http://ag/480881 changed RouteInfo.getDestination() to return an IpPrefix
instead of a LinkAddress. This makes the equals() comparison always fail.
So, when ConnectivityService.updateRoutes() is given identical routes, instead
of realizing that there's no diff, it would consider them different, and thus
add and remove the same route. The add would fail, since the route already
existed in netd, but the remove would succeed, leaving the system with no routes
and thus no connectivity.
Bug: 15564210
Change-Id: I2003b0fcb809cc20837dc489c58af37891ca4556
- socketFactory() renamed to getSocketFactory()
- Make sure bindProcess() documentation points developers to getSocketFactory() as the preferred approach
- Move bindProcess() and unbindProcess() to ConnectivityManager.setProcessBoundNetwork() -- passing null clears it.
- Move getProcessBoundNetwork() to ConnectivityManager.getProcessBoundNetwork().
Bug:15142362
Bug:13885501
Change-Id: Ia55c59d52e1ec8bf10dd0d9d037bd04c0998bc71
(cherry picked from commit 5ca1e6675bf4182b6e9ca76a7696bf2e38e96c4f)
This change uses IpPrefix only in the public API and continues
to use LinkAddress for everything else. It does not change the
callers to use the new APIs, with the exception of changing
all current uses of getDestination to getDestinationLinkAddress
to make room for the new getDestination method that returns an
IpPrefix.
Based on Sreeram's earlier change:
https://googleplex-android-review.git.corp.google.com/#/c/477874/
but a bit simplified and with a bit more documentation.
Bug: 15142362
Bug: 13885501
Change-Id: Ib4cd96b22cbff4ea31bb26a7853989f50da8de4e
(cherry picked from commit 7d3b4b9a3d4de9673119632da0ebd583e50126f7)
Applying API council comments.
bug: 15142362
(cherry picked from commit Ie0bde68b72656a676d90c0343b9756fe9268d8d6)
Change-Id: Ie0bde68b72656a676d90c0343b9756fe9268d8d6
1. Rename getNetworkPrefixLength to getPrefixLength. Update all
callers in frameworks/base and add a shim method and a TODO
for the rest.
2. @hide isSameAddressAs. It doesn't add much, and it's just
one-liner that callers can implement if they want.
3. Fix the alignment of the initial paragraph (<ul> should have
been </ul>).
4. Remove the documentation that talks about creating
LinkAddresses, since there's no public API for creating them.
With these changes I think LinkAddress is fine as a public API.
Bug: 15142362
Change-Id: Iaf3b1db577745bb68a9e1dd7f96d666dd3f3ec7c
(cherry picked from commit 9ab53650cfcd91a2a151b44b3fd1381841f76269)
This is a first order approx of what we want - should probably be good enough in most cases.
bug:15277751
Change-Id: I10e3b25f6ad5c7e022ba966ed514d4e6a999180d
(cherry picked from commit 94a5c61cb82401dd777d0a7ac43183d92d955323)
When guessing whether a network is restricted or not (e.g., when
constructing a NetworkCapabilities object from an APN type, or
when constructing a request using startUsingNetworkFeature),
only assume the network is restricted if all the capabilities it
provides are typically provided by restricted networks (e.g.,
IMS, FOTA, etc.).
Previous code would conclude a network was restricted even if it
supported one "restricted" capability, so for example an APN
that provides both Internet connectivity and FOTA was marked as
restricted. This caused it to become ineligible to provide the
default Internet connection, because that must be unrestricted.
Also expand the list of restricted APN types a bit.
Bug: 15417453
Change-Id: I8c385f2cc83c695449dc8cf943d918321716fe58
Currently, calling startUsingNetworkFeature for a restricted APN
type (e.g., IMS or FOTA) will create a request that requires
NET_CAPABILITY_NOT_RESTRICTED. Because these APNs are restricted,
when we bring them up we conclude that it does not match the
unrestricted requirement, and we tear them down.
1. Clear the NET_CAPABILITY_NOT_RESTRICTED capability when
creating requests in startUsingNetworkFeature.
2. Refactor the code to a common function so this cannot happen
again.
Bug: 15191336
Change-Id: Id1ec79c58ff79b1a83457ffaecc57d50b61ed4e4
Two fixes. First make sure we mark the request as handled by the network handling it.
Second, convert ensureRouteToHostForAddress to use the new legacyNetworkForType.
bug:14993207
Change-Id: I230968938ca0ed91f834b36a2af60caff2eab682
Make the connectivity changed broadcasts send correct NetworkInfos.
Also update the results of getNetwork.
bug:15290306
bug:15191336
bug:14993207
Change-Id: Ie99ad25f3ebb90d18348e7013761b139e7481866
(cherry picked from commit 16fe1c18289de200d2249e51db8c0986619f487b)
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 9a17b9c5a256cb4bb14821c5ee89b03b99c045e8)
This is implemented by calling through to netd_client.
Included are functions to bind-to-network-for-process strictly for DNS to
facilitate startUsingNetworkFeature() reimplementation.
bug: 13885501
Change-Id: Ib22c7d02ea81d251bdfeeb0f64a47ce32eefcb1b
(cherry picked from commit dbf76f898f1f57eb74722358087c926d2f529bda)
Currently, if a network goes from CONNECTED to some other "live"
state (e.g., CONNECTING, because it's VERIFYING_POOR_LINK) and
back, ConnectivityService treats it as if a new network had
connected. This causes it to attempt to create the network
(which fails, since a network with that netid already exists), to
trigger verification, and if the verification succeeds, to tear
down the network because the request it's satisfying is already
satisfied by the network itself.
Instead, if creating the network fails, assume it's because the
network had already been created, and bail out.
Also, when validation completes, ignore NetworkRequests that were
being served by the same NetworkAgent as before.
Bug: 15244052
(cherry picked from commit cfff026ec47afc7e31f60f80e3deea7f4e2f9be5)
Change-Id: I52c2220e8f1d98fca765880be3040593e92722ed
Currently, once a NetworkAgent has decided to connect, there's
no way to disconnect without first connecting and having
ConnectivityService tear down the connection. This is suboptimal
because it causes the transport to keep retrying even if it knows
that it will not be able to connect.
Instead, allow the transport to abort a connection request that's
in progress, as long as the agent is not yet registered with
ConnectivityService.
Also add locking to evalScores. evalScores should already have
been taking a lock, because it accesses member variables that are
also accessed by the send*methods.
Bug: 15295359
Change-Id: I913c341bdfc50be9c23b632399f53168e754c1c0
(cherry picked from commit 081e99c3ac341734467b65458f94adc160f50d6c)
ConnectivityService doesn't do this anymore.
bug:15077247
Change-Id: I3208c91b2c0369b594987f39ca29da7478435513
(cherry picked from commit 53013c87496980b534e447e717a32698fbd4bca0)
- Improve monitoring of level changes to not be confused
when it goes up while draining or down while charging.
- Put back in connectivity service code to tell battery
stats about the interfaces.
- Turn back on reporting of mobile radio active state
from the RIL.
- Fix bug in marshalling/unmarshalling that would cause
the UI to show bad data.
Change-Id: I733ef52702894cca81a0813eccdfc1023e546fce
Code search says these are the only two files that use it. The
tracker will be resurrected in a slightly different form in
frameworks/opt/net/ethernet.
Bug: 14993642
Bug: 14981801
Change-Id: I2477668ca78dfe46661dda1d97c7f786fd7eba35
Some Factories come and go (Telephony) and so they need to be able to unregister.
Also, debugging is tough when the factories are anonymous, so add names for logging.
Lastly, only send single set of NetworkRequests to a newly registered NetworkFactory
and only send the requests.
Change-Id: I717d63363f25c446f8ecf38d933b1a35d744af6e
Also moved the requestId serial number out of this public class into CS.
Had to leave NetworkRequest hidden for now because the docs refer to things still hidden
in ConnectivityManager.
Change-Id: I14d1fe52d992adf5e4dc197b8f5433e40b0adfe6
If a runtime restart happens while clatd was running, we try to
start clatd, which causes a fatal exception because netd returns
a 400 error (clatd already started.
Bug: 13450716
Bug: 15012035
Change-Id: I102a06d6193fb5f4a1ebe5ad52e5647ff72ca0da
