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 # include <linux/ctype.h>
39 # include <linux/init.h>
40 #else /* __KERNEL__ */
41 # include <liblustre.h>
45 #include <libcfs/kp30.h>
46 #include <linux/lustre_net.h>
48 #include <linux/lustre_ha.h>
49 #include <linux/obd_support.h> /* for OBD_FAIL_CHECK */
50 #include <linux/lprocfs_status.h>
54 unsigned long pc_flags;
56 struct completion pc_starting;
57 struct completion pc_finishing;
58 struct list_head pc_req_list;
59 wait_queue_head_t pc_waitq;
60 struct ptlrpc_request_set *pc_set;
68 static struct ptlrpcd_ctl ptlrpcd_pc;
69 static struct ptlrpcd_ctl ptlrpcd_recovery_pc;
71 static DECLARE_MUTEX(ptlrpcd_sem);
72 static int ptlrpcd_users = 0;
74 void ptlrpcd_wake(struct ptlrpc_request *req)
76 struct ptlrpcd_ctl *pc = req->rq_ptlrpcd_data;
80 wake_up(&pc->pc_waitq);
83 void ptlrpcd_add_req(struct ptlrpc_request *req)
85 struct ptlrpcd_ctl *pc;
87 if (req->rq_send_state == LUSTRE_IMP_FULL)
90 pc = &ptlrpcd_recovery_pc;
92 do_gettimeofday(&req->rq_rpcd_start);
93 ptlrpc_set_add_new_req(pc->pc_set, req);
94 req->rq_ptlrpcd_data = pc;
99 static int ptlrpcd_check(struct ptlrpcd_ctl *pc)
101 struct list_head *tmp, *pos;
102 struct ptlrpc_request *req;
107 if (test_bit(LIOD_STOP, &pc->pc_flags))
110 spin_lock_irqsave(&pc->pc_set->set_new_req_lock, flags);
111 list_for_each_safe(pos, tmp, &pc->pc_set->set_new_requests) {
112 req = list_entry(pos, struct ptlrpc_request, rq_set_chain);
113 list_del_init(&req->rq_set_chain);
114 ptlrpc_set_add_req(pc->pc_set, req);
115 rc = 1; /* need to calculate its timeout */
117 spin_unlock_irqrestore(&pc->pc_set->set_new_req_lock, flags);
119 if (pc->pc_set->set_remaining) {
120 rc = rc | ptlrpc_check_set(pc->pc_set);
122 /* XXX our set never completes, so we prune the completed
123 * reqs after each iteration. boy could this be smarter. */
124 list_for_each_safe(pos, tmp, &pc->pc_set->set_requests) {
125 req = list_entry(pos, struct ptlrpc_request,
127 if (req->rq_phase != RQ_PHASE_COMPLETE)
130 list_del_init(&req->rq_set_chain);
132 ptlrpc_req_finished (req);
136 /* If new requests have been added, make sure to wake up */
137 spin_lock_irqsave(&pc->pc_set->set_new_req_lock, flags);
138 rc = !list_empty(&pc->pc_set->set_new_requests);
139 spin_unlock_irqrestore(&pc->pc_set->set_new_req_lock, flags);
145 /* ptlrpc's code paths like to execute in process context, so we have this
146 * thread which spins on a set which contains the io rpcs. llite specifies
147 * ptlrpcd's set when it pushes pages down into the oscs */
148 static int ptlrpcd(void *arg)
150 struct ptlrpcd_ctl *pc = arg;
154 kportal_daemonize(pc->pc_name);
156 SIGNAL_MASK_LOCK(current, flags);
157 sigfillset(¤t->blocked);
159 SIGNAL_MASK_UNLOCK(current, flags);
161 complete(&pc->pc_starting);
163 /* this mainloop strongly resembles ptlrpc_set_wait except
164 * that our set never completes. ptlrpcd_check calls ptlrpc_check_set
165 * when there are requests in the set. new requests come in
166 * on the set's new_req_list and ptlrpcd_check moves them into
169 wait_queue_t set_wait;
170 struct l_wait_info lwi;
173 timeout = ptlrpc_set_next_timeout(pc->pc_set) * HZ;
174 lwi = LWI_TIMEOUT(timeout, ptlrpc_expired_set, pc->pc_set);
176 /* ala the pinger, wait on pc's waitqueue and the set's */
177 init_waitqueue_entry(&set_wait, current);
178 add_wait_queue(&pc->pc_set->set_waitq, &set_wait);
179 l_wait_event(pc->pc_waitq, ptlrpcd_check(pc), &lwi);
180 remove_wait_queue(&pc->pc_set->set_waitq, &set_wait);
182 if (test_bit(LIOD_STOP, &pc->pc_flags))
185 /* wait for inflight requests to drain */
186 if (!list_empty(&pc->pc_set->set_requests))
187 ptlrpc_set_wait(pc->pc_set);
188 complete(&pc->pc_finishing);
193 int ptlrpcd_check_async_rpcs(void *arg)
195 struct ptlrpcd_ctl *pc = arg;
198 /* single threaded!! */
201 if (pc->pc_recurred == 1) {
202 rc = ptlrpcd_check(pc);
204 ptlrpc_expired_set(pc->pc_set);
212 static int ptlrpcd_start(char *name, struct ptlrpcd_ctl *pc)
216 memset(pc, 0, sizeof(*pc));
217 init_completion(&pc->pc_starting);
218 init_completion(&pc->pc_finishing);
219 init_waitqueue_head(&pc->pc_waitq);
221 spin_lock_init(&pc->pc_lock);
222 INIT_LIST_HEAD(&pc->pc_req_list);
223 snprintf (pc->pc_name, sizeof (pc->pc_name), name);
225 pc->pc_set = ptlrpc_prep_set();
226 if (pc->pc_set == NULL)
227 GOTO(out, rc = -ENOMEM);
230 if (kernel_thread(ptlrpcd, pc, 0) < 0) {
231 ptlrpc_set_destroy(pc->pc_set);
232 GOTO(out, rc = -ECHILD);
235 wait_for_completion(&pc->pc_starting);
238 liblustre_register_wait_callback(&ptlrpcd_check_async_rpcs, pc);
244 static void ptlrpcd_stop(struct ptlrpcd_ctl *pc)
246 set_bit(LIOD_STOP, &pc->pc_flags);
247 wake_up(&pc->pc_waitq);
249 wait_for_completion(&pc->pc_finishing);
251 liblustre_deregister_wait_callback(pc->pc_callback);
253 ptlrpc_set_destroy(pc->pc_set);
256 int ptlrpcd_addref(void)
262 if (++ptlrpcd_users != 1)
265 rc = ptlrpcd_start("ptlrpcd", &ptlrpcd_pc);
271 rc = ptlrpcd_start("ptlrpcd-recov", &ptlrpcd_recovery_pc);
273 ptlrpcd_stop(&ptlrpcd_pc);
282 void ptlrpcd_decref(void)
285 if (--ptlrpcd_users == 0) {
286 ptlrpcd_stop(&ptlrpcd_pc);
287 ptlrpcd_stop(&ptlrpcd_recovery_pc);