Re: [Announce] Strace Test

[Nikita Danilov <Nikita@Namesys.COM>]

Andrew Morton writes:
 > Nikita Danilov <Nikita@Namesys.COM> wrote:
 > >
 > >  > Strace Test uses a modified version of strace 4.5.1.  
 > >   > Instead of printing out information about system calls, 
 > >   > the modified version calls the syscalls with improper 
 > >   > values.
 > > 
 > >  It immediately DoSes kernel by calling sys_sysctl() with huge nlen:
 > >  printk() consumes all CPU.
 > 
 > Something like this?

On slow console (serial kgdb) this still would be problematic. I think
printk_ratelimit() is needed. But why this loop is needed at all? It
seems strange that syscall prints its arguments instead of just
returning -EINVAL.

 > 
 > --- 25/kernel/sysctl.c~sysctl-nlen-check	2004-02-19 02:36:20.000000000 -0800
 > +++ 25-akpm/kernel/sysctl.c	2004-02-19 02:37:40.000000000 -0800
 > @@ -913,6 +913,9 @@ asmlinkage long sys_sysctl(struct __sysc
 >  
 >  	if (copy_from_user(&tmp, args, sizeof(tmp)))
 >  		return -EFAULT;
 > +
 > +	if (tmp.nlen < 0 || tmp.nlen > CTL_MAXNAME)
 > +		return -EINVAL;
 >  	
 >  	if (tmp.nlen != 2 || copy_from_user(name, tmp.name, sizeof(name)) ||
 >  	    name[0] != CTL_KERN || name[1] != KERN_VERSION) { 

Nikita.

 > 
 > _
 >