kern/61513: kernel gets into a panic to put invalid value in setsockopt

[Redmar Kerkhoff]

The following reply was made to PR kern/61513; it has been noted by GNATS.

From: Redmar Kerkhoff <redmar at interrupt.nl>
To: freebsd-gnats-submit at FreeBSD.org, katsuhisa.abe at nifty.com
Cc: redmar at interrupt.nl
Subject: Re: kern/61513: kernel gets into a panic to put invalid value in setsockopt
Date: Fri, 23 Jan 2004 15:44:42 +0100

 >Description:
 "When I put unproper value (uninitialzed pointer) into the 4th argument, "optval" for setsockopt,
 then kernel starts to reboot (not generate core)."
 
 i have traced the problem down to ip6_setpktoption() but the actual
 problem lies in unitialized pointer assignment in setsockopt.
 (found in /sys/kern/uipc_syscalls.c) The setsockopt call is forwarding
 de optval information from userland without handling any address errors.
 ( sopt.sopt_val = uap->val; ) finally exploding in this case at 
 ip6_setpktoption(). 
 
 i looked at netbsd and openbsd and they we're both resistent to
 this bug. the fix is more or less derived from their getsockopt handling.
 
 here's the fix against 5.2-RELEASE. (but can also be applied to current afaik)
 (the only little problem now is the double assignment of sopt.sopt_val, options are
 removing the assignment in the mutex lock or putting the copyin in the mutex
 lock )
 
 
 >Fix:
 
 
 --- uipc_syscalls.orig  Fri Jan 23 10:38:46 2004
 +++ uipc_syscalls.c     Fri Jan 23 10:27:33 2004
 @@ -1308,6 +1308,10 @@
                 return (EFAULT);
         if (uap->valsize < 0)
                 return (EINVAL);
 +       
 +       error = copyin(uap->val, &sopt.sopt_val, sizeof(sopt.sopt_val));
 +       if (error)
 +               return (error);
  
         mtx_lock(&Giant);
         if ((error = fgetsock(td, uap->s, &so, NULL)) == 0) {