Re: [PATCH v2 11/17] gdbstub: Try unlinking the unix socket before binding

["Alex Bennée" <alex.bennee@linaro.org>]

Michael Tokarev <mjt@tls.msk.ru> writes:

> 07.02.2025 18:31, Alex Bennée wrote:
>> From: Ilya Leoshkevich <iii@linux.ibm.com>
>> In case an emulated process execve()s another emulated process,
>> bind()
>> will fail, because the socket already exists. So try deleting it. Use
>> the existing unix_listen() function which does this. Link qemu-user
>> with qemu-sockets.c and add the monitor_get_fd() stub.
>> Note that it is not possible to handle this in do_execv(): deleting
>> gdbserver_user_state.socket_path before safe_execve() is not correct,
>> because the latter may fail, and afterwards we may lose control.
>
> Please note: this is linux-user stuff, which is usually linked statically.
> By linking it with qemu-sockets, we basically broke static linking, because
> qemu-sockets uses getaddrinfo() &Co.  The previous code, I think, was there
> for a reason, - to avoid this linkage.

Oops, how come we didn't notice? We do have a bunch of --static targets
in the CI.

> How do you think about reverting this one and addressing the original
> problem without using qemu-sockets?
>
> Alternatively, it might be possible to split qemu-sockets.c into unix-related
> stuff and generic stuff.

You mean move all the unix_XXX() functions into a new unit that ensures
we don't need getaddrinfo()?

>
> Thanks,
>
> /mjt

-- 
Alex Bennée
Virtualisation Tech Lead @ Linaro