Re: + sysctl-return-einval-if-val-violates-minmax.patch added to -mm tree

Re: + sysctl-return-einval-if-val-violates-minmax.patch added to -mm tree

[Luis Chamberlain]

On Mon, Feb 11, 2019 at 10:19:19PM +0100, Christian Brauner wrote:
> On Tue, Feb 12, 2019 at 12:17:16AM +0300, Alexey Dobriyan wrote:
> > On Mon, Feb 11, 2019 at 01:06:32PM -0800, akpm@linux-foundation.org wrote:
> > 
> > > @@ -2848,8 +2848,10 @@ static int __do_proc_doulongvec_minmax(v
> > 
> > > -			if ((min && val < *min) || (max && val > *max))
> > > -				continue;
> > > +			if ((min && val < *min) || (max && val > *max)) {
> > > +				err = -EINVAL;
> > 
> > I was asked to return ERANGE in kstrto*().
> 
> I think we discussed ERANGE vs EINVAL and decided EINVAL because there
> was precedence for other sysctls already.

Can you do a proper audit and see?

linux-api folks may care.

 Luis

Re: + sysctl-return-einval-if-val-violates-minmax.patch added to -mm tree

[Christian Brauner]

On Mon, Feb 11, 2019 at 01:25:23PM -0800, Luis Chamberlain wrote:
> On Mon, Feb 11, 2019 at 10:19:19PM +0100, Christian Brauner wrote:
> > On Tue, Feb 12, 2019 at 12:17:16AM +0300, Alexey Dobriyan wrote:
> > > On Mon, Feb 11, 2019 at 01:06:32PM -0800, akpm@linux-foundation.org wrote:
> > > 
> > > > @@ -2848,8 +2848,10 @@ static int __do_proc_doulongvec_minmax(v
> > > 
> > > > -			if ((min && val < *min) || (max && val > *max))
> > > > -				continue;
> > > > +			if ((min && val < *min) || (max && val > *max)) {
> > > > +				err = -EINVAL;
> > > 
> > > I was asked to return ERANGE in kstrto*().
> > 
> > I think we discussed ERANGE vs EINVAL and decided EINVAL because there
> > was precedence for other sysctls already.
> 
> Can you do a proper audit and see?

If you look at proc_get_long() right now you can see that when the
buffer we use to parse the number is exceeded we return EINVAL. In short
if you do right now:

echo 1844674407370955161600000 > /proc/sys/fs/file-max

that would exceed the buffer in proc_get_long() and you already get
EINVAL for all such cases. If we now change this to ERANGE we would
return:

echo 18446744073709551616      > /proc/sys/fs/file-max -> ERANGE
echo 1844674407370955161600000 > /proc/sys/fs/file-max -> EINVAL

which would be very confusing. For consistency we should use EINVAL.

See kernel/sysctl.c:

/* We don't know if the next char is whitespace thus we may accept
 * invalid integers (e.g. 1234...a) or two integers instead of one
 * (e.g. 123...1). So lets not allow such large numbers. */
if (len == TMPBUFLEN - 1)
        return -EINVAL;

Christian

Re: + sysctl-return-einval-if-val-violates-minmax.patch added to -mm tree

[Luis Chamberlain]

On Mon, Feb 11, 2019 at 10:32:50PM +0100, Christian Brauner wrote:
> On Mon, Feb 11, 2019 at 01:25:23PM -0800, Luis Chamberlain wrote:
> > On Mon, Feb 11, 2019 at 10:19:19PM +0100, Christian Brauner wrote:
> > > On Tue, Feb 12, 2019 at 12:17:16AM +0300, Alexey Dobriyan wrote:
> > > > On Mon, Feb 11, 2019 at 01:06:32PM -0800, akpm@linux-foundation.org wrote:
> > > > 
> > > > > @@ -2848,8 +2848,10 @@ static int __do_proc_doulongvec_minmax(v
> > > > 
> > > > > -			if ((min && val < *min) || (max && val > *max))
> > > > > -				continue;
> > > > > +			if ((min && val < *min) || (max && val > *max)) {
> > > > > +				err = -EINVAL;
> > > > 
> > > > I was asked to return ERANGE in kstrto*().
> > > 
> > > I think we discussed ERANGE vs EINVAL and decided EINVAL because there
> > > was precedence for other sysctls already.
> > 
> > Can you do a proper audit and see?
> 
> If you look at proc_get_long() right now you can see that when the
> buffer we use to parse the number is exceeded we return EINVAL. In short
> if you do right now:
> 
> echo 1844674407370955161600000 > /proc/sys/fs/file-max
> 
> that would exceed the buffer in proc_get_long() and you already get
> EINVAL for all such cases. If we now change this to ERANGE we would
> return:
> 
> echo 18446744073709551616      > /proc/sys/fs/file-max -> ERANGE
> echo 1844674407370955161600000 > /proc/sys/fs/file-max -> EINVAL
> 
> which would be very confusing. For consistency we should use EINVAL.
> 
> See kernel/sysctl.c:
> 
> /* We don't know if the next char is whitespace thus we may accept
>  * invalid integers (e.g. 1234...a) or two integers instead of one
>  * (e.g. 123...1). So lets not allow such large numbers. */
> if (len == TMPBUFLEN - 1)
>         return -EINVAL;

Thanks this works for me.

  Luis