Only computing the reassignment does not actually change the
default network.
Test: FrameworksNetTests
Change-Id: I21ddf5cc1e3d3817055dbda4246e38ceb0732407
This check is now unnecessary, seeing how the code adding these
changes is now guaranteed to only add at most one change for
each request.
Test: FrameworksNetTests
Change-Id: Ia0443602d9c89ee413e956df9c7b79f8f74813f7
This patch marks the most important turning point of this
refactoring. Cleanups removing unused code will follow.
Replace the old network-major reassignment computation with a
much simpler and faster loop that takes each request and assigns
it the highest-scoring network.
All tests pass, of course.
Bug: 113554781
Test: FrameworksNetTests
Change-Id: Ie143802995155151a38a4eb1d2f26c3f29e556bd
This makes the rematchAllNetworksAndRequests function, which
is the nexus of the rematching code, very straightforward and
easy to read.
Bug: 113554781
Test: FrameworksNetTests
Change-Id: I5cea4ed7e06439494700d88ab202b696402fa360
This patch finally separates completely computing the rematch
from all the side effects. A collateral effect of this is to
compute correctly the background network state in
rematchNetworkAndRequests, which compensates the breakage
from the previous patch.
Bug: 113554781
Test: ConnectivityServiceTest
Change-Id: I998c729c385940168fcd6ba3f2e01911f1844ce1
To compute accurately whether a network is in the background, the
linger state needs to be updated. Do that before computing whether
a network is in the background and possibly calling
applyBackgroundChangeForRematch.
However ! As of this patch, rematchNetworksAndRequests computes
a wrong value when adding to the list of affected networks, because
it is looking at intermediate global state. Somehow this used to
compensate exactly for the way reading back the state was wrong.
There might have been a few undetected bugs there, but none is
known.
As such, as of this patch, rematchNetworksAndRequests still
computes a wrong value while the computation when applying that
state now computes the right one, so the tests do not pass.
This patch must be checked in together with A44 2/2 which will
fix the computation in rematchNetworksAndRequests, but is kept
separate for ease of review.
Bug: 113554781
Test: ConnectivityServiceTest
Change-Id: Iaeed0d11bfa09f292f232ae020e944e430bc0184
This will be necessary because updating the linger timer is
what lets the network agent know whether the network is
Test: ConnectivityServiceTest
Change-Id: I9b5a36f6d5eac5a404eff6740cd8f4dcbb0c2786
When ConnectivityService#reportNetworkTested is called, the platform
needs to fire ConnectivityDiagnostics callbacks for registered callbacks
that are permissioned for the network being reported on. This adds a new
event to ConnectivityDiagnosticsHandler for invoking these callbacks.
Bug: 143187964
Test: compiles
Test: atest CtsNetTestCases ConnectivityServiceTest
Change-Id: Icc6bcf7a2411133d8ecd7477bc351dad9333f24f
Merged-In: Icc6bcf7a2411133d8ecd7477bc351dad9333f24f
INetworkMonitorCallbacks defines notifyDataStallSuspected() for
notifying ConnectivityService of networks encountering a potential data
stall. A new event is introduced for ConnectivityDiagnosticsHandler to
process the notification and invoke the relevant
ConnectivityDiagnosticsCallbacks.
Bug: 143187964
Test: compiles
Test: atest CtsNetTestCases FrameworksNetTests
Change-Id: I70320bdda9855dced31e08e6a0b25329fb5cb535
Merged-In: I70320bdda9855dced31e08e6a0b25329fb5cb535
INetworkMonitorCallbacks defines notifyNetworkTestedWithExtras() for
notifying ConnectivityService of networks being tested along with a
PersistableBundle of extras. A new event is introduced for
NetworkStateTrackerHandler to notify the ConnectivityDiagnosticsHandler
before continuing with the normal processing for "network tested"
notifications. The event is also used in the
ConnectivityDiagnosticsHandler.
Bug: 143187964
Bug: 147391402
Test: compiles.
Test: atest CtsNetTestCases FrameworksNetTests
Change-Id: Iab29da790c0f5faae68227770bc3a84bbc94f124
Merged-In: Iab29da790c0f5faae68227770bc3a84bbc94f124
This change adds a new VPN user consent flow (using the same text) for
granting the lesser OP_ACTIVATE_PLATFORM_VPN. A new
PlatformVpnConfirmDialog is created as a subclass to preserve all logic,
but ensure the right appop is granted for the relevant dialog.
Intent extras were considered, but are inherently unsafe, since the
caller may add any extras that they would want.
Bug: 144246835
Test: FrameworksNetTests passing
Change-Id: Ia6f36207d43c3748f938430c2780dcf29e5623f3
Merged-In: Ia6f36207d43c3748f938430c2780dcf29e5623f3
The reassignment state has the future value, but is transient
instead of global. With this, the reassignment loop does not
read from global state any more, so it's not affected by the
side effects from the loop.
This means future patches will be able to move the side effects
out of the computation.
Test: atest ConnectivityServiceTest
Change-Id: I43499f19730bcd94b691cfea83aa1ca1d3de75db
If newNetwork is satisfying this request, it means it is the old
satisfier. Plain and simple.
Test: ConnectivityServiceTest
Change-Id: Ic1a5d032801bac476b1c1f53da6f1c4c6056bff0
This is more expensive for now but it will allow subsequent patches
to read relevant state from the reassignment instead of from the
global state, which will stop the computation of the reassignment
from reading state that is mutated by the same loop. Eventually
this lets us completely split the computation from the side effects.
The ugly parts of this patch will be cleaned up later as a result,
namely in patches [NS B04] and [NS B05].
Test: ConnectivityServiceTest
Change-Id: I271e7f4d4bc81493c1ea212025b7130619592a8a
This is an optimization that skips doing intermediate assignments
of networks to requests that will undergo multiple changes during
the recomputation.
It happens to fix a bug where some of these intermediate states
used to have a visible, transient side effect.
Bug: 113554781
Test: ConnectivityServiceTest
Change-Id: I7af3728152a1cf7571de67f394088a5970ee3c1e
makeDefault() will be called in rematchAllNetworksAndRequests
in all cases and its first job is to set mDefaultNetwork.
The old code checks if the currently processing request is
the default request and assigns mDefaultNetwork if it is,
doing it earlier than in the new code iff the new default is
not null. However there is no good reason to assign this
member earlier in the non-null case than in the null case,
it's simpler if the same code path is used in both cases.
mDefaultNetwork is also not used between the old place where
it was set and the new place where it is set, so changing
the timing of the assignment has no observable side effects.
Test: ConnectivityServiceTest
Change-Id: Id47c19b73650ba66bff73b07edb8fd95c707e699
The condition this is testing for cannot actually be false.
The only place where the code writes a null value into this
map is at the end of computeRequestReassignmentForNetwork :
reassignedRequests.put(nri, null).
This proves the code the if() block, which proves that
newNetwork.isSatisfyingRequest(nri.request.requestId) is true.
By definition newNetwork.isSatisfyingRequest(nri) implies that
nri.mSatifier == newNetwork, which proves that
previousSatisfier == newNetwork whenever newSatisfier is null.
Fixes: 146482072
Test: FrameworksNetTests
Change-Id: Ifd6faedce7d49757b82a5f341076ab208b0ccfcb
ConnectivityDiagnosticsManager will send callbacks to
ConnectivityService for registering and unregistering them with the
system. ConnectivityService needs to do the processing for persisting
(and deleting) these callbacks on the ConnectivityService Thread, so
messages are sent to the Connectivity Diagnostics Handler, which runs
on the ConnectivityService Thread.
Bug: 146444622
Bug: 143187964
Bug: 147848028
Test: compiles
Test: atest FrameworksNetTests
Change-Id: Ia5c8f90a60c050504e8676de9564a7607a9b03bc
This change adds stubs for the Platform built-in VPNs, along with
implementing some basic permissions checks.
Bug: 144246837
Test: FrameworksNetTests passing, new tests added
Change-Id: I68d2293fc1468544f0d9f64d02ea7e1c80c8d18c
These logs haven't found a bug in a long time and we now have
some structural guarantees that the conditions they check for
can't happen (like the checks that everything is happening on
the same thread).
Maybe we'll reinstate similar checks later, but for now they
are in the way and removing them is a small sacrifice for the
intended benefit.
The local was simply not used any more.
Test: FrameworksNetTests
Change-Id: If8c8d1f3eb883ffcf0fbdb70824b87dd70da507c
