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 #include <linux/config.h>
26 #include <linux/module.h>
27 #include <linux/kernel.h>
29 #define DEBUG_SUBSYSTEM S_RPC
31 #include <linux/obd_support.h>
32 #include <linux/obd_class.h>
33 #include <linux/lustre_net.h>
35 ptl_handle_eq_t sent_pkt_eq, rcvd_rep_eq, bulk_source_eq, bulk_sink_eq;
36 static const ptl_handle_ni_t *socknal_nip = NULL, *qswnal_nip = NULL;
39 * Free the packet when it has gone out
41 static int sent_packet_callback(ptl_event_t *ev, void *data)
47 PtlEQGet(sent_pkt_eq, &junk_ev);
49 if (ev->type == PTL_EVENT_SENT) {
50 OBD_FREE(ev->mem_desc.start, ev->mem_desc.length);
52 // XXX make sure we understand all events, including ACK's
53 CERROR("Unknown event %d\n", ev->type);
62 * Wake up the thread waiting for the reply once it comes in.
64 static int rcvd_reply_callback(ptl_event_t *ev, void *data)
66 struct ptlrpc_request *rpc = ev->mem_desc.user_ptr;
69 if (ev->type == PTL_EVENT_PUT) {
70 rpc->rq_repbuf = ev->mem_desc.start + ev->offset;
72 wake_up_interruptible(&rpc->rq_wait_for_rep);
74 // XXX make sure we understand all events, including ACK's
75 CERROR("Unknown event %d\n", ev->type);
83 int server_request_callback(ptl_event_t *ev, void *data)
85 struct ptlrpc_service *service = data;
88 if (ev->rlength != ev->mlength)
89 CERROR("Warning: Possibly truncated rpc (%d/%d)\n",
90 ev->mlength, ev->rlength);
92 /* The ME is unlinked when there is less than 1024 bytes free
93 * on its MD. This ensures we are always able to handle the rpc,
94 * although the 1024 value is a guess as to the size of a
95 * large rpc (the known safe margin should be determined).
97 * NOTE: The portals API by default unlinks all MD's associated
98 * with an ME when it's unlinked. For now, this behavior
99 * has been commented out of the portals library so the
100 * MD can be unlinked when its ref count drops to zero.
101 * A new MD and ME will then be created that use the same
102 * kmalloc()'ed memory and inserted at the ring tail.
105 spin_lock(&service->srv_lock);
106 if ( ev->mem_desc.start !=
107 service->srv_md[service->srv_md_active].start ) {
111 service->srv_ref_count[service->srv_md_active]++;
112 CDEBUG(D_INODE, "event offset %d buf size %d\n",
113 ev->offset, service->srv_buf_size);
114 if (ev->offset >= (service->srv_buf_size - 1024)) {
115 CDEBUG(D_INODE, "Unlinking ME %d\n", service->srv_md_active);
117 rc = PtlMEUnlink(service->srv_me_h[service->srv_md_active]);
118 service->srv_me_h[service->srv_md_active] = 0;
121 CERROR("PtlMEUnlink failed - DROPPING soon: %d\n", rc);
123 spin_unlock(&service->srv_lock);
127 service->srv_md_active = (service->srv_md_active + 1) %
128 service->srv_ring_length;
130 if (service->srv_me_h[service->srv_md_active] == 0) {
131 CERROR("All %d ring ME's are unlinked!\n",
132 service->srv_ring_length);
137 spin_unlock(&service->srv_lock);
138 if (ev->type == PTL_EVENT_PUT) {
139 wake_up(&service->srv_waitq);
141 CERROR("Unexpected event type: %d\n", ev->type);
147 static int bulk_source_callback(ptl_event_t *ev, void *data)
149 struct ptlrpc_bulk_desc *bulk = ev->mem_desc.user_ptr;
153 if (ev->type == PTL_EVENT_SENT) {
154 CDEBUG(D_NET, "got SENT event\n");
155 } else if (ev->type == PTL_EVENT_ACK) {
156 CDEBUG(D_NET, "got ACK event\n");
157 bulk->b_flags = PTL_BULK_SENT;
158 wake_up_interruptible(&bulk->b_waitq);
160 CERROR("Unexpected event type!\n");
168 static int bulk_sink_callback(ptl_event_t *ev, void *data)
170 struct ptlrpc_bulk_desc *bulk = ev->mem_desc.user_ptr;
174 if (ev->type == PTL_EVENT_PUT) {
175 if (bulk->b_buf != ev->mem_desc.start + ev->offset)
176 CERROR("bulkbuf != mem_desc -- why?\n");
177 bulk->b_flags = PTL_BULK_RCVD;
178 if (bulk->b_cb != NULL)
179 bulk->b_cb(bulk, data);
180 wake_up_interruptible(&bulk->b_waitq);
182 CERROR("Unexpected event type!\n");
186 /* FIXME: This should happen unconditionally */
187 if (bulk->b_cb != NULL) {
188 OBD_FREE(bulk, sizeof(*bulk));
195 int ptlrpc_init_portals(void)
200 socknal_nip = inter_module_get_request("ksocknal_ni", "ksocknal");
201 qswnal_nip = inter_module_get_request("kqswnal_ni", "kqswnal");
202 if (socknal_nip == NULL && qswnal_nip == NULL) {
203 CERROR("get_ni failed: is a NAL module loaded?\n");
207 /* Use the qswnal if it's there */
208 if (qswnal_nip != NULL)
213 rc = PtlEQAlloc(ni, 128, sent_packet_callback, NULL, &sent_pkt_eq);
215 CERROR("PtlEQAlloc failed: %d\n", rc);
217 rc = PtlEQAlloc(ni, 128, rcvd_reply_callback, NULL, &rcvd_rep_eq);
219 CERROR("PtlEQAlloc failed: %d\n", rc);
221 rc = PtlEQAlloc(ni, 128, bulk_source_callback, NULL, &bulk_source_eq);
223 CERROR("PtlEQAlloc failed: %d\n", rc);
225 rc = PtlEQAlloc(ni, 128, bulk_sink_callback, NULL, &bulk_sink_eq);
227 CERROR("PtlEQAlloc failed: %d\n", rc);
232 void ptlrpc_exit_portals(void)
234 PtlEQFree(sent_pkt_eq);
235 PtlEQFree(rcvd_rep_eq);
236 PtlEQFree(bulk_source_eq);
237 PtlEQFree(bulk_sink_eq);
239 if (qswnal_nip != NULL)
240 inter_module_put("kqswnal_ni");
241 if (socknal_nip != NULL)
242 inter_module_put("ksocknal_ni");