NAME
encapsulate - multiplex several channels over a single socket with
sampling of remote process exit status, and provide conversation
termination without closing the socket.
netpipes 4.2
SYNOPSIS
encapsulate --fd n [ --verbose ] [ --subproc [ --infd n[=sid] ] [
--outfd n[=sid] ] [ --duplex n[=sid] ] [ --Duplex n[=sid] ] [ --DUPLEX
n[=sid] ] [ --prefer-local ] [ --prefer-remote ] [ --local-only ] [
--remote-only ] ] [ --client ] [ --server ]
-[#n][v][s[in][on][dn][ion][oin][l][r][L][R]] command args ...
DESCRIPTION
encapsulate implements the Session Control Protocol (SCP) in a limited
manner. encapsulate multiplexes several virtual channels over a single
socket using SCP. encapsulate transmits the exit status of the local
program to the remote end over a reserved SCP channel and receives the
remote exit status back. encapsulate provides conversation boundaries
without closing the socket.
Flags may appear in any order. The first argument that isn’t a flag is
the command to spawn (assuming --subproc is specified, an error
otherwise).
OPTIONS
--fd n, -#n : specify the file descriptor of the socket we will be
multiplexing subprocess channels over. This argument is required
--verbose, -v : Print extra information (including a copyright notice)
to stderr.
--subproc, -s : spawn a subprocess. You must supply a command and
args. If you omit this flag, then you must not supply a command and
args. If you omit this flag, encapsulate will copy input from stdin to
an outgoing channel in the SCP-muxed socket and copy to stdout from an
incoming channel in the SCP-muxed socket. If you omit this flag, all
of the input and output channel flags are illegal.
--infd n, -in : specify an input channel. If there is a subprocess, it
will be able to read from descriptor n. If there is no subprocess
encapsulate will read from its descriptor n (these are opposite
polarities for the SCP channel).
--outfd n, -on : specify an output channel. If there is a subprocess,
it will be able to write to descriptor n. If there is no subprocess
encapsulate will write to its descriptor n (these are opposite
polarities for the SCP channel).
--duplex n, -ion : specify a bidirectional channel. The remote
encapsulate will send the SCP SYN packet, and the local will respond
with a SYN for the same session. The subprocess will be able to read
and write to file descriptor n. The subprocess should use the
sockdown(1) program if it must close one direction while leaving the
other direction open.
--Duplex n, -dn : specify a bidirectional channel. The --client end of
the encapsulate connection sends the SCP SYN packet and --server
responds with a SYN for the same session. The subprocess will be able
to read and write to file descriptor n. The subprocess should use the
sockdown(1) program if it must close one direction while leaving the
other direction open.
--DUPLEX n, -oin : specify a bidirectional channel. The local
encapsulate will send the SCP SYN packet, and the remote will respond
with a SYN for the same session. The subprocess will be able to read
and write to file descriptor n. The subprocess should use the
sockdown(1) program if it must close one direction while leaving the
other direction open.
All of the long forms of the bidirectional channel have an optional
=sid component that can be used to specify the SCP Session ID. This is
not very useful when connecting encapsulate to another instance of
itself, but could be handy when connecting to another piece of software
that implements SCP.
--prefer-local, -l : if both the remote and local subprocesses exit
with non-zero (erroneous) codes, encapsulate will exit with the same
code as the local subprocess. This is the default.
--prefer-remote, -r : if both the remote and local subprocesses exit
with non-zero (erroneous) codes, encapsulate will exit with the same
code as the remote subprocess.
--local-only, -L : encapsulate exits with the local status and ignores
the remote status.
--remote-only, -R : encapsulate exits with the remote status and
ignores the local status.
SESSION IDs AND SUBPROCESS CHANNELS
When specifying channels for the subprocess, the order of the flags is
very important. Every flag to the local encapsulate must have a
corresponding flag on the remote encapsulate that is in the exact same
position (in the list of channels). The descriptor numbers need not
correspond, but the position and type of each channel must.
A lamentably complicating factor is that the data flow implied by
--infd and --outfd are different when you specify a subprocess.
Local Remote
--infd w/subproc --outfd w/subproc
--infd w/subproc --infd
--infd --infd w/subproc
--infd --outfd
--outfd w/subproc --infd w/subproc
--outfd w/subproc --outfd
--outfd --outfd w/subproc
--outfd --infd
--duplex --DUPLEX
--Duplex --Duplex
--DUPLEX --duplex
RIGHT:
l$ encapsulate --infd 0 --duplex 5
r$ encapsulate --outfd 1 --DUPLEX 5
WRONG:
l$ encapsulate --infd 0 --duplex 5
r$ encapsulate --outfd 1 --duplex 5
--duplex must have a corresponding --DUPLEX on the remote end.
l$ encapsulate --infd 0 --duplex 5
r$ encapsulate --DUPLEX 5 --outfd 1
--infd must have a corresponding --outfd on the remote end. It’s out
of order and the channels will be allocated incorrectly leading to
protocol errors.
If you understand the source code for encapsulate, you can violate
these guidelines, but it is unnecessary, error-prone, and ill-advised;
besides, you don’t really understand the source code. Don’t do it.
CLIENT -VS- SERVER
The SCP has an implicit polarity. One end is the server and the other
end is the client. You can specify which end is which using --client
and --server. If you do not specify one, then encapsulate will compare
the addresses of both ends of the socket (specified with --fd) and use
a deterministic algorithm to pick one to be the server and one to be
the client. If the remote address of the socket does not correspond to
the remote encapsulate (e.g. the packets are being forwarded through a
plugged gateway, the addresses are being masqueraded, or are otherwise
percieved inconsistently by the two ends) then this algorithm has a
good chance of "failing" and assigning both to be server or both to be
client.
The only time you should ever let encapsulate choose between client and
server is in interactive situations. It is very likely that a software
system built around encapsulate will be reused in a situation where the
automatic polarity assignment fails.
EXAMPLES
Here’s a simple file transfer daemon:
server$ faucet 3001 --once --fd3 \
sh -c ’while ~/src/netpipes4.0/encapsulate --fd 3 -so5i4 \
sh -c "fname=‘cat 0<&4‘; echo \$fname; cat < \$fname 1>&5"; \
do true; done’
client$ hose server 3001 --retry 10 --delay 1 --fd3 \
sh -c ’while read fname; do \
~/src/netpipes4.0/encapsulate --fd 3 -si4o5 \
sh -c "echo $fname 1>&5; exec 5>&-; cat 0<&4" \
|| break; done’
Just type the name of the file you want to retrieve into the hose and
press return. It will be dumped to stdout. Repeat until enlightened
or bored.
TROUBLESHOOTING
Did you specify --client and --server properly? One side should be
server, the other side should be client. If you specify them both as
server or both as client, you have made a mistake. Do not rely on the
automatic polarity detection. While it is theoretically a very good
algorithm, it is fooled very easily.
Do all of your channel assignments (--infd et al) match up? If you get
these wrong, encapsulate will freak out and drip spooge all over your
shoes.
For deadlock avoidance, make sure you are closing channels when you
don’t need them anymore. Use the >&- redirection operator in sh or
bash. Make sure you close it in all of the background processes as
well.
Unable to read stdin from a process that has been backgrounded with & ?
Bash closes file descriptor 0 for any subprocess that is backgrounded
(e.g. (command&) ). You can get around this by copying 0 onto another
descriptor, and then copying it back within the backgrounded process.
( ( cat 0<&3 ) & ) 3<&0
SEE ALSO
netpipes (1), http://sunsite.unc.edu/ses/scp.html
The Session Control Protocol document on SunSite was a draft. There is
a more recent one that doesn’t specify header compression (which I
don’t use anyway). It may eventually become an RFC. Then again,
encapsulate may be the only program which ever implements SCP.
BUGS
encapsulate is not hard to deadlock. Until I add unbounded buffering
inside encapsulate, avoid constructing deadlock-vulnerable systems.
The encapsulate included with netpipes 4.0 totally failed to handle the
case where no subprocess was specified. No error message would be
issued, and the program would do absolutely nothing. The 4.1 version
should work.
encapsulate has no other known bugs. I’m sure there are unknown ones
because this software is not yet mature; in fact, it’s totally wet
behind the ears. Break it and send me the pieces.
Well, the command-line argument style is inconsistent with faucet &
hose. I’ll be updating faucet & hose.
The Linux kernel from the beginning of time up through version 2.0.29
has a problem with sockets being shut down "too fast". This results in
loss of data at the end of a stream and an "Error: connection reset by
peer" during reads. 2.0.30 supposedly fixes this. This state machine
flaw is very likely present in many other OSes, because the strange
conditions that exercise it are almost nonexistent in normal
applications, but happen all the time in some applications of the
NetPipes package. encapsulate can be used to work around this bug in
some cases because encapsulate does not perform a shutdown on the
network socket ever (it doesn’t even do a "close").
CREDITS
Hi Mom! Hi Dad!
COPYRIGHT
Copyright (C) 1997-98 Robert Forsman
This program is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 2 of the License, or (at your
option) any later version.
This program is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
675 Mass Ave, Cambridge, MA 02139, USA.
AUTHOR
Robert Forsman
thoth@purplefrog.com
Purple Frog Software
http://web.purplefrog.com/~thoth/
June 19, 1997