Fixed how Asterisk destroys a dialog on channel hangup before invite receives a response.
If an ast_channel with a SIP tech pvt hangs up before the sip dialog gets a response
to its outgoing INVITE, Asterisk used to pretend_ack the INVITE. This is not rfc
compliant and results in confusion at the other endpoint. sip_pretend_ack will ack
and remove all the packets in the retransmit queue. This means that the INVITE will
stop retransmitting, and that any response to that INVITE that comes after the pretend_ack
occurs will be ignored.
Instead of faking any sort of acknowledgement for an outgoing INVITE during an internal
hangup, we should let the protocol stack process the INVITE transaction and terminate
the dialog properly. This is achieved by setting the PENDING_BYE flag. When this flag
is used, once the dialog proceeds to an escapable state the transaction will either be
canceled with a SIP_CANCEL or completed followed immediately by a BYE. Attempting to do
this any other way is incorrect. If the endpoint is not responding to the INVITE request,
the INVITE must continue to be retransmitted until it times out which will result in the
dialog being destroyed.
........
David Vossel [Wed, 25 Aug 2010 22:56:42 +0000 (22:56 +0000)]
Add to and from tags to NOTIFY dialog-info xml body so pickup can occur.
When pedantic mode is used, the dialog-info xml generated during a
ringing event must contain the to and from tag values. Otherwise if
a pickup occurs using INVITE with replaces, Astrisk will not be able
to locate the subscription.
David Vossel [Wed, 25 Aug 2010 15:52:54 +0000 (15:52 +0000)]
Asterisk will not advertise session timers are supported when 'session-timers=refuse' is used.
Asterisk now dynamically builds the "Supported" header depending
on what is enabled/disabled in sip.conf. Session timers used
to always be advertised as being supported even when they were disabled
in the configuration. This caused problems with some end points.
This fix makes sure the ast_channel hangs up correctly when the dialog's PENDING_BYE flag is set.
When the pending bye flag is used, it is possible that the dialog will terminate
and leave the sip_pvt->owner channel up. This is because we never hangup the
ast_channel after sending the SIP_BYE request. When we receive the response for
the SIP_BYE we set need_destroy which we would expect to destroy the dialog on the
next do_monitor loop, but this is not the case. The dialog will only be destroyed
once the owner is hungup even with the need_destroy flag set. This patch sets the
softhangup flag on the ast_channel when a SIP_BYE request is sent as a result of the
pending bye flag.
........
Q931 - Sending PROGRESS after sending ALERTING is a protocol error
The PRI layer in chan_dadhi will check if a PROGRESS message has already
been sent, and not allow sending another (although that is technically
allowed by the Q931 spec), however it does not protect against sending an
ALERTING and then sending a PROGRESS message, which is a violation of the
specification.
Most switches don't seem to care too deeply about this, but some do, and
will disconnect the call when receiving this invalid sequence.
Protocol specification reference: T-REC-Q.931-199805-I page 223, "Figure
A.5/Q.931 -- Overview protocol control (network side) point-point
(sheet 3 of 8)"
(closes issue #17874)
Reported by: nic_bellamy
Patches:
asterisk-1.4-r282537_no-progress-after-alerting.patch uploaded by nic bellamy (license 299)
asterisk-1.6.2-r282537_no-progress-after-alerting.patch uploaded by nic bellamy (license 299)
asterisk-trunk-r282537_no-progress-after-alerting.patch uploaded by nic bellamy (license 299)
........
When tos_sip is used, the tos of the sip socket is only set
correctly if the socket binding changes on a reload. If the binding
stays the same but the TOS changes, the new tos value would not take
into effect. This patch fixes that.
Leif Madsen [Mon, 16 Aug 2010 18:00:09 +0000 (18:00 +0000)]
Add information about creating sounds files using
the sounds tools publically available so that others can create their
own sounds prompts using the same tools we use to generate sounds releases.
This allows people creating their own prompts to sound consistent with
the prompts available from the open source project.
Masquerading a channel means that the src of the audio is potentially
changing, so send a SRCCHANGE so that RTP-based media streams can get
a new SSRC generated to reflect the change. Original patch by addix
(along with lots of testing--thanks!).
David Vossel [Fri, 13 Aug 2010 18:54:53 +0000 (18:54 +0000)]
only do magic pickup when notifycid is enabled
A new way of doing BLF pickup was introduced into 1.6.2. This feature
adds a call-id value into the XML of a SIP_NOTIFY message sent to alert
a subscriber that a device is ringing. This option should only be enabled
when the new 'notifycid' option is set... but this was not the case. Instead
the call-id value was included for every RINGING Notify message, which
caused a regression for people who used other methods for call pickup.
Ensure SSRC is changed when media source is changed to resolve audio delay.
This change causes the SSRC to change right before the channels are bridged,
which is what used to happen. It seems that fixes were made to attempt limiting
SSRC changes, targeted mainly at sending DTMF. DTMF is not affecting the SSRC
with this change.
There are two other control frames sent in ast_channel_bridge that probably
should also be changed to AST_CONTROL_SRCCHANGE as well, but I'm going to leave
this change up to the discretion of resolving issue #17007.
For reference - old review implementing new control frame SRCCHANGE:
https://reviewboard.asterisk.org/r/540
David Vossel [Mon, 9 Aug 2010 20:46:50 +0000 (20:46 +0000)]
fixes SIP peers memory leak
We zeroed out the peer's addr before it was removed from the
peers_by_ip container. This made it impossible to be removed
from the container as the addr is the key used by the container
to find the peer.
(closes issue #17774)
Reported by: kkm
Patches:
017774-sip-peer-leak-1.6.2.10.diff uploaded by kkm (license 888)
017774-sip-peer-leak-1.8.diff uploaded by kkm (license 888)
Prevent loss of Caller ID information set on local channel after masquerade.
Caller ID set on the channel before a masquerade occurs when using a local
channel would cause the information to be lost. The problem was that the
information was set on a channel destined to be hung up. The somewhat confusing
fix is to detect if any Caller ID has been set on the channel and if so
preswap the Caller ID data so that basically the masquerade puts the data back.
There is a scheduler item in chan_sip that keeps sending the
last provisional message in response to an INVITE Request for
a period of time until a final response to that INVITE is
sent. Because of the way this scheduler item works, it requires
a reference to a sip_pvt pointer to work properly. The problem
with this is that it is currently possible (but rare) for the
sip_pvt to get destroyed and that scheduler item to still
exist. When this occurs, the scheduler event fires and attempts
to access a freed sip_pvt which causes a crash.
Jeff Peeler [Mon, 2 Aug 2010 21:14:20 +0000 (21:14 +0000)]
Change SIP NOTIFY requests to expect a response so authentication will work.
This changes the request to be sent with the transmit type XMIT_RELIABLE so that
sip_ack doesn't return false and cause the 401 to be ignored in cases where
authentication is required.
If a channel has an audiohook write list created on it, that
list stays on the channel until the channel is destroyed. There
is no reason to keep that list on the channel if it becomes empty.
If it is empty that just means we are doing needless translating
for every ast_read and ast_write. This patch removes the audiohook
list from the channel once it is detected to be empty on either a
read or write. If a audiohook is added back to the channel after
this list is destroyed, the list just gets recreated as if it never
existed to begin with.
Mark Michelson [Tue, 27 Jul 2010 15:13:24 +0000 (15:13 +0000)]
Fix bad behavior of dynamic_exclude_static option in sip.conf.
We were attempting to create a contactdeny rule based on the peer's
IP address before the peer's IP address had been set. By moving the
processing further down in the function, we can ensure stuff works
as we expect for it to.
Provide a default value for DAHDI_TRANSCODE so when DAHDI is not installed
menuselect doesn't get confused:
Unknown value '' found in build_tools/menuselect-deps for DAHDI_TRANSCODE
........
SIP promiscuous redirect could fail to dial the redirect.
The ast_channel was created with one variable to ast_request() but the
call to ast_call() that initiates the outgoing call was using a different
variable. The two variables are not equivalent if the call_forward string
included a channel technology specifier. e.g., SIP/200
........
Add method for finding XML doc files for systems that don't support GLOB_BRACE.
In particular, Solaris and perhaps others do not support the above mentioned
GNU extension. In this case the paths are simply expanded without the braces
and the calls to glob are made separately.
Note: I could not explain memory allocation failures that were being reported
from within libxml itself when making calls to glob without using GLOB_NOCHECK.
This is the only reason why that flag is being used.
(closes issue #15402)
Reported by: snuffy
Patches:
bug_xmlpatt-v3.diff uploaded by snuffy (license 35),
modified by me
........
Richard Mudgett [Thu, 22 Jul 2010 19:32:16 +0000 (19:32 +0000)]
DNID does not get cleard on a new call when using immediate=yes with ISDN signaling.
When you are using chan_dahdi ISDN signaling with immediate=yes and a call
comes in without a DNID then you get the DNID of a previous call.
Chan_dahdi does not touch the DNID field on a new call if it does not have
a DNID.
Made always copy the DNID from the new call.
The patches backport the relevant changes from trunk -r210387.
(closes issue #17568)
Reported by: wuwu
Patches:
issue17568_v1.4.patch uploaded by rmudgett (license 664)
issue17568_v1.6.2.patch uploaded by rmudgett (license 664)
Allow PLC to function properly when channels use SLIN for audio.
If a channel involved in a bridge was using SLIN audio, then translation
paths were not guaranteed to be set up properly since in all likelihood
the number of translation steps was only 1.
This patch enforces the transcode_via_slin behavior if transcode_via_slin
or generic_plc is enabled and one of the formats to make compatible is
SLIN.
This fixes some cases of no outgoing calls on FXO before an incoming call.
Remove an unnecessary testing of an "off-hook" bit from DAHDI for FXO
(KS/GS) channels.In some cases the bit would not be initialized properly
before the first inbound call and thus prevent an outgoing call.
If those tests are actually required by anybody, they should define
DAHDI_CHECK_HOOKSTATE in channels/sig_analog.c .
(closes issue #14577)
Reported by: jkroon
Patches:
asterisk_chan_dahdi_hookstate_fix.diff uploaded by frawd (license 610)
Tested by: frawd
Use poll() instead of select() in res_timing_pthread to avoid stack corruption.
This code did not properly check FD_SETSIZE to ensure that it did not try to
select() on fds that were too large. Switching to poll() removes the limitation
on the maximum fd value.
Avoid trying to pickup a parked extension before the park operation is completed.
A crash could occur if the extension is picked up while the parking extension is
being announced. Testing pu->notquiteyet while searching for a parked extension
resolves this crash.
Save and restore AST_FLAG_BRIDGE_HANGUP_DONT on attended transfer.
ast_bridge_call() clears AST_FLAG_BRIDGE_HANGUP_DONT. But during an attended
transfer, ast_bridge_call() is called for a second bridge on the same channel,
and it clears that flag, which still needs to get set for when the original
ast_bridge_call() gets control back and checks it.
Default to no udptl error correction so that error correction will be disabled in the event that the remote end indicates that they do not support the error correction mode we requested.
priexclusive in chan_dahdi.conf ignored when reloading dahdi module
During a reload, the priexclusive and outsignalling parameters are not
read in from the config file as intended. Unfortunately, they get set to
defaults as a result. This patch makes sure that they do not get set to
defaults during a reload.
(closes issue #17441)
Reported by: mtryfoss
Patches:
issue17441_v1.4.patch uploaded by rmudgett (license 664)
issue17441_v1.6.2.patch uploaded by rmudgett (license 664)
issue17441_trunk.patch uploaded by rmudgett (license 664)
Tested by: rmudgett
........
................
For pass through DTMF tones, measure the actual duration between the begin and end packets on the wire. If it is detected to be less than AST_MIN_DTMF_DURATION, trigger dtmf emulation.
Correct not setting the bindport before attempting to open the socket.
Related to changes from 276571, I was accidentally testing with a port set in
my configuration causing me to miss this. Also moved the TCP handling as well
to occur before build_peer is called.
........
Fix MWI notification transmission problems over SIP.
MWI updates were not being sent if no messages were found in the event cache.
This was corrected since a phone may need to clear its MWI status configured
previously from another mailbox.
Upon module or sip reload, MWI updates could not be sent due to the sipsock
socket not being set early enough in reload_config. The code handling the
descriptor assignment and such has simply been moved before the call to
build_peer.
Issuing a sip reload cleared the IP address of the peer, but skipped checking
the database for registration information. The database is now checked both
for sip reload and actually reloading the module.
If a transmission occurs before the do_monitor thread has started, do not
attempt to send a signal to it.
Make user removals and traversals thread safe in meetme.
Race conditions present in meetme involving the user list where a lack of
locking has the potential for a user to be removed during a traversal or as in
the case of the reporter after checking if the list is empty could cause a
crash. Fixing this was done by convering the userlist to an ao2 container.
Access peer->cdr directly instead of through a saved off reference.
At this point in the code, it is possible that peer_cdr may be invalid.
Specifically, in the blind transfer code, CDRs are swapped between channels.
So, peer_cdr is no longer == peer->cdr.
The scenario that exposed a crash in this code was a blind transfer that hit
the system call limit, causing the transferee channel to get destroyed after
the transfer attempt failed. Even if it succeeds and this code doesn't crash,
this code was still trying to reset a CDR on a channel that was now owned by
a different thread, which is a BadThing(tm).
Change ast_write to not stop generator when called from ast_prod.
For SIP channels configured with the progressinband option on, the ringback was
being immediately stopped. This problem was due to ast_prod being moved for a
deadlock fix in 259858. Prodding the channel after setting up the generator
triggered the check in ast_write to stop the generator. The fix here should
write the frame the same as was done before the call to ast_prod was moved.
This change adds an ERROR message to let you know when a failure exists to
get the columns from the pgsql database, which typically means that the
table does not exist.
Remove useless sip options related to hash table size.
First off, these options weren't actually doing anything.
By the time the options were parsed, the peer and dialog
containers had already been allocated with their default
values.
Second, hash table size is something that doesn't really
make sense to change in a config file. If a user is that
interested in changing the hashtable size, he can modify
the source itself.
I have removed the parsing of the hash_peer, hash_user,
and hash_dialog options. I have removed the hash_user_size
variable altogether since it is not used at all. I also
changed hash_peer_size and hash_dialog_size to be constant,
and have changed the symbols to be in all caps as constants
typically are. I have also removed the entire section in
sip.conf.sample regarding configurable hashtable sizes.
........
projects / thirdparty / asterisk.git / log
