amd64/134786: [patch] vfs.bufspace sysctl wideness on amd64

[John Baldwin]

The following reply was made to PR amd64/134786; it has been noted by GNATS.

From: John Baldwin <jhb at freebsd.org>
To: freebsd-amd64 at freebsd.org
Cc: Emil Mikulic <emikulic at gmail.com>,
 FreeBSD-gnats-submit at freebsd.org
Subject: Re: amd64/134786: [patch] vfs.bufspace sysctl wideness on amd64
Date: Thu, 21 May 2009 12:16:26 -0400

 On Thursday 21 May 2009 10:48:25 am Emil Mikulic wrote:
 > 
 > >Number:         134786
 > >Category:       amd64
 > >Synopsis:       [patch] vfs.bufspace sysctl wideness on amd64
 > >Confidential:   no
 > >Severity:       non-critical
 > >Priority:       low
 > >Responsible:    freebsd-amd64
 > >State:          open
 > >Quarter:        
 > >Keywords:       
 > >Date-Required:
 > >Class:          sw-bug
 > >Submitter-Id:   current-users
 > >Arrival-Date:   Thu May 21 14:50:04 UTC 2009
 > >Closed-Date:
 > >Last-Modified:
 > >Originator:     Emil Mikulic
 > >Release:        FreeBSD 8-CURRENT
 > >Organization:
 > >Environment:
 > >Description:
 > 
 > On amd64, providing a 64-bit buffer to get the vfs.bufspace sysctl will
 > return a 64-bit (long) quantity, but providing a *larger* buffer will
 > only yield a 32-bit (int) quantity.
 > 
 > >How-To-Repeat:
 > 
 > len = 8;
 > sysctlbyname("vfs.bufspace", &buf, &len, NULL, 0);
 > /* len is still 8 */
 > 
 > len = 10;
 > sysctlbyname("vfs.bufspace", &buf, &len, NULL, 0);
 > /* len is 4! */
 > 
 > >Fix:
 > 
 > --- sys/kern/vfs_bio.c	2009-04-17 10:01:39.000000000 +0000
 > +++ sys/kern/vfs_bio.c.2	2009-05-09 09:27:16.000000000 +0000
 > @@ -288,15 +288,15 @@
 >      defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
 >  static int
 >  sysctl_bufspace(SYSCTL_HANDLER_ARGS)
 >  {
 >  	long lvalue;
 >  	int ivalue;
 >  
 > -	if (sizeof(int) == sizeof(long) || req->oldlen == sizeof(long))
 > +	if (sizeof(int) == sizeof(long) || req->oldlen >= sizeof(long))
 >  		return (sysctl_handle_long(oidp, arg1, arg2, req));
 >  	lvalue = *(long *)arg1;
 >  	if (lvalue > INT_MAX)
 >  		/* On overflow, still write out a long to trigger ENOMEM. */
 >  		return (sysctl_handle_long(oidp, &lvalue, 0, req));
 >  	ivalue = lvalue;
 >  	return (sysctl_handle_int(oidp, &ivalue, 0, req));
 
 Hummm.  I guess that is correct, though that is a bit odd.
 
 -- 
 John Baldwin