Linux implements the IPv4 protocol
described in RFC791 and RFC1122.
ip
contains a level 2
multicasting implementation comforming to RFC1112. It also contains an IP
router including a packet filter.
The protocol is implemented in the
kernel on the basis of a BSD compatible socket interface.
For more information on sockets, see
socket(4).
An IP socket is created by calling the
socket(2)
function with a
PF_INET
socket family argument. Valid socket types are
SOCK_STREAM
to open a
tcp(4)
socket,
SOCK_DGRAM
to open a
udp(4)
socket, or
SOCK_RAW
to open a raw socket.
protocol
is the IP protocol in the IP header to be received or sent. For TCP and UDP
sockets, only 0,
IPPROTO_TCP ,
or
IPPROTO_UDP
are valid. For
SOCK_RAW
you may specify
a valid IANA IP protocol defined in
RFC1700
assigned numbers.
Raw sockets may only be opened by a process with effective user id 0 or when the process has the
CAP_NET_RAW
capability.
When a process wants to receive new incoming packets or connections, it should be
bound to a local interface address using
bind(2).
When
INADDR_ANY
is specified it will bind to any local interface.
A bound TCP socket is unavailable for some time after closing,
unless the
SO_REUSEADDR
flag is set.
ADDRESS FORMAT
An IP socket address is defined as a combination of an IP interface address
and a port number.
struct sockaddr_in {
sa_family_t sin_family;/* address family: AF_INET */
u_int16_t sin_port; /* port in network byte order */
struct in_addr sin_addr;/* internet address */
};
/* Internet address. */
struct in_addr {
u_int32_t s_addr; /* IPv4 address in network byte order */
};
sin_family
is always set to
AF_INET.
This is required; in Linux 2.2 most networking functions return
EINVAL
when this setting is missing.
sin_port
contains the port in network byte order. The port numbers below 1024 are called
reserved ports.
Only processes with the effective user id 0 or the
CAP_NET_BIND_SERVICE
attribute
set may
bind(2)
to these sockets. Note that the raw IPv4 protocol as such has no concept of a
port, they are only implemented by higher protocols like
tcp(4)
and
udp(4).
sin_addr
is the host address.
The
addr
member of
struct in_addr
contains the host interface address in network order.
in_addr
should be only accessed using the
inet_aton(3),
inet_addr(3),
inet_makeaddr(3)
library functions or directly with the name resolver (see
gethostbyname(3)
). IPv4 addresses are divided into unicast, broadcast
and multicast addresses. Unicast addresses specify a single interface of a host,
broadcast addresses specify all host on a network and multicast addresses
address all hosts in a multicast group. Datagrams to broadcast
addresses are only passed to the user when the socket broadcast flag is set.
To send datagrams to broadcast addresses it has to be set too.
Connection oriented sockets are only allowed to use unicast addresses.
Note that the address and the port are always stored in network order, this particulary
means that you need to call
htons(3)
on the number that is assigned to a port. All address/port manipulation functions
in the standard library automatically convert to network order.
SOCKET OPTIONS
IP supports some protocol specific socket options that can be set with
setsockopt(2)
and read by
getsockopt(2).
The socket option level for IP is
SOL_IP
IP_OPTIONS
Sets or get the IP options to be sent with every packet from this
socket. The arguments are a pointer to a memory buffer contained the options
and the option length. Setsockopt sets the IP options associated with a socket.
Maximum option size for IPv4 is 40 bytes. See RFC791 for the allowed
options. When the initial connection request packet for a SOCK_STREAM
socket contains IP options the outgoing IP options will be automatically
set to the received options with routing headers reversed.
Thus, outgoing packets will echo the received options then.
After the connection is established
incoming packets are
not allowed to change options anymore. The processing of all incoming source
routing options
can be disabled using the accept_source_route sysctl, which is off
by default.
For datagram sockets IP options
can be only set by the local user. getsockopt returns the current send IP options.
IP_PKTINFO
Pass a IP_PKTINFO ancillary message that contains a
pktinfo
structure that supplies some information about the incoming packet. This
only works for datagram oriented sockets.
ipi_ifindex
is the index of the interface the packet was received on.
The ipi_spec_dst address is the RFC specified destination address and may
differ from ipi_addr when the packet contains source routing options.
If
IP_PKTINFO
is passed to
sendmsg(2)
then the outgoing packet will be sent over the interface
specified in
ipi_ifindex
with the destination address set to
ipi_spec_dst
IP_RECVTOS
If enabled the
IP_TOS
ancillary message is passed with incomming packets. It contains a byte with
the Type of Service/Precedence field of the packet header as a byte.
Expects a boolean integer flag.
IP_RECVTTL
Set or read a flag to
pass a
IP_RECVTTL
ancillary message that contains the time to live
field of the received packet as a byte. Not supported for SOCK_STREAM sockets.
IP_RECVOPTS
Pass all incoming IP options to the user in a
IP_OPTIONS
control message. The routing header and other options are already filled in
for the local host. Not supported for
SOCK_STREAM
sockets.
IP_RETOPTS
Identical to
IP_RECVOPTS
but returns raw unprocessed options with timestamp and route record
options not filled in for this hop.
IP_TOS
Set or receive the Type-Of-Service (TOS) field that is sent with every IP packet
originating from this socket. It is used to prioritize packets on the network.
TOS is a byte. There are some standard TOS flags defined:
IPTOS_LOWDELAY
to minimize delays for interactive traffic,
IPTOS_THROUGHPUT
to optimize throughput,
IPTOS_RELIABILITY
to optimize for reliability,
IPTOS_MINCOST
should be used for "filler data" where slow transmission doesn't matter.
At most one of these TOS values can be specified. Other bits are invalid and
shall be cleared.
Linux per default sends
IPTOS_LOWDELAY
datagrams first,
but the exact behaviour depends on the configured queueing discipline.
Some high priority levels may require an effective user id 0 or the
CAP_NET_ADMIN
attribute set.
The priority can also be set in a protocol independent way by the
(SOL_SOCKET, SO_PRIORITY)
socket option (see
socket(4)
).
IP_TTL
Set or receive the time to live field for every outgoing IP packet.
IP_HDRINCL
If enabled
the user supplies his own ip header in front of the user data. Only valid
for
SOCK_RAW
sockets. See
raw(4)
for more information. When this flag is enabled the values set by
IP_OPTIONS,IP_TTL,IP_TOS
are ignored.
IP_RECVERR
Enable extended reliable error message passing.
When enabled on a datagram socket all
generated errors will be queued in a per-socket error queue. When the user
gets an error (by a error return of a socket operation) then the errors can
be received by calling
recvmsg(2)
with the
MSG_ERRQUEUE
flag set. The
sock_extended_err
structure describing the error will be passed in a ancillary message with
the type
IP_RECVERR
and the level
SOL_IP.
This is useful for reliable error handling on unconnected sockets.
The received data portion of the error queue
contains the error packet.
IP uses the
sock_extended_err
structure as follows:
ee_origin
set to
SO_EE_ORIGIN_ICMP
for errors received as an ICMP packet, or
SO_EE_ORIGIN_LOCAL
for locally generated errors.
ee_type
and
ee_code
are set from the type and code fields of the ICMP header.
ee_info
contains the discovered MTU for
EMSGSIZE
errors.
ee_data
is currently not used. When the error originated from the network, all IP options
(IP_OPTIONS, IP_TTL,
etc.) enabled on the socket and contained in the
error packet are passed as control messages. The payload of the packet
causing the error is returned as normal data.
On
SOCK_STREAM
TCP sockets,
IP_RECVERR
has a slightly different semantic. Instead of
queueing the errors reliably, it passes all incoming errors immediately to the
user. This might be useful for very short-lived TCP connection that
need quick error handling. Use this option with care: it makes TCP unreliable by not allowing it to recover properly from routing shifts and other normal
conditions.
Note that TCP has no error queue;
MSG_ERRQUEUE
is not invalid on
SOCK_STREAM
sockets.
All errors are passed by return value only.
For raw sockets,
IP_RECVERR
enables passing of all received ICMP errors to the
application. This is turned off by default for compatibility.
It sets or receives an integer boolean flag.
IP_RECVERR
defaults to off.
IP_PMTU_DISCOVER
Sets or receives the Path MTU Discovery setting
for a socket. When enabled, Linux will perform Path MTU Discovery
as defined in RFC1191
on this socket. The system-wide default is controlled by the
ip_no_pmtu_disc
sysctl for
SOCK_STREAM
sockets, and disabled on all others. The user can retrieve the
path MTU
using the
IP_MTU
or the
IP_RECVERR
options.
Path MTU discovery flags
Meaning
IP_PMTUDISC_WANT
Use per-route settings.
IP_PMTUDISC_DONT
Never do Path MTU Discovery.
IP_PMTUDISC_DO
Always do Path MTU Discovery.
When PMTU discovery is enabled the kernel automatically keeps track of
the path MTU. For TCP sockets the outgoing packets are automatically sized
based on the path MTU, for datagram oriented sockets the user has to size
the datagrams appropiately. When it is enabled the kernel rejects packets
bigger than the path MTU with
EMSGSIZEraw(4)
and
udp(4)
for more information.
IP_MTU
Retrieve the current known path MTU of the current socket.
Only valid when the socket has been connected. Returns an integer. Only valid
as a
getsockopt(2).
IP_ROUTER_ALERT
Pass all forwarded packets with the
IP Router Alert
option
set to this socket. Only valid for raw sockets. This is useful, for instance, for user
space RSVP daemons. Expects an integer argument.
IP_MULTICAST_TTL
Set or reads the time-to-live value of outgoing multicast packets for this
socket. It is
very important for multicast packets to set the smallest TTL possible.
The default is 1 which means that multicast packets don't leave the local
network unless the user program explicitly requests it. Argument is an
integer.
IP_MULTICAST_LOOP
Sets or reads a boolean integer argument whether sent multicast packets should be
looped back to the local sockets.
IP_ADD_MEMBERSHIP
Join a multicast group. Argument is a
struct ip_mreqn
structure.
struct ip_mreqn
{
struct in_addr imr_multiaddr;/* IP multicast group address */
struct in_addr imr_address;/* IP address of local interface */
int imr_ifindex;/* interface index */
};
imr_multiaddr
contains the address of the multicast group the application wants to join or leave.
It must be a valid multicast address.
imr_address
is the address of the local interface with which the system should join the multicast
group; if it is equal to
INADDR_ANY
an appropriate interface is chosen by the system.
imr_ifindex
is the interface index of the interface that should join/leave the
imr_multiaddr
group, or 0 to indicate any interface.
For compatibility, the old
ip_mreq
structure is still supported. It differs from
ip_mreqn
only by not including
the
imr_ifindex
field. Only valid as a
setsockopt(2).
IP_DROP_MEMBERSHIP
Leave a multicast group. Argument is an
ip_mreqn
or
ip_mreq
structure similar to
IP_ADD_MEMBERSHIP.
IP_MULTICAST_IF
Set the local device for a multicast socket. Argument is an
ip_mreqn
or
ip_mreq
structure similar to
IP_ADD_MEMBERSHIP.
When an invalid socket option is passed,
ENOPROTOOPT
is returned.
SYSCTLS
The IP protocol
supports the sysctl interface to configure some global options. The sysctls
can be accessed by reading or writing the
/proc/sys/net/ipv4/*
files or using the
sysctl(2)
interface.
ip_default_ttl
Set the default time-to-live value of outgoing packets. This can be changed
per socket with the
IP_TTL
option.
ip_forward
Enable IP forwarding with a boolean flag. IP forwarding can be also set on a
per interface basis.
ip_dynaddr
Enable dynamic socket address rewriting on interface address change. This is
useful for dialup interface with changing IP addresses.
ip_autoconfig
Not documented.
ip_local_port_range
Contains two integers that define the default local port range allocated to
sockets. Allocation starts with the first number and ends with the second number.
ip_no_pmtu_disc
If enabled, don't do Path MTU Discovery for TCP sockets by default. Path MTU
discovery may fail if misconfigured firewalls (that drop all ICMP packets) or
misconfigured interfaces (e.g., a point-to-point link where the both ends don't
agree on the MTU) are on the path. It is better to fix the broken routers on
the path than to turn off Path MTU Discovery globally, because not doing it
incurs a high cost to the network.
ipfrag_high_thresh and ipfrag_low_thresh
If the amount of queued IP fragments reaches
ipfrag_high_thresh,
the queue
is pruned down to
ipfrag_low_thresh.
Contains an integer with the number of
bytes.
IOCTLS
These ioctls can be accessed using
ioctl(2).
The correct syntax is:
error = ioctl(ip_socket, ioctl_type, value_ptr);
SIOCGSTAMP
Return a
struct timeval
with the receive timestamp of the last packet passed to the user. This is useful
for accurate round trip time measurements. See
setitimer(2)
for a description of
struct timeval.
FIOCSETOWN and SIOCSPGRP
Set the process or process group (negative value passed with a process group id
of the absolute value) to send
SIGIO
or
SIGURG
signals
to when an
asynchronous I/O operation has finished or urgent data is available.
Argument is a
pid_t.
Only processes with effective user id 0 may set this value to
an arbitrary process/group id; all others only to processes/groups with a
matching effective group id or user id.
FIOASYNC
Set a flag to enable or disable asynchronous mode of the socket. Asynchronous
mode means that
SIGIO
is raised when a new I/O event occurs.
See
socket(4)
for a description of the valid IO events.
FIOCGETOWN and SIOCGPGRP
Get the current process or process group that receive
SIGIO
or
SIGURG
signals,
or 0
when none is set. Argument is a
pid_t.
The ioctls to configure firewalling are documented in
ipfw(4)
from the
ipchains
package.
Ioctls to configure generic device parameters are described in
netdevice(4).
NOTES
Be very careful with the
SO_BROADCAST
option - it is not privileged in Linux. It is easy to overload the network
with careless broadcasts. For new application protocols
it is better to use a multicast group instead of broadcasting. Broadcasting
is discouraged.
Some other BSD sockets implementations provide
IP_RCVDSTADDR
and
IP_RECVIF
socket options to get the destination address and the interface of
received datagrams. Linux has the more general
IP_PKTINFO
for the same task.
ERRORS
ENOTCONN
The operation is only defined on a connected socket, but the socket wasn't
connected.
EINVAL
Invalid argument passed.
EMSGSIZE
Datagram is bigger than an MTU on the path and it cannot be fragmented.
EACCES
The user tried to execute an operation without the necessary permissions. These
include sending to a broadcast address without having the broadcast flag set,
trying to modify the firewall settings without effective user id 0 or
CAP_NET_ADMIN,
or trying
to bind to a reserved port without effective user id 0 or
CAP_NET_BIND_SERVICE.
EADDRINUSE
Tried to bind to an address already in use.
ENOMEM and ENOBUFS
Not enough memory available.
ENOPROTOOPT and EOPNOTSUPP
Invalid socket option passed.
EPERM
User doesn't have permission to set high priority, change configuration,
or send signals to the requested process or group,
EADDRNOTAVAIL
A non-existent interface was requested or the requested source address was
not local.
EAGAIN
Operation on a non-blocking socket would block.
ESOCKTNOSUPPORT
The socket is not configured or an unknown socket type was requested.
EISCONN
connect(2)
was called on an already connected socket.
EALREADY
An connection operation on a non-blocking socket is already in progress.
The connection was unexpectedly closed or shut down by the other end.
ENOENT
SIOCGSTAMP
was called on a socket where no packet arrived.
EHOSTUNREACH
No routing table entry matches the destination address.
ENODEV
Network device not available or not capable of sending IP.
ENOPKG
A kernel subsystem was not configured.
Other errors may be generated by the underlying protocols; see
tcp(4),
raw(4),
udp(4)
or the generic socket layer.
VERSIONS
IP_PKTINFO,
IP_MTU,
IP_PMTU_DISCOVER,
IP_PKTINFO,
IP_RECVERR,
and
IP_ROUTER_ALERT
are new options in Linux 2.2.
struct ip_mreqn
is new in Linux 2.2. Linux 2.0 only supported
ip_mreq.
The sysctls were introduced with Linux 2.2.
COMPATIBILITY
For compatibility with Linux 2.0, the obsolete
socket(PF_INET, SOCK_RAW, protocol)
syntax is still supported to open a
packet(4)
socket. This is deprecated and should be replaced by
socket(PF_PACKET, SOCK_RAW, protocol)
instead. The main difference is the
new
sockaddr_ll
address structure for generic link layer information instead of the old
sockaddr_pkt.
BUGS
There are too many inconsistent error values.
The ioctls to configure IP-specific interface options and ARP tables are
not described.
