Support adding NATT keepalive packet filter to APF
filter.
Generating APF program will be addressed in another CL.
Bug: 33530442
Test: - atest NetworkStackTests
- atest FrameworksNetTests
Change-Id: I4961d5da343d8700600269632787c28112f0f9f1
Mock out PackageManager and returns correct information corresponding
to the test app package itself.
Test: atest --generate-new-metrics 10 com.android.server.ConnectivityServiceTest
Bug: 114231106
Bug: 130397860
Change-Id: Ib921700eda417f411d7a2c77c1140fba9ab50bbb
Since socket keepalive APIs for UDP encapsulation sockets are
public to generic app. In order to ensure the given fd is valid,
this change verifies the resource id inside the UDP
encapsulation socket by using methods provided by IpSecService.
Bug: 125517194
Fix: 123968920
Test: 1. atest FrameworksNetTests --generate-new-metrics 10
2. atestcom.android.server.ConnectivityServiceTest \
#testNattSocketKeepalives --generate-new-metrics 100
Change-Id: I408aacc19b364683854d15a095c34e72389a6e5b
Currntly, keepalive slot is released when stop() is called. Next
starting keepalive can use the same slot number while previous
keepalive is still stopping. When the previous keepalive is
stopped, the incoming as will be processed by the new keepalive.
This change release keepalive slot after the result of stopping
has returned. Thus, newly created keepalive cannot allocate the
same slot number while lower layer is still processing stop event.
This change also disable flaky assertions that are caused by
test port has been occupied by other process.
Bug: 129512753
Test: 1. atest com.android.server.ConnectivityServiceTest \
#testNattSocketKeepalives --generate-new-metrics 100
2. atest FrameworksNetTests --generate-new-metrics 10
3. simulate the fail case manually.
Change-Id: I1991627545519ee5cb408a3df3a006f710f4af7b
A mocked PackageManager caused test failures in existing tests.
Revert that for now to make tests pass again.
Bug: 114231106
Bug: 130397860
Test: atest FrameworksNetTests
Change-Id: I4f181789152438f18e6cd2d235d76fabe3872ea3
This reverts commit 3897df2e36.
Reason for revert: Rolling forward, will fix tests in same CL stack.
Bug: 114231106
Bug: 130397860
Test: FrameworksNetTests
Change-Id: Ia8a0c99b4e1fd5dff26c881715cd876618ca4321
This reverts commit fd8f96d719.
This change does not have any topic: not reverting the other 2 commits in the original topic.
Reason for revert: broke FrameworksNetTests presubmit: b/130397860
Change-Id: Iff41d9fe97fafea44680c8d67d1ce19277548cc0
When a fully-routed VPN is running, we want to prevent normal apps
under the VPN from receiving packets originating from any local non-VPN
interfaces. This is achieved by using eBPF to create a per-UID input
interface whitelist and populate the whitelist such that all
non-bypassable apps under a VPN can only receive packets from the VPN's
TUN interface (and loopback implicitly)
This is the framework part of the change that build the whitelist.
The whitelist needs to be updated in the following cases:
* When a VPN is connected and disconnected
This will cover the change to allowBypass bit, since that can't be
changed without reconnecting.
* When a VPN's NetworkCapabilites is changed (whitelist/blacklist app changes)
* When a new app is installed
* When an existing app is removed
* When a VPN becomes fully-routed or is no longer fully-routed
New user/profile creation will automatically result in a whitelist app change
transition so it doesn't need to be handled specially here.
Due to the limitation of the kernel IPSec interacting with eBPF (sk_buf->ifindex
does not point to the virtual tunnel interface for kernel IPSec), the whitelist
will only apply to app VPNs but not legacy VPN connections, to prevent breaking
connectivity with kernel IPSec entirely.
Test: atest PermissionMonitorTest
Test: atest android.net.RouteInfoTest
Test: atest com.android.server.ConnectivityServiceTest
Test: atest HostsideVpnTests
Bug: 114231106
Change-Id: I143b03d60e46cb1b04732b4a4034f5847b4d1b1a
- Restrict unprivileged apps to use
NetworkRequest.Builder#setSignalStrength.
- Remove the "throws NullPointerException" in
CaptivePortalProbeSpec constructor.
- Remove the null check in LinkProperties.
- Add annotataion into all ConnectivityManager.NetworkCallback
methods.
Change-Id: Id275cac1d6a30d7515cd7b113394f5e8a0179314
Fix: 129097486
Test: atest FrameworksNetTests
NetworkMonitor obtained LinkProperties and NetworkCapabilities via
synchronous calls to ConnectivityManager after receiving an asynchronous
notification, which is prone to races: the network could be gone before
the LinkProperties/NetworkCapabilities can be fetched.
Fix the race by passing LinkProperties/NetworkCapabilities directly to
NetworkMonitor in the asynchronous notifications.
Test: atest FrameworksNetTests NetworkStackTests
Test: booted, WiFi works
Bug: 129375892
Change-Id: I200ac7ca6ff79590b11c9be705f650c92fd3cb63
Currently, socketKeepalive implementation is accepting null fd
due to backward compatibility with legacy packet keepalive API.
However, due to lack of the fd, the service cannot guarantee the
port is not reused by another app if the caller release the port
for any reason.
Thus, grant the null fd access only for priviledged apps.
This commit also address some comments from aosp/918533.
Bug: 126699232
Test: atest FrameworksNetTests
Change-Id: I0baf582ff4ca8af6082c3754e8dfbcd867f39792
There are 2 problems will make testPartialConnectivity flaky:
1. If we call setNetworkValid() before expectCapabilitiesWith(),
there may be a timing issue that network will become VALID before
NetworkMonitor send PARTIAL_CONNECTIVITY to ConnectivityService.
Solution:
We should set network to valid after ConnectivityService received
NETWORK_TEST_RESULT_PARTIAL_CONNECTIVITY to ensure NetworkMonitor
will send PARTIAL_CONNECTIVITY to ConnectivityService first then
send VALID.
2. When test case call explicitlySelected(true) first then call
connect(true), NetworkMonitor will report the network validation
test result twice because ConnectivityServiceTest() will trigger
notifyNetworkTested() when setAcceptPartialConnectivity() is
called, it may cause a timing that before the second test result
send to ConnectivityService, connect() already called
setNetworkInvalid. So, NET_CAPABILITY_VALIDATED will be removed
and ConnectivityService will trigger onCapabilitiesChanged()
unexpectedly.
Solution:
Don't trigger notifyNetworkTested() when
setAcceptPartialConnectivity() is called. If there is needed,
use mCm.reportNetworkConnectivity() to report the test result
instead.
Bug: 128426024
Test: 1. atest FrameworksNetTests: \
ConnectivityServiceTest#testPartialConnectivity \
--generate-new-metrics 1000
Change-Id: I7200528378201a3c7c09a78ff827b41f2741dfa1
Currently, the fails in testTcpSocketKeepalives are triggered by
fail() inside the executor, which is hiding the actual call trace
but only message remains. And it made the fail case hard to
debug.
So this commit is to bubble up the Exception by using a custom
functional interface.
Bug: 123987272
Test: 1. atest FrameworksNetTests
2. manually fail the test case and see the call trace
Change-Id: I125e673938a5e9d1de86f83c1a732227a4bd3207
Per API review, change the use of FileDescriptor to
ParcelFileDescriptor.
This change also fix nullability according to API review
feedbacks.
Fix: 126698610
Fix: 126699425
Fix: 126699232
Fix: 126700278
Test: 1. m -j
2. atest FrameworksNetTests --generate-new-metrics 50
3. m -j doc-comment-check-docs
Change-Id: I19476c50dd1ca290bf3f41973829da2bd229796a
With the @JavaOnlyStableParcelable annotation, Network can now be used
as-is as a framework parcelable.
Test: atest FrameworksNetTests NetworkStackTests
Bug: 126477266
Change-Id: I5d9695a31b3ac48d10edcf08ccd0c34375353b79
Currently, client side of keepalive event handling rely on a
newly created thread, looper, messenger and handler per object.
However, by creating oneway AIDL interface with the executor,
the callbacks can be invoked on the binder thread with user
specified context, which not only greatly simplify the design
but also reduce the cost of current thread modeling.
Bug: 114151147
Bug: 123969871
Test: 1. atest FrameworksNetTests --generate-new-metric 10
2. atest-deflake.sh
Change-Id: I27504074cd28d5b5eb94a7ec0e97ebaaaaa1ae3d
This is a cherry-pick of ag/607226 that has been rebased on
top of four years of changes and with comments addressed.
Gives each factory a serial number and propagates it to every
NetworkAgent so when a score comes back indicating a request is
being handled the factory can account for it properly.
Without this, a new request that's already handled by a network
offered by a factory will not cause an increment of the factorys
ref count. Concretely this results in issues like the RAT icon
not being displayed in spite of the network actually being up
and usable.
This will be ported to AOSP as soon as possible, but immediately
some master-only WiFi tests need to be adjusted with this change
which would not let me submit to AOSP.
Bug: 18637384
Bug: 29030667
Test: manual
Test: atest frameworks/opt/telephony/tests/telephonytests
Test: atest frameworks-net
Test: atest CtsNetTestCases CtsHostsideNetworkTests
Change-Id: I597ac588f76dd507512ff02868fd1310b7e63f7e
Merged-In: I597ac588f76dd507512ff02868fd1310b7e63f7e
In some networks, network validation may only get success
result for http probe but fail result for https probe.
For this kind of network, it may still work at some websites
or apps, but user didn't know about that. In order to fix this
issue, we will check if network has partial connectivity and
notify user to make a choice if they want to use this partial
connectivity or not.
Bug: 113450764
Test: 1. Build pass.
2. Fake partial connectivity case for testing.
3. atest FrameworksNetTests
4. atest NetworkStackTests
Change-Id: I69ed00ac4850904ff708c9fef22e148879a10e92
Add a mechanism by which a factory can declare "instant failure" for
a request - which would result in it getting an OnUnavailable()
(even without a timeout).
Factories may only do this iff:
1. They know they are the only factory which may fulfill this
request (common for transport-specific requests).
2. The know that the request can definitely not be
fulfilled at any point in the future.
Bug: 31382922
Test: atest ConnectivityServiceTest
Merged-In: I9bce0f4d85fa8cad7f8a9998819f945b778c5ac5
Change-Id: I9bce0f4d85fa8cad7f8a9998819f945b778c5ac5
NSS needed it for getting VpnInfo[], NetworkState[] and
activeLinkProperties which it used to query via ConnectivityManager.
For VpnInfo[], this was racy as NSS may ignore intermediate changes to a
VPN's underlying networks. See http://b/123961098 for more context.
It may also lead to deadlocks b/w ConnectivityService and
NetworkStatsService. See http://b/126245192 for more info.
This change will ensure that NSS is never contending on any of
ConnectivityService locks.
Bug: 123961098
Bug: 126245192
Bug: 120145746
Test: atest FrameworksNetTests
Change-Id: Id1da446b54d95ee68ed14079107b1a10318bcf8b
Merged-In: I57e117bb4e9efe491b19d6b5a479f2d58d1c58e6
This change is basically a revert of http://ag/3580901.
It was made because previously VPN capabilities did not use to update
based on its underlying networks. That is no longer the case anymore.
This was previously returning meteredness on the basis of VPN's first
underlying network which is incorrect in cases such as VPN using
multiple underlying networks, or VPN that has explicitly marked itself
as metered via VpnService.Builder#setMetered API.
Bug: 123936838
Test: atest FrameworksNetTests
Change-Id: Ia54b8570fbad4a638a6d43a95e0271c6baf66685
Currently NAT64 prefix discovery, which runs in netd, is started
by netd itself when a network is programmed with all-IPv6 DNS
servers. Unfortunately this is not correct because in many cases
we program DNS servers before the network is connected and it's
actually possible to send packets to them.
In general netd does not have enough visibility into network
lifecycle management to decide when to start and stop prefix
discovery. So move it into the framework with the rest of the
464xlat control plane.
Bug: 65674744
Test: atest FrameworksNetTests
Change-Id: I8fa051a9c216d9c05082bf7d0bbb0cbd56000162
Currently, both netd and clatd do NAT64 prefix detection, and we
start clatd on all IPv6-only networks regardless of whether netd
has detected a NAT64 prefix or not.
Instead, track the NAT64 prefix in the framework, and only start
clatd if the prefix is detected. This will allow us to remove
DNS64 detection in clatd, and pass the NAT64 prefix and the IPv6
address to clatd on the command line instead of clatd finding it
itself. That way, netd and the framework will always know how
464xlat is configured, and we'll be able to use that information
in netd.
Test: builds, boots
Test: atest FrameworksNetTests
Change-Id: Ida32d5760c5aecf7aeebef08fdb596291b2ce14a
1. Remove "throws Exception" from the signature so it can be used
in tests that do not already throw Exception.
2. Update comment to reflect the fact that TCP buffer sizes are
set on default network switch, not on connect.
3. Move into the method a class-level static constant that is
only used in that method.
Test: atest ConnectivityServiceTest
Change-Id: Ic2e4fbedc23065efc20e45ea84996b577a8e94b6
