2 * Copyright (c) 2008, 2009, 2010, 2011 Nicira, Inc.
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 "byte-order.h"
24 #include "collectors.h"
26 #include "lib/netflow.h"
29 #include "ofproto/netflow.h"
31 #include "poll-loop.h"
32 #include "socket-util.h"
37 VLOG_DEFINE_THIS_MODULE(netflow);
40 uint8_t engine_type; /* Value of engine_type to use. */
41 uint8_t engine_id; /* Value of engine_id to use. */
42 long long int boot_time; /* Time when netflow_create() was called. */
43 struct collectors *collectors; /* NetFlow collectors. */
44 bool add_id_to_iface; /* Put the 7 least significiant bits of
45 * 'engine_id' into the most significant
46 * bits of the interface fields. */
47 uint32_t netflow_cnt; /* Flow sequence number for NetFlow. */
48 struct ofpbuf packet; /* NetFlow packet being accumulated. */
49 long long int active_timeout; /* Timeout for flows that are still active. */
50 long long int next_timeout; /* Next scheduled active timeout. */
51 long long int reconfig_time; /* When we reconfigured the timeouts. */
55 netflow_mask_wc(struct flow *flow, struct flow_wildcards *wc)
57 if (flow->dl_type != htons(ETH_TYPE_IP)) {
60 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
61 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
62 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
63 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
64 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
65 wc->masks.nw_tos |= IP_DSCP_MASK;
69 gen_netflow_rec(struct netflow *nf, struct netflow_flow *nf_flow,
70 struct ofexpired *expired,
71 uint32_t packet_count, uint32_t byte_count)
73 struct netflow_v5_header *nf_hdr;
74 struct netflow_v5_record *nf_rec;
76 if (!nf->packet.size) {
79 time_wall_timespec(&now);
81 nf_hdr = ofpbuf_put_zeros(&nf->packet, sizeof *nf_hdr);
82 nf_hdr->version = htons(NETFLOW_V5_VERSION);
83 nf_hdr->count = htons(0);
84 nf_hdr->sysuptime = htonl(time_msec() - nf->boot_time);
85 nf_hdr->unix_secs = htonl(now.tv_sec);
86 nf_hdr->unix_nsecs = htonl(now.tv_nsec);
87 nf_hdr->flow_seq = htonl(nf->netflow_cnt++);
88 nf_hdr->engine_type = nf->engine_type;
89 nf_hdr->engine_id = nf->engine_id;
90 nf_hdr->sampling_interval = htons(0);
93 nf_hdr = nf->packet.data;
94 nf_hdr->count = htons(ntohs(nf_hdr->count) + 1);
96 nf_rec = ofpbuf_put_zeros(&nf->packet, sizeof *nf_rec);
97 nf_rec->src_addr = expired->flow.nw_src;
98 nf_rec->dst_addr = expired->flow.nw_dst;
99 nf_rec->nexthop = htonl(0);
100 if (nf->add_id_to_iface) {
101 uint16_t iface = (nf->engine_id & 0x7f) << 9;
102 nf_rec->input = htons(iface
103 | (ofp_to_u16(expired->flow.in_port.ofp_port) & 0x1ff));
104 nf_rec->output = htons(iface
105 | (ofp_to_u16(nf_flow->output_iface) & 0x1ff));
107 nf_rec->input = htons(ofp_to_u16(expired->flow.in_port.ofp_port));
108 nf_rec->output = htons(ofp_to_u16(nf_flow->output_iface));
110 nf_rec->packet_count = htonl(packet_count);
111 nf_rec->byte_count = htonl(byte_count);
112 nf_rec->init_time = htonl(nf_flow->created - nf->boot_time);
113 nf_rec->used_time = htonl(MAX(nf_flow->created, expired->used)
115 if (expired->flow.nw_proto == IPPROTO_ICMP) {
116 /* In NetFlow, the ICMP type and code are concatenated and
117 * placed in the 'dst_port' field. */
118 uint8_t type = ntohs(expired->flow.tp_src);
119 uint8_t code = ntohs(expired->flow.tp_dst);
120 nf_rec->src_port = htons(0);
121 nf_rec->dst_port = htons((type << 8) | code);
123 nf_rec->src_port = expired->flow.tp_src;
124 nf_rec->dst_port = expired->flow.tp_dst;
126 nf_rec->tcp_flags = (uint8_t)nf_flow->tcp_flags;
127 nf_rec->ip_proto = expired->flow.nw_proto;
128 nf_rec->ip_tos = expired->flow.nw_tos & IP_DSCP_MASK;
130 /* NetFlow messages are limited to 30 records. */
131 if (ntohs(nf_hdr->count) >= 30) {
137 netflow_expire(struct netflow *nf, struct netflow_flow *nf_flow,
138 struct ofexpired *expired)
140 uint64_t pkt_delta = expired->packet_count - nf_flow->packet_count_off;
141 uint64_t byte_delta = expired->byte_count - nf_flow->byte_count_off;
143 nf_flow->last_expired += nf->active_timeout;
145 /* NetFlow only reports on IP packets and we should only report flows
146 * that actually have traffic. */
147 if (expired->flow.dl_type != htons(ETH_TYPE_IP) || pkt_delta == 0) {
151 if ((byte_delta >> 32) <= 175) {
152 /* NetFlow v5 records are limited to 32-bit counters. If we've wrapped
153 * a counter, send as multiple records so we don't lose track of any
154 * traffic. We try to evenly distribute the packet and byte counters,
155 * so that the bytes-per-packet lengths don't look wonky across the
158 int n_recs = (byte_delta + UINT32_MAX - 1) / UINT32_MAX;
159 uint32_t pkt_count = pkt_delta / n_recs;
160 uint32_t byte_count = byte_delta / n_recs;
162 gen_netflow_rec(nf, nf_flow, expired, pkt_count, byte_count);
164 pkt_delta -= pkt_count;
165 byte_delta -= byte_count;
168 /* In 600 seconds, a 10GbE link can theoretically transmit 75 * 10**10
169 * == 175 * 2**32 bytes. The byte counter is bigger than that, so it's
170 * probably a bug--for example, the netdev code uses UINT64_MAX to
171 * report "unknown value", and perhaps that has leaked through to here.
173 * We wouldn't want to hit the loop above in this case, because it
174 * would try to send up to UINT32_MAX netflow records, which would take
177 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
179 VLOG_WARN_RL(&rl, "impossible byte counter %"PRIu64, byte_delta);
182 /* Update flow tracking data. */
183 nf_flow->created = 0;
184 nf_flow->packet_count_off = expired->packet_count;
185 nf_flow->byte_count_off = expired->byte_count;
186 nf_flow->tcp_flags = 0;
189 /* Returns true if it's time to send out a round of NetFlow active timeouts,
190 * false otherwise. */
192 netflow_run(struct netflow *nf)
194 if (nf->packet.size) {
195 collectors_send(nf->collectors, nf->packet.data, nf->packet.size);
199 if (nf->active_timeout && time_msec() >= nf->next_timeout) {
200 nf->next_timeout = time_msec() + 1000;
208 netflow_wait(struct netflow *nf)
210 if (nf->active_timeout) {
211 poll_timer_wait_until(nf->next_timeout);
213 if (nf->packet.size) {
214 poll_immediate_wake();
219 netflow_set_options(struct netflow *nf,
220 const struct netflow_options *nf_options)
223 long long int old_timeout;
225 nf->engine_type = nf_options->engine_type;
226 nf->engine_id = nf_options->engine_id;
227 nf->add_id_to_iface = nf_options->add_id_to_iface;
229 collectors_destroy(nf->collectors);
230 collectors_create(&nf_options->collectors, 0, &nf->collectors);
232 old_timeout = nf->active_timeout;
233 if (nf_options->active_timeout >= 0) {
234 nf->active_timeout = nf_options->active_timeout;
236 nf->active_timeout = NF_ACTIVE_TIMEOUT_DEFAULT;
238 nf->active_timeout *= 1000;
239 if (old_timeout != nf->active_timeout) {
240 nf->reconfig_time = time_msec();
241 nf->next_timeout = time_msec();
250 struct netflow *nf = xzalloc(sizeof *nf);
253 nf->boot_time = time_msec();
254 nf->collectors = NULL;
255 nf->add_id_to_iface = false;
257 ofpbuf_init(&nf->packet, 1500);
262 netflow_destroy(struct netflow *nf)
265 ofpbuf_uninit(&nf->packet);
266 collectors_destroy(nf->collectors);
271 /* Initializes a new 'nf_flow' given that the caller has already cleared it to
274 netflow_flow_init(struct netflow_flow *nf_flow OVS_UNUSED)
280 netflow_flow_clear(struct netflow_flow *nf_flow)
282 ofp_port_t output_iface = nf_flow->output_iface;
284 memset(nf_flow, 0, sizeof *nf_flow);
285 nf_flow->output_iface = output_iface;
289 netflow_flow_update_time(struct netflow *nf, struct netflow_flow *nf_flow,
292 if (!nf_flow->created) {
293 nf_flow->created = used;
296 if (!nf || !nf->active_timeout || !nf_flow->last_expired ||
297 nf->reconfig_time > nf_flow->last_expired) {
298 /* Keep the time updated to prevent a flood of expiration in
300 nf_flow->last_expired = time_msec();
305 netflow_flow_update_flags(struct netflow_flow *nf_flow, uint16_t tcp_flags)
307 nf_flow->tcp_flags |= tcp_flags;
311 netflow_active_timeout_expired(struct netflow *nf, struct netflow_flow *nf_flow)
313 if (nf->active_timeout) {
314 return time_msec() > nf_flow->last_expired + nf->active_timeout;