sysklogd.8

制作2.6内核的CLFS时 sysklogd-1.5.tar.gz包

files.  You must specifically
modify the configuration files to obtain the enhanced behavior.
.SH SUPPORT FOR REMOTE LOGGING
These modifications provide network support to the syslogd facility.
Network support means that messages can be forwarded from one node
running syslogd to another node running syslogd where they will be
actually logged to a disk file.

To enable this you have to specify the
.B "\-r"
option on the command line.  The default behavior is that
.B syslogd
won't listen to the network.

The strategy is to have syslogd listen on a unix domain socket for
locally generated log messages.  This behavior will allow syslogd to
inter-operate with the syslog found in the standard C library.  At the
same time syslogd listens on the standard syslog port for messages
forwarded from other hosts.  To have this work correctly the
.BR services (5)
files (typically found in
.IR /etc )
must have the following
entry:
.IP
.nf
	syslog          514/udp
.fi
.PP
If this entry is missing
.B syslogd
neither can receive remote messages nor send them, because the UDP
port cant be opened.  Instead 
.B syslogd
will die immediately, blowing out an error message.

To cause messages to be forwarded to another host replace
the normal file line in the
.I syslog.conf
file with the name of the host to which the messages is to be sent
prepended with an @.
.IP
For example, to forward
.B ALL
messages to a remote host use the
following
.I syslog.conf
entry:
.IP
.nf
	# Sample syslogd configuration file to
	# messages to a remote host forward all.
	*.*			@hostname
.fi

To forward all \fBkernel\fP messages to a remote host the
configuration file would be as follows:
.IP
.nf
	# Sample configuration file to forward all kernel
	# messages to a remote host.
	kern.*		@hostname
.fi
.PP

If the remote hostname cannot be resolved at startup, because the
name-server might not be accessible (it may be started after syslogd)
you don't have to worry.
.B Syslogd
will retry to resolve the name ten times and then complain.  Another
possibility to avoid this is to place the hostname in
.IR /etc/hosts .

With normal
.BR syslogd s
you would get syslog-loops if you send out messages that were received
from a remote host to the same host (or more complicated to a third
host that sends it back to the first one, and so on).  In my domain
(Infodrom Oldenburg) we accidently got one and our disks filled up
with the same single message. :-(

To avoid this no messages received from a
remote host are sent out to another (or the same) remote host
anymore.  If you experience are setup in which you need this behaviour,
please use the
.B \-h
command line switch.
However, this option needs to be handled with caution since a syslog
loop can fill up hard disks quite fast.

If the remote host is located in the same domain as the host, 
.B syslogd
is running on, only the simple hostname will be logged instead of
the whole fqdn.

In a local network you may provide a central log server to have all
the important information kept on one machine.  If the network consists
of different domains you don't have to complain about logging fully
qualified names instead of simple hostnames.  You may want to use the
strip-domain feature
.B \-s
of this server.  You can tell the
.B syslogd
to strip off several domains other than the one the server is located
in and only log simple hostnames.

Using the
.B \-l
option there's also a possibility to define single hosts as local
machines.  This, too, results in logging only their simple hostnames
and not the fqdns.

The UDP socket used to forward messages to remote hosts or to receive
messages from them is only opened when it is needed.  In releases
prior to 1.3-23 it was opened every time but not opened for reading or
forwarding respectively.
.SH OUTPUT TO NAMED PIPES (FIFOs)
This version of syslogd has support for logging output to named pipes
(fifos).  A fifo or named pipe can be used as a destination for log
messages by prepending a pipy symbol (``|'') to the name of the
file.  This is handy for debugging.  Note that the fifo must be created
with the mkfifo command before syslogd is started.
.IP
The following configuration file routes debug messages from the
kernel to a fifo:
.IP
.nf
	# Sample configuration to route kernel debugging
	# messages ONLY to /usr/adm/debug which is a
	# named pipe.
	kern.=debug			|/usr/adm/debug
.fi
.LP
.SH INSTALLATION CONCERNS
There is probably one important consideration when installing this
version of syslogd.  This version of syslogd is dependent on proper
formatting of messages by the syslog function.  The functioning of the
syslog function in the shared libraries changed somewhere in the
region of libc.so.4.[2-4].n.  The specific change was to
null-terminate the message before transmitting it to the 
.I /dev/log
socket.  Proper functioning of this version of syslogd is dependent on
null-termination of the message.

This problem will typically manifest itself if old statically linked
binaries are being used on the system.  Binaries using old versions of
the syslog function will cause empty lines to be logged followed by
the message with the first character in the message removed.
Relinking these binaries to newer versions of the shared libraries
will correct this problem.

Both the
.BR syslogd "(8) and the " klogd (8)
can either be run from
.BR init (8)
or started as part of the rc.*
sequence.  If it is started from init the option \fI\-n\fR must be set,
otherwise you'll get tons of syslog daemons started.  This is because 
.BR init (8)
depends on the process ID.
.SH SECURITY THREATS
There is the potential for the syslogd daemon to be
used as a conduit for a denial of service attack.  Thanks go to John
Morrison ([e-mail]) for alerting me to this potential.
A rogue program(mer) could very easily flood the syslogd daemon with
syslog messages resulting in the log files consuming all the remaining
space on the filesystem.  Activating logging over the inet domain
sockets will of course expose a system to risks outside of programs or
individuals on the local machine.

There are a number of methods of protecting a machine:
.IP 1.
Implement kernel firewalling to limit which hosts or networks have
access to the 514/UDP socket.
.IP 2.
Logging can be directed to an isolated or non-root filesystem which,
if filled, will not impair the machine.
.IP 3.
The ext2 filesystem can be used which can be configured to limit a
certain percentage of a filesystem to usage by root only.  \fBNOTE\fP
that this will require syslogd to be run as a non-root process.
\fBALSO NOTE\fP that this will prevent usage of remote logging since
syslogd will be unable to bind to the 514/UDP socket.
.IP 4.
Disabling inet domain sockets will limit risk to the local machine.
.IP 5.
Use step 4 and if the problem persists and is not secondary to a rogue
program/daemon get a 3.5 ft (approx. 1 meter) length of sucker rod*
and have a chat with the user in question.

Sucker rod def. \(em 3/4, 7/8 or 1in. hardened steel rod, male
threaded on each end.  Primary use in the oil industry in Western
North Dakota and other locations to pump 'suck' oil from oil wells.
Secondary uses are for the construction of cattle feed lots and for
dealing with the occasional recalcitrant or belligerent individual.
.SH DEBUGGING
When debugging is turned on using
.B "\-d"
option then
.B syslogd
will be very verbose by writing much of what it does on stdout.  Whenever
the configuration file is reread and re-parsed you'll see a tabular,
corresponding to the internal data structure.  This tabular consists of
four fields:
.TP
.I number
This field contains a serial number starting by zero.  This number
represents the position in the internal data structure (i.e. the
array).  If one number is left out then there might be an error in the
corresponding line in
.IR /etc/syslog.conf .
.TP
.I pattern
This field is tricky and represents the internal structure
exactly.  Every column stands for a facility (refer to
.BR syslog (3)).
As you can see, there are still some facilities left free for former
use, only the left most are used.  Every field in a column represents
the priorities (refer to
.BR syslog (3)).
.TP
.I action
This field describes the particular action that takes place whenever a
message is received that matches the pattern.  Refer to the
.BR syslog.conf (5)
manpage for all possible actions.
.TP
.I arguments
This field shows additional arguments to the actions in the last
field.  For file-logging this is the filename for the logfile; for
user-logging this is a list of users; for remote logging this is the
hostname of the machine to log to; for console-logging this is the
used console; for tty-logging this is the specified tty; wall has no
additional arguments.
.SH FILES
.PD 0
.TP
.I /etc/syslog.conf
Configuration file for
.BR syslogd .
See
.BR syslog.conf (5)
for exact information.
.TP
.I /dev/log
The Unix domain socket to from where local syslog messages are read.
.TP
.I /var/run/syslogd.pid
The file containing the process id of 
.BR syslogd .
.PD
.SH BUGS
If an error occurs in one line the whole rule is ignored.

.B Syslogd
doesn't change the filemode of opened logfiles at any stage of
process.  If a file is created it is world readable.  If you want to
avoid this, you have to create it and change permissions on your own.
This could be done in combination with rotating logfiles using the 
.BR savelog (8)
program that is shipped in the 
.B smail
3.x distribution.  Remember that it might be a security hole if
everybody is able to read auth.* messages as these might contain
passwords.
.SH SEE ALSO
.BR syslog.conf (5),
.BR klogd (8),
.BR logger (1),
.BR syslog (2),
.BR syslog (3),
.BR services (5),
.BR savelog (8).
.SH AUTHORS
The system log daemon
.B syslogd
is originally  taken from BSD sources, Greg Wettstein <[e-mail]>
performed the port to Linux, Martin Schulze <[e-mail]>
fixed some bugs, added several new features and took over maintenance.