Here is a copy and paste of the details from my request on
reviewboard that dealt with these changes:
Fix 1. The first change in place is to fix Mantis issue 15811, which deals with a situation where Asterisk will incorrectly interpret out of order RFC2833 frames as duplicate DTMF digits. For instance, we would receive a sequence like:
Prior to this patch when we received the frame with seqno 5, we would interpret this as a new DTMF 1. With this patch, we will check the seqno of the incoming digit and not process the frame if the seqno is lower than the last recorded seqno. Note that we do not record the seqno of the dropped DTMF frame for future processing. While the above situation is what was designed to be fixed, the patch is written in such a way that the following would also be fixed too:
In this second situation, the beginning of the DTMF 2 arrives before the final end frame of the DTMF 1. With the patch, seqno 12 is no processed and thus we properly interpret the DTMF.
Fix 2. The second change in place is to fix an issue like the following:
When we receive seqno 6, we had code in place that was supposed to properly end the previously unended DTMF 1. The problem was that the code was essentially a no-op. The code would set up an end frame for the DTMF 1 but would immediately overwrite the frame with the begin for DTMF 2. I changed process_dtmf_rfc2833() so that instead of returning a single frame, it is given as an output parameter a list of frames. Each frame that needs to be returned is appended to this list.
Fix 3. The final change is a minor one where an AST_CONTROL_SRCCHANGE frame could get lost. If we process a cisco DTMF or an RFC 3389 frame and no frame was returned, then we would return &ast_null_frame. The problem is that earlier in the function, we may have generated an AST_CONTROL_SRCCHANGE frame and put it in the list of frames we wish to return. This frame would be lost in such a case. The patch fixes this problem
........
................
Ensure that monitor recordings are written to the correct location (again)
This is an extension to 248860. As such the dialplan test has been extended:
; non absolute path, not combined
exten => 5040, 1, monitor(wav,tmp/jeff/monitor_test)
exten => 5040, n, dial(sip/5001)
; absolute path, not combined
exten => 5041, 1, monitor(wav,/tmp/jeff/monitor_test2)
exten => 5041, n, dial(sip/5001)
; no path, not combined
exten => 5042, 1, monitor(wav,monitor_test3)
exten => 5042, n, dial(sip/5001)
; combined: changemonitor from non absolute to no path (leaves tmp/jeff)
exten => 5043, 1, monitor(wav,tmp/jeff/monitor_test4,m)
exten => 5043, n, changemonitor(monitor_test5)
exten => 5043, n, dial(sip/5001)
; combined: changemonitor from no path to non absolute path
exten => 5044, 1, monitor(wav,monitor_test6,m)
exten => 5044, n, changemonitor(tmp/jeff/monitor_test7) ; this wasn't possible before
exten => 5044, n, dial(sip/5001)
; non absolute path, combined
exten => 5045, 1, monitor(wav,tmp/jeff/monitor_test8,m)
exten => 5045, n, dial(sip/5001)
; absolute path, combined
exten => 5046, 1, monitor(wav,/tmp/jeff/monitor_test9,m)
exten => 5046, n, dial(sip/5001)
; no path, combined
exten => 5047, 1, monitor(wav,monitor_test10,m)
exten => 5047, n, dial(sip/5001)
; combined: changemonitor from non absolute to absolute (leaves tmp/jeff)
exten => 5048, 1, monitor(wav,tmp/jeff/monitor_test11,m)
exten => 5048, n, changemonitor(/tmp/jeff/monitor_test12)
exten => 5048, n, dial(sip/5001)
; combined: changemonitor from absolute to non absolute (leaves /tmp/jeff)
exten => 5049, 1, monitor(wav,/tmp/jeff/monitor_test13,m)
exten => 5049, n, changemonitor(tmp/jeff/monitor_test14)
exten => 5049, n, dial(sip/5001)
; combined: changemonitor from no path to absolute
exten => 5050, 1, monitor(wav,monitor_test15,m)
exten => 5050, n, changemonitor(/tmp/jeff/monitor_test16)
exten => 5050, n, dial(sip/5001)
; combined: changemonitor from absolute to no path (leaves /tmp/jeff)
exten => 5051, 1, monitor(wav,/tmp/jeff/monitor_test17,m)
exten => 5051, n, changemonitor(monitor_test18)
exten => 5051, n, dial(sip/5001)
; not combined: changemonitor from non absolute to no path (leaves tmp/jeff)
exten => 5052, 1, monitor(wav,tmp/jeff/monitor_test19)
exten => 5052, n, changemonitor(monitor_test20)
exten => 5052, n, dial(sip/5001)
; not combined: changemonitor from no path to non absolute
exten => 5053, 1, monitor(wav,monitor_test21)
exten => 5053, n, changemonitor(tmp/jeff/monitor_test22)
exten => 5053, n, dial(sip/5001)
; not combined: changemonitor from non absolute to absolute (leaves tmp/jeff)
exten => 5054, 1, monitor(wav,tmp/jeff/monitor_test23)
exten => 5054, n, changemonitor(/tmp/jeff/monitor_test24)
exten => 5054, n, dial(sip/5001)
; not combined: changemonitor from absolute to non absolute (leaves /tmp/jeff)
exten => 5055, 1, monitor(wav,/tmp/jeff/monitor_test24)
exten => 5055, n, changemonitor(tmp/jeff/monitor_test25)
exten => 5055, n, dial(sip/5001)
; not combined: changemonitor from no path to absolute
exten => 5056, 1, monitor(wav,monitor_test26)
exten => 5056, n, changemonitor(/tmp/jeff/monitor_test27)
exten => 5056, n, dial(sip/5001)
; not combined: changemonitor from absolute to no path (leaves /tmp/jeff)
exten => 5057, 1, monitor(wav,/tmp/jeff/monitor_test28)
exten => 5057, n, changemonitor(monitor_test29)
exten => 5057, n, dial(sip/5001)
........
................
Exit native bridging early for greater timing accuracy with warnings
This changes native bridging to break one millisecond early so that the more
accurate timeval calculations done in the generic bridge can be performed using
the bridge config. Currently the time between exiting native bridging slightly
late can sometimes cause a large enough discrepancy for warnings to be missed.
For the record, 1.4 does not attempt to native bridge at all when warnings are
enabled.
prevent segfault if bad magic number is encountered.
internal_ao2_ref uses INTERNAL_OBJ which mzy report 'bad magic number', but
internal_ao2_ref continues on, causing segfault.
Although AO2_MAGIC number is checked by INTERNAL_OBJ before internal_ao2_ref is
called, A02_MAGIC is being destroyed (or a wrong pointer) by the time
internal_ao2_ref uses INTERNAL_OBJ.
internal_ao2_ref now returns -1 if INTERNAL_OBJ encouters a bad magic number.
Mask out previous arguments on each nested invocation of Gosub.
(closes issue #16758)
Reported by: wdoekes
Patches:
20100316__issue16758.diff.txt uploaded by tilghman (license 14)
Russell Bryant [Tue, 16 Mar 2010 19:01:50 +0000 (19:01 +0000)]
Merged revisions 252767 via svnmerge from
https://origsvn.digium.com/svn/asterisk/trunk
................
r252767 | russell | 2010-03-16 14:01:04 -0500 (Tue, 16 Mar 2010) | 13 lines
Merged revisions 252766 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r252766 | russell | 2010-03-16 14:00:43 -0500 (Tue, 16 Mar 2010) | 6 lines
Don't treat warnings as errors for muted.
muted supports OS X, but uses functions marked as deprecated in 10.6. However,
the functions are still supported, so just ignore the warnings for now and
allow the build to proceed.
........
................
Fixing up a couple more overlapping global variable namespaces shared with
extensions.conf.sample. Also noticed a few of the lines that were commented
out didn't have the closing semi-colon so I added that as well.
Update extensions.ael file to not overlap extensions.conf.
Updated the extensions.ael file so the global variables don't overlap
those that we have in extensions.conf (sample files). This way unexpected
things won't happed hopefully if both pbx_ael and res_config are loaded.
Only change the RTP ssrc when we see that it has changed
This change basically reverts the change reviewed in
https://reviewboard.asterisk.org/r/374/ and instead limits the
updating of the RTP synchronization source to only those times when we
detect that the other side of the conversation has changed the ssrc.
The problem is that SRCUPDATE control frames are sent many times where
we don't want a new ssrc, including whenever Asterisk has to send DTMF
in a normal bridge. This is also not the first time that this mistake
has been made. The initial implementation of the ast_rtp_new_source
function also changed the ssrc--and then it was removed because of
this same issue. Then, we put it back in again to fix a different
issue. This patch attempts to only change the ssrc when we see that
the other side of the conversation has changed the ssrc.
It also renames some functions to make their purpose more clear.
Tilghman Lesher [Thu, 11 Mar 2010 19:21:19 +0000 (19:21 +0000)]
Verify whether the created buffer was actually large enough to hold the expanded value.
For certain types of queries, where the size of the substituted query was much
larger than the template, it was possible for the substitution buffer to be too
small. This is only an issue in 1.6.1, as previously we used a static buffer
anyway, and we have a substitution routine in 1.6.2 forward that automatically
sizes itself appropriately to handle larger expansions.
Fix ParkAndAnnounce not respecting parking options.
The patch ensures that if a peer does not exist, parking settings are read from
the channel. A unit test has been written to ensure proper operation for both
standard parking and parking using masquerades.
Fix Debian init script to not use -c.
When using the init script as-is currently, it could cause issues on Debian
such as high CPU usage. This fix has worked for several people so I'm
implementing the change.
A complete re-write of the Local channel documentation has been performed, with
the existing information from localchannel.txt and localchannel.tex merged in.
Make sure to clear red alarm after polarity reversal.
From the issue:
The automatic overnight line tests (or manual ones) used on UK (BT) lines causes
a red alarm on a dahdi / TDM400P connected channel. This is because the line
uses voltage tests (battery loss) and polarity reversal. The polarity reversal
causes chan_dahdi to initiate v23 CallerID processing but during this the event
DAHDI_EVENT_NOALARM is ignored so that the alarm is never cleared.
When Asterisk receives an IAX2 TXREQ packet, try_transfer()
will call store_by_transfercallno() to link the chan_iax2_pvt
struct into iax_transfercallno_pvts. If a duplicate TXREQ
packet is received for the same call, the pvt struct will be
linked into iax_transfercallno_pvts multiple times. This patch
fixes this. Thanks rain for debugging this and providing a patch!
Update IMAP documentation.
Update the IMAP documentation to make it clear that storing voicemails
in the same folder as a large number of emails could potentially cause
significant slow downs when writing or retrieving voicemails.
When configuring the adaptive jitterbuffer, the target_extra
value not only could not be set from the configuration, but was
not even being set to its proper default. This value is required
in order for the adaptive jitterbuffer to work correctly. To resolve
this a config option has been added to expose this value to the conf
files, and a default value is provided when no config specific value
is present.
........
Modify queued frames from local channels to not set the other side to up
In this case, attended transfers were broken due to ast_feature_request_and_dial
detecting the channel being set to up before the answer frame could be read and
therefore failing to mark the channel as ready. This fix is a regression fix for
244785, which should continue to work properly as well.
Alec L Davis [Sat, 27 Feb 2010 23:37:16 +0000 (23:37 +0000)]
overlap receiving: automatically send CALL PROCEEDING when dialplan starts
Following Q.931 5.2.4
When the user has determined that sufficient call information has been received the
user shall stop T302 and send CALL PROCEEDING to the network.
Previously timeouts were possible if the dialplan took a long time to issue any
response back to the network.
Cleanups to fix bugs in the VM count API functions.
- Urgent voicemails were not attached, because the attachment code looked in the wrong folder.
- Urgent voicemails were sometimes counted twice when displaying the count of new messages.
- Backends were inconsistent as to which voicemails each API counted.
Some platforms clear /var/run at boot, which makes connecting a remote console... difficult.
Previously, we only created the default /var/run/asterisk directory at install
time. While we could create it in the init script, that would not work for
those who start asterisk manually from the command line. So the safest thing
to do is to create it as part of the Asterisk boot process. This also changes
the ownership of the directory, because the pid and ctl files are created after
we setuid/setgid.
(closes issue #16802)
Reported by: Brian
Patches:
20100224__issue16802.diff.txt uploaded by tilghman (license 14)
Tested by: tzafrir
........
................
Remove color code sequences from verbose messages that go to logfiles.
(closes issue #16786)
Reported by: dodo
Patches:
logger2.patch uploaded by dodo (license 989)
Tested by: tilghman
........
................
Fix various problems detected with Valgrind.
* chan_console accessed pvts after deallocation.
* The module loader did not check usecount on shutdown, which led to chan_iax2
reading a timer that was already unloaded.
(closes issue #16062)
Reported by: alexanderheinz
Patches:
20091109__issue16062.diff.txt uploaded by tilghman (license 14)
Tested by: tilghman
........
Make chan_misdn DTMF processing consistent with other channel technologies.
The processing of DTMF tones on the receiving side of an ISDN channel is
inconsistent with the way it is handled in other channels, especially
DAHDI analog. This causes DTMF tones sent from an ISDN phone to be
doubled at the connected party.
We are using the following 2 options of misdn.conf
1) astdtmf=yes
2) senddtmf=yes
Option one is necessary because the asterisk DSP DTMF detection is better
than mISDN's internal DSP. Not as many false positives.
Option two is necessary to transmit DTMF tones end to end when mISDN
channels are connected to SIP channels with out of band DTMF for example.
The symptom is that DTMF tones sent by an ISDN phone are doubled on the
way through asterisk when two mISDN channels are connected with a Local
channel in between or if it is bridged to an analog channel.
The doubling of DTMF tones is because DTMF is passed inband to asterisk by
the mISDN channel and passed out of band once again after the release of
the DTMF tone. Passing it inband is wrong. Neither an analog channel nor
SIP channel passes DTMF inband if configured to inband DTMF. Analog and
SIP channels filter out the DTMF tones because they use the voice frames
returned by ast_dsp_process. But chan_misdn passes the unfiltered input
voice frames instead.
To overcome one aspect of the problem, the doubling of DTMF tones when two
mISDN channels are directly bridged, someone made an 'optimization', where
in that case the DTMF tone passed out-of-band to the peer channel is not
translated to an inband tone at the transmit side. This optimization is
bad because it does not work in general. For example, analog channels or
mISDN channels when bridged through an intermediary local channel will
generate DTMF tones from out-of-band information. Also, of course, it
must not be done when there is no inband DTMF available.
This patch fixes the issue. Now chan_misdn will filter the received
inband DTMF signal the same as other channel types.
Another change included: No need to build an extra translation path
because ast_process_dsp does it if required.
I'm working with this code right now trying to analyze a deadlock.
This change is just to clean up a few things before I make a more
complex patch.
........
If the peer record is from realtime, it could be set to 0, due to MySQL not representing NULL well in integer columns.
NULL means the value is not specified for the column, which normally means
the driver uses whatever is the default value. However, on MySQL, placing
a NULL in either a float or integer column results in a retrieval of the 0
value. Hence, users get an errant error on load. This patch suppresses
that error and makes the value as if it was not there.
Note that this cannot be done in the realtime driver, because the lack of
difference between NULL and 0 can only be intepreted correctly by the
driver itself. If we did it in the realtime driver, then it would be
effectively impossible to set any realtime field to 0, because it would act
as if the field were unspecified and possibly take on a different value.
Russell Bryant [Thu, 18 Feb 2010 04:21:07 +0000 (04:21 +0000)]
Merged revisions 247423 via svnmerge from
https://origsvn.digium.com/svn/asterisk/trunk
................
r247423 | russell | 2010-02-17 22:20:11 -0600 (Wed, 17 Feb 2010) | 17 lines
Merged revisions 247422 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r247422 | russell | 2010-02-17 22:19:01 -0600 (Wed, 17 Feb 2010) | 10 lines
Tweak argument handling for wget in the sounds Makefile.
1) Fix the check to see if we are using wget to not be full of fail. The
configure script populates this variable with the absolute path to wget if
it is found, so it didn't work.
2) Allow some extra arguments to be passed in for wget. This is just a simple
change to allow our Bamboo build script to tell wget to be quiet and not fill
up our logs with download status output.
........
................
Fix two problems in ast_str functions found while writing a unit test.
1. The documentation for ast_str_set and ast_str_append state that
the max_len parameter may be -1 in order to limit the size of the
ast_str to its current allocated size. The problem was that the max_len
parameter in all cases was a size_t, which is unsigned. Thus a -1 was
interpreted as UINT_MAX instead of -1. Changing the max_len parameter
to be ssize_t fixed this issue.
2. Once issue 1 was fixed, there was an off-by-one error in the case
where we attempted to write a string larger than the current allotted
size to a string when -1 was passed as the max_len parameter. When trying
to write more than the allotted size, the ast_str's __AST_STR_USED was
set to 1 higher than it should have been. Thanks to Tilghman for quickly
spotting the offending line of code.
Oh, and the unit test that I referenced in the top line of this commit
will be added to reviewboard shortly. Sit tight...
........
According to the man page for stdarg(3),
"Each invocation of va_copy() must be matched by a
corresponding invocation of va_end() in the same
function."
There were several cases in __ast_str_helper where
va_copy was not matched with a corresponding call
to va_end.
........
fixes sample rate conversion issue with Monitor application
When using ast_seekstream with the read/write streams of a monitor,
the number of samples we are seeking must be of the same rate as the
stream or the jump calculation will be incorrect. This patch adds logic
to correctly convert the number of samples to jump to the sample rate
the read/write stream is using.
For example, if the call is G722 (16khz) and the read/write stream is
recording a 8khz wav, seeking 320 samples of 16khz audio is not the
same as seeking 320 samples of 8khz audio when performing the ast_seekstream
on the stream.
On channel destruction the channel's datastores are removed and
destroyed. Since there are public API calls to find and remove
datastores on a channel, a lock should be held whenever datastores are
removed and destroyed. This resolves a crash caused by a race
condition in app_chanspy.c.
(closes issue #16678)
Reported by: tim_ringenbach
Patches:
datastore_destroy_race.diff uploaded by tim ringenbach (license 540)
Tested by: dvossel
........
................
Change channel state on local channels for busy,answer,ring.
Previously local channels channel state never changed. This became problematic
when the state of the other side of the local channel was lost, for example
during a masquerade. Changing the state of the local channel allows for the
scenario to be detected when the channel state is set to ringing, but the peer
isn't ringing. The specific problem scenario is described in 164201. Although
this was noted on one of the issues, here is the tested dialplan verified to
work:
projects / thirdparty / asterisk.git / log
