On Thu, Dec 06, 2007 at 01:36:16PM -0600, Greg Oster wrote:
> Manuel Bouyer writes:
> > Hi,
> > running raidctl -c on current with debug options I get:
> > panic: kernel diagnostic assertion "(l) == curlwp" failed: file "/dsk/l1/misc
> > /bouyer/current/src/sys/dev/dksubr.c", line 640
> > Stopped in pid 8.1 (raidctl) at netbsd:breakpoint+0x1:  ret
> > breakpoint() at netbsd:breakpoint+0x1
> > __kernassert() at netbsd:__kernassert+0x2d
> > dk_lookup() at netbsd:dk_lookup+0x198
> > rf_ConfigureDisk() at netbsd:rf_ConfigureDisk+0x10a
> > rf_ConfigureDisks() at netbsd:rf_ConfigureDisks+0xb9
> > rf_Configure() at netbsd:rf_Configure+0x6fc
> > raidioctl() at netbsd:raidioctl+0xf05
> > VOP_IOCTL() at netbsd:VOP_IOCTL+0x2e
> > vn_ioctl() at netbsd:vn_ioctl+0x79
> > sys_ioctl() at netbsd:sys_ioctl+0x1bc
> > syscall() at netbsd:syscall+0xe5
> > 
> > Reading the code I understand why we get a panic (rf_ConfigureDisk pass
> > raidPtr->engine_thread to dk_lookup()), the question is what would be
> > the fix ...
> 
> I'd like to know too....  IIRC, the code was changed (some years ago)
> to what it is now because curproc->lwp0->cred (or some such) being passed
> in wasn't sufficient to match the credentials of the kernel thread.  
> If it is now the case that l->l_cred == curlwp->l_cred == 
> some_other_kernel_lwp->l_cred , then we probably don't need to be 
> passing 'engine_thread' into dk_lookup() anymore, and just curlwp 
> will be sufficient... 

I think it's that dk_lookup() is now more strict than it was. You can't
call it with other credential than curproc(). Maybe it's the check in
dk_lookup() which is wrong, I don't know ...
Note that once the raid is setup and use autoconf, all works fine ...

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--