wireshark/doc/README.developer

$Id$

This file is a HOWTO for Ethereal developers. It describes how to start coding
a Ethereal protocol dissector and the use some of the important functions and
variables.

1. Setting up your protocol dissector code.

This section provides skeleton code for a protocol dissector. It also explains
the basic functions needed to enter values in the traffic summary columns,
add to the protocol tree, and work with registered header fields.

1.1 Code style.

1.1.1 Portability.

Ethereal runs on many platforms, and can be compiled with a number of
different compilers; here are some rules for writing code that will work
on multiple platforms.

Don't use C++-style comments (comments beginning with "//" and running
to the end of the line); Ethereal's dissectors are written in C, and
thus run through C rather than C++ compilers, and not all C compilers
support C++-style comments (GCC does, but IBM's C compiler for AIX, for
example, doesn't do so by default).

Don't use zero-length arrays; not all compilers support them.  If an
array would have no members, just leave it out.

Don't declare variables in the middle of executable code; not all C
compilers support that.  Variables should be declared outside a
function, or at the beginning of a function or compound statement.

Don't use "inline"; not all compilers support it.  If you want to have a
function be an inline function if the compiler supports it, use
G_INLINE_FUNC, which is declared by <glib.h>.  This may not work with
functions declared in header files; if it doesn't work, don't declare
the function in a header file, even if this requires that you not make
it inline on any platform.

Don't use "uchar", "u_char", "ushort", "u_short", "uint", "u_int",
"ulong", or "u_long"; they aren't defined on all platforms.  If you want
an 8-bit unsigned quantity, use "guint8"; if you want an 8-bit character
value with the 8th bit not interpreted as a sign bit, use "guchar"; if
you want a 16-bit unsigned quantity, use "guint16"; if you want a 32-bit
unsigned quantity, use "guint32"; and if you want an "int-sized"
unsigned quantity, use "guint".  Use "%d", "%u", "%x", and "%o" to print
those types; don't use "%ld", "%lu", "%lx", or "%lo", as longs are 64
bits long on many platforms, but "guint32" is 32 bits long.

Don't use "long" to mean "signed 32-bit integer", and don't use
"unsigned long" to mean "unsigned 32-bit integer"; "long"s are 64 bits
long on many platforms.  Use "gint32" for signed 32-bit integers and use
"guint32" for unsigned 32-bit integers.

Don't use "long" to mean "signed 64-bit integer" and don't use "unsigned
long" to mean "unsigned 64-bit integer"; "long"s are 32 bits long on
other many platforms.  Don't use "long long" or "unsigned long long",
either, as not all platforms support them; use "gint64" or "guint64",
which will be defined as the appropriate types for 64-bit signed and
unsigned integers.

When printing or displaying the values of 64-bit integral data types,
don't assume use "%lld", "%llu", "%llx", or "%llo" - not all platforms
support "%ll" for printing 64-bit integral data types.  Instead, use
PRId64, PRIu64, PRIx64, and PRIo64, for example

    proto_tree_add_text(tree, tvb, offset, 8,
			"Sequence Number: %" PRIu64, sequence_number);

When specifying an integral constant that doesn't fit in 32 bits, don't
use "LL" at the end of the constant - not all compilers use "LL" for
that.  Instead, put the constant in a call to the "G_GINT64_CONSTANT()"
macro, e.g.

	G_GINT64_CONSTANT(11644473600U)

rather than

	11644473600ULL

Don't use a label without a statement following it.  For example,
something such as

	if (...) {

		...

	done:
	}
	
will not work with all compilers - you have to do

	if (...) {

		...

	done:
		;
	}

with some statement, even if it's a null statement, after the label.

Don't use "bzero()", "bcopy()", or "bcmp()"; instead, use the ANSI C
routines

	"memset()" (with zero as the second argument, so that it sets
	all the bytes to zero);

	"memcpy()" or "memmove()" (note that the first and second
	arguments to "memcpy()" are in the reverse order to the
	arguments to "bcopy()"; note also that "bcopy()" is typically
	guaranteed to work on overlapping memory regions, while
	"memcpy()" isn't, so if you may be copying from one region to a
	region that overlaps it, use "memmove()", not "memcpy()" - but
	"memcpy()" might be faster as a result of not guaranteeing
	correct operation on overlapping memory regions);

	and "memcmp()" (note that "memcmp()" returns 0, 1, or -1, doing
	an ordered comparison, rather than just returning 0 for "equal"
	and 1 for "not equal", as "bcmp()" does).

Not all platforms necessarily have "bzero()"/"bcopy()"/"bcmp()", and
those that do might not declare them in the header file on which they're
declared on your platform.

Don't use "index()" or "rindex()"; instead, use the ANSI C equivalents,
"strchr()" and "strrchr()".  Not all platforms necessarily have
"index()" or "rindex()", and those that do might not declare them in the
header file on which they're declared on your platform.

Don't fetch data from packets by getting a pointer to data in the packet
with "tvb_get_ptr()", casting that pointer to a pointer to a structure,
and dereferencing that pointer.  That point won't necessarily be aligned
on the proper boundary, which can cause crashes on some platforms (even
if it doesn't crash on an x86-based PC); furthermore, the data in a
packet is not necessarily in the byte order of the machine on which
Ethereal is running.  Use the tvbuff routines to extract individual
items from the packet, or use "proto_tree_add_item()" and let it extract
the items for you.

Don't use "ntohs()", "ntohl()", "htons()", or "htonl()"; the header
files required to define or declare them differ between platforms, and
you might be able to get away with not including the appropriate header
file on your platform but that might not work on other platforms. 
Instead, use "g_ntohs()", "g_ntohl()", "g_htons()", and "g_htonl()";
those are declared by <glib.h>, and you'll need to include that anyway,
as Ethereal header files that all dissectors must include use stuff from
<glib.h>.

Don't put a comma after the last element of an enum - some compilers may
either warn about it (producing extra noise) or refuse to accept it.

Don't include <unistd.h> without protecting it with

	#ifdef HAVE_UNISTD_H

		...

	#endif

and, if you're including it to get routines such as "open()", "close()",
"read()", and "write()" declared, also include <io.h> if present:

	#ifdef HAVE_IO_H
	#include <io.h>
	#endif

in order to declare the Windows C library routines "_open()",
"_close()", "_read()", and "_write()".  Your file must include <glib.h>
- which many of the Ethereal header files include, so you might not have
to include it explicitly - in order to get "open()", "close()",
"read()", "write()", etc. mapped to "_open()", "_close()", "_read()",
"_write()", etc..

When opening a file with "fopen()", "freopen()", or "fdopen()", if the
file contains ASCII text, use "r", "w", "a", and so on as the open mode
- but if it contains binary data, use "rb", "wb", and so on.  On
Windows, if a file is opened in a text mode, writing a byte with the
value of octal 12 (newline) to the file causes two bytes, one with the
value octal 15 (carriage return) and one with the value octal 12, to be
written to the file, and causes bytes with the value octal 15 to be
discarded when reading the file (to translate between C's UNIX-style
lines that end with newline and Windows' DEC-style lines that end with
carriage return/line feed).

In addition, that also means that when opening or creating a binary
file, you must use "open()" (with O_CREAT and possibly O_TRUNC if the
file is to be created if it doesn't exist), and OR in the O_BINARY flag. 
That flag is not present on most, if not all, UNIX systems, so you must
also do

	#ifndef O_BINARY
	#define O_BINARY	0
	#endif

to properly define it for UNIX (it's not necessary on UNIX).

Don't use forward declarations of static arrays without a specified size
in a fashion such as this:

	static const value_string foo_vals[];

		...

	static const value_string foo_vals[] = {
		{ 0,		"Red" },
		{ 1,		"Green" },
		{ 2,		"Blue" },
		{ 0,		NULL }
	};

as some compilers will reject the first of those statements.  Instead,
initialize the array at the point at which it's first declared, so that
the size is known.

Don't put declarations in the middle of a block; put them before all
code.  Not all compilers support declarations in the middle of code,
such as

	int i;

	i = foo();

	int j;

For #define names and enum member names, prefix the names with a tag so
as to avoid collisions with other names - this might be more of an issue
on Windows, as it appears to #define names such as DELETE and
OPTIONAL.

Don't use the "numbered argument" feature that many UNIX printf's
implement, e.g.:

	sprintf(add_string, " - (%1$d) (0x%1$04x)", value);

as not all UNIX printf's implement it, and Windows printf doesn't appear
to implement it.  Use something like

	sprintf(add_string, " - (%d) (0x%04x)", value, value);

instead.

Don't use "variadic macros", such as

	#define DBG(format, args...)	fprintf(stderr, format, ## args)

as not all C compilers support them.  Use macros that take a fixed
number of arguments, such as

	#define DBG0(format)		fprintf(stderr, format)
	#define DBG1(format, arg1)	fprintf(stderr, format, arg1)
	#define DBG2(format, arg1, arg2) fprintf(stderr, format, arg1, arg2)

		...

or something such as

	#define DBG(args)		printf args

snprintf() -> g_snprintf()
snprintf() is not available on all platforms, so it's a good idea to use the 
g_snprintf() function declared by <glib.h> instead.

tmpnam() -> mkstemp()
tmpnam is insecure and should not be used any more. Ethereal brings its
own mkstemp implementation for use on platforms that lack mkstemp.
Note: mkstemp does not accept NULL as a parameter.

The pointer retured by a call to "tvb_get_ptr()" is not guaranteed to be
aligned on any particular byte boundary; this means that you cannot
safely cast it to any data type other than a pointer to "char",
"unsigned char", "guint8", or other one-byte data types.  You cannot,
for example, safely cast it to a pointer to a structure, and then access
the structure members directly; on some systems, unaligned accesses to
integral data types larger than 1 byte, and floating-point data types,
cause a trap, which will, at best, result in the OS slowly performing an
unaligned access for you, and will, on at least some platforms, cause
the program to be terminated.

Ethereal supports both platforms with GLib 1.2[.x]/GTK+ 1.2[.x] and GLib
2.x/GTK+ 1.3[.x] and 2.x.  If at all possible, either use only
mechanisms that are present in GLib 1.2[.x] and GTK+ 1.2[.x], use #if's
to conditionally use older or newer mechanisms depending on the platform
on which Ethereal is being built, or, if the code in GLib or GTK+ that
implements that mechanism will build with GLib 1.2[.x]/GTK+ 1.2[.x],
conditionally include that code as part of the Ethereal source and use
the included version with GLib 1.2[.x] or GTK+ 1.2[.x].  In particular,
if the GLib 2.x or GTK+ 2.x mechanism indicates that a routine is
deprecated and shouldn't be used in new code, and that it was renamed in
GLib 2.x or GTK+ 2.x and the new name should be used, disregard that and
use the old name - it'll still work with GLib 2.x or GTK+ 2.x, but will
also work with GLib 1.2[.x] and GTK+ 1.2[.x].

When different code must be used on UN*X and Win32, use a #if or #ifdef
that tests _WIN32, not WIN32.  Try to write code portably whenever
possible, however; note that there are some routines in Ethereal with
platform-dependent implementations and platform-independent APIs, such
as the routines in epan/filesystem.c, allowing the code that calls it to
be written portably without #ifdefs.

1.1.2 Robustness.

Ethereal is not guaranteed to read only network traces that contain
correctly-formed packets; in fact, one of the reasons why Ethereal is
used is to track down networking problems, and the problems might be due
to a buggy protocol implementation sending out bad packets.

Therefore, protocol dissectors not only have to be able to handle
correctly-formed packets without, for example, crashing or looping
infinitely, they also have to be able to handle *incorrectly*-formed
packets without crashing or looping infinitely.

Here are some suggestions for making dissectors more robust in the face
of incorrectly-formed packets:

If you are allocating a chunk of memory to contain data from a packet,
or to contain information derived from data in a packet, and the size of
the chunk of memory is derived from a size field in the packet, make
sure all the data is present in the packet before allocating the buffer.
Doing so means that

	1) Ethereal won't leak that chunk of memory if an attempt to
	   fetch data not present in the packet throws an exception

and

	2) it won't crash trying to allocate an absurdly-large chunk of
	   memory if the size field has a bogus large value.

If you're fetching into such a chunk of memory a string from the buffer,
and the string has a specified size, you can use "tvb_get_string()",
which will check whether the entire string is present before allocating
a buffer for the string, and will also put a trailing '\0' at the end of
the buffer.

If you're fetching into such a chunk of memory a 2-byte Unicode string
from the buffer, and the string has a specified size, you can use
"tvb_fake_unicode()", which will check whether the entire string is
present before allocating a buffer for the string, and will also put a
trailing '\0' at the end of the buffer.  The resulting string will be a
sequence of single-byte characters; the only Unicode characters that
will be handled correctly are those in the ASCII range.  (Ethereal's
ability to handle non-ASCII strings is limited; it needs to be
improved.)

If you're fetching into such a chunk of memory a sequence of bytes from
the buffer, and the sequence has a specified size, you can use
"tvb_memdup()", which will check whether the entire sequence is present
before allocating a buffer for it.

Otherwise, you can check whether the data is present by using
"tvb_ensure_bytes_exist()" or by getting a pointer to the data by using
"tvb_get_ptr()", although note that there might be problems with using
the pointer from "tvb_get_ptr()" (see the item on this in the
Portability section above, and the next item below).

If you have gotten a pointer using "tvb_get_ptr()", you must make sure
that you do not refer to any data past the length passed as the last
argument to "tvb_get_ptr()"; while the various "tvb_get" routines
perform bounds checking and throw an exception if you refer to data not
available in the tvbuff, direct references through a pointer gotten from
"tvb_get_ptr()" do not do any bounds checking.

If you have a loop that dissects a sequence of items, each of which has
a length field, with the offset in the tvbuff advanced by the length of
the item, then, if the length field is the total length of the item, and
thus can be zero, you *MUST* check for a zero-length item and abort the
loop if you see one.  Otherwise, a zero-length item could cause the
dissector to loop infinitely.  You should also check that the offset,
after having the length added to it, is greater than the offset before
the length was added to it, if the length field is greater than 24 bits
long, so that, if the length value is *very* large and adding it to the
offset causes an overflow, that overflow is detected.

Any tvbuff offset that is added to as processing is done on a packet
should be stored in a 32-bit variable, such as an "int"; if you store it
in an 8-bit or 16-bit variable, you run the risk of the variable
overflowing.

sprintf() -> g_snprintf()
Prevent yourself from using the sprintf() function, as it does not test the 
length of the given output buffer and might be writing into memory areas not 
intended for. This function is one of the main causes of security problems 
like buffer exploits and many other bugs that are very hard to find. It's 
much better to use the g_snprintf() function declared by <glib.h> instead.

1.1.3 Name convention.

Ethereal uses the underscore_convention rather than the InterCapConvention for
function names, so new code should probably use underscores rather than
intercaps for functions and variable names. This is especially important if you
are writing code that will be called from outside your code.  We are just
trying to keep things consistent for other users.

1.1.4 White space convention.

Avoid using tab expansions different from 8 spaces, as not all text editors in
use by the developers support this.

When creating a new file, you are free to choose an indentation logic. Most of
the files in Ethereal tend to use 2-space or 4-space indentation. You are
encouraged to write a short comment on the indentation logic at the beginning
of this new file.

When editing an existing file, try following the existing indentation logic.

1.2 Skeleton code.

Ethereal requires certain things when setting up a protocol dissector. 
Below is skeleton code for a dissector that you can copy to a file and
fill in.  Your dissector should follow the naming convention of packet-
followed by the abbreviated name for the protocol.  It is recommended
that where possible you keep to the IANA abbreviated name for the
protocol, if there is one, or a commonly-used abbreviation for the
protocol, if any.

Dissectors that use the dissector registration to tell a lower level
dissector don't need to define a prototype in the .h file. For other
dissectors the main dissector routine should have a prototype in a header
file whose name is "packet-", followed by the abbreviated name for the
protocol, followed by ".h"; any dissector file that calls your dissector
should be changed to include that file.

You may not need to include all the headers listed in the skeleton
below, and you may need to include additional headers.  For example, the
code inside

	#ifdef HAVE_LIBPCRE

		...

	#endif

is needed only if you are using a function from libpcre, e.g. the
"pcre_compile()" function.

The "$Id$"
in the comment will be updated by CVS when the file is
checked in; it will allow the RCS "ident" command to report which
version of the file is currently checked out.

When creating a new file, it is fine to just write "$Id$" as RCS will
automatically fill in the identifier at the time the file will be added to the
CVS repository (checked in).

------------------------------------Cut here------------------------------------
/* packet-PROTOABBREV.c
 * Routines for PROTONAME dissection
 * Copyright 2000, YOUR_NAME <YOUR_EMAIL_ADDRESS>
 *
 * $Id$
 *
 * Ethereal - Network traffic analyzer
 * By Gerald Combs <gerald@ethereal.com>
 * Copyright 1998 Gerald Combs
 *
 * Copied from WHATEVER_FILE_YOU_USED (where "WHATEVER_FILE_YOU_USED"
 * is a dissector file; if you just copied this from README.developer,
 * don't bother with the "Copied from" - you don't even need to put
 * in a "Copied from" if you copied an existing dissector, especially
 * if the bulk of the code in the new dissector is your code)
 * 
 * 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., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
 */

#ifdef HAVE_CONFIG_H
# include "config.h"
#endif

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include <glib.h>

#include <epan/packet.h>

/* IF PROTO exposes code to other dissectors, then it must be exported
   in a header file. If not, a header file is not needed at all. */
#include "packet-PROTOABBREV.h"

/* Initialize the protocol and registered fields */
static int proto_PROTOABBREV = -1;
static int hf_PROTOABBREV_FIELDABBREV = -1;

/* Initialize the subtree pointers */
static gint ett_PROTOABBREV = -1;

/* Code to actually dissect the packets */
static void
dissect_PROTOABBREV(tvbuff_t *tvb, packet_info *pinfo, proto_tree *tree)
{

/* Set up structures needed to add the protocol subtree and manage it */
	proto_item *ti;
	proto_tree *PROTOABBREV_tree;

/* Make entries in Protocol column and Info column on summary display */
	if (check_col(pinfo->cinfo, COL_PROTOCOL)) 
		col_set_str(pinfo->cinfo, COL_PROTOCOL, "PROTOABBREV");
    
/* This field shows up as the "Info" column in the display; you should make
   it, if possible, summarize what's in the packet, so that a user looking
   at the list of packets can tell what type of packet it is. See section 1.5
   for more information.

   If you are setting it to a constant string, use "col_set_str()", as
   it's more efficient than the other "col_set_XXX()" calls.

   If you're setting it to a string you've constructed, or will be
   appending to the column later, use "col_add_str()".

   "col_add_fstr()" can be used instead of "col_add_str()"; it takes
   "printf()"-like arguments.  Don't use "col_add_fstr()" with a format
   string of "%s" - just use "col_add_str()" or "col_set_str()", as it's
   more efficient than "col_add_fstr()".

   If you will be fetching any data from the packet before filling in
   the Info column, clear that column first, in case the calls to fetch
   data from the packet throw an exception because they're fetching data
   past the end of the packet, so that the Info column doesn't have data
   left over from the previous dissector; do

	if (check_col(pinfo->cinfo, COL_INFO)) 
		col_clear(pinfo->cinfo, COL_INFO);

   */

	if (check_col(pinfo->cinfo, COL_INFO)) 
		col_set_str(pinfo->cinfo, COL_INFO, "XXX Request");

/* A protocol dissector can be called in 2 different ways:

	(a) Operational dissection

		In this mode, Ethereal is only interested in the way protocols
		interact, protocol conversations are created, packets are reassembled
		and handed over to higher-level protocol dissectors.
		This way Ethereal does not build a so-called "protocol tree".

	(b) Detailed dissection

		In this mode, Ethereal is also interested in all details of a given
		protocol, so a "protocol tree" is created.

   Ethereal distinguishes between the 2 modes with the proto_tree pointer:
	(a) <=> tree == NULL
	(b) <=> tree != NULL

   In the interest of speed, if "tree" is NULL, avoid building a
   protocol tree and adding stuff to it, or even looking at any packet
   data needed only if you're building the protocol tree, if possible.

   Note, however, that you must fill in column information, create
   conversations, reassemble packets, build any other persistent state
   needed for dissection, and call subdissectors regardless of whether
   "tree" is NULL or not.  This might be inconvenient to do without
   doing most of the dissection work; the routines for adding items to
   the protocol tree can be passed a null protocol tree pointer, in
   which case they'll return a null item pointer, and
   "proto_item_add_subtree()" returns a null tree pointer if passed a
   null item pointer, so, if you're careful not to dereference any null
   tree or item pointers, you can accomplish this by doing all the
   dissection work.  This might not be as efficient as skipping that
   work if you're not building a protocol tree, but if the code would
   have a lot of tests whether "tree" is null if you skipped that work,
   you might still be better off just doing all that work regardless of
   whether "tree" is null or not. */
	if (tree) {

/* NOTE: The offset and length values in the call to
   "proto_tree_add_item()" define what data bytes to highlight in the hex
   display window when the line in the protocol tree display
   corresponding to that item is selected.

   Supplying a length of -1 is the way to highlight all data from the
   offset to the end of the packet. */

/* create display subtree for the protocol */
		ti = proto_tree_add_item(tree, proto_PROTOABBREV, tvb, 0, -1, FALSE);

		PROTOABBREV_tree = proto_item_add_subtree(ti, ett_PROTOABBREV);

/* add an item to the subtree, see section 1.6 for more information */
		proto_tree_add_item(PROTOABBREV_tree,
		    hf_PROTOABBREV_FIELDABBREV, tvb, offset, len, FALSE)


/* Continue adding tree items to process the packet here */


	}

/* If this protocol has a sub-dissector call it here, see section 1.8 */
}


/* Register the protocol with Ethereal */

/* this format is require because a script is used to build the C function
   that calls all the protocol registration.
*/

void
proto_register_PROTOABBREV(void)
{                 

/* Setup list of header fields  See Section 1.6.1 for details*/
	static hf_register_info hf[] = {
		{ &hf_PROTOABBREV_FIELDABBREV,
			{ "FIELDNAME",           "PROTOABBREV.FIELDABBREV",
			FIELDTYPE, FIELDBASE, FIELDCONVERT, BITMASK,          
			"FIELDDESCR", HFILL }
		},
	};

/* Setup protocol subtree array */
	static gint *ett[] = {
		&ett_PROTOABBREV,
	};

/* Register the protocol name and description */
	proto_PROTOABBREV = proto_register_protocol("PROTONAME",
	    "PROTOSHORTNAME", "PROTOABBREV");

/* Required function calls to register the header fields and subtrees used */
	proto_register_field_array(proto_PROTOABBREV, hf, array_length(hf));
	proto_register_subtree_array(ett, array_length(ett));
}


/* If this dissector uses sub-dissector registration add a registration routine.
   This format is required because a script is used to find these routines and
   create the code that calls these routines.
*/
void
proto_reg_handoff_PROTOABBREV(void)
{
	dissector_handle_t PROTOABBREV_handle;

	PROTOABBREV_handle = create_dissector_handle(dissect_PROTOABBREV,
	    proto_PROTOABBREV);
	dissector_add("PARENT_SUBFIELD", ID_VALUE, PROTOABBREV_handle);
}

------------------------------------Cut here------------------------------------

1.3 Explanation of needed substitutions in code skeleton.

In the above code block the following strings should be substituted with
your information.

YOUR_NAME	Your name, of course.  You do want credit, don't you?
		It's the only payment you will receive....
YOUR_EMAIL_ADDRESS	Keep those cards and letters coming.
WHATEVER_FILE_YOU_USED	Add this line if you are using another file as a
		starting point.
PROTONAME	The name of the protocol; this is displayed in the
		top-level protocol tree item for that protocol.
PROTOSHORTNAME	An abbreviated name for the protocol; this is displayed
		in the "Preferences" dialog box if your dissector has
		any preferences, and in the dialog box for filter fields
		when constructing a filter expression.
PROTOABBREV	A name for the protocol for use in filter expressions;
		it should contain only lower-case letters, digits, and
		hyphens.
FIELDNAME	The displayed name for the header field.
FIELDABBREV	The abbreviated name for the header field. (NO SPACES)
FIELDTYPE	FT_NONE, FT_BOOLEAN, FT_UINT8, FT_UINT16, FT_UINT24,
		FT_UINT32, FT_UINT64, FT_INT8, FT_INT16, FT_INT24, FT_INT32,
		FT_INT64, FT_FLOAT, FT_DOUBLE, FT_ABSOLUTE_TIME,
		FT_RELATIVE_TIME, FT_STRING, FT_STRINGZ, FT_UINT_STRING,
		FT_ETHER, FT_BYTES, FT_IPv4, FT_IPv6, FT_IPXNET,
		FT_FRAMENUM
FIELDBASE	BASE_NONE, BASE_DEC, BASE_HEX, BASE_OCT
FIELDCONVERT	VALS(x), TFS(x), NULL
BITMASK		Usually 0x0 unless using the TFS(x) field conversion.
FIELDDESCR	A brief description of the field.
PARENT_SUBFIELD	Lower level protocol field used for lookup, i.e. "tcp.port"
ID_VALUE	Lower level protocol field value that identifies this protocol
		For example the TCP or UDP port number

If, for example, PROTONAME is "Internet Bogosity Discovery Protocol",
PROTOSHORTNAME would be "IBDP", and PROTOABBREV would be "ibdp".  Try to
conform with IANA names.

1.4 The dissector and the data it receives.


1.4.1 Header file.

This is only needed if the dissector doesn't use self-registration to
register itself with the lower level dissector, or if the protocol dissector
wants/needs to expose code to other subdissectors.

The dissector has the following header that must be placed into
packet-PROTOABBREV.h.

void
dissect_PROTOABBREV(tvbuff_t *tvb, packet_info *pinfo, proto_tree *tree);


1.4.2 Extracting data from packets.

NOTE: See the README.tvbuff for more details

The "tvb" argument to a dissector points to a buffer containing the raw
data to be analyzed by the dissector; for example, for a protocol
running atop UDP, it contains the UDP payload (but not the UDP header,
or any protocol headers above it).  A tvbuffer is a opaque data
structure, the internal data structures are hidden and the data must be
access via the tvbuffer accessors.

The accessors are:

Single-byte accessor:

guint8  tvb_get_guint8(tvbuff_t*, gint offset);

Network-to-host-order accessors for 16-bit integers (guint16), 32-bit
integers (guint32), and 24-bit integers:

guint16 tvb_get_ntohs(tvbuff_t*, gint offset);
guint32 tvb_get_ntohl(tvbuff_t*, gint offset);
guint32 tvb_get_ntoh24(tvbuff_t*, gint offset);

Network-to-host-order accessors for single-precision and
double-precision IEEE floating-point numbers:

gfloat tvb_get_ntohieee_float(tvbuff_t*, gint offset);
gdouble tvb_get_ntohieee_double(tvbuff_t*, gint offset);

Little-Endian-to-host-order accessors for 16-bit integers (guint16),
32-bit integers (guint32), and 24-bit integers:

guint16 tvb_get_letohs(tvbuff_t*, gint offset);
guint32 tvb_get_letohl(tvbuff_t*, gint offset);
guint32 tvb_get_letoh24(tvbuff_t*, gint offset);

Little-Endian-to-host-order accessors for single-precision and
double-precision IEEE floating-point numbers:

gfloat tvb_get_letohieee_float(tvbuff_t*, gint offset);
gdouble tvb_get_letohieee_double(tvbuff_t*, gint offset);

NOTE: IPv4 addresses are not to be converted to host byte order before
being passed to "proto_tree_add_ipv4()".  You should use "tvb_memcpy()"
to fetch them, not "tvb_get_ntohl()" *OR* "tvb_get_letohl()" - don't,
for example, try to use "tvb_get_ntohl()", find that it gives you the
wrong answer on the PC on which you're doing development, and try
"tvb_get_letohl()" instead, as "tvb_get_letohl()" will give the wrong
answer on big-endian machines.

String accessors:

guint8 *tvb_get_string(tvbuff_t*, gint offset, gint length);

Returns a null-terminated buffer, allocated with "g_malloc()" (so it
must be freed with "g_free()"), containing data from the specified
tvbuff, starting at the specified offset, and containing the specified
length worth of characters (the length of the buffer will be length+1,
as it includes a null character to terminate the string).

guint8 *tvb_get_stringz(tvbuff_t *tvb, gint offset, gint *lengthp);

Returns a null-terminated buffer, allocated with "g_malloc()",
containing data from the specified tvbuff, starting with at the
specified offset, and containing all characters from the tvbuff up to
and including a terminating null character in the tvbuff.  "*lengthp"
will be set to the length of the string, including the terminating null.

Copying memory:
guint8* tvb_memcpy(tvbuff_t*, guint8* target, gint offset, gint length);

Copies into the specified target the specified length's worth of data
from the specified tvbuff, starting at the specified offset.

guint8* tvb_memdup(tvbuff_t*, gint offset, gint length);

Returns a buffer, allocated with "g_malloc()", containing the specified
length's worth of data from the specified tvbuff, starting at the
specified offset.

Pointer-retrieval:
/* WARNING! This function is possibly expensive, temporarily allocating
 * another copy of the packet data. Furthermore, it's dangerous because once
 * this pointer is given to the user, there's no guarantee that the user will
 * honor the 'length' and not overstep the boundaries of the buffer.
 */ 
guint8* tvb_get_ptr(tvbuff_t*, gint offset, gint length);

The reason that tvb_get_ptr() might have to allocate a copy of its data
only occurs with TVBUFF_COMPOSITES, data that spans multiple tvbuffers. 
If the user request a pointer to a range of bytes that spans the member
tvbuffs that make up the TVBUFF_COMPOSITE, the data will have to be
copied to another memory region to assure that all the bytes are
contiguous.



1.5 Functions to handle columns in the traffic summary window.

The topmost pane of the main window is a list of the packets in the
capture, possibly filtered by a display filter.

Each line corresponds to a packet, and has one or more columns, as
configured by the user.

Many of the columns are handled by code outside individual dissectors;
most dissectors need only specify the value to put in the "Protocol" and
"Info" columns.

Columns are specified by COL_ values; the COL_ value for the "Protocol"
field, typically giving an abbreviated name for the protocol (but not
the all-lower-case abbreviation used elsewhere) is COL_PROTOCOL, and the
COL_ value for the "Info" field, giving a summary of the contents of the
packet for that protocol, is COL_INFO. 

A value for a column should only be added if the user specified that it
be displayed; to check whether a given column is to be displayed, call
'check_col' with the COL_ value for that field as an argument - it will
return TRUE if the column is to be displayed and FALSE if it is not to
be displayed.

The value for a column can be specified with one of several functions,
all of which take the 'fd' argument to the dissector as their first
argument, and the COL_ value for the column as their second argument.

1.5.1 The col_set_str function.

'col_set_str' takes a string as its third argument, and sets the value
for the column to that value.  It assumes that the pointer passed to it
points to a string constant or a static "const" array, not to a
variable, as it doesn't copy the string, it merely saves the pointer
value; the argument can itself be a variable, as long as it always
points to a string constant or a static "const" array.

It is more efficient than 'col_add_str' or 'col_add_fstr'; however, if
the dissector will be using 'col_append_str' or 'col_append_fstr" to
append more information to the column, the string will have to be copied
anyway, so it's best to use 'col_add_str' rather than 'col_set_str' in
that case.

For example, to set the "Protocol" column
to "PROTOABBREV":

	if (check_col(pinfo->cinfo, COL_PROTOCOL)) 
		col_set_str(pinfo->cinfo, COL_PROTOCOL, "PROTOABBREV");


1.5.2 The col_add_str function.

'col_add_str' takes a string as its third argument, and sets the value
for the column to that value.  It takes the same arguments as
'col_set_str', but copies the string, so that if the string is, for
example, an automatic variable that won't remain in scope when the
dissector returns, it's safe to use.


1.5.3 The col_add_fstr function.

'col_add_fstr' takes a 'printf'-style format string as its third
argument, and 'printf'-style arguments corresponding to '%' format
items in that string as its subsequent arguments.  For example, to set
the "Info" field to "<XXX> request, <N> bytes", where "reqtype" is a
string containing the type of the request in the packet and "n" is an
unsigned integer containing the number of bytes in the request:

	if (check_col(pinfo->cinfo, COL_INFO)) 
		col_add_fstr(pinfo->cinfo, COL_INFO, "%s request, %u bytes",
		    reqtype, n);

Don't use 'col_add_fstr' with a format argument of just "%s" -
'col_add_str', or possibly even 'col_set_str' if the string that matches
the "%s" is a static constant string, will do the same job more
efficiently.


1.5.4 The col_clear function.

If the Info column will be filled with information from the packet, that
means that some data will be fetched from the packet before the Info
column is filled in.  If the packet is so small that the data in
question cannot be fetched, the routines to fetch the data will throw an
exception (see the comment at the beginning about tvbuffers improving
the handling of short packets - the tvbuffers keep track of how much
data is in the packet, and throw an exception on an attempt to fetch
data past the end of the packet, so that the dissector won't process
bogus data), causing the Info column not to be filled in.

This means that the Info column will have data for the previous
protocol, which would be confusing if, for example, the Protocol column
had data for this protocol.

Therefore, before a dissector fetches any data whatsoever from the
packet (unless it's a heuristic dissector fetching data to determine
whether the packet is one that it should dissect, in which case it
should check, before fetching the data, whether there's any data to
fetch; if there isn't, it should return FALSE), it should set the
Protocol column and the Info column.

If the Protocol column will ultimately be set to, for example, a value
containing a protocol version number, with the version number being a
field in the packet, the dissector should, before fetching the version
number field or any other field from the packet, set it to a value
without a version number, using 'col_set_str', and should later set it
to a value with the version number after it's fetched the version
number.

If the Info column will ultimately be set to a value containing
information from the packet, the dissector should, before fetching any
fields from the packet, clear the column using 'col_clear' (which is
more efficient than clearing it by calling 'col_set_str' or
'col_add_str' with a null string), and should later set it to the real
string after it's fetched the data to use when doing that.


1.5.5 The col_append_str function.

Sometimes the value of a column, especially the "Info" column, can't be
conveniently constructed at a single point in the dissection process;
for example, it might contain small bits of information from many of the
fields in the packet.  'col_append_str' takes, as arguments, the same
arguments as 'col_add_str', but the string is appended to the end of the
current value for the column, rather than replacing the value for that
column.  (Note that no blank separates the appended string from the
string to which it is appended; if you want a blank there, you must add
it yourself as part of the string being appended.)


1.5.6 The col_append_fstr function.

'col_append_fstr' is to 'col_add_fstr' as 'col_append_str' is to
'col_add_str' - it takes, as arguments, the same arguments as
'col_add_fstr', but the formatted string is appended to the end of the
current value for the column, rather than replacing the value for that
column.

1.5.7 The col_append_sep_str and col_append_sep_fstr functions.

In specific situations the developer knows that a column's value will be
created in a stepwise manner, where the appended values are listed. Both
'col_append_sep_str' and 'col_append_sep_fstr' functions will add an item
separator between two consecutive items, and will not add the separator at the
beginning of the column. The remainder of the work both functions do is
identical to what 'col_append_str' and 'col_append_fstr' do.

1.6 Constructing the protocol tree.

The middle pane of the main window, and the topmost pane of a packet
popup window, are constructed from the "protocol tree" for a packet.

The protocol tree, or proto_tree, is a GNode, the N-way tree structure
available within GLIB. Of course the protocol dissectors don't care
what a proto_tree really is; they just pass the proto_tree pointer as an
argument to the routines which allow them to add items and new branches
to the tree.

When a packet is selected in the packet-list pane, or a packet popup
window is created, a new logical protocol tree (proto_tree) is created. 
The pointer to the proto_tree (in this case, 'protocol tree'), is passed
to the top-level protocol dissector, and then to all subsequent protocol
dissectors for that packet, and then the GUI tree is drawn via
proto_tree_draw().

The logical proto_tree needs to know detailed information about the
protocols and fields about which information will be collected from the
dissection routines. By strictly defining (or "typing") the data that can
be attached to a proto tree, searching and filtering becomes possible.
This means that the for every protocol and field (which I also call
"header fields", since they are fields in the protocol headers) which
might be attached to a tree, some information is needed.

Every dissector routine will need to register its protocols and fields
with the central protocol routines (in proto.c). At first I thought I
might keep all the protocol and field information about all the
dissectors in one file, but decentralization seemed like a better idea.
That one file would have gotten very large; one small change would have
required a re-compilation of the entire file. Also, by allowing
registration of protocols and fields at run-time, loadable modules of
protocol dissectors (perhaps even user-supplied) is feasible.

To do this, each protocol should have a register routine, which will be
called when Ethereal starts.  The code to call the register routines is
generated automatically; to arrange that a protocol's register routine
be called at startup:

	the file containing a dissector's "register" routine must be
	added to "DISSECTOR_SRC" in "epan/dissectors/Makefile.common";
 
	the "register" routine must have a name of the form
	"proto_register_XXX";
  
	the "register" routine must take no argument, and return no
	value;
 
	the "register" routine's name must appear in the source file
	either at the beginning of the line, or preceded only by "void "
	at the beginning of the line (that'd typically be the
	definition) - other white space shouldn't cause a problem, e.g.:
 
void proto_register_XXX(void) {
 
	...
 
}
 
and
 
void
proto_register_XXX( void )
{
 
	...
 
}
 
	and so on should work.

For every protocol or field that a dissector wants to register, a variable of
type int needs to be used to keep track of the protocol. The IDs are
needed for establishing parent/child relationships between protocols and
fields, as well as associating data with a particular field so that it
can be stored in the logical tree and displayed in the GUI protocol
tree.

Some dissectors will need to create branches within their tree to help
organize header fields. These branches should be registered as header
fields. Only true protocols should be registered as protocols. This is
so that a display filter user interface knows how to distinguish
protocols from fields.

A protocol is registered with the name of the protocol and its
abbreviation.

Here is how the frame "protocol" is registered.

	int proto_frame;

        proto_frame = proto_register_protocol (
                /* name */            "Frame",
                /* short name */      "Frame",
                /* abbrev */          "frame" );

A header field is also registered with its name and abbreviation, but
information about the its data type is needed. It helps to look at
the header_field_info struct to see what information is expected:

struct header_field_info {
	char				*name;
	char				*abbrev;
	enum ftenum			type;
	int				display;
	void				*strings;
	guint				bitmask;
	char				*blurb;

	int				id;	  /* calculated */
	int				parent;
	int				bitshift; /* calculated */
};

name
----
A string representing the name of the field. This is the name
that will appear in the graphical protocol tree.

abbrev
------
A string with an abbreviation of the field. We concatenate the
abbreviation of the parent protocol with an abbreviation for the field,
using a period as a separator. For example, the "src" field in an IP packet
would have "ip.src" as an abbreviation. It is acceptable to have
multiple levels of periods if, for example, you have fields in your
protocol that are then subdivided into subfields. For example, TRMAC
has multiple error fields, so the abbreviations follow this pattern:
"trmac.errors.iso", "trmac.errors.noniso", etc.

The abbreviation is the identifier used in a display filter.

type
----
The type of value this field holds. The current field types are:

	FT_NONE			No field type. Used for fields that
				aren't given a value, and that can only
				be tested for presence or absence; a
				field that represents a data structure,
				with a subtree below it containing
				fields for the members of the structure,
				or that represents an array with a
				subtree below it containing fields for
				the members of the array, might be an
				FT_NONE field.
	FT_BOOLEAN		0 means "false", any other value means
				"true".
	FT_FRAMENUM		A frame number; if this is used, the "Go
				To Corresponding Frame" menu item can
				work on that field.
	FT_UINT8		An 8-bit unsigned integer.
	FT_UINT16		A 16-bit unsigned integer.
	FT_UINT24		A 24-bit unsigned integer.
	FT_UINT32		A 32-bit unsigned integer.
	FT_UINT64		A 64-bit unsigned integer.
	FT_INT8			An 8-bit signed integer.
	FT_INT16		A 16-bit signed integer.
	FT_INT24		A 24-bit signed integer.
	FT_INT32		A 32-bit signed integer.
	FT_INT64		A 64-bit signed integer.
	FT_FLOAT		A single-precision floating point number.
	FT_DOUBLE		A double-precision floating point number.
	FT_ABSOLUTE_TIME	Seconds (4 bytes) and nanoseconds (4 bytes)
				of time displayed as month name, month day,
				year, hours, minutes, and seconds with 9
				digits after the decimal point.
	FT_RELATIVE_TIME	Seconds (4 bytes) and nanoseconds (4 bytes)
				of time displayed as seconds and 9 digits
				after the decimal point.
	FT_STRING		A string of characters, not necessarily
				NUL-terminated, but possibly NUL-padded.
				This, and the other string-of-characters
				types, are to be used for text strings,
				not raw binary data.
	FT_STRINGZ		A NUL-terminated string of characters.
	FT_UINT_STRING		A counted string of characters, consisting
				of a count (represented as an integral
				value) followed immediately by the
				specified number of characters.
	FT_ETHER		A six octet string displayed in
				Ethernet-address format.
	FT_BYTES		A string of bytes with arbitrary values;
				used for raw binary data.
	FT_IPv4			A version 4 IP address (4 bytes) displayed
				in dotted-quad IP address format (4
				decimal numbers separated by dots).
	FT_IPv6			A version 6 IP address (16 bytes) displayed
				in standard IPv6 address format.
	FT_IPXNET		An IPX address displayed in hex as a 6-byte
				network number followed by a 6-byte station
				address. 

Some of these field types are still not handled in the display filter
routines, but the most common ones are. The FT_UINT* variables all
represent unsigned integers, and the FT_INT* variables all represent
signed integers; the number on the end represent how many bits are used
to represent the number.

display
-------
The display field has a couple of overloaded uses. This is unfortunate,
but since we're C as an application programming language, this sometimes
makes for cleaner programs. Right now I still think that overloading
this variable was okay.

For integer fields (FT_UINT* and FT_INT*), this variable represents the
base in which you would like the value displayed.  The acceptable bases
are:

	BASE_DEC,
	BASE_HEX,
	BASE_OCT

BASE_DEC, BASE_HEX, and BASE_OCT are decimal, hexadecimal, and octal,
respectively.

For FT_BOOLEAN fields that are also bitfields, 'display' is used to tell
the proto_tree how wide the parent bitfield is.  With integers this is
not needed since the type of integer itself (FT_UINT8, FT_UINT16,
FT_UINT24, FT_UINT32, etc.) tells the proto_tree how wide the parent
bitfield is.

Additionally, BASE_NONE is used for 'display' as a NULL-value. That is,
for non-integers and non-bitfield FT_BOOLEANs, you'll want to use BASE_NONE
in the 'display' field.  You may not use BASE_NONE for integers.

It is possible that in the future we will record the endianness of
integers. If so, it is likely that we'll use a bitmask on the display field
so that integers would be represented as BEND|BASE_DEC or LEND|BASE_HEX.
But that has not happened yet.

strings
-------
Some integer fields, of type FT_UINT*, need labels to represent the true
value of a field.  You could think of those fields as having an
enumerated data type, rather than an integral data type.

A 'value_string' structure is a way to map values to strings. 

	typedef struct _value_string {
		guint32  value;
		gchar   *strptr;
	} value_string;

For fields of that type, you would declare an array of "value_string"s:

	static const value_string valstringname[] = {
		{ INTVAL1, "Descriptive String 1" }, 
		{ INTVAL2, "Descriptive String 2" }, 
		{ 0,       NULL },
	};

(the last entry in the array must have a NULL 'strptr' value, to
indicate the end of the array).  The 'strings' field would be set to
'VALS(valstringname)'.

If the field has a numeric rather than an enumerated type, the 'strings'
field would be set to NULL.

FT_BOOLEANS have a default map of 0 = "False", 1 (or anything else) = "True".
Sometimes it is useful to change the labels for boolean values (e.g.,
to "Yes"/"No", "Fast"/"Slow", etc.).  For these mappings, a struct called
true_false_string is used. (This struct is new as of Ethereal 0.7.6).

	typedef struct true_false_string {
		char	*true_string;
		char	*false_string;
	} true_false_string;

For Boolean fields for which "False" and "True" aren't the desired
labels, you would declare a "true_false_string"s:

	static const true_false_string boolstringname = {
		"String for True",
		"String for False"
	};

Its two fields are pointers to the string representing truth, and the
string representing falsehood.  For FT_BOOLEAN fields that need a
'true_false_string' struct, the 'strings' field would be set to
'TFS(&boolstringname)'. 

If the Boolean field is to be displayed as "False" or "True", the
'strings' field would be set to NULL.

bitmask
-------
If the field is a bitfield, then the bitmask is the mask which will
leave only the bits needed to make the field when ANDed with a value.
The proto_tree routines will calculate 'bitshift' automatically
from 'bitmask', by finding the rightmost set bit in the bitmask.
If the field is not a bitfield, then bitmask should be set to 0.

blurb
-----
This is a string giving a proper description of the field.
It should be at least one grammatically complete sentence.
It is meant to provide a more detailed description of the field than the
name alone provides. This information will be used in the man page, and
in a future GUI display-filter creation tool. We might also add tooltips
to the labels in the GUI protocol tree, in which case the blurb would
be used as the tooltip text.


1.6.1 Field Registration.

Protocol registration is handled by creating an instance of the
header_field_info struct (or an array of such structs), and
calling the registration function along with the registration ID of
the protocol that is the parent of the fields. Here is a complete example:

	static int proto_eg = -1;
	static int hf_field_a = -1;
	static int hf_field_b = -1;

	static hf_register_info hf[] = {

		{ &hf_field_a,
		{ "Field A",	"proto.field_a", FT_UINT8, BASE_HEX, NULL,
			0xf0, "Field A represents Apples", HFILL }},

		{ &hf_field_b,
		{ "Field B",	"proto.field_b", FT_UINT16, BASE_DEC, VALS(vs),
			0x0, "Field B represents Bananas", HFILL }}
	};

	proto_eg = proto_register_protocol("Example Protocol",
	    "PROTO", "proto");
	proto_register_field_array(proto_eg, hf, array_length(hf));

Be sure that your array of hf_register_info structs is declared 'static',
since the proto_register_field_array() function does not create a copy
of the information in the array... it uses that static copy of the
information that the compiler created inside your array. Here's the
layout of the hf_register_info struct:

typedef struct hf_register_info {
	int			*p_id;	/* pointer to parent variable */
	header_field_info	hfinfo;
} hf_register_info;

Also be sure to use the handy array_length() macro found in packet.h
to have the compiler compute the array length for you at compile time.

If you don't have any fields to register, do *NOT* create a zero-length
"hf" array; not all compilers used to compile Ethereal support them. 
Just omit the "hf" array, and the "proto_register_field_array()" call,
entirely.

It is OK to have header fields with a different format be registered with
the same abbreviation. For instance, the following is valid:

	static hf_register_info hf[] = {

		{ &hf_field_8bit, /* 8-bit version of proto.field */
		{ "Field (8 bit)", "proto.field", FT_UINT8, BASE_DEC, NULL,
			0x00, "Field represents FOO", HFILL }},

		{ &hf_field_32bit, /* 32-bit version of proto.field */
		{ "Field (32 bit)", "proto.field", FT_UINT32, BASE_DEC, NULL,
			0x00, "Field represents FOO", HFILL }}
	};

This way a filter expression can match a header field, irrespective of the
representation of it in the specific protocol context. This is interesting
for protocols with variable-width header fields.

The HFILL macro at the end of the struct will set resonable default values
for internally used fields.

1.6.2 Adding Items and Values to the Protocol Tree.

A protocol item is added to an existing protocol tree with one of a
handful of proto_XXX_DO_YYY() functions.

Remember that it only makes sense to add items to a protocol tree if its
proto_tree pointer is not null. Should you add an item to a NULL tree, then
the proto_XXX_DO_YYY() function will immediately return. The cost of this
function call can be avoided by checking for the tree pointer.

Subtrees can be made with the proto_item_add_subtree() function:

	item = proto_tree_add_item(....);
	new_tree = proto_item_add_subtree(item, tree_type);

This will add a subtree under the item in question; a subtree can be
created under an item made by any of the "proto_tree_add_XXX" functions,
so that the tree can be given an arbitrary depth.

Subtree types are integers, assigned by
"proto_register_subtree_array()".  To register subtree types, pass an
array of pointers to "gint" variables to hold the subtree type values to
"proto_register_subtree_array()":

	static gint ett_eg = -1;
	static gint ett_field_a = -1;

	static gint *ett[] = {
		&ett_eg,
		&ett_field_a,
	};

	proto_register_subtree_array(ett, array_length(ett));

in your "register" routine, just as you register the protocol and the
fields for that protocol.

There are several functions that the programmer can use to add either
protocol or field labels to the proto_tree:

	proto_item*
	proto_tree_add_item(tree, id, tvb, start, length, little_endian);

	proto_item*
	proto_tree_add_item_hidden(tree, id, tvb, start, length, little_endian);

	proto_item*
	proto_tree_add_none_format(tree, id, tvb, start, length, format, ...);

	proto_item*
	proto_tree_add_protocol_format(tree, id, tvb, start, length,
	    format, ...);

	proto_item *
	proto_tree_add_bytes(tree, id, tvb, start, length, start_ptr);

	proto_item *
	proto_tree_add_bytes_hidden(tree, id, tvb, start, length, start_ptr);

	proto_item *
	proto_tree_add_bytes_format(tree, id, tvb, start, length, start_ptr,
	    format, ...);

	proto_item *
	proto_tree_add_time(tree, id, tvb, start, length, value_ptr);

	proto_item *
	proto_tree_add_time_hidden(tree, id, tvb, start, length, value_ptr);

	proto_item *
	proto_tree_add_time_format(tree, id, tvb, start, length, value_ptr,
	    format, ...);

	proto_item *
	proto_tree_add_ipxnet(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_ipxnet_hidden(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_ipxnet_format(tree, id, tvb, start, length, value,
	    format, ...);

	proto_item *
	proto_tree_add_ipv4(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_ipv4_hidden(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_ipv4_format(tree, id, tvb, start, length, value,
	    format, ...);

	proto_item *
	proto_tree_add_ipv6(tree, id, tvb, start, length, value_ptr);

	proto_item *
	proto_tree_add_ipv6_hidden(tree, id, tvb, start, length, value_ptr);

	proto_item *
	proto_tree_add_ipv6_format(tree, id, tvb, start, length, value_ptr,
	    format, ...);

	proto_item *
	proto_tree_add_ether(tree, id, tvb, start, length, value_ptr);

	proto_item *
	proto_tree_add_ether_hidden(tree, id, tvb, start, length, value_ptr);

	proto_item *
	proto_tree_add_ether_format(tree, id, tvb, start, length, value_ptr,
	    format, ...);

	proto_item *
	proto_tree_add_string(tree, id, tvb, start, length, value_ptr);

	proto_item *
	proto_tree_add_string_hidden(tree, id, tvb, start, length, value_ptr);

	proto_item *
	proto_tree_add_string_format(tree, id, tvb, start, length, value_ptr,
	    format, ...);

	proto_item *
	proto_tree_add_boolean(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_boolean_hidden(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_boolean_format(tree, id, tvb, start, length, value,
	    format, ...);

	proto_item *
	proto_tree_add_float(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_float_hidden(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_float_format(tree, id, tvb, start, length, value,
	    format, ...);

	proto_item *
	proto_tree_add_double(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_double_hidden(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_double_format(tree, id, tvb, start, length, value,
	    format, ...);

	proto_item *
	proto_tree_add_uint(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_uint_hidden(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_uint_format(tree, id, tvb, start, length, value,
	    format, ...);

	proto_item *
	proto_tree_add_uint64(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_uint64_format(tree, id, tvb, start, length, value,
	    format, ...);

	proto_item *
	proto_tree_add_int(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_int_hidden(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_int_format(tree, id, tvb, start, length, value,
	    format, ...);

	proto_item *
	proto_tree_add_int64(tree, id, tvb, start, length, value);

	proto_item *
	proto_tree_add_int64_format(tree, id, tvb, start, length, value,
	    format, ...);

	proto_item*
	proto_tree_add_text(tree, tvb, start, length, format, ...);

	proto_item*
	proto_tree_add_text_valist(tree, tvb, start, length, format, ap);

The 'tree' argument is the tree to which the item is to be added.  The
'tvb' argument is the tvbuff from which the item's value is being
extracted; the 'start' argument is the offset from the beginning of that
tvbuff of the item being added, and the 'length' argument is the length,
in bytes, of the item.

The length of some items cannot be determined until the item has been
dissected; to add such an item, add it with a length of -1, and, when the
dissection is complete, set the length with 'proto_item_set_len()':

	void
	proto_item_set_len(ti, length);

The "ti" argument is the value returned by the call that added the item
to the tree, and the "length" argument is the length of the item.

proto_tree_add_item()
---------------------
proto_tree_add_item is used when you wish to do no special formatting. 
The item added to the GUI tree will contain the name (as passed in the
proto_register_*() function) and a value.  The value will be fetched
from the tvbuff by proto_tree_add_item(), based on the type of the field
and, for integral and Boolean fields, the byte order of the value; the
byte order is specified by the 'little_endian' argument, which is TRUE
if the value is little-endian and FALSE if it is big-endian.

Now that definitions of fields have detailed information about bitfield
fields, you can use proto_tree_add_item() with no extra processing to
add bitfield values to your tree.  Here's an example.  Take the Format
Identifer (FID) field in the Transmission Header (TH) portion of the SNA
protocol.  The FID is the high nibble of the first byte of the TH.  The
FID would be registered like this:

	name		= "Format Identifer"
	abbrev		= "sna.th.fid"
	type		= FT_UINT8
	display		= BASE_HEX
	strings		= sna_th_fid_vals
	bitmask		= 0xf0

The bitmask contains the value which would leave only the FID if bitwise-ANDed
against the parent field, the first byte of the TH.

The code to add the FID to the tree would be;

	proto_tree_add_item(bf_tree, hf_sna_th_fid, tvb, offset, 1, TRUE);

The definition of the field already has the information about bitmasking
and bitshifting, so it does the work of masking and shifting for us!
This also means that you no longer have to create value_string structs
with the values bitshifted.  The value_string for FID looks like this,
even though the FID value is actually contained in the high nibble. 
(You'd expect the values to be 0x0, 0x10, 0x20, etc.)

/* Format Identifier */
static const value_string sna_th_fid_vals[] = {
	{ 0x0,	"SNA device <--> Non-SNA Device" },
	{ 0x1,	"Subarea Node <--> Subarea Node" },
	{ 0x2,	"Subarea Node <--> PU2" },
	{ 0x3,	"Subarea Node or SNA host <--> Subarea Node" },
	{ 0x4,	"?" },
	{ 0x5,	"?" },
	{ 0xf,	"Adjaced Subarea Nodes" },
	{ 0,	NULL }
};

The final implication of this is that display filters work the way you'd
naturally expect them to. You'd type "sna.th.fid == 0xf" to find Adjacent
Subarea Nodes. The user does not have to shift the value of the FID to
the high nibble of the byte ("sna.th.fid == 0xf0") as was necessary
before Ethereal 0.7.6.

proto_tree_add_item_hidden()
----------------------------
proto_tree_add_item_hidden is used to add fields and values to a tree,
but not show them on a GUI tree.  The caller may want a value to be
included in a tree so that the packet can be filtered on this field, but
the representation of that field in the tree is not appropriate.  An
example is the token-ring routing information field (RIF).  The best way
to show the RIF in a GUI is by a sequence of ring and bridge numbers. 
Rings are 3-digit hex numbers, and bridges are single hex digits:

	RIF: 001-A-013-9-C0F-B-555

In the case of RIF, the programmer should use a field with no value and
use proto_tree_add_none_format() to build the above representation. The
programmer can then add the ring and bridge values, one-by-one, with
proto_tree_add_item_hidden() so that the user can then filter on or
search for a particular ring or bridge. Here's a skeleton of how the
programmer might code this.

	char *rif;
	rif = create_rif_string(...);

	proto_tree_add_none_format(tree, hf_tr_rif_label, ..., "RIF: %s", rif);

	for(i = 0; i < num_rings; i++) {
		proto_tree_add_item_hidden(tree, hf_tr_rif_ring, ..., FALSE);
	}
	for(i = 0; i < num_rings - 1; i++) {
		proto_tree_add_item_hidden(tree, hf_tr_rif_bridge, ..., FALSE);
	}

The logical tree has these items:

	hf_tr_rif_label, text="RIF: 001-A-013-9-C0F-B-555", value = NONE
	hf_tr_rif_ring,  hidden, value=0x001
	hf_tr_rif_bridge, hidden, value=0xA
	hf_tr_rif_ring,  hidden, value=0x013
	hf_tr_rif_bridge, hidden, value=0x9
	hf_tr_rif_ring,  hidden, value=0xC0F
	hf_tr_rif_bridge, hidden, value=0xB
	hf_tr_rif_ring,  hidden, value=0x555

GUI or print code will not display the hidden fields, but a display
filter or "packet grep" routine will still see the values. The possible
filter is then possible:

	tr.rif_ring eq 0x013

proto_tree_add_protocol_format()
----------------------------
proto_tree_add_protocol_format is used to add the top-level item for the
protocol when the dissector routines wants complete control over how the
field and value will be represented on the GUI tree.  The ID value for
the protocol is passed in as the "id" argument; the rest of the
arguments are a "printf"-style format and any arguments for that format. 
The caller must include the name of the protocol in the format; it is
not added automatically as in proto_tree_add_item().

proto_tree_add_none_format()
----------------------------
proto_tree_add_none_format is used to add an item of type FT_NONE.
The caller must include the name of the field in the format; it is
not added automatically as in proto_tree_add_item().

proto_tree_add_bytes()
proto_tree_add_time()
proto_tree_add_ipxnet()
proto_tree_add_ipv4()
proto_tree_add_ipv6()
proto_tree_add_ether()
proto_tree_add_string()
proto_tree_add_boolean()
proto_tree_add_float()
proto_tree_add_double()
proto_tree_add_uint()
proto_tree_add_uint64()
proto_tree_add_int()
proto_tree_add_int64()
----------------------------
These routines are used to add items to the protocol tree if either:

	the value of the item to be added isn't just extracted from the
	packet data, but is computed from data in the packet;

	the value was fetched into a variable.

The 'value' argument has the value to be added to the tree.

NOTE: in all cases where the 'value' argument is a pointer, a copy is
made of the object pointed to; if you have dynamically allocated a
buffer for the object, that buffer will not be freed when the protocol
tree is freed - you must free the buffer yourself when you don't need it
any more.

For proto_tree_add_bytes(), the 'value_ptr' argument is a pointer to a
sequence of bytes.

For proto_tree_add_time(), the 'value_ptr' argument is a pointer to an
"nstime_t", which is a structure containing the time to be added; it has
'secs' and 'nsecs' members, giving the integral part and the fractional
part of a time in units of seconds, with 'nsecs' being the number of
nanoseconds.  For absolute times, "secs" is a UNIX-style seconds since
January 1, 1970, 00:00:00 GMT value.

For proto_tree_add_ipxnet(), the 'value' argument is a 32-bit IPX
network address.

For proto_tree_add_ipv4(), the 'value' argument is a 32-bit IPv4
address, in network byte order.

For proto_tree_add_ipv6(), the 'value_ptr' argument is a pointer to a
128-bit IPv6 address.

For proto_tree_add_ether(), the 'value_ptr' argument is a pointer to a
48-bit MAC address.

For proto_tree_add_string(), the 'value_ptr' argument is a pointer to a
text string.

For proto_tree_add_boolean(), the 'value' argument is a 32-bit integer;
zero means "false", and non-zero means "true".

For proto_tree_add_float(), the 'value' argument is a 'float' in the
host's floating-point format.

For proto_tree_add_double(), the 'value' argument is a 'double' in the
host's floating-point format.

For proto_tree_add_uint(), the 'value' argument is a 32-bit unsigned
integer value, in host byte order.  (This routine cannot be used to add
64-bit integers.)

For proto_tree_add_uint64(), the 'value' argument is a 64-bit unsigned
integer value, in host byte order.

For proto_tree_add_int(), the 'value' argument is a 32-bit signed
integer value, in host byte order.  (This routine cannot be used to add
64-bit integers.)

For proto_tree_add_int64(), the 'value' argument is a 64-bit signed
integer value, in host byte order.

proto_tree_add_bytes_hidden()
proto_tree_add_time_hidden()
proto_tree_add_ipxnet_hidden()
proto_tree_add_ipv4_hidden()
proto_tree_add_ipv6_hidden()
proto_tree_add_ether_hidden()
proto_tree_add_string_hidden()
proto_tree_add_boolean_hidden()
proto_tree_add_float_hidden()
proto_tree_add_double_hidden()
proto_tree_add_uint_hidden()
proto_tree_add_int_hidden()
----------------------------
These routines add fields and values to a tree, but don't show them in
the GUI tree.  They are used for the same reason that
proto_tree_add_item() is used.

proto_tree_add_bytes_format()
proto_tree_add_time_format()
proto_tree_add_ipxnet_format()
proto_tree_add_ipv4_format()
proto_tree_add_ipv6_format()
proto_tree_add_ether_format()
proto_tree_add_string_format()
proto_tree_add_boolean_format()
proto_tree_add_float_format()
proto_tree_add_double_format()
proto_tree_add_uint_format()
proto_tree_add_uint64_format()
proto_tree_add_int_format()
proto_tree_add_int64_format()
----------------------------
These routines are used to add items to the protocol tree when the
dissector routines wants complete control over how the field and value
will be represented on the GUI tree.  The argument giving the value is
the same as the corresponding proto_tree_add_XXX() function; the rest of
the arguments are a "printf"-style format and any arguments for that
format.  The caller must include the name of the field in the format; it
is not added automatically as in the proto_tree_add_XXX() functions.

proto_tree_add_text()
---------------------
proto_tree_add_text() is used to add a label to the GUI tree.  It will
contain no value, so it is not searchable in the display filter process. 
This function was needed in the transition from the old-style proto_tree
to this new-style proto_tree so that Ethereal would still decode all
protocols w/o being able to filter on all protocols and fields. 
Otherwise we would have had to cripple Ethereal's functionality while we
converted all the old-style proto_tree calls to the new-style proto_tree
calls.

This can also be used for items with subtrees, which may not have values
themselves - the items in the subtree are the ones with values.

For a subtree, the label on the subtree might reflect some of the items
in the subtree.  This means the label can't be set until at least some
of the items in the subtree have been dissected.  To do this, use
'proto_item_set_text()' or 'proto_item_append_text()':

	void
	proto_item_set_text(proto_item *ti, ...);

	void
	proto_item_append_text(proto_item *ti, ...);

'proto_item_set_text()' takes as an argument the value returned by
'proto_tree_add_text()', a 'printf'-style format string, and a set of
arguments corresponding to '%' format items in that string, and replaces
the text for the item created by 'proto_tree_add_text()' with the result
of applying the arguments to the format string. 

'proto_item_append_text()' is similar, but it appends to the text for
the item the result of applying the arguments to the format string.

For example, early in the dissection, one might do:

	ti = proto_tree_add_text(tree, tvb, offset, length, <label>);

and later do

	proto_item_set_text(ti, "%s: %s", type, value);

after the "type" and "value" fields have been extracted and dissected. 
<label> would be a label giving what information about the subtree is
available without dissecting any of the data in the subtree.

Note that an exception might thrown when trying to extract the values of
the items used to set the label, if not all the bytes of the item are
available.  Thus, one should create the item with text that is as
meaningful as possible, and set it or append additional information to
it as the values needed to supply that information is extracted.

proto_tree_add_text_valist()
---------------------
This is like proto_tree_add_text(), but takes, as the last argument, a
'va_list'; it is used to allow routines that take a printf-like
variable-length list of arguments to add a text item to the protocol
tree.

1.7 Utility routines

1.7.1 match_strval and val_to_str

A dissector may need to convert a value to a string, using a
'value_string' structure, by hand, rather than by declaring a field with
an associated 'value_string' structure; this might be used, for example,
to generate a COL_INFO line for a frame.

'match_strval()' will do that:

	gchar*
	match_strval(guint32 val, const value_string *vs)

It will look up the value 'val' in the 'value_string' table pointed to
by 'vs', and return either the corresponding string, or NULL if the
value could not be found in the table.  Note that, unless 'val' is
guaranteed to be a value in the 'value_string' table ("guaranteed" as in
"the code has already checked that it's one of those values" or "the
table handles all possible values of the size of 'val'", not "the
protocol spec says it has to be" - protocol specs do not prevent invalid
packets from being put onto a network or into a purported packet capture
file), you must check whether 'match_strval()' returns NULL, and arrange
that its return value not be dereferenced if it's NULL.  In particular,
don't use it in a call to generate a COL_INFO line for a frame such as

	col_add_fstr(COL_INFO, ", %s", match_strval(val, table));

unless is it certain that 'val' is in 'table'.

'val_to_str()' can be used to generate a string for values not found in
the table:

	gchar*
	val_to_str(guint32 val, const value_string *vs, const char *fmt)

If the value 'val' is found in the 'value_string' table pointed to by
'vs', 'val_to_str' will return the corresponding string; otherwise, it
will use 'fmt' as an 'sprintf'-style format, with 'val' as an argument,
to generate a string, and will return a pointer to that string. 
(Currently, it has three 64-byte static buffers, and cycles through
them; this permits the results of up to three calls to 'val_to_str' to
be passed as arguments to a routine using those strings.)


1.8 Calling Other Dissector 

NOTE: This is discussed in the README.tvbuff file.  For more 
information on tvbuffers consult that file.

As each dissector completes its portion of the protocol analysis, it
is expected to create a new tvbuff of type TVBUFF_SUBSET which
contains the payload portion of the protocol (that is, the bytes
that are relevant to the next dissector).

The syntax for creating a new TVBUFF_SUBSET is:

next_tvb = tvb_new_subset(tvb, offset, length, reported_length)

Where:
	tvb is the tvbuff that the dissector has been working on. It
	can be a tvbuff of any type.

	next_tvb is the new TVBUFF_SUBSET.

	offset is the byte offset of 'tvb' at which the new tvbuff
	should start.  The first byte is the 0th byte.

	length is the number of bytes in the new TVBUFF_SUBSET. A length
	argument of -1 says to use as many bytes as are available in
	'tvb'.

	reported_length is the number of bytes that the current protocol
	says should be in the payload. A reported_length of -1 says that
	the protocol doesn't say anything about the size of its payload.


An example from packet-ipx.c -

void
dissect_ipx(tvbuff_t *tvb, packet_info *pinfo, proto_tree *tree)
{
        tvbuff_t        *next_tvb;
        int             reported_length, available_length;

 
        /* Make the next tvbuff */

/* IPX does have a length value in the header, so calculate report_length */
   Set this to -1 if there isn't any length information in the protocol
*/
        reported_length = ipx_length - IPX_HEADER_LEN;

/* Calculate the available data in the packet, 
   set this to -1 to use all the data in the tv_buffer
*/
        available_length = tvb_length(tvb) - IPX_HEADER_LEN;

/* Create the tvbuffer for the next dissector */
        next_tvb = tvb_new_subset(tvb, IPX_HEADER_LEN,
                        MIN(available_length, reported_length),
                        reported_length);

/* call the next dissector */
	dissector_next( next_tvb, pinfo, tree);


1.9 Editing Makefile.common to add your dissector.

To arrange that your dissector will be built as part of Ethereal, you
must add the name of the source file for your dissector to the
'DISSECTOR_SRC' macro in the 'Makefile.common' file in the 'epan/dissectors'
directory.  (Note that this is for modern versions of UNIX, so there
is no 14-character limitation on file names, and for modern versions of
Windows, so there is no 8.3-character limitation on file names.)

If your dissector also has its own header file or files, you must add
them to the 'DISSECTOR_INCLUDES' macro in the 'Makefile.common' file in
the 'epan/dissectors' directory, so that it's included when release source
tarballs are built (otherwise, the source in the release tarballs won't
compile).

1.10 Using the CVS source code tree.

1.11 Submitting code for your new dissector.

2. Advanced dissector topics.

2.1 ?? 

2.2 Following "conversations".

In ethereal a conversation is defined as a series of data packet between two
address:port combinations.  A conversation is not sensitive to the direction of
the packet.  The same conversation will be returned for a packet bound from
ServerA:1000 to ClientA:2000 and the packet from ClientA:2000 to ServerA:1000.

There are five routines that you will use to work with a conversation:
conversation_new, find_conversation, conversation_add_proto_data,
conversation_get_proto_data, and conversation_delete_proto_data.


2.2.1 The conversation_init function.

This is an internal routine for the conversation code.  As such the you
will not have to call this routine.  Just be aware that this routine is
called at the start of each capture and before the packets are filtered
with a display filter.  The routine will destroy all stored
conversations.  This routine does NOT clean up any data pointers that are
passed in the conversation_new 'data' variable.  You are responsible for
this clean up if you pass a malloc'ed pointer in this variable.

See item 2.2.7 for more information about the 'data' pointer.


2.2.2 The conversation_new function.

This routine will create a new conversation based upon two address/port
pairs.  If you want to associate with the conversation a pointer to a
private data structure you must use the conversation_add_proto_data
function.  The ptype variable is used to differentiate between
conversations over different protocols, i.e. TCP and UDP.  The options
variable is used to define a conversation that will accept any destination
address and/or port.  Set options = 0 if the destination port and address
are know when conversation_new is called.  See section 2.4 for more
information on usage of the options parameter.

The conversation_new prototype:
	conversation_t *conversation_new(address *addr1, address *addr2,
	    port_type ptype, guint32 port1, guint32 port2, guint options);

Where:
	address* addr1 	= first data packet address
	address* addr2 	= second data packet address
	port_type ptype = port type, this is defined in packet.h
	guint32 port1	= first data packet port
	guint32 port2	= second data packet port
	guint options	= conversation options, NO_ADDR2 and/or NO_PORT2

"addr1" and "port1" are the first address/port pair; "addr2" and "port2"
are the second address/port pair.  A conversation doesn't have source
and destination address/port pairs - packets in a conversation go in
both directions - so "addr1"/"port1" may be the source or destination
address/port pair; "addr2"/"port2" would be the other pair.

If NO_ADDR2 is specified, the conversation is set up so that a
conversation lookup will match only the "addr1" address; if NO_PORT2 is
specified, the conversation is set up so that a conversation lookup will
match only the "port1" port; if both are specified, i.e.
NO_ADDR2|NO_PORT2, the conversation is set up so that the lookup will
match only the "addr1"/"port1" address/port pair.  This can be used if a
packet indicates that, later in the capture, a conversation will be
created using certain addresses and ports, in the case where the packet
doesn't specify the addresses and ports of both sides.

2.2.3 The find_conversation function.

Call this routine to look up a conversation.  If no conversation is found,
the routine will return a NULL value.

The find_conversation prototype:

	conversation_t *find_conversation(address *addr_a, address *addr_b,
	    port_type ptype, guint32 port_a, guint32 port_b, guint options);

Where:
	address* addr_a = first address
	address* addr_b = second address
	port_type ptype = port type
	guint32 port_a	= first data packet port
	guint32 port_b	= second data packet port
	guint options	= conversation options, NO_ADDR_B and/or NO_PORT_B

"addr_a" and "port_a" are the first address/port pair; "addr_b" and
"port_b" are the second address/port pair.  Again, as a conversation
doesn't have source and destination address/port pairs, so
"addr_a"/"port_a" may be the source or destination address/port pair;
"addr_b"/"port_b" would be the other pair.  The search will match the
"a" address/port pair against both the "1" and "2" address/port pairs,
and match the "b" address/port pair against both the "2" and "1"
address/port pairs; you don't have to worry about which side the "a" or
"b" pairs correspond to.

If the NO_ADDR_B flag was specified to "find_conversation()", the
"addr_b" address will be treated as matching any "wildcarded" address;
if the NO_PORT_B flag was specified, the "port_b" port will be treated
as matching any "wildcarded" port.  If both flags are specified, i.e. 
NO_ADDR_B|NO_PORT_B, the "addr_b" address will be treated as matching
any "wildcarded" address and the "port_b" port will be treated as
matching any "wildcarded" port.


2.2.4 The conversation_add_proto_data function.

Once you have created a conversation with conversation_new, you can
associate data with it using this function.

The conversation_add_proto_data prototype:

	void conversation_add_proto_data(conversation_t *conv, int proto,
	    void *proto_data);

Where:
	conversation_t *conv = the conversation in question
	int proto	     = registered protocol number
	void *data	     = dissector data structure

"conversation" is the value returned by conversation_new.  "proto" is a
unique protocol number created with proto_register_protocol.  Protocols
are typically registered in the proto_register_XXXX section of your
dissector.  "data" is a pointer to the data you wish to associate with the
conversation.  Using the protocol number allows several dissectors to
associate data with a given conversation.


2.2.5 The conversation_get_proto_data function.

After you have located a conversation with find_conversation, you can use
this function to retrieve any data associated with it.

The conversation_get_proto_data prototype:

	void *conversation_get_proto_data(conversation_t *conv, int proto);

Where:
	conversation_t *conv = the conversation in question
	int proto	     = registered protocol number
	
"conversation" is the conversation created with conversation_new.  "proto"
is a unique protocol number acreated with proto_register_protocol,
typically in the proto_register_XXXX portion of a dissector.  The function
returns a pointer to the data requested, or NULL if no data was found.


2.2.6 The conversation_delete_proto_data function.

After you are finished with a conversation, you can remove your assocation
with this function.  Please note that ONLY the conversation entry is
removed.  If you have allocated any memory for your data, you must free it
as well.

The conversation_delete_proto_data prototype:

	void conversation_delete_proto_data(conversation_t *conv, int proto);
	
Where:
	conversation_t *conv = the conversation in question
	int proto	     = registered protocol number

"conversation" is the conversation created with conversation_new.  "proto"
is a unique protocol number acreated with proto_register_protocol,
typically in the proto_register_XXXX portion of a dissector.


2.2.7 The example conversation code with GMemChunk's

For a conversation between two IP addresses and ports you can use this as an
example.  This example uses the GMemChunk to allocate memory and stores the data
pointer in the conversation 'data' variable.

NOTE: Remember to register the init routine (my_dissector_init) in the
protocol_register routine.


/************************ Globals values ************************/

/* the number of entries in the memory chunk array */
#define my_init_count 10

/* define your structure here */
typedef struct {

}my_entry_t;

/* the GMemChunk base structure */
static GMemChunk *my_vals = NULL;

/* Registered protocol number
static int my_proto = -1;


/********************* in the dissector routine *********************/

/* the local variables in the dissector */

conversation_t *conversation;
my_entry_t *data_ptr


/* look up the conversation */

conversation = find_conversation( &pinfo->src, &pinfo->dst, pinfo->ptype,
	pinfo->srcport, pinfo->destport, 0);

/* if conversation found get the data pointer that you stored */
if ( conversation)
    data_ptr = (my_entry_t*)conversation_get_proto_data(conversation,
    	    my_proto);
else {

    /* new conversation create local data structure */

    data_ptr = g_mem_chunk_alloc(my_protocol_vals);

    /*** add your code here to setup the new data structure ***/

    /* create the conversation with your data pointer  */

    conversation_new( &pinfo->src, &pinfo->dst, pinfo->ptype,
	    pinfo->srcport, pinfo->destport, 0);
    conversation_add_proto_data(conversation, my_proto, (void *) data_ptr);
}

/* at this point the conversation data is ready */


/******************* in the dissector init routine *******************/

#define my_init_count 20

static void
my_dissector_init( void){

    /* destroy memory chunks if needed */

    if ( my_vals)
	g_mem_chunk_destroy(my_vals);

    /* now create memory chunks */

    my_vals = g_mem_chunk_new( "my_proto_vals",
	    sizeof( _entry_t),
	    my_init_count * sizeof( my_entry_t),
	    G_ALLOC_AND_FREE);
}

/***************** in the protocol register routine *****************/

/* register re-init routine */

register_init_routine( &my_dissector_init);

my_proto = proto_register_protocol("My Protocol", "My Protocol", "my_proto");


2.2.8 The example conversation code using conversation index field

Sometimes the conversation isn't enough to define a unique data storage
value for the network traffic.  For example if you are storing information
about requests carried in a conversation, the request may have an
identifier that is used to  define the request. In this case the
conversation and the identifier are required to find the data storage
pointer.  You can use the conversation data structure index value to
uniquely define the conversation.  

See packet-afs.c for an example of how to use the conversation index.  In
this dissector multiple requests are sent in the same conversation.  To store
information for each request the dissector has an internal hash table based
upon the conversation index and values inside the request packets. 


/* in the dissector routine */

/* to find a request value, first lookup conversation to get index */
/* then used the conversation index, and request data to find data */
/* in the local hash table */

	conversation = find_conversation(&pinfo->src, &pinfo->dst, pinfo->ptype,
	    pinfo->srcport, pinfo->destport, 0);
	if (conversation == NULL) {
		/* It's not part of any conversation - create a new one. */
		conversation = conversation_new(&pinfo->src, &pinfo->dst, pinfo->ptype,
		    pinfo->srcport, pinfo->destport, NULL, 0);
	}

	request_key.conversation = conversation->index;	
	request_key.service = pntohs(&rxh->serviceId);
	request_key.callnumber = pntohl(&rxh->callNumber);

	request_val = (struct afs_request_val *) g_hash_table_lookup(
		afs_request_hash, &request_key);

	/* only allocate a new hash element when it's a request */
	opcode = 0;
	if ( !request_val && !reply)
	{
		new_request_key = g_mem_chunk_alloc(afs_request_keys);
		*new_request_key = request_key;

		request_val = g_mem_chunk_alloc(afs_request_vals);
		request_val -> opcode = pntohl(&afsh->opcode);
		opcode = request_val->opcode;

		g_hash_table_insert(afs_request_hash, new_request_key,
			request_val);
	}



2.3 Dynamic conversation dissector registration


NOTE: This sections assumes that all information is available to
	create a complete conversation, source port/address and
	destination port/address.  If either the destination port or
	address is know, see section 2.4 Dynamic server port dissector
	registration.

For protocols that negotiate a secondary port connection, for example
packet-msproxy.c, a conversation can install a dissector to handle 
the secondary protocol dissection.  After the conversation is created
for the negotiated ports use the conversation_set_dissector to define
the dissection routine.

The second argument to conversation_set_dissector is a dissector handle,
which is created with a call to create_dissector_handle or
register_dissector.

create_dissector_handle takes as arguments a pointer to the dissector
function and a protocol ID as returned by proto_register_protocol;
register_dissector takes as arguments a string giving a name for the
dissector, a pointer to the dissector function, and a protocol ID.

The protocol ID is the ID for the protocol dissected by the function. 
The function will not be called if the protocol has been disabled by the
user; instead, the data for the protocol will be dissected as raw data.

An example -

/* the handle for the dynamic dissector *
static dissector_handle_t sub_dissector_handle;

/* prototype for the dynamic dissector */
static void sub_dissector( tvbuff_t *tvb, packet_info *pinfo,
                proto_tree *tree);

/* in the main protocol dissector, where the next dissector is setup */

/* if conversation has a data field, create it and load structure */

        new_conv_info = g_mem_chunk_alloc( new_conv_vals);
        new_conv_info->data1 = value1;

/* create the conversation for the dynamic port */
        conversation = conversation_new( &pinfo->src, &pinfo->dst, protocol,
                src_port, dst_port, new_conv_info, 0);

/* set the dissector for the new conversation */
        conversation_set_dissector(conversation, sub_dissector_handle);

		...

void
proto_register_PROTOABBREV(void)
{                 
	...

	sub_dissector_handle = create_dissector_handle(sub_dissector,
	    proto);

	...
}

2.4 Dynamic server port dissector registration

NOTE: While this example used both NO_ADDR2 and NO_PORT2 to create a
conversation with only one port and address set, this isn't a
requirement.  Either the second port or the second address can be set
when the conversation is created.

For protocols that define a server address and port for a secondary
protocol, a conversation can be used to link a protocol dissector to
the server port and address.  The key is to create the new 
conversation with the second address and port set to the "accept
any" values.  

There are two support routines that will allow the second port and/or
address to be set latter.  

conversation_set_port2( conversation_t *conv, guint32 port);
conversation_set_addr2( conversation_t *conv, address addr);

These routines will change the second address or port for the
conversation.  So, the server port conversation will be converted into a
more complete conversation definition.  Don't use these routines if you
want create a conversation between the server and client and retain the
server port definition, you must create a new conversation.


An example -

/* the handle for the dynamic dissector *
static dissector_handle_t sub_dissector_handle;

	...

/* in the main protocol dissector, where the next dissector is setup */

/* if conversation has a data field, create it and load structure */

        new_conv_info = g_mem_chunk_alloc( new_conv_vals);
        new_conv_info->data1 = value1;

/* create the conversation for the dynamic server address and port 	*/
/* NOTE: The second address and port values don't matter because the	*/
/* NO_ADDR2 and NO_PORT2 options are set. 				*/

        conversation = conversation_new( &server_src_addr, 0, protocol,
                server_src_port, 0, new_conv_info, NO_ADDR2 | NO_PORT2);

/* set the dissector for the new conversation */
        conversation_set_dissector(conversation, sub_dissector_handle);


2.5 Per packet information

Information can be stored for each data packet that is process by the dissector.
The information is added with the p_add_proto_data function and retreived with the 
p_get_proto_data function.  The data pointers passed into the p_add_proto_data are
not managed by the proto_data routines. If you use malloc or any other dynamic 
memory allocation scheme, you must release the data when it isn't required.

void
p_add_proto_data(frame_data *fd, int proto, void *proto_data)
void *
p_get_proto_data(frame_data *fd, int proto)

Where: 
	fd         - The fd pointer in the pinfo structure, pinfo->fd
	proto      - Protocol id returned by the proto_register_protocol call during initialization
	proto_data - pointer to the dissector data.


2.6 User Preferences

If the dissector has user options, there is support for adding these preferences
to a configuration dialog.

You must register the module with the preferences routine with -

module_t *prefs_register_protocol(proto_id, void (*apply_cb)(void))

Where: proto_id   - the value returned by "proto_register_protocol()" when
		    the protocol was registered
	 apply_cb - Callback routine that is call when preferences are applied


Then you can register the fields that can be configured by the user with these routines -

	/* Register a preference with an unsigned integral value. */
	void prefs_register_uint_preference(module_t *module, const char *name,
	    const char *title, const char *description, guint base, guint *var);

	/* Register a preference with an Boolean value. */
	void prefs_register_bool_preference(module_t *module, const char *name,
	    const char *title, const char *description, gboolean *var);

	/* Register a preference with an enumerated value. */
	void prefs_register_enum_preference(module_t *module, const char *name,
	    const char *title, const char *description, gint *var,
	    const enum_val_t *enumvals, gboolean radio_buttons)

	/* Register a preference with a character-string value. */
	void prefs_register_string_preference(module_t *module, const char *name,
	    const char *title, const char *description, char **var)

Where: module - Returned by the prefs_register_protocol routine
	 name     - This is appended to the name of the protocol, with a
		    "." between them, to construct a name that identifies
		    the field in the preference file; the name itself
		    should not include the protocol name, as the name in
		    the preference file will already have it
	 title    - Field title in the preferences dialog
	 description - Comments added to the preference file above the 
		       preference value
	 var	  - pointer to the storage location that is updated when the
		    field is changed in the preference dialog box
	 enumvals - an array of enum_val_t structures.  This must be
		    NULL-terminated; the members of that structure are:

			a short name, to be used with the "-o" flag - it
			should not contain spaces or upper-case letters,
			so that it's easier to put in a command line;

			a description, which is used in the GUI (and
			which, for compatibility reasons, is currently
			what's written to the preferences file) - it can
			contain spaces, capital letters, punctuation,
			etc.;

			the numerical value corresponding to that name
			and description
	 radio_buttons - TRUE if the field is to be displayed in the
			 preferences dialog as a set of radio buttons,
			 FALSE if it is to be displayed as an option
			 menu

An example from packet-beep.c -
	
  proto_beep = proto_register_protocol("Blocks Extensible Exchange Protocol",
				       "BEEP", "beep");

	...

  /* Register our configuration options for BEEP, particularly our port */

  beep_module = prefs_register_protocol(proto_beep, proto_reg_handoff_beep);

  prefs_register_uint_preference(beep_module, "tcp.port", "BEEP TCP Port",
				 "Set the port for BEEP messages (if other"
				 " than the default of 10288)",
				 10, &global_beep_tcp_port);

  prefs_register_bool_preference(beep_module, "strict_header_terminator", 
				 "BEEP Header Requires CRLF", 
				 "Specifies that BEEP requires CRLF as a "
				 "terminator, and not just CR or LF",
				 &global_beep_strict_term);

This will create preferences "beep.tcp.port" and
"beep.strict_header_terminator", the first of which is an unsigned
integer and the second of which is a Boolean.

2.7 Reassembly/desegmentation for protocols running atop TCP

There are two main ways of reassembling PDUs spanning across multiple
TCP segmentss.  The first one is simpler, but assumes you are running
atop of TCP when this occurs (but your dissector might run atop of UDP,
too, for example), and that your PDUs consist of a fixed amount of data
that includes enough information to determine the PDU length, possibly
followed by additional data.  The second one is more generic but
requires more code and is less efficient.

For the first method, you register two different dissection methods, on
for the TCP case, and one for the other cases.  It is a good idea to
have a dissect_PROTO_common function which will parse the generic
content that you can find in all PDUs which is called from
dissect_PROTO_tcp when the reassembly is complete and from
dissect_PROTO_udp (or dissect_PROTO_other).

To register the distinct dissector functions, consider the following
example, stolen from packet-dns.c:

	dissector_handle_t dns_udp_handle;
	dissector_handle_t dns_tcp_handle;
	dissector_handle_t mdns_udp_handle;

	dns_udp_handle = create_dissector_handle(dissect_dns_udp,
	    proto_dns);
	dns_tcp_handle = create_dissector_handle(dissect_dns_tcp,
	    proto_dns);
	mdns_udp_handle = create_dissector_handle(dissect_mdns_udp,
	    proto_dns);

	dissector_add("udp.port", UDP_PORT_DNS, dns_udp_handle);
	dissector_add("tcp.port", TCP_PORT_DNS, dns_tcp_handle);
	dissector_add("udp.port", UDP_PORT_MDNS, mdns_udp_handle);
	dissector_add("tcp.port", TCP_PORT_MDNS, dns_tcp_handle);

The dissect_dns_udp function does very little work and calls
dissect_dns_common, while dissect_dns_tcp calls tcp_dissect_pdus with a
reference to a callback which will be called with reassembled data:

	static void
	dissect_dns_tcp(tvbuff_t *tvb, packet_info *pinfo, proto_tree *tree)
	{
		tcp_dissect_pdus(tvb, pinfo, tree, dns_desegment, 2,
		    get_dns_pdu_len, dissect_dns_tcp_pdu);
	}

(The dissect_dns_tcp_pdu function acts similarly to dissect_dns_udp.) 
The arguments to tcp_dissect_pdus are:

	the tvbuff pointer, packet_info pointer, and proto_tree pointer
	passed to the dissector;

	a gboolean flag indicating whether desegmentation is enabled for
	your protocol;

	the number of bytes of PDU data required to determine the length
	of the PDU;

	a routine that takes as arguments a tvbuff pointer and an offset
	value representing the offset into the tvbuff at which a PDU
	begins and should return - *without* throwing an exception (it
	is guaranteed that the number of bytes specified by the previous
	argument to tcp_dissect_pdus is available, but more data might
	not be available, so don't refer to any data past that) - the
	total length of the PDU, in bytes;

	a routine that's passed a tvbuff pointer, packet_info pointer,
	and proto_tree pointer, with the tvbuff containing a
	possibly-reassembled PDU, and that should dissect that PDU.

The second method is to return a modified pinfo structure when
dissect_PROTO is called.  In this case, you have to check if you have
collected enough bytes: if you have enough, you parse the PDU, and if
don't have enough bytes, you return from the dissector supplying
information to the caller on how many bytes you need to proceed.  This
is done by indicating the offset where you would like to start again and
the number of bytes that you need in pinfo->desegment_*:

	if (i_miss_five_bytes) {
		pinfo->desegment_offset = offset;
		pinfo->desegment_len = 5;
	}

You can repeat this procedure until you've got enough bytes; for
example, you can request one byte more until you've got the byte you're
searching for if the data to be dissected consists of a sequence of
bytes ending with a particular byte value.

3. Plugins

See the README.plugins for more information on how to "pluginize" 
a dissector.

4.0 Extending Wiretap.

5.0 How the Display Filter Engine works

code:
epan/dfilter/* - the display filter engine, including
		scanner, parser, syntax-tree semantics checker, DFVM bytecode
		generator, and DFVM engine.
epan/ftypes/* - the definitions of the various FT_* field types.
epan/proto.c   - proto_tree-related routines

5.1 Parsing text

The scanner/parser pair read the string representing the display filter
and convert it into a very simple syntax tree.  The syntax tree is very
simple in that it is possible that many of the nodes contain unparsed
chunks of text from the display filter.

5.1 Enhancing the syntax tree.

The semantics of the simple syntax tree are checked to make sure that
the fields that are being compared are being compared to appropriate
values.  For example, if a field is an integer, it can't be compared to
a string, unless a value_string has been defined for that field.

During the process of checking the semantics, the simple syntax tree is
fleshed out and no longer contains nodes with unparsed information.  The
syntax tree is no longer in its simple form, but in its complete form.

5.2 Converting to DFVM bytecode

The syntax tree is analyzed to create a sequence of bytecodes in the
"DFVM" language.  "DFVM" stands for Display Filter Virtual Machine.  The
DFVM is similar in spirit, but not in definition, to the BPF VM that
libpcap uses to analyze packets.

A virtual bytecode is created and used so that the actual process of
filtering packets will be fast.  That is, it should be faster to process
a list of VM bytecodes than to attempt to filter packets directly from
the syntax tree.  (heh...  no measurement has been made to support this
supposition)

5.3 Filtering

Once the DFVM bytecode has been produced, it's a simple matter of
running the DFVM engine against the proto_tree from the packet
dissection, using the DFVM bytecodes as instructions.  If the DFVM
bytecode is known before packet dissection occurs, the
proto_tree-related code can be "primed" to store away pointers to
field_info structures that are interesting to the display filter.  This
makes lookup of those field_info structures during the filtering process
faster.


6.0 Adding new capabilities.




James Coe <jammer@cin.net>
Gilbert Ramirez <gram@alumni.rice.edu>
Jeff Foster <jfoste@woodward.com>
Olivier Abad <oabad@cybercable.fr>
Laurent Deniel <laurent.deniel@free.fr>
Gerald Combs <gerald@ethereal.com>
Guy Harris <guy@alum.mit.edu>