/*
- * Copyright (c) 2008, 2009, 2010 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <errno.h>
#include <stdlib.h>
#include <unistd.h>
+#include "byte-order.h"
#include "collectors.h"
#include "flow.h"
-#include "netflow.h"
+#include "lib/netflow.h"
#include "ofpbuf.h"
#include "ofproto.h"
+#include "ofproto/netflow.h"
#include "packets.h"
+#include "poll-loop.h"
#include "socket-util.h"
-#include "svec.h"
#include "timeval.h"
#include "util.h"
#include "vlog.h"
-#include "xtoxll.h"
-
-VLOG_DEFINE_THIS_MODULE(netflow)
-#define NETFLOW_V5_VERSION 5
-
-/* Every NetFlow v5 message contains the header that follows. This is
- * followed by up to thirty records that describe a terminating flow.
- * We only send a single record per NetFlow message.
- */
-struct netflow_v5_header {
- uint16_t version; /* NetFlow version is 5. */
- uint16_t count; /* Number of records in this message. */
- uint32_t sysuptime; /* System uptime in milliseconds. */
- uint32_t unix_secs; /* Number of seconds since Unix epoch. */
- uint32_t unix_nsecs; /* Number of residual nanoseconds
- after epoch seconds. */
- uint32_t flow_seq; /* Number of flows since sending
- messages began. */
- uint8_t engine_type; /* Engine type. */
- uint8_t engine_id; /* Engine id. */
- uint16_t sampling_interval; /* Set to zero. */
-};
-BUILD_ASSERT_DECL(sizeof(struct netflow_v5_header) == 24);
-
-/* A NetFlow v5 description of a terminating flow. It is preceded by a
- * NetFlow v5 header.
- */
-struct netflow_v5_record {
- uint32_t src_addr; /* Source IP address. */
- uint32_t dst_addr; /* Destination IP address. */
- uint32_t nexthop; /* IP address of next hop. Set to 0. */
- uint16_t input; /* Input interface index. */
- uint16_t output; /* Output interface index. */
- uint32_t packet_count; /* Number of packets. */
- uint32_t byte_count; /* Number of bytes. */
- uint32_t init_time; /* Value of sysuptime on first packet. */
- uint32_t used_time; /* Value of sysuptime on last packet. */
-
- /* The 'src_port' and 'dst_port' identify the source and destination
- * port, respectively, for TCP and UDP. For ICMP, the high-order
- * byte identifies the type and low-order byte identifies the code
- * in the 'dst_port' field. */
- uint16_t src_port;
- uint16_t dst_port;
-
- uint8_t pad1;
- uint8_t tcp_flags; /* Union of seen TCP flags. */
- uint8_t ip_proto; /* IP protocol. */
- uint8_t ip_tos; /* IP TOS value. */
- uint16_t src_as; /* Source AS ID. Set to 0. */
- uint16_t dst_as; /* Destination AS ID. Set to 0. */
- uint8_t src_mask; /* Source mask bits. Set to 0. */
- uint8_t dst_mask; /* Destination mask bits. Set to 0. */
- uint8_t pad[2];
-};
-BUILD_ASSERT_DECL(sizeof(struct netflow_v5_record) == 48);
+VLOG_DEFINE_THIS_MODULE(netflow);
struct netflow {
uint8_t engine_type; /* Value of engine_type to use. */
uint32_t netflow_cnt; /* Flow sequence number for NetFlow. */
struct ofpbuf packet; /* NetFlow packet being accumulated. */
long long int active_timeout; /* Timeout for flows that are still active. */
+ long long int next_timeout; /* Next scheduled active timeout. */
long long int reconfig_time; /* When we reconfigured the timeouts. */
};
static void
gen_netflow_rec(struct netflow *nf, struct netflow_flow *nf_flow,
- struct ofexpired *expired,
+ struct ofexpired *expired,
uint32_t packet_count, uint32_t byte_count)
{
struct netflow_v5_header *nf_hdr;
nf_rec = ofpbuf_put_zeros(&nf->packet, sizeof *nf_rec);
nf_rec->src_addr = expired->flow.nw_src;
nf_rec->dst_addr = expired->flow.nw_dst;
- nf_rec->nexthop = htons(0);
+ nf_rec->nexthop = htonl(0);
if (nf->add_id_to_iface) {
uint16_t iface = (nf->engine_id & 0x7f) << 9;
nf_rec->input = htons(iface | (expired->flow.in_port & 0x1ff));
nf_rec->init_time = htonl(nf_flow->created - nf->boot_time);
nf_rec->used_time = htonl(MAX(nf_flow->created, expired->used)
- nf->boot_time);
- if (expired->flow.nw_proto == IP_TYPE_ICMP) {
+ if (expired->flow.nw_proto == IPPROTO_ICMP) {
/* In NetFlow, the ICMP type and code are concatenated and
* placed in the 'dst_port' field. */
uint8_t type = ntohs(expired->flow.tp_src);
}
nf_rec->tcp_flags = nf_flow->tcp_flags;
nf_rec->ip_proto = expired->flow.nw_proto;
- nf_rec->ip_tos = expired->flow.nw_tos;
+ nf_rec->ip_tos = expired->flow.nw_tos & IP_DSCP_MASK;
/* NetFlow messages are limited to 30 records. */
if (ntohs(nf_hdr->count) >= 30) {
return;
}
- /* NetFlow v5 records are limited to 32-bit counters. If we've
- * wrapped a counter, send as multiple records so we don't lose
- * track of any traffic. We try to evenly distribute the packet and
- * byte counters, so that the bytes-per-packet lengths don't look
- * wonky across the records. */
- while (byte_delta) {
- int n_recs = (byte_delta + UINT32_MAX - 1) / UINT32_MAX;
- uint32_t pkt_count = pkt_delta / n_recs;
- uint32_t byte_count = byte_delta / n_recs;
-
- gen_netflow_rec(nf, nf_flow, expired, pkt_count, byte_count);
-
- pkt_delta -= pkt_count;
- byte_delta -= byte_count;
+ if ((byte_delta >> 32) <= 175) {
+ /* NetFlow v5 records are limited to 32-bit counters. If we've wrapped
+ * a counter, send as multiple records so we don't lose track of any
+ * traffic. We try to evenly distribute the packet and byte counters,
+ * so that the bytes-per-packet lengths don't look wonky across the
+ * records. */
+ while (byte_delta) {
+ int n_recs = (byte_delta + UINT32_MAX - 1) / UINT32_MAX;
+ uint32_t pkt_count = pkt_delta / n_recs;
+ uint32_t byte_count = byte_delta / n_recs;
+
+ gen_netflow_rec(nf, nf_flow, expired, pkt_count, byte_count);
+
+ pkt_delta -= pkt_count;
+ byte_delta -= byte_count;
+ }
+ } else {
+ /* In 600 seconds, a 10GbE link can theoretically transmit 75 * 10**10
+ * == 175 * 2**32 bytes. The byte counter is bigger than that, so it's
+ * probably a bug--for example, the netdev code uses UINT64_MAX to
+ * report "unknown value", and perhaps that has leaked through to here.
+ *
+ * We wouldn't want to hit the loop above in this case, because it
+ * would try to send up to UINT32_MAX netflow records, which would take
+ * a long time.
+ */
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
+
+ VLOG_WARN_RL(&rl, "impossible byte counter %"PRIu64, byte_delta);
}
/* Update flow tracking data. */
nf_flow->tcp_flags = 0;
}
-void
+/* Returns true if it's time to send out a round of NetFlow active timeouts,
+ * false otherwise. */
+bool
netflow_run(struct netflow *nf)
{
if (nf->packet.size) {
collectors_send(nf->collectors, nf->packet.data, nf->packet.size);
nf->packet.size = 0;
}
+
+ if (nf->active_timeout && time_msec() >= nf->next_timeout) {
+ nf->next_timeout = time_msec() + 1000;
+ return true;
+ } else {
+ return false;
+ }
+}
+
+void
+netflow_wait(struct netflow *nf)
+{
+ if (nf->active_timeout) {
+ poll_timer_wait_until(nf->next_timeout);
+ }
+ if (nf->packet.size) {
+ poll_immediate_wake();
+ }
}
int
nf->active_timeout *= 1000;
if (old_timeout != nf->active_timeout) {
nf->reconfig_time = time_msec();
+ nf->next_timeout = time_msec();
}
return error;
struct netflow *
netflow_create(void)
{
- struct netflow *nf = xmalloc(sizeof *nf);
+ struct netflow *nf = xzalloc(sizeof *nf);
nf->engine_type = 0;
nf->engine_id = 0;
nf->boot_time = time_msec();
}
}
+/* Initializes a new 'nf_flow' given that the caller has already cleared it to
+ * all-zero-bits. */
+void
+netflow_flow_init(struct netflow_flow *nf_flow OVS_UNUSED)
+{
+ /* Nothing to do. */
+}
+
void
netflow_flow_clear(struct netflow_flow *nf_flow)
{