Le 19 août 2013 à 13:24, Evan Huus <eapache () gmail com> a écrit :
On 2013-08-19, at 6:42 AM, Pascal Quantin <pascal.quantin () gmail com>
2013/8/17 Jakub Zawadzki <darkjames-ws () darkjames pl>
By the way from ROHC code:
2296 data = (guint8 *)ep_alloc(len);
2299 next_tvb = tvb_new_child_real_data(tvb, data, len, len);
2300 add_new_data_source(pinfo, next_tvb, "Payload");
It's very bad idea to use ep-allocated tvb as data source.
I fixed several dissectors using ep allocated memory for data source in
r51423. I hope I did not miss any.
That's great Pascal. However, there are some where if an exception gets
thrown before the tvb_set_free call then it could be a memory leak. If it
is not possible to move the set_free to immediately after the allocation
(so there is nothing to throw an exception) then I suggest allocating in
wmem's pinfo->pool instead.
I already used the pinfo pool for two dissectors (one because wmem was
already used and the opensafety one because ther ewas far too many code
lines and branches between allocation and the add as data source) and I
will do a quick pass to see if it can be used for other dissectors as well
when I have time.