(also registry and srvsvc?) pipe dissectors.
Also added some helper routines for decoding a range of uint8s. These map
nicely to the dissect_{dcerpc,ndr}_foo() format with the addition of a
length parameter, but aren't really part of the NDR specification and are
probably only going to be used in the NT dcerpc dissectors.
svn path=/trunk/; revision=5421
frame number arguments, and elements in data structures, unsigned,
display them with "%u" rather than "%d", and use 0, rather than -1, as
"not known".
svn path=/trunk/; revision=5223
"packet-dcerpc-nt.c", and registers "dcerpc_smb_init()" as an
initialization routine. Take the ett_ registration out of the latter
routine, and also take out the "do this only once" stuff.
Get rid of the initialization routines for netlogon, samr, and spoolss;
they just call "dcerpc_smb_init()", which is now an initialization
routine of its own.
The policy hash initialization should be done before every capture, so
it should be done in an initialization routine, and should not do any
"do this only once" stuff. It should also be called only once before
every capture, rather than 3 times.
The ett_ initialization should, however, be done at the same time all
other ett_ initialization is done - at protocol registration time - so
it should be done in a "proto_register_" routine.
This fixes a bug I saw wherein
1) the tree for Unicode strings was open by default
and
2) if you closed one and then exited, Ethereal would crash.
The problem is that "proto_register_subtree_array()" doesn't expand the
array, it just bumps the number of registered ett_ values; the array is
allocated in "proto_init()". As such, if you register ett_ values with
"proto_register_subtree_array()" *after* "proto_init()" is called - and,
even for the first capture, initialization routines are called after
"proto_init()" is called - you will get ett_ numbers that go past the
number of elements in the array.
Move the declaration of "ett_nt_unicode_string" to "packet-dcerpc-nt.h",
as it's exported from "packet-dcerpc-nt.c".
Get rid of the declaration of "dcerpc_smb_init()" in
"packet-dcerpc-nt.h", and make it static, as it's no longer called from
outside "packet-dcerpc-nt.c".
svn path=/trunk/; revision=5196
- rewrote request/response hashing code and moved it into
packet-dcerpc-nt.c for use by other dcerpc/smb dissectors
- rewrote policy handle hashing code as above, with nifty improvements
- moved check for unparsed data and end of frame into a function
- fixed matching of printername to handle in addprinterex
svn path=/trunk/; revision=5014
neatly aligned on a 2-byte or a 4-byte boundary, and there is no
guarantee that a misaligned pointer can be dereferenced without getting
a fault.
Furthermore, there is no guarantee that, even if you *can* dereference a
pointer to a 2-byte or 4-byte quantity in a packet, the resulting number
you get back is in the right byte order; the data in the packet might
have a different byte order from the machine on which you're running.
Therefore, we change "prs_uint8s()", "prs_uint16s()", and
"prs_uint32s()" to return the starting offset, in the tvbuff, of the
collection of 8-bit, 16-bit, or 32-bit integral values, rather than a
pointer to the raw packet data, and change their callers to fetch the
data using "tvb_get_guint8()", "tvb_get_letohs()", and
"tvb_get_letohl()" (the stuff in all the NT protocols is presumed to be
little-endian here). We also change "fake_unicode()" to take a tvbuff
and an offset, rather than a data pointer, as arguments, and to use
"tvb_get_letohs()" to fetch the Unicode characters (again, we assume
little-endian Unicode).
This requires "fake_unicode()" to establish a cleanup handler, so we
don't leak memory if it throws an exception.
We also make "fake_unicode()" use "g_malloc()" to allocate its buffer
(we weren't checking for allocation failures in any case; with
"g_malloc()", we'll abort on an allocation failure - if we can come up
with a cleverer way of handling them, fine), and the matching frees to
use "g_free()". (We also insert some missing frees....)
Fix some formats to print unsigned quantities with "%u", not "%d".
Don't append text to items in the tree for non-string values in
"dissect_ndr_nt_STRING_string()".
svn path=/trunk/; revision=4986
"dissect_ndr_nt_UNICODE_STRING_string()", in
"samr_dissect_connect2_server()"; that eliminates an unnecessary extra
level of protocol tree.
That removes the last call to "dissect_ndr_nt_UNICODE_STRING_string()";
eliminate that routine.
In "dissect_ndr_nt_UNICODE_STRING()", initially create the subtree with
the name of the field as a string, so that if an exception is thrown
before the name is set, the subtree won't show up as blank when
displayed or printed. Also pass in the name to "dissect_ndr_pointer()",
so the same happens for subtrees below it. Append only the string data,
not its name, to items up the tree, as the name was put in when the item
was created. Also, when adding a colon before the string, put a space
after the colon, as is done elsewhere in Ethereal.
When appending additional strings, put the blank before the new string,
not after it.
In "dissect_ndr_nt_STRING()", put the subtree into the string with the
name of the field, rather than just "String". Pass in that name to
"dissect_ndr_pointer()", so subtrees below it get a name when they're
initially created.
Get rid of colons in the name string passed to "dissect_ndr_pointer()"
in some calls. Supply a non-null name string in more calls to
"dissect_ndr_pointer()", and fix some calls to pass in the name of the
field being handed to "dissect_ndr_pointer()".
There's no need to fetch the entire "header_field_info" structure for a
protocol field in order to get the field's name - just use
"proto_registrar_get_name()" to get the name.
Use a length of -1, not 0, when creating a subtree whose length will be
set when the dissection of the items under the subtree is complete; that
way, if an exception is thrown while dissecting the items - which means
the item goes past the end of the tvbuff - the item will refer to all
data to the end of the tvbuff, rather than referring to nothing.
Fix a typo in the name of the "hf_samr_unknown_string" field.
svn path=/trunk/; revision=4912
bitmap and make the function not static.
This bitmap is present in other RPC protocols for NT services as well
besides SAMR.
svn path=/trunk/; revision=4889
DOS error codes to the table of them, and exporting that table to other
dissectors for protocols using DOS error codes.
svn path=/trunk/; revision=4470
Tim Potter