Deadlock when registration retry and IP change handling triggered simultaneously

[nanangizz]

Registration retry acquires the PJSUA lock followed by the transport manager lock, whereas IP change acquires them in the reverse order.

Call stack trace

IP change:

pj_mutex_lock + 136
PJSUA_LOCK + 28
pjsua_acc_on_tp_state_changed + 112
on_tp_state_callback + 140
pjsip_transport_shutdown2 + 308
pjsip_tpmgr_shutdown_all + 312
pjsua_handle_ip_change + 604
pj::Endpoint::handleIpChange + 64

Registration retry:

pj_mutex_lock + 136
pj_lock_acquire + 116
pjsip_tpmgr_acquire_transport2 + 60
pjsip_endpt_acquire_transport2 + 72
pjsua_acc_get_uac_addr + 2076
pjsua_acc_create_uac_contact + 244
pjsua_regc_init + 384
pjsua_acc_set_registration + 564
auto_rereg_timer_cb + 496
pj_timer_heap_poll + 640
pjsip_endpt_handle_events2 + 92
pjsua_handle_events + 72
worker_thread + 44
thread_main + 188

Thanks to Yeuda for the report.

[Copilot]

Pull Request Overview

This PR fixes a deadlock that occurs when registration retry and IP change handling are triggered simultaneously due to inconsistent lock ordering between the two code paths.

Key Changes:

• Adds PJSUA lock acquisition before calling pjsip_tpmgr_shutdown_all() in IP change handling
• Ensures consistent lock ordering: PJSUA lock → transport manager lock