This CL also updates not to use isTetheringFeatureEnabled for
TETHER_FORCE_UPSTREAM_AUTOMATIC_VERSION which is wrongly configured in
Connectivity namespace.
Test: m
Bug: 279108992
Change-Id: If4f3a57c749bbaeb85454c31248c5bc5feaab936
Closing TCP sockets sends RST packets. If the cellular modem is idle,
sending RST packets will wake the modem up and consume battery.
This CL adds delay_destroy_frozen_sockets_version flag.
When this flag and destroy_frozen_sockets_version is enabled,
ConnectivityService delays closing socket until the cellular modem wakes up.
Pending frozen sockets are closed also when cellular network becomes no
longer the default network.
This CL also adds flag status and pending uids to the dump.
Bug: 284900338
Test: FrameworksNetTests
Change-Id: I2562568390dda36d02f72afb3a96f824788964c0
DeviceConfigUtils has isFeatureEnabled for NetworkStack and Tethering
which were confusing since the difference was only the arguments.
This CL renames isFeatureEnabled for Tethering module to
isTetheringFeatureEnabled to avoid confusion.
Bug: 279108992
Test: m
Change-Id: I05255bb38619afbe621d29524d7910d206a149f1
aosp/2605757 updated isDefaultNetworkActive returns false when there is
no default network.
But this change was not merged to U release branch and this change was
not well tested by dogfooding.
So this CL reverts the isDefaultNetworkActivity behavior change.
Bug: 279380356
Bug: 291870075
Test: atest FrameworksNetTests
Change-Id: Id88662faea9eeaba93d59ab2729f6204a3631ab1
aosp/2605757 changed onNetworkActive callback behavior on U+ but this
change was not merged to U release branch.
This CL reverts the behavior change and keep the current behavior on U+.
Bug: 279380356
Bug: 291870075
Test: atest FrameworksNetTests
Change-Id: I91ccc7ee93f0827986401f316997b4a2df11318f
This patch flags off the functionality added by aosp/2162425
in the wake of b/286649301 where a race in ConnectivityService
breaks WiFi connectivity until reboot.
Bug: 286649301
Test: NetworkAgentTest
ConnectivityServiceTest
Change-Id: I96d755445f6e1f88bb71a7d32742e87dae185250
This reverts commit 4bc9fa6b8c.
Reason for revert: b/288450518
Some iterations over mNetworkAgentInfos result in networks being
disconnected, which removes them from mNetworkAgentInfos during
iteration. This crashes with NPE or OOB exceptions.
Bug: 286649301
Bug: 288149251
Bug: 288450518
Change-Id: I6e0b5b614d9e88267db77cb807ae4bf09f88c0f6
Jean Chalardauthored and Chalard Jeancommited 2023-06-27 19:19:11 +09:00
When a network was explicitly selected, the user just tapped it and
is likely waiting for it to connect. A shorter timeout is therefore
more appropriate for these.
Bug: 281970908
Test: New test in ConnectivityServiceTest
Change-Id: I99a263960cbc9292868fc4bb46a4f3b2f3c71541
This CL changes the behavior as follows.
Before this CL, isDefaultNetworkActive returns true if there is no
default network.
After this CL, isDefaultNetworkActive returns false if there is no
default network on all releases.
Before this CL, LegacyNetworkActivityTracker does not call
onNetworkActive callback for networks that tracker does not add the idle
timer to (e.g. ethernet, cell network configured with activity tracking
disabled).
After this CL,
On T-, onNetworkActive keeps the current behavior.
On U+, onNetworkActive is called regardless of the idle timer existence. This behavior is consistent with ConnectivityManager#isDefaultNetworkActive, which always returns true for a network that has no activity tracking.
Bug: 267870186
Bug: 279380356
Test: atest FrameworksNetTests
Change-Id: I7b1d493fea0ab028b53a3d640a58a00ebbdcc143
Following CLs update LegacyNetworkActivityTracker to support multiple
network activity tracking on U+ devices.
This CL is a preparation for them.
Reasons for having locks in LegacyNetworkActivityTracker were 1)
mNetworkActive can be updated from non handler thread 2)
isDefaultNetworkActive can be called from non handler thread and this
return boolean based on mNetworkActive and mActiveIdleTimers.
aosp/2606673 moves the activity change processing to handler thread
and resolved the first reason.
This CL updates isDefaultNetworkActive to just return
mNetworkActive and resolved the second reason.
So, now LegacyNetworkActivityTracker doesn't need locks and this CL
removed the locks.
Bug: 267870186
Bug: 279380356
Test: atest FrameworksNetTests
Change-Id: I12e3a00c6f8c4a0c40b45b9461860fe2e82fe22a
Currently, various codepaths in ConnectivityService disconnect
networks using NetworkAgentInfo#disconnect. This posts a message
to the NetworkAgent to disconnect and also posts a message to
ConnectivityService to call disconnectAndDestroyNetwork, which
performs all cleanup in ConnectivityService. These two messages
race and the order is non-deterministic.
Instead, always disconnect using disconnectAndDestroyNetwork,
and have disconnectAndDestroyNetwork post the message to the
agent to disconnect.
This fixes a bug where if wifi uses unregisterAfterReplacement
twice in quick succession, when the third agent connects it
doesn't work because the interface is still being used by the
second network.
Also remove the import of IllegalArgumentException to keep the
linter happy (java.lang.* never needs to be imported).
Bug: 286649301
Test: covered by existing tests
Change-Id: Ie01f5589d6839ac6db25f0ba98fc929fbb5b0a96
Lorenzo Colitticommited 2023-06-14 18:03:02 +09:00
One of the reasons for having locks in LegacyNetworkActivityTracker is
mNetworkActive is updated from non handler thread.
aosp/2606673 moves network activity change processing to handler thread.
All the methods that update mNetworkActive and mActiveIdleTimers should
be called from handler thread now.
Bug: 267870186
Bug: 279380356
Test: atest FrameworksNetTests
Change-Id: I5c8cd0b253a77a2740ade3a0145ef80336b3b133
IdleTimer tracking can be disabled if configured timeout is 0 or
negative value.
In this case, setupDataActivityTracking does not add idleTimer and
removeDataActivityTracking causes NPE.
This did not cause crash since the code to cause NPE is wrapped by try
catch.
This CL adds check to avoid causing NPE.
Bug: 267870186
Bug: 279380356
Test: atest FrameworksNetTests
Change-Id: Id774c5ee3d0318d9736f287c50ccf4c7c5743aef
Following CLs update LegacyNetworkActivityTracker to support multiple
network activity tracking on U+ devices.
This CL is a preparation for them.
Bug: 267870186
Bug: 279380356
Test: atest FrameworksNetTests
Change-Id: I2598feef93ef34e10bf576560f687ff7fc449ab9
aosp/2162425 added a WTF if a NetworkAgent changed state before
its native network has been created. This was intended to apply
only to the new U+ codepath, where networks are always created
immediately as soon as their agents register them. But the WTF is
logged on pre-U devices as well.
Fix this so that the WTF is only logged when the new codepath is
running.
Fix: 285999421
Test: TH
Change-Id: I28bda08a006978ab760fe8891bf5cf15b536201a
Lorenzo Colitticommited 2023-06-07 11:42:20 +09:00
Test: new test for this behavior in the preliminary change
Test: FrameworksNetTests NetworkStackTests
Fixes: 138810051
Fixes: 140610528
Change-Id: I95a979d232fb60ece2e33e972bf5d66d20357a1f
Aaron Huangauthored and Chalard Jeancommited 2023-06-05 20:24:59 +09:00
See system/netd/include/Fwmark.h which reserves the bottom 21 bits
of skb->mark for netid, etc... so out of necessity these have to be
in the top 11 bits.
In practice the only values that really make sense are either:
mark == mask == 0 (disabled)
or (mark == mask) being one of the top 11 available bits, for example:
mark == mask == 0x80000000 (enabled, using top-most bit)
(only a single bit makes sense, as this is really just a boolean signal,
and only the topmost is known to be used on any real devices)
Let's just force the use of 0x80000000 for ecosystem consistency.
Bug: 284334830
Test: TreeHugger
Signed-off-by: Maciej Żenczykowski <maze@google.com>
(cherry picked from https://android-review.googlesource.com/q/commit:767cf7f1cfc6d612bc8554fef8c354fdd660695b)
Merged-In: I199e9b00de845b3747940426ea6644426ab72e87
Change-Id: I199e9b00de845b3747940426ea6644426ab72e87
Maciej Żenczykowskicommited 2023-06-05 07:44:18 +00:00
See system/netd/include/Fwmark.h which reserves the bottom 21 bits
of skb->mark for netid, etc... so out of necessity these have to be
in the top 11 bits.
In practice the only values that really make sense are either:
mark == mask == 0 (disabled)
or (mark == mask) being one of the top 11 available bits, for example:
mark == mask == 0x80000000 (enabled, using top-most bit)
(only a single bit makes sense, as this is really just a boolean signal,
and only the topmost is known to be used on any real devices)
Let's just force the use of 0x80000000 for ecosystem consistency.
Bug: 284334830
Test: TreeHugger
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I199e9b00de845b3747940426ea6644426ab72e87
Maciej Żenczykowskicommited 2023-06-05 07:41:00 +00:00
aosp/2490881 updated to use InetDiagMessage.destroyLiveTcpSocket for all
devices.
But it is possible that netd socketDestory is modified in T- devices.
So this CL revert changes to keep using netd socketDestroy in T-
devices.
Test: atest FrameworksNetTests
Bug: 284253763
Change-Id: I9b61f10e975d2e38e9829a8c01d3af706e2518ef
Motomu Utsumi
authored and
Automerger Merge Worker
commited 
Xin Li
commited 
Treehugger Robot
authored and
Gerrit Code Review
commited 
Chalard Jean
commited 
Lorenzo Colitti
authored and
Maciej Żenczykowski
authored and
Aaron Huang
authored and
Treehugger Robot
authored and