sysconf(_SC_IOV_MAX) returns -1 in f27 and f28/rawhide (glibc-2.26-8)

[Kaleb S. KEITHLEY]

perusing the glibc source and headers it seems to come down to the
    #undef __IOV_MAX
line in bits/uio_lim.h.

The Changelog (2017-06-14) speaks rather obliquely to the issue, but I 
can't quite figure out if

   sysconf(_SC_IOV_MAX);

now returning -1 (errno = 0) is an unintentional side effect of the 
changes or not. (I don't follow glibc development to know what the 
history is behind this change.) In f26, glibc-2.25-10 it returns 1024.
Off hand I don't see any way that glibc could be built now to make the 
call return anything but -1.

Anyone know more of the story? Or want to venture a guess?

Thanks

--

Kaleb

[Florian Weimer]

On 10/19/2017 04:45 PM, Kaleb S. KEITHLEY wrote:
>  
>  perusing the glibc source and headers it seems to come down to the
>      #undef __IOV_MAX
>  line in bits/uio_lim.h.
>  
>  The Changelog (2017-06-14) speaks rather obliquely to the issue, but I 
>  can't quite figure out if
>  
>     sysconf(_SC_IOV_MAX);
>  
>  now returning -1 (errno = 0) is an unintentional side effect of the 
>  changes or not. (I don't follow glibc development to know what the 
>  history is behind this change.) In f26, glibc-2.25-10 it returns 1024.
>  Off hand I don't see any way that glibc could be built now to make the 
>  call return anything but -1.
>  
>  Anyone know more of the story? Or want to venture a guess? 
Looks like a simple bug.  I'm investigating.

Do you know Linux actually has a limit?

Thanks,
Florian

[Florian Weimer]

On 10/19/2017 05:11 PM, Florian Weimer wrote:
>  On 10/19/2017 04:45 PM, Kaleb S. KEITHLEY wrote:
> >
> > perusing the glibc source and headers it seems to come down to the
> >     #undef __IOV_MAX
> > line in bits/uio_lim.h.
> >
> > The Changelog (2017-06-14) speaks rather obliquely to the issue, but I 
> > can't quite figure out if
> >
> >    sysconf(_SC_IOV_MAX);
> >
> > now returning -1 (errno = 0) is an unintentional side effect of the 
> > changes or not. (I don't follow glibc development to know what the 
> > history is behind this change.) In f26, glibc-2.25-10 it returns 1024.
> > Off hand I don't see any way that glibc could be built now to make the 
> > call return anything but -1.
> >
> > Anyone know more of the story? Or want to venture a guess?
>  
>  Looks like a simple bug.  I'm investigating.
>  
>  Do you know Linux actually has a limit? 
It's still a real limit.  I've posted a fix upstream and will update 
rawhide and F27 once it's reviewed there.

Thanks,
Florian