Re: + sysctl-remove-binary-sysctl-support-where-it-clearly-doesnt-work.patch added to -mm tree

[Alexey Dobriyan <adobriyan@sw.ru>]

On Fri, Aug 10, 2007 at 03:26:52AM -0600, Eric W. Biederman wrote:
> Alexey Dobriyan <adobriyan@sw.ru> writes:
> 
> >> @@ -1124,7 +1118,6 @@ static struct ctl_table fs_table[] = {
> >>  		.proc_handler	= &proc_dointvec,
> >>  	},
> >>  	{
> >> -		.ctl_name	= FS_NRFILE,
> >>  		.procname	= "file-nr",
> >>  		.data		= &files_stat,
> >>  		.maxlen		= 3*sizeof(int),
> >
> > Why? It will work just fine through default sysctl(2) writeback.
> 
> 
> Well write doesn't happen.  But even more so proc_nr_files() dynamically
> generates files_stat.nr_files.  That doesn't happen on the generic
> sysctl path, and thus it's broken.

I see now, thanks. CC'ing Dipankar who probably want to fix this.
Dipankar, fs.file-nr always contains stale data in nr_files field unless
you regenerate it by reading /proc/sys/fs/file-nr :)

> Yes.  I'm being picky, because at some point in the past before
> that was a per cpu variable the code worked, and won't look broken
> now unless you examine the contents of the data.

More than year passed, nobody noticed until now, probably FS_NRFILE
should go.