ofproto/netflow.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 "netflow.h"
  19 #include <arpa/inet.h>
  20 #include <errno.h>
  21 #include <stdlib.h>
  22 #include <unistd.h>
  23 #include "byte-order.h"
  24 #include "collectors.h"
  25 #include "flow.h"
  26 #include "netflow.h"
  27 #include "ofpbuf.h"
  28 #include "ofproto.h"
  29 #include "packets.h"
  30 #include "socket-util.h"
  31 #include "timeval.h"
  32 #include "util.h"
  33 #include "vlog.h"
  34
  35 VLOG_DEFINE_THIS_MODULE(netflow);
  36
  37 #define NETFLOW_V5_VERSION 5
  38
  39 /* Every NetFlow v5 message contains the header that follows.  This is
  40  * followed by up to thirty records that describe a terminating flow.
  41  * We only send a single record per NetFlow message.
  42  */
  43 struct netflow_v5_header {
  44     ovs_be16 version;              /* NetFlow version is 5. */
  45     ovs_be16 count;                /* Number of records in this message. */
  46     ovs_be32 sysuptime;            /* System uptime in milliseconds. */
  47     ovs_be32 unix_secs;            /* Number of seconds since Unix epoch. */
  48     ovs_be32 unix_nsecs;           /* Number of residual nanoseconds
  49                                       after epoch seconds. */
  50     ovs_be32 flow_seq;             /* Number of flows since sending
  51                                       messages began. */
  52     uint8_t  engine_type;          /* Engine type. */
  53     uint8_t  engine_id;            /* Engine id. */
  54     ovs_be16 sampling_interval;    /* Set to zero. */
  55 };
  56 BUILD_ASSERT_DECL(sizeof(struct netflow_v5_header) == 24);
  57
  58 /* A NetFlow v5 description of a terminating flow.  It is preceded by a
  59  * NetFlow v5 header.
  60  */
  61 struct netflow_v5_record {
  62     ovs_be32 src_addr;             /* Source IP address. */
  63     ovs_be32 dst_addr;             /* Destination IP address. */
  64     ovs_be32 nexthop;              /* IP address of next hop.  Set to 0. */
  65     ovs_be16 input;                /* Input interface index. */
  66     ovs_be16 output;               /* Output interface index. */
  67     ovs_be32 packet_count;         /* Number of packets. */
  68     ovs_be32 byte_count;           /* Number of bytes. */
  69     ovs_be32 init_time;            /* Value of sysuptime on first packet. */
  70     ovs_be32 used_time;            /* Value of sysuptime on last packet. */
  71
  72     /* The 'src_port' and 'dst_port' identify the source and destination
  73      * port, respectively, for TCP and UDP.  For ICMP, the high-order
  74      * byte identifies the type and low-order byte identifies the code
  75      * in the 'dst_port' field. */
  76     ovs_be16 src_port;
  77     ovs_be16 dst_port;
  78
  79     uint8_t  pad1;
  80     uint8_t  tcp_flags;            /* Union of seen TCP flags. */
  81     uint8_t  ip_proto;             /* IP protocol. */
  82     uint8_t  ip_tos;               /* IP TOS value. */
  83     ovs_be16 src_as;               /* Source AS ID.  Set to 0. */
  84     ovs_be16 dst_as;               /* Destination AS ID.  Set to 0. */
  85     uint8_t  src_mask;             /* Source mask bits.  Set to 0. */
  86     uint8_t  dst_mask;             /* Destination mask bits.  Set to 0. */
  87     uint8_t  pad[2];
  88 };
  89 BUILD_ASSERT_DECL(sizeof(struct netflow_v5_record) == 48);
  90
  91 struct netflow {
  92     uint8_t engine_type;          /* Value of engine_type to use. */
  93     uint8_t engine_id;            /* Value of engine_id to use. */
  94     long long int boot_time;      /* Time when netflow_create() was called. */
  95     struct collectors *collectors; /* NetFlow collectors. */
  96     bool add_id_to_iface;         /* Put the 7 least signficiant bits of
  97                                    * 'engine_id' into the most signficant
  98                                    * bits of the interface fields. */
  99     uint32_t netflow_cnt;         /* Flow sequence number for NetFlow. */
 100     struct ofpbuf packet;         /* NetFlow packet being accumulated. */
 101     long long int active_timeout; /* Timeout for flows that are still active. */
 102     long long int reconfig_time;  /* When we reconfigured the timeouts. */
 103 };
 104
 105 static void
 106 gen_netflow_rec(struct netflow *nf, struct netflow_flow *nf_flow,
 107                 struct ofexpired *expired,
 108                 uint32_t packet_count, uint32_t byte_count)
 109 {
 110     struct netflow_v5_header *nf_hdr;
 111     struct netflow_v5_record *nf_rec;
 112
 113     if (!nf->packet.size) {
 114         struct timespec now;
 115
 116         time_wall_timespec(&now);
 117
 118         nf_hdr = ofpbuf_put_zeros(&nf->packet, sizeof *nf_hdr);
 119         nf_hdr->version = htons(NETFLOW_V5_VERSION);
 120         nf_hdr->count = htons(0);
 121         nf_hdr->sysuptime = htonl(time_msec() - nf->boot_time);
 122         nf_hdr->unix_secs = htonl(now.tv_sec);
 123         nf_hdr->unix_nsecs = htonl(now.tv_nsec);
 124         nf_hdr->flow_seq = htonl(nf->netflow_cnt++);
 125         nf_hdr->engine_type = nf->engine_type;
 126         nf_hdr->engine_id = nf->engine_id;
 127         nf_hdr->sampling_interval = htons(0);
 128     }
 129
 130     nf_hdr = nf->packet.data;
 131     nf_hdr->count = htons(ntohs(nf_hdr->count) + 1);
 132
 133     nf_rec = ofpbuf_put_zeros(&nf->packet, sizeof *nf_rec);
 134     nf_rec->src_addr = expired->flow.nw_src;
 135     nf_rec->dst_addr = expired->flow.nw_dst;
 136     nf_rec->nexthop = htonl(0);
 137     if (nf->add_id_to_iface) {
 138         uint16_t iface = (nf->engine_id & 0x7f) << 9;
 139         nf_rec->input = htons(iface | (expired->flow.in_port & 0x1ff));
 140         nf_rec->output = htons(iface | (nf_flow->output_iface & 0x1ff));
 141     } else {
 142         nf_rec->input = htons(expired->flow.in_port);
 143         nf_rec->output = htons(nf_flow->output_iface);
 144     }
 145     nf_rec->packet_count = htonl(packet_count);
 146     nf_rec->byte_count = htonl(byte_count);
 147     nf_rec->init_time = htonl(nf_flow->created - nf->boot_time);
 148     nf_rec->used_time = htonl(MAX(nf_flow->created, expired->used)
 149                              - nf->boot_time);
 150     if (expired->flow.nw_proto == IPPROTO_ICMP) {
 151         /* In NetFlow, the ICMP type and code are concatenated and
 152          * placed in the 'dst_port' field. */
 153         uint8_t type = ntohs(expired->flow.tp_src);
 154         uint8_t code = ntohs(expired->flow.tp_dst);
 155         nf_rec->src_port = htons(0);
 156         nf_rec->dst_port = htons((type << 8) | code);
 157     } else {
 158         nf_rec->src_port = expired->flow.tp_src;
 159         nf_rec->dst_port = expired->flow.tp_dst;
 160     }
 161     nf_rec->tcp_flags = nf_flow->tcp_flags;
 162     nf_rec->ip_proto = expired->flow.nw_proto;
 163     nf_rec->ip_tos = expired->flow.tos_frag & IP_DSCP_MASK;
 164
 165     /* NetFlow messages are limited to 30 records. */
 166     if (ntohs(nf_hdr->count) >= 30) {
 167         netflow_run(nf);
 168     }
 169 }
 170
 171 void
 172 netflow_expire(struct netflow *nf, struct netflow_flow *nf_flow,
 173                struct ofexpired *expired)
 174 {
 175     uint64_t pkt_delta = expired->packet_count - nf_flow->packet_count_off;
 176     uint64_t byte_delta = expired->byte_count - nf_flow->byte_count_off;
 177
 178     nf_flow->last_expired += nf->active_timeout;
 179
 180     /* NetFlow only reports on IP packets and we should only report flows
 181      * that actually have traffic. */
 182     if (expired->flow.dl_type != htons(ETH_TYPE_IP) || pkt_delta == 0) {
 183         return;
 184     }
 185
 186     if ((byte_delta >> 32) <= 175) {
 187         /* NetFlow v5 records are limited to 32-bit counters.  If we've wrapped
 188          * a counter, send as multiple records so we don't lose track of any
 189          * traffic.  We try to evenly distribute the packet and byte counters,
 190          * so that the bytes-per-packet lengths don't look wonky across the
 191          * records. */
 192         while (byte_delta) {
 193             int n_recs = (byte_delta + UINT32_MAX - 1) / UINT32_MAX;
 194             uint32_t pkt_count = pkt_delta / n_recs;
 195             uint32_t byte_count = byte_delta / n_recs;
 196
 197             gen_netflow_rec(nf, nf_flow, expired, pkt_count, byte_count);
 198
 199             pkt_delta -= pkt_count;
 200             byte_delta -= byte_count;
 201         }
 202     } else {
 203         /* In 600 seconds, a 10GbE link can theoretically transmit 75 * 10**10
 204          * == 175 * 2**32 bytes.  The byte counter is bigger than that, so it's
 205          * probably a bug--for example, the netdev code uses UINT64_MAX to
 206          * report "unknown value", and perhaps that has leaked through to here.
 207          *
 208          * We wouldn't want to hit the loop above in this case, because it
 209          * would try to send up to UINT32_MAX netflow records, which would take
 210          * a long time.
 211          */
 212         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
 213
 214         VLOG_WARN_RL(&rl, "impossible byte counter %"PRIu64, byte_delta);
 215     }
 216
 217     /* Update flow tracking data. */
 218     nf_flow->created = 0;
 219     nf_flow->packet_count_off = expired->packet_count;
 220     nf_flow->byte_count_off = expired->byte_count;
 221     nf_flow->tcp_flags = 0;
 222 }
 223
 224 void
 225 netflow_run(struct netflow *nf)
 226 {
 227     if (nf->packet.size) {
 228         collectors_send(nf->collectors, nf->packet.data, nf->packet.size);
 229         nf->packet.size = 0;
 230     }
 231 }
 232
 233 int
 234 netflow_set_options(struct netflow *nf,
 235                     const struct netflow_options *nf_options)
 236 {
 237     int error = 0;
 238     long long int old_timeout;
 239
 240     nf->engine_type = nf_options->engine_type;
 241     nf->engine_id = nf_options->engine_id;
 242     nf->add_id_to_iface = nf_options->add_id_to_iface;
 243
 244     collectors_destroy(nf->collectors);
 245     collectors_create(&nf_options->collectors, 0, &nf->collectors);
 246
 247     old_timeout = nf->active_timeout;
 248     if (nf_options->active_timeout >= 0) {
 249         nf->active_timeout = nf_options->active_timeout;
 250     } else {
 251         nf->active_timeout = NF_ACTIVE_TIMEOUT_DEFAULT;
 252     }
 253     nf->active_timeout *= 1000;
 254     if (old_timeout != nf->active_timeout) {
 255         nf->reconfig_time = time_msec();
 256     }
 257
 258     return error;
 259 }
 260
 261 struct netflow *
 262 netflow_create(void)
 263 {
 264     struct netflow *nf = xmalloc(sizeof *nf);
 265     nf->engine_type = 0;
 266     nf->engine_id = 0;
 267     nf->boot_time = time_msec();
 268     nf->collectors = NULL;
 269     nf->add_id_to_iface = false;
 270     nf->netflow_cnt = 0;
 271     ofpbuf_init(&nf->packet, 1500);
 272     return nf;
 273 }
 274
 275 void
 276 netflow_destroy(struct netflow *nf)
 277 {
 278     if (nf) {
 279         ofpbuf_uninit(&nf->packet);
 280         collectors_destroy(nf->collectors);
 281         free(nf);
 282     }
 283 }
 284
 285 /* Initializes a new 'nf_flow' given that the caller has already cleared it to
 286  * all-zero-bits. */
 287 void
 288 netflow_flow_init(struct netflow_flow *nf_flow OVS_UNUSED)
 289 {
 290     /* Nothing to do. */
 291 }
 292
 293 void
 294 netflow_flow_clear(struct netflow_flow *nf_flow)
 295 {
 296     uint16_t output_iface = nf_flow->output_iface;
 297
 298     memset(nf_flow, 0, sizeof *nf_flow);
 299     nf_flow->output_iface = output_iface;
 300 }
 301
 302 void
 303 netflow_flow_update_time(struct netflow *nf, struct netflow_flow *nf_flow,
 304                          long long int used)
 305 {
 306     if (!nf_flow->created) {
 307         nf_flow->created = used;
 308     }
 309
 310     if (!nf || !nf->active_timeout || !nf_flow->last_expired ||
 311         nf->reconfig_time > nf_flow->last_expired) {
 312         /* Keep the time updated to prevent a flood of expiration in
 313          * the future. */
 314         nf_flow->last_expired = time_msec();
 315     }
 316 }
 317
 318 void
 319 netflow_flow_update_flags(struct netflow_flow *nf_flow, uint8_t tcp_flags)
 320 {
 321     nf_flow->tcp_flags |= tcp_flags;
 322 }
 323
 324 bool
 325 netflow_active_timeout_expired(struct netflow *nf, struct netflow_flow *nf_flow)
 326 {
 327     if (nf->active_timeout) {
 328         return time_msec() > nf_flow->last_expired + nf->active_timeout;
 329     }
 330
 331     return false;
 332 }