am: 37efc112e2 -s ours
am skip reason: change_id I004e78a33689e2208918d4316bcf9a8f50a0bac3 with SHA1 a5c68348d8 is in history
Change-Id: I11daaf075229a2891eb2d8b3df54a91a285d52fb
am: bc16cf3d72 -s ours
am skip reason: change_id I262ca41fe0225660551c9a421562405366b6acac with SHA1 06d81d2c62 is in history
Change-Id: I367d5afaa5ee254064d61f7f807af838b37583fe
Also make it extendable for mocking and update atoms.proto comments.
Bug: 131833599
Test: compiled
Change-Id: I6a0e8a6175e7e9e46572ea4aa50ce7669a02588b
In addition to the NetworkStack app monitoring, have PackageWatchdog
register an observer to NetworkStackClient to receive severe failure
notifications, and attempt a rollback if available.
The callback is registered in onPackagesReady(), which is called in the
boot sequence just before starting the NetworkStack.
Test: installed new networkstack, killed it twice, observe rollback
Test: unit test in change on top
Bug: 133725814
Change-Id: I2cb4200b78c2482cacc4bfe2ace1581b869be512
Instead of always crashing on userdebug builds, do the following on all
builds:
- If the device did not observe a NetworkStack crash in the last 6h
crash the system server. This is to handle spurious crashes of the
NetworkStack, so that the system can recover instead of staying
without connectivity.
- Otherwise, the device has had a recent crash. Notify the listeners
(watchdog listener to be added) that something is seriously wrong,
but do not crash to avoid bootlooping the device. This allows the
watchdog to do its job, and avoids bricking the device in
situations where the user could need to make emergency calls.
Bug: 133725814
Test: Killed the network stack, observe reboot of framework.
Test: Kill network stack again, observe nothing.
Test: Wipe data, install new network stack, kill it twice: rollback.
(with patch on top applied)
Change-Id: If1fa00bed769eb60ca4832609006bdf15ceddb80
1. Previously, getDnsNetId doesn't handle all the cases.
Fix it with cosidering bypass private DNS flag.
2. Make getDnsNetId return Network instead of netId,
and change name from getDnsNetId to getDnsNetwork
Bug: 129530368
Test: atest DnsResolverTest DnsUtilsTest
Merged-In: Ibb5080acd3c296650d56532fc7da525e9fa95e8f
(cherry picked from commit 3854966dc9)
Change-Id: I37353642088bcc17da0cf17f78a5ed9efc9aefc3
No internet notification may be prompted before partial
connectivity being detected. Partial connectivity status will
be set into NAI and prompted in the Setting. Behavior is not
aligned between Setting and notification. Thus, update
notification again if partial connectivity is detected.
Also, sliently show the updated notification if no internet
notification has already been shown to user to prevent alerting
user in short time.
Bug: 130683832
Test: Verified with simulated partial connectivity
Test: atest FrameworksNetTests
Change-Id: Ie16a8ce6e0fa437048e8c1eea240314ca30e9520
Merged-In: I004e78a33689e2208918d4316bcf9a8f50a0bac3
Merged-In: I14385a39d99a45c4a6a50a665f456f589c2f4da3
(cherry picked from commit a5c68348d8)
am: 709c46a60a -s ours
am skip reason: change_id I64b05adcb46baf3fca116dbdc5177bc31c71e06d with SHA1 c0274d1315 is in history
Change-Id: I8b637490b79bc202530e095644ad7e23c434e59f
Split-tunnel VPN (which are the only ones affected by this change)
always fall through to the default network for routes they don't
handle, and even if the underlying network(s) don't provide access
this may be a pinhole that can actually reach the broader network.
In practice this behaves like the original release of P and is the
safest thing to do for Q. In R we should evaluate giving the VPN
app the ability to simply tell the network stack whether it does
provide Internet access or not.
Bug: 119216095
Test: FrameworksNetTests NetworkStackTests
Change-Id: I262ca41fe0225660551c9a421562405366b6acac
Merged-In: I262ca41fe0225660551c9a421562405366b6acac
(cherry picked from commit 6d5a4a3b1d)
Add a check that callers with UID 1000 always have the same PID. This is
a proxy for checking that no system is designed to bind to the network
stack unless it is the system_server, as otherwise either the
system_server would start crashing, or that system would not have access
to binder calls.
Also remove access from PHONE_UID as it is not being used.
Test: Flashed, WiFi working, Bluetooth reverse tethering shows no
permission issue.
Bug: 133209255
Change-Id: Ib848aaaedfd599c1d4437378846c7dda74352019
No internet notification may be prompted before partial
connectivity being detected. Partial connectivity status will
be set into NAI and prompted in the Setting. Behavior is not
aligned between Setting and notification. Thus, update
notification again if partial connectivity is detected.
Also, sliently show the updated notification if no internet
notification has already been shown to user to prevent alerting
user in short time.
Bug: 130683832
Test: Verified with simulated partial connectivity
Test: atest FrameworksNetTests
Change-Id: Ie16a8ce6e0fa437048e8c1eea240314ca30e9520
System server | NetworkStack
|
NetworkMonitorCallbacks ←----|--- NetworkMonitorCallbacks$Stub$Proxy
↓ | ↑
NetworkAgentInfo | NetworkMonitor
↓ | ↑
NetworkMonitor$Stub$Proxy ----|---→ NetworkMonitorImpl
Bug: b/133174607
Test: Manual. The simplest artifact is observed by watching the output of
adb shell dumpsys meminfo -d com.android.networkstack | grep 'Proxy Binders'
while connecting and disconnecting multiple times to any network.
This will display the number of binder proxies. Before this, the binder
proxy count increases by 1 with each connection and never goes down (there
is some noise, as proxy objects are sometimes created for other reasons,
and get GC'd eventually). After this, the binder proxy count is always
eventually stable at 27 + connected network count.
See the bug for the complete analysis.
Change-Id: Ide2428dab3fcd6d7cd00aa2a9fd99d6c99b815a4
Split-tunnel VPN (which are the only ones affected by this change)
always fall through to the default network for routes they don't
handle, and even if the underlying network(s) don't provide access
this may be a pinhole that can actually reach the broader network.
In practice this behaves like the original release of P and is the
safest thing to do for Q. In R we should evaluate giving the VPN
app the ability to simply tell the network stack whether it does
provide Internet access or not.
Bug: 119216095
Test: FrameworksNetTests NetworkStackTests
Change-Id: I262ca41fe0225660551c9a421562405366b6acac
am: 8608b9e115 -s ours
am skip reason: change_id I43640d87c4fcd0a8211dc04c4203a7b80b150f58 with SHA1 5ec20d472e is in history
Change-Id: I0f0851a45797278eb8092242c2b5c7d5b453a9d8
1. Previously, getDnsNetId doesn't handle all the cases.
Fix it with cosidering bypass private DNS flag.
2. Make getDnsNetId return Network instead of netId,
and change name from getDnsNetId to getDnsNetwork
Bug: 129530368
Test: atest DnsResolverTest DnsUtilsTest
Change-Id: Ic8d45b802d72a266f9ec97bfea620c1ede6967e5
Captive portal overrides two shouldOverrideUrlLoading methods,
but show certificate logic only be implemented in deprecated one.
Bug: 133714666
Test: -build, flash, boot
-atest FramworksNetTests
-manual test
Change-Id: I2656bb709cb0bf259279ca75ad8411cd8c92fb43
Merged-In: I64b05adcb46baf3fca116dbdc5177bc31c71e06d
(cherry picked from commit 9161b5e845)
