DEVELOPMENT - socat - Git at Google


 This file should help you to add new address types and address options to
 socat.

 NOTE:
 socat will in future releases be split into a library "libxio" containing all
 the address stuff, useful also for many other purposes, and the socat main()
 and data shuffler. If you intend to perform major changes to the xio part and
 to publish them, please contact me before!


 ADDING A NEW ADDRESS TYPE:

 * Create new files xio-newaddr.c and xio-newaddr.h

 * Create a new record of struct addrdesc in xio-newaddr.c, with declaration in xio-newaddr.h.

 * Make a new entry to addressnames[] in xioopen.c with the addresses main name
 and maybe with alias names. Keep this array ASCII sorted, without uppercase
 chars.

 * config.h.in: #undef WITH_NEWADDR

 * configure.in: Copy the disable part of, e.g., WITH_SOCKS4 and adapt it to
 NEWADDR

 * In socat.c, add to socat_version

 * Write a function xioopen_newaddr() in xio-newaddr.c, declaration in
 xio-newaddr.h
 Do not forget the following option processing calls:
 All groups: _xio_openlate()
 Group FD: applyopts_cloexec()
 Group NAMED: applyopts_file() for phases PREOPEN, OPEN, and FD

 * Describe a tested example in file EXAMPLES, and maybe in the socat manpage
 source.

 * Try to define a test for this address type in test.sh

 * Update file CHANGES


 ADDING A NEW ADDRESS OPTION:

 xioopen.c:

 * If this option depends on a #define that is probably not available on all
 platforms, make all new code for this option dependent on the existence of this
 C header define:
 #ifdef PREFIX_NEWOPTION
 ...
 #endif

 * Add an OPT_NEWOPTION to enum e_optcode in xioopts.h, preferably keeping
 alphabetic order

 * Add a struct optdesc opt_newoption record in xio-newaddr.c and its
 declaration in xio-newaddr.h. The complete structure definition must be in one
 line without breaks for automatic docu extraction.
 Build the record from the following components:
 . A canonical default name (e.g. "newoption")
 . A short, preferable name (e.g. "newopt") or NULL
 . OPT_NEWOPTION (from enum e_optcode, see above)
 . A group membership that restricts appliance of the new option to matching
 address types (e.g., one of GROUP_ANY, GROUP_IP_TCP, GROUP_EXEC)
 . A phase specification that positions this option within address processing.
 Note that the function code can override this value.
 . A representation type for option arguments (e.g., TYPE_INT, TYPE_STRING etc.;
 use TYPE_BOOL if this option just triggers an action)
 . A function or action definition for applying this option. If it does not use
 one of the standard functions (open(), ioctl(), setsockopt()...), then use
 OFUNC_SPEC (specific).

 * For the canonical name and all its aliases and abbreviations, add entries to
 the array optionnames in xioopts.c. KEEP STRICT ALPHABETIC (ASCII) ORDER!
 The entries must be embedded in an IF_... macro of their group for conditional
 compiling.

 * For options using some predefined action (see OFUNC above), this might be
 enough - test the option and document it in doc/socat.yo!
 For OFUNC_SPEC, it might suffice to add another "case" to the OFUNC_SPEC branch
 in applyopts() in xioopts.c. If you need more special handling, you should try
 to understand the address specific functions and add your code there.

 * If you use system or low level C library calls or library calls that might
 hang or induce problems, please invoke them with capitalized name; if no such
 name is defined, add an appropriate debug function to sycls.c, and a header
 entry and a wrapper "define" to sycls.h

 * Update file CHANGES


 INFO ABOUT ADDRESS PHASES:

 Each option entry has a field specifying a default phase for its application.
 Of course, the code that analyses and applies an address may override this
 default phase.

 Depending on the type of address there are several major phase sequences:


 OPEN addresses:

 PH_INIT		retrieving info from original state
 PH_EARLY	before any other processing
 PH_PREOPEN	before file creation/opening (not UNIX sockets)
 PH_OPEN		during file creation/opening (not UNIX sockets)
 PH_PASTOPEN	past file creation/opening (not UNIX sockets)
 PH_FD		soon after FD creation or identification
 PH_LATE		FD is ready, before start of data loop
 PH_LATE2	FD is ready, dropping privileges


 SOCKET addresses:

 PH_INIT		retrieving info from original state
 PH_EARLY	before any other processing
 PH_PRESOCKET	before socket call
 PH_SOCKET	for socket call
 PH_PASTSOCKET	after socket call
 PH_FD		soon after FD creation or identification
 PH_PREBIND	before socket bind()
 PH_BIND		during socket bind()
 PH_PASTBIND	past socket bind()
 PH_PRECONNECT	before connect()
 PH_CONNECT	during connect()
 PH_PASTCONNECT	after connect()
 PH_CONNECTED	phase common with listen
 PH_LATE		FD is ready, before start of data loop
 PH_LATE2	FD is ready, dropping privileges


 SOCKET with LISTEN and ACCEPT:

 PH_INIT		retrieving info from original state
 PH_EARLY	before any other processing
 PH_PRESOCKET	before socket call
 PH_SOCKET	for socket call
 PH_PREBIND	before socket bind()
 PH_BIND		during socket bind()
 PH_PASTBIND	past socket bind()
 PH_PRELISTEN	before listen()
 PH_LISTEN	during listen()
 PH_PASTLISTEN	after listen()
 PH_PREACCEPT	before accept()
 PH_ACCEPT	during accept()
 PH_PASTACCEPT	after accept()
 # and the following on the new FD:
 PH_FD		soon after FD creation or identification
 PH_PASTSOCKET	after socket call
 PH_CONNECTED	phase common with connect
 PH_PREFORK	before forking
 PH_FORK		during fork()
 PH_PASTFORK	after fork()
 PH_LATE		FD is ready, before start of data loop
 PH_LATE2	FD is ready, dropping privileges


 Passive UNIX socket addresses; this is a mix of socket phases and file system phases:
 PH_INIT		retrieving info from original state
 PH_EARLY	before any other processing
 PH_PRESOCKET	before socket call
 PH_SOCKET	for socket call
 PH_PASTSOCKET	after socket call
 PH_FD		soon after FD creation or identification
 PH_PREOPEN	before file creation/opening
 PH_PREBIND	before socket bind()
 PH_BIND		during socket bind()
 PH_PASTOPEN	past file creation/opening
 PH_PASTBIND	past socket bind(), not used up to 1.7.3.4
 PH_PRECONNECT	before connect()
 PH_CONNECT	during connect()
 PH_PASTCONNECT	after connect()
 PH_CONNECTED	phase common with listen
 PH_LATE		FD is ready, before start of data loop
 PH_LATE2	FD is ready, dropping privileges


 FD addresses:

 PH_INIT		retrieving info from original state
 PH_EARLY	before any other processing
 PH_FD		soon after FD identification
 PH_LATE		FD is ready, before start of data loop
 PH_LATE2	FD is ready, dropping privileges


 EXEC addresses:

 PH_INIT		retrieving info from original state
 PH_EARLY	before any other processing
 PH_PREBIGEN	before socketpair() pipe() openpty()
 PH_BIGEN	during socketpair() pipe() openpty()
 PH_PASTBIGEN	past socketpair() pipe() openpty()
 PH_PASTSOCKET	for socketpair()
 PH_FD		soon after FD creation or identification
 PH_PREFORK	before forking
 PH_FORK		during fork()
 PH_PASTFORK	after fork()
 PH_LATE		FD is ready, before start of data loop
 PH_LATE2	FD is ready, dropping privileges
 PH_PREEXEC	before exec() or system()
 PH_EXEC		during exec() or system()


 There are lots of semantic relations between group, phase, and func fields of
 an option.


 There exists something like an overall phase sequence:
 PH_INIT						# su-d.1
 PH_EARLY					# chroot-early
 PH_PREOPEN,	PH_OPEN,	PH_PASTOPEN	# (chroot before/after?)
 PH_PRESOCKET,	PH_SOCKET,	PH_PASTSOCKET	# (su after (root for raw)?)
 PH_PREBIGEN,	PH_BIGEN,	PH_PASTBIGEN	# (chroot before/after (/dev..)?)
 PH_FD
 PH_PREBIND,	PH_BIND,	PH_PASTBIND	# (su after(before?))
 PH_PRELISTEN,	PH_LISTEN,	PH_PASTLISTEN
 PH_PRECONNECT,	PH_CONNECT,	PH_PASTCONNECT	# (chroot before/after (AF_UNIX)?)
 PH_PREACCEPT,	PH_ACCEPT,	PH_PASTACCEPT
 PH_CONNECTED
 PH_PREFORK,	PH_FORK,	PH_PASTFORK	# (all before/after?)
 PH_LATE						# chroot
 PH_LATE2					# su, su-d.2
 PH_PREEXEC,	PH_EXEC				# (all before)

 ===============================================================================
 // Up to 1.7.2.4 socat used non async signal safe system and library calls in signal handlers, mostly for logging purposes. This problem was fixed in release 1.7.3.0 with the following concepts:

 Signal handlers set on entry and unset on return the diag_in_handler global variable. The logging system, when this variable is set, queues the text message together with errno and exit info in a UNIX datagram socket. When invoked with unset diag_in_handler it first checks if there are messages in that queue and prints them first.

 A async signal safe but minimal version of vsnprintf, named vsnprintf_r, was written so no value arguments need to be queued.

 Because strerror is not async signal safe a new function snprinterr was written that replaces the (glibc compatible) %m format with strerror output. The original errno is passed in the message queue, snprinterr is called when dequeuing messages outside of signal handler.

 // List of signal handlers in socat
 socat.c:socat_signal (generic, just logs and maybe exits)
 xioshutdown.c:signal_kill_pid (SIGALRM, kill child process)
 xiosigchld.c:childdied (SIGCHLD: get info, log; possibly close channel)
 xiosignal.c:socatsignalpass: cascades signal to channel child processes; w/ options sighup,sigint,sigquit
 xio-socket.c:xiosigaction_hasread: SIGUSR1,SIGCHLD, tells parent that datagram has been consumed

	This file should help you to add new address types and address options to
	socat.

	NOTE:
	socat will in future releases be split into a library "libxio" containing all
	the address stuff, useful also for many other purposes, and the socat main()
	and data shuffler. If you intend to perform major changes to the xio part and
	to publish them, please contact me before!


	ADDING A NEW ADDRESS TYPE:

	* Create new files xio-newaddr.c and xio-newaddr.h

	* Create a new record of struct addrdesc in xio-newaddr.c, with declaration in xio-newaddr.h.

	* Make a new entry to addressnames[] in xioopen.c with the addresses main name
	and maybe with alias names. Keep this array ASCII sorted, without uppercase
	chars.

	* config.h.in: #undef WITH_NEWADDR

	* configure.in: Copy the disable part of, e.g., WITH_SOCKS4 and adapt it to
	NEWADDR

	* In socat.c, add to socat_version

	* Write a function xioopen_newaddr() in xio-newaddr.c, declaration in
	xio-newaddr.h
	Do not forget the following option processing calls:
	All groups: _xio_openlate()
	Group FD: applyopts_cloexec()
	Group NAMED: applyopts_file() for phases PREOPEN, OPEN, and FD

	* Describe a tested example in file EXAMPLES, and maybe in the socat manpage
	source.

	* Try to define a test for this address type in test.sh

	* Update file CHANGES


	ADDING A NEW ADDRESS OPTION:

	xioopen.c:

	* If this option depends on a #define that is probably not available on all
	platforms, make all new code for this option dependent on the existence of this
	C header define:
	#ifdef PREFIX_NEWOPTION
	...
	#endif

	* Add an OPT_NEWOPTION to enum e_optcode in xioopts.h, preferably keeping
	alphabetic order

	* Add a struct optdesc opt_newoption record in xio-newaddr.c and its
	declaration in xio-newaddr.h. The complete structure definition must be in one
	line without breaks for automatic docu extraction.
	Build the record from the following components:
	. A canonical default name (e.g. "newoption")
	. A short, preferable name (e.g. "newopt") or NULL
	. OPT_NEWOPTION (from enum e_optcode, see above)
	. A group membership that restricts appliance of the new option to matching
	address types (e.g., one of GROUP_ANY, GROUP_IP_TCP, GROUP_EXEC)
	. A phase specification that positions this option within address processing.
	Note that the function code can override this value.
	. A representation type for option arguments (e.g., TYPE_INT, TYPE_STRING etc.;
	use TYPE_BOOL if this option just triggers an action)
	. A function or action definition for applying this option. If it does not use
	one of the standard functions (open(), ioctl(), setsockopt()...), then use
	OFUNC_SPEC (specific).

	* For the canonical name and all its aliases and abbreviations, add entries to
	the array optionnames in xioopts.c. KEEP STRICT ALPHABETIC (ASCII) ORDER!
	The entries must be embedded in an IF_... macro of their group for conditional
	compiling.

	* For options using some predefined action (see OFUNC above), this might be
	enough - test the option and document it in doc/socat.yo!
	For OFUNC_SPEC, it might suffice to add another "case" to the OFUNC_SPEC branch
	in applyopts() in xioopts.c. If you need more special handling, you should try
	to understand the address specific functions and add your code there.

	* If you use system or low level C library calls or library calls that might
	hang or induce problems, please invoke them with capitalized name; if no such
	name is defined, add an appropriate debug function to sycls.c, and a header
	entry and a wrapper "define" to sycls.h

	* Update file CHANGES


	INFO ABOUT ADDRESS PHASES:

	Each option entry has a field specifying a default phase for its application.
	Of course, the code that analyses and applies an address may override this
	default phase.

	Depending on the type of address there are several major phase sequences:


	OPEN addresses:

	PH_INIT retrieving info from original state
	PH_EARLY before any other processing
	PH_PREOPEN before file creation/opening (not UNIX sockets)
	PH_OPEN during file creation/opening (not UNIX sockets)
	PH_PASTOPEN past file creation/opening (not UNIX sockets)
	PH_FD soon after FD creation or identification
	PH_LATE FD is ready, before start of data loop
	PH_LATE2 FD is ready, dropping privileges


	SOCKET addresses:

	PH_INIT retrieving info from original state
	PH_EARLY before any other processing
	PH_PRESOCKET before socket call
	PH_SOCKET for socket call
	PH_PASTSOCKET after socket call
	PH_FD soon after FD creation or identification
	PH_PREBIND before socket bind()
	PH_BIND during socket bind()
	PH_PASTBIND past socket bind()
	PH_PRECONNECT before connect()
	PH_CONNECT during connect()
	PH_PASTCONNECT after connect()
	PH_CONNECTED phase common with listen
	PH_LATE FD is ready, before start of data loop
	PH_LATE2 FD is ready, dropping privileges


	SOCKET with LISTEN and ACCEPT:

	PH_INIT retrieving info from original state
	PH_EARLY before any other processing
	PH_PRESOCKET before socket call
	PH_SOCKET for socket call
	PH_PREBIND before socket bind()
	PH_BIND during socket bind()
	PH_PASTBIND past socket bind()
	PH_PRELISTEN before listen()
	PH_LISTEN during listen()
	PH_PASTLISTEN after listen()
	PH_PREACCEPT before accept()
	PH_ACCEPT during accept()
	PH_PASTACCEPT after accept()
	# and the following on the new FD:
	PH_FD soon after FD creation or identification
	PH_PASTSOCKET after socket call
	PH_CONNECTED phase common with connect
	PH_PREFORK before forking
	PH_FORK during fork()
	PH_PASTFORK after fork()
	PH_LATE FD is ready, before start of data loop
	PH_LATE2 FD is ready, dropping privileges


	Passive UNIX socket addresses; this is a mix of socket phases and file system phases:
	PH_INIT retrieving info from original state
	PH_EARLY before any other processing
	PH_PRESOCKET before socket call
	PH_SOCKET for socket call
	PH_PASTSOCKET after socket call
	PH_FD soon after FD creation or identification
	PH_PREOPEN before file creation/opening
	PH_PREBIND before socket bind()
	PH_BIND during socket bind()
	PH_PASTOPEN past file creation/opening
	PH_PASTBIND past socket bind(), not used up to 1.7.3.4
	PH_PRECONNECT before connect()
	PH_CONNECT during connect()
	PH_PASTCONNECT after connect()
	PH_CONNECTED phase common with listen
	PH_LATE FD is ready, before start of data loop
	PH_LATE2 FD is ready, dropping privileges


	FD addresses:

	PH_INIT retrieving info from original state
	PH_EARLY before any other processing
	PH_FD soon after FD identification
	PH_LATE FD is ready, before start of data loop
	PH_LATE2 FD is ready, dropping privileges


	EXEC addresses:

	PH_INIT retrieving info from original state
	PH_EARLY before any other processing
	PH_PREBIGEN before socketpair() pipe() openpty()
	PH_BIGEN during socketpair() pipe() openpty()
	PH_PASTBIGEN past socketpair() pipe() openpty()
	PH_PASTSOCKET for socketpair()
	PH_FD soon after FD creation or identification
	PH_PREFORK before forking
	PH_FORK during fork()
	PH_PASTFORK after fork()
	PH_LATE FD is ready, before start of data loop
	PH_LATE2 FD is ready, dropping privileges
	PH_PREEXEC before exec() or system()
	PH_EXEC during exec() or system()


	There are lots of semantic relations between group, phase, and func fields of
	an option.


	There exists something like an overall phase sequence:
	PH_INIT # su-d.1
	PH_EARLY # chroot-early
	PH_PREOPEN, PH_OPEN, PH_PASTOPEN # (chroot before/after?)
	PH_PRESOCKET, PH_SOCKET, PH_PASTSOCKET # (su after (root for raw)?)
	PH_PREBIGEN, PH_BIGEN, PH_PASTBIGEN # (chroot before/after (/dev..)?)
	PH_FD
	PH_PREBIND, PH_BIND, PH_PASTBIND # (su after(before?))
	PH_PRELISTEN, PH_LISTEN, PH_PASTLISTEN
	PH_PRECONNECT, PH_CONNECT, PH_PASTCONNECT # (chroot before/after (AF_UNIX)?)
	PH_PREACCEPT, PH_ACCEPT, PH_PASTACCEPT
	PH_CONNECTED
	PH_PREFORK, PH_FORK, PH_PASTFORK # (all before/after?)
	PH_LATE # chroot
	PH_LATE2 # su, su-d.2
	PH_PREEXEC, PH_EXEC # (all before)

	===============================================================================
	// Up to 1.7.2.4 socat used non async signal safe system and library calls in signal handlers, mostly for logging purposes. This problem was fixed in release 1.7.3.0 with the following concepts:

	Signal handlers set on entry and unset on return the diag_in_handler global variable. The logging system, when this variable is set, queues the text message together with errno and exit info in a UNIX datagram socket. When invoked with unset diag_in_handler it first checks if there are messages in that queue and prints them first.

	A async signal safe but minimal version of vsnprintf, named vsnprintf_r, was written so no value arguments need to be queued.

	Because strerror is not async signal safe a new function snprinterr was written that replaces the (glibc compatible) %m format with strerror output. The original errno is passed in the message queue, snprinterr is called when dequeuing messages outside of signal handler.

	// List of signal handlers in socat
	socat.c:socat_signal (generic, just logs and maybe exits)
	xioshutdown.c:signal_kill_pid (SIGALRM, kill child process)
	xiosigchld.c:childdied (SIGCHLD: get info, log; possibly close channel)
	xiosignal.c:socatsignalpass: cascades signal to channel child processes; w/ options sighup,sigint,sigquit
	xio-socket.c:xiosigaction_hasread: SIGUSR1,SIGCHLD, tells parent that datagram has been consumed