Hi,
I've just found the following behavior of my -current i386 box ( an old
current: 95 04 26). This box is a p60 whith a ncr53810 scsi adapter, and
3 cdroms, which are read from nfs. One of the cdrom make a read error
on a file. When readed from the local box, scsi layer says one or two
"medium error", and the reading process get a I/O error.
When this file is read from nfs, i get a page fault trap, or sometimes
a "Panic: free: multiple free", but the kernel always panic just after
the first message "medium error".
Has this been fixed ? Does someone have any idea ? I don't have much time
to track -current and debug for now. Below is the debbuger trace.
I thick this is not i386 specific, and that this behavior could happen
whith other disk devices giving i/o errors (i don't have any scsi disk giving
me errors to try this :).
Here is the debbuger trace (copyed by hand):
kernel: page fault trap, code=0
stopped at _free +0x122: movl 0x8(%edi),edx
db>trace
_free(f8759100,f81c08a8) at _free +0x122
_m_freem(f8759100) at _m_freem +0xc1
_nfsrv_read(f874dc00,f8759100,f8759180,f87591d8,f874dc2c) at _nfsrv_read +0xb6d
_nfssvc_nfsd(f7b8fea0,14cb0,f874ff00,8,f874ff00 at _nfssvc_nfsd +0x3eb
_nfssvc(f874ff00,f7bfff98,f7bfff90,0,2) at _nfssvc +0x410
_syscall() at _syscall +0x20a
--- syscall (number 155) ---

--
Manuel Bouyer, Ecole Nationale Superieure de Techniques Avancees, Paris
email: bouyer@ensta.fr
--