BpfNetMapsReader is loaded to bootclasspath. So this class should not
contain unnecessary codes.
To make this clear, this CL renames to NetowrkStackBpfNetMaps and add
comments
Bug: 328732146
Test: TH
Change-Id: I557627c9fae087eb6119c2ae31d46526fa9e5d7e
netd.o bpf program was mainlined in T
Test: TreeHugger
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I6d2b34e26f5640840b74d4c761b01668e44dc9d6
Maciej Żenczykowskicommited 2024-01-25 15:15:26 -08:00
we can't use java for this, because pre-U-QPR2 does not include:
https://android-review.googlesource.com/c/platform/system/sepolicy/+/2821590
system_server dontaudit key_socket getopt
so we'll spew lots of:
11-08 07:52:43.776 1469 1469 I auditd : type=1400 audit(0.0:4): avc: denied { getopt } for comm="system_server" scontext=u:r:system_server:s0 tcontext=u:r:system_server:s0 tclass=key_socket permissive=0
11-08 07:52:44.360 1469 1469 I auditd : type=1400 audit(0.0:5): avc: denied { getopt } for comm="NetworkStats" scontext=u:r:system_server:s0 tcontext=u:r:system_server:s0 tclass=key_socket permissive=0
11-08 07:52:44.508 1469 1469 I auditd : type=1400 audit(0.0:7): avc: denied { getopt } for comm="android.bg" scontext=u:r:system_server:s0 tcontext=u:r:system_server:s0 tclass=key_socket permissive=0
and the like.
This is due to Java's Os.close() in:
libcore/luni/src/main/java/libcore/io/BlockGuardOs.java;l=100
calling:
if (fd.isSocket$()) if (isLingerSocket(fd)) ...
Test: TreeHugger
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ib87fa5e8f0dfd9fbbccb25fb58a9fa78d6a07111
Maciej Żenczykowskicommited 2023-11-09 11:08:58 +00:00
skDestroyListener is always running with Java code so removed the
unnecessary check.
Bug: 217624062
Test: TH
Change-Id: Ie1a36c3c736005ec4821c5e9c9983998f7042afa
BpfNetMaps uses java library for updating bpf map as the default code
path.
This is already released to 100% production without any issues.
So this CL removes the old code path from BpfNetMaps.
Note that native_init only opens the bpf maps if sEnableJavaBpfMap is
true.
But if sEnableJavaBpfMap is true, no native code in TrafficController
access the bpf map.
So this CL can remove native_init.
Following CL will remove the nativce codes for old code path.
Bug: 217624062
Test: NetworkStaticLibsTests
Change-Id: Ifba198d9e6f93b53fd1dbf3b2aafb644da0b147d
This is needed for data stall detection mechanism in NetworkStack
to get the information about whether the network is blocked for
a given uid and conditions. Because the API will be called
frequently from NetworkStack to resolve all status for all uids
on the device, the API cannot call into the service which
creates IPC. Instead, the API need to directly access bpf maps
in the user process to retrieve the status. In this case the
user process is the network stack, the access control is provided
by linux file permission and selinux.
Test: atest FrameworksNetTests:android.net.connectivity.android.net.BpfNetMapsReaderTest
Test: atest FrameworksNetTests:android.net.connectivity.android.net.ConnectivityManagerTest
NO_IFTTT=Refactor only change for firewall chains definitions
Bug: 297836825
Change-Id: Iaf983b71ec98cbfe5152dcfade8a3120f938f135
The information is needed by modules who want to know whether a
specific UID is blocked by Data Saver feature.
1. Add a one-element map data_saver_enabled_map.
2. Update current data saver setting to the map.
Bug: 288340533
Test: atest FrameworksNetTests:android.net.connectivity.com.android.serv
er.BpfNetMapsTest
Test: atest bpf_existence_test
Change-Id: I981da4b569247c33cba2d365cb6f2691f673474e
A helper class to *read* java BpfMaps. This is designed to
provide direct bpf access in the caller process through
ConnectivityManager APIs.
The change also removes any statical link to
net-utils-device-common-struct from service-connectivity.
This is because net-utils-device-common-struct is already
included in framework-connectivity. Including it again in
service-connectivity would create a r8 build fail by circular
dependency.
Test: atest FrameworksNetTests:android.net.connectivity.com.android.server.BpfNetMapsTest
Test: atest ConnectivityCoverageTests:android.net.connectivity.com.android.net.module.util.StructTest
Test: atest FrameworksNetTests:android.net.connectivity.android.net.BpfNetMapsReaderTest
Bug: 297836825
Change-Id: I7a6d2eb816d0dc7343167bddd672806b199f44fe
If the flag value is unset or 0, isFeatureEnabled return false and
isFeatureNotChickenedOut return true.
If the flag value is -1 (force disable), both return false.
If the flag value is other values, both query the package version and
return `flagValue <= packageVersion`
Now the only difference is a default behavior when the flag is not set
or 0.
So isFeatureEnabled and isFeatureNotChickenedOut can use the same flag.
This CL also fixes the issue in rollback.
Before this CL, isFeatureNotChickenedOut did not check the module
version and could have a issue if there is a rollback.
Test: NetworkStaticLibsTests
Bug: 279108992
Change-Id: I12d6ebadff3aee7b7c614aca4eb0a34ef0db9857
This is a no-op refactoring that splits constants and utility
methods into standalone classes which will be shared with
a bpf reader class in subsequent CLs.
NO_IFTTT=No-op refactoring
Test: atest FrameworksNetTests:android.net.connectivity.com.android.server.BpfNetMapsTest
Test: atest ConnectivityCoverageTests:android.net.connectivity.com.android.net.module.util.StructTest
Bug: 297836825
Change-Id: I6d7ea044e43180ae001573009a166be74ebe6a5d
Experiment to verify bpf map operations refactoring showed no
change of map size metrics and no system health regressions.
So this CL update BpfNetMaps to always use new code path with kill
switch.
Test: m
Bug: 217624062
Change-Id: Ic17470ac5c075884871ea2df2c13c054f5ac2929
Build/TestCop - revert this if you see failures (just document where).
It erroneously(?) doesn't appear to cause any failures in TH
on aosp/master.
Bug: 291519274
Change-Id: I78b049302ae9bd2a8db8c03e79388dc4e51e110f
Maciej Żenczykowskicommited 2023-07-18 20:09:20 +00:00
From service-connectivity, NAMESPACE_TETHERING must be used, as
NAMESPACE_CONNECTIVITY is used for the NetworkStack module. Also the
isFeatureEnabled with module name must be used, otherwise the "android"
package version is queried instead of the tethering module version.
BpfNetMaps is using flag value 1 to workaround this issue.
flags in NsdService, AutomaticOnOffKeepaliveTracker, and ConnectivityFlags
were not pushed currently.
So, this change has no effect to devices.
Bug: 279108992
Test: TH
Change-Id: I2b4b6a13c048c20beef52b1f43b9e21aca0c637a
Upcoming CLs use this method to add chain status to failure log of
ConnectivityManagerTest#testFirewallBlocking which is a flaky test
Test: atest BpfNetMapsTest
Bug: 262141231
Change-Id: Ib130a28c708f675c79b889bf293dfe6b0f8d7f60
Having SdkLevel.isAtLeastU() as a default value of
DeviceConfigUtils.isFeatureEnabled could result confusing behavior.
For example, when flag value is pushed to only the T device and the T
device with the flag is ugraded to U, this device could disable java bpf
map while the device keeps the flag value.
This CL prevents above confusing behavior.
Bug: 246218229
Test: atest BpfNetMaps
Change-Id: I615603c337bb94101dccac77ea0e6fa857041259
Information in the dump does not change
....
current ownerMatch configuration: 516 DOZABLE_MATCH OEM_DENY_1_MATCH
....
Bug: 217624062
Test: dumpsys connectivity trafficcontroller, atest BpfNetMapsTest
Change-Id: I5f7b56e4ba03256414f49d0e82d65477fb97f05a
Information in the dump does not change
....
current statsMap configuration: 0 SELECT_MAP_A
....
Bug: 217624062
Test: dumpsys connectivity trafficcontroller, atest BpfNetMapsTest
Change-Id: I42a844227f258b91ba5b368d8e8abf82a773a801
Also remove the dump of mPrivilegedUser because this is not used now and
the information is duplicated with uidPermissionMap dump.
Before this CL
....
mUidPermissionMap:
10049 PERMISSION_NONE
10027 BPF_PERMISSION_INTERNET BPF_PERMISSION_UPDATE_DEVICE_STATS
1041 BPF_PERMISSION_UPDATE_DEVICE_STATS
....
After this CL
....
sUidPermissionMap:
10049 PERMISSION_NONE
10027 PERMISSION_INTERNET PERMISSION_UPDATE_DEVICE_STATS
1041 PERMISSION_UPDATE_DEVICE_STATS
....
Bug: 217624062
Test: dumpsys connectivity trafficcontroller, atest BpfNetMapsTest
Change-Id: I416501d5a28a5443f954f9c8c58ea255e5cdc897
This converts the key, which is an index in an array, of sConfigurationMap from U32 to S32.
Since the array has exactly 2 elements, there's no real difference.
In addition while the kernel does indeed index arrays with U32s,
it doesn't matter, because you'd need to have over 2 billion
elements in the bpf map (at which point you'd need GBs of ram for
the bpf map itself)...
Test: TreeHugger
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: If17efc25d910af642e519f9ecb6d0e8695c38d42
Maciej Żenczykowskicommited 2022-09-17 06:16:34 +00:00
This converts the key, which is a uid, of sUidOwnerMap & sUidPermissionMap from U32 to S32.
The kernel is actually not consistent in whether uids & gids
are signed or unsigned, and neither is our Java code, which
also commonly uses just 'int' for uid. In practice values
greater or equal to 2**31 often don't quite work right.
For example icmp sockets are enabled via a sysctl that
takes a minimum and maximum gid - and these are signed int32s.
It's worth pointing out that these functions already accept int's
as arguments, so when converting them to U32 (and thus long) they
technically previously did the wrong thing (though it's unlikely
it ever mattered).
Test: TreeHugger
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I7b389a1cf822c56134b20df5ded269963f5bf69f
Maciej Żenczykowskicommited 2022-09-17 06:06:23 +00:00
These are allocated in order by the kernel, if we go
over 2 billion, we've got other problems... besides
U32 to S32 conversion will work just fine anyway.
Test: TreeHugger
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I6e1da81c03bd36d8c6e7bd2b33f535d1fabfad0a
Maciej Żenczykowskicommited 2022-09-16 18:34:39 +00:00
This is a preparation for the upcoming CL that will add the test for
BpfNetMaps#dump.
Bug: 217624062
Test: dumpsys connectivity trafficcontroller, atest BpfNetMapsTest
Change-Id: I79aa6f6153dc94f806411c642d72f4801915a588
Upcoming CL will update BpfNetMaps to pull NetworkBpfMapInfo atom that
contains the cookieTagMap size field.
Test: m, atest BpfNetMaps
Bug: 217624062
Change-Id: I84bb0aa56e3916fc6fc0ca31ba5962f3b9be4b6f
Switch existing C SkDestroyListener and new Java SkDestroyListener based
on the experiment flag.
Bug: 217624062
Test: atest SkDestroyListenerTest
Change-Id: I7031ba7c155edb43e0e7b77449587058d6ba38fb
This CL prepares for upcoming CL.
Upcoming CL will add SkDestroyListener with Java BpfMap and switch
current C SkDestroyListener and new Java SkDestroyListener based on the
experiment flag.
Bug: 217624062
Test: atest SkDestroyListenerTest
Change-Id: I5ebb319d1b2262199d4ef6a3549894fee24c4ccf
Currently, production code uses BpfMap type and TestBpfMap extends
BpfMap but this makes it diffcult to test because TestBpfMap loads the
JNI.
This CL updates to use IBpfMap type in the production code.
Upcoming CL updates TestBpfMap to implement IBpfMap instead of extending
BpfMap so that test can still use TestBpfMap but tests do not need to
load JNI.
Bug: 217624062
Test: atest BpfCoordinatorTest BpfNetMapsTest
Change-Id: Ie67e14bf5519fb4427474ecc0fda441877a9555f
Motomu Utsumi
commited 
Maciej Żenczykowski
commited 
Gerrit Code Review
commited 
Junyu Lai
commited 
Ken Chen
commited