freebsd/share/man/man4/syncache.4

.\"
.\" syncache - TCP SYN caching to handle SYN flood DoS.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions
.\" are met:
.\" 1. Redistributions of source code must retain the above copyright
.\"    notice, this list of conditions and the following disclaimer.
.\" 2. Redistributions in binary form must reproduce the above copyright
.\"    notice, this list of conditions and the following disclaimer in the
.\"    documentation and/or other materials provided with the distribution.
.\"
.\" $FreeBSD$
.\"
.Dd August 31, 2001
.Dt SYNCACHE 4
.Os
.Sh NAME
.Nm syncache , syncookies
.Nd
.Xr sysctl 8
MIBs for controlling TCP SYN caching
.Sh SYNOPSIS
.Bl -item -compact
.It
.Nm sysctl Cm net.inet.tcp.syncookies
.El
.Pp
.Bl -item -compact
.It
.Nm sysctl Cm net.inet.tcp.syncache.hashsize
.It
.Nm sysctl Cm net.inet.tcp.syncache.bucketlimit
.It
.Nm sysctl Cm net.inet.tcp.syncache.cachelimit
.It
.Nm sysctl Cm net.inet.tcp.syncache.rexmtlimit
.It
.Nm sysctl Cm net.inet.tcp.syncache.count
.El
.Sh DESCRIPTION
The
.Nm
.Xr sysctl 8
MIB is used to control the TCP SYN caching in the system, which
is intended to handle SYN flood Denial of Service attacks.
.Pp
When a TCP SYN segment is received on a port corresponding to a listen
socket, an entry is made in the
.Nm ,
and a SYN,ACK segment is
returned to the peer.
The
.Nm
entry holds the TCP options from the initial SYN,
enough state to perform a SYN,ACK retransmission, and takes up less
space than a TCP control block endpoint.
An incoming segment which contains an ACK for the SYN,ACK
and matches a
.Nm
entry will cause the system to create a TCP control block
with the options stored in the
.Nm
entry, which is then released.
.Pp
The
.Nm
protects the system from SYN flood DoS attacks by minimizing
the amount of state kept on the server, and by limiting the overall size
of the
.Nm .
.Pp
Transaction TCP accelerated opens (see
.Xr ttcp 4 )
are also supported, and bypass the
.Nm .
.Pp
.Nm Syncookies
provides a way to virtually expand the size of the
.Nm
by keeping state regarding the initial SYN in the network.
Enabling
.Nm syncookies
sends a cryptographic value in the SYN,ACK reply to
the client machine, which is then returned in the client's ACK.
If the corresponding entry is not found in the
.Nm ,
but the value
passes specific security checks, the connection will be accepted.
This is only used if the
.Nm
is unable to handle the volume of
incoming connections, and a prior entry has been evicted from the cache.
.Pp
.Nm Syncookies
have a certain number of disadvantages that a paranoid
administrator may wish to take note of.
Since the TCP options from the initial SYN are not saved, they are not
applied to the connection, precluding use of features like window scale,
timestamps, or exact MSS sizing.
As the returning ACK establishes the connection, it may be possible for
an attacker to ACK flood a machine in an attempt to create a connection.
While steps have been taken to militate this risk, this may provide a way
to bypass firewalls which filter incoming segments with the SYN bit set.
.Pp
The
.Nm
implements a number of variables in
the
.Va net.inet.tcp.syncache
branch of the
.Xr sysctl 3
MIB.
Several of these may be tuned by setting the corresponding
variable in the
.Xr loader 8 .
.Bl -tag -width ".Va bucketlimit"
.It Va hashsize
Size of the
.Nm
hash table, must be a power of 2.
Read-only, tunable via
.Xr loader 8 .
.It Va bucketlimit
Limit on the number of entries permitted in each bucket of the hash table.
This should be left at a low value to minimize search time.
Read-only, tunable via
.Xr loader 8 .
.It Va cachelimit
Limit on the total number of entries in the
.Nm .
Defaults to
.Va ( hashsize No \(mu Va bucketlimit ) ,
may be set lower to minimize memory
consumption.
Read-only, tunable via
.Xr loader 8 .
.It Va rexmtlimit
Maximum number of times a SYN,ACK is retransmitted before being discarded.
The default of 3 retransmits corresponds to a 15 second timeout, this value
may be increased depending on the RTT to client machines.
Tunable via
.Xr sysctl 3 .
.It Va count
Number of entries present in the
.Nm
(read-only).
.El
.Pp
Statistics on the performance of the
.Nm
may be obtained via
.Xr netstat 1 ,
which provides the following counts:
.Bl -tag -width ".Li cookies received"
.It Li "syncache entries added"
Entries successfully inserted in the
.Nm .
.It Li retransmitted
SYN,ACK retransmissions due to a timeout expiring.
.It Li dupsyn
Incoming SYN segment matching an existing entry.
.It Li dropped
SYNs dropped because SYN,ACK could not be sent.
.It Li completed
Successfully completed connections.
.It Li "bucket overflow"
Entries dropped for exceeding per-bucket size.
.It Li "cache overflow"
Entries dropped for exceeding overall cache size.
.It Li reset
RST segment received.
.It Li stale
Entries dropped due to maximum retransmissions or listen socket disappearance.
.It Li aborted
New socket allocation failures.
.It Li badack
Entries dropped due to bad ACK reply.
.It Li unreach
Entries dropped due to ICMP unreachable messages.
.It Li "zone failures"
Failures to allocate new
.Nm
entry.
.It Li "cookies received"
Connections created from segment containing ACK.
.El
.Sh SEE ALSO
.Xr netstat 1 ,
.Xr tcp 4 ,
.Xr ttcp 4 ,
.Xr loader 8 ,
.Xr sysctl 8
.Sh HISTORY
The existing
.Nm
implementation
first appeared in
.Fx 4.5 .
The original concept of a
.Nm
originally appeared in
.Bsx ,
and was later modified by
.Nx ,
then further extended here.
.Sh AUTHORS
The
.Nm
code and manual page were written by
.An Jonathan Lemon Aq jlemon@FreeBSD.org .