Consumers like media_cache have been running into issues with
the previous astdb "/family/key" limit of 253 bytes when needing
to store things like long URIs. An Amazon S3 URI is a good example
of this. Now, instead of using a static 256 byte buffer for
"/family/key", we use ast_asprintf() to dynamically create it.
Both test_db.c and test_media_cache.c were also updated to use
keys/URIs over the old 253 character limit.
Resolves: #881
UserNote: The `ast_db_*()` APIs have had the 253 byte limit on
"/family/key" removed and will now accept families and keys with a
total length of up to SQLITE_MAX_LENGTH (currently 1e9!). This
affects the `DB*` dialplan applications, dialplan functions,
manager actions and `databse` CLI commands. Since the
media_cache also uses the `ast_db_*()` APIs, you can now store
resources with URIs longer than 253 bytes.
verification.c had an include for jansson.h left over from previous
versions of the module. Since res_stir_shaken no longer has a
dependency on jansson, the bundled version wasn't added to GCC's
include path so if you didn't also have a jansson development package
installed, the compile would fail. Removing the stale include
was the only thing needed.
Resolves: #889
* If the call to ast_config_load() returns CONFIG_STATUS_FILEINVALID,
check_for_old_config() now returns LOAD_DECLINE instead of continuing
on with a bad pointer.
* If CONFIG_STATUS_FILEMISSING is returned, check_for_old_config()
assumes the config is being loaded from realtime and now returns
LOAD_SUCCESS. If it's actually not being loaded from realtime,
sorcery will catch that later on.
* Also refactored the error handling in load_module() a bit.
Resolves: #884
For both attestation and verification, we now check whether they've
been disabled either globally or by the profile before validating
things like callerid, orig_tn, dest_tn, etc. This prevents useless
error messages.
Resolves: #879
The ub_result pointer passed to unbound_resolver_callback by
libunbound can be NULL if the query was for something malformed
like `.1` or `[.1]`. If it is, we now set a 'ns_r_formerr' result
and return instead of crashing with a SEGV. This causes pjproject
to simply cancel the transaction with a "No answer record in the DNS
response" error. The existing "off nominal" unit test was also
updated to check this condition.
Although not necessary for this fix, we also made
ast_dns_resolver_completed() tolerant of a NULL result.
Resolves: GHSA-v428-g3cw-7hv9
When commit e8c9cb80 was cherry-picked in from master, the
fact that the 20 and 18 branches still had the old "macrocontext"
column wasn't taken into account so the number of named parameters
didn't match the number of '?' placeholders. They do now.
We also now use ast_asprintf to create the full mailbox query SQL
statement instead of trying to calculate the proper length ourselves.
Resolves: #831
* A static array of security mechanism type names was created.
* ast_sip_str_to_security_mechanism_type() was refactored to do
a lookup in the new array instead of using fixed "if/else if"
statments.
* security_mechanism_to_str() and ast_sip_security_mechanisms_to_str()
were refactored to use ast_str instead of a fixed length buffer
to store the result.
* ast_sip_security_mechanism_type_to_str was removed in favor of
just referencing the new type name array. Despite starting with
"ast_sip_", it was a static function so removing it doesn't affect
ABI.
* Speaking of "ast_sip_", several other static functions that
started with "ast_sip_" were renamed to avoid confusion about
their public availability.
* A few VECTOR free loops were replaced with AST_VECTOR_RESET().
* Fixed a meomry leak in pjsip_configuration.c endpoint_destructor
caused by not calling ast_sip_security_mechanisms_vector_destroy().
* Fixed a memory leak in res_pjsip_outbound_registration.c
add_security_headers() caused by not specifying OBJ_NODATA in
an ao2_callback.
* Fixed a few ao2_callback return code misuses.
Resolves: #845
Add dialplan application PJSIPNOTIFY to send either pre-configured
NOTIFY messages from pjsip_notify.conf or with headers defined in
dialplan.
Also adds the ability to send pre-configured NOTIFY commands to a
channel via the CLI.
Resolves: #799
UserNote: A new dialplan application PJSIPNotify is now available
which can send SIP NOTIFY requests from the dialplan.
The pjsip send notify CLI command has also been enhanced to allow
sending NOTIFY messages to a specific channel. Syntax:
pjsip send notify <option> channel <channel>
If you run an AMI CoreShowChannelMap on a channel that isn't in a
bridge and you're in DEVMODE, you can get a FRACK because the
bridge id is empty. We now simply return an empty list for that
request.
This patch introduces a new identifier for channels: tenantid. It's
a stringfield on the channel that can be used for general purposes. It
will be inherited by other channels the same way that linkedid is.
You can set tenantid in a few ways. The first is to set it in the
dialplan with the Set and CHANNEL functions:
exten => example,1,Set(CHANNEL(tenantid)=My tenant ID)
It can also be accessed via CHANNEL:
exten => example,2,NoOp(CHANNEL(tenantid))
Another method is to use the new tenantid option for pjsip endpoints in
pjsip.conf:
[my_endpoint]
type=endpoint
tenantid=My tenant ID
This is considered the best approach since you will be able to see the
tenant ID as early as the Newchannel event.
It can also be set using set_var in pjsip.conf on the endpoint like
setting other channel variable:
set_var=CHANNEL(tenantid)=My tenant ID
Note that set_var will not show tenant ID on the Newchannel event,
however.
Tenant ID has also been added to CDR. It's read-only and can be accessed
via CDR(tenantid). You can also get the tenant ID of the last channel
communicated with via CDR(peertenantid).
Tenant ID will also show up in CEL records if it has been set, and the
version number has been bumped accordingly.
Fixes: #740
UserNote: tenantid has been added to channels. It can be read in
dialplan via CHANNEL(tenantid), and it can be set using
Set(CHANNEL(tenantid)=My tenant ID). In pjsip.conf, it is recommended to
use the new tenantid option for pjsip endpoints (e.g., tenantid=My
tenant ID) so that it will show up in Newchannel events. You can set it
like any other channel variable using set_var in pjsip.conf as well, but
note that this will NOT show up in Newchannel events. Tenant ID is also
available in CDR and can be accessed with CDR(tenantid). The peer tenant
ID can also be accessed with CDR(peertenantid). CEL includes tenant ID
as well if it has been set.
UpgradeNote: A new versioned struct (ast_channel_initializers) has been
added that gets passed to __ast_channel_alloc_ap. The new function
ast_channel_alloc_with_initializers should be used when creating
channels that require the use of this struct. Currently the only value
in the struct is for tenantid, but now more fields can be added to the
struct as necessary rather than the __ast_channel_alloc_ap function. A
new option (tenantid) has been added to endpoints in pjsip.conf as well.
CEL has had its version bumped to include tenant ID.
Added Reload and DBdeltree to the list of dialplan application that
can't be executed via the Originate manager action without also
having write SYSTEM permissions.
Added CURL, DB*, FILE, ODBC and REALTIME* to the list of dialplan
functions that can't be executed via the Originate manager action
without also having write SYSTEM permissions.
If the Queue application is attempted to be run by the Originate
manager action and an AGI parameter is specified in the app data,
it'll be rejected unless the manager user has either the AGI or
SYSTEM permissions.
Resolves: #GHSA-c4cg-9275-6w44
Previously, on command execution, the control thread was awoken by
sending a SIGURG. It was found that this still resulted in some
instances where the thread was not immediately awoken.
This change instead sends a null frame to awaken the control thread,
which awakens the thread more consistently.
Resolves: #801
* Fixed a bug in crypto_show_cli_store that was causing asterisk
to crash if there were certificate revocation lists in the
verification certificate store. We're also now prefixing
certificates with "Cert:" and CRLs with "CRL:" to distinguish them
in the list.
* Added 'untrusted_cert_file' and 'untrusted_cert_path' options
to both verification and profile objects. If you have CRLs that
are signed by a different CA than the incoming X5U certificate
(indirect CRL), you'll need to provide the certificate of the
CRL signer here. Thse will show up as 'Untrusted" when showing
the verification or profile objects.
* Fixed loading of crl_path. The OpenSSL API we were using to
load CRLs won't actually load them from a directory, only a file.
We now scan the directory ourselves and load the files one-by-one.
* Fixed the verification flags being set on the certificate store.
- Removed the CRL_CHECK_ALL flag as this was causing all certificates
to be checked for CRL extensions and failing to verify the cert if
there was none. This basically caused all certs to fail when a CRL
was provided via crl_file or crl_path.
- Added the EXTENDED_CRL_SUPPORT flag as it is required to handle
indirect CRLs.
* Added a new CLI command...
`stir_shaken verify certificate_file <certificate_file> [ <profile> ]`
which will assist troubleshooting certificate problems by allowing
the user to manually verify a certificate file against either the
global verification certificate store or the store for a specific
profile.
* Updated the XML documentation and the sample config file.
Resolves: #809
The way we have been initializing the config wizard prevented it
from registering its objects if res_pjsip happened to load
before it.
* We now use the object_type_registered sorcery observer to kick
things off instead of the wizard_mapped observer.
* The load_module function now checks if res_pjsip has been loaded
already and if it was it fires the proper observers so the objects
load correctly.
Resolves: #816
UserNote: The res_pjsip_config_wizard.so module can now be reloaded.
softmix_bridge_write_control() now calls ast_bridge_queue_everyone_else()
with the bridge_channel so the VIDUPDATE control frame isn't echoed back.
softmix_bridge_write_control() was setting bridge_channel to NULL
when calling ast_bridge_queue_everyone_else() for VIDUPDATE control
frames. This was causing the frame to be echoed back to the
channel it came from. In certain cases, like when two channels or
bridges are being recorded, this can cause a ping-pong effect that
floods the system with VIDUPDATE control frames.
Resolves: #780
This commit adds a new voicemail.conf option 'odbc_audio_on_disk'
which when set causes the ODBC variant of app_voicemail to leave
the message and greeting audio files on disk and only store the
message metadata in the database. This option came from a concern
that the database could grow to large and cause remote access
and/or replication to become slow. In a clustering situation
with this option, all asterisk instances would share the same
database for the metadata and either use a shared filesystem
or other filesystem replication service much more suitable
for synchronizing files.
The changes to app_voicemail to implement this feature were actually
quite small but due to the complexity of the module, the actual
source code changes were greater. They fall into the following
categories:
* Tracing. The module is so complex that it was impossible to
figure out the path taken for various scenarios without the addition
of many SCOPE_ENTER, SCOPE_EXIT and ast_trace statements, even in
code that's not related to the functional change. Making this worse
was the fact that many "if" statements in this module didn't use
braces. Since the tracing macros add multiple statements, many "if"
statements had to be converted to use braces.
* Excessive use of PATH_MAX. Previous maintainers of this module
used PATH_MAX to allocate character arrays for filesystem paths
and SQL statements as though they cost nothing. In fact, PATH_MAX
is defined as 4096 bytes! Some functions had (and still have)
multiples of these. One function has 7. Given that the vast
majority of installations use the default spool directory path
`/var/spool/asterisk/voicemail`, the actual path length is usually
less than 80 bytes. That's over 4000 bytes wasted. It was the
same for SQL statement buffers. A 4K buffer for statement that
only needed 60 bytes. All of these PATH_MAX allocations in the
ODBC related code were changed to dynamically allocated buffers.
The rest will have to be addressed separately.
* Bug fixes. During the development of this feature, several
pre-existing ODBC related bugs were discovered and fixed. They
had to do with leaving orphaned files on disk, not preserving
original message ids when moving messages between folders,
not honoring the "formats" config parameter in certain circumstances,
etc.
UserNote: This commit adds a new voicemail.conf option
'odbc_audio_on_disk' which when set causes the ODBC variant of
app_voicemail_odbc to leave the message and greeting audio files
on disk and only store the message metadata in the database.
Much more information can be found in the voicemail.conf.sample
file.
The `Require: mediasec` and `Proxy-Require: mediasec` headers need
to be sent whenever we send `Security-Client` or `Security-Verify`
headers but the logic to do that was only in add_security_headers()
in res_pjsip_outbound_register. So while we were sending them on
REGISTER requests, we weren't sending them on INVITE requests.
This commit moves the logic to send the two headers out of
res_pjsip_outbound_register:add_security_headers() and into
security_agreement:ast_sip_add_security_headers(). This way
they're always sent when we send `Security-Client` or
`Security-Verify`.
Resolves: #789
Whenver a new channel snapshot is created or when a channel is
destroyed, we need to delete any existing channel snapshot from
the snapshot cache. Historically, we used the channel->snapshot
pointer to delete any existing snapshots but this has two issues.
First, if something (possibly ast_channel_internal_swap_snapshots)
sets channel->snapshot to NULL while there's still a snapshot in
the cache, we wouldn't be able to delete it and it would be orphaned
when the channel is destroyed. Since we use the cache to list
channels from the CLI, AMI and ARI, it would appear as though the
channel was still there when it wasn't.
Second, since there are actually two caches, one indexed by the
channel's uniqueid, and another indexed by the channel's name,
deleting from the caches by pointer requires a sequential search of
all of the hash table buckets in BOTH caches to find the matching
snapshots. Not very efficient.
So, we now delete from the caches using the channel's uniqueid
and name. This solves both issues.
This doesn't address how channel->snapshot might have been set
to NULL in the first place because although we have concrete
evidence that it's happening, we haven't been able to reproduce it.
Resolves: #783
Two functions are deprecated as of libxml2 2.12:
* xmlSubstituteEntitiesDefault
* xmlParseMemory
So we update those with supported API.
Additionally, `res_calendar_caldav` has been updated to use libxml2's
xmlreader API instead of the SAX2 API which has always felt a little
hacky (see deleted comment block in `res_calendar_caldav.c`).
The xmlreader API has been around since libxml2 2.5.0 which was
released in 2003.
Fixes#725
* Fixed possible memory leak in tn_config:tn_get_etn() where we
weren't releasing etn if tn or eprofile were null.
* We now canonicalize TNs before using them for lookups or adding
them to Identity headers.
* Fixed a typo in stir_shaken.conf.sample.
Resolves: #716
If you're tracing a large function that may call another function
multiple times in different circumstances, it can be difficult to
see from the trace output exactly which location that function
was called from. There's no good way to automatically determine
the calling location. SCOPE_CALL and SCOPE_CALL_WITH_RESULT
simply print out a trace line before and after the call.
The difference between SCOPE_CALL and SCOPE_CALL_WITH_RESULT is
that SCOPE_CALL ignores the function's return value (if any) where
SCOPE_CALL_WITH_RESULT allows you to specify the type of the
function's return value so it can be assigned to a variable.
SCOPE_CALL_WITH_INT_RESULT is just a wrapper for SCOPE_CALL_WITH_RESULT
and the "int" return type.
* OpenSSL 1.0.2 doesn't support X509_get0_pubkey so we now use
X509_get_pubkey. The difference is that X509_get_pubkey requires
the caller to free the EVP_PKEY themselves so we now let
RAII_VAR do that.
* OpenSSL 1.0.2 doesn't support upreffing an X509_STORE so we now
wrap it in an ao2 object.
* OpenSSL 1.0.2 doesn't support X509_STORE_get0_objects to get all
the certs from an X509_STORE and there's no easy way to polyfill
it so the CLI commands that list profiles will show a "not
supported" message instead of listing the certs in a store.
Resolves: #676
SQLAlchemy 2.0 changed the way that commits/rollbacks are handled
causing the final `UPDATE` to our `alembic_version_<whatever>` tables
to be rolled back instead of committed.
We now use one connection to determine which
`alembic_version_<whatever>` table to use and another to run the
actual migrations. This prevents the erroneous rollback.
This change is compatible with both SQLAlchemy 1.4 and 2.0.
It is possible for dialplan to result in an infinite
recursion of variable substitution, which eventually
leads to stack overflow. If we detect this, abort
substitution and log an error for the user to fix
the broken dialplan.
Resolves: #480
UpgradeNote: The maximum amount of dialplan recursion
using variable substitution (such as by using EVAL_EXTEN)
is capped at 15.
Send "RELOADING=1" instead of "RELOAD=1" to follow the format
expected by systemd (see sd_notify(3) man page).
Do not send STOPPING=1 in remote console mode:
attempting to execute "asterisk -rx" by the main process leads to
a warning if NotifyAccess=main (the default) or to a forced termination
if NotifyAccess=all.
If the hostname field of the ast_tcptls_session_args structure is
set (which it is for websocket client connections), that hostname
will now automatically be used in an SNI TLS extension in the client
hello.
Resolves: #713
UserNote: Secure websocket client connections now send SNI in
the TLS client hello.
Since DETECT_DEADLOCKS is now split from DEBUG_THREADS, it must
always be included in buildopts.h instead of only when
ADD_CFLAGS_TO_BUILDOPTS_H is defined. A SEGV will result otherwise.
Resolves: #719
rtp_engine.c and stun.c were calling ast_register_cleanup which
is skipped if any loadable module can't be cleanly unloaded
when asterisk shuts down. Since this will always be the case,
their cleanup functions never get run. In a practical sense
this makes no difference since asterisk is shutting down but if
you're in development mode and trying to use the leak sanitizer,
the leaks from both of those modules clutter up the output.
There were a few references in the embedded documentation XML
where the case didn't match or where the referenced app or function
simply didn't exist any more. These were causing 404 responses
in docs.asterisk.org.
