problem with implementation of sock_set_flag

[eyuval]

I am using sockets in non-blocking mode, i.e. I set O_NONBLOCK flag for my listening socket and for all newly accepted sockets. The sock_set_flag function in subsys/net/lib/sockets/sockets_internal.h uses the low bits of the net_context user_data pointer to store the SOCK_EOF/SOCK_NONBLOCK flags. I encounter problems, e.g. in zsock_accepted_cb, because the user_data pointer is used without discarding the flag bits, resulting in a pointer, which is off from the true pointer by some bytes, dependent on the flags.
I am working in the native posix system:

CONFIG_BOARD="native_posix_64"
CONFIG_BOARD_NATIVE_POSIX_64BIT=y

[jukkar]

I quickly went through the code and it was not clear to me where the issue is. Only place that I noticed setting ctx->user_data was in zsock_socket_internal() which is the socket() call that initializes the ctx->user_data to NULL. The socket_internal.h has sock_set_flag() and sock_get_flag() use the ctx->user_data as you mentioned. Then sendto() calls either net_context_send() or net_context_sendto() with ctx->user_data and in net_context.c:context_sendto() sets it to itself.

Could you elaborate what kind of issues you were seeing?

[eyuval]

My scenario is as following:

1. I create a listening tcp socket and set it to non-blocking mode.
2. For debbuging I set a breakpoint at zsock_accepted_cb and connect to the listening socket.
3. When the zsock_accepted_cb breakpoint is hit I see the following:
   • The zsock_accepted_cb is called from _tcp_syn_rcvd with the user_data of the listening context passed as user data. In zsock_accepted_cb, the user_data pointer is simply cast to the listener context. But because of the flag bits, the passed in user_data pointer is not equal to the listener context. In my scenario this results in a crash zsock_accepted_cb, when we try to put the new context to the accept_q.

[jukkar]

Yep, got the issue, thanks for the analysis. I will investigate how to solve it.

[jukkar]

@eyuval can you try the proposed PR? It worked in my tests but I would like to get confirmation from you too.

[eyuval]

I tested the patch and for me it fixes the issue. Thanks.