If a tc ebpf program writes into a packet using direct packet access
then the packet will automatically be uncloned and pulled by
additional prologue inserted by the kernel itself. See
tc_cls_act_prologue() & bpf_unclone_prologue() in kernel sources
(this is how the clat ebpf program works, which does DPA writes).
However in the forwarding programs we only *read* from the packets
using direct packet access, but never write. All writes happen via
kernel bpf helpers (this is mostly an implementation detail: since
we need to use helpers for checksum updates, I decided to also use
checksums for the writes themselves). As such the insert 'automatic
unclone/pull' logic doesn't trigger.
It is thus possible (it depends on the skb layout delivered by the
nic driver) for 0 bytes of the packet to be accessible for read
using direct packet access. We thus need to explicitly try to pull
in the header of the packet so that we can inspect it.
In most cases (on most drivers for most packet types) this will
end up being a no-op (because the headers will already be in
the linear portion of the skb). But on some drivers for some
packet types it ends up mattering.
Test: TreeHugger, makes icmpv6 tether forwarding work on bramble
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I4b07e57728ce544ffb908527ea11ecc315e5acec
When forwarding is on, DadProxyTest fails because forwarding
disables optimistic addresses. This results in the DAD proxy
getting EADDRNOTAVAIL.
Forwarding is on quite often on real hardware because many
solutions for wifi calling use forwarding.
Fix the test to retry after two seconds if the packet is not
forwarded on the first attempt.
It would also be possible to make this work more reliably in the
test by simply disabling DAD or disabling forwarding. However,
because DAD does happen in real usage (downstream interfaces
always disable DAD, but upstream interfaces don't), that seems
risky. For example, if the test disabled DAD, it would not catch
bugs where the DAD proxy crashed or stopped forwarding if it got
EADDRNOTAVAIL.
Test: atest TetheringPrivilegedTests on coral, which has forwarding on
Change-Id: I58280ef7c0e40371cd770ead4c8baa7190c288fd
Current upstream selection code suffers from a race where if the
CONNECTIVITY_ACTION broadcast for a given network switch is
received and processed before the NetworkCallbacks for that
network switch, upstream selection just re-selects the same
upstream it had before. The incorrect upstream persists until
another CONNECTIVITY_ACTION is received.
Fix this by defining a new EVENT_DEFAULT_SWITCHED message code
communicated from UpstreamNetworkMonitor to Tethering, and send
that whenever the default network switches.
The message is sent in onLinkPropertiesChanged, because the
tethering code stores all information about networks in an
UpstreamNetworkState structure that contains Network,
LinkProperties and NetworkCapabilities. When a network switch
occurs, onLinkPropertiesChanged always follows onAvailable and
onCapabilitiesChanged, and thus marks the first point in time
when all the information is available.
This CL tries not to change existing codepaths too much, but
it does move the update of mDefaultInternetNetwork from
onCapabilitiesChanged to onLinkPropertiesChanged. This should
not be a problem because the only thing that reads
mDefaultInternetNetwork is getCurrentPreferredUpstream, which,
in the case of a default network switch, will be run by the
onLinkPropertiesChanged which will immediately follow.
Bug: 173068192
Test: changes to existing unit tests show bug is fixed
Change-Id: Ic9196bc92892811b25bda463ffd839ee5c19d294
In the current tethering code, upstream selection is only
triggered by CONNECTIVITY_ACTION. But in automatic mode, the
upstream network is selected by listening to a NetworkCallback
that tracks the default network.
This causes a race where if the CONNECTIVITY_ACTION for a network
switch is received and processed before the callbacks for that
network switch, upstream selection just re-selects the upstream
currently in use.
Make it possible to test this by giving TestConnectivityManager
the ability to choose the ordering between NetworkCallbacks and
CONNECTIVITY_ACTION, and to run an arbitrary Runnable between
calling one and calling the other. TetheringTest passes a
Runnable that calls mLooper.dispatchAll(), which ensures that
the tethering code fully processes the first set of information
it receives (either the broadcast (or the callbacks) before
receiving any more information.
Add test coverage to testAutomaticUpstreamSelection that
exercises various orderings, and make the test pass by expecting
the buggy behaviour of the current code.
An upcoming CL will fix the bug and update the tests.
Bug: 173068192
Test: test-only change
Change-Id: I7805444dcf59f6d5f8517fbcf2f2b1641783d50b
