Huge thanks to our Platinum Members Endace and LiveAction,
and our Silver Member Veeam, for supporting the Wireshark Foundation and project.

Wireshark-dev: Re: [Wireshark-dev] Babel: prevent an infinite loop while parsing sub-TLV

From: Pascal Quantin <pascal@xxxxxxxxxxxxx>
Date: Fri, 18 Oct 2019 21:21:59 +0200
Le ven. 18 oct. 2019 à 21:16, Pascal Quantin <pascal@xxxxxxxxxxxxx> a écrit :


Le ven. 18 oct. 2019 à 20:58, Pascal Quantin <pascal@xxxxxxxxxxxxx> a écrit :
Hi Juliusz,

Le ven. 18 oct. 2019 à 20:51, Juliusz Chroboczek <jch@xxxxxxx> a écrit :
Dear Pascal,

I've just seen your commit dd15b2, which I believe is incorrect.

  https://code.wireshark.org/review/gitweb?p=wireshark.git;a=commit;h=dd15b203c7ec8a8ec5c930cf018c838991ee3182

In dissect_babel_subtlvs, sublen is the length of the TLV body; the full
TLV is of length sublen+2.  At the end of the loop, the code is correct:

        beg += (sublen+2);

Hence, it is perfectly fine to have sublen=0.  Unless I'm wrong, your
commit is incorrect.

Maybe, honestly I do not remember the details as it was done a few months ago. It would require me to dive in once I have the time (so not now).


Could you please either explain why I'm wrong, or else revert commit dd15b2?

I will not revert this commit as it solved/workarounded an infinite loop, as seen in https://bugs.wireshark.org/bugzilla/show_bug.cgi?id=15856
But is you can come up with a better solution that ensures that we do not have the infinite loop while still being correct, you are welcome and I will try to review it

Without even looking at the capture from bug 15856, I can see an infinite loop when subtype = MESSAGE_SUB_PAD1 and sublen = 0:
        if(subtype == MESSAGE_SUB_PAD1){
            beg += sublen;
            continue;
        }
Maybe the test for sublen value should be put only when the subtype is MESSAGE_SUB_PAD1? Or when subtype is different from MESSAGE_SUB_PADN?
As you seem to know this protocol, your feedback is welcome.

And the code below seems to assume that sublen cannot be 0:
        if (message_tree) {
            subtlv_tree = proto_item_add_subtree(sub_item, ett_subtlv);
            proto_tree_add_item(subtlv_tree, hf_babel_subtlv_type,
                                tvb, beg+2, 1, ENC_BIG_ENDIAN);
        }
This is probably why I added the check just before, as those lines do not consider a case where sublen would be 0. But after rereading the code, I guess it is currently selecting the wrong byte and that beg+2 should be replaced by beg?
And in the same area, in the code below:
        sub_item =
          proto_tree_add_uint_format(message_tree, hf_babel_subtlv,
                                     tvb, beg, sublen, subtype,
                                     "Sub TLV %s (%u)",
                                     val_to_str_const(subtype, subtlvs, "unknown"),
                                     subtype);
I guess sublen should be replaced by sublen+2?


Best regards,
Pascal.