As it turns out, the C standard forbids passing va_list by value and then
continue using it in the parent function (C Standard, 7.16, paragraph 3).
Luckily, there's a footnote stating:
"It is permitted to create a pointer to a va_list and pass that
pointer to another function, in which case the original function may
take further use of the original list after the other function
returns."
Therefore, we're safe passing va_list by reference and thus can keep the
current coding style.
This fixes weird bugs on ARM32 which really doesn't allow passing va_list
by value.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Forward group-formation-failure events via dbus so sinkctl can restart
p2p-scans immediately. Unfortunately, the event itself does not contain
any useful information at all. Therefore, we have to track the connection
attempts ourselves.
Signed-off-by: Andrey Gusakov <adnrey.gusakov@cogentembedded.com>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
sd_bus never uses 'bool', and given the weird implicit type promotions, it
is totally right to do so. Make sure we never pass pointers to 'bool' to
va_arg sd_bus functions to avoid overwriting stack space.
Signed-off-by: Andrey Gusakov <adnrey.gusakov@cogentembedded.com>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
The libsystemd-bus API changed recently. Update our calls to use the new
API so it compiles with newer systemd versions.
Signed-off-by: Andrey Gusakov <adnrey.gusakov@cogentembedded.com>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Once we have a source connection to our local sink, spawn gstreamer so the
actual video data is rendered.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
miracle-sinkctl is a very basic Miracast-Sink implementation. It is meant
for debugging and as proof-of-concept.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
If we advertise changes in Peer.Connected _before_ updating other props,
they're not available in the callbacks. Delay the notifications after we
parsed all props.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
We need to drop the remote-cookie flag when matching replies. Otherwise,
we will never find the local request.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
If we delay device-parsing until P2P_PEER returns, we might receive other
device events in between which we cannot handle as the peer is unknown.
Avoid that by immediately creating the peer and appending more detailed
data later once P2P_PEER returns.
As P2P_PEER keys and P2P_DEVICE_FOUND keys are different (wholy shit, what
a fucking interface), we need to parse both. Lets hope there won't be any
conflicts.
WARNING: The wfd_subelements field is currently very inconsistent.
Sometimes we only use dev_info fields, somethings the whole IE. Users
shall use it as boolean value, only (set vs. unset). Once we add a proper
IE parser, we will provide more advanced fields.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
WifiDisplay uses RTSP for stream-setups. This adds a basic rtsp-bus
implementations that we can use for sinks and sources.
Note that the implementation is optimized for usability, not speed. RTSP
is used for control-data, not streaming-data so there's no need to
over-optimize it. In case inlined RTP is used, we still provide proper
speed, even though that's usually not used by WifiDisplay implementations
(due to the TCP requirement).
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
The event-loop might be needed in later ctl additions for timeouts, so
make it global. Same for the bus-logic.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
wpa_supplicant is horrible in notifying us about p2p-scans. During failed
connection attempts or other timeouts, we usually are not notified about
scan-stops. Therefore, a following p2p-scan will simply be ignored as we
think it's still running.
Avoid that problem by always forwarding the P2P_FIND command.
wpa_supplicant can deal with redundant requests so we're fine. Note that
this requires the caller to issue these calls in intervals or/and after
any issues wifi request. Once wpas fixes this issue, we can add this guard
again and the redundant requests will be suppressed.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Delay object-broadcasts so we first parse properties and then advertise
the objects. Otherwise, the caller will get almost empty objects which are
rather useless.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Copy over the "make memcheck" target from libshl. It runs the test-suite
again via valgrind and fails on any valgrind warnings. This is very useful
to find memory-leaks and invalid memory-accesses via the test-suite.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
We _must_ aggressively drain input queues before we signal any HUP. There
might be queued messages that tell us important information about the
termination of wpa_supplicant.
Therefore, if a write() operation fails, we only signal HUP if the input
queues are empty. However, we cannot rely on EPOLLIN to be set, as the
input data might have arrived in between epoll_wait() and write(),
therefore, always run the non-blocking read() in case write() failed.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
If you run "make check -jX", the test-suites will run in parallel. This
currently works fine, but someone might copy the wpas tests into something
more complex. In that case, the /tmp/XY files will be shared, which breaks
the test-suite. Therefore, use pid-suffixes so these files are exclusive
to the given test-process.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
The shl_ring helpers manage a dynamically-growing ring-buffer. We need
that for the following RTSP stream parsers, so add them to src/shared/.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Add u64 helpers to shl_htable. They're fairly trivial and just copied from
unsigned-long, however, in case size_t is not 64bit wide we need to do
some trivial hashing.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
The miracle-wifictl program can be used to manage miracle-wifid. It
provides low-level P2P helpers, so the main miraclectl can skip thos.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
The miracle-wifid daemon is a rewrite of the wifi-related parts of
miracled. Unlike miracled, we no longer integrate the wifi-part directly
into the core daemon. This way, we can easily replace the wifi-bridge with
other network managers once they gain P2P capabilities.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
The wpas-helpers provide easy _asynchronous_ access to wpa_supplicant
control interfaces. Compared to the old wpa_ctrl_* stuff it's no longer
synchronous. Thus, it doesn't block our daemon while wpa_supplicant runs
some heavy work again..
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Use a common subdirectory for gdhcp so we can easily replace it once
sd-dhcp has server-side support.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
If we get ADD events from udev for a new link but run in unmanaged mode,
skip the event. It is fairly unlikely that another manager has already
started wpa_supplicant on it (and even if, it would be racy).
Hence, lets not support hotplugging in these cases until external managers
provide us a notification API.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
If, after 5 times (which is 1s), wpa_supplicant is still not reachable, we
now drop the link. This avoids hangup issues during wpas startup that
might happen from time to time.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
If miracled is spawned with --manage-wifi, we run our own copy of
wpa_supplicant on managed devices. This is meant for testing purposes
only. Hopefully, network-manages like NM or connman will provide a P2P API
at some time so we can switch. Until then, we allow the user to control
who is in charge of wpa_supplicant.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
This is very handy to catch wrong usage of shl_strjoin(). We now get a gcc
warning if the sentinel (NULL) is omitted.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
