as raw TCP segment data under the TCP protocol tree item, rather than as
a top-level data item - and do so even for the last of the segments
reassembled into that packet.
svn path=/trunk/; revision=4754
RPC and NDMP.
Show the RPC-over-TCP fragment header as a tree with bitfields below it.
Add a routine to show a reported bounds error as an "Unreassembled
Packet" or a "Malformed Packet" depending on whether "pinfo->fragmented"
is set, and have NBNS and RPC use that.
Add "ett_ndmp_file_stats" to the list of ett_ values to be initialized
(it wasn't in that list, and wasn't getting initialized).
When freeing up various hash tables and memory chunks in the RPC
dissector, zero out the pointers to them, just to make sure we don't try
to free them again.
Always destroy the TCP segment key and address memory chunks in
"tcp_desegment_init()", regardless of whether TCP desegmentation is
enabled - we don't *allocate* them if TCP desegmentation isn't enabled,
but we should free them even if it's not enabled. Also, when we free
them, set the pointers to them to null, so we don't double-free them.
Supply to subdissectors called from the TCP dissector the sequence
number of the first byte handed to the sub dissector.
svn path=/trunk/; revision=4753
"data source" has a name and a top-level tvbuff, and frames can have a
list of data sources associated with them.
Use the tvbuff pointer to determine which data source is the data source
for a given field; this means we don't have to worry about multiple data
sources with the same name - the only thing the name does is label the
notebook tab for the display of the data source, and label the hex dump
of the data source in print/Tethereal output.
Clean up a bunch of things discovered in the process of doing the above.
svn path=/trunk/; revision=4749
reassembled TCP data being able to indicate that they need still more
reassembly, so that, for example, a dissector can indicate that it needs
reassembly in order to dissect a header that says how long the PDU is
and, when that reassembly is done and it dissects the header, it can
then indicate that it needs more reassembly to get the entire PDU.
svn path=/trunk/; revision=4694
"epan/..." pathnames, so as to avoid collisions with header files in any
of the directories in which we look (e.g., "proto.h", as some other
package has its own "proto.h" file which it installs in the top-level
include directory).
Don't add "-I" flags to search "epan", as that's no longer necessary
(and we want includes of "epan" headers to fail if the "epan/" is left
out, so that we don't re-introduce includes lacking "epan/").
svn path=/trunk/; revision=4586
items to the protocol tree; it's interpreted as "the rest of the data in
the tvbuff". This can be used if
1) the item covers the entire packet or the remaining payload in
the packet
or
2) the item's length won't be known until it's dissected, and
will be then set with "proto_item_set_len()" - if an
exception is thrown in the dissection, it means the item ran
*past* the end of the tvbuff, so saying it runs to the end of
the tvbuff is reasonable.
Convert a number of "proto_tree_add_XXX()" calls using
"tvb_length_remaining()", values derived from the result of
"tvb_length()", or 0 (in the case of items whose length is unknown) to
use -1 instead (using 0 means that if an exception is thrown, selecting
the item highlights nothing; using -1 means it highlights all the data
for that item that's available).
In some places where "tvb_length()" or "tvb_length_remaining()" was used
to determine how large a packet is, use "tvb_reported_length()" or
"tvb_reported_length_remaining()", instead - the first two calls
indicate how much captured data was in the packet, the latter two calls
indicate how large the packet actually was (and the fact that using the
latter could cause BoundsError exceptions to be thrown is a feature - if
such an exception is thrown, the frame really *was* short, and it should
be tagged as such).
Replace some "proto_tree_add_XXX()" calls with equivalent
"proto_tree_add_item()" calls.
Fix some indentation.
svn path=/trunk/; revision=4578
desegmentation even though we don't know whether the checksum is valid).
I've seen packets with bad TCP checksums in Solaris network traces, but
the traffic appears to indicate that the packet *was* received; I
suspect the packets were sent by the host on which the capture was being
done, on a network interface to which checksumming was offloaded, so
that DLPI supplied an un-checksummed packet to the capture program but a
checksummed packet got put onto the wire.
svn path=/trunk/; revision=4571
setting the "pinfo->fragmented" flag.
If a ReportedBoundsError occurs, flag the frame as being an
unreassembled packet, not an unreassembled fragmented packet, as it may
have been segmented across TCP segment boundaries rather than being part
of an IPv4/IPv6/CLNP/etc. fragmented/segmented packet.
svn path=/trunk/; revision=4558
the list of segments in a desegmented PDU as unsigned, rather than
signed.
Fix some other displays of unsigned quantities with "%d" while we're at
it.
svn path=/trunk/; revision=4516
structure to the "packet_info" structure; only stuff that's permanently
stored with each frame should be in the "frame_data" structure, and the
"column_info" structure is not guaranteed to hold the column values for
that frame at all times - it was only in the "frame_data" structure so
that it could be passed to dissectors, and, as all dissectors are now
passed a pointer to a "packet_info" structure, it could just as well be
put in the "packet_info" structure.
That saves memory, by shrinking the "frame_data" structure (there's one
of those per frame), and also lets us clean up the code a bit.
svn path=/trunk/; revision=4370
specifies how the selector values used as keys in those tables are to be
displayed, and the title to use when displaying the table.
Use that information in the code to display the initial and current
entries of various dissector tables.
Have the dissector for BACnet APDUs register itself by name, and have
the BACnet NPDU dissector call it iff the BAC_CONTROL_NET bit isn't set,
rather than doing it with a dissector table.
svn path=/trunk/; revision=4358
dissector table contain both a hash table, to use to look up port
numbers to find a dissector, and a list of all dissectors that *could*
be assigned to ports in that hash table, to be used by user interface
code.
Make the "Decode As" dialog box code use that.
Also make it *not* let you choose whether to set the dissector for both
the UDP and TCP versions of a port; some protocols run only atop TCP,
some run only atop UDP, and even those that can run atop both may have
different dissector handles to use over TCP and UDP, so handling a
single merged list would be a mess. (If the user is setting the
dissector for a TCP port, only those protocols that Ethereal can handle
over TCP should be listed; if the user is setting the dissector for a
UDP port, only those protocols that Ethereal can handle over TCP should
be listed; if the user is setting a dissector for both, only those
protocols that Ethereal can handle over *both* TCP *and* UDP should be
listed, *and* there needs to be a way to let the "Decode As" code get
both the TCP handle *and* the UDP handle and use the right ones. If
somebody really wants that, they need to implement all of the above if
they want the code to be correct.)
Fix the code that handles setting the dissection for the IP protocol
number to correctly update the lists of protocols being dissected as TCP
and as UDP; the code before this change wasn't updating the single such
list to add new protocols.
svn path=/trunk/; revision=4311
take a dissector handle as an argument, rather than a pointer to a
dissector function and a protocol ID. Associate dissector handles with
dissector table entries.
svn path=/trunk/; revision=4308
1. Changes how can_desegment works so that can_desegment is
only != 0 for whichever dissector is running immediately on
top of whoever offers the can_desegment service.
Thus DCERPC needs no special handling to see if it can trust
can_desegment (which is currently only available ontop of TCP
and not ontop of tcp->nbss->smb).
2. Changes fragment reassembly of transaction smb to only show
the defragmented packet for the transaction smb holding the
first fragment.
To see why, test it with a transaction SMB containing a ~60kb
PDU or larger. The old behaviour had approximately quadratic
behaviour regarding runtime for dissecting such PDUs.
(example: NetShareEnum is a command which can grow really really
large if the number of shares and comments are large)
svn path=/trunk/; revision=4296
structure, we may have to worry about it in more places than the places
that *used* to set "pi.len" and "pi.captured_len", so there's no point
in just saving and restoring it there. We'll remove those
saves/restores, and worry about saves and restores when we find a
problem.
svn path=/trunk/; revision=4245
structure, the check for a null tvbuff pointer in "alloc_field_info()",
and the "tvb_create_from_top()" macro; they're no longer needed, as
there's no non-tvbuffified dissector code remaining.
svn path=/trunk/; revision=4205
stuff currently being dissected is part of a packet included in an error
packet (e.g., an ICMP Unreachable packet). Have the TCP dissector not
bother doing reassembly if the TCP segment is part of an error packet,
rather than an actual TCP transmission; other dissectors might want to
treat those packets specially as well.
Add to the "tcpinfo" structure a flag indicating whether the URG flag
was set, rather than having the zero or non-zero value of the urgent
pointer indicate that. (Yes, at least as I read RFC 793, a zero urgent
pointer value isn't useful, as it means "the stuff before this segment
is urgent", but it's certainly possible to put onto the wire a TCP
segment with URG set and a zero urgent pointer.)
Don't dissect the TCP header by grabbing the entire header with
"tvb_memcpy()" and then pulling stuff out of it - extract stuff with
individual tvbuff calls, and put stuff into the protocol tree and the
Info column as we extract it, so that we can dissect a partial header.
This lets us, for example, get the source and destination ports from the
TCP header of the part of a TCP segment included in a minimum-length
ICMPv4 error packet.
svn path=/trunk/; revision=3986
variable that holds it an "int" rather than a "guint16".
Further strengthen the heuristics the NBSS dissector uses to distinguish
NBSS messages from continuations of NBSS messages.
If an frame contains an NBSS continuation, put the protocol tree item
for the continuation data under an NBSS protocol tree item.
Have the TCP dissector supply information to subdissectors via a "struct
tcpinfo" pointed to by "pinfo->private"; move the urgent pointer value
from a global variable into that structure, and add a Boolean flag that
indicates whether the data it's handing to a subdissector is reassembled
data or not.
Make the NBSS dissector check for continuations only in non-reassembled
data.
Fix the computation, in the TCP dissector, of the offset into the tvbuff
handed to the subdissector of the first byte of stuff that needs further
reassembly, and fix the computation of the sequence number corresponding
to that byte.
svn path=/trunk/; revision=3984
dissectors to use it, from Ronnie Sahlberg, with additional changes to
handle the case where a frame contains messages that don't run past the
end followed by one that does and where a reassembled chunk has, at the
end, a message that runs past the end of that chunk (because the
reassembly was for an earlier message).
svn path=/trunk/; revision=3923
"proto_tree_add_item_hidden()", to add the "checksum bad" flags to
packets; the value should be "TRUE", not the numerical value of the
checksum field.
svn path=/trunk/; revision=3202
Initialize the "hf_" value for "icmp.checksum_bad" to -1, the way all
other "hf_" values are initialized, and declare it and "ip.checksum_bad"
to have base BASE_NONE, not 4.
svn path=/trunk/; revision=3087
requires that the dfilter code be initialized before the plugins are
added; this required us to *re*-initialize the dfilter code after
reading in all the plugins, as the plugins may themselves have added new
filterable fields - that was a bit of a mess), and make the
"Tools->Plugins" dialog box show the new-style plugins.
svn path=/trunk/; revision=2950
statements.
Move the setting of the Protocol column in various dissectors before
anything is fetched from the packet, and also clear the Info column at
that point in those and some other dissectors, so that if an exception
is thrown, the columns don't reflect the previous protocol.
Make the IP dissector static, as it's called only via dissector tables
or dissector handles. Also make the "dissect the TOS field as the
DiffServ DS field" flag static, as it's not referred to outside of
"packet-ip.c".
In the NCP dissector, refer to the port type through "pinfo" rather than
through the global "pi", as it's a tvbuffified dissector.
svn path=/trunk/; revision=2929
"{old_}heur_dissector_add()", "{old_}conv_dissector_add()", and
"register_dissector()", so that an entry in those tables has associated
with it the protocol index of the protocol the dissector handles (or -1,
if there is no protocol index for it).
This is for future use in a number of places.
(Arguably, "proto_register_protocol()" should take a dissector pointer
as an argument, but
1) it'd have to handle both regular and heuristic dissectors;
2) making it take either a "dissector_t" or a union of that and
a "heur_dissector_t" introduces some painful header-file
interdependencies
so I'm punting on that for now. As with other Ethereal internal APIs,
these APIs are subject to change in the future, at least until Ethereal
1.0 comes out....)
svn path=/trunk/; revision=2849
particular protocols, and which keep track of all dissectors that could
be associated with conversations using those particular protocols - for
example, the RTP and RTCP dissectors could be assigned to UDP
conversations.
This is for future use with UI features allowing the dissector for a
given conversation to be set from the UI, to allow
1) conversations between two ports, both of which have
dissectors associated with them, that have been given to the
wrong dissector to be given to the right dissector;
2) conversations between two ports, neither of which have
dissectors associated with them, to be given to a dissector
(RTP and RTCP, for example, typically run on random ports,
and if you don't have, in a capture, traffic that would say
"OK, traffic between these two hosts and ports will be RTP
traffic", you may have to tell Ethereal explicitly what
protocol the conversation is).
svn path=/trunk/; revision=2848
"prefs_register_module()" except that it takes a protocol index as
returned by "proto_register_protocol()" as its first argument, rather
than taking two character strings as arguments as its first two
arguments, and uses the protocol's abbreviation as the name to use for
preferences in the preferences file and the "-o" flag and uses the
protocol's short name as the name to use in the tabs in the
"Edit->Preferences" window.
svn path=/trunk/; revision=2812
protocols, in addition to adding structures to the list of filterable
fields. Give it an extra argument that specifies a "short name" for the
protocol, for use in such places as
pinfo->current_proto;
the dialog box for constructing filters;
the preferences tab for the protocol;
and so on (although we're not yet using it in all those places).
Make the preference name that appears in the preferences file and the
command line for the DIAMETER protocol "diameter", not "Diameter"; the
convention is that the name in question be all-lower-case.
Make some routines and variables that aren't exported static.
Update a comment in the ICP dissector to make it clear that the
dissector won't see fragments other than the first fragment of a
fragmented datagram.
svn path=/trunk/; revision=2810
replace the existing checksummer with a modified version of the BSD
checksumming code. Add a flag to the "packet_info" structure to
indicate that a packet is the first fragment of a fragmented datagram,
so that the checksummers won't try to checksum those.
(It doesn't seem to add a lot of CPU overhead, so we don't introduce a
flag to disable it, yet. Further checks may be necessary to see whether
the overhead is just swamped by other overheads when scanning through a
capture dissecting all frames, or if it truly is negligible.)
Make the Boolean preference option controlling whether to make the
top-level protocol tree item for TCP display a packet summary static to
the TCP dissector (it doesn't need to be accessible outside the TCP
dissector).
svn path=/trunk/; revision=2751
doesn't just seem to be a profiling artifact) that, at least on FreeBSD
3.4, it's significantly more efficient to clear out a column by stuffing
a '\0' into the first byte of the column data than to do so by copying a
null string (I guess when copying one byte, the fixed overhead of the
procedure call and of "strcpy()" is significant).
Have the TCP dissector set the Protocol column, and clear the Info
column, before doing anything that might cause an exception to be
thrown, so that if we *do* get an exception thrown, the frame at least
shows up as TCP.
Instead of, in the TCP dissector, constructing a string and then
stuffing it into the Info column, just append to the Info column, which
avoids one string copy.
Pass a "frame_data" pointer to dissectors for TCP and IP (and PPP)
options, so they can use it to append to the Info column.
svn path=/trunk/; revision=2744
can be put, and a pointer to the string for the column, which might or
might not point to that buffer.
Add a routine "col_set_str()", which sets the string for the column to
the string passed to it as an argument; it should only be handed a
static string (a string constant would be ideal). It doesn't do any
copying, so it's faster than "col_add_str()".
Make the routines that append to columns check whether the pointer to
the string for the column points to the buffer for the column and, if
not, copy the string for the column to the buffer for the column so that
you can append to it (so you can use "col_set_str()" and then use
"col_append_str()" or "col_append_fstr()").
Convert a bunch of "col_add_str()" calls that take a string constant as
an argument to "col_set_str()" calls.
Convert some "col_add_fstr()" calls that take a string constant as the
only argument - i.e., the format string doesn't have any "%" slots into
which to put strings for subsequent arguments to "col_set_str()" calls
(those calls are just like "col_add_str()" calls).
Replace an END_OF_FRAME reference in a tvbuffified dissector with a
"tvb_length(tvb)" call.
svn path=/trunk/; revision=2670
ESIS dissectors.
Register the IP dissector and have dissectors that call it directly
(rather than through a port table) call it through a handle.
Add a routine "tvb_set_reported_length()" which a dissector can use if
it was handed a tvbuff that contains more data than is actually in its
part of the packet - for example, handing a padded Ethernet frame to IP;
the routine sets the reported length of the tvbuff (and also adjusts the
actual length, as appropriate). Then use it in IP.
Given that, "ethertype()" can determine how much of the Ethernet frame
was actually part of an IP datagram (and can do the same for other
protocols under Ethernet that use "tvb_set_reported_length()"; have it
return the actual length, and have "dissect_eth()" and "dissect_vlan()"
use that to mark trailer data in Ethernet II frames as well as in 802.3
frames.
svn path=/trunk/; revision=2658
after a TCP segment, so I can see what stuff some other segment is
ACKing, I'll go crazy. Add a "Next sequence number" field to the TCP
dissection, giving exactly that (well, giving exactly that unless the
TCP segment is in a fragmented IP datagram, but hopefully those are
rare; when we support IP fragment reassembly, we can fix that).
svn path=/trunk/; revision=2453