A periodic CLCC polling is started when there is an ongoing multiparty
call and a new call appears in the system. A simple way to reproduce
the crashing scenario is:
1. Place a call.
2. Place a second call.
3. Create a multiparty call with both calls.
4. Place a third call (incoming or outgoing does not matter).
5. Disconnect HFP from the modem.
Within the function ciev_callheld_notify, the AT+CLCC command is also
invoked, thus a new cyclic CLCC polling is started, and it overwrites
the timer resource identifier stored in voicecall_data.clcc_source.
This means that there are several timers doing the CLCC polling, but
only one of those is under control, i.e. it can be removed through its
source identifier, hence a timer source leak.
This has a fatal consequence when the HFP modem is disconnected. The
function hfp_voicecall_remove stops the timer that is under control
before freeing the voicecall_data struct. However there are other timers
that are still active and will execute its handler poll_clcc afterwards.
Inside poll_clcc the driver_data is accessed, which is already NULL.
A solution for this is to avoid starting a CLCC polling if there is
already one active, i.e. clcc_source is not 0. By doing this the
uncontrolled timers will not cycle forever.
According to the standard "3GPP 27.007 v6.8.0" Appendix C.2.11,
when sending multiple DTMF characters, these must go in individual
+VTS commands for each tone. This adopts the AT modem approach.
Before: AT+VTS=1234\r
After: AT+VTS=1;+VTS=2;+VTS=3;+VTS=4\r
The affected call types for +CHUP were set to only ACTIVE calls.
Instead the affected set should include INCOMING, DIALING, ALERTING and
ACTIVE calls.
Thanks to Ionut Dediu for the diagnosing and reporting this issue.
Before, the AT+BAC command was being sent with fixed information,
now we send the command (that inform the AG of the codecs supported by
the HF) with the codecs supported by the registered Handsfree Audio
Agent.
This patch fixes segmentation fault when the network registration
watch is called without being initialized. CIEV GAtChat callback can
be called before ofono_netreg_register().
==15101== Invalid read of size 8
==15101== at 0x492B56: ofono_netreg_register (network.c:2073)
==15101== by 0x47245E: hfp_netreg_probe (network-registration.c:311)
==15101== by 0x492A8D: ofono_netreg_create (network.c:1881)
==15101== by 0x4849D5: hfp_pre_sim (hfp_hf_bluez5.c:288)
==15101== by 0x48C486: ofono_modem_set_powered (modem.c:1194)
==15101== by 0x484E9D: slc_established (hfp_hf_bluez5.c:85)
==15101== by 0x4702AD: chld_cb (slc.c:147)
==15101== by 0x440457: at_chat_finish_command (gatchat.c:461)
==15101== by 0x44109F: new_bytes (gatchat.c:532)
==15101== by 0x4433B7: received_data (gatio.c:122)
==15101== by 0x3CBAA47824: g_main_context_dispatch (gmain.c:2539)
==15101== by 0x3CBAA47B57: g_main_context_iterate.isra.23
(gmain.c:3146)
==15101== Address 0x18 is not stack'd, malloc'd or (recently) free'd
Some phones do not send the corresponding call state update (+CIEV)
after a successful release-and-swap operation (AT+CHLD=1).
This has been observed with a Nokia 500, while testing ReleaseAndSwap()
while an active and a held call exist:
ofonod[20414]: > AT+CLCC\r
ofonod[20414]: < \r\n+CLCC: 1,0,1,0,0,"<number1>",145\r\n
ofonod[20414]: < \r\n+CLCC: 2,0,0,0,0,"<number2>",145\r\n
ofonod[20414]: < \r\nOK\r\n
ofonod[20414]: > AT+CHLD=1\r
ofonod[20414]: < \r\nOK\r\n
After this, no +CIEV is received, but the call has been hung up.
The proposed approach to solve this consists of using AT+CLCC, unless
a call release has been received within a specific time period.
The result fixes the problem as can be seen below:
ofonod[20847]: < \r\n+CLCC: 1,0,1,0,0,"<number1>",145\r\n
ofonod[20847]: < \r\n+CLCC: 2,0,0,0,0,"<number2>",145\r\n
ofonod[20847]: < \r\nOK\r\n
ofonod[20847]: > AT+CHLD=1\r
ofonod[20847]: < \r\nOK\r\n
ofonod[20847]: > AT+CLCC\r
ofonod[20847]: < \r\n+CLCC: 1,0,0,0,0,"<number1>",145\r\n
ofonod[20847]: < \r\nOK\r\n
ofonod[20847]: < \r\n+CIEV: 5,2\r\n
ofonod[20847]: < \r\n+CIEV: 5,0\r\n
While processing the result of AT+CLCC, process the differences in a way
that disconnections are reported first, then call state changes and
finally new calls.
The goal is to avoid unnecessary transitional states such as two active
calls existing at the same time.
Right now, only the mandatory CVSD codec is supported. The mSBC
mandatory codec is "temporarily" not supported.
The spec alows this, HFP 1.6 Spec Section 4.34.1 page 92: "If wide band
speech is supported then the mandatory codec (mSBC) shall be included
unless it is temporarily unavailable."
A call that moves from the dialing to active state before the +CLCC response
will not properly be added as a voicecall. This is because the dialing callback
was using simplified handling and only looked for calls in the dialing or
alerting state.
AT sequence that exhibited the failure (AG device was an iPhone accessing
visual voicemail):
> +CIND: ("service",(0-1)),("call",(0-1)),("callsetup",(0-3)),
("battchg",(0-5)),("signal",(0-5)),("roam",(0-1)),("callheld",(0-2))
...
> +CIEV: 3,2
< AT+CLCC
> +CIEV: 2,1
> +CIEV: 3,0
> +CLCC: 1,0,0,0,0,"**21153**",129,"Voicemail"