ofproto/fail-open.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
   3  *
   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:
   7  *
   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.
  15  */
  16
  17 #include <config.h>
  18 #include "fail-open.h"
  19 #include <inttypes.h>
  20 #include <stdlib.h>
  21 #include "classifier.h"
  22 #include "connmgr.h"
  23 #include "flow.h"
  24 #include "mac-learning.h"
  25 #include "odp-util.h"
  26 #include "ofp-util.h"
  27 #include "ofpbuf.h"
  28 #include "ofproto.h"
  29 #include "pktbuf.h"
  30 #include "poll-loop.h"
  31 #include "private.h"
  32 #include "rconn.h"
  33 #include "timeval.h"
  34 #include "vconn.h"
  35 #include "vlog.h"
  36
  37 VLOG_DEFINE_THIS_MODULE(fail_open);
  38
  39 /*
  40  * Fail-open mode.
  41  *
  42  * In fail-open mode, the switch detects when the controller cannot be
  43  * contacted or when the controller is dropping switch connections because the
  44  * switch does not pass its admission control policy.  In those situations the
  45  * switch sets up flows itself using the "normal" action.
  46  *
  47  * There is a little subtlety to implementation, to properly handle the case
  48  * where the controller allows switch connections but drops them a few seconds
  49  * later for admission control reasons.  Because of this case, we don't want to
  50  * just stop setting up flows when we connect to the controller: if we did,
  51  * then new flow setup and existing flows would stop during the duration of
  52  * connection to the controller, and thus the whole network would go down for
  53  * that period of time.
  54  *
  55  * So, instead, we add some special cases when we are connected to a
  56  * controller, but not yet sure that it has admitted us:
  57  *
  58  *     - We set up flows immediately ourselves, but simultaneously send out an
  59  *       OFPT_PACKET_IN to the controller.  We put a special bogus buffer-id in
  60  *       these OFPT_PACKET_IN messages so that duplicate packets don't get sent
  61  *       out to the network when the controller replies.
  62  *
  63  *     - We also send out OFPT_PACKET_IN messages for totally bogus packets
  64  *       every so often, in case no real new flows are arriving in the network.
  65  *
  66  *     - We don't flush the flow table at the time we connect, because this
  67  *       could cause network stuttering in a switch with lots of flows or very
  68  *       high-bandwidth flows by suddenly throwing lots of packets down to
  69  *       userspace.
  70  */
  71
  72 struct fail_open {
  73     struct ofproto *ofproto;
  74     struct connmgr *connmgr;
  75     int last_disconn_secs;
  76     long long int next_bogus_packet_in;
  77     struct rconn_packet_counter *bogus_packet_counter;
  78 };
  79
  80 static void fail_open_recover(struct fail_open *);
  81
  82 /* Returns the number of seconds of disconnection after which fail-open mode
  83  * should activate. */
  84 static int
  85 trigger_duration(const struct fail_open *fo)
  86 {
  87     if (!connmgr_has_controllers(fo->connmgr)) {
  88         /* Shouldn't ever arrive here, but if we do, never fail open. */
  89         return INT_MAX;
  90     } else {
  91         /* Otherwise, every controller must have a chance to send an
  92          * inactivity probe and reconnect before we fail open, so take the
  93          * maximum probe interval and multiply by 3:
  94          *
  95          *  - The first interval is the idle time before sending an inactivity
  96          *    probe.
  97          *
  98          *  - The second interval is the time allowed for a response to the
  99          *    inactivity probe.
 100          *
 101          *  - The third interval is the time allowed to reconnect after no
 102          *    response is received.
 103          */
 104         return connmgr_get_max_probe_interval(fo->connmgr) * 3;
 105     }
 106 }
 107
 108 /* Returns true if 'fo' is currently in fail-open mode, otherwise false. */
 109 bool
 110 fail_open_is_active(const struct fail_open *fo)
 111 {
 112     return fo->last_disconn_secs != 0;
 113 }
 114
 115 static void
 116 send_bogus_packet_ins(struct fail_open *fo)
 117 {
 118     uint8_t mac[ETH_ADDR_LEN];
 119     struct ofpbuf *opi;
 120     struct ofpbuf b;
 121
 122     /* Compose ofp_packet_in. */
 123     ofpbuf_init(&b, 128);
 124     eth_addr_nicira_random(mac);
 125     compose_benign_packet(&b, "Open vSwitch Controller Probe", 0xa033, mac);
 126     opi = make_packet_in(pktbuf_get_null(), OFPP_LOCAL, OFPR_NO_MATCH, &b, 64);
 127     ofpbuf_uninit(&b);
 128
 129     /* Send. */
 130     connmgr_broadcast(fo->connmgr, opi);
 131 }
 132
 133 /* Enter fail-open mode if we should be in it. */
 134 void
 135 fail_open_run(struct fail_open *fo)
 136 {
 137     int disconn_secs = connmgr_failure_duration(fo->connmgr);
 138
 139     /* Enter fail-open mode if 'fo' is not in it but should be.  */
 140     if (disconn_secs >= trigger_duration(fo)) {
 141         if (!fail_open_is_active(fo)) {
 142             VLOG_WARN("Could not connect to controller (or switch failed "
 143                       "controller's post-connection admission control "
 144                       "policy) for %d seconds, failing open", disconn_secs);
 145             fo->last_disconn_secs = disconn_secs;
 146
 147             /* Flush all OpenFlow and datapath flows.  We will set up our
 148              * fail-open rule from fail_open_flushed() when
 149              * ofproto_flush_flows() calls back to us. */
 150             ofproto_flush_flows(fo->ofproto);
 151         } else if (disconn_secs > fo->last_disconn_secs + 60) {
 152             VLOG_INFO("Still in fail-open mode after %d seconds disconnected "
 153                       "from controller", disconn_secs);
 154             fo->last_disconn_secs = disconn_secs;
 155         }
 156     }
 157
 158     /* Schedule a bogus packet-in if we're connected and in fail-open. */
 159     if (fail_open_is_active(fo)) {
 160         if (connmgr_is_any_controller_connected(fo->connmgr)) {
 161             bool expired = time_msec() >= fo->next_bogus_packet_in;
 162             if (expired) {
 163                 send_bogus_packet_ins(fo);
 164             }
 165             if (expired || fo->next_bogus_packet_in == LLONG_MAX) {
 166                 fo->next_bogus_packet_in = time_msec() + 2000;
 167             }
 168         } else {
 169             fo->next_bogus_packet_in = LLONG_MAX;
 170         }
 171     }
 172
 173 }
 174
 175 /* If 'fo' is currently in fail-open mode and its rconn has connected to the
 176  * controller, exits fail open mode. */
 177 void
 178 fail_open_maybe_recover(struct fail_open *fo)
 179 {
 180     if (fail_open_is_active(fo)
 181         && connmgr_is_any_controller_admitted(fo->connmgr)) {
 182         fail_open_recover(fo);
 183     }
 184 }
 185
 186 static void
 187 fail_open_recover(struct fail_open *fo)
 188 {
 189     struct cls_rule rule;
 190
 191     VLOG_WARN("No longer in fail-open mode");
 192     fo->last_disconn_secs = 0;
 193     fo->next_bogus_packet_in = LLONG_MAX;
 194
 195     cls_rule_init_catchall(&rule, FAIL_OPEN_PRIORITY);
 196     ofproto_delete_flow(fo->ofproto, &rule);
 197 }
 198
 199 void
 200 fail_open_wait(struct fail_open *fo)
 201 {
 202     if (fo->next_bogus_packet_in != LLONG_MAX) {
 203         poll_timer_wait_until(fo->next_bogus_packet_in);
 204     }
 205 }
 206
 207 void
 208 fail_open_flushed(struct fail_open *fo)
 209 {
 210     int disconn_secs = connmgr_failure_duration(fo->connmgr);
 211     bool open = disconn_secs >= trigger_duration(fo);
 212     if (open) {
 213         union ofp_action action;
 214         struct cls_rule rule;
 215
 216         /* Set up a flow that matches every packet and directs them to
 217          * OFPP_NORMAL. */
 218         memset(&action, 0, sizeof action);
 219         action.type = htons(OFPAT_OUTPUT);
 220         action.output.len = htons(sizeof action);
 221         action.output.port = htons(OFPP_NORMAL);
 222
 223         cls_rule_init_catchall(&rule, FAIL_OPEN_PRIORITY);
 224         ofproto_add_flow(fo->ofproto, &rule, &action, 1);
 225     }
 226 }
 227
 228 /* Creates and returns a new struct fail_open for 'ofproto' and 'mgr'. */
 229 struct fail_open *
 230 fail_open_create(struct ofproto *ofproto, struct connmgr *mgr)
 231 {
 232     struct fail_open *fo = xmalloc(sizeof *fo);
 233     fo->ofproto = ofproto;
 234     fo->connmgr = mgr;
 235     fo->last_disconn_secs = 0;
 236     fo->next_bogus_packet_in = LLONG_MAX;
 237     fo->bogus_packet_counter = rconn_packet_counter_create();
 238     return fo;
 239 }
 240
 241 /* Destroys 'fo'. */
 242 void
 243 fail_open_destroy(struct fail_open *fo)
 244 {
 245     if (fo) {
 246         if (fail_open_is_active(fo)) {
 247             fail_open_recover(fo);
 248         }
 249         /* We don't own fo->connmgr. */
 250         rconn_packet_counter_destroy(fo->bogus_packet_counter);
 251         free(fo);
 252     }
 253 }