cvs commit: src/sys/vfs/hammer hammer.h hammer_flusher.c hammer_inode.c hammer_io.c hammer_object.c hammer_prune.c hammer_reblock.c hammer_recover.c hammer_transaction.c hammer_vfsops.c hammer_vnops.c

[Matthew Dillon]

:I think an "extern int hidirtybuffers" will do. It's declared in vfs_bio.c.
:
:Regards,
:
:   Michael

    Fixed!

    53D coming today.  There are still a few assertions related to
    reblocking and pruning, and there may be one read/flush race left
    too.  The assertions will be dealt with today.

    fsx alone doesn't report problems, a buildworld loop alone doesn't
    report problems, but a buildworld loop + blogbench caused an internal
    compiler error once during my testing (which typically means the
    demand-paged executable got corrupted).

    I think the race is related to data records in memory being flushed
    to disk while the frontend is simultaniously trying to do a read.  It
    may take a few days to track it down.

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>