Deadlock caused by ICE media transport (thanks Alain Totouom for the report)

[bennylp]

The mutex locking in ICE transport may cause deadlock in application. Consider this scenario:

Thread 1:

• RTP packet arrives
• the pj_ice_sess_on_rx_pkt() is called
• ICE session's mutex is acquired
• ice->cb.on_rx_data() callback is called
• application acquires some application mutex B (e.g. conference bridge's mutex).

Thread 2:

• application acquires some application mutex B (e.g. conference bridge's mutex)
• RTP packet needs to be transmitted
• the pj_ice_sess_send_data() function is called
• thread is acquiring ICE session's mutex, and it will deadlock.

(Stable branch: the corresponding ticket for the 1.0 branch is ticket #691)

[bennylp]

Fixed in r2419:

• don't hold the ICE session's mutex when calling ice->cb.on_rx_data() callback. We don't need to hold this since ICE session's lifetime is already protected by the ICE transport (ice_strans).