ofproto/fail-open.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2015, 2016 Nicira, Inc.
   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 <inttypes.h>
  19 #include <stdlib.h>
  20 #include "classifier.h"
  21 #include "connmgr.h"
  22 #include "dp-packet.h"
  23 #include "fail-open.h"
  24 #include "flow.h"
  25 #include "mac-learning.h"
  26 #include "odp-util.h"
  27 #include "openvswitch/ofp-actions.h"
  28 #include "openvswitch/ofpbuf.h"
  29 #include "openvswitch/vconn.h"
  30 #include "openvswitch/vlog.h"
  31 #include "ofproto.h"
  32 #include "ofproto-provider.h"
  33 #include "openvswitch/poll-loop.h"
  34 #include "openvswitch/rconn.h"
  35 #include "timeval.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     bool fail_open_active;
  79 };
  80
  81 static void fail_open_recover(struct fail_open *) OVS_REQUIRES(ofproto_mutex);
  82
  83 /* Returns the number of seconds of disconnection after which fail-open mode
  84  * should activate. */
  85 static int
  86 trigger_duration(const struct fail_open *fo)
  87 {
  88     if (!connmgr_has_controllers(fo->connmgr)) {
  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          */
 105         return connmgr_get_max_probe_interval(fo->connmgr) * 3;
 106     }
 107 }
 108
 109 /* Returns true if 'fo' is currently in fail-open mode, otherwise false. */
 110 bool
 111 fail_open_is_active(const struct fail_open *fo)
 112 {
 113     return fo->last_disconn_secs != 0;
 114 }
 115
 116 static void
 117 send_bogus_packet_ins(struct fail_open *fo)
 118 {
 119     struct eth_addr mac;
 120     struct dp_packet b;
 121
 122     dp_packet_init(&b, 128);
 123     eth_addr_nicira_random(&mac);
 124     compose_rarp(&b, mac);
 125
 126     struct ofproto_async_msg am = {
 127         .oam = OAM_PACKET_IN,
 128         .pin = {
 129             .up = {
 130                 .base = {
 131                     .packet = dp_packet_data(&b),
 132                     .packet_len = dp_packet_size(&b),
 133                     .flow_metadata.flow.in_port.ofp_port = OFPP_LOCAL,
 134                     .flow_metadata.wc.masks.in_port.ofp_port
 135                     = u16_to_ofp(UINT16_MAX),
 136                     .reason = OFPR_NO_MATCH,
 137                     .cookie = OVS_BE64_MAX,
 138                 },
 139             },
 140             .max_len = UINT16_MAX,
 141         }
 142     };
 143     connmgr_send_async_msg(fo->connmgr, &am);
 144
 145     dp_packet_uninit(&b);
 146 }
 147
 148 static void
 149 fail_open_del_normal_flow(struct fail_open *fo)
 150     OVS_REQUIRES(ofproto_mutex)
 151 {
 152     struct match match;
 153
 154     match_init_catchall(&match);
 155     ofproto_delete_flow(fo->ofproto, &match, FAIL_OPEN_PRIORITY);
 156 }
 157
 158 static void
 159 fail_open_add_normal_flow(struct fail_open *fo)
 160 {
 161     struct ofpbuf ofpacts;
 162     struct match match;
 163
 164     /* Set up a flow that matches every packet and directs them to
 165      * OFPP_NORMAL. */
 166     ofpbuf_init(&ofpacts, sizeof(struct ofpact_output));
 167     ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
 168
 169     match_init_catchall(&match);
 170     ofproto_add_flow(fo->ofproto, &match, FAIL_OPEN_PRIORITY,
 171             ofpacts.data, ofpacts.size);
 172
 173     ofpbuf_uninit(&ofpacts);
 174 }
 175
 176 /* Enter fail-open mode if we should be in it. */
 177 void
 178 fail_open_run(struct fail_open *fo)
 179 {
 180     int disconn_secs = connmgr_failure_duration(fo->connmgr);
 181
 182     /* Enter fail-open mode if 'fo' is not in it but should be.  */
 183     if (disconn_secs >= trigger_duration(fo)) {
 184         if (!fail_open_is_active(fo)) {
 185             VLOG_WARN("Could not connect to controller (or switch failed "
 186                       "controller's post-connection admission control "
 187                       "policy) for %d seconds, failing open", disconn_secs);
 188             fo->last_disconn_secs = disconn_secs;
 189
 190             /* Flush all OpenFlow and datapath flows.  We will set up our
 191              * fail-open rule from fail_open_flushed() when
 192              * ofproto_flush_flows() calls back to us. */
 193             ofproto_flush_flows(fo->ofproto);
 194         } else if (disconn_secs > fo->last_disconn_secs + 60) {
 195             VLOG_INFO("Still in fail-open mode after %d seconds disconnected "
 196                       "from controller", disconn_secs);
 197             fo->last_disconn_secs = disconn_secs;
 198         }
 199     }
 200
 201     /* Schedule a bogus packet-in if we're connected and in fail-open. */
 202     if (fail_open_is_active(fo)) {
 203         if (connmgr_is_any_controller_connected(fo->connmgr)) {
 204             bool expired = time_msec() >= fo->next_bogus_packet_in;
 205             if (expired) {
 206                 send_bogus_packet_ins(fo);
 207             }
 208             if (expired || fo->next_bogus_packet_in == LLONG_MAX) {
 209                 fo->next_bogus_packet_in = time_msec() + 2000;
 210             }
 211         } else {
 212             fo->next_bogus_packet_in = LLONG_MAX;
 213         }
 214     }
 215
 216 }
 217
 218 /* If 'fo' is currently in fail-open mode and its rconn has connected to the
 219  * controller, exits fail open mode. */
 220 void
 221 fail_open_maybe_recover(struct fail_open *fo)
 222     OVS_EXCLUDED(ofproto_mutex)
 223 {
 224     if (fail_open_is_active(fo)
 225         && connmgr_is_any_controller_admitted(fo->connmgr)) {
 226         ovs_mutex_lock(&ofproto_mutex);
 227         fail_open_recover(fo);
 228         ovs_mutex_unlock(&ofproto_mutex);
 229     }
 230 }
 231
 232 static void
 233 fail_open_recover(struct fail_open *fo)
 234     OVS_REQUIRES(ofproto_mutex)
 235 {
 236     VLOG_WARN("No longer in fail-open mode");
 237     fo->last_disconn_secs = 0;
 238     fo->next_bogus_packet_in = LLONG_MAX;
 239
 240     fail_open_del_normal_flow(fo);
 241 }
 242
 243 void
 244 fail_open_wait(struct fail_open *fo)
 245 {
 246     if (fo->next_bogus_packet_in != LLONG_MAX) {
 247         poll_timer_wait_until(fo->next_bogus_packet_in);
 248     }
 249 }
 250
 251 void
 252 fail_open_flushed(struct fail_open *fo)
 253     OVS_EXCLUDED(ofproto_mutex)
 254 {
 255     int disconn_secs = connmgr_failure_duration(fo->connmgr);
 256     bool open = disconn_secs >= trigger_duration(fo);
 257     if (open) {
 258         fail_open_add_normal_flow(fo);
 259     }
 260     fo->fail_open_active = open;
 261 }
 262
 263 /* Returns the number of fail-open rules currently installed in the flow
 264  * table. */
 265 int
 266 fail_open_count_rules(const struct fail_open *fo)
 267 {
 268     return fo->fail_open_active != 0;
 269 }
 270
 271 /* Creates and returns a new struct fail_open for 'ofproto' and 'mgr'. */
 272 struct fail_open *
 273 fail_open_create(struct ofproto *ofproto, struct connmgr *mgr)
 274 {
 275     struct fail_open *fo = xmalloc(sizeof *fo);
 276     fo->ofproto = ofproto;
 277     fo->connmgr = mgr;
 278     fo->last_disconn_secs = 0;
 279     fo->next_bogus_packet_in = LLONG_MAX;
 280     fo->bogus_packet_counter = rconn_packet_counter_create();
 281     fo->fail_open_active = false;
 282     return fo;
 283 }
 284
 285 /* Destroys 'fo'. */
 286 void
 287 fail_open_destroy(struct fail_open *fo)
 288     OVS_REQUIRES(ofproto_mutex)
 289 {
 290     if (fo) {
 291         if (fail_open_is_active(fo)) {
 292             fail_open_recover(fo);
 293         }
 294         /* We don't own fo->connmgr. */
 295         rconn_packet_counter_destroy(fo->bogus_packet_counter);
 296         free(fo);
 297     }
 298 }