1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #define DEBUG_SUBSYSTEM S_RPC
25 #include <linux/obd_support.h>
26 #include <linux/obd_class.h>
27 #include <linux/lustre_net.h>
29 extern int request_in_callback(ptl_event_t *ev);
31 static int ptlrpc_check_event(struct ptlrpc_service *svc,
32 struct ptlrpc_thread *thread, ptl_event_t *event)
37 spin_lock(&svc->srv_lock);
38 if (thread->t_flags & SVC_STOPPING)
41 LASSERT ((thread->t_flags & SVC_EVENT) == 0);
43 if (ptl_is_valid_handle(&svc->srv_eq_h)) {
45 err = PtlEQGet(svc->srv_eq_h, event);
48 thread->t_flags |= SVC_EVENT;
52 if (err != PTL_EQ_EMPTY) {
53 CERROR("BUG: PtlEQGet returned %d\n", rc);
62 spin_unlock(&svc->srv_lock);
66 struct ptlrpc_service *
67 ptlrpc_init_svc(__u32 bufsize, int nbuffs, int req_portal, int rep_portal,
68 char *uuid, svc_handler_t handler, char *name)
72 struct ptlrpc_service *service;
75 OBD_ALLOC(service, sizeof(*service));
81 service->srv_name = name;
82 spin_lock_init(&service->srv_lock);
83 INIT_LIST_HEAD(&service->srv_reqs);
84 INIT_LIST_HEAD(&service->srv_threads);
85 init_waitqueue_head(&service->srv_waitq);
87 service->srv_buf_size = bufsize;
88 service->srv_nbuffs = nbuffs;
89 service->srv_rep_portal = rep_portal;
90 service->srv_req_portal = req_portal;
91 service->srv_handler = handler;
93 err = kportal_uuid_to_peer(uuid, &service->srv_self);
95 CERROR("cannot get peer for uuid '%s'\n", uuid);
96 OBD_FREE(service, sizeof(*service));
100 /* NB We need exactly 1 event for each buffer we queue */
101 rc = PtlEQAlloc(service->srv_self.peer_ni, service->srv_nbuffs,
102 request_in_callback, &(service->srv_eq_h));
105 CERROR("PtlEQAlloc failed: %d\n", rc);
107 OBD_FREE(service, sizeof(*service));
111 OBD_ALLOC(service->srv_rqbds, service->srv_nbuffs *
112 sizeof(struct ptlrpc_request_buffer_desc));
113 if (service->srv_rqbds == NULL) {
114 CERROR("no memory\n");
119 for (i = 0; i < service->srv_nbuffs; i++) {
120 struct ptlrpc_request_buffer_desc *rqbd =&service->srv_rqbds[i];
122 rqbd->rqbd_service = service;
123 ptl_set_inv_handle (&rqbd->rqbd_me_h);
124 OBD_ALLOC(rqbd->rqbd_buffer, service->srv_buf_size);
125 if (rqbd->rqbd_buffer == NULL) {
126 CERROR("no memory\n");
130 ptlrpc_link_svc_me(rqbd);
133 CDEBUG(D_NET, "Starting service listening on portal %d\n",
134 service->srv_req_portal);
138 ptlrpc_unregister_service(service);
142 static int handle_incoming_request(struct obd_device *obddev,
143 struct ptlrpc_service *svc,
146 struct ptlrpc_request_buffer_desc *rqbd = event->mem_desc.user_ptr;
147 struct ptlrpc_request request;
150 /* FIXME: If we move to an event-driven model, we should put the request
151 * on the stack of mds_handle instead. */
152 LASSERT ((event->mem_desc.options & PTL_MD_IOV) == 0);
153 LASSERT (rqbd->rqbd_service == svc);
154 LASSERT (rqbd->rqbd_buffer == event->mem_desc.start);
155 LASSERT (event->offset == 0);
157 memset(&request, 0, sizeof(request));
158 request.rq_svc = svc;
159 request.rq_obd = obddev;
160 request.rq_xid = event->match_bits;
161 request.rq_reqmsg = event->mem_desc.start + event->offset;
162 request.rq_reqlen = event->mem_desc.length;
164 if (request.rq_reqlen < sizeof(struct lustre_msg)) {
165 CERROR("incomplete request (%d): ptl %d from "LPX64" xid "LPD64"\n",
166 request.rq_reqlen, svc->srv_req_portal,
167 event->initiator.nid, request.rq_xid);
168 spin_unlock(&svc->srv_lock);
172 if (NTOH__u32(request.rq_reqmsg->type) != PTL_RPC_MSG_REQUEST) {
173 CERROR("wrong packet type received (type=%u)\n",
174 request.rq_reqmsg->type);
176 spin_unlock(&svc->srv_lock);
180 if (request.rq_reqmsg->magic != PTLRPC_MSG_MAGIC) {
181 CERROR("wrong lustre_msg magic %d: ptl %d from "LPX64" xid "LPD64"\n",
182 request.rq_reqmsg->magic, svc->srv_req_portal,
183 event->initiator.nid, request.rq_xid);
184 spin_unlock(&svc->srv_lock);
188 if (request.rq_reqmsg->version != PTLRPC_MSG_VERSION) {
189 CERROR("wrong lustre_msg version %d: ptl %d from "LPX64" xid "LPD64"\n",
190 request.rq_reqmsg->version, svc->srv_req_portal,
191 event->initiator.nid, request.rq_xid);
192 spin_unlock(&svc->srv_lock);
196 CDEBUG(D_NET, "got req "LPD64"\n", request.rq_xid);
198 request.rq_peer.peer_nid = event->initiator.nid;
199 /* FIXME: this NI should be the incoming NI.
200 * We don't know how to find that from here. */
201 request.rq_peer.peer_ni = svc->srv_self.peer_ni;
203 request.rq_export = class_conn2export((struct lustre_handle *)
206 if (request.rq_export) {
207 request.rq_connection = request.rq_export->exp_connection;
208 ptlrpc_connection_addref(request.rq_connection);
211 spin_unlock(&svc->srv_lock);
213 rc = svc->srv_handler(&request);
214 ptlrpc_put_connection(request.rq_connection);
216 ptlrpc_link_svc_me (rqbd);
220 static int ptlrpc_main(void *arg)
223 struct ptlrpc_svc_data *data = (struct ptlrpc_svc_data *)arg;
224 struct obd_device *obddev = data->dev;
225 struct ptlrpc_service *svc = data->svc;
226 struct ptlrpc_thread *thread = data->thread;
232 spin_lock_irq(¤t->sigmask_lock);
233 sigfillset(¤t->blocked);
234 recalc_sigpending(current);
235 spin_unlock_irq(¤t->sigmask_lock);
237 sprintf(current->comm, data->name);
240 /* Record that the thread is running */
241 thread->t_flags = SVC_RUNNING;
242 wake_up(&thread->t_ctl_waitq);
244 /* XXX maintain a list of all managed devices: insert here */
246 /* And now, loop forever on requests */
250 wait_event(svc->srv_waitq,
251 ptlrpc_check_event(svc, thread, &event));
253 spin_lock(&svc->srv_lock);
255 if (thread->t_flags & SVC_STOPPING) {
256 thread->t_flags &= ~SVC_STOPPING;
257 spin_unlock(&svc->srv_lock);
262 if (thread->t_flags & SVC_EVENT) {
263 LASSERT (event.sequence != 0);
264 rc = handle_incoming_request(obddev, svc, &event);
265 thread->t_flags &= ~SVC_EVENT;
269 CERROR("unknown break in service");
270 spin_unlock(&svc->srv_lock);
275 thread->t_flags = SVC_STOPPED;
276 wake_up(&thread->t_ctl_waitq);
277 CDEBUG(D_NET, "service thread exiting, process %d\n", current->pid);
281 static void ptlrpc_stop_thread(struct ptlrpc_service *svc,
282 struct ptlrpc_thread *thread)
284 spin_lock(&svc->srv_lock);
285 thread->t_flags = SVC_STOPPING;
286 spin_unlock(&svc->srv_lock);
288 wake_up(&svc->srv_waitq);
289 wait_event(thread->t_ctl_waitq, (thread->t_flags & SVC_STOPPED));
292 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc)
294 spin_lock(&svc->srv_lock);
295 while (!list_empty(&svc->srv_threads)) {
296 struct ptlrpc_thread *thread;
297 thread = list_entry(svc->srv_threads.next, struct ptlrpc_thread,
299 spin_unlock(&svc->srv_lock);
300 ptlrpc_stop_thread(svc, thread);
301 spin_lock(&svc->srv_lock);
302 list_del(&thread->t_link);
303 OBD_FREE(thread, sizeof(*thread));
305 spin_unlock(&svc->srv_lock);
308 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc,
311 struct ptlrpc_svc_data d;
312 struct ptlrpc_thread *thread;
316 OBD_ALLOC(thread, sizeof(*thread));
317 if (thread == NULL) {
321 init_waitqueue_head(&thread->t_ctl_waitq);
328 spin_lock(&svc->srv_lock);
329 list_add(&thread->t_link, &svc->srv_threads);
330 spin_unlock(&svc->srv_lock);
332 rc = kernel_thread(ptlrpc_main, (void *) &d,
333 CLONE_VM | CLONE_FS | CLONE_FILES);
335 CERROR("cannot start thread\n");
336 OBD_FREE(thread, sizeof(*thread));
339 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_RUNNING);
344 int ptlrpc_unregister_service(struct ptlrpc_service *service)
348 /* NB service->srv_nbuffs gets set before we attempt (and possibly
349 * fail) to allocate srv_rqbds.
351 if (service->srv_rqbds != NULL) {
352 for (i = 0; i < service->srv_nbuffs; i++) {
353 struct ptlrpc_request_buffer_desc *rqbd =
354 &service->srv_rqbds[i];
356 if (rqbd->rqbd_buffer == NULL) /* no buffer allocated */
357 continue; /* => never initialised */
359 /* Buffer allocated => got linked */
360 LASSERT (ptl_is_valid_handle (&rqbd->rqbd_me_h));
362 rc = PtlMEUnlink(rqbd->rqbd_me_h);
364 CERROR("PtlMEUnlink failed: %d\n", rc);
366 OBD_FREE(rqbd->rqbd_buffer, service->srv_buf_size);
369 OBD_FREE(service->srv_rqbds, service->srv_nbuffs *
370 sizeof (struct ptlrpc_request_buffer_desc));
373 rc = PtlEQFree(service->srv_eq_h);
375 CERROR("PtlEQFree failed: %d\n", rc);
377 if (!list_empty(&service->srv_reqs)) {
378 // XXX reply with errors and clean up
379 CERROR("Request list not empty!\n");
383 OBD_FREE(service, sizeof(*service));