On 10/15/13 12:00, Christopher Maynard wrote:
What is the proper, safe and recommended way to use pinfo->private_data?
If you look at the TCP dissector, you see the following:
pd_save = pinfo->private_data;
TRY {
(*dissect_pdu)(next_tvb, pinfo, tree);
}
CATCH_NONFATAL_ERRORS {
/* Restore the private_data structure in case one of the
* called dissectors modified it (and, due to the exception,
* was unable to restore it).
*/
pinfo->private_data = pd_save;
show_exception(tvb, pinfo, tree, EXCEPT_CODE, GET_MESSAGE);
}
ENDTRY;
I did not check all dissectors, but this is certainly not done for all of
them. I guess the question I have is, should it? Is this the recommended
approach?
I started adding that back in r34408 (because I was getting a crash
after a malformed packet exception). At the time I went through most of
the dissectors and tried to make the same change (e.g., r34436 but I
think there were more revisions). Of course I may have missed some and
of course that was a long time ago and new dissectors have arrived...
But, yes, I think that when using private_data the restoration is
necessary: otherwise if you have multiple PDUs in one frame then an
exception in one of the early PDUs will cause the private_data to be
wrong when the next PDU is dissected.
This is one of the reasons not to use private_data at all.