Russell Bryant [Sat, 1 Aug 2009 10:59:05 +0000 (10:59 +0000)]
Modify how Playtones() is used in Milliwatt() to resolve gain issue.
When Milliwatt() was changed internally to use Playtones() so that the proper
tone was used, it introduced a drop in gain in the output signal. So, use
the playtones API directly and specify a volume argument such that the output
matches the gain of the original Milliwatt() code.
(closes issue #15386)
Reported by: rue_mohr
Patches:
issue_15386.rev2.diff uploaded by russell (license 2)
Tested by: rue_mohr
Minor changes inspired by testing with latest GCC.
The latest GCC (what will become 4.5.x) has a few new warnings, that in these
cases found some either downright buggy code, or at least seriously poorly
designed code that could be improved.
Mark Michelson [Mon, 27 Jul 2009 17:44:06 +0000 (17:44 +0000)]
Allow for UDPTL to use only even-numbered ports if desired.
There are some VoIP providers out there that will not accept SDP
offers with odd numbered UDPTL ports. While it is my personal opinion
that these VoIP providers are misinterpreting RFC 2327, it really is
not a big deal to play along with their silly little games. Of course,
since restricting UDPTL ports to only even numbers reduces the range
of available ports by half, so the option to use only even port numbers
is off by default. A user can enable the behavior by setting
use_even_ports=yes in udptl.conf.
Jeff Peeler [Sat, 25 Jul 2009 06:19:50 +0000 (06:19 +0000)]
Fix compiling under dev-mode with gcc 4.4.0.
Mostly trivial changes, but I did not know of any other way to fix the
"dereferencing type-punned pointer will break strict-aliasing rules" error
without creating a tmp variable in chan_skinny.
Mark Michelson [Fri, 24 Jul 2009 19:24:28 +0000 (19:24 +0000)]
Don't impose an arbitrary limit on member lines in queues.conf
I know what some of you are thinking: "UGH! Mark, why are you using
ast_strdup and ast_free for the string when you can just use ast_strdupa
and let the memory free itself?! Have the bats been chewing on your brain
again?"
Based on past experiences, I don't like using ast_strdupa inside a loop.
It's a good way to potentially exhaust stack space. Also, since this only
happens when reloading queues, I don't think that heap allocations and
frees are going to be a huge problem.
Mark Michelson [Fri, 24 Jul 2009 18:26:50 +0000 (18:26 +0000)]
Only send a BYE when hanging up a channel that is up.
For cases where Asterisk sends an INVITE and receives a non 2XX final
response, Asterisk would follow the INVITE transaction by immediately
sending a BYE, which was unnecessary.
Mark Michelson [Thu, 23 Jul 2009 19:24:21 +0000 (19:24 +0000)]
Fix a problem where a 491 response could be sent out of dialog.
This generalizes the fix for issue 13849. The initial fix corrected the
problem that Asterisk would reply with a 491 if a reinvite were received
from an endpoint and we had not yet received an ACK from that endpoint
for the initial INVITE it had sent us. This expansion also allows Asterisk
to appropriately handle an INVITE with authorization credentials if Asterisk
had not received an ACK from the previous transaction in which Asterisk had
responded to an unauthorized INVITE with a 407.
Force an error if a blank is passed to QUOTE (because the documentation states the argument is not optional).
This change makes URIENCODE and QUOTE behave similarly, since the documentation
states that the argument is not optional, for both.
(closes issue #15439)
Reported by: pkempgen
Patches:
20090706__issue15439.diff.txt uploaded by tilghman (license 14)
Jeff Peeler [Tue, 21 Jul 2009 20:16:55 +0000 (20:16 +0000)]
Wait for wink before dialing when using E&M wink signaling
There was already code for other signaling types in dahdi_handle_event to
handle dialing if a dial operation dial string was present. Simply add
SIG_EMWINK to the list.
Kevin P. Fleming [Tue, 21 Jul 2009 13:04:44 +0000 (13:04 +0000)]
Ensure that user-provided CFLAGS and LDFLAGS are honored.
This commit changes the build system so that user-provided flags (in ASTCFLAGS
and ASTLDFLAGS) are supplied to the compiler/linker *after* all flags provided
by the build system itself, so that the user can effectively override the
build system's flags if desired. In addition, ASTCFLAGS and ASTLDFLAGS can now
be provided *either* in the environment before running 'make', or as variable
assignments on the 'make' command line. As a result, the use of COPTS and LDOPTS
is no longer necessary, so they are no longer documented, but are still supported
so as not to break existing build systems that supply them when building Asterisk.
Mark Michelson [Mon, 20 Jul 2009 19:39:59 +0000 (19:39 +0000)]
Answer video SDP offers properly when videosupport is not enabled.
Copied from Review board:
In issue 12434, the reporter describes a situation in which audio and video
is offered on the call, but because videosupport is disabled in sip.conf,
Asterisk gives no response at all to the video offer. According to RFC 3264,
all media offers should have a corresponding answer. For offers we do not
intend to actually reply to with meaningful values, we should still reply
with the port for the media stream set to 0.
In this patch, we take note of what types of media have been offered and
save the information on the sip_pvt. The SDP in the response will take into
account whether media was offered. If we are not otherwise going to answer
a media offer, we will insert an appropriate m= line with the port set to 0.
It is important to note that this patch is pretty much a bandage being
applied to a broken bone. The patch *only* helps for situations where video
is offered but videosupport is disabled and when udptl_pt is disabled but
T.38 is offered. Asterisk is not guaranteed to respond to every media offer.
Notable cases are when multiple streams of the same type are offered.
The 2 media stream limit is still present with this patch, too.
In trunk and the 1.6.X branches, things will be a bit different since Asterisk
also supports text in SDPs as well.
Russell Bryant [Mon, 20 Jul 2009 16:26:24 +0000 (16:26 +0000)]
Only do the chan->fdno check in ast_read() in a developer build.
I changed this check to only happen in a dev-mode build. I also added a
comment explaining what is going on. I also made it so that detection of
this situation does not affect ast_read() operation.
Jeff Peeler [Fri, 17 Jul 2009 19:13:27 +0000 (19:13 +0000)]
Enhance configuration option for overlapdial allowing direction choice
Previously overlap dialing could only be turned on or off for both incoming and
outgoing calls. New parameters incoming, outgoing, and both have been added to
allow further control. There is no change in default behavior with these new
options and allows in band DTMF to be accepted in one direction if required.
David Vossel [Fri, 17 Jul 2009 16:05:06 +0000 (16:05 +0000)]
SIP incorrect From: header information when callpres is prohib
Some ITSP make use of the "Anonymous" display name to detect a
requirement to withhold caller id across the PSTN. This does
not work if the display name is "Unknown".
(closes issue #14465)
Reported by: Nick_Lewis
Patches:
chan_sip.c-callerpres.patch uploaded by Nick (license 657)
chan_sip.c-callerpres_trunk.patch uploaded by dvossel (license 671)
Tested by: Nick_Lewis, dvossel
David Vossel [Thu, 16 Jul 2009 21:24:16 +0000 (21:24 +0000)]
avoid segfault caused by user error
If the CALLERPRES() dialplan function is set to nothing,
a segfault occurs. This is user error to begin with, but
I'd rather see a cli warning message than have Asterisk
crash on me.
Richard Mudgett [Wed, 15 Jul 2009 20:44:55 +0000 (20:44 +0000)]
Merged revision 206700 from
https://origsvn.digium.com/svn/asterisk/be/branches/C.2-...
..........
Fixed chan_misdn crash because mISDNuser library is not thread safe.
With Asterisk the mISDNuser library is driven by two threads concurrently:
1. channels/misdn/isdn_lib.c::manager_event_handler()
2. channels/misdn/isdn_lib.c::misdn_lib_isdn_event_catcher()
Calls into the library are done concurrently and recursively from
isdn_lib.c.
Both threads can fiddle with the master/child layer3_proc_t lists. One
thread may traverse the list when the other interrupts it and then removes
the list element which the first thread was currently handling. This is
exactly what caused the crash. About 60 calls were needed to a Gigaset
CX475 before it occurred once.
This patch adds locking when calling into the mISDNuser library.
This also fixes some cb_log calls with wrong port parameter.
JIRA ABE-1913
Patches: misdn-locking.patch (Modified with mostly cosmetic changes)
..........
Richard Mudgett [Tue, 14 Jul 2009 16:44:47 +0000 (16:44 +0000)]
Fixes several call transfer issues with chan_misdn.
* issue #14355 - Crash if attempt to transfer a call to an application.
Masquerade the other pair of the four asterisk channels involved in the
two calls. The held call already must be a bridged call (not an
applicaton) or it would have been rejected.
* issue #14692 - Held calls are not automatically cleared after transfer.
Allow the core to initate disconnect of held calls to the ISDN port. This
also fixes a similar case where the party on hold hangs up before being
transferred or taken off hold.
* JIRA ABE-1903 - Orphaned held calls left in music-on-hold.
Do not simply block passing the hangup event on held calls to asterisk
core.
* Fixed to allow held calls to be transferred to ringing calls.
Previously, held calls could only be transferred to connected calls.
* Eliminated unused call states to simplify hangup code.
* Eliminated most uses of "holded" because it is not a word.
Ensure apathetic replies are sent out on the proper socket.
chan_iax2 supports multiple address bindings. The send_apathetic_reply()
function did not attempt to send its response on the same socket that the
incoming message came in on.
........
Mark Michelson [Fri, 10 Jul 2009 17:39:13 +0000 (17:39 +0000)]
Properly ACK 487 responses to canceled INVITEs.
From the review board request:
The fix from review 298 has exposed a new bug in chan_sip.
When we hang up an outgoing call, we first will dump all the outstanding
packets on the sip_pvt using __sip_pretend_ack. Then, if we can, we send
a CANCEL. The problem with this is that since destroyed all the outstanding
packets on the dialog, we cannot match the incoming 487 response to our
INVITE. Because we cannot match the response, we do not send an ACK.
To correct this, instead of using __sip_pretend_ack, I have changed the code
to loop through the list of packets and call __sip_semi_ack on each one
instead. This causes us to stop retransmitting the requests, but we still have
them around in case we get responses for them.
When discussing this earlier today with Josh Colp, we both agreed that in the
majority of cases, this would be enough of a fix. However, we also agreed that
we should have a safety net in place in case we never receive a response to
our initial INVITE. To handle this, I have modified __sip_autodestruct to
behave similar to the way it does in Asterisk 1.4. If there are outstanding
packets on the sip_pvt, the needdestroy flag is not set, and the last request
on the dialog was either a CANCEL or BYE, then we set the needdestroy flag and
reschedule destruction for 10 seconds in the future. If, though, the
needdestroy flag is set, then we use __sip_pretend_ack to kill the remaining
outstanding packets so that the monitor thread can destroy the sip_pvt.
I ran two separate tests. First, I placed a call from my Aastra phone to my
Polycom phone. I hung up the Aastra before the Polycom answered. I verified
through sip debug output that Asterisk properly ACKed the 487 received from the
Polycom.
For my second test, I set up a SIPp UAS scenario so that it would send a 200 OK
in response to a CANCEL but would not send a 487 for the original INVITE. I
verified that after about 40 seconds, Asterisk properly cleans up the outgoing
sip_pvt for the call.
David Vossel [Fri, 10 Jul 2009 16:23:59 +0000 (16:23 +0000)]
SIP registration auth loop caused by stale nonce
If an endpoint sends two registration requests in a very short
period of time with the same nonce, both receive 401 responses
from Asterisk, each with a different nonce (the second 401
containing the current nonce and the first one being stale).
If the endpoint responds to the first 401, it does not match
the current nonce so Asterisk sends a third 401 with a newly
generated nonce (which updates the current nonce)... Now if
the endpoint responds to the second 401, it does not match the
current nonce either and Asterisk sends a fourth 401 with a
newly generated nonce... This loop goes on and on.
There appears to be a simple fix for this. If the nonce from
the request does not match our nonce, but is a good response
to a previous nonce, instead of sending a 401 with a newly
generated nonce, use the current one instead. This breaks
the loop as the nonce is not updated until a response is
received. Additional logic has been added to make sure no
nonce can be responded to twice though.
Mark Michelson [Fri, 10 Jul 2009 15:51:36 +0000 (15:51 +0000)]
Ensure that outbound NOTIFY requests are properly routed through stateful proxies.
With this change, we make note of Record-Route headers present in any SUBSCRIBE
request that we receive so that our outbound NOTIFY requests will have the proper
Route headers in them.
David Vossel [Wed, 8 Jul 2009 23:15:54 +0000 (23:15 +0000)]
Fixes 8khz assumptions
Many calculations assume 8khz is the codec rate. This
is not always the case. This patch only addresses chan_iax.c
and res_rtp_asterisk.c, but I am sure there are other areas
that make this assumption as well.
Mark Michelson [Wed, 8 Jul 2009 19:26:13 +0000 (19:26 +0000)]
Prevent phantom calls to queue members.
If a caller were to hang up while a periodic announcement or position
were being said, the return value for those functions would incorrectly
indicate that the caller was still in the queue. With these changes,
the problem does not occur.
(closes issue #14631)
Reported by: latinsud
Patches:
queue_announce_ghost_call2.diff uploaded by latinsud (license 745)
(with small modification from me)
David Vossel [Wed, 8 Jul 2009 16:53:40 +0000 (16:53 +0000)]
ast_samp2tv needs floating point for 16khz audio
In ast_samp2tv(), (1000000 / _rate) = 62.5 when _rate is 16000.
The .5 is currently stripped off because we don't calculate
using floating points. This causes madness with 16khz audio.
Russell Bryant [Wed, 8 Jul 2009 15:54:21 +0000 (15:54 +0000)]
Make OpenSSL usage thread-safe.
OpenSSL is not thread-safe by default. However, making it thread safe is
very easy. We just have to provide a couple of callbacks. One callback
returns a thread ID. The other handles locking. For more information,
start with the "Is OpenSSL thread-safe?" question on the FAQ page of
openssl.org.
Richard Mudgett [Thu, 2 Jul 2009 21:59:43 +0000 (21:59 +0000)]
Removed confusing warning message "Got Busy in Connected State"
If an incoming mISDN call is answered with the Answer application and a
subsequent Dial gets a busy endpoint then it is valid for that already
connected channel to get the busy indication. Asterisk will play the busy
tones until the dialplan plays something else or hangs up the call.
Tilghman Lesher [Tue, 30 Jun 2009 20:23:51 +0000 (20:23 +0000)]
More incorrect language codes, plus ensuring that regionalizations use the specified language, and not English for grammar.
(closes issue #15022)
Reported by: greenfieldtech
Patches:
20090519__issue15022.diff.txt uploaded by tilghman (license 14)
Mark Michelson [Mon, 29 Jun 2009 21:23:43 +0000 (21:23 +0000)]
Fix a problem where chan_sip would ignore "old" but valid responses.
chan_sip has had a problem for quite a long time that would manifest when
Asterisk would send multiple SIP responses on the same dialog before receiving
a response. The problem occurred because chan_sip only kept track of the highest
outgoing sequence number used on the dialog. If Asterisk sent two requests out,
and a response arrived for the first request sent, then Asterisk would ignore
the response. The result was that Asterisk would continue retransmitting the
requests and ignoring the responses until the maximum number of retransmissions
had been reached.
The fix here is to rearrange the code a bit so that instead of simply comparing
the sequence number of the response to our latest outgoing sequence number, we
walk our list of outstanding packets and determine if there is a match. If there is,
we continue. If not, then we ignore the response.
In doing this, I found a few completely useless variables that I have now removed.
David Vossel [Mon, 29 Jun 2009 17:04:04 +0000 (17:04 +0000)]
segfault after SPINLOCK schedule delete
Using the SPINLOCK schedule delete macro can result in the iax_pvt lock
being given up. This makes it possible for the iax_pvt to dissappear
when we thought we held the mutex the entire time. To resolve this, the
iax_pvt's ref count is incremented.
Richard Mudgett [Sat, 27 Jun 2009 00:55:12 +0000 (00:55 +0000)]
The ISDN CPE side should not exclusively pick B channels normally.
Before this patch, Asterisk unconditionally picked B channels exclusively
on the CPE side and normally allowed alternative B channels on the network
side. Now Asterisk does the opposite.
Reasons for the CPE side to normally not pick B channels exclusively:
* For CPE point-to-multipoint mode (i.e. phone side), the CPE side does
not have enough information to exclusively pick B channels. (There may be
other devices on the line.)
* Q.931 gives preference to the network side picking B channels.
* Some telcos require the CPE side to not pick B channels exclusively.
Russell Bryant [Fri, 26 Jun 2009 21:16:39 +0000 (21:16 +0000)]
Don't fast forward past the end of a message.
This is nice change for users of the voicemail application. If someone gets a
little carried away with fast forwarding through a message, they can easily
get to the end and accidentally exit the voicemail application by hitting the
fast forward key during the following prompt.
This adds some safety by not allowing a fast forward past the end of a message.
David Brooks [Fri, 26 Jun 2009 20:03:42 +0000 (20:03 +0000)]
Fixing voicemail's error in checking max silence vs min message length
Max silence was represented in milliseconds, yet vmminsecs (minmessage) was represented
as seconds.
Also, the inequality was reversed. The warning, if triggered, was "Max silence should
be less than minmessage or you may get empty messages", which should have been logged
if max silence was greater than minmessage, but the check was for less than.
Also, conforming if statement to coding guidelines.
Russell Bryant [Thu, 25 Jun 2009 16:02:16 +0000 (16:02 +0000)]
Resolve a crash related to a T.38 reinvite race condition.
This change resolves a crash observed locally during some T.38 testing.
A call was set up using a call file, and when the T.38 reinvite came in,
the channel state was still AST_STATE_DOWN. The reason is explained by
a comment in the code that previously lived in the handling of
AST_STATE_RINGING. This change modifies the logic to handle the same
race condition for any channel state that is not UP.
Mark Michelson [Mon, 22 Jun 2009 14:42:55 +0000 (14:42 +0000)]
Fix a situation in which Asterisk would not stop retransmitting 487s.
If a CANCEL were received by Asterisk, we would send a 487 in response
to the original INVITE and a 200 OK for the CANCEL. If there were a network
hiccup which caused the 200 OK and the 487 to be lost, then the UA communicating
with Asterisk may try to retransmit its CANCEL. Asterisk's response to this used
to be to try sending another 487 to the canceled INVITE and another 200 OK to the
CANCEL.
The problem here is that the originally-sent 487 was sent "reliably" meaning that
it will be retransmitted until it is received properly. So when we receive the second
CANCEL it is likely that the first batch of 487s we sent is still going strong and
reaches the UA. The result was that the second set of 487s would be retransmitted
constantly until the maximum number of retries had been reached.
The fix for this is that if we receive a second CANCEL for an INVITE, then we cancel
the retransmission of the first set of 487s and start a second set. This causes the
dialog to be terminated reasonably.
Mark Michelson [Mon, 22 Jun 2009 14:34:05 +0000 (14:34 +0000)]
Fix a possible infinite loop in SDP parsing during glare situation.
There was a while loop in get_ip_and_port_from_sdp which was controlled
by a call to get_sdp_iterate. The loop would exit either if what we were
searching for was found or if the return was NULL. The problem is that
get_sdp_iterate never returns NULL. This means that if what we were searching
for was not present, the loop would run infinitely. This modification of the
loop fixes the problem.
Tilghman Lesher [Fri, 19 Jun 2009 00:40:41 +0000 (00:40 +0000)]
If the "h" extension fails, give it another chance in main/pbx.c.
If the "h" extension fails, give it another chance in main/pbx.c, when it
returns from the bridge code. Fixes an issue where the "h" extension may
occasionally not fire, when a Dial is executed from a Macro.
Debugged in #asterisk with user tompaw.
Russell Bryant [Thu, 18 Jun 2009 15:24:31 +0000 (15:24 +0000)]
Fix memory corruption and leakage related reloads of non files mode MoH classes.
For Music on Hold classes that are not files mode, meaning that we are executing
an application that will feed us audio data, we use a thread to monitor the
external application and read audio from it. This thread also makes use of the
MoH class object. In the MoH class destructor, we used pthread_cancel() to ask
the thread to exit. Unfortunately, the code did not wait to ensure that the
thread actually went away. What needed to be done is a pthread_join() to ensure
that the thread fully cleans up before we proceed. By adding this one line, we
resolve two significant problems:
1) Since the thread was never joined, it never fully goes away. So, on every
reload of non-files mode MoH, an unused thread was sticking around.
2) There was a race condition here where the application monitoring thread
could still try to access the MoH class, even though the thread executing
the MoH reload has already destroyed it.
Mark Michelson [Wed, 17 Jun 2009 19:59:31 +0000 (19:59 +0000)]
Change the datastore traversal in ast_do_masquerade to use a safe list traversal.
It is possible for datastore fixup functions to remove the datastore from the list
and free it. In particular, the queue_transfer_fixup in app_queue does this. While
I don't yet know of this causing any crashes, it certainly could.
Found while discussing a separate issue with Brian Degenhardt.
David Vossel [Wed, 17 Jun 2009 19:28:12 +0000 (19:28 +0000)]
StopMixMonitor race condition (not giving up file immediately)
StopMixMonitor only indicates to the MixMonitor thread to stop
writing to the file. It does not guarantee that the recording's
file handle is available to the dialplan immediately after execution.
This results in a race condition. To resolve this, the filestream
pointer is placed in a datastore on the channel. When StopMixMonitor
is called, the datastore is retrieved from the channel and the
filestream is closed immediately before returning to the dialplan.
Documentation indicating the use of StopMixMonitor to free files
has been updated as well.
Kevin P. Fleming [Wed, 17 Jun 2009 12:03:25 +0000 (12:03 +0000)]
Correct AST_LIST_APPEND_LIST behavior when list to be appended is empty.
When the list to be appended is empty, and the list to be appended to is *not*,
AST_LIST_APPEND_LIST would actually cause the target list to become broken,
and no longer have a pointer to its last entry. This patch fixes the problem.
(reported by Stanislaw Pitucha on the asterisk-dev mailing list)
Kevin P. Fleming [Tue, 16 Jun 2009 17:05:38 +0000 (17:05 +0000)]
Improve support for media paths that can generate multiple frames at once.
There are various media paths in Asterisk (codec translators and UDPTL, primarily)
that can generate more than one frame to be generated when the application calling
them expects only a single frame. This patch addresses a number of those cases,
at least the primary ones to solve the known problems. In addition it removes the
broken TRACE_FRAMES support, fixes a number of bugs in various frame-related API
functions, and cleans up various code paths affected by these changes.
Sean Bright [Mon, 8 Jun 2009 19:24:32 +0000 (19:24 +0000)]
Increase the size of our thread stack on 64 bit processors.
We were setting the stack size for each thread to 240KB regardless of
architecture, which meant that in some scenarios we actually had less available
stack space on 64 bit processors (pointers use 8 bytes instead of 4). So now we
calculate the stack size we reserve based on the platform's __WORDSIZE, which
gives us:
32 bit -> 240KB
64 bit -> 496KB
128 bit -> 1008KB (that's right, we're ready for 128 bit processors)
Patch typed by me but written by several members of #asterisk-dev, including
Kevin, Tilghman, and Qwell.
Sean Bright [Thu, 4 Jun 2009 14:14:57 +0000 (14:14 +0000)]
Safely handle AMI connections/reload requests that occur during startup.
During asterisk startup, a lock on the list of modules is obtained by the
primary thread while each module is initialized. Issue 13778 pointed out a
problem with this approach, however. Because the AMI is loaded before other
modules, it is possible for a module reload to be issued by a connected client
(via Action: Command), causing a deadlock.
The resolution for 13778 was to move initialization of the manager to happen
after the other modules had already been lodaded. While this fixed this
particular issue, it caused a problem for users (like FreePBX) who call AMI
scripts via an #exec in a configuration file (See issue 15189).
The solution I have come up with is to defer any reload requests that come in
until after the server is fully booted. When a call comes in to
ast_module_reload (from wherever) before we are fully booted, the request is
added to a queue of pending requests. Once we are done booting up, we then
execute these deferred requests in turn.
Note that I have tried to make this a bit more intelligent in that it will not
queue up more than 1 request for the same module to be reloaded, and if a
general reload request comes in ('module reload') the queue is flushed and we
only issue a single deferred reload for the entire system.
As for how this will impact existing installations - Before 13778, a reload
issued before module initialization was completed would result in a deadlock.
After 13778, you simply couldn't connect to the manager during startup (which
causes problems with #exec-that-calls-AMI configuration files). I believe this
is a good general purpose solution that won't negatively impact existing
installations.