Re: WAPBL silent file corruption

[Juergen Hannken-Illjes <hannken%eis.cs.tu-bs.de@localhost>]

On Fri, Jan 21, 2011 at 02:32:38PM +0100, Rhialto wrote:
> I was running build.sh on my amd64 laptop last night. It is running a
> -current kernel from last October, since release-kernels don't properly
> support its hdaudio. It has WAPBL enabled. Userland is 5.1.
> 
> Sometimes it suffers from spontaneous reboots. That is annoying. But it
> gets worse.
> 
> I had one of those reboots while build.sh was running. When I restarted
> it this morning, a whole bunch of object files was corrupted. I removed
> a few directories worth of files, because many of them were filled with
> 00 instead of being valid object files.
> 
> A re-restarted build failed again, since some library was apparenty
> corrupt too. (The symptom was unresolved symbols when linking)
> 
> If WAPBL doesn't guarantee that my files are either correct or absent, I
> want Soft Dependencies back! Suppose I had made a backup of silently
> corrupted files, discarding valid originals!

Maybe revision 1.38 of sys/kern/vfs_wapbl.c could help:

date: 2010/11/09 16:30:26;  author: hannken;  state: Exp;  lines: +8 -5
Wapbl_register_deallocation(): the taken reader lock is not sufficient to
protect wl_dealloc* members.  Take the mutex here and change the lock
requirements of these fields to "writer lock or mutex".

This error lead to file system corruption and "freeing free block" panics.

-- 
Juergen Hannken-Illjes - hannken%eis.cs.tu-bs.de@localhost - TU Braunschweig 
(Germany)