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/ofp-util.h"
  29 #include "openvswitch/ofpbuf.h"
  30 #include "openvswitch/vconn.h"
  31 #include "openvswitch/vlog.h"
  32 #include "ofproto.h"
  33 #include "ofproto-provider.h"
  34 #include "openvswitch/poll-loop.h"
  35 #include "openvswitch/rconn.h"
  36 #include "timeval.h"
  37
  38 VLOG_DEFINE_THIS_MODULE(fail_open);
  39
  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  *
  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:
  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
  73 struct fail_open {
  74     struct ofproto *ofproto;
  75     struct connmgr *connmgr;
  76     int last_disconn_secs;
  77     long long int next_bogus_packet_in;
  78     struct rconn_packet_counter *bogus_packet_counter;
  79     bool fail_open_active;
  80 };
  81
  82 static void fail_open_recover(struct fail_open *) OVS_REQUIRES(ofproto_mutex);
  83
  84 /* Returns the number of seconds of disconnection after which fail-open mode
  85  * should activate. */
  86 static int
  87 trigger_duration(const struct fail_open *fo)
  88 {
  89     if (!connmgr_has_controllers(fo->connmgr)) {
  90         /* Shouldn't ever arrive here, but if we do, never fail open. */
  91         return INT_MAX;
  92     } else {
  93         /* Otherwise, every controller must have a chance to send an
  94          * inactivity probe and reconnect before we fail open, so take the
  95          * maximum probe interval and multiply by 3:
  96          *
  97          *  - The first interval is the idle time before sending an inactivity
  98          *    probe.
  99          *
 100          *  - The second interval is the time allowed for a response to the
 101          *    inactivity probe.
 102          *
 103          *  - The third interval is the time allowed to reconnect after no
 104          *    response is received.
 105          */
 106         return connmgr_get_max_probe_interval(fo->connmgr) * 3;
 107     }
 108 }
 109
 110 /* Returns true if 'fo' is currently in fail-open mode, otherwise false. */
 111 bool
 112 fail_open_is_active(const struct fail_open *fo)
 113 {
 114     return fo->last_disconn_secs != 0;
 115 }
 116
 117 static void
 118 send_bogus_packet_ins(struct fail_open *fo)
 119 {
 120     struct eth_addr mac;
 121     struct dp_packet b;
 122
 123     dp_packet_init(&b, 128);
 124     eth_addr_nicira_random(&mac);
 125     compose_rarp(&b, mac);
 126
 127     struct ofproto_async_msg am = {
 128         .oam = OAM_PACKET_IN,
 129         .pin = {
 130             .up = {
 131                 .base = {
 132                     .packet = dp_packet_data(&b),
 133                     .packet_len = dp_packet_size(&b),
 134                     .flow_metadata.flow.in_port.ofp_port = OFPP_LOCAL,
 135                     .flow_metadata.wc.masks.in_port.ofp_port
 136                     = u16_to_ofp(UINT16_MAX),
 137                     .reason = OFPR_NO_MATCH,
 138                     .cookie = OVS_BE64_MAX,
 139                 },
 140             },
 141             .max_len = UINT16_MAX,
 142         }
 143     };
 144     connmgr_send_async_msg(fo->connmgr, &am);
 145
 146     dp_packet_uninit(&b);
 147 }
 148
 149 /* Enter fail-open mode if we should be in it. */
 150 void
 151 fail_open_run(struct fail_open *fo)
 152 {
 153     int disconn_secs = connmgr_failure_duration(fo->connmgr);
 154
 155     /* Enter fail-open mode if 'fo' is not in it but should be.  */
 156     if (disconn_secs >= trigger_duration(fo)) {
 157         if (!fail_open_is_active(fo)) {
 158             VLOG_WARN("Could not connect to controller (or switch failed "
 159                       "controller's post-connection admission control "
 160                       "policy) for %d seconds, failing open", disconn_secs);
 161             fo->last_disconn_secs = disconn_secs;
 162
 163             /* Flush all OpenFlow and datapath flows.  We will set up our
 164              * fail-open rule from fail_open_flushed() when
 165              * ofproto_flush_flows() calls back to us. */
 166             ofproto_flush_flows(fo->ofproto);
 167         } else if (disconn_secs > fo->last_disconn_secs + 60) {
 168             VLOG_INFO("Still in fail-open mode after %d seconds disconnected "
 169                       "from controller", disconn_secs);
 170             fo->last_disconn_secs = disconn_secs;
 171         }
 172     }
 173
 174     /* Schedule a bogus packet-in if we're connected and in fail-open. */
 175     if (fail_open_is_active(fo)) {
 176         if (connmgr_is_any_controller_connected(fo->connmgr)) {
 177             bool expired = time_msec() >= fo->next_bogus_packet_in;
 178             if (expired) {
 179                 send_bogus_packet_ins(fo);
 180             }
 181             if (expired || fo->next_bogus_packet_in == LLONG_MAX) {
 182                 fo->next_bogus_packet_in = time_msec() + 2000;
 183             }
 184         } else {
 185             fo->next_bogus_packet_in = LLONG_MAX;
 186         }
 187     }
 188
 189 }
 190
 191 /* If 'fo' is currently in fail-open mode and its rconn has connected to the
 192  * controller, exits fail open mode. */
 193 void
 194 fail_open_maybe_recover(struct fail_open *fo)
 195     OVS_EXCLUDED(ofproto_mutex)
 196 {
 197     if (fail_open_is_active(fo)
 198         && connmgr_is_any_controller_admitted(fo->connmgr)) {
 199         ovs_mutex_lock(&ofproto_mutex);
 200         fail_open_recover(fo);
 201         ovs_mutex_unlock(&ofproto_mutex);
 202     }
 203 }
 204
 205 static void
 206 fail_open_recover(struct fail_open *fo)
 207     OVS_REQUIRES(ofproto_mutex)
 208 {
 209     struct match match;
 210
 211     VLOG_WARN("No longer in fail-open mode");
 212     fo->last_disconn_secs = 0;
 213     fo->next_bogus_packet_in = LLONG_MAX;
 214
 215     match_init_catchall(&match);
 216     ofproto_delete_flow(fo->ofproto, &match, FAIL_OPEN_PRIORITY);
 217 }
 218
 219 void
 220 fail_open_wait(struct fail_open *fo)
 221 {
 222     if (fo->next_bogus_packet_in != LLONG_MAX) {
 223         poll_timer_wait_until(fo->next_bogus_packet_in);
 224     }
 225 }
 226
 227 void
 228 fail_open_flushed(struct fail_open *fo)
 229     OVS_EXCLUDED(ofproto_mutex)
 230 {
 231     int disconn_secs = connmgr_failure_duration(fo->connmgr);
 232     bool open = disconn_secs >= trigger_duration(fo);
 233     if (open) {
 234         struct ofpbuf ofpacts;
 235         struct match match;
 236
 237         /* Set up a flow that matches every packet and directs them to
 238          * OFPP_NORMAL. */
 239         ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE);
 240         ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
 241
 242         match_init_catchall(&match);
 243         ofproto_add_flow(fo->ofproto, &match, FAIL_OPEN_PRIORITY,
 244                          ofpacts.data, ofpacts.size);
 245
 246         ofpbuf_uninit(&ofpacts);
 247     }
 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;
 257 }
 258
 259 /* Creates and returns a new struct fail_open for 'ofproto' and 'mgr'. */
 260 struct fail_open *
 261 fail_open_create(struct ofproto *ofproto, struct connmgr *mgr)
 262 {
 263     struct fail_open *fo = xmalloc(sizeof *fo);
 264     fo->ofproto = ofproto;
 265     fo->connmgr = mgr;
 266     fo->last_disconn_secs = 0;
 267     fo->next_bogus_packet_in = LLONG_MAX;
 268     fo->bogus_packet_counter = rconn_packet_counter_create();
 269     fo->fail_open_active = false;
 270     return fo;
 271 }
 272
 273 /* Destroys 'fo'. */
 274 void
 275 fail_open_destroy(struct fail_open *fo)
 276     OVS_REQUIRES(ofproto_mutex)
 277 {
 278     if (fo) {
 279         if (fail_open_is_active(fo)) {
 280             fail_open_recover(fo);
 281         }
 282         /* We don't own fo->connmgr. */
 283         rconn_packet_counter_destroy(fo->bogus_packet_counter);
 284         free(fo);
 285     }
 286 }