[rtems-libbsd commit] sys/kern: Add VFS support

[Chris Johns]

On 8/2/22 12:48 am, Sebastian Huber wrote:
> Hello Chris,
> 
> sorry, this slipped through may review list.

Thanks for the review.

> 
> On 02/09/2021 04:43, Chris Johns wrote:
>> @@ -1232,9 +993,6 @@ osendmsg(struct thread *td, struct osendmsg_args *uap)
>>   #endif
>>   #endif /* __rtems__ */
>>   -#ifdef __rtems__
>> -static
>> -#endif /* __rtems__ */
>>   int
>>   sys_sendmsg(struct thread *td, struct sendmsg_args *uap)
>>   {
>> @@ -1257,35 +1015,7 @@ sys_sendmsg(struct thread *td, struct sendmsg_args *uap)
>>       free(iov, M_IOV);
>>       return (error);
>>   }
>> -#ifdef __rtems__
>> -ssize_t
>> -sendmsg(int socket, const struct msghdr *message, int flags)
>> -{
>> -    struct thread *td = rtems_bsd_get_curthread_or_null();
>> -    struct sendmsg_args ua = {
>> -        .s = socket,
>> -        .msg = message,
>> -        .flags = flags
>> -    };
>> -    int error;
>> -
>> -    if (td != NULL) {
>> -        error = sys_sendmsg(td, &ua);
>> -    } else {
>> -        error = ENOMEM;
>> -    }
>> -
>> -    if (error == 0) {
>> -        return td->td_retval[0];
>> -    } else {
>> -        rtems_set_errno_and_return_minus_one(error);
>> -    }
>> -}
>> -#endif /* __rtems__ */
> 
> Why did you move all these system call implementations for RTEMS into a separate
> file? The sys_sendmsg() was a static function so the compiler was able to
> optimize the use of struct sendmsg_args away and there was no function call
> overhead.  In the successful case the return value was determined by
> td->td_retval[0]. I don't think this now the case?
> 
> ssize_t
> sendmsg(int socket, const struct msghdr *message, int flags)
> {
>     struct thread *td = rtems_bsd_get_curthread_or_null();
>     struct sendmsg_args ua;
>     int ffd;
>     int error;
>     if (RTEMS_BSD_SYSCALL_TRACE) {
>         printf("bsd: sys: sendmsg: %d\n", socket);
>     }
>     if (td == NULL) {
>         return rtems_bsd_error_to_status_and_errno(ENOMEM);
>     }
>     ffd = rtems_bsd_libio_iop_hold(socket, NULL);
>     if (ffd < 0) {
>         return rtems_bsd_error_to_status_and_errno(EBADF);
>     }
>     ua.s = ffd;
>     ua.msg = message;
>     ua.flags = flags;
>     error = sys_sendmsg(td, &ua);
>     rtems_bsd_libio_iop_drop(socket);
>     return rtems_bsd_error_to_status_and_errno(error);> }
> 
> Unfortunately syscalls01 only tests error conditions and not a good case.

It looks like a bug. I think there should be a check of `error` after the drop
and the call's result returned.

> It seems that the commit is not present in the master branch. This means all the
> work will be lost when we update to a newer FreeBSD baseline.

Yes. It is an out standing task that I need to get funding for.

What are your plans and if you are looking to update, timeline?

Chris