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.
25 #define DEBUG_SUBSYSTEM S_RPC
27 #include <linux/lustre_net.h>
29 extern ptl_handle_eq_t request_out_eq,
34 static ptl_process_id_t local_id = {PTL_ID_ANY, PTL_ID_ANY};
37 int ptlrpc_check_bulk_sent(struct ptlrpc_bulk_desc *bulk)
41 if (bulk->b_flags == PTL_BULK_SENT)
44 if (sigismember(&(current->pending.signal), SIGKILL) ||
45 sigismember(&(current->pending.signal), SIGINT)) {
46 bulk->b_flags = PTL_RPC_INTR;
50 CDEBUG(D_NET, "no event yet\n");
54 static int ptl_send_buf(struct ptlrpc_request *request,
55 struct ptlrpc_connection *conn, int portal)
58 ptl_process_id_t remote_id;
62 request->rq_req_md.user_ptr = request;
64 switch (request->rq_type) {
66 request->rq_req_md.start = request->rq_bulkbuf;
67 request->rq_req_md.length = request->rq_bulklen;
68 request->rq_req_md.eventq = bulk_source_eq;
69 request->rq_req_md.threshold = 2; /* SENT and ACK events */
73 request->rq_req_md.start = request->rq_reqmsg;
74 request->rq_req_md.length = request->rq_reqlen;
75 request->rq_req_md.eventq = request_out_eq;
76 request->rq_req_md.threshold = 1;
80 request->rq_req_md.start = request->rq_repmsg;
81 request->rq_req_md.length = request->rq_replen;
82 request->rq_req_md.eventq = reply_out_eq;
83 request->rq_req_md.threshold = 1;
88 return -1; /* notreached */
90 request->rq_req_md.options = PTL_MD_OP_PUT;
91 request->rq_req_md.user_ptr = request;
93 rc = PtlMDBind(conn->c_peer.peer_ni, request->rq_req_md, &md_h);
94 //CERROR("MDBind (outgoing req/rep/bulk): %Lu\n", (__u64)md_h);
96 CERROR("PtlMDBind failed: %d\n", rc);
101 remote_id.nid = conn->c_peer.peer_nid;
104 CDEBUG(D_NET, "Sending %d bytes to portal %d, xid %d\n",
105 request->rq_req_md.length, portal, request->rq_reqmsg->xid);
107 rc = PtlPut(md_h, ack, remote_id, portal, 0, request->rq_reqmsg->xid,
110 CERROR("PtlPut(%d, %d, %d) failed: %d\n", remote_id.nid,
111 portal, request->rq_reqmsg->xid, rc);
118 int ptlrpc_send_bulk(struct ptlrpc_bulk_desc *bulk, int portal)
121 ptl_process_id_t remote_id;
123 bulk->b_md.start = bulk->b_buf;
124 bulk->b_md.length = bulk->b_buflen;
125 bulk->b_md.eventq = bulk_source_eq;
126 bulk->b_md.threshold = 2; /* SENT and ACK events */
127 bulk->b_md.options = PTL_MD_OP_PUT;
128 bulk->b_md.user_ptr = bulk;
130 rc = PtlMDBind(bulk->b_connection->c_peer.peer_ni, bulk->b_md,
133 CERROR("PtlMDBind failed: %d\n", rc);
138 remote_id.nid = bulk->b_connection->c_peer.peer_nid;
141 CDEBUG(D_NET, "Sending %d bytes to portal %d, xid %d\n",
142 bulk->b_md.length, portal, bulk->b_xid);
144 rc = PtlPut(bulk->b_md_h, PTL_ACK_REQ, remote_id, portal, 0,
147 CERROR("PtlPut(%d, %d, %d) failed: %d\n", remote_id.nid,
148 portal, bulk->b_xid, rc);
149 PtlMDUnlink(bulk->b_md_h);
156 int ptlrpc_register_bulk(struct ptlrpc_bulk_desc *bulk)
161 rc = PtlMEAttach(bulk->b_connection->c_peer.peer_ni, bulk->b_portal,
162 local_id, bulk->b_xid, 0, PTL_UNLINK, PTL_INS_AFTER,
165 CERROR("PtlMEAttach failed: %d\n", rc);
170 bulk->b_md.start = bulk->b_buf;
171 bulk->b_md.length = bulk->b_buflen;
172 bulk->b_md.threshold = 1;
173 bulk->b_md.options = PTL_MD_OP_PUT;
174 bulk->b_md.user_ptr = bulk;
175 bulk->b_md.eventq = bulk_sink_eq;
177 rc = PtlMDAttach(bulk->b_me_h, bulk->b_md, PTL_UNLINK, &bulk->b_md_h);
178 //CERROR("MDAttach (bulk sink): %Lu\n", (__u64)bulk->b_md_h);
180 CERROR("PtlMDAttach failed: %d\n", rc);
185 CDEBUG(D_NET, "Setup bulk sink buffer: %u bytes, xid %u, portal %u\n",
186 bulk->b_buflen, bulk->b_xid, bulk->b_portal);
189 // XXX Confirm that this is safe!
191 PtlMEUnlink(bulk->b_me_h);
195 int ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *bulk)
199 rc = PtlMEUnlink(bulk->b_me_h);
201 CERROR("PtlMEUnlink failed: %d\n", rc);
206 int ptlrpc_reply(struct ptlrpc_service *svc, struct ptlrpc_request *req)
208 /* FIXME: we need to increment the count of handled events */
209 req->rq_type = PTL_RPC_REPLY;
210 req->rq_repmsg->conn = req->rq_connection->c_remote_conn;
211 req->rq_repmsg->token = req->rq_connection->c_remote_token;
212 req->rq_repmsg->xid = HTON__u32(req->rq_reqmsg->xid);
213 req->rq_repmsg->status = HTON__u32(req->rq_status);
214 req->rq_reqmsg->type = HTON__u32(req->rq_type);
215 return ptl_send_buf(req, req->rq_connection, svc->srv_rep_portal);
218 int ptlrpc_error(struct ptlrpc_service *svc, struct ptlrpc_request *req)
223 if (req->rq_repmsg) {
224 CERROR("req already has repmsg\n");
228 rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
232 req->rq_repmsg->type = HTON__u32(PTL_RPC_ERR);
234 rc = ptlrpc_reply(svc, req);
238 int ptl_send_rpc(struct ptlrpc_request *request)
240 ptl_process_id_t local_id;
246 if (NTOH__u32(request->rq_reqmsg->type) != PTL_RPC_REQUEST) {
247 CERROR("wrong packet type sent %d\n",
248 NTOH__u32(request->rq_reqmsg->type));
252 if (request->rq_replen == 0) {
253 CERROR("request->rq_replen is 0!\n");
257 /* request->rq_repmsg is set only when the reply comes in, in
258 * client_packet_callback() */
259 OBD_ALLOC(repbuf, request->rq_replen);
263 local_id.nid = PTL_ID_ANY;
264 local_id.pid = PTL_ID_ANY;
266 down(&request->rq_client->cli_rpc_sem);
268 rc = PtlMEAttach(request->rq_connection->c_peer.peer_ni,
269 request->rq_client->cli_reply_portal,
270 local_id, request->rq_reqmsg->xid, 0, PTL_UNLINK,
271 PTL_INS_AFTER, &request->rq_reply_me_h);
273 CERROR("PtlMEAttach failed: %d\n", rc);
278 request->rq_type = PTL_RPC_REQUEST;
279 request->rq_reply_md.start = repbuf;
280 request->rq_reply_md.length = request->rq_replen;
281 request->rq_reply_md.threshold = 1;
282 request->rq_reply_md.options = PTL_MD_OP_PUT;
283 request->rq_reply_md.user_ptr = request;
284 request->rq_reply_md.eventq = reply_in_eq;
286 rc = PtlMDAttach(request->rq_reply_me_h, request->rq_reply_md,
287 PTL_UNLINK, &request->rq_reply_md_h);
289 CERROR("PtlMDAttach failed: %d\n", rc);
294 CDEBUG(D_NET, "Setup reply buffer: %u bytes, xid %u, portal %u\n",
295 request->rq_replen, request->rq_reqmsg->xid,
296 request->rq_client->cli_request_portal);
298 spin_lock(&request->rq_client->cli_lock);
299 list_add(&request->rq_list, &request->rq_client->cli_sending_head);
300 spin_unlock(&request->rq_client->cli_lock);
302 rc = ptl_send_buf(request, request->rq_connection,
303 request->rq_client->cli_request_portal);
307 PtlMEUnlink(request->rq_reply_me_h);
309 OBD_FREE(repbuf, request->rq_replen);
310 up(&request->rq_client->cli_rpc_sem);
315 void ptlrpc_link_svc_me(struct ptlrpc_service *service, int i)
319 ptl_handle_md_t md_h;
321 /* Attach the leading ME on which we build the ring */
322 rc = PtlMEAttach(service->srv_self.peer_ni, service->srv_req_portal,
323 service->srv_id, 0, ~0, PTL_RETAIN, PTL_INS_BEFORE,
324 &(service->srv_me_h[i]));
326 CERROR("PtlMEAttach failed: %d\n", rc);
330 if (service->srv_ref_count[i])
333 dummy.start = service->srv_buf[i];
334 dummy.length = service->srv_buf_size;
335 dummy.max_offset = service->srv_buf_size;
336 dummy.threshold = PTL_MD_THRESH_INF;
337 dummy.options = PTL_MD_OP_PUT | PTL_MD_AUTO_UNLINK;
338 dummy.user_ptr = service;
339 dummy.eventq = service->srv_eq_h;
340 dummy.max_offset = service->srv_buf_size;
342 rc = PtlMDAttach(service->srv_me_h[i], dummy, PTL_UNLINK, &md_h);
345 CERROR("PtlMDAttach failed: %d\n", rc);
350 /* ptl_handled_rpc() should be called by the sleeping process once
351 * it finishes processing an event. This ensures the ref count is
352 * decremented and that the rpc ring buffer cycles properly.
354 int ptl_handled_rpc(struct ptlrpc_service *service, void *start)
358 spin_lock(&service->srv_lock);
360 for (index = 0; index < service->srv_ring_length; index++)
361 if (service->srv_buf[index] == start)
364 if (index == service->srv_ring_length)
367 CDEBUG(D_INFO, "MD index=%d Ref Count=%d\n", index,
368 service->srv_ref_count[index]);
369 service->srv_ref_count[index]--;
371 if (service->srv_ref_count[index] < 0)
374 if (service->srv_ref_count[index] == 0 &&
375 !ptl_is_valid_handle(&(service->srv_me_h[index]))) {
376 CDEBUG(D_NET, "relinking %d\n", index);
377 ptlrpc_link_svc_me(service, index);
380 spin_unlock(&service->srv_lock);