Re: [PATCH] Implement the mlock2 function

[Florian Weimer <fweimer at redhat dot com>]

On 11/24/2017 06:24 PM, Adhemerval Zanella wrote:

> > +int
> > +mlock2 (const void *addr, size_t length, unsigned int flags)
> > +{
> > +#ifdef __ASSUME_MLOCK2
> > +  return INLINE_SYSCALL (mlock2, 3, addr, length, flags);
> > +#else
> > +  if (flags == 0)
> > +    return INLINE_SYSCALL (mlock, 2, addr, length);
> > +# ifdef __NR_mlock2
> > +  int ret = INLINE_SYSCALL (mlock2, 3, addr, length, flags);
> > +  if (ret == 0 || errno != ENOSYS)
> > +    return ret;
> > +# endif /* __NR_mlock2 */
> > +  /* Treat the missing system call as an invalid (non-zero) flag
> > +     argument.  */
> > +  __set_errno (EINVAL);
> > +  return -1;
> > +#endif /* __ASSUME_MLOCK2 */
> > +}
>
> We have the INLINE_SYSCALL_CALL to simplify and avoid issue with mismatch
> input number and arguments (which is not this case).

I'll switch to that, thanks.

> I am not sure if it is better to advertise EINVAL for ENOSYS mainly
> because it won't be transparent on a syscall trace.  But I do not have
> a strong opinion here.

It's for consistency.  If you specify a (yet unsupported) flag, you get 
EINVAL with the kernel implementation, but you would get ENOSYS with the 
userspace implementation.  This matters if more flags are added.  It 
would be just another error case for the application to check and a 
potential for application bugs (say because it is only tested on 
mlock2-with-new-flag and mlock2 kernels, not on mlock2-less kernels).

In case this isn't obvious (and you're objecting on principle), I'll add 
a comment. 8-)

Thanks,
Florian