1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001-2003 Cluster File Systems, Inc.
5 * Author Peter Braam <braam@clusterfs.com>
7 * This file is part of Lustre, http://www.lustre.org.
9 * Lustre is free software; you can redistribute it and/or
10 * modify it under the terms of version 2 of the GNU General Public
11 * License as published by the Free Software Foundation.
13 * Lustre is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with Lustre; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_RPC
27 # include <linux/version.h>
28 # include <linux/module.h>
29 # include <linux/mm.h>
30 # include <linux/highmem.h>
31 # include <linux/lustre_dlm.h>
32 # if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
33 # include <linux/workqueue.h>
34 # include <linux/smp_lock.h>
36 # include <linux/locks.h>
38 #else /* __KERNEL__ */
39 # include <liblustre.h>
42 #include <linux/kp30.h>
43 #include <linux/lustre_net.h>
46 # include <linux/init.h>
51 #include <linux/lustre_ha.h>
52 #include <linux/obd_support.h> /* for OBD_FAIL_CHECK */
53 #include <linux/lprocfs_status.h>
57 unsigned long pc_flags;
59 struct completion pc_starting;
60 struct completion pc_finishing;
61 struct list_head pc_req_list;
62 wait_queue_head_t pc_waitq;
63 struct ptlrpc_request_set *pc_set;
71 static struct ptlrpcd_ctl ptlrpcd_pc;
72 static struct ptlrpcd_ctl ptlrpcd_recovery_pc;
74 static DECLARE_MUTEX(ptlrpcd_sem);
75 static int ptlrpcd_users = 0;
77 void ptlrpcd_wake(struct ptlrpc_request *req)
79 struct ptlrpcd_ctl *pc = req->rq_ptlrpcd_data;
83 wake_up(&pc->pc_waitq);
86 void ptlrpcd_add_req(struct ptlrpc_request *req)
88 struct ptlrpcd_ctl *pc;
90 if (req->rq_send_state == LUSTRE_IMP_FULL)
93 pc = &ptlrpcd_recovery_pc;
95 ptlrpc_set_add_new_req(pc->pc_set, req);
96 req->rq_ptlrpcd_data = pc;
101 static int ptlrpcd_check(struct ptlrpcd_ctl *pc)
103 struct list_head *tmp, *pos;
104 struct ptlrpc_request *req;
109 if (test_bit(LIOD_STOP, &pc->pc_flags))
112 spin_lock_irqsave(&pc->pc_set->set_new_req_lock, flags);
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_irqrestore(&pc->pc_set->set_new_req_lock, flags);
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);
142 /* ptlrpc's code paths like to execute in process context, so we have this
143 * thread which spins on a set which contains the io rpcs. llite specifies
144 * ptlrpcd's set when it pushes pages down into the oscs */
145 static int ptlrpcd(void *arg)
147 struct ptlrpcd_ctl *pc = arg;
151 kportal_daemonize(pc->pc_name);
153 SIGNAL_MASK_LOCK(current, flags);
154 sigfillset(¤t->blocked);
156 SIGNAL_MASK_UNLOCK(current, flags);
158 complete(&pc->pc_starting);
160 /* this mainloop strongly resembles ptlrpc_set_wait except
161 * that our set never completes. ptlrpcd_check calls ptlrpc_check_set
162 * when there are requests in the set. new requests come in
163 * on the set's new_req_list and ptlrpcd_check moves them into
166 wait_queue_t set_wait;
167 struct l_wait_info lwi;
170 timeout = ptlrpc_set_next_timeout(pc->pc_set) * HZ;
171 lwi = LWI_TIMEOUT(timeout, ptlrpc_expired_set, pc->pc_set);
173 /* ala the pinger, wait on pc's waitqueue and the set's */
174 init_waitqueue_entry(&set_wait, current);
175 add_wait_queue(&pc->pc_set->set_waitq, &set_wait);
176 l_wait_event(pc->pc_waitq, ptlrpcd_check(pc), &lwi);
177 remove_wait_queue(&pc->pc_set->set_waitq, &set_wait);
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);
190 int ptlrpcd_check_async_rpcs(void *arg)
192 struct ptlrpcd_ctl *pc = arg;
195 /* single threaded!! */
198 if (pc->pc_recurred == 1)
199 rc = ptlrpcd_check(pc);
206 static int ptlrpcd_start(char *name, struct ptlrpcd_ctl *pc)
210 memset(pc, 0, sizeof(*pc));
211 init_completion(&pc->pc_starting);
212 init_completion(&pc->pc_finishing);
213 init_waitqueue_head(&pc->pc_waitq);
215 spin_lock_init(&pc->pc_lock);
216 INIT_LIST_HEAD(&pc->pc_req_list);
217 snprintf (pc->pc_name, sizeof (pc->pc_name), name);
219 pc->pc_set = ptlrpc_prep_set();
220 if (pc->pc_set == NULL)
221 GOTO(out, rc = -ENOMEM);
224 if (kernel_thread(ptlrpcd, pc, 0) < 0) {
225 ptlrpc_set_destroy(pc->pc_set);
226 GOTO(out, rc = -ECHILD);
229 wait_for_completion(&pc->pc_starting);
232 liblustre_register_wait_callback(&ptlrpcd_check_async_rpcs, pc);
238 static void ptlrpcd_stop(struct ptlrpcd_ctl *pc)
240 set_bit(LIOD_STOP, &pc->pc_flags);
241 wake_up(&pc->pc_waitq);
243 wait_for_completion(&pc->pc_finishing);
245 liblustre_deregister_wait_callback(pc->pc_callback);
247 ptlrpc_set_destroy(pc->pc_set);
250 int ptlrpcd_addref(void)
256 if (++ptlrpcd_users != 1)
259 rc = ptlrpcd_start("ptlrpcd", &ptlrpcd_pc);
265 rc = ptlrpcd_start("ptlrpcd-recov", &ptlrpcd_recovery_pc);
267 ptlrpcd_stop(&ptlrpcd_pc);
276 void ptlrpcd_decref(void)
279 if (--ptlrpcd_users == 0) {
280 ptlrpcd_stop(&ptlrpcd_pc);
281 ptlrpcd_stop(&ptlrpcd_recovery_pc);