]> git.proxmox.com Git - ovs.git/blame - ofproto/fail-open.c
hash: Introduce an implementation of murmurhash.
[ovs.git] / ofproto / fail-open.c
CommitLineData
064af421 1/*
e0edde6f 2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
064af421 3 *
a14bc59f
BP
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
064af421 7 *
a14bc59f
BP
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
064af421
BP
15 */
16
17#include <config.h>
18#include "fail-open.h"
19#include <inttypes.h>
20#include <stdlib.h>
cf3fad8a 21#include "classifier.h"
19a87e36 22#include "connmgr.h"
064af421
BP
23#include "flow.h"
24#include "mac-learning.h"
25#include "odp-util.h"
f25d0cf3 26#include "ofp-actions.h"
fa37b408 27#include "ofp-util.h"
7778bd15 28#include "ofpbuf.h"
064af421 29#include "ofproto.h"
5bee6e26 30#include "ofproto-provider.h"
7778bd15
BP
31#include "pktbuf.h"
32#include "poll-loop.h"
064af421 33#include "rconn.h"
064af421 34#include "timeval.h"
7778bd15 35#include "vconn.h"
064af421
BP
36#include "vlog.h"
37
d98e6007 38VLOG_DEFINE_THIS_MODULE(fail_open);
5136ce49 39
7778bd15
BP
40/*
41 * Fail-open mode.
42 *
43 * In fail-open mode, the switch detects when the controller cannot be
44 * contacted or when the controller is dropping switch connections because the
45 * switch does not pass its admission control policy. In those situations the
46 * switch sets up flows itself using the "normal" action.
47 *
48 * There is a little subtlety to implementation, to properly handle the case
49 * where the controller allows switch connections but drops them a few seconds
50 * later for admission control reasons. Because of this case, we don't want to
51 * just stop setting up flows when we connect to the controller: if we did,
52 * then new flow setup and existing flows would stop during the duration of
53 * connection to the controller, and thus the whole network would go down for
54 * that period of time.
55 *
4a2026dd
BP
56 * So, instead, we add some special cases when we are connected to a
57 * controller, but not yet sure that it has admitted us:
7778bd15
BP
58 *
59 * - We set up flows immediately ourselves, but simultaneously send out an
60 * OFPT_PACKET_IN to the controller. We put a special bogus buffer-id in
61 * these OFPT_PACKET_IN messages so that duplicate packets don't get sent
62 * out to the network when the controller replies.
63 *
64 * - We also send out OFPT_PACKET_IN messages for totally bogus packets
65 * every so often, in case no real new flows are arriving in the network.
66 *
67 * - We don't flush the flow table at the time we connect, because this
68 * could cause network stuttering in a switch with lots of flows or very
69 * high-bandwidth flows by suddenly throwing lots of packets down to
70 * userspace.
71 */
72
064af421
BP
73struct fail_open {
74 struct ofproto *ofproto;
19a87e36 75 struct connmgr *connmgr;
064af421 76 int last_disconn_secs;
7778bd15
BP
77 long long int next_bogus_packet_in;
78 struct rconn_packet_counter *bogus_packet_counter;
064af421
BP
79};
80
2f6d3445
BP
81static void fail_open_recover(struct fail_open *);
82
76ce9432
BP
83/* Returns the number of seconds of disconnection after which fail-open mode
84 * should activate. */
85static int
86trigger_duration(const struct fail_open *fo)
064af421 87{
19a87e36 88 if (!connmgr_has_controllers(fo->connmgr)) {
76ce9432
BP
89 /* Shouldn't ever arrive here, but if we do, never fail open. */
90 return INT_MAX;
91 } else {
92 /* Otherwise, every controller must have a chance to send an
93 * inactivity probe and reconnect before we fail open, so take the
94 * maximum probe interval and multiply by 3:
95 *
96 * - The first interval is the idle time before sending an inactivity
97 * probe.
98 *
99 * - The second interval is the time allowed for a response to the
100 * inactivity probe.
101 *
102 * - The third interval is the time allowed to reconnect after no
103 * response is received.
104 */
19a87e36 105 return connmgr_get_max_probe_interval(fo->connmgr) * 3;
76ce9432 106 }
7778bd15
BP
107}
108
109/* Returns true if 'fo' is currently in fail-open mode, otherwise false. */
110bool
111fail_open_is_active(const struct fail_open *fo)
112{
113 return fo->last_disconn_secs != 0;
114}
064af421 115
7778bd15 116static void
19a87e36 117send_bogus_packet_ins(struct fail_open *fo)
7778bd15 118{
0b0517ef 119 struct ofputil_packet_in pin;
7778bd15 120 uint8_t mac[ETH_ADDR_LEN];
7778bd15 121 struct ofpbuf b;
064af421 122
7778bd15 123 ofpbuf_init(&b, 128);
70150daf 124 eth_addr_nicira_random(mac);
2ea838ac 125 compose_rarp(&b, mac);
7778bd15 126
0b0517ef
BP
127 memset(&pin, 0, sizeof pin);
128 pin.packet = b.data;
129 pin.packet_len = b.size;
130 pin.reason = OFPR_NO_MATCH;
131 pin.send_len = b.size;
132 pin.fmd.in_port = OFPP_LOCAL;
133 connmgr_send_packet_in(fo->connmgr, &pin);
134
135 ofpbuf_uninit(&b);
7778bd15
BP
136}
137
76ce9432 138/* Enter fail-open mode if we should be in it. */
7778bd15
BP
139void
140fail_open_run(struct fail_open *fo)
141{
19a87e36 142 int disconn_secs = connmgr_failure_duration(fo->connmgr);
76ce9432 143
7778bd15 144 /* Enter fail-open mode if 'fo' is not in it but should be. */
76ce9432 145 if (disconn_secs >= trigger_duration(fo)) {
7778bd15 146 if (!fail_open_is_active(fo)) {
5ec6690c
JP
147 VLOG_WARN("Could not connect to controller (or switch failed "
148 "controller's post-connection admission control "
149 "policy) for %d seconds, failing open", disconn_secs);
064af421
BP
150 fo->last_disconn_secs = disconn_secs;
151
152 /* Flush all OpenFlow and datapath flows. We will set up our
153 * fail-open rule from fail_open_flushed() when
154 * ofproto_flush_flows() calls back to us. */
155 ofproto_flush_flows(fo->ofproto);
7778bd15
BP
156 } else if (disconn_secs > fo->last_disconn_secs + 60) {
157 VLOG_INFO("Still in fail-open mode after %d seconds disconnected "
158 "from controller", disconn_secs);
159 fo->last_disconn_secs = disconn_secs;
064af421 160 }
064af421 161 }
7778bd15
BP
162
163 /* Schedule a bogus packet-in if we're connected and in fail-open. */
164 if (fail_open_is_active(fo)) {
19a87e36 165 if (connmgr_is_any_controller_connected(fo->connmgr)) {
7778bd15
BP
166 bool expired = time_msec() >= fo->next_bogus_packet_in;
167 if (expired) {
76ce9432 168 send_bogus_packet_ins(fo);
7778bd15
BP
169 }
170 if (expired || fo->next_bogus_packet_in == LLONG_MAX) {
171 fo->next_bogus_packet_in = time_msec() + 2000;
172 }
173 } else {
174 fo->next_bogus_packet_in = LLONG_MAX;
175 }
176 }
177
064af421
BP
178}
179
7778bd15
BP
180/* If 'fo' is currently in fail-open mode and its rconn has connected to the
181 * controller, exits fail open mode. */
064af421 182void
7778bd15 183fail_open_maybe_recover(struct fail_open *fo)
064af421 184{
19a87e36
BP
185 if (fail_open_is_active(fo)
186 && connmgr_is_any_controller_admitted(fo->connmgr)) {
2f6d3445
BP
187 fail_open_recover(fo);
188 }
189}
190
191static void
192fail_open_recover(struct fail_open *fo)
193{
81a76618 194 struct match match;
7778bd15 195
19a87e36
BP
196 VLOG_WARN("No longer in fail-open mode");
197 fo->last_disconn_secs = 0;
198 fo->next_bogus_packet_in = LLONG_MAX;
7778bd15 199
81a76618
BP
200 match_init_catchall(&match);
201 ofproto_delete_flow(fo->ofproto, &match, FAIL_OPEN_PRIORITY);
7778bd15
BP
202}
203
204void
205fail_open_wait(struct fail_open *fo)
206{
207 if (fo->next_bogus_packet_in != LLONG_MAX) {
7cf8b266 208 poll_timer_wait_until(fo->next_bogus_packet_in);
7778bd15 209 }
064af421
BP
210}
211
212void
213fail_open_flushed(struct fail_open *fo)
214{
19a87e36 215 int disconn_secs = connmgr_failure_duration(fo->connmgr);
76ce9432 216 bool open = disconn_secs >= trigger_duration(fo);
064af421 217 if (open) {
f25d0cf3 218 struct ofpbuf ofpacts;
81a76618 219 struct match match;
064af421
BP
220
221 /* Set up a flow that matches every packet and directs them to
222 * OFPP_NORMAL. */
f25d0cf3
BP
223 ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE);
224 ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
225 ofpact_pad(&ofpacts);
cf3fad8a 226
81a76618
BP
227 match_init_catchall(&match);
228 ofproto_add_flow(fo->ofproto, &match, FAIL_OPEN_PRIORITY,
229 ofpacts.data, ofpacts.size);
f25d0cf3
BP
230
231 ofpbuf_uninit(&ofpacts);
064af421
BP
232 }
233}
234
19a87e36 235/* Creates and returns a new struct fail_open for 'ofproto' and 'mgr'. */
064af421 236struct fail_open *
19a87e36 237fail_open_create(struct ofproto *ofproto, struct connmgr *mgr)
064af421
BP
238{
239 struct fail_open *fo = xmalloc(sizeof *fo);
240 fo->ofproto = ofproto;
19a87e36 241 fo->connmgr = mgr;
064af421 242 fo->last_disconn_secs = 0;
7778bd15
BP
243 fo->next_bogus_packet_in = LLONG_MAX;
244 fo->bogus_packet_counter = rconn_packet_counter_create();
064af421
BP
245 return fo;
246}
247
76ce9432 248/* Destroys 'fo'. */
064af421
BP
249void
250fail_open_destroy(struct fail_open *fo)
251{
252 if (fo) {
19a87e36
BP
253 if (fail_open_is_active(fo)) {
254 fail_open_recover(fo);
255 }
256 /* We don't own fo->connmgr. */
7778bd15 257 rconn_packet_counter_destroy(fo->bogus_packet_counter);
064af421
BP
258 free(fo);
259 }
260}