NG_NETFLOW(4)

📦 version 15.1, in freebsd 14.5, freebsd 15.1, nixpkgs unstable

🦋 NG_NETFLOW(4) — the netflow node, explained by your local Unix bestie

okay so this is a manpage from the Kernel Interfaces Manual and it's about ng_netflow, which is genuinely one of those things that sounds scary and then turns out to be iconic once someone explains it. that someone is me. you're welcome 💅

✨ NAME ✨

ng_netflow — Cisco's NetFlow implementation 🎀

✨ SYNOPSIS ✨

#include <sys/types.h>
#include <netinet/in.h>
#include <netgraph/netflow/ng_netflow.h>

✨ DESCRIPTION ✨

Bestie. Listen. The ng_netflow node implements Cisco's NetFlow export protocol on a router that's running FreeBSD. It literally just sits there listening to incoming traffic and figuring out which flows are unique in it. That's the whole job. Main character energy, honestly 👑

So how does it tell flows apart? A flow is distinguished by:

  • endpoint IP addresses
  • TCP/UDP port numbers
  • ToS
  • input interface

Expired flows get exported out of the node in NetFlow version 5/9 UDP datagrams. And why does a flow expire? Three reasons, and here's the list:

  • An RST or FIN TCP segment. bye boo 👋
  • Active timeout. Flows literally cannot live longer than the specified period of time. The default is 1800 seconds (that's 30 minutes).
  • Inactive timeout. The flow was inactive for the specified period of time. The default is 15 seconds.

The node also supports IPv6 accounting (NetFlow v9 only, no cap) and it's aware of multiple fibs. Different fibs are mapped to different domain_id in NetFlow V9, and different engine_id in NetFlow V5. it's giving organized ✨

🫶 HOOKS 🫶

This node type supports up to NG_NETFLOW_MAXIFACES hooks — and that default is 65536, yes really — named iface0, iface1, etc., plus the same number of hooks named out0, out1, etc. Oh and also two export hooks: export (for NetFlow version 5) and export9 (for NetFlow version 9). Export can happen simultaneously for all supported export hooks, because we contain multitudes.

By default (that's with ingress NetFlow enabled), the node does NetFlow accounting of data received on iface* hooks. If the corresponding out hook is connected, the unmodified data is bypassed to it — otherwise the data is just freed. 💀

If data is received on an out hook, it's bypassed to the corresponding iface hook without any processing (that's egress NetFlow, which is disabled by default).

When a full export datagram for an export protocol is built, it gets sent to the export or export9 hook. In normal operation, one (or more) export hook is connected to the inet/dgram/udp hook of the ng_ksocket(4) node. slay.

💅 CONTROL MESSAGES 💅

This node type supports the generic control messages, plus the following:

  • NGM_NETFLOW_INFO (info) — returns some node statistics and the current timeout values in a struct ng_netflow_info.
  • NGM_NETFLOW_IFINFO (ifinfo) — returns information about the ifaceN hook. The hook number is passed as an argument.
  • NGM_NETFLOW_SETDLT (setdlt) — sets data link type on the ifaceN hook. Currently the supported types are DLT_RAW (raw IP datagrams) and DLT_EN10MB (Ethernet). DLT_ definitions can be found in the <net/bpf.h> header. Currently used values are 1 for DLT_EN10MB and 12 for DLT_RAW. This message type uses struct ng_netflow_setdlt as an argument:
struct ng_netflow_setdlt {
        uint16_t iface;         /* which iface dlt change */
        uint8_t  dlt;           /* DLT_XXX from bpf.h */
};
heads up: the requested ifaceN hook must already be connected, otherwise the message send operation will return an error. no hook, no service 🙈
  • NGM_NETFLOW_SETIFINDEX (setifindex) — okay so sometimes ng_netflow may be unable to determine the input interface index of a packet. This can happen if traffic enters the ng_netflow node before it comes to the system interface's input queue. An example of such a setup is capturing traffic between a synchronous data line and ng_iface(4). In this case, the input index should be associated with a given hook. The interface's index can be determined via if_nametoindex(3) from userland. This message requires struct ng_netflow_setifindex as an argument:
struct ng_netflow_setifindex {
        uint16_t iface;         /* which iface index change */
        uint16_t index;         /* new index */
};
same rule bestie: the requested ifaceN hook must already be connected, otherwise the message send operation will return an error.
  • NGM_NETFLOW_SETTIMEOUTS (settimeouts) — sets values in seconds for the NetFlow active/inactive timeouts. This message requires struct ng_netflow_settimeouts as an argument:
struct ng_netflow_settimeouts {
        uint32_t inactive_timeout;      /* flow inactive timeout */
        uint32_t active_timeout;        /* flow active timeout */
};
  • NGM_NETFLOW_SETCONFIG (setconfig) — sets configuration for the specified interface. This message requires struct ng_netflow_setconfig as an argument:
struct ng_netflow_setconfig {
        uint16_t iface;         /* which iface config change */
        uint32_t conf;          /* new config */
#define NG_NETFLOW_CONF_INGRESS         1
#define NG_NETFLOW_CONF_EGRESS          2
#define NG_NETFLOW_CONF_ONCE            4
#define NG_NETFLOW_CONF_THISONCE        8
#define NG_NETFLOW_CONF_NOSRCLOOKUP     16
#define NG_NETFLOW_CONF_NODSTLOOKUP     32
};

Configuration is a bitmask of several options. Let's go through them one by one, because this is the part everyone gets confused about and you're not gonna be everyone:

  • Option NG_NETFLOW_CONF_INGRESS is enabled by default and enables ingress NetFlow generation (for data coming from the ifaceX hook).
  • Option NG_NETFLOW_CONF_EGRESS enables egress NetFlow (for data coming from the outX hook).
  • Option NG_NETFLOW_CONF_ONCE defines that a packet should be accounted only once if it several times passes via the netflow node.
  • Option NG_NETFLOW_CONF_THISONCE defines that a packet should be accounted only once if it several times passes via exactly this netflow node.
  • These two options are important to avoid duplicate accounting when both ingress and egress NetFlow are enabled. fr fr, double counting is not the vibe.
  • Option NG_NETFLOW_CONF_NOSRCLOOKUP skips radix lookup on flow source address, which is used to fill in network mask.
  • Option NG_NETFLOW_CONF_NODSTLOOKUP skips radix lookup on destination (which fills egress interface id, destination mask and gateway).

If one doesn't need the data provided by lookups, he/she can disable them, to reduce load on routers. efficiency queen 🌸

  • NGM_NETFLOW_SETTEMPLATE (settemplate) — sets various timeouts to announce data flow templates (NetFlow v9-specific). This message requires struct ng_netflow_settemplate as an argument:
struct ng_netflow_settemplate {
        uint16_t time;          /* max time between announce */
        uint16_t packets;       /* max packets between announce */
};

The value of the time field represents time in seconds to re-announce data templates. The value of the packets field represents the maximum packets count between re-announcing data templates.

  • NGM_NETFLOW_SETMTU (setmtu) — sets export interface MTU to build packets of specified size (NetFlow v9-specific). This message requires struct ng_netflow_setmtu as an argument:
struct ng_netflow_setemtu {
        uint16_t mtu;           /* MTU for packet */
};

The default is 1500 bytes. (yes the struct in the header is spelled ng_netflow_setemtu — that's not me, that's canon, iykyk 🫶)

  • NGM_NETFLOW_SHOW — this control message asks a node to dump the entire contents of the flow cache. Important: it is called from flowctl(8), not directly from ngctl(8). respect the chain of command bestie.
  • NGM_NETFLOW_V9INFO (v9info) — returns some NetFlow v9 related values in a:
struct ng_netflow_v9info {
    uint16_t        templ_packets;  /* v9 template packets */
    uint16_t        templ_time;     /* v9 template time */
    uint16_t        mtu;            /* v9 MTU */
};

🌈 SHUTDOWN 🌈

This node shuts down upon receipt of a NGM_SHUTDOWN control message, or when all hooks have been disconnected. that's it. that's the whole shutdown policy. ✨

🍓 EXAMPLES 🍓

The simplest possible configuration is one Ethernet interface, where flow collecting is enabled. Look at her go:

/usr/sbin/ngctl -f- <<-SEQ
        mkpeer fxp0: netflow lower iface0
        name fxp0:lower netflow
        connect fxp0: netflow: upper out0
        mkpeer netflow: ksocket export inet/dgram/udp
        msg netflow:export connect inet/10.0.0.1:4444
SEQ

And this is a more complicated example of a router with 2 NetFlow-enabled interfaces fxp0 and ng0. Note that the ng0: node in this example is connected to ng_tee(4) — the latter sends us a copy of IP packets, which we analyze and free. On fxp0: we do not use tee, but send packets back to either node. love that for us 💜

/usr/sbin/ngctl -f- <<-SEQ
        # connect ng0's tee to iface0 hook
        mkpeer ng0:inet netflow right2left iface0
        name ng0:inet.right2left netflow
        # set DLT to raw mode
        msg netflow: setdlt { iface=0 dlt=12 }
        # set interface index (5 in this example)
        msg netflow: setifindex { iface=0 index=5 }

        # Connect fxp0: to iface1 and out1 hook
        connect fxp0: netflow: lower iface1
        connect fxp0: netflow: upper out1

        # Create ksocket node on export hook, and configure it
        # to send exports to proper destination
        mkpeer netflow: ksocket export inet/dgram/udp
        msg netflow:export connect inet/10.0.0.1:4444
SEQ

👑 SEE ALSO 👑

setfib(2), netgraph(4), ng_ether(4), ng_iface(4), ng_ksocket(4), ng_tee(4), flowctl(8), ngctl(8)

B. Claise, Ed, Cisco Systems NetFlow Services Export Version 9, RFC 3954.

http://www.cisco.com/en/US/docs/ios/solutions_docs/netflow/nfwhite.html

🧚 AUTHORS 🧚

The ng_netflow node type was written by Gleb Smirnoff <glebius@FreeBSD.org>, Alexander Motin <mav@FreeBSD.org>, and Alexander Chernikov <melifaro@ipfw.ru>. The initial code was based on ng_ipacct, written by Roman V. Palagin <romanp@unshadow.net>. absolute legends, all of them 💖

💀 BUGS 💀

Cache snapshot obtained via the NGM_NETFLOW_SHOW command may lack some percentage of entries under severe load. 😭

The ng_netflow node type does not fill in AS numbers. This is due to the lack of necessary information in the kernel routing table. However — plot twist — this information can be injected into the kernel from a routing daemon such as GNU Zebra. This functionality may become available in future releases. 🙈

and that's the whole page bestie ✨ this beauty came out of the Debian corner of the world, dated December 10, 2012 — yes it's vintage, yes it still ate. go configure some flows, you've got this 🦋

📦 where this page comes from