NETWORK_STACK_UID check was introduced by aosp/1134635, and it
assumed that wifi factories/agents will be running on the network
stack process, but it didn't eventually. It runs in the system
server instead. So remove this check.
Bug: 170598012
Test: atest FrameworksNetTests
Change-Id: Ie11268738e364e252f18876f7a4a0dc064ca8b9c
Connectivity is becoming a mainline module in S but mainline
modules are not allowed to use non-formal APIs. Thus, replace
internal okhttp APIs to stable libcore APIs which are created for
using HttpURLConnectionFactory.
Bug: 182238821
Test: atest FrameworksNetTests
Change-Id: I56ba1b9e6e94f9c6519c3f1c8f0c5993fccbe185
* changes:
Using jniThrowErrnoException in android_util_Process
Using jniThrowErrnoException in android_os_SharedMemory
Using jniThrowErrnoException in android_net_NetworkUtils
As attested by numerous TODOs in the code, a new way of
representing network quality and policy is needed instead
of an int.
An int representing the quality of the network requires
all parties using it to know how all other parties are
using it, and implementation details about the decision
algorithm. For all intents and purposes, the selection
is left to individual network factories who try to
achieve a desired result while piecing together all
possible states of the system.
As the number of such cases and desires increases, this
becomes both intractable and unmaintainable. Indeed, at
this time in the codebase nobody can really predict exactly
how a given change in score will affect selection across
the board, and it is essentially impossible to figure out
the behavior of network selection by inspecting the code
because the moving parts are scattered throughout the
entire codebase.
Having an object encapsulating policy and quality values
will let us centralize the selection and make it again
possible to maintain without knowledge of all behaviors
of all network factories. It will also provide better
guarantees of respecting policy, and allow bugfixes that
were not possible before because they'd touch too many
parts of the code.
Test: FrameworksNetTests FrameworksWifiTests NetworkStackTests
Change-Id: I3185a6412b9b659798faf0c6882699e9c63cc115
PacProxyInstaller class is running a thread all the time and is
listening to intent ACTION_PAC_REFRESH so it would be better to
make it be a system service with a manager class PacProxyManager
which is obtained with getSystemService(PacProxyManager.class).
Besides, rename PacProxyInstaller to PacProxyService will
be easier to know it's the service for PacProxyManager.
ConnectivityService is going to be a mainline module and it
needs constructor of PacProxyService to be SystemApi.
However, in current design, it needs to pass a handler and
an int arguments to the constructor which would be difficult
to maintain if just expose the constructor directly.
So, define a listener for the event that the current PAC
proxy has been installed so that the handler and the int
arguments can be removed from the constructor.
Bug: 177035719
Test: FrameworksNetTests
Change-Id: I2abff75ec59a17628ef006aad348c53fadbed076
ConnectivityService is going to become a mainline module, and
it will not able to use hidden method anymore. Using PlatformProperties
as a static library instead of hidden API.
Bug: 170917042
Test: atest FrameworksNetTests
Merged-In: I3a3deca5d2e0f690db8c0061de2db2217376d268
Change-Id: I3a3deca5d2e0f690db8c0061de2db2217376d268
The caller to create a TestNetworkService via
ConnectivityManager#startOrGetTestNetworkManager should have
passed the MANAGE_TEST_NETWORKS permission check and the caller
doesn't need to declare the permission to pass the permission
check of registerNetworkProvider. So call clearCallingIdentity
when calling registerNetworkProvider in TestNetworkService.
Bug: 181573283
Test: atest CtsNetTestCases:ConnectivityManagerTest#testRequestBackgroundNetwork
Change-Id: Ia28627dacf933d1937978ed9709b975c9a4660ce
