pread and pwrite

pread and pwrite

[Carsten Langgaard]

I'm running some tests from LTP, which tests pread and pwrite.
It look like pread/pwrite doesn't do any check, if they are called with
'buf =NULL' or 'offset < 0', and no error is return.
If I look in glibc in sysdeps/generic/pread.c it look like this:

ssize_t
__libc_pread (int fd, void *buf, size_t nbytes, off_t offset)
{
  if (nbytes == 0)
    return 0;
  if (fd < 0)
    {
      __set_errno (EBADF);
      return -1;
    }
  if (buf == NULL || offset < 0)
    {
      __set_errno (EINVAL);
      return -1;
    }

  __set_errno (ENOSYS);
  return -1;
}

Here there is some checking for sane values and a proper error value is
return.
I guess this routine is replaced, if we have the syscall implemented
with the sysdeps/unix/sysv/linux/mips/pread.c file.
Here there is no check for sane values, is there any reason why ?
The same thing goes for pwrite.

/Carsten

--
_    _ ____  ___   Carsten Langgaard   Mailto:carstenl@mips.com
|\  /|||___)(___   MIPS Denmark        Direct: +45 4486 5527
| \/ |||    ____)  Lautrupvang 4B      Switch: +45 4486 5555
  TECHNOLOGIES     2750 Ballerup       Fax...: +45 4486 5556
                   Denmark             http://www.mips.com

Re: pread and pwrite

[Ralf Baechle]

On Wed, Jul 17, 2002 at 11:03:55AM +0200, Carsten Langgaard wrote:

> 
> Here there is some checking for sane values and a proper error value is
> return.
> I guess this routine is replaced, if we have the syscall implemented
> with the sysdeps/unix/sysv/linux/mips/pread.c file.
> Here there is no check for sane values, is there any reason why ?
> The same thing goes for pwrite.

The kernel does it's own error checking.  No need to duplicate that in
userspace.

  Ralf

Re: pread and pwrite

[Carsten Langgaard]

Ralf Baechle wrote:

> On Wed, Jul 17, 2002 at 11:03:55AM +0200, Carsten Langgaard wrote:
>
> >
> > Here there is some checking for sane values and a proper error value is
> > return.
> > I guess this routine is replaced, if we have the syscall implemented
> > with the sysdeps/unix/sysv/linux/mips/pread.c file.
> > Here there is no check for sane values, is there any reason why ?
> > The same thing goes for pwrite.
>
> The kernel does it's own error checking.  No need to duplicate that in
> userspace.

The kernel doesn't do this a proper check then.
The pread/pwrite parameters is also convert in glibc, the 'offset' is
convert from a 'long' to a 'long long', but it isn't sign extended.
So when pread is call with offset -1, then kernel won't see it as -1.

>
>   Ralf

--
_    _ ____  ___   Carsten Langgaard   Mailto:carstenl@mips.com
|\  /|||___)(___   MIPS Denmark        Direct: +45 4486 5527
| \/ |||    ____)  Lautrupvang 4B      Switch: +45 4486 5555
  TECHNOLOGIES     2750 Ballerup       Fax...: +45 4486 5556
                   Denmark             http://www.mips.com

Re: pread and pwrite

[H. J. Lu]

On Wed, Jul 17, 2002 at 04:18:17PM +0200, Carsten Langgaard wrote:
> Ralf Baechle wrote:
> 
> > On Wed, Jul 17, 2002 at 11:03:55AM +0200, Carsten Langgaard wrote:
> >
> > >
> > > Here there is some checking for sane values and a proper error value is
> > > return.
> > > I guess this routine is replaced, if we have the syscall implemented
> > > with the sysdeps/unix/sysv/linux/mips/pread.c file.
> > > Here there is no check for sane values, is there any reason why ?
> > > The same thing goes for pwrite.
> >
> > The kernel does it's own error checking.  No need to duplicate that in
> > userspace.
> 
> The kernel doesn't do this a proper check then.
> The pread/pwrite parameters is also convert in glibc, the 'offset' is
> convert from a 'long' to a 'long long', but it isn't sign extended.
> So when pread is call with offset -1, then kernel won't see it as -1.
> 

Please check it out:

http://sources.redhat.com/ml/libc-alpha/2002-07/msg00188.html


H.J.

Re: pread and pwrite

[Carsten Langgaard]

"H. J. Lu" wrote:

> On Wed, Jul 17, 2002 at 04:18:17PM +0200, Carsten Langgaard wrote:
> > Ralf Baechle wrote:
> >
> > > On Wed, Jul 17, 2002 at 11:03:55AM +0200, Carsten Langgaard wrote:
> > >
> > > >
> > > > Here there is some checking for sane values and a proper error value is
> > > > return.
> > > > I guess this routine is replaced, if we have the syscall implemented
> > > > with the sysdeps/unix/sysv/linux/mips/pread.c file.
> > > > Here there is no check for sane values, is there any reason why ?
> > > > The same thing goes for pwrite.
> > >
> > > The kernel does it's own error checking.  No need to duplicate that in
> > > userspace.
> >
> > The kernel doesn't do this a proper check then.
> > The pread/pwrite parameters is also convert in glibc, the 'offset' is
> > convert from a 'long' to a 'long long', but it isn't sign extended.
> > So when pread is call with offset -1, then kernel won't see it as -1.
> >
>
> Please check it out:
>
> http://sources.redhat.com/ml/libc-alpha/2002-07/msg00188.html
>
> H.J.

So the same issue has been raised today on the glibc list, amazing. I guess the
problem has existed quite some time.
But it look like the patch will fix the problem. Do you know if the patch has
been committed and which version of glibc will it then be fixed in ?

Thanks,
/Carsten