[Bug 764744] Crashes when multiple udpsrc are created for each client on a shared media, misses tracking and cleanup

[GStreamer (GNOME Bugzilla)]

https://bugzilla.gnome.org/show_bug.cgi?id=764744

Jake Foytik <jake.foytik at ipconfigure.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
 Attachment #326051|none                        |needs-work
             status|                            |

--- Comment #15 from Jake Foytik <jake.foytik at ipconfigure.com> ---
Created attachment 326051
  --> https://bugzilla.gnome.org/attachment.cgi?id=326051&action=edit
initial patch

Sebastian, here is an initial patch with the changes. I've provided some notes
below and there are a couple sections I would like your input on.

Notes:
 -- I've assumed that there can exist multiple unicast udpsrcs but only a
single set of multicast udpsrcs (one for IPV4 and one for IPV6).
   - The multicast udpsrcs remain the same.
   - The unicast udpsrcs are now maintained in a hash table that uses the
RTSPTransport as a key. It didn't look like there was any special logic
required for handling IPV4 v. IPV6 except for the creation of the sockets, so
the hash table can contain both types.
   - When unicast sockets are allocated, the udpsrcs are stored in the hash
table. See the gst_rtsp_stream_allocate_udp_sockets() function.
   - When a client disconnects and the connection is unicast, we now cleanup
the udpsrcs related to the RTSPTransport.  See update_transport() function.
   - In the gst_rtsp_stream_leave_bin() function, all udpsrcs are now cleaned
up.
 -- I've made some fundamental changes to the
gst_rtsp_stream_allocate_udp_sockets() function that I think are correct, but
you should verify. They include:
    - IPV6 unicast sockets were only allowed to be allocated one time, in the
same manner as multicast. This is different for IPV4 unicast. I assumed that we
wanted to allow multiple IPV6 unicast connections, so I removed the 'have_ipv6'
check that would skip allocation.
    - I changed the logic for calculating the return 'result'. Before, it
performed 'result = have_ipv4 || have_ipv4_mcast || have_ipv6 ||
have_ipv6_mcast'. This would return TRUE if a client was trying to create an
IPV4 unicast connection and it failed to allocate the unicast udpsrcs, but a
client had previously connected with IPV6 multicast successfully.
 -- There are two sections of the code where the code has been commented out
with '#ifdef NOTHING'. Please review these portions. I think they are no longer
needed since we now postpone the creation of udpsrc elements to after the
join_bin function, but there may be a scenario I am missing.

Please review the patch and make recommendations. I'll make changes as needed.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.