1. Remove the IP provisioning thread and just attempt
provisioning indefinitely whenever we have an interface.
2. Make all methods run on the passed-in handler thread. This
makes it easier to verify correctness by code inspection.
3. Remove the code that changes the factory score depending on
whether we're tracking an interface and have link. This is
unnecessary complexity, as there is no penalty to accepting a
request even if we don't have an interface.
4. Remove code duplication and only have one codepath for
stopping layer 3.
Tested the following are tested with this CL:
- Booting with an interface connected.
- Disconnecting/reconnecting the Ethernet cable repeatedly,
particularly at inconvenient times (e.g., during provisioning).
- Similarly, disconnecting/reconnecting USB Ethernet interfaces.
- Falling back to another Ethernet interface if the currently
tracked Ethernet interface is unplugged.
- Disconnecting and restarting provisioning when provisioning is
lost (e.g., if the default route is deleted).
- Crashing the system server causes Ethernet to reconnect on
restart.
- The above while running watch -n 0.1 adb shell dumpsys ethernet
(cherry picked from commit 9effed6f9e)
Bug: 62308954
Test: tested on marlin with USB ethernet adapters, as described
Test: smoketested on aosp_bullhead
Change-Id: Ie3041f1d453d15c81eee81adfc8593269c499d79
1. Remove the IP provisioning thread and just attempt
provisioning indefinitely whenever we have an interface.
2. Make all methods run on the passed-in handler thread. This
makes it easier to verify correctness by code inspection.
3. Remove the code that changes the factory score depending on
whether we're tracking an interface and have link. This is
unnecessary complexity, as there is no penalty to accepting a
request even if we don't have an interface.
4. Remove code duplication and only have one codepath for
stopping layer 3.
Tested the following are tested with this CL:
- Booting with an interface connected.
- Disconnecting/reconnecting the Ethernet cable repeatedly,
particularly at inconvenient times (e.g., during provisioning).
- Similarly, disconnecting/reconnecting USB Ethernet interfaces.
- Falling back to another Ethernet interface if the currently
tracked Ethernet interface is unplugged.
- Disconnecting and restarting provisioning when provisioning is
lost (e.g., if the default route is deleted).
- Crashing the system server causes Ethernet to reconnect on
restart.
- The above while running watch -n 0.1 adb shell dumpsys ethernet
Bug: 62308954
Test: tested on marlin with USB ethernet adapters, as described
Change-Id: Iad12a52a903bfaccf7e245dfe499652c752c31e9
This fixes a longstanding bug where after a DHCP timeout, we
would never restart the DHCP client and get an IP address until
the link bounced.
Also, two minor improvements:
1. Dump IpManager info when dump() is called.
2. When onLinkPropertiesChange is called, also update
mLinkProperties. We were already sending the updated
LinkProperties to the NetworkAgent, so this is really only
useful for dump(), but it's just one line and safe because
onLinkPropertiesChange already grabs the lock.
Bug: 17733693
Change-Id: I42c3319cb4bc151c547ed721baf5e83f97e23862
Requiring CHANGE_NETWORK_STATE is pointless, since immediately
after doing that we require CONNECTIVITY_INTERNAL, which is
a system-only permission.
Bug: 23597341
Change-Id: If4c8ecbb319e547c2c81d2dcc3c7bbb63e0bc2af
The intent of the code was to catch exceptions, but it wasn't
actually doing that because when a netd command fails we get an
IllegalStateException, not a RemoteException.
Bug: 23581852
Change-Id: I66c7c33bdc7b0d0de4fa5cd92dff054797cf860f
Replaced the method where we used to enforce static permission
CHANGE_NETWORK_STATE to method calls that also check for WRITE_SETTINGS.
This is due to the merging of CHANGE_NETWORK_STATE permission with
WRITE_SETTINGS in M.
Change-Id: I104ac41733663b4f6b5cd30e7314574ec4acdc07
Previously we did not stop DHCP or try to force a disconnect.
This made it impossible to reconnect by unplugging and
replugging an USB Ethernet adapter.
Bug: 17736828
Change-Id: Idd8576a96b063b5ca726ee9149800bdcb0f62bd1
Network Factories are allowed to go below, but networks need to be
constrained. Allowing the network to go below 0 meant that -1 could
sometimes leak through and foul the logic.
bug:17361330
Change-Id: Ife34ca0f9c233dd3c3df80f6fea580af43afcdeb
Also make static IP configuration more robust (e.g., tear down
our NetworkAgent when we switch between static and DHCP).
Bug: 16114392
Bug: 16893413
Change-Id: Ib33f35c004e30b6067bb20235ffa43c247d174df
NetworkFactory only indicates if we're interested in new connections. It
shouldn't be used to tear down existing connections (they have unwanted callbacks
for that).
Supports linger properly as well as dealing with tie scores.
bug:15612739
Change-Id: Ib3dfe673d3645b9dc4756c176958409a64ec32e4
(cherry picked from commit 8b07ee36a2)
NetworkFactory only indicates if we're interested in new connections. It
shouldn't be used to tear down existing connections (they have unwanted callbacks
for that).
Supports linger properly as well as dealing with tie scores.
bug:15612739
Change-Id: Ib3dfe673d3645b9dc4756c176958409a64ec32e4
It was attempting to update the NetworkAgent before the NetworkAgent
was created, causing NullPointerExceptions.
bug:15407938
Change-Id: Ie36b906df4b12f095a16d1b30c847a55b074d60c
Make NetworkFactory a concrete class and divide responsibilites between it and NetworkAgent.
Factory will track requests and by default give a single connect/disconnect api for ease
of use. Then NetworkAgent is created and destroyed as needed with very simple logic.
Change-Id: I401c14a6e5466f2fc63b04219b97ff85bb9af291
(cherry picked from commit 8af038d056)
Make NetworkFactory a concrete class and divide responsibilites between it and NetworkAgent.
Factory will track requests and by default give a single connect/disconnect api for ease
of use. Then NetworkAgent is created and destroyed as needed with very simple logic.
Change-Id: I401c14a6e5466f2fc63b04219b97ff85bb9af291
