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/lustre_net.h>
27 extern ptl_handle_eq_t request_out_eq, reply_in_eq, reply_out_eq,
28 bulk_source_eq, bulk_sink_eq;
29 static ptl_process_id_t local_id = {PTL_NID_ANY, PTL_PID_ANY};
31 int ptlrpc_check_bulk_sent(struct ptlrpc_bulk_desc *bulk)
35 if (bulk->b_flags & PTL_BULK_FL_SENT)
38 if (sigismember(&(current->pending.signal), SIGKILL) ||
39 sigismember(&(current->pending.signal), SIGINT)) {
40 bulk->b_flags |= PTL_RPC_FL_INTR;
44 CDEBUG(D_NET, "no event yet\n");
48 static int ptl_send_buf(struct ptlrpc_request *request,
49 struct ptlrpc_connection *conn, int portal)
52 ptl_process_id_t remote_id;
56 request->rq_req_md.user_ptr = request;
58 switch (request->rq_type) {
59 case PTL_RPC_TYPE_REQUEST:
60 request->rq_req_md.start = request->rq_reqmsg;
61 request->rq_req_md.length = request->rq_reqlen;
62 request->rq_req_md.eventq = request_out_eq;
63 request->rq_req_md.threshold = 1;
66 case PTL_RPC_TYPE_REPLY:
67 request->rq_req_md.start = request->rq_repmsg;
68 request->rq_req_md.length = request->rq_replen;
69 request->rq_req_md.eventq = reply_out_eq;
70 request->rq_req_md.threshold = 1;
75 return -1; /* notreached */
77 request->rq_req_md.options = PTL_MD_OP_PUT;
78 request->rq_req_md.user_ptr = request;
80 rc = PtlMDBind(conn->c_peer.peer_ni, request->rq_req_md, &md_h);
81 //CERROR("MDBind (outgoing req/rep/bulk): %Lu\n", (__u64)md_h);
83 CERROR("PtlMDBind failed: %d\n", rc);
88 remote_id.nid = conn->c_peer.peer_nid;
91 CDEBUG(D_NET, "Sending %d bytes to portal %d, xid %Ld\n",
92 request->rq_req_md.length, portal, request->rq_xid);
94 rc = PtlPut(md_h, ack, remote_id, portal, 0, request->rq_xid,
97 CERROR("PtlPut(%Lu, %d, %Ld) failed: %d\n", remote_id.nid,
98 portal, request->rq_xid, rc);
105 int ptlrpc_send_bulk(struct ptlrpc_bulk_desc *desc)
108 struct list_head *tmp, *next;
109 ptl_process_id_t remote_id;
112 list_for_each_safe(tmp, next, &desc->b_page_list) {
113 struct ptlrpc_bulk_page *bulk;
114 bulk = list_entry(tmp, struct ptlrpc_bulk_page, b_link);
116 bulk->b_md.start = bulk->b_buf;
117 bulk->b_md.length = bulk->b_buflen;
118 bulk->b_md.eventq = bulk_source_eq;
119 bulk->b_md.threshold = 2; /* SENT and ACK */
120 bulk->b_md.options = PTL_MD_OP_PUT;
121 bulk->b_md.user_ptr = bulk;
123 rc = PtlMDBind(desc->b_connection->c_peer.peer_ni, bulk->b_md,
126 CERROR("PtlMDBind failed: %d\n", rc);
131 remote_id.nid = desc->b_connection->c_peer.peer_nid;
134 CDEBUG(D_NET, "Sending %d bytes to portal %d, xid %d\n",
135 bulk->b_md.length, desc->b_portal, bulk->b_xid);
137 rc = PtlPut(bulk->b_md_h, PTL_ACK_REQ, remote_id,
138 desc->b_portal, 0, bulk->b_xid, 0, 0);
140 CERROR("PtlPut(%Lu, %d, %d) failed: %d\n",
141 remote_id.nid, desc->b_portal, bulk->b_xid, rc);
142 PtlMDUnlink(bulk->b_md_h);
151 int ptlrpc_register_bulk(struct ptlrpc_bulk_desc *desc)
153 struct list_head *tmp, *next;
157 list_for_each_safe(tmp, next, &desc->b_page_list) {
158 struct ptlrpc_bulk_page *bulk;
159 bulk = list_entry(tmp, struct ptlrpc_bulk_page, b_link);
161 rc = PtlMEAttach(desc->b_connection->c_peer.peer_ni,
162 desc->b_portal, local_id, bulk->b_xid, 0,
163 PTL_UNLINK, PTL_INS_AFTER, &bulk->b_me_h);
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,
180 CERROR("PtlMDAttach failed: %d\n", rc);
185 CDEBUG(D_NET, "Setup bulk sink buffer: %u bytes, xid %u, "
186 "portal %u\n", bulk->b_buflen, bulk->b_xid,
193 ptlrpc_abort_bulk(desc);
198 int ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc)
200 struct list_head *tmp, *next;
202 list_for_each_safe(tmp, next, &desc->b_page_list) {
203 struct ptlrpc_bulk_page *bulk;
204 bulk = list_entry(tmp, struct ptlrpc_bulk_page, b_link);
206 /* This should be safe: these handles are initialized to be
207 * invalid in ptlrpc_prep_bulk_page() */
208 PtlMDUnlink(bulk->b_md_h);
209 PtlMEUnlink(bulk->b_me_h);
215 int ptlrpc_reply(struct ptlrpc_service *svc, struct ptlrpc_request *req)
217 /* FIXME: we need to increment the count of handled events */
218 req->rq_type = PTL_RPC_TYPE_REPLY;
219 //req->rq_repmsg->conn = req->rq_connection->c_remote_conn;
220 //req->rq_repmsg->token = req->rq_connection->c_remote_token;
221 req->rq_repmsg->status = HTON__u32(req->rq_status);
222 req->rq_reqmsg->type = HTON__u32(req->rq_type);
223 return ptl_send_buf(req, req->rq_connection, svc->srv_rep_portal);
226 int ptlrpc_error(struct ptlrpc_service *svc, struct ptlrpc_request *req)
231 if (req->rq_repmsg) {
232 CERROR("req already has repmsg\n");
236 rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
240 req->rq_repmsg->type = HTON__u32(PTL_RPC_MSG_ERR);
242 rc = ptlrpc_reply(svc, req);
247 int ptl_send_rpc(struct ptlrpc_request *request)
254 if (NTOH__u32(request->rq_reqmsg->type) != PTL_RPC_MSG_REQUEST) {
255 CERROR("wrong packet type sent %d\n",
256 NTOH__u32(request->rq_reqmsg->type));
260 if (request->rq_replen == 0) {
261 CERROR("request->rq_replen is 0!\n");
265 /* request->rq_repmsg is set only when the reply comes in, in
266 * client_packet_callback() */
267 if (request->rq_reply_md.start)
268 OBD_FREE(request->rq_reply_md.start, request->rq_replen);
270 OBD_ALLOC(repbuf, request->rq_replen);
276 down(&request->rq_client->cli_rpc_sem);
278 rc = PtlMEAttach(request->rq_connection->c_peer.peer_ni,
279 request->rq_client->cli_reply_portal,
280 local_id, request->rq_xid, 0, PTL_UNLINK,
281 PTL_INS_AFTER, &request->rq_reply_me_h);
283 CERROR("PtlMEAttach failed: %d\n", rc);
288 request->rq_type = PTL_RPC_TYPE_REQUEST;
289 request->rq_reply_md.start = repbuf;
290 request->rq_reply_md.length = request->rq_replen;
291 request->rq_reply_md.threshold = 1;
292 request->rq_reply_md.options = PTL_MD_OP_PUT;
293 request->rq_reply_md.user_ptr = request;
294 request->rq_reply_md.eventq = reply_in_eq;
296 rc = PtlMDAttach(request->rq_reply_me_h, request->rq_reply_md,
297 PTL_UNLINK, &request->rq_reply_md_h);
299 CERROR("PtlMDAttach failed: %d\n", rc);
304 CDEBUG(D_NET, "Setup reply buffer: %u bytes, xid %Lu, portal %u\n",
305 request->rq_replen, request->rq_xid,
306 request->rq_client->cli_reply_portal);
308 rc = ptl_send_buf(request, request->rq_connection,
309 request->rq_client->cli_request_portal);
313 PtlMEUnlink(request->rq_reply_me_h);
315 OBD_FREE(repbuf, request->rq_replen);
316 up(&request->rq_client->cli_rpc_sem);
321 void ptlrpc_link_svc_me(struct ptlrpc_service *service, int i)
325 ptl_handle_md_t md_h;
327 /* Attach the leading ME on which we build the ring */
328 rc = PtlMEAttach(service->srv_self.peer_ni, service->srv_req_portal,
329 local_id, 0, ~0, PTL_RETAIN, PTL_INS_BEFORE,
330 &(service->srv_me_h[i]));
332 CERROR("PtlMEAttach failed: %d\n", rc);
336 if (service->srv_ref_count[i])
339 dummy.start = service->srv_buf[i];
340 dummy.length = service->srv_buf_size;
341 dummy.max_offset = service->srv_buf_size;
342 dummy.threshold = PTL_MD_THRESH_INF;
343 dummy.options = PTL_MD_OP_PUT | PTL_MD_AUTO_UNLINK;
344 dummy.user_ptr = service;
345 dummy.eventq = service->srv_eq_h;
346 dummy.max_offset = service->srv_buf_size;
348 rc = PtlMDAttach(service->srv_me_h[i], dummy, PTL_UNLINK, &md_h);
351 CERROR("PtlMDAttach failed: %d\n", rc);
356 /* ptl_handled_rpc() should be called by the sleeping process once
357 * it finishes processing an event. This ensures the ref count is
358 * decremented and that the rpc ring buffer cycles properly.
360 int ptl_handled_rpc(struct ptlrpc_service *service, void *start)
364 spin_lock(&service->srv_lock);
365 for (index = 0; index < service->srv_ring_length; index++)
366 if (service->srv_buf[index] == start)
369 if (index == service->srv_ring_length)
372 CDEBUG(D_INFO, "MD index=%d Ref Count=%d\n", index,
373 service->srv_ref_count[index]);
374 service->srv_ref_count[index]--;
376 if (service->srv_ref_count[index] < 0)
379 if (service->srv_ref_count[index] == 0 &&
380 !ptl_is_valid_handle(&(service->srv_me_h[index]))) {
381 CDEBUG(D_NET, "relinking %d\n", index);
382 ptlrpc_link_svc_me(service, index);
385 spin_unlock(&service->srv_lock);