Scroll to navigation

BPF(9) Kernel Developer's Manual BPF(9)


bpfBerkeley Packet Filter


#include <net/bpf.h>

bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen);

bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp);

bpfdetach(struct ifnet *ifp);

bpf_tap(struct ifnet *ifp, u_char *pkt, u_int *pktlen);

bpf_mtap(struct ifnet *ifp, struct mbuf *m);

bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m);

bpf_filter(const struct bpf_insn *pc , u_char *pkt, u_int wirelen, u_int buflen);

bpf_validate(const struct bpf_insn *fcode, int flen);


The Berkeley Packet Filter provides a raw interface, that is protocol independent, to data link layers. It allows all packets on the network, even those destined for other hosts, to be passed from a network interface to user programs. Each program may specify a filter, in the form of a bpf filter machine program. The bpf(4) manual page describes the interface used by user programs. This manual page describes the functions used by interfaces to pass packets to bpf and the functions for testing and running bpf filter machine programs.

The () function attaches a network interface to bpf. The ifp argument is a pointer to the structure that defines the interface to be attached to an interface. The dlt argument is the data link-layer type: DLT_NULL (no link-layer encapsulation), DLT_EN10MB (Ethernet), DLT_IEEE802_11 (802.11 wireless networks), etc. The rest of the link layer types can be found in <net/bpf.h>. The hdrlen argument is the fixed size of the link header; variable length headers are not yet supported. The bpf system will hold a pointer to ifp->if_bpf. This variable will set to a non-NULL value when bpf requires packets from this interface to be tapped using the functions below.

The () function allows multiple bpf instances to be attached to a single interface, by registering an explicit if_bpf rather than using ifp->if_bpf. It is then possible to run tcpdump(1) on the interface for any data link-layer types attached.

The () function detaches a bpf instance from an interface, specified by ifp. The bpfdetach() function should be called once for each bpf instance attached.

The () function is used by an interface to pass the packet to bpf. The packet data (including link-header), pointed to by pkt, is of length pktlen, which must be a contiguous buffer. The ifp argument is a pointer to the structure that defines the interface to be tapped. The packet is parsed by each processes filter, and if accepted, it is buffered for the process to read.

The () function is like bpf_tap() except that it is used to tap packets that are in an mbuf chain, m. The ifp argument is a pointer to the structure that defines the interface to be tapped. Like bpf_tap(), bpf_mtap() requires a link-header for whatever data link layer type is specified. Note that bpf only reads from the mbuf chain, it does not free it or keep a pointer to it. This means that an mbuf containing the link-header can be prepended to the chain if necessary. A cleaner interface to achieve this is provided by bpf_mtap2().

The () function allows the user to pass a link-header data, of length dlen, independent of the mbuf m, containing the packet. This simplifies the passing of some link-headers.

The () function executes the filter program starting at pc on the packet pkt. The wirelen argument is the length of the original packet and buflen is the amount of data present. The buflen value of 0 is special; it indicates that the pkt is actually a pointer to an mbuf chain (struct mbuf *).

The () function checks that the filter code fcode, of length flen, is valid.


The bpf_filter() function returns -1 (cast to an unsigned integer) if there is no filter. Otherwise, it returns the result of the filter program.

The bpf_validate() function returns 0 when the program is not a valid filter program.


bpf invokes bpf_track EVENTHANDLER(9) event each time listener attaches to or detaches from an interface. Pointer to (struct ifnet *) is passed as the first argument, interface dlt follows. Last argument indicates listener is attached (1) or detached (0). Note that handler is invoked with bpf global lock held, which implies restriction on sleeping and calling bpf subsystem inside EVENTHANDLER(9) dispatcher. Note that handler is not called for write-only listeners.


tcpdump(1), bpf(4), EVENTHANDLER(9)


The Enet packet filter was created in 1980 by Mike Accetta and Rick Rashid at Carnegie-Mellon University. Jeffrey Mogul, at Stanford, ported the code to BSD and continued its development from 1983 on. Since then, it has evolved into the Ultrix Packet Filter at DEC, a STREAMS NIT module under SunOS 4.1, and BPF.


Steven McCanne, of Lawrence Berkeley Laboratory, implemented BPF in Summer 1990. Much of the design is due to Van Jacobson. This manpage was written by Orla McGann.

May 11, 2012 Debian