When publishing device state check the local cache for the
existing device state. If the new device state is unchanged
from the prior one, don't bother publishing the update. This
can reduce the work done by consumers of device state, such
as hints and app_queue, by not publishing a message to them.
These messages would most often occur with devices that are
seeing numerous simultaneous channels. The underlying device
state would remain as in use throughout, but an update would
be published as channels are created and hung up.
Although the ISDN/Q.850/Q.931 hangup cause code is already part of the ARI
and AMI hangup and channel destroyed events, it can be helpful to know what
the actual channel technology code was if the call was unsuccessful.
For PJSIP, it's the SIP response code.
* A new "tech_hangupcause" field was added to the ast_channel structure along
with ast_channel_tech_hangupcause() and ast_channel_tech_hangupcause_set()
functions. It should only be set for off-nominal terminations.
* chan_pjsip was modified to set the tech hangup cause in the
chan_pjsip_hangup() and chan_pjsip_session_end() functions. This is a bit
tricky because these two functions aren't always called in the same order.
The channel that hangs up first will get chan_pjsip_session_end() called
first which will trigger the core to call chan_pjsip_hangup() on itself,
then call chan_pjsip_hangup() on the other channel. The other channel's
chan_pjsip_session_end() function will get called last. Unfortunately,
the other channel's HangupRequest events are sent before chan_pjsip has had a
chance to set the tech hangupcause code so the HangupRequest events for that
channel won't have the cause code set. The ChannelDestroyed and Hangup
events however will have the code set for both channels.
* A new "tech_cause" field was added to the ast_channel_snapshot_hangup
structure. This is a public structure so a bit of refactoring was needed to
preserve ABI compatibility.
* The ARI ChannelHangupRequest and ChannelDestroyed events were modified to
include the "tech_cause" parameter in the JSON for off-nominal terminations.
The parameter is suppressed for nominal termination.
* The AMI SoftHangupRequest, HangupRequest and Hangup events were modified to
include the "TechCause" parameter for off-nominal terminations. Like their ARI
counterparts, the parameter is suppressed for nominal termination.
DeveloperNote: A "tech_cause" parameter has been added to the
ChannelHangupRequest and ChannelDestroyed ARI event messages and a "TechCause"
parameter has been added to the HangupRequest, SoftHangupRequest and Hangup
AMI event messages. For chan_pjsip, these will be set to the last SIP
response status code for off-nominally terminated calls. The parameter is
suppressed for nominal termination.
When an endpoint is created in the core of Asterisk a subscription
was previously created alongside it to monitor any channels being
destroyed that were related to it. This was done by receiving all
channel snapshot updates for every channel and only reacting when
it was indicated that the channel was dead.
This change removes this logic and instead provides an API call
for directly removing a channel from an endpoint. This is called
when channels are destroyed. This operation is fast, so blocking
the calling thread for a short period of time doesn't have any
noticeable impact.
Dial() already preserves the ADSI capability by copying it to the new
channel, but since Local channel pairs consist of two channels, we
also need to copy the capability to the second channel.
Resolves: #1517
As soon as SIP call may end with several Reason headers, we
want to make all of them available through the HAGUPCAUSE() function.
This implementation uses the same ao2 hash for cause codes storage
and adds a flag to make difference between last processed sip
message and content of reason headers.
UserNote: Added a new option to HANGUPCAUSE to access additional
information about hangup reason. Reason headers from pjsip
could be read using 'tech_extended' cause type.
Among the lower-quality voice codecs, some of the quality scores did
not make sense relative to each other.
For instance, quality-wise, G.729 > G.723 > PLC10.
However, current scores do not uphold these relationships.
Tweak the scores slightly to reflect more accurate relationships.
Resolves: #1501
When we retrieve a channel from a C++ map, we actually get back a wrapper
object that points to the channel then right after we retrieve it, we bump its
reference count. There's a tiny chance however that between those two
statements a delete and/or unref might happen which would cause the wrapper
object or the channel itself to become invalid resulting in a SEGV. To avoid
this we now perform a read lock on the driver around those statements.
Resolves: #1491
Commit dc8e3eeaaf improved the debug log
messages in dsp.c. This makes two minor corrections to it:
* Properly guard an added log statement in a conditional.
* Don't add one to the hit count if there was no hit (however, we do
still want to do this for the case where this is one).
Resolves: #1496
When running "config show help <module>", if no XML documentation exists
for the specified module, "Module <module> not found." is returned,
which is misleading if the module is loaded but simply has no XML
documentation for its config. Improve the message to clarify that the
module may simply have no config documentation.
Resolves: #1489
This change moves observer invocation from the use of
a threadpool to a taskpool. The taskpool options have also
been adjusted to ensure that at least one taskprocessor
remains available at all times.
Follow-on to #244 and #960 regarding how the ast_config_XXX APIs
handle template inheritance.
ast_config_text_file_save2() incorrectly suppressed variables if they
matched any ancestor template. This broke deep chains (dropping values
based on distant parents) and wide inheritance (ignoring last-wins order
across multiple parents).
The function now inspects the full template hierarchy to find the nearest
effective parent (last occurrence wins). Earlier inherited duplicates are
collapsed, explicit overrides are kept unless they exactly match the parent,
and PreserveEffectiveContext avoids writing redundant lines.
Resolves: #1451
When handling SIP transfers via ARI, the `referred_by` field in
`transfer_ari_state` may be null, since SIP REFER requests are not
required to include a `Referred-By` header. Without this check, a null
value caused the transfer to fail and triggered a NOTIFY with a 500
Internal Server Error.
This change introduces a new API called taskpool. This is a pool
of taskprocessors. It provides the following functionality:
1. Task pushing to a pool of taskprocessors
2. Synchronous tasks
3. Serializers for execution ordering of tasks
4. Growing/shrinking of number of taskprocessors in pool
This functionality already exists through the combination of
threadpool+taskprocessors but through investigating I determined
that this carries substantial overhead for short to medium duration
tasks. The threadpool uses a single queue of work, and for management
of threads it involves additional tasks.
I wrote taskpool to eliminate the extra overhead and management
as much as possible. Instead of a single queue of work each
taskprocessor has its own queue and at push time a selector chooses
the taskprocessor to queue the task to. Each taskprocessor also
has its own thread like normal. This spreads out the tasks immediately
and reduces contention on shared resources.
Using the included efficiency tests the number of tasks that can be
executed per second in a taskpool is 6-12 times more than an equivalent
threadpool+taskprocessor setup.
Stasis has been moved over to using this new API as it is a heavy consumer
of threadpool+taskprocessors and produces a lot of tasks.
UpgradeNote: The threadpool_* options in stasis.conf have now been deprecated
though they continue to be read and used. They have been replaced with taskpool
options that give greater control over the underlying taskpool used for stasis.
DeveloperNote: The taskpool API has been added for common usage of a
pool of taskprocessors. It is suggested to use this API instead of the
threadpool+taskprocessor approach.
In a previous commit, a change was made to
ast_rtp_codecs_payload_code_tx_sample_rate to check for differing sample
rates. This ended up returning an invalid payload int for comfort noise.
A check has been added that returns early if the payload is in fact
supposed to be comfort noise.
Fixes: #1340
"dialplan eval function" has been using a dummy channel for function
evaluation, much like many of the unit tests. However, sometimes, this
can cause issues for functions that are not expecting dummy channels.
As an example, ast_channel_tech(chan) is NULL on such channels, and
ast_channel_tech(chan)->type consequently results in a NULL dereference.
Normally, functions do not worry about this since channels executing
dialplan aren't dummy channels.
While some functions are better about checking for these sorts of edge
cases, use a real channel with a dummy technology to make this CLI
command inherently safe for any dialplan function that could be evaluated
from the CLI.
Resolves: #1434
Currently, the 'd' option will play dial tone while waiting
for digits. Allow it to accept an argument for any tone from
indications.conf.
Resolves: #1396
UserNote: The tone used while waiting for digits in WaitExten
can now be overridden by specifying an argument for the 'd'
option.
This patch resolves two issues in Sorcery objectset handling with multiple
backends:
1. Prevent duplicate objects:
When an object exists in more than one backend (e.g., a contact in both
'astdb' and 'realtime'), the objectset previously returned multiple instances
of the same logical object. This caused logic failures in components like the
PJSIP registrar, where duplicate contact entries led to overcounting and
incorrect deletions, when max_contacts=1 and remove_existing=yes.
This patch ensures only one instance of an object with a given key is added
to the objectset, avoiding these duplicate-related side effects.
2. Ensure missing objects are created:
When using multiple writable backends, a temporary backend failure can lead
to objects missing permanently from that backend.
Currently, .update() silently fails if the object is not present,
and no .create() is attempted.
This results in inconsistent state across backends (e.g. astdb vs. realtime).
This patch introduces a new global option in sorcery.conf:
[general]
update_or_create_on_update_miss = yes|no
Default: no (preserves existing behavior).
When enabled: if .update() fails with no data found, .create() is attempted
in that backend. This ensures that objects missing due to temporary backend
outages are re-synchronized once the backend is available again.
Added a new CLI command:
sorcery show settings
Displays global Sorcery settings, including the current value of
update_or_create_on_update_miss.
Updated tests to validate both flag enabled/disabled behavior.
Fixes: #1289
UserNote: Users relying on Sorcery multiple writable backends configurations
(e.g., astdb + realtime) may now enable update_or_create_on_update_miss = yes
in sorcery.conf to ensure missing objects are recreated after temporary backend
failures. Default behavior remains unchanged unless explicitly enabled.
* Added a new option to the WebSocket dial string to capture the additional
URI parameters.
* Added a new API ast_uri_verify_encoded() that verifies that a string
either doesn't need URI encoding or that it has already been encoded.
* Added a new API ast_websocket_client_add_uri_params() to add the params
to the client websocket session.
* Added XML documentation that will show up with `core show application Dial`
that shows how to use it.
Resolves: #1352
UserNote: A new WebSocket channel driver option `v` has been added to the
Dial application that allows you to specify additional URI parameters on
outgoing connections. Run `core show application Dial` from the Asterisk CLI
to see how to use it.
The debug logging during DSP processing has always been kind
of overwhelming and annoying to troubleshoot. Simplify and
improve the logging in a few ways to aid DSP debugging:
* If we had a DSP hit, don't also emit the previous debug message that
was always logged. It is duplicated by the hit message, so this can
reduce the number of debug messages during detection by 50%.
* Include the hit count and required number of hits in the message so
on partial detections can be more easily troubleshot.
* Use debug level 9 for hits instead of 10, so we can focus on hits
without all the noise from the per-frame debug message.
* 1-index the hit count in the debug messages. On the first hit, it
currently logs '0', just as when we are not detecting anything,
which can be confusing.
Resolves: #1375
If you do a `core show application Dial`, you'll see it's kind of a mess.
Indents are wrong is some places, examples are printed in black which makes
them invisible on most terminals, and the lack of line breaks in some cases
makes it hard to follow.
* Fixed the rendering of examples so they are indented properly and changed
the color so they can be seen.
* There is now a line break before each option.
* Options are now printed on their own line with all option content indented
below them.
Example from Dial before fixes:
```
Example: Dial 555-1212 on first available channel in group 1, searching
from highest to lowest
Example: Ringing FXS channel 4 with ring cadence 2
Example: Dial 555-1212 on channel 3 and require answer confirmation
...
O([mode]):
mode - With <mode> either not specified or set to '1', the originator
hanging up will cause the phone to ring back immediately.
- With <mode> set to '2', when the operator flashes the trunk, it will ring
their phone back.
Enables *operator services* mode. This option only works when bridging a DAHDI
channel to another DAHDI channel only. If specified on non-DAHDI interfaces, it
will be ignored. When the destination answers (presumably an operator services
station), the originator no longer has control of their line. They may hang up,
but the switch will not release their line until the destination party (the
operator) hangs up.
p: This option enables screening mode. This is basically Privacy mode
without memory.
```
After:
```
Example: Dial 555-1212 on first available channel in group 1, searching
from highest to lowest
same => n,Dial(DAHDI/g1/5551212)
Example: Ringing FXS channel 4 with ring cadence 2
same => n,Dial(DAHDI/4r2)
Example: Dial 555-1212 on channel 3 and require answer confirmation
same => n,Dial(DAHDI/3c/5551212)
...
O([mode]):
mode - With <mode> either not specified or set to '1', the originator
hanging up will cause the phone to ring back immediately.
With <mode> set to '2', when the operator flashes the trunk, it will
ring their phone back.
Enables *operator services* mode. This option only works when bridging
a DAHDI channel to another DAHDI channel only. If specified on
non-DAHDI interfaces, it will be ignored. When the destination answers
(presumably an operator services station), the originator no longer has
control of their line. They may hang up, but the switch will not
release their line until the destination party (the operator) hangs up.
p:
This option enables screening mode. This is basically Privacy mode
without memory.
```
There are still things we can do to make this more readable but this is a
start.
If the BRIDGE_NOANSWER variable is set on a channel, it is not supposed
to answer when another channel bridges to it using Bridge(), and this is
checked when ast_bridge_call* is called. However, another path exists
(bridge_exec -> ast_bridge_add_channel) where this variable was not
checked and channels would be answered. We now check the variable there.
Resolves: #401Resolves: #1364
With `sounds_search_custom_dir = yes`, we are supposed to search for sounds
in the `AST_DATA_DIR/sounds/custom` directory before searching the normal
directories. Unfortunately, a recent change
(https://github.com/asterisk/asterisk/pull/1172) had a typo resulting in
the "custom" directory not being searched. This change restores this
expected behavior.
Resolves: #1353
Fixes: #1280
UserNote: Enabling the tracking of the
STREAM_BEGIN and the STREAM_END event
types in cel.conf will log media files and
music on hold played to each channel.
The STREAM_BEGIN event's extra field will
contain a JSON with the file details (path,
format and language), or the class name, in
case of music on hold is played. The DTMF
event's extra field will contain a JSON with
the digit and the duration in milliseconds.
Remove the deprecated 'rotatetimestamp' config option, as this
was deprecated by 'rotatestrategy' in 1.6 by commit
f5a14167f3.
Resolves: #1345
UpgradeNote: The deprecated rotatetimestamp option has been removed.
Use rotatestrategy instead.
The "no debug channel" command has been deprecated since
1.6 (commit 691363656f),
as it is replaced by "core set debug channel", which also
supports tab-completion on channels. Remove the redundant
command.
Resolves: #1343
UpgradeNote: The deprecated "no debug channel" command has
now been removed; use "core set debug channel" instead.
The fact that deleting an object from a map invalidates any iterator
that happens to currently point to that object was overlooked in the initial
implementation. Unfortunately, there's no way to detect that an iterator
has been invalidated so the result was an occasional SEGV triggered by modules
like app_chanspy that opens an iterator and can keep it open for a long period
of time. The new implementation doesn't keep the underlying C++ iterator
open across calls to ast_channel_iterator_next() and uses a read lock
on the map to ensure that, even for the few microseconds we use the
iterator, another thread can't delete a channel from under it. Even with
this change, the iterators are still WAY faster than the ao2_legacy
storage driver.
Full details about the new implementation are located in the comments for
iterator_next() in channelstorage_cpp_map_name_id.cc.
Resolves: #1309
In the original implementation, both CANCEL and NO ANSWER states were
consolidated under the NO ANSWER disposition. This patch introduces a
separate CANCEL disposition, with an optional configuration switch to
enable this new disposition.
Resolves: #1323
UserNote: A new CDR option "canceldispositionenabled" has been added
that when set to true, the NO ANSWER disposition will be split into
two dispositions: CANCEL and NO ANSWER. The default value is 'no'
DeveloperNote: The 32-bit ast_options has no room left to accomodate new
options and so has been converted to an ast_flags64 structure. All internal
references to ast_options have been updated to use the 64-bit flag
manipulation macros. External module references to the 32-bit ast_options
should continue to work on little-endian systems because the
least-significant bytes of a 64 bit integer will be in the same location as a
32-bit integer. Because that's not the case on big-endian systems, we've
swapped the bytes in the flags manupulation macros on big-endian systems
so external modules should still work however you are encouraged to test.
This patch fixes an issue in the ODBC Realtime engine where Asterisk incorrectly
falls back to the next configured backend when the current one returns
SQL_NO_DATA (i.e., no record found).
This is a logical error and performance risk in multi-backend configurations.
Solution:
Introduced CONFIG_RT_NOT_FOUND ((void *)-1) as a special return marker.
ODBC Realtime backend now return CONFIG_RT_NOT_FOUND when no data is found.
Core engine stops iterating on this marker, avoiding unnecessary fallback.
Notes:
Other Realtime backends (PostgreSQL, LDAP, etc.) can be updated similarly.
This patch only covers ODBC.
Fixes: #1305
Under certain circumstances the context/extens/prio are set in the ast_async_goto, for example action Redirect.
In the situation that action Redirect is broken by pbx_extension_helper this info is changed.
This will cause the current dialplan location to be executed twice.
In other words, the Redirect action does not take effect.
Resolves: #1315
The CDR tenantid was being set in cdr_object_alloc from the channel->base
snapshot. Since this happens at channel creation before the dialplan is even
reached, calls to `CHANNEL(tenantid)=<something>` in the dialplan were being
ignored. Instead we now take tenantid from party_a when
cdr_object_create_public_records() is called which is after the call has
ended and all channel snapshots rebuilt. This is exactly how accountcode
and amaflags, which can also be set in tha dialplpan, are handled.
Resolves: #1259
When the callback() API was invoked but no channel passed the test, callback
would return the last channel tested instead of NULL. It now correctly
returns NULL when no channel matches.
Resolves: #1288
users.conf was deprecated in Asterisk 21 and is now being removed
for Asterisk 23, in accordance with the Asterisk deprecation policy.
This consists of:
* Removing integration with app_directory, app_voicemail, chan_dahdi,
chan_iax2, and AMI.
* users.conf was also partially used for res_phoneprov, and this remaining
functionality is consolidated to a separate phoneprov_users.conf,
used only by res_phoneprov.
Resolves: #1292
UpgradeNote: users.conf has been removed and all channel drivers must
be configured using their specific configuration files. The functionality
previously in users.conf for res_phoneprov is now in phoneprov_users.conf.
* Created chan_websocket which can exchange media over both inbound and
outbound websockets which the driver will frame and time.
See http://s.asterisk.net/mow for more information.
* res_http_websocket: Made defines for max message size public and converted
a few nuisance verbose messages to debugs.
* main/channel.c: Changed an obsolete nuisance error to a debug.
* ARI channels: Updated externalMedia to include chan_websocket as a supported
transport.
UserNote: A new channel driver "chan_websocket" is now available. It can
exchange media over both inbound and outbound websockets and will both frame
and re-time the media it receives.
See http://s.asterisk.net/mow for more information.
UserNote: The ARI channels/externalMedia API now includes support for the
WebSocket transport provided by chan_websocket.
DEBUG_FD_LEAKS replaces calls to "open" and "close" with functions that keep
track of file descriptors, even when those calls are actually callbacks
defined in structures like ast_channelstorage_instance->open and don't touch
file descriptors. This causes compilation failures. Those callbacks
have been renamed to "open_instance" and "close_instance" respectively.
Resolves: #1287
This patch adjusts the read/write synchronization logic in audiohook_read_frame_both()
to better handle calls where participants use different codecs or sample sizes
(e.g., alaw vs G.722). The previous hard threshold of 2 * samples caused MixMonitor
recordings to break or stutter when frames were not aligned between both directions.
The new logic uses a more tolerant limit (1.5 * samples), which prevents audio tearing
without causing excessive buffer overruns. This fix specifically addresses issues
with MixMonitor when recording directly on a channel in a bridge using mixed codecs.
Reported-by: Michal Hajek <michal.hajek@daktela.com>
Resolves: #1276Resolves: #1279
Under certain circumstances the context/extens/prio are set in the ast_async_goto, for example action Redirect.
In the situation that action Redirect is broken by GotoIf this info is changed.
that will causes confusion in dialplan execution.
Resolves: #1273
Asterisk can now establish websocket sessions _to_ your ARI applications
as well as accepting websocket sessions _from_ them.
Full details: http://s.asterisk.net/ari-outbound-ws
Code change summary:
* Added an ast_vector_string_join() function,
* Added ApplicationRegistered and ApplicationUnregistered ARI events.
* Converted res/ari/config.c to use sorcery to process ari.conf.
* Added the "outbound-websocket" ARI config object.
* Refactored res/ari/ari_websockets.c to handle outbound websockets.
* Refactored res/ari/cli.c for the sorcery changeover.
* Updated res/res_stasis.c for the sorcery changeover.
* Updated apps/app_stasis.c to allow initiating per-call outbound websockets.
* Added CLI commands to manage ARI websockets.
* Added the new "outbound-websocket" object to ari.conf.sample.
* Moved the ARI XML documentation out of res_ari.c into res/ari/ari_doc.xml
UserNote: Asterisk can now establish websocket sessions _to_ your ARI applications
as well as accepting websocket sessions _from_ them.
Full details: http://s.asterisk.net/ari-outbound-ws
UserNote: A new asterisk.conf option 'disable_remote_console_shell' has
been added that, when set, will prevent remote consoles from executing
shell commands using the '!' prefix.
Resolves: #GHSA-c7p6-7mvq-8jq2
We have a use-case where we generate a *lot* of events on the AMI, and
then when doing `manager show eventq` we would see some events which
would linger for hours or days in there. Obviously something was leaking.
Testing allowed us to track down this logic bug in the ref-counting on
the event purge.
Reproducing the bug was not super trivial, we managed to do it in a
production-like load testing environment with multiple AMI consumers.
The race condition itself:
1. something allocates and links `session`
2. `purge_sessions` iterates over that `session` (takes ref)
3. `purge_session` correctly de-referencess that session
4. `purge_session` re-evaluates the while() loop, taking a reference
5. `purge_session` exits (`n_max > 0` is false)
6. whatever allocated the `session` deallocates it, but a reference is
now lost since we exited the `while` loop before de-referencing.
7. since the destructor is never called, the session->last_ev->usecount
is never decremented, leading to events lingering in the queue
The impact of this bug does not seem major. The events are small and do
not seem, from our testing, to be causing meaningful additional CPU
usage. Mainly we wanted to fix this issue because we are internally
adding prometheus metrics to the eventq and those leaked events were
causing the metrics to show garbage data.
Full details: http://s.asterisk.net/dc679ec3
The previous proof-of-concept showed that the cpp_map_name_id alternate
storage backed performed better than all the others so this final PR
adds only that option. You still need to enable it in menuselect under
the "Alternate Channel Storage Backends" category.
To select which one is used at runtime, set the "channel_storage_backend"
option in asterisk.conf to one of the values described in
asterisk.conf.sample. The default remains "ao2_legacy".
UpgradeNote: With this release, you can now select an alternate channel
storage backend based on C++ Maps. Using the new backend may increase
performance and reduce the chances of deadlocks on heavily loaded systems.
For more information, see http://s.asterisk.net/dc679ec3
