
186 timeout = ptlrpc_set_next_timeout(pc->pc_set);
187 lwi = LWI_TIMEOUT(cfs_time_seconds(timeout ? timeout : 1),
188 ptlrpc_expired_set, pc->pc_set);
190 l_wait_event(pc->pc_set->set_waitq, ptlrpcd_check(pc), &lwi);
193 * Abort inflight rpcs for forced stop case.
195 if (test_bit(LIOD_STOP, &pc->pc_flags)) {
196 if (test_bit(LIOD_FORCE, &pc->pc_flags))
197 ptlrpc_abort_set(pc->pc_set);
202 * Let's make one more loop to make sure that ptlrpcd_check()
203 * copied all raced new rpcs into the set so we can kill them.
208 * Wait for inflight requests to drain.
210 if (!list_empty(&pc->pc_set->set_requests))
211 ptlrpc_set_wait(pc->pc_set);
213 complete(&pc->pc_finishing);
215 clear_bit(LIOD_START, &pc->pc_flags);
216 clear_bit(LIOD_STOP, &pc->pc_flags);
217 clear_bit(LIOD_FORCE, &pc->pc_flags);
223 int ptlrpcd_check_async_rpcs(void *arg)
225 struct ptlrpcd_ctl *pc = arg;
233 if (pc->pc_recurred == 1) {
234 rc = ptlrpcd_check(pc);
236 ptlrpc_expired_set(pc->pc_set);
238 * XXX: send replay requests.
240 if (pc == &ptlrpcd_recovery_pc)
241 rc = ptlrpcd_check(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_bit(LIOD_START, &pc->pc_flags)) {
267 CERROR("Starting second thread (%s) for same pc %p\n",
272 set_bit(LIOD_START, &pc->pc_flags);
273 init_completion(&pc->pc_starting);
274 init_completion(&pc->pc_finishing);
275 spin_lock_init(&pc->pc_lock);
276 snprintf (pc->pc_name, sizeof (pc->pc_name), name);
278 pc->pc_set = ptlrpc_prep_set();
279 if (pc->pc_set == NULL)
280 GOTO(out, rc = -ENOMEM);
283 rc = cfs_kernel_thread(ptlrpcd, pc, 0);
285 ptlrpc_set_destroy(pc->pc_set);
289 wait_for_completion(&pc->pc_starting);
291 pc->pc_wait_callback =
292 liblustre_register_wait_callback("ptlrpcd_check_async_rpcs",
293 &ptlrpcd_check_async_rpcs, pc);
294 pc->pc_idle_callback =
295 liblustre_register_idle_callback("ptlrpcd_check_idle_rpcs",
300 clear_bit(LIOD_START, &pc->pc_flags);
304 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force)
306 if (!test_bit(LIOD_START, &pc->pc_flags)) {
307 CERROR("Thread for pc %p was not started\n", pc);
311 set_bit(LIOD_STOP, &pc->pc_flags);
313 set_bit(LIOD_FORCE, &pc->pc_flags);
314 cfs_waitq_signal(&pc->pc_set->set_waitq);
316 wait_for_completion(&pc->pc_finishing);
318 liblustre_deregister_wait_callback(pc->pc_wait_callback);
319 liblustre_deregister_idle_callback(pc->pc_idle_callback);
321 ptlrpc_set_destroy(pc->pc_set);
324 int ptlrpcd_addref(void)
329 mutex_down(&ptlrpcd_sem);
330 if (++ptlrpcd_users != 1)
333 rc = ptlrpcd_start("ptlrpcd", &ptlrpcd_pc);
339 rc = ptlrpcd_start("ptlrpcd-recov", &ptlrpcd_recovery_pc);
341 ptlrpcd_stop(&ptlrpcd_pc, 0);
346 mutex_up(&ptlrpcd_sem);
350 void ptlrpcd_decref(void)
352 mutex_down(&ptlrpcd_sem);
353 if (--ptlrpcd_users == 0) {
354 ptlrpcd_stop(&ptlrpcd_pc, 0);
355 ptlrpcd_stop(&ptlrpcd_recovery_pc, 0);
357 mutex_up(&ptlrpcd_sem);