1.\" $FreeBSD$ 2.\" 3.Dd October 22, 2004 4.Dt DIVERT 4 5.Os 6.Sh NAME 7.Nm divert 8.Nd kernel packet diversion mechanism 9.Sh SYNOPSIS 10.In sys/types.h 11.In sys/socket.h 12.In netinet/in.h 13.Ft int 14.Fn socket PF_INET SOCK_RAW IPPROTO_DIVERT 15.Sh DESCRIPTION 16Divert sockets are similar to raw IP sockets, except that they 17can be bound to a specific 18.Nm 19port via the 20.Xr bind 2 21system call. 22The IP address in the bind is ignored; only the port 23number is significant. 24A divert socket bound to a divert port will receive all packets diverted 25to that port by some (here unspecified) kernel mechanism(s). 26Packets may also be written to a divert port, in which case they 27re-enter kernel IP packet processing. 28.Pp 29Divert sockets are normally used in conjunction with 30.Fx Ns 's 31packet filtering implementation and the 32.Xr ipfw 8 33program. 34By reading from and writing to a divert socket, matching packets 35can be passed through an arbitrary ``filter'' as they travel through 36the host machine, special routing tricks can be done, etc. 37.Sh READING PACKETS 38Packets are diverted either as they are ``incoming'' or ``outgoing.'' 39Incoming packets are diverted after reception on an IP interface, 40whereas outgoing packets are diverted before next hop forwarding. 41.Pp 42Diverted packets may be read unaltered via 43.Xr read 2 , 44.Xr recv 2 , 45or 46.Xr recvfrom 2 . 47In the latter case, the address returned will have its port set to 48some tag supplied by the packet diverter, (usually the ipfw rule number) 49and the IP address set to the (first) address of 50the interface on which the packet was received (if the packet 51was incoming) or 52.Dv INADDR_ANY 53(if the packet was outgoing). 54The interface name (if defined 55for the packet) will be placed in the 8 bytes following the address, 56if it fits. 57.Sh WRITING PACKETS 58Writing to a divert socket is similar to writing to a raw IP socket; 59the packet is injected ``as is'' into the normal kernel IP packet 60processing and minimal error checking is done. 61Packets are written as either incoming or outgoing: 62if 63.Xr write 2 64or 65.Xr send 2 66is used to deliver the packet, or if 67.Xr sendto 2 68is used with a destination IP address of 69.Dv INADDR_ANY , 70then the packet is treated as if it were outgoing, i.e., destined 71for a non-local address. 72Otherwise, the packet is assumed to be 73incoming and full packet routing is done. 74.Pp 75In the latter case, the 76IP address specified must match the address of some local interface, 77or an interface name 78must be found after the IP address. 79If an interface name is found, 80that interface will be used and the value of the IP address will be 81ignored (other than the fact that it is not 82.Dv INADDR_ANY ) . 83This is to indicate on which interface the packet 84.Dq arrived . 85.Pp 86Normally, packets read as incoming should be written as incoming; 87similarly for outgoing packets. 88When reading and then writing back 89packets, passing the same socket address supplied by 90.Xr recvfrom 2 91unmodified to 92.Xr sendto 2 93simplifies things (see below). 94.Pp 95The port part of the socket address passed to the 96.Xr sendto 2 97contains a tag that should be meaningful to the diversion module. 98In the 99case of 100.Xr ipfw 8 101the tag is interpreted as the rule number 102.Em after which 103rule processing should restart. 104.Sh LOOP AVOIDANCE 105Packets written into a divert socket 106(using 107.Xr sendto 2 ) 108re-enter the packet filter at the rule number 109following the tag given in the port part of the socket address, which 110is usually already set at the rule number that caused the diversion 111(not the next rule if there are several at the same number). 112If the 'tag' 113is altered to indicate an alternative re-entry point, care should be taken 114to avoid loops, where the same packet is diverted more than once at the 115same rule. 116.Sh DETAILS 117To enable divert sockets, your kernel must be compiled with the option 118.Dv IPDIVERT 119or you have to load the 120.Dv IPDIVERT 121module. 122.Pp 123You can load the 124.Dv IPDIVERT 125module at runtime by issuing the following command: 126.Bd -literal -offset indent 127kldload ipdivert 128.Ed 129.Pp 130If a packet is diverted but no socket is bound to the 131port, or if 132.Dv IPDIVERT 133is not enabled or loaded in the kernel, the packet is dropped. 134.Pp 135Incoming packet fragments which get diverted are fully reassembled 136before delivery; the diversion of any one fragment causes the entire 137packet to get diverted. 138If different fragments divert to different ports, 139then which port ultimately gets chosen is unpredictable. 140.Pp 141Note that packets arriving on the divert socket by the 142.Xr ipfw 8 143.Cm tee 144action are delivered as-is and packet fragments do not get reassembled 145in this case. 146.Pp 147Packets are received and sent unchanged, except that 148packets read as outgoing have invalid IP header checksums, and 149packets written as outgoing have their IP header checksums overwritten 150with the correct value. 151Packets written as incoming and having incorrect checksums will be dropped. 152Otherwise, all header fields are unchanged (and therefore in network order). 153.Pp 154Binding to port numbers less than 1024 requires super-user access, as does 155creating a socket of type SOCK_RAW. 156.Sh ERRORS 157Writing to a divert socket can return these errors, along with 158the usual errors possible when writing raw packets: 159.Bl -tag -width Er 160.It Bq Er EINVAL 161The packet had an invalid header, or the IP options in the packet 162and the socket options set were incompatible. 163.It Bq Er EADDRNOTAVAIL 164The destination address contained an IP address not equal to 165.Dv INADDR_ANY 166that was not associated with any interface. 167.El 168.Sh SEE ALSO 169.Xr bind 2 , 170.Xr recvfrom 2 , 171.Xr sendto 2 , 172.Xr socket 2 , 173.Xr ipfw 8 174.Sh BUGS 175This is an attempt to provide a clean way for user mode processes 176to implement various IP tricks like address translation, but it 177could be cleaner, and it's too dependent on 178.Xr ipfw 8 . 179.Pp 180It's questionable whether incoming fragments should be reassembled 181before being diverted. 182For example, if only some fragments of a 183packet destined for another machine don't get routed through the 184local machine, the packet is lost. 185This should probably be 186a settable socket option in any case. 187.Sh AUTHORS 188.An Archie Cobbs Aq archie@FreeBSD.org , 189Whistle Communications Corp. 190