15 May
2023
15 May
'23
8:13 a.m.
Hello,
While troubleshooting a memory leak, I noticed that we are mixing SHM
memory with PKG memory during hep tracing.
When the dispatcher timer is scheduled for OPTION pings, the
ds_ping_set() is called and the tm module is engaged to handle the
ping.
dispatcher: dispatcher.c
ds_ping_set()
tmb.t_request
Inside the tm module, a new cell is created and the SIP message is
cloned for further processing via sip_msg_cloner() and later on the
onsend_route is engaged:
tm: uac.c:
request()
t_uac()
t_uac_with_ids()
t_uac_prepare()
new_cell->uas.request = sip_msg_cloner(&lreq, &sip_msg_len);
core: sip_msg_shm_clone() (we are cloning the msg in SHM)
send_prepared_request_impl(SHM pointer)
run_onsend(SHM pointer)
Inside the onsend_route we call siptrace() and we go like this:
siptrace: siptrace_send.c
sip_trace_prepare(SHM pointer)
parser:
parse_from_header(SHM pointer)
we allocate PKG mem that is leaked
parse_to()
parse_addr_spec()
we allocate PKG mem that is leaked
Later on, the SHM memory block is released by the free_cell_helper():
sip_msg_free_unsafe(dead_cell->uas.request);
This was a misconfiguration, because the siptrace was configured to
automatically mirror all SIP traffic and the call for siptrace() in
the onsend_route was not necessary.
Perhaps siptrace() should be a noop when trace_mode is not 0.
Here are the siptrace params:
loadmodule "siptrace.so"
modparam("siptrace", "db_url", DBMYSQL)
modparam("siptrace", "duplicate_uri", SIPTRACE_SVR_IP)
modparam("siptrace", "send_sock_addr", SIPTRACE_OUT_IP)
modparam("siptrace", "hep_mode_on", 1)
modparam("siptrace", "hep_version", 3)
modparam("siptrace", "trace_to_database", 0)
modparam("siptrace", "trace_flag", 22)
modparam("siptrace", "trace_on", 1)
modparam("siptrace", "evcb_msg", "ksr_siptrace_msg")
modparam("siptrace", "trace_mode", 1)
This is happening on the latest version of kamailio 5.6.
-ovidiu