resource-cleanup responsibilities clearly between the 2 channels involved in the bridge
- Each channel is responsible for clearning its own peer_data and event queue
at the end of the call (when moving to DOWN state)
- Each channel dequeues messages only from its own queue and enqueues messages
in the peer's queue, with the only exception being messages received before
the bridge is stablished (IAM for sure and possible SAM messages) because
if the bridge is not yet stablished the messages must be queued by the channel
in its own queue temporarily until the bridge is ready
- When the bridge is ready it is the responsibility of the incoming channel to
move the messages that stored temporarily in its own queue to the bridged peer queue
- During hangup, each channel is responsible for moving itself to DOWN. The procedure
however differs slightly depending on the hangup conditions
If the user requests hangup (ie, FreeSWITCH) the request will be noted by setting the
FTDM_CHANNEL_USER_HANGUP flag but will not be processed yet because call control is
driven only by the link messages (so no hangup from ESL or command line allowed)
When REL message comes, the channel receiving it must move to TERMINATING state and:
- If the user has not hangup yet (FTDM_CHANNEL_USER_HANGUP flag not set) then
notify the user via SIGEVENT_STOP and wait for the user to move to HANGUP
state by calling ftdm_channel_call_hangup() before sending RLC
- If the user did hangup already (FTDM_CHANNEL_USER_HANGUP flag is set) then
skip user notification and move to HANGUP state directly where the RLC message
will be sent
- On HANGUP state the RLC is sent and the channel is moved to DOWN, final state
The peer channel will forward the REL message and wait for RLC from the network, when
RLC is received the channel can move straight to DOWN itself because the peer channel
is completing its own shutdown procedure when it received the REL message
Thanks to Phil Zimmermann for the code and for the license exception
we needed to include it.
There remains some build system integration work to be done before
this code will build properly in the FreeSWITCH tree.
This fixes an issue where ss7 native bridge was accidentally enabled
any time two freetdm channels were bridged regardless of the freetdm_native_sigbridge
variable value.
Use the amount of audio data received in misdn_read() to determine how many
bytes we need to send to the b-channel (= how much free space is left
in the b-channel tx queue). (This is how libosmo-abis and LCR handle it too.)
A pipe is used as a poll()-able audio tx buffer (filled in misdn_write()):
FTDM_WRITE wait requests are currently poll()-ed on the input side of the pipe,
whereas FTDM_READ and _EVENT requests are poll()-ed on the b-channel socket itself.
For every N-bytes of audio data read from the b-channel in misdn_read(),
we try to get as much out of the tx pipe, convert it into the ISDN_P_B_RAW
format and send it to the b-channel socket.
If there's less than N-bytes left in the pipe, we fill the remaining buffer
with silence to avoid buffer underflows.
B-Channel handling overview:
- misdn_wait(FTDM_WRITE) on audio pipe
- misdn_write() put audio data into pipe
- misdn_wait(FTDM_READ) for next incoming mISDN
message on b-channel socket
- misdn_read() handle mISDN event, for PH_DATA_IND:
- Write data into channel buffer and convert
to a/u-law using misdn_convert_audio_bits()
- Try to fetch N-bytes from audio pipe
- If not enough bytes in pipe: fill remaining space with silence
- Convert audio to raw format
- Send to b-channel (PH_DATA_REQ)
Known problems / bugs / further investigation:
1. Bridge aborted by "Write Buffer 0 bytes Failed!" error from switch_core_io.c.
This is "fixed" by _not_ setting the b-channel sockfd to non-blocking mode.
2. Audio glitches (maybe caused by FTDM_WRITE misdn_wait() handling or blocking I/O on sockfd?)
3. misdn_read() EBUSY error messages from sending data to b-channel sockfd after enabling channel.
Signed-off-by: Stefan Knoblich <stkn@openisdn.net>
The former fixes a strange "bug" with hfcsusb, where a b-channel deactivation
on a inactive channel (caused by a reset cycle) would cause the port to
lock up and stop processing events.
NOTE: this still needs to be investigated further, but this workaround will
at least prevent it from breaking completely.
We'll now keep track of the channel activation state and not send any
PH_ACTIVATE_REQ / PH_DEACTIVATE_REQ requests, if the channel already has the
desired state.
Signed-off-by: Stefan Knoblich <stkn@openisdn.net>
Use POLLIN on the socket instead, the b-channel should be able
to write when there is something to read.
Several other projects handle it this way, e.g. libosmo-abis.
Signed-off-by: Stefan Knoblich <stkn@openisdn.net>
The code was improperly using peer_data as an indicator that the sigbridge
ss7 mode was enabled. The channel flag FTDM_CHANNEL_NATIVE_SIGBRDIGE should
be used instead
- The outgoing tdm leg should not move to UP until
after the IAM is sent at the end of the function
- The UP state should be processed immediately otherwise
the state processor is not run due to the way the main
ss7 processing loop currently works
Use FTDM_SIZE_FMT where needed, don't treat ftdm_event_t as an int
(even if the e_type enum is the first member), datalen vs. *datalen fix
and other warnings.
All reported by __check_printf() (GCC + __attribute__((format(printf,x,y))) ).
Signed-off-by: Stefan Knoblich <stkn@openisdn.net>
... that could cause segmentation faults.
Caught while working on __check_printf() support for ftdm_log().
Signed-off-by: Stefan Knoblich <stkn@openisdn.net>
For compilers that seem to do the wrong thing(tm).
Speculative fix for:
segfault at 1 ip b72145d3 sp b58f8bfc error 4 in libc-2.11.3.so
#0 0xb7a5d5d3 in vfprintf () from /lib/i686/cmov/libc.so.6
#1 0xb7a7cec7 in vasprintf () from /lib/i686/cmov/libc.so.6
#2 0xb7dd7c5b in switch_vasprintf (...)
#3 0xb6296de2 in ftdm_logger (...)
#4 0xb621625d in misdn_handle_mph_information_ind (...) at ftmod_misdn.c:658
Signed-off-by: Stefan Knoblich <stkn@openisdn.net>
Only two mISDN hardware drivers emit MPH_INFORMATION_IND messages and both use a different payload:
- hfcsusb (HFC-based USB dongle) sends a set of ph_info + ph_info_ch structures
which contain the complete state information of the port
(including internal hw-specific state and flags).
- hfcmulti which sends a single integer, a single L1_SIGNAL_* event.
We now try to guess the type of message from the payload length.
The hfcmulti signals are converted to FreeTDM alarm flags; the hfcsusb
state/flags are defined in kernel internal hw-specific headers and are ignored ATM (todo).
Signed-off-by: Stefan Knoblich <stkn@openisdn.net>
Add MISDN_MSG_DATA() helper macro for easy access to mISDN message
payload.
Add forward declaration of misdn_handle_mph_information_ind() and use
it in misdn_activate_channel().
Signed-off-by: Stefan Knoblich <stkn@openisdn.net>
Add missing mISDN event/message types (e.g. MPH_INFORMATION_IND)
and use a helper macro (MISDN_EVENT_TYPE) to define the entries,
like we already do for misdn_control_types[].
Signed-off-by: Stefan Knoblich <stkn@openisdn.net>
FS-3815 --resolve
This is a workaround for the fact that libedit counts terminal control
characters when calculating the length of the prompt. By not using
absolute positioning, we avoid the issue.
Thanks to Ivan Isaev for the workaround and testing.
- The queue size has been bumped again, long calls could potentially require more elements (multiple resume/suspend)
- The queue is only used when there is a call active
- redirect number in Transparent IAM
- redirect information in Transparent IAM
- called party number in Transparent IAM
- adding incoming uuid to x-header to check loop calls
cc1: warnings being treated as errors
libs/esl/src/esl.c: In function "esl_recv_event":
libs/esl/src/esl.c:1190: error: array subscript is above array bounds
libs/esl/src/esl.c:1227: error: array subscript is above array bounds
Clamp handle_recv() return value to safe values.
Signed-off-by: Stefan Knoblich <stkn@openisdn.net>
On SIG Down we must not fail a call instead try hunting for another.
The only time we can fail the call and not bother hunting is
if sng_cc_resource_check fails.
Took out configuration retry as the config code is now
fixed in sng_ss7 library. Transaction id fix.
Unit Tested:
NSG UP -- start full load
kill NSG
NSG UP again on full load
make sure it comes up fine.
before hanging up the freetdm channel by force
seems to have a memory leak. I have increased the timeout
to 30sec and made the print statement WARNING level.
Where FS does not hangup the channel with in 3sec.
Without this fix, it was possible that 2 FS sessions
use the single span/chan freetdm session.
Merged the feature from git master of freetdm
SUSPEND will not only change the return state in few
cases.
I have retested all test cases from previous 2 commits.
What is missing is timer for each Tx Event. RSC,BLO,UBL.
Its possible to loose a message from Slave via relay to Master.
If that ever occours we will not know what state we are in.
Thus we need a timer that will re-transmit a packet if
it does not get acknowledged.
changed parameter name from cpu_reset_alarm_threshold to cpu_clear_alarm_threshold
(old parameter name still valid for backwards compatibility in configuration files)
James Zhang
David Yat Sin
Anthony Minessale
Travis Cross
Steve Underwood
Stefan Knoblich
Ken Rice
Moises Silva
Jeff Lenk
Christopher Rienzo
Brian West
Michael Jerris
Marc Olivier Chouinard
William Adam
Nenad Corbic