This patch groups connect() events per netId. It adds netid and
transport information to serialized ConnectStatistics events.
Test: updated NetdEventListenerServiceTest
updated IpConnectivityMetricsTest
$ runtest frameworks-net passes
Bug: 34901696
Change-Id: Id0d536ff723ded5c26eafe0bb138ba75ba2856c5
Merged-In: I4769496383943e714a1d350c298e093c2ed57477
(cherry picked from commit dfc2cc5857)
This patch changes how DnsEvents are logged in IpConnectivityMetrics.
The following changes are made:
- DnsEventBatch are not logged after 100 queries on the same network
- this allows to merge DnsEvent and DnsEventBatch into one class
- DnsEventBatch are not logged after a network disconnect
- this allows to remove the NetworkCallback
- DnsEvent are now logged similarly to ConnectStats when statistics are
flushed, in a direct call from IpConnectivityMetrics into
NetdEventListenerService, in a direct call from IpConnectivityMetrics
into NetdEventListenerService.
- this allows to remove the Parcelable implementation of DnsEvent
- transports information is added to DnsEvent.
Test: - simplified NetdEventListenerServiceTest covering dns logging
- updated IpConnectivityEventBuilderTest
- updated IpConnectivityMetricsTest
- $ runtest frameworks-net passes
- manually verified $ adb shell dumpsys connmetrics list proto
Bug: 34901696
Change-Id: I4fcd0ad7a7b85d587647f471a90c1e53a18fc95a
Merged-In: Ia4b33fd4212741152662a2adbb0533bd1b4902ee
(cherry picked from commit 0699cf9804)
This patch also
- partially reverts commit f927f0c52e
that exposed a getTransports method on NetworkCapabilities.
- moves enumerateBits to BitUtils (as unpackBits), and adds the
reverse packBit method.
Bug: 34901696
Test: manually looked at $ adb shell dumpsys connmetrics list
Change-Id: I1650daf8fc9c1b6e0d986d2285f81e888be8847f
Merged-In: Id04f9080e7f75608deeb49306aec34941e71794c
(cherry picked from commit df456e13a1)
Documented the requirements for becoming a network recommendation
provider.
Test: Built
Bug: 33632378
Change-Id: I8ec037c8688b250514cbe25a13434c7b8bef8327
Because there is no way using the Java sockets API to actually
get a socket of AF_INET on mode machines, it is necessary to
provide a way to apply transforms to sockets made using the
native wrapper API, which uses POSIX APIs and will create a
socket that is AF_INET.
Bug: 36073210
Test: b/34811227
Change-Id: I28ac7cc4f36045ce523a54111e5be975b0331356
-Add a reserveSecurityParamterIndex() function that allows the
system to select an SPI.
-Disallow INVALID_SECURITY_PARAMETER_INDEX from being passed as
an explicit SPI request.
-Remove the ALGO_ prefix from constants in IpSecAlgorithm
Bug: 36073210
Test: Updated CTS tests still pass on bullhead
Change-Id: Ic94809996076b0718f153f550b82192fe7048a2e
This allows an application that knows how to provide seamless
network connectivity (e.g., using QUIC multipath) to find out if
doing so is desired.
(cherry picked from commit 2de4925f5c)
Test: builds, boots, runtest frameworks-net passes.
Bug: 34630278
Change-Id: Ic7fd0b9e1cd879fdfaf84009d7125391895e9087
API visibility change: unhide allowing NetworkSpecifier
to be an arbitrary object.
Bug: 27533960
Bug: 36053921
Bug: 36275276
Test: builds and runs
Change-Id: I1d1705cca7ece077ef8d7c674c62d5369fedbb03
-Remove Int-based SPI usage from the IpSecTransform.Builder
This is essentially a less-safe method overload, and it is both
unnecessary and difficult to implement: the cross-validation
between SPI and Transform is actually useful, and the kernel
requires two different mechanisms to use an unreserved vs a
reserved (alloc'd) SPI: CREATESA vs UPDATESA, which makes this
hard to support. API Council has questioned the value of this,
and they are right: everything points to "remove this". In the
future, if we find that SPI reservation is overhead, we can
always add it back.
-Hiding the TunnelMode builder method and application/remove
methods. These will not land by the time the next API
stabilizes, so better to hide them now that this is a
near-certainty. Expectation is to un-hide them in the subsequent
API bump.
Bug: 36073210
Test: Compilation, verified nobody is calling these stubs
Change-Id: Ic1a3f2cf7128633318ac175d6b56b45eb8d21cab
(cherry picked from commit 48b566557d)
To make the SPI reservation more semantically consistent with the
transform creation API, and to ensure that we always create SPI
reservations relative to a well-known remote, we should take the
SPI request relative to a remote (rather than to a destination).
This necessitates that we now consider direction separately, which
is used for keying the SA-Id.
Bug: 36073210
Test: compilation
Change-Id: I81e955c20128c1f8e04fd68eb26669561f827a78
(cherry picked from commit c4f879925b)
-Add IpSecService with the necessary glue to connect to netd
-Add code to retrieve IpSecService from System Server
Bug: 30984788
Test: b/34812052, b/34811227
Change-Id: I4cdcb643421141202f77a0e2f87a37012de0cd92
(cherry picked from commit 28084d89ec)
This patch adds basic logging to NsdManager and NsdService, and improves
the facilities for pretty printing the event ids defined in NsdManager.
It also includes a few minor cleanups:
- adding 'final' on effectively final instance variables of NsdManager
and NsdService.
- similarly, adding 'static' on effectively static class fields.
- regrouping instance variables together.
Test: no functional changes
Bug: 33074219
Change-Id: I360d539e73cc8e4b45d4e0d20b2e345455fdb10c
-Plumb IpSecManager APIs to NetD
-Add Resource Management to IpSecService
Bug: 33695893
Test: CTS verifies nearly all of these paths
Change-Id: Ic43965c6158f28cac53810adbf5cf50d2c54f920
-Remove Int-based SPI usage from the IpSecTransform.Builder
This is essentially a less-safe method overload, and it is both
unnecessary and difficult to implement: the cross-validation
between SPI and Transform is actually useful, and the kernel
requires two different mechanisms to use an unreserved vs a
reserved (alloc'd) SPI: CREATESA vs UPDATESA, which makes this
hard to support. API Council has questioned the value of this,
and they are right: everything points to "remove this". In the
future, if we find that SPI reservation is overhead, we can
always add it back.
-Hiding the TunnelMode builder method and application/remove
methods. These will not land by the time the next API
stabilizes, so better to hide them now that this is a
near-certainty. Expectation is to un-hide them in the subsequent
API bump.
Bug: 36073210
Test: Compilation, verified nobody is calling these stubs
Change-Id: Ic1a3f2cf7128633318ac175d6b56b45eb8d21cab
This patch removes from ConnectivityService the logic involved in
deciding if a uid has access to networking based on networking policies.
This logic is moved into NetworkPolicyManagerService which is the source
of truth with regards to the state of networking policie, both for
existing networks and uids.
Instead ConnectivityService directly queries NetworkPolicyManagerService
in a synchronous fashion for a specific uid or a (uid, network) pair.
This eliminates the need to keep a copy of the uid policy rules inside
ConnectivityService and ensures that ConnectivityService takes
networking decisions based on the correct state of networking policies,
and therefore eliminates certain data races in ConnectivityManager API
that applications are exposed to.
Test: $ runtest frameworks-net
$ runtest -x frameworks/base/services/tests/../NetworkPolicyManagerServiceTest.java
$ runtest -c com.android.server.net.ConnOnActivityStartTest frameworks-services
Bug: 32069544, 30919851
Change-Id: Ic75d4f7a8853e6be20e51262c4b59805ec35093a
To make the SPI reservation more semantically consistent with the
transform creation API, and to ensure that we always create SPI
reservations relative to a well-known remote, we should take the
SPI request relative to a remote (rather than to a destination).
This necessitates that we now consider direction separately, which
is used for keying the SA-Id.
Bug: 36073210
Test: compilation
Change-Id: I81e955c20128c1f8e04fd68eb26669561f827a78
-Add IpSecService with the necessary glue to connect to netd
-Add code to retrieve IpSecService from System Server
Bug: 34811227
Test: Service boots (and dumpsys works), more via b/34811227
Merged-In: I4cdcb643421141202f77a0e2f87a37012de0cd92
Change-Id: I4cdcb643421141202f77a0e2f87a37012de0cd92
- Add this new meta-data field on NetworkRecommendationProvider to NetworkScorerAppData
Bug: 36571359
Test: runtest frameworks-services
Change-Id: Ic8c594bea406fc5183a4919b808bce5159912650
This patch adds transports info to ValidationProbeEvent and migrates
netId logging for this event to the topt-level netId field in
ConnectivityMetricsEvent.
Test: modified unit tests. $ runtest frameworks-net passes
Bug: 3490169
Change-Id: Ibf51049ba8901ae5ca4ea86e2f500944a4738b5c
This patch deprecates the ifname field for specific metrics events of
types DhcpClientEvent, DhcpErrorEvent, IpReachabilityEvent and
IpManagerEvent.
Instead ifnames are logged in ConnectivityMetricsEvent, allowing for
link layer inference.
Test: updated unit tests, $ runtest frameworks-net passes
Bug: 34901696
Change-Id: I8bfabcb115bbd5289471d653c153a40bb48f28cd
This patch adds translation from ConnectivityMetricsEvent to
IpConnectivityEvent of recently added fields:
- top-level network id
- top-level ifname
- transports
Also adds inference of link layer from transports or ifname.
At the moment these new fields are not populated in
ConnectivityMetricsEvent. Follow-up patches will fill this gap for
the events of the android.net.metrics package.
Test: new unit tests, $ runtest frameworks-net passes
Bug: 34901696
Change-Id: I563a6a3183470bdfaabb7c781a1beaf6b1058bf0
This patch adds new fields to ConnectivityMetricsEvent to make it more
symmetric to IpConnectivityEvent in ipconnectivity.proto.
Follow-up patches will start populating these fields for users of
IpConnectivityLog.
Test: unit tests updated, $ runtest frameworks-net passes
Bug: 34901696
Change-Id: I396767cdfcf38cce893c0d6e1f4524f12e3fdc64
Now that ConnectivityMetricsEvent is only used for core networking
metrics and is not @SystemApi anymore, remove unused fields and prepare
for additional new fields.
Test: updated unit tests, $ runtest frameworks-net passes
Bug: 34901696
Change-Id: I15abad19981d491f16f2a3afe401f1e833079907
This patch adds a few missing counters to APF events:
- an actual lifetime duration to ApfProgramEvent.
- counters for total number of updates to ApfStatistics.
ApfProgramEvents are now recorded at program removal in order to
populate the actual lifetime of the program. ApfProgramEvents whose
actual lifetime was less than 1 second are filtered out.
Finally, instance fields of ApfProgramEvent and ApfStats classes are
made mutable to allow for simple record-like creation. This was not
possible when these classes were tagged @SystemApi.
Test: - manually verified output of $ dumpsys connmetrics list
- unit tests updated.
Bug: 34901696
Change-Id: I02694ebb9421ce1c2aa757fa6aa209d19a654dcd
