Currently, NetworkStats use int to handle uid, set and tag,
while native side using unsigned int mixing with signed int
with that. This commit make necessary changes in JNI part while
libnetdbpf unifying the types of fields.
Bug: 112226716
Bug: 119193941
Test: 1. manually reconnect vpn
2. update apps from playstore
3. atest libnetdbpf_test
4. runtest frameworks-net
5. cts-tradefed run cts -m CtsUsageStatsTestCases -t \
android.app.usage.cts.NetworkUsageStatsTest
Change-Id: I6c27124db8292e2825fba51b8994f013897cb566
Merged-In: I6c27124db8292e2825fba51b8994f013897cb566
(cherry picked from commit bf7803eed8283d5a1caa2efe90aa449590ddb94f)
(cherry picked from commit 5a43a6735e)
When using xt_qtaguid to count per uid stats,
NetworkStatsService needs to adjust the 464xlat traffic since
iptables module would double count for ipv4 and ipv6 packet.
But for eBPF, the per uid stats is collected in a different
hook, so the adjustment on root uid would only be needed in tx
direction.
Bug: 112226716
Test: 1. Make ipv4 traffic in ipv6-only network and check data
usage.
2. Make ipv4 traffic in a client which connect to
ipv6-only hotspot.
3. runtest frameworks-net
4. cts-tradefed run cts -m CtsNetTestCases -t \
android.net.cts.TrafficStatsTest
5. cts-tradefed run cts -m CtsUsageStatsTestCases
Change-Id: Ic9a84f5446eddc943c255d5f3b89dad171f53cac
Merged-In: Ic9a84f5446eddc943c255d5f3b89dad171f53cac
(cherry picked from commit c33ac0d43b594f6154accf03ae7e3fd34dedc79d)
(cherry picked from commit 3da40c4828)
Currently, NetworkStats use int to handle uid, set and tag,
while native side using unsigned int mixing with signed int
with that. This commit make necessary changes in JNI part while
libnetdbpf unifying the types of fields.
Bug: 112226716
Bug: 119193941
Test: 1. manually reconnect vpn
2. update apps from playstore
3. atest libnetdbpf_test
4. runtest frameworks-net
5. cts-tradefed run cts -m CtsUsageStatsTestCases -t \
android.app.usage.cts.NetworkUsageStatsTest
Change-Id: I6c27124db8292e2825fba51b8994f013897cb566
Merged-In: I6c27124db8292e2825fba51b8994f013897cb566
(cherry picked from commit bf7803eed8283d5a1caa2efe90aa449590ddb94f)
When using xt_qtaguid to count per uid stats,
NetworkStatsService needs to adjust the 464xlat traffic since
iptables module would double count for ipv4 and ipv6 packet.
But for eBPF, the per uid stats is collected in a different
hook, so the adjustment on root uid would only be needed in tx
direction.
Bug: 112226716
Test: 1. Make ipv4 traffic in ipv6-only network and check data
usage.
2. Make ipv4 traffic in a client which connect to
ipv6-only hotspot.
3. runtest frameworks-net
4. cts-tradefed run cts -m CtsNetTestCases -t \
android.net.cts.TrafficStatsTest
5. cts-tradefed run cts -m CtsUsageStatsTestCases
Change-Id: Ic9a84f5446eddc943c255d5f3b89dad171f53cac
Merged-In: Ic9a84f5446eddc943c255d5f3b89dad171f53cac
(cherry picked from commit c33ac0d43b594f6154accf03ae7e3fd34dedc79d)
This is a cherry-pick of aosp/735725 with a minor conflict.
By skipping updates if an alert is already scheduled, this avoids firing
the global alert too often on devices with high transfer speeds and high
quota.
Test: with tethering watching videos. Also runtest frameworks-net
Bug: 117243748
Change-Id: Ia0ed76141dfb33546529786eb0118138ef3708c2
Merged-In: Ie0ba5c1d1a2139606e63fcfe46ec77a89accd8d9
Merged-In: Iaabb31586e9e747871a526dc7ebb7deeb65d8f83
The current networkStats getIfaceStats implementation check if
bpf is enabled, and use bpf to get all traffic stats. However,
the bpf implementation did not contain tcp packet counts. So
data stall detection in DcTracker could not get the packet
count to trigger data stall. Hence the data stall never
triggers for device that enables bpf.
This solution is for short term solution that rollback the
design to use xt_qtaguid for bpf enabled device.
Bug: 110443385
Test: 1. fake data stall to trigger data stall recovery
2. enable debug log to make sure tcp packet count is
correct
3. runtest frameworks-net
4. run cts -m CtsUsageStatsTestCases
Change-Id: I1ce9e92fe194da2ea0a3eec014fd50bb50cdd44a
We've seen reports of negative values flowing through to attempt
being recorded, which will outright crash. This change does one
last-ditch check to see if we're about to work with negative values,
reporting any trouble and clamping them to zero so we don't crash.
This gives us the data we need to continue investigating without
triggering runtime restarts in the field.
Bug: 80057433
Test: atest android.net.NetworkStatsTest
Change-Id: I8174391c6cf5dadc2c2c10a8d841ee07e1f7d934
The current implementation check for bpf map existance whenever a
NetworkStats request comes in. The check is not efficient and may
require additional permission for threads. So a member variable is added
to NetworkStatsService to store the bpf configuration status and is
initialized when NetworkStatsService start.
Test: -m CtsNetTestCases -t android.net.cts.TrafficStatsTest
-m CtsUsageStatsTestCases -t android.app.usage.cts.NetworkUsageStatsTest
Bug: 79994577
Change-Id: I872ae18fb3e9631a4e6c4d773fefbba32e60a03b
Convert to using a constant in INetd to ensure
that there is a consistent tunnel prefix between
Java and native code.
Bug: 74560705
Test: atest FramworksNetTest; atest CtsNetTestCases
Change-Id: Ida233aac2e6c6b26567463964e0ebac9d52eff1e
With the new xt_bpf support for iface stats. We no longer need to parse
the per interface stats from /proc/net/dev. And since the old xt_qtaguid
code path also not depend on it, we can completly remove that helper
function since no caller is depending on it now.
Bug: 72111305
Test: runtest frameworks-net -c com.android.internal.net.NetworkStatsFactoryTest
Change-Id: Icb7eaeef0eeb9fdffd32a90316c76ee05bafffbe
Merged-In: Icb7eaeef0eeb9fdffd32a90316c76ee05bafffbe
(cherry picked from aosp commit b815c978b81eee4b1494bd0b9d25bfad52f08b72)
After adding the xt_bpf module and ifaceStatsMap, eBPF tool can now
support reading per interface data. So networkStatsFactory should move
away from parsing proc/net/dev and use the eBPF map stats instead.
Bug: 72111305
Test: atest com.android.server.net.NetworkStatsServiceTest
Change-Id: Ibcc6150d00835b3bd33af22a72e4a86e172581cf
Merged-In: Ibcc6150d00835b3bd33af22a72e4a86e172581cf
(cherry picked from aosp commit dcc56783f5b945b16c9f5e50333fba557a7c3603)
In order to properly support EOPNOTSUPP this CL
applies a consistent approach to handling Exceptions.
Hereafter, all exceptions that aren't of a special
method-specific type (such as SpiUnavailableException)
will all be returned to the calling process unchanged.
At the API call site, the ServiceSpecificException,
which is really an Errno, will be inspected and either
converted to an unchecked exception for types we know,
or it will be converted to an IOException in cases where
that method can return a checked exception. In cases
where we do not expect an errno, we will simply throw
a generic RuntimeException. This means all API calls
will now properly throw UnsupportedOperationException
and may be CTS tested accordingly.
Bug: 72420898
Test: runtest frameworks-net
Change-Id: I4a00e221618896223fcdb4b4279fb14cd14e34d8
Collecting network statistics is pretty heavy, which is why we're
throttling callers. However, to keep CTS running fast, we provide a
way for tests to force a poll event, instead of making them wait for
the throttle timeout.
Bug: 77908520
Test: atest cts/tests/tests/app.usage/src/android/app/usage/cts/NetworkUsageStatsTest.java
Change-Id: Ia792f0cd495023366ff8c4839df54e7da2ae8331
We've seen devices where heavy communication between "system_server"
and the "phone" process can exhuast Binder threads, especially when
calling while holding locks. To mitigate this, we now interact with
the "phone" process before acquiring any locks.
Update our internal data structures either when we see a connectivity
change, or when SubscriptionManager tells us something changed.
Fix bug in resolveSubscriptionPlan() that always picked the 0'th
SubscriptionPlan instead of looking for the currently active plan;
we now use the same logic for both NSS and NPMS.
Bug: 77908520, 77154412
Test: atest com.android.server.NetworkPolicyManagerServiceTest
Test: atest com.android.server.net.NetworkStatsServiceTest
Change-Id: I177d3fa6cddc78d745b35a9ede12451d458b892c
These trace points surround areas of heavy lifting to help us debug
code that might be taking longer than expected. (For example, slow
getting data from eBPF, slow recording into data structures, slow
writing to disk, or slow evaluating policy.)
Typical timings on a taimen:
performPollLocked: 25.5ms
snapshotUid: 4.4ms
snapshotXt: 1.4ms
[persisting]: 13.4ms
updateNotifications: 3.3ms
Bug: 77908520, 77808546, 77853238, 77154412
Test: builds, boots
Change-Id: I072bfecffee01eeec7e4cbad439bd1cdc166febc
Someone started setting mSystemReady too early, when we really want
to know SystemServer's real ready state. Add a variable to track
that, and don't process broadcasts until that's set.
Bug: 78020762
Test: builds, boots
Change-Id: I65213e46044c95fb0a8a4b09b9aa463bb15c1844
Per email feedback, we should be using "noteOp" instead of "checkOp"
when testing if caller holds OP_GET_USAGE_STATS, so that we record
that caller used the operation.
Bug: 77662908
Test: builds, boots
Exempt-From-Owner-Approval: keep tests passing
Change-Id: I3a60345d590534fdbc2c1248e0d30dc85a5d6772
