2 * Copyright (c) 2008, 2009, 2010 Nicira Networks.
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:
8 * http://www.apache.org/licenses/LICENSE-2.0
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.
19 #include <arpa/inet.h>
23 #include "collectors.h"
29 #include "socket-util.h"
36 VLOG_DEFINE_THIS_MODULE(netflow)
38 #define NETFLOW_V5_VERSION 5
40 /* Every NetFlow v5 message contains the header that follows. This is
41 * followed by up to thirty records that describe a terminating flow.
42 * We only send a single record per NetFlow message.
44 struct netflow_v5_header {
45 uint16_t version; /* NetFlow version is 5. */
46 uint16_t count; /* Number of records in this message. */
47 uint32_t sysuptime; /* System uptime in milliseconds. */
48 uint32_t unix_secs; /* Number of seconds since Unix epoch. */
49 uint32_t unix_nsecs; /* Number of residual nanoseconds
50 after epoch seconds. */
51 uint32_t flow_seq; /* Number of flows since sending
53 uint8_t engine_type; /* Engine type. */
54 uint8_t engine_id; /* Engine id. */
55 uint16_t sampling_interval; /* Set to zero. */
57 BUILD_ASSERT_DECL(sizeof(struct netflow_v5_header) == 24);
59 /* A NetFlow v5 description of a terminating flow. It is preceded by a
62 struct netflow_v5_record {
63 uint32_t src_addr; /* Source IP address. */
64 uint32_t dst_addr; /* Destination IP address. */
65 uint32_t nexthop; /* IP address of next hop. Set to 0. */
66 uint16_t input; /* Input interface index. */
67 uint16_t output; /* Output interface index. */
68 uint32_t packet_count; /* Number of packets. */
69 uint32_t byte_count; /* Number of bytes. */
70 uint32_t init_time; /* Value of sysuptime on first packet. */
71 uint32_t used_time; /* Value of sysuptime on last packet. */
73 /* The 'src_port' and 'dst_port' identify the source and destination
74 * port, respectively, for TCP and UDP. For ICMP, the high-order
75 * byte identifies the type and low-order byte identifies the code
76 * in the 'dst_port' field. */
81 uint8_t tcp_flags; /* Union of seen TCP flags. */
82 uint8_t ip_proto; /* IP protocol. */
83 uint8_t ip_tos; /* IP TOS value. */
84 uint16_t src_as; /* Source AS ID. Set to 0. */
85 uint16_t dst_as; /* Destination AS ID. Set to 0. */
86 uint8_t src_mask; /* Source mask bits. Set to 0. */
87 uint8_t dst_mask; /* Destination mask bits. Set to 0. */
90 BUILD_ASSERT_DECL(sizeof(struct netflow_v5_record) == 48);
93 uint8_t engine_type; /* Value of engine_type to use. */
94 uint8_t engine_id; /* Value of engine_id to use. */
95 long long int boot_time; /* Time when netflow_create() was called. */
96 struct collectors *collectors; /* NetFlow collectors. */
97 bool add_id_to_iface; /* Put the 7 least signficiant bits of
98 * 'engine_id' into the most signficant
99 * bits of the interface fields. */
100 uint32_t netflow_cnt; /* Flow sequence number for NetFlow. */
101 struct ofpbuf packet; /* NetFlow packet being accumulated. */
102 long long int active_timeout; /* Timeout for flows that are still active. */
103 long long int reconfig_time; /* When we reconfigured the timeouts. */
107 gen_netflow_rec(struct netflow *nf, struct netflow_flow *nf_flow,
108 struct ofexpired *expired,
109 uint32_t packet_count, uint32_t byte_count)
111 struct netflow_v5_header *nf_hdr;
112 struct netflow_v5_record *nf_rec;
114 if (!nf->packet.size) {
117 time_wall_timespec(&now);
119 nf_hdr = ofpbuf_put_zeros(&nf->packet, sizeof *nf_hdr);
120 nf_hdr->version = htons(NETFLOW_V5_VERSION);
121 nf_hdr->count = htons(0);
122 nf_hdr->sysuptime = htonl(time_msec() - nf->boot_time);
123 nf_hdr->unix_secs = htonl(now.tv_sec);
124 nf_hdr->unix_nsecs = htonl(now.tv_nsec);
125 nf_hdr->flow_seq = htonl(nf->netflow_cnt++);
126 nf_hdr->engine_type = nf->engine_type;
127 nf_hdr->engine_id = nf->engine_id;
128 nf_hdr->sampling_interval = htons(0);
131 nf_hdr = nf->packet.data;
132 nf_hdr->count = htons(ntohs(nf_hdr->count) + 1);
134 nf_rec = ofpbuf_put_zeros(&nf->packet, sizeof *nf_rec);
135 nf_rec->src_addr = expired->flow.nw_src;
136 nf_rec->dst_addr = expired->flow.nw_dst;
137 nf_rec->nexthop = htons(0);
138 if (nf->add_id_to_iface) {
139 uint16_t iface = (nf->engine_id & 0x7f) << 9;
140 nf_rec->input = htons(iface | (expired->flow.in_port & 0x1ff));
141 nf_rec->output = htons(iface | (nf_flow->output_iface & 0x1ff));
143 nf_rec->input = htons(expired->flow.in_port);
144 nf_rec->output = htons(nf_flow->output_iface);
146 nf_rec->packet_count = htonl(packet_count);
147 nf_rec->byte_count = htonl(byte_count);
148 nf_rec->init_time = htonl(nf_flow->created - nf->boot_time);
149 nf_rec->used_time = htonl(MAX(nf_flow->created, expired->used)
151 if (expired->flow.nw_proto == IP_TYPE_ICMP) {
152 /* In NetFlow, the ICMP type and code are concatenated and
153 * placed in the 'dst_port' field. */
154 uint8_t type = ntohs(expired->flow.tp_src);
155 uint8_t code = ntohs(expired->flow.tp_dst);
156 nf_rec->src_port = htons(0);
157 nf_rec->dst_port = htons((type << 8) | code);
159 nf_rec->src_port = expired->flow.tp_src;
160 nf_rec->dst_port = expired->flow.tp_dst;
162 nf_rec->tcp_flags = nf_flow->tcp_flags;
163 nf_rec->ip_proto = expired->flow.nw_proto;
164 nf_rec->ip_tos = expired->flow.nw_tos;
166 /* NetFlow messages are limited to 30 records. */
167 if (ntohs(nf_hdr->count) >= 30) {
173 netflow_expire(struct netflow *nf, struct netflow_flow *nf_flow,
174 struct ofexpired *expired)
176 uint64_t pkt_delta = expired->packet_count - nf_flow->packet_count_off;
177 uint64_t byte_delta = expired->byte_count - nf_flow->byte_count_off;
179 nf_flow->last_expired += nf->active_timeout;
181 /* NetFlow only reports on IP packets and we should only report flows
182 * that actually have traffic. */
183 if (expired->flow.dl_type != htons(ETH_TYPE_IP) || pkt_delta == 0) {
187 if ((byte_delta >> 32) <= 175) {
188 /* NetFlow v5 records are limited to 32-bit counters. If we've wrapped
189 * a counter, send as multiple records so we don't lose track of any
190 * traffic. We try to evenly distribute the packet and byte counters,
191 * so that the bytes-per-packet lengths don't look wonky across the
193 while (byte_delta > UINT32_MAX) {
194 uint32_t n_recs = byte_delta >> 32;
195 uint32_t pkt_count = pkt_delta / n_recs;
196 uint32_t byte_count = byte_delta / n_recs;
198 gen_netflow_rec(nf, nf_flow, expired, pkt_count, byte_count);
200 pkt_delta -= pkt_count;
201 byte_delta -= byte_count;
203 if (byte_delta > 0) {
204 gen_netflow_rec(nf, nf_flow, expired, pkt_delta, byte_delta);
207 /* In 600 seconds, a 10GbE link can theoretically transmit 75 * 10**10
208 * == 175 * 2**32 bytes. The byte counter is bigger than that, so it's
209 * probably a bug--for example, the netdev code uses UINT64_MAX to
210 * report "unknown value", and perhaps that has leaked through to here.
212 * We wouldn't want to hit the loop above in this case, because it
213 * would try to send up to UINT32_MAX netflow records, which would take
216 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
218 VLOG_WARN_RL(&rl, "impossible byte counter %"PRIu64, byte_delta);
221 /* Update flow tracking data. */
222 nf_flow->created = 0;
223 nf_flow->packet_count_off = expired->packet_count;
224 nf_flow->byte_count_off = expired->byte_count;
225 nf_flow->tcp_flags = 0;
229 netflow_run(struct netflow *nf)
231 if (nf->packet.size) {
232 collectors_send(nf->collectors, nf->packet.data, nf->packet.size);
238 netflow_set_options(struct netflow *nf,
239 const struct netflow_options *nf_options)
242 long long int old_timeout;
244 nf->engine_type = nf_options->engine_type;
245 nf->engine_id = nf_options->engine_id;
246 nf->add_id_to_iface = nf_options->add_id_to_iface;
248 collectors_destroy(nf->collectors);
249 collectors_create(&nf_options->collectors, 0, &nf->collectors);
251 old_timeout = nf->active_timeout;
252 if (nf_options->active_timeout >= 0) {
253 nf->active_timeout = nf_options->active_timeout;
255 nf->active_timeout = NF_ACTIVE_TIMEOUT_DEFAULT;
257 nf->active_timeout *= 1000;
258 if (old_timeout != nf->active_timeout) {
259 nf->reconfig_time = time_msec();
268 struct netflow *nf = xmalloc(sizeof *nf);
271 nf->boot_time = time_msec();
272 nf->collectors = NULL;
273 nf->add_id_to_iface = false;
275 ofpbuf_init(&nf->packet, 1500);
280 netflow_destroy(struct netflow *nf)
283 ofpbuf_uninit(&nf->packet);
284 collectors_destroy(nf->collectors);
290 netflow_flow_clear(struct netflow_flow *nf_flow)
292 uint16_t output_iface = nf_flow->output_iface;
294 memset(nf_flow, 0, sizeof *nf_flow);
295 nf_flow->output_iface = output_iface;
299 netflow_flow_update_time(struct netflow *nf, struct netflow_flow *nf_flow,
302 if (!nf_flow->created) {
303 nf_flow->created = used;
306 if (!nf || !nf->active_timeout || !nf_flow->last_expired ||
307 nf->reconfig_time > nf_flow->last_expired) {
308 /* Keep the time updated to prevent a flood of expiration in
310 nf_flow->last_expired = time_msec();
315 netflow_flow_update_flags(struct netflow_flow *nf_flow, uint8_t tcp_flags)
317 nf_flow->tcp_flags |= tcp_flags;
321 netflow_active_timeout_expired(struct netflow *nf, struct netflow_flow *nf_flow)
323 if (nf->active_timeout) {
324 return time_msec() > nf_flow->last_expired + nf->active_timeout;