Raise the poll signal when the socket is being closed to prevent users
of `zsock_poll` blocking after the socket is no more.
Signed-off-by: Jordan Yates <jordan@embeint.com>
Handle AF_UNIX family sockets for NSOS offloaded driver
Note that the size of struct sockaddr_un is done conditionnaly based on
CONFIG_NET_NATIVE_OFFLOADED_SOCKETS
Also, NET_SOCKADDR_PTR_MAX_SIZE needs to be updated only if
CONFIG_NET_SOCKETS_PACKET is not set. Otherwise, for a AF_PACKET socket,
a struct net_context variable can be corrupted, as local would have be on
16 bytes instead of 20 bytes.
Signed-off-by: Noemie Gillet <ngillet@sequans.com>
Some of the network tests require that source and destination
addresses are not swapped so allow test to control the address
swapping from the test.
Signed-off-by: Jukka Rissanen <jukka.rissanen@nordicsemi.no>
Move the network buffer header file from zephyr/net/buf.h to
zephyr/net_buf.h as the implementation now lives outside of the networking
subsystem.
Add (deprecated) zephyr/net/buf.h header to maintain compatibility with old
file path.
Signed-off-by: Henrik Brix Andersen <henrik@brixandersen.dk>
Close file descriptor returned by accept() in error path of
nsos_adapt_accept(), so that file descriptor is not leaked.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
There are 2-level switch statements (one switch inside another) which were
not properly terminated with 'break' statements, leading to implicit
fallthrough. Fix that.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
For each of the fdtable.h functions listed below, convert the
z_ prefixed semi-private functions to use the zvfs_ prefix.
ZVFS stands for Zephyr Virtual File System and
is intended to be a common library used by the C library,
POSIX API, Networking, Filesystem, and other areas.
There are already a few functions in fdtable.h that use the
zvfs_ prefix, so this change is mostly about unifying them in
a way that uses a suitable prefix ("namespace") so that it can
be considered a public API.
- z_alloc_fd
- z_fdtable_call_ioctl
- z_finalize_fd
- z_finalize_typed_fd
- z_free_fd
- z_get_fd_obj
- z_get_fd_obj_and_vtable
- z_get_obj_lock_and_cond
- z_reserve_fd
Signed-off-by: Chris Friedt <cfriedt@tenstorrent.com>
Fill-in the mode field of the fd_entry so that the
implementation can be made aware that the specific file
descriptors created are sockets.
Signed-off-by: Chris Friedt <cfriedt@tenstorrent.com>
Use poll() to wait for connect attempt to complete or timeout. That way
connect() does not block Native Simulator on host syscall.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
So far only a single blocking API could be handled simultaneously, due to
epoll_ctl(..., EPOLL_CTL_ADD, ...) returning -EEXIST when same file
descriptor was added twice.
Follow 'man epoll' advice about using dup() syscall to create a duplicate
file descriptor, which can be used with different events masks. Use such
duplicate for each blocking API except ioctl() (for handling Zephyr poll()
syscall).
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
Introduce nsos_poll_if_blocking(), which replaces common code in:
* nsos_accept_with_poll()
* nsos_recvfrom_with_poll()
This will allow to introduce similar behavior for other blocking APIs in
subsequent commits.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
The gsm_ppp driver is deprecated, and will be removed, along with
its dependencies UART_MUX and GSM_MUX.
This commit removes the integration with GSM_MUX from ppp.c, making
it solely dependent on the chosen node zephyr,ppp-uart.
Signed-off-by: Bjarki Arge Andreasen <bjarki@arge-andreasen.me>
net_pkt_alloc_buffer() deducts the free space from the requested number
of bytes.
As ppp_save_byte() calls net_pkt_alloc_buffer() when the packet has one
byte available still, this causes all but the first net_buf in the
packet to be of size CONFIG_NET_BUF_DATA_SIZE - 1.
Consequences:
- With CONFIG_NET_BUF_FIXED_DATA_SIZE enabled, one byte per net_buf
gets wasted.
- CONFIG_NET_BUF_DATA_SIZE has typically an even, likely even a power
of two value. Using exactly one byte less per buffer causes
operations that require aligned memory (e.g. DMA) to become
inefficient or to not work at all.
Signed-off-by: Reto Schneider <reto.schneider@husqvarnagroup.com>
ioctl() API expects error codes to be returned as negative value directly,
instead of using 'errno'. Fix ZFD_IOCTL_POLL_PREPARE handling to respect
that.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
There is a goto statement with conversion using errno_from_nsos_mid(). Use
NSOS_MID_ERRNO instead of ERRNO to return proper error code.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
Definitions of fcntl flags are different between native libc and Zephyr. In
order to correctly map those values to the host, include Zephyr's fcntl.h
header, instead of the one bundled with libc.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
There is no reason to pass addrlen by pointer, since it is a read-only in
the context of sockaddr_to_nsos_mid().
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
zsock_poll_internal() iterates through all fds to call
ZFD_IOCTL_POLL_UPDATE on them. In cases when -EAGAIN is returned from one
of such syscalls (which happens for example for TLS sockets when in the
middle of a TLS handshake) whole fds array is iterated once again. This
means that ZFD_IOCTL_POLL_UPDATE can be called more than once for a single
preceding ZFD_IOCTL_POLL_PREPARE operation. This resulted in error in
nsos_adapt_poll_remove() call, which was not intended to be called twice.
In case ZFD_IOCTL_POLL_UPDATE is called second time, update 'revents' just
by calling poll() syscall on the host (Linux) side with 0 timeout.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
User is able to enable packet capture for low level HDLC data
frames received from and sent to network. This data can sent to
remote system for analysis. The captured data is encapsulated
into Linux cooked mode SLL packets.
Signed-off-by: Jukka Rissanen <jukka.rissanen@nordicsemi.no>
Improve code consistency and fix CONFIG_POSIX_API=n compatibility with use
of ZSOCK_* and DNS_* macros.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
So far only non-blocking accept() and recvfrom() were suported. This patch
implements blocking behavior, with the use of poll(fd, POLLIN) as helper
mechanism.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
Use NSI_HW_EVENT() in order to periodically check for events in host
sockets. Whenever there is a socket event ready to be processed by Zephyr,
raise native_sim (newly introduced) CPU interrupt, so that Zephyr driver
can signal readiness with k_poll().
Maintain a list of Zephyr poll() executions in Zephyr context. Iterate
through them whenever there is some event to be processed.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
Add driver for 'native_sim' target that implements offloaded socket
networking by the use of host networking stack and wrapped BSD sockets API.
This driver has following advantages over existing networking drivers for
emulated platforms that are already in tree:
* no TUN/TAP use means that no additional setup is required on the host
side:
* possible to use it within unpriviledged Docker containers, either for
development or in CI
* possibility to use and test offloaded sockets
(CONFIG_NET_SOCKETS_OFFLOAD=y) with emulated target, which allows
to increase tests coverage of this feature, without requirement of using
hardware
Native Simulator host libc has different error codes than embedded libc
used by Zephyr. Convert between those.
Signed-off-by: Marcin Niestroj <m.niestroj@emb.dev>
Its default value (100 ms) resulted in PDUs big enough to never make it
through on a slow enough UART (e.g. ~1152-byte PDUs on a UART@115200).
The UART TXs were silently aborted.
A no-timeout value is now allowed and made the default.
Additional warnings are logged when it is likely that a UART TX
was aborted due to a too low timeout for the used baud rate.
Signed-off-by: Tomi Fontanilles <tomi.fontanilles@nordicsemi.no>
- Disable UART when the PPP interface is brought down.
This prevents an error when it is next brought up.
- Change the level of certain logs to be less concerning
and less verbose.
- Fix warnings regarding the passed parameter types of %p conversions.
Signed-off-by: Tomi Fontanilles <tomi.fontanilles@nordicsemi.no>
When we are sending a network pkt, do not tweak the original
packet but the cloned one. The original behavior is ok too, but
logically we should adjust the cloned packet only that is being
received by the stack. This also means that we avoid one extra
copy to tmp variable when sending the packet.
Signed-off-by: Jukka Rissanen <jukka.rissanen@nordicsemi.no>
A proper netmask should be set on the loopback interface, so that
source address selection work properly when there are multiple
interfaces in the system.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
rand32.h does not make much sense, since the random subsystem
provides more APIs than just getting a random 32 bits value.
Rename it to random.h and get consistently with other
subsystems.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Fix few mismatched CONTAINER_OF, one missing k_work_delayable_from_work
conversion and few cases where the target should be pointing at the
first element explicitly.
Signed-off-by: Fabio Baltieri <fabiobaltieri@google.com>
Fix few instances of delayable work handlers using the k_work pointer
directly in a CONTAINER_OF pointing to a k_work_delayable.
This is harmless since the k_work is the first element in
k_work_delayable, but using k_work_delayable_from_work is the right way
of handling it.
Change a couple of explicit CONTAINER_OF doing the same work as the
macro in the process.
Signed-off-by: Fabio Baltieri <fabiobaltieri@google.com>
