The OPTIONS support in PJSIP has organically grown, like many things in
Asterisk. It has been tweaked, changed, and adapted based on situations
run into. Unfortunately this has taken its toll. Configuration file
based objects have poor performance and even dynamic ones aren't that
great.
This change scraps the existing code and starts fresh with new eyes. It
leverages all of the APIs made available such as sorcery observers and
serializers to provide a better implementation.
1. The state of contacts, AORs, and endpoints relevant to the qualify
process is maintained. This state can be updated by external forces (such
as a device registering/unregistering) and also the reload process. This
state also includes the association between endpoints and AORs.
2. AORs are scheduled and not contacts. This reduces the amount of work
spent juggling scheduled items.
3. Manipulation of which AORs are being qualified and the endpoint states
all occur within a serializer to reduce the conflict that can occur with
multiple threads attempting to modify things.
4. Operations regarding an AOR use a serializer specific to that AOR.
5. AORs and endpoint state act as state compositors. They take input
from lower level objects (contacts feed AORs, AORs feed endpoint state)
and determine if a sufficient enough change has occurred to be fed further
up the chain.
6. Realtime is supported by using observers to know when a contact has
been registered. If state does not exist for the associated AOR then it
is retrieved and becomes active as appropriate.
The end result of all of this is best shown with a configuration file of
3000 endpoints each with an AOR that has a static contact. In the old
code it would take over a minute to load and use all 8 of my cores. This
new code takes 2-3 seconds and barely touches the CPU even while dealing
with all of the OPTIONS requests.
ASTERISK-26806
Change-Id: I6a5ebbfca9001dfe933eaeac4d3babd8d2e6f082
Use of extended stringfields is a temporary mechanism to avoid ABI
breakage in released branches without resorting to more inconvienient
methods.
* Collect existing extended stringfields into the parent stringfield
section of the struct.
Change-Id: I8d46d037801b4518837c3ea4b6df95ceadc9436b
When a scheduled task is created you can pass in the
AST_SIP_SCHED_TASK_TRACK flag. This new flag causes scheduling events to
be logged.
Change-Id: I91967eb3d5a220915ce86881a28af772f9a7f56b
ast_sip_push_task_synchronous() did not necessarily execute the passed in
task under the specified serializer. If the current thread is any
registered pjsip thread then it would execute the task immediately instead
of under the specified serializer. Reentrancy issues could result if the
task does not execute with the right serializer.
The original reason ast_sip_push_task_synchronous() checked to see if the
current thread was a registered pjsip thread was because of a deadlock
with masquerades and the channel technology's fixup callback
(ASTERISK_22936). A subsequent masquerade deadlock fix (ASTERISK_24356)
involving call pickups avoided the original deadlock situation entirely.
The PJSIP channel technology's fixup callback no longer needed to call
ast_sip_push_task_synchronous().
However, there are a few places where this unexpected behavior is still
required to avoid deadlocks. The pjsip monitor thread executes callbacks
that do calls to ast_sip_push_task_synchronous() that would deadlock if
the task were actually pushed to the specified serializer. I ran into one
dealing with the pubsub subscriptions where an ao2 destructor called
ast_sip_push_task_synchronous().
* Split ast_sip_push_task_synchronous() into
ast_sip_push_task_wait_servant() and ast_sip_push_task_wait_serializer().
ast_sip_push_task_wait_servant() has the old behavior of
ast_sip_push_task_synchronous(). ast_sip_push_task_wait_serializer() has
the new behavior where the task is always executed by the specified
serializer or a picked serializer if one is not passed in. Both functions
behave the same if the current thread is not a SIP servant.
* Redirected ast_sip_push_task_synchronous() to
ast_sip_push_task_wait_servant() to preserve API for released branches.
ASTERISK_26806
Change-Id: Id040fa42c0e5972f4c8deef380921461d213b9f3
* Fix the periodic interval wander because it may take significant time
between the sched thread queueing the task in the serializer and the
serializer actually executing the task. The time it takes to actually
execute the task was already taken into account.
* Pass a schtd ref to the serializer when we queue a scheduled task on
the serializer. We don't want it going away on us while it is in the
serializer queue.
* Skip the scheduled task if the task was canceled between queueing the
task to the serializer and the serializer actually executing the task.
* Reorder struct ast_sip_sched_task to avoid unnecessary padding. Removed
task_id and added next_periodic.
* Hold a ref to the passed in serializer so the serializer cannot go away
on the scheduled task.
ASTERISK_26806
Change-Id: I6c8046b75f6953792c8c30e55b836a4291143f24
* A side benefit is that the scheduled tasks are not completely blocked
while the CLI command executes.
* Adjusted the "Task Name" column width to have more room for longer
names.
Change-Id: Iec64aa463ee8b10eef90120e00c38b1fb444087e
* Removed several invalid uses of OBJ_NOLOCK. These uses resulted in the
'tasks' container being accessed without a lock in a multi-threaded
environment. A recipe for crashes.
* Removed needlessly obtaining schtd object references. If the caller
providing you a pointer to an object doesn't have a valid reference then
you cannot safely get one from it.
* Getting a ref to 'tasks' when you aren't copying the pointer into
another location is useless. The 'tasks' container pointer is global.
* Removed many unnecessary uses of RAII_VAR.
* Make ast_sip_schedule_task() name parameter const.
ASTERISK_26806
Change-Id: I5c62488e651314e2a1dbc01f5b078a15512d73db
Apparently it is possible for the transport to be destroyed without
triggering the transport callback logic. As a result the transport gets
destroyed and we have a stale pointer in the active_transports container.
* Invoke the transport monitor callback checks when the transport is
destroyed in addition to when it is disconnected and shutdown.
ASTERISK-27688
Change-Id: Ia9b5469fea8f2b3f2d8476fae6b748a4d23e7261
The pool cache gets in the way of finding use after free errors of memory
pool contents. Tools like valgrind and MALLOC_DEBUG don't know when a
pool is released because it gets put into the cache instead of being
freed.
* Added the "cache_pools" option to pjproject.conf. Disabling the option
helps track down pool content mismanagement when using valgrind or
MALLOC_DEBUG. The cache gets in the way of determining if the pool
contents are used after free and who freed it.
To disable the pool caching simply disable the cache_pools option in
pjproject.conf and restart Asterisk.
Sample pjproject.conf setting:
[startup]
cache_pools=no
* Made current users of the caching pool factory initialization and
destruction calls call common routines to create and destroy cached pools.
ASTERISK-27704
Change-Id: I64d5befbaeed2532f93aa027a51eb52347d2b828
Since res_pjsip_transport_management provides several attack
mitigation features, its functionality moved to res_pjsip and
this module has been removed. This way the features will always
be available if res_pjsip is loaded.
ASTERISK-27618
Reported By: Sandro Gauci
Change-Id: I21a2d33d9dda001452ea040d350d7a075f9acf0d
pjsip_distributor:
authenticate() creates a tdata and uses it to send a challenge or
failure response. When pjsip_endpt_send_response2() succeeds, it
automatically decrements the tdata ref count but when it fails, it
doesn't. Since we weren't checking for a return status, we weren't
decrementing the count ourselves on error and were therefore leaking
tdatas.
res_pjsip_session:
session_reinvite_on_rx_request wasn't decrementing the ref count
if an error happened while sending a 491 response.
pre_session_setup wasn't decrementing the ref count if
while sending an error after a pjsip_inv_verify_request failure.
res_pjsip:
ast_sip_send_response wasn't decrementing the ref count on error.
ASTERISK-27618
Reported By: Sandro Gauci
Change-Id: Iab33a6c7b6fba96148ed465b690ba8534ac961bf
In an earlier release, inbound registrations on a reliable transport
were pruned on Asterisk restart since the TCP connection would have
been torn down and become unusable when Asterisk stopped. This same
process is now also applied to inbound subscriptions.
Also fixed issues in res_pjsip_registrar where it wasn't handling the
monitoring correctly when multiple registrations came in over the same
transport.
To accomplish this, the pjsip_transport_event feature needed to
be refactored to allow multiple monitors (multiple subcriptions or
registrations from the same endpoint) to exist on the same transport.
Since this changed the API, any external modules that may have used the
transport monitor feature (highly unlikey) will need to be changed.
ASTERISK-27612
Reported by: Ross Beer
Change-Id: Iee87cf4eb9b7b2b93d5739a72af52d6ca8fbbe36
This removes references that are no longer needed due to automatic
references created by module dependencies.
In addition this removes most calls to ast_module_check as they were
checking modules which are listed as dependencies.
Change-Id: I332a6e8383d4c72c8e89d988a184ab8320c4872e
I've audited all modules that include any header which includes
asterisk/optional_api.h. All modules which use OPTIONAL_API now declare
those dependencies in AST_MODULE_INFO using requires or optional_modules
as appropriate.
In addition ARI dependency declarations have been reworked. Instead of
declaring additional required modules in res/ari/resource_*.c we now add
them to an optional array "requiresModules" in api-docs for each module.
This allows the AST_MODULE_INFO dependencies to include those missing
modules.
Change-Id: Ia0c70571f5566784f63605e78e1ceccb4f79c606
The type=identify endpoint identification method can match by IP address
and by SIP header. However, the SIP header matching has limited
usefulness because you cannot specify the SIP header matching priority
relative to the IP address matching. All the matching happens at the same
priority and the order of evaluating the identify sections is
indeterminate. e.g., If you had two type=identify sections where one
matches by IP address for endpoint alice and the other matches by SIP
header for endpoint bob then you couldn't predict which endpoint is
matched when a request comes in that matches both.
* Extract the SIP header matching criteria into its own "header" endpoint
identification method so the user can specify the relative priority of the
SIP header and the IP address matching criteria in the global
endpoint_identifier_order option. The "ip" endpoint identification method
now only matches by IP address.
ASTERISK-27491
Change-Id: I9df142a575b7e1e3471b7cda5d3ea156cef08095
We did this for TCP transports already but I'm not sure why we
didn't do it for TLS transports.
ASTERISK_27474 #not_final_fix
Change-Id: I5b1ef4b882f7b859e718236686b7898751dbb262
* Extracted sip_endpoint_identifier_type2str() and
sip_endpoint_identifier_str2type() to simplify the calling functions.
* Fixed pjsip_configuration.c:ident_to_str() building the endpoint's
identify_by value string.
Change-Id: Ide876768a8d5d828b12052e2a75008b0563fc509
Those SIP messages that create dialogs require a contact header to be present.
If the contact header was missing from the message it could cause Asterisk to
crash.
This patch checks to make sure SIP messages that create a dialog contain the
contact header. If the message does not and it is required Asterisk now returns
a "400 Missing Contact header" response. Also added NULL checks when retrieving
the contact header that were missing as a "just in case".
ASTERISK-27480 #close
Change-Id: I1810db87683fc637a9e3e1384a746037fec20afe
Fix instances of:
* Retreive
* Recieve
* other then
* different then
* Repeated words ("the the", "an an", "and and", etc).
* othterwise, teh
ASTERISK-24198 #close
Change-Id: I3809a9c113b92fd9d0d9f9bac98e9c66dc8b2d31
Support for these protocols was added in the same commit as the 'proto'
field, so we can safely use the same ./configure check.
For reference: https://trac.pjsip.org/repos/changeset/4968
Change-Id: Icf4975d785d6bfb8f30ac7ffa695a0adf9382dac
A couple of places were setting the status to "UNKNOWN" when qualifies were
being disabled. Instead this should be set to the "CREATED" status that
represents when a contact is given (uri available), but the qualify frequency
is set to zero so we don't know the status.
This patch updates the relevant places with "CREATED". It also updates the
"CREATED" status description (value shown in CLI/AMI/ARI output) to a value
of "NonQualified"/"NonQual" as this description is hopefully less confusing.
ASTERISK-27467
Change-Id: Id67509d25df92a72eb3683720ad2a95a27b50c89
Use the new ast_cli_completion_add() function to improve completion
performance for commands like 'pjsip show endpoint.'
Change-Id: I76d802294d2ac1766110dc75f7d117c8541ce348
For both dynamic and static contacts it was possible that potential AOR
changes were not being applied to all contacts. This was because the qualify
and schedule code was only retrieving AOR's, and contacts with frequencies
greater than zero.
For instance the following could happen: and AOR/contact has a frequency of 5,
it then gets set to 0, and then a reload occurs. All scheduled OPTIONS are
stopped, a list of AOR's is retrieved with frequency > 0, but none are
selected since in this scenario all are 0. The contact for the one previously
set to 5 though does not get updated, so it's status remains "AVAILABLE".
This patch makes it so all contacts (static and dynamic) are selected, and
appropriately updated if need be.
ASTERISK-27467 #close
Change-Id: I7a920170f89c683af9505d4723a44fc6841decdb
Dynamic contacts were not being properly updated on reload. As a matter of
fact any changes to the AOR that a dynamic contact was associated with were
not being applied.
On reload, this patch makes it so for each dynamic contact, the associated
AOR is now retrieved and the AOR's fields are applied to the contact.
ASTERISK-27467
Change-Id: I8e3165dc6a745218c1c9db837f77fafa0516985d
The SuccessfulAuth using_password field was declared as a pointer to a
uint32_t when the field was later read as a uint32_t value. This resulted
in unnecessary casts and a non-portable field value reinterpret in
main/security_events.c:add_json_object(). i.e., It would work on a 32 bit
architecture but not on a 64 bit big endian architecture.
Change-Id: Ia08bc797613a62f07e5473425f9ccd8d77c80935
Mac doesn't like the comparison of -1 to an enum, so store the result of
ast_sip_str_to_dtmf to an int so we can check for the negative return
value. ast_sip_str_to_dtmf returns an int so this is only delaying the
implicit type cast.
Change-Id: I0c262c1719ee951aae1f437d733a301cf5f8ad29
