[Pascal Quantin <pascal@xxxxxxxxxxxxx>]

Hi Moshe,

Le mer. 21 juil. 2021 à 17:56, Moshe Kaplan <mosheekaplan@xxxxxxxxx> a écrit :

Coverity is complaining that some of the allocations made with pinfo -> pool are leaking. Is it possible that the pinfo->pool based allocations are not always cleaned up?

As an example, CoverityID 1487512 complains about packet-tcp.c's calls to port_with_resolution_to_str leaking: https://gitlab.com/wireshark/wireshark/-/blob/master/epan/dissectors/packet-tcp.c#L6500 .

Most likely Coverity's analysis capabilities cannot properly handle a custom memory allocator like the one we use, where the garbage collector is performed out of scope of the current functions.

Looking at CID 1487512, this really seems to be the case: Coverity cannot know that the dynamically allocated buffers that are not stored in any local/global variable can be freed later on in epan_dissect_cleanup(). So at first glance it seems like a false positive (like many other ones as far as I know, maybe there is a way to provide directives to avoid those false reports). The problem is that using a NULL scope will fallback to g_malloc'ed memory that will leak, and this might be too subtle for a simple directive.

Best regards,

Pascal.

Moshe

On Wed, Jul 21, 2021 at 11:31 AM Evan Huus <eapache@xxxxxxxxx> wrote:

FYI this migration has now begun. Going forward, please use pinfo->pool instead of wmem_packet_scope() in new code when possible. And if anybody has some time, there are lots of existing dissectors left to convert. I expect most of them to be pretty straightforward, just adding pinfo to a few more method signatures as needed.

Thanks,

Evan

On Mon, Jul 12, 2021 at 11:52 Evan Huus <eapache@xxxxxxxxx> wrote:

I've been thinking recently about starting the process of getting rid
of the "global" wmem scope methods (wmem_packet_scope,
wmem_file_scope, etc) in favour of passing them around in arguments
(or in pinfo, or something). This would let us drop a bunch of
in-scope/out-of-scope tracking and assertion, as well as make the code
more amenable to future refactors like (potentially) concurrency.

At a first glance, we already have pinfo->pool which maintains the
lifetime of the packet_info object. As far as I can reason, this is
almost/effectively the same as the existing wmem_packet_scope - it
gets cleaned up later in the dissection flow, but there's still only
ever one which gets reused for each packet.

Is this correct? If so, does it make sense to start replacing
`wmem_packet_scope()` calls with `pinfo->pool` when pinfo is already
in scope?

Thanks,
Evan

___________________________________________________________________________
Sent via:    Wireshark-dev mailing list <wireshark-dev@xxxxxxxxxxxxx>
Archives:    https://www.wireshark.org/lists/wireshark-dev
Unsubscribe: https://www.wireshark.org/mailman/options/wireshark-dev
             mailto:wireshark-dev-request@xxxxxxxxxxxxx?subject=unsubscribe

___________________________________________________________________________
Sent via:    Wireshark-dev mailing list <wireshark-dev@xxxxxxxxxxxxx>
Archives:    https://www.wireshark.org/lists/wireshark-dev
Unsubscribe: https://www.wireshark.org/mailman/options/wireshark-dev
             mailto:wireshark-dev-request@xxxxxxxxxxxxx?subject=unsubscribe