Update XPACK.cc to fix use-of-uninitialized-pointer-field case

[shukitchan]

Fuzzing reveals a use-of-uninitialized-pointer-field case

https://oss-fuzz.com/testcase-detail/6285050312196096

I think this should fix the problem

[maskit]

I'm not too sure if we should do this, the memset. Use of uninitialized field implies reading an entry that does not exist. I think we should check index numbers instead so such access does not happen. Touching the entire buffer would suppress the report, but it can hide real issues.

[shukitchan]

Let me give it more thoughts

[shukitchan]

@maskit
I double checked and i think it should be the fix we should do.

If we start from the beginning and insert an oversized entry, _make_space() will always look at the tail entry. So memset to 0 will make we fail gracefully. That's the problem the fuzz test is having.

If we already have entries and now add an oversized entry, we need _make_space() to at least clear out all entries first before giving up. So we can't have a precondition check for an oversized entry. That is implied in the unit test - https://github.com/apache/trafficserver/blob/master/src/proxy/hdrs/unit_tests/test_XPACK.cc#L318

[maskit]

Shouldn't we call is_empty() at the beginning of _make_space() and return false if the table is empty?

[shukitchan]

Shouldn't we call is_empty() at the beginning of _make_space() and return false if the table is empty?

I agree. I just made a change for that.

[shukitchan]

[approve ci fedora]

[shukitchan]

[approve ci autest]

[shukitchan]

@maskit it is ready for review again. Thanks

[cmcfarlen]

Cherry-picked to v10.0.x