]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
9bfe9334 | 2 | * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2015, 2016 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" | |
cf62fa4c | 26 | #include "ofpbuf.h" |
f25d0cf3 | 27 | #include "ofp-actions.h" |
fa37b408 | 28 | #include "ofp-util.h" |
064af421 | 29 | #include "ofproto.h" |
5bee6e26 | 30 | #include "ofproto-provider.h" |
7778bd15 | 31 | #include "pktbuf.h" |
cf62fa4c | 32 | #include "dp-packet.h" |
7778bd15 | 33 | #include "poll-loop.h" |
064af421 | 34 | #include "rconn.h" |
064af421 | 35 | #include "timeval.h" |
4a1f523f | 36 | #include "openvswitch/vconn.h" |
e6211adc | 37 | #include "openvswitch/vlog.h" |
064af421 | 38 | |
d98e6007 | 39 | VLOG_DEFINE_THIS_MODULE(fail_open); |
5136ce49 | 40 | |
7778bd15 BP |
41 | /* |
42 | * Fail-open mode. | |
43 | * | |
44 | * In fail-open mode, the switch detects when the controller cannot be | |
45 | * contacted or when the controller is dropping switch connections because the | |
46 | * switch does not pass its admission control policy. In those situations the | |
47 | * switch sets up flows itself using the "normal" action. | |
48 | * | |
49 | * There is a little subtlety to implementation, to properly handle the case | |
50 | * where the controller allows switch connections but drops them a few seconds | |
51 | * later for admission control reasons. Because of this case, we don't want to | |
52 | * just stop setting up flows when we connect to the controller: if we did, | |
53 | * then new flow setup and existing flows would stop during the duration of | |
54 | * connection to the controller, and thus the whole network would go down for | |
55 | * that period of time. | |
56 | * | |
4a2026dd BP |
57 | * So, instead, we add some special cases when we are connected to a |
58 | * controller, but not yet sure that it has admitted us: | |
7778bd15 BP |
59 | * |
60 | * - We set up flows immediately ourselves, but simultaneously send out an | |
61 | * OFPT_PACKET_IN to the controller. We put a special bogus buffer-id in | |
62 | * these OFPT_PACKET_IN messages so that duplicate packets don't get sent | |
63 | * out to the network when the controller replies. | |
64 | * | |
65 | * - We also send out OFPT_PACKET_IN messages for totally bogus packets | |
66 | * every so often, in case no real new flows are arriving in the network. | |
67 | * | |
68 | * - We don't flush the flow table at the time we connect, because this | |
69 | * could cause network stuttering in a switch with lots of flows or very | |
70 | * high-bandwidth flows by suddenly throwing lots of packets down to | |
71 | * userspace. | |
72 | */ | |
73 | ||
064af421 BP |
74 | struct fail_open { |
75 | struct ofproto *ofproto; | |
19a87e36 | 76 | struct connmgr *connmgr; |
064af421 | 77 | int last_disconn_secs; |
7778bd15 BP |
78 | long long int next_bogus_packet_in; |
79 | struct rconn_packet_counter *bogus_packet_counter; | |
3fbbcba7 | 80 | bool fail_open_active; |
064af421 BP |
81 | }; |
82 | ||
2f6d3445 BP |
83 | static void fail_open_recover(struct fail_open *); |
84 | ||
76ce9432 BP |
85 | /* Returns the number of seconds of disconnection after which fail-open mode |
86 | * should activate. */ | |
87 | static int | |
88 | trigger_duration(const struct fail_open *fo) | |
064af421 | 89 | { |
19a87e36 | 90 | if (!connmgr_has_controllers(fo->connmgr)) { |
76ce9432 BP |
91 | /* Shouldn't ever arrive here, but if we do, never fail open. */ |
92 | return INT_MAX; | |
93 | } else { | |
94 | /* Otherwise, every controller must have a chance to send an | |
95 | * inactivity probe and reconnect before we fail open, so take the | |
96 | * maximum probe interval and multiply by 3: | |
97 | * | |
98 | * - The first interval is the idle time before sending an inactivity | |
99 | * probe. | |
100 | * | |
101 | * - The second interval is the time allowed for a response to the | |
102 | * inactivity probe. | |
103 | * | |
104 | * - The third interval is the time allowed to reconnect after no | |
105 | * response is received. | |
106 | */ | |
19a87e36 | 107 | return connmgr_get_max_probe_interval(fo->connmgr) * 3; |
76ce9432 | 108 | } |
7778bd15 BP |
109 | } |
110 | ||
111 | /* Returns true if 'fo' is currently in fail-open mode, otherwise false. */ | |
112 | bool | |
113 | fail_open_is_active(const struct fail_open *fo) | |
114 | { | |
115 | return fo->last_disconn_secs != 0; | |
116 | } | |
064af421 | 117 | |
7778bd15 | 118 | static void |
19a87e36 | 119 | send_bogus_packet_ins(struct fail_open *fo) |
7778bd15 | 120 | { |
74ff3298 | 121 | struct eth_addr mac; |
cf62fa4c | 122 | struct dp_packet b; |
064af421 | 123 | |
cf62fa4c | 124 | dp_packet_init(&b, 128); |
74ff3298 | 125 | eth_addr_nicira_random(&mac); |
2ea838ac | 126 | compose_rarp(&b, mac); |
7778bd15 | 127 | |
a2b53dec BP |
128 | struct ofproto_async_msg am = { |
129 | .oam = OAM_PACKET_IN, | |
130 | .pin = { | |
131 | .up = { | |
77ab5fd2 BP |
132 | .public = { |
133 | .packet = dp_packet_data(&b), | |
134 | .packet_len = dp_packet_size(&b), | |
135 | .flow_metadata = MATCH_CATCHALL_INITIALIZER, | |
136 | .flow_metadata.flow.in_port.ofp_port = OFPP_LOCAL, | |
137 | .flow_metadata.wc.masks.in_port.ofp_port | |
a2b53dec | 138 | = u16_to_ofp(UINT16_MAX), |
77ab5fd2 BP |
139 | .reason = OFPR_NO_MATCH, |
140 | .cookie = OVS_BE64_MAX, | |
141 | }, | |
a2b53dec BP |
142 | }, |
143 | .max_len = UINT16_MAX, | |
144 | } | |
9bfe9334 | 145 | }; |
a2b53dec | 146 | connmgr_send_async_msg(fo->connmgr, &am); |
0b0517ef | 147 | |
cf62fa4c | 148 | dp_packet_uninit(&b); |
7778bd15 BP |
149 | } |
150 | ||
76ce9432 | 151 | /* Enter fail-open mode if we should be in it. */ |
7778bd15 BP |
152 | void |
153 | fail_open_run(struct fail_open *fo) | |
154 | { | |
19a87e36 | 155 | int disconn_secs = connmgr_failure_duration(fo->connmgr); |
76ce9432 | 156 | |
7778bd15 | 157 | /* Enter fail-open mode if 'fo' is not in it but should be. */ |
76ce9432 | 158 | if (disconn_secs >= trigger_duration(fo)) { |
7778bd15 | 159 | if (!fail_open_is_active(fo)) { |
5ec6690c JP |
160 | VLOG_WARN("Could not connect to controller (or switch failed " |
161 | "controller's post-connection admission control " | |
162 | "policy) for %d seconds, failing open", disconn_secs); | |
064af421 BP |
163 | fo->last_disconn_secs = disconn_secs; |
164 | ||
165 | /* Flush all OpenFlow and datapath flows. We will set up our | |
166 | * fail-open rule from fail_open_flushed() when | |
167 | * ofproto_flush_flows() calls back to us. */ | |
168 | ofproto_flush_flows(fo->ofproto); | |
7778bd15 BP |
169 | } else if (disconn_secs > fo->last_disconn_secs + 60) { |
170 | VLOG_INFO("Still in fail-open mode after %d seconds disconnected " | |
171 | "from controller", disconn_secs); | |
172 | fo->last_disconn_secs = disconn_secs; | |
064af421 | 173 | } |
064af421 | 174 | } |
7778bd15 BP |
175 | |
176 | /* Schedule a bogus packet-in if we're connected and in fail-open. */ | |
177 | if (fail_open_is_active(fo)) { | |
19a87e36 | 178 | if (connmgr_is_any_controller_connected(fo->connmgr)) { |
7778bd15 BP |
179 | bool expired = time_msec() >= fo->next_bogus_packet_in; |
180 | if (expired) { | |
76ce9432 | 181 | send_bogus_packet_ins(fo); |
7778bd15 BP |
182 | } |
183 | if (expired || fo->next_bogus_packet_in == LLONG_MAX) { | |
184 | fo->next_bogus_packet_in = time_msec() + 2000; | |
185 | } | |
186 | } else { | |
187 | fo->next_bogus_packet_in = LLONG_MAX; | |
188 | } | |
189 | } | |
190 | ||
064af421 BP |
191 | } |
192 | ||
7778bd15 BP |
193 | /* If 'fo' is currently in fail-open mode and its rconn has connected to the |
194 | * controller, exits fail open mode. */ | |
064af421 | 195 | void |
7778bd15 | 196 | fail_open_maybe_recover(struct fail_open *fo) |
feb8a80b | 197 | OVS_EXCLUDED(ofproto_mutex) |
064af421 | 198 | { |
19a87e36 BP |
199 | if (fail_open_is_active(fo) |
200 | && connmgr_is_any_controller_admitted(fo->connmgr)) { | |
2f6d3445 BP |
201 | fail_open_recover(fo); |
202 | } | |
203 | } | |
204 | ||
205 | static void | |
206 | fail_open_recover(struct fail_open *fo) | |
feb8a80b | 207 | OVS_EXCLUDED(ofproto_mutex) |
2f6d3445 | 208 | { |
81a76618 | 209 | struct match match; |
7778bd15 | 210 | |
19a87e36 BP |
211 | VLOG_WARN("No longer in fail-open mode"); |
212 | fo->last_disconn_secs = 0; | |
213 | fo->next_bogus_packet_in = LLONG_MAX; | |
7778bd15 | 214 | |
81a76618 BP |
215 | match_init_catchall(&match); |
216 | ofproto_delete_flow(fo->ofproto, &match, FAIL_OPEN_PRIORITY); | |
7778bd15 BP |
217 | } |
218 | ||
219 | void | |
220 | fail_open_wait(struct fail_open *fo) | |
221 | { | |
222 | if (fo->next_bogus_packet_in != LLONG_MAX) { | |
7cf8b266 | 223 | poll_timer_wait_until(fo->next_bogus_packet_in); |
7778bd15 | 224 | } |
064af421 BP |
225 | } |
226 | ||
227 | void | |
228 | fail_open_flushed(struct fail_open *fo) | |
15aaf599 | 229 | OVS_EXCLUDED(ofproto_mutex) |
064af421 | 230 | { |
19a87e36 | 231 | int disconn_secs = connmgr_failure_duration(fo->connmgr); |
76ce9432 | 232 | bool open = disconn_secs >= trigger_duration(fo); |
064af421 | 233 | if (open) { |
f25d0cf3 | 234 | struct ofpbuf ofpacts; |
81a76618 | 235 | struct match match; |
064af421 BP |
236 | |
237 | /* Set up a flow that matches every packet and directs them to | |
238 | * OFPP_NORMAL. */ | |
f25d0cf3 BP |
239 | ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE); |
240 | ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL; | |
cf3fad8a | 241 | |
81a76618 BP |
242 | match_init_catchall(&match); |
243 | ofproto_add_flow(fo->ofproto, &match, FAIL_OPEN_PRIORITY, | |
6fd6ed71 | 244 | ofpacts.data, ofpacts.size); |
f25d0cf3 BP |
245 | |
246 | ofpbuf_uninit(&ofpacts); | |
064af421 | 247 | } |
3fbbcba7 BP |
248 | fo->fail_open_active = open; |
249 | } | |
250 | ||
251 | /* Returns the number of fail-open rules currently installed in the flow | |
252 | * table. */ | |
253 | int | |
254 | fail_open_count_rules(const struct fail_open *fo) | |
255 | { | |
256 | return fo->fail_open_active != 0; | |
064af421 BP |
257 | } |
258 | ||
19a87e36 | 259 | /* Creates and returns a new struct fail_open for 'ofproto' and 'mgr'. */ |
064af421 | 260 | struct fail_open * |
19a87e36 | 261 | fail_open_create(struct ofproto *ofproto, struct connmgr *mgr) |
064af421 BP |
262 | { |
263 | struct fail_open *fo = xmalloc(sizeof *fo); | |
264 | fo->ofproto = ofproto; | |
19a87e36 | 265 | fo->connmgr = mgr; |
064af421 | 266 | fo->last_disconn_secs = 0; |
7778bd15 BP |
267 | fo->next_bogus_packet_in = LLONG_MAX; |
268 | fo->bogus_packet_counter = rconn_packet_counter_create(); | |
3fbbcba7 | 269 | fo->fail_open_active = false; |
064af421 BP |
270 | return fo; |
271 | } | |
272 | ||
76ce9432 | 273 | /* Destroys 'fo'. */ |
064af421 BP |
274 | void |
275 | fail_open_destroy(struct fail_open *fo) | |
feb8a80b | 276 | OVS_EXCLUDED(ofproto_mutex) |
064af421 BP |
277 | { |
278 | if (fo) { | |
19a87e36 BP |
279 | if (fail_open_is_active(fo)) { |
280 | fail_open_recover(fo); | |
281 | } | |
282 | /* We don't own fo->connmgr. */ | |
7778bd15 | 283 | rconn_packet_counter_destroy(fo->bogus_packet_counter); |
064af421 BP |
284 | free(fo); |
285 | } | |
286 | } |