On Fri, 24 Apr 2009, Simon Burge wrote:
Stephen Borrill wrote:Sorry, that's not true, I miscounted lines. The problem is actually that=20 the ffs_hashalloc() call in ffs_alloc() is returning 0.I recall finding a 32-bit overflow problem in the block allocation code last year some time... I wonder if you're not hitting some similar problem.
Yes, I am. Using gdb and single step rump_ffs, I found the offending lines. I've commited a fix to -current and a pullup has been submitted for netbsd-5. It will not make 5.0 though, so if anyone requires >2TB filesystems it will be best to use a 5.0_STABLE kernel once the fix has gone in.
In the meantime the patch can be found at: ftp://ftp.netbsd.org/pub/NetBSD/misc/sborrill/ffs_alloc.diff -- Stephen