Wenpeng Song authored 1 year ago and Grzegorz Sluja committed 10 months ago

There is a remaining channel when ref>2 or Asterisk may crash when ref<2 in ast_hangup().

Wenpeng Song authored 1 year ago and Grzegorz Sluja committed 10 months ago

Crash due to a race condition between this ref correction
https://dev.iopsys.eu/voice/asterisk/-/blob/devel/channels/chan_pjsip.c#L2088-2095
and the unref under chan_voicemngr_attended_call_transfer
https://dev.iopsys.eu/voice/asterisk-chan-voicemngr/-/blob/devel/src/channels/chan_voicemngr.c?ref_type=heads#L5646-5651.
When the correction happened and finished before the unref during the transfer, which leading
the result of ref=1 after the unref, then the crash happened during hangup the channel, add an
extra correction for ref<2 could resolve this issue.

Yalu Zhang authored 1 year ago and Grzegorz Sluja committed 10 months ago

Asterisk can crash if two FXS ports on the same device act as transferor and
transfer target when asterisk.tel_options.hold_target_before_refer='yes',
which is the default config. The issues happens after a couple of calls.

The solution is to revert d38e42d8 about
reference count decreasing.
Some validation check is also added to chan_pjsip_hangup().

TODO: After this commit, there will be issues if
asterisk.tel_options.hold_target_before_refer='no'.
- Remaining pjsip channels after a couple of attended transfers in the above
  scenario
- Transferor might not be able to call transfer target because the maximum
  sessions per line exceed the specified number
These issues will be solved in a new commit.

George Yang authored 1 year ago and Yalu Zhang committed 1 year ago

The pjsip channel was not released after call transfer with setting hold_target_before_refer='no'
before the fix.

This is to avoid memory corruption issue that happens when more than one thread calls ubus_free() which frees the global buffer which is defined in libubus.

Wenpeng Song authored 1 year ago and Yalu Zhang committed 1 year ago

When SIP un-registration request got an 200 OK with expiration value which is not 0 (due to the
server setting and network), it is treated as registration success based on the non zero expiration value. This made the other procedures following and it went into an uncontrollable
stage, i.e. unstoppable loop of the un-registration requests until crash.

Solution
Retrieve the requested expiration if possible, and check it together with the received expiration
in response to identify if the 200 OK is to register or unregister.

This "merge" merges the proper history of Asterisk 20.3.0 by reverting
an accidentally squashed merge and doing a proper merge commit after.
Thus, there is no change to the actual code.

Grzegorz Sluja authored 1 year ago and Andreas Gnau committed 1 year ago

This commit is identical to commit 3c8116c8 (Upgrading asterisk to
the version 20.3.0, 2023-10-03) with the difference that it retains
proper history and attribution.

Signed-off-by: Andreas Gnau <andreas.gnau@iopsys.eu>

This reverts commit 3c8116c8. It has
been a merge that has been acidentally squashed into one commit with the
consequence that all history has been lost. The next commit will be a
proper merge commit rectifying this.

Grzegorz Sluja authored 1 year ago and Yalu Zhang committed 1 year ago

There was no audio for 3-way conference when sRTP is used.
For 2-way calls frame->seqno is taken from DSP and is used by asterisk for the sequence number
in RTP headers. However for 3-way conference the sequence number is generated by asterisk and
it has to be greater than the previous value, otherwise libsrtp refuses to forward 'too old'
RTP packets.

George Joseph authored 1 year ago and Wenpeng Song committed 1 year ago

Commit f66f77fa last year prevents the res_pjsip_exten_state and
res_pjsip_mwi modules from unloading due to possible pjproject
asserts if the modules are reloaded. A side effect of the
implementation is that the taskprocessors these modules use aren't
being released. When asterisk is doing a graceful shutdown, it
waits AST_TASKPROCESSOR_SHUTDOWN_MAX_WAIT seconds for all
taskprocessors to stop but since those 2 modules don't release
theirs, the shutdown hangs for that amount of time.

This change allows the modules to be unloaded and their resources to
be released when ast_shutdown_final is true.

Resolves: #379

Wenpeng Song authored 1 year ago and Yalu Zhang committed 1 year ago

This hook function is supposed to return PJ_SUCCESS in order to let a SIP request be
transmitted.

This reverts commit 0f97e870.

Wenpeng Song authored 1 year ago and Yalu Zhang committed 1 year ago

The purpose is to avoid crash during closing.
Also enhance the pointer validation check in sipaddress_on_outgoing_inv_request().

Lukasz Kotasa authored 1 year ago and Yalu Zhang committed 1 year ago

The ast_sip_dialog_get_session function returns the session with reference count
increased. This was not taken into account and was causing sessions to remain
around when they should not be.

Wenpeng Song authored 1 year ago and Yalu Zhang committed 1 year ago

Session invitation is not ready (NULL) on "session_on_tx_request" (before sending the request)
when the function for setting SIPIPAdress is called in the current version.
Call the function under "session_inv_on_tsx_state_changed" (just after the request be sent) 
instead.

Grzegorz Sluja authored 1 year ago and Yalu Zhang committed 1 year ago

Our commits to the previous version have been rebased.

In ast_sip_session_terminate(), reset both active and pending media to prevent
non-freed resources after a call. Otherwise RTP/RTCP ports won't be closed after
the call in some call scenarios, e.g. a non-answered incoming call. In that case,
there are no active media sessions before the swapping.

This fixes a segment fault caused by ast_channel_name(NULL) for internal CBBS.

Wenpeng Song authored 1 year ago and Yalu Zhang committed 1 year ago

- Disable the trans-coding steps from set_cap which has been only used for outgoing call
  negotiation. Ignore trans-coding path failure
- Add codec sync for early media with AST_CONTROL_PROGRES

Lukasz Kotasa authored 1 year ago and Yalu Zhang committed 1 year ago

Reduce the retry-timer from 60s to 1s.

The SUBSCRIBE request for MWI is sent in the following conditions:
* mwi is enabled in asterisk configuration: option mwi_enable '1'
* SUBSCRIBE is sent after registration succeeds with expiration set to the same value as REGISTER.