(Illustration by Gaich Muramatsu)
I have seen indications of a similar bug. In my case, it has been 'vref used where vget required' on NetBSD-1.4.2, tending to happen when I do a 'cfs mkm'. I think the crash happens when that item in the minicache gets freed/flushed. My ideas are some sort of log of all operations on cfs0 that can be examined post-mortem implement some stronger invariant checking that can verify the vnode refcounts put in recovery code for these cases that can adjust refcounts rather than panicing, and seeing what the pattern of errors is [dangerous and risks some corruption, but might provide useful insight] a thorough code review of the kernel code with explicit documentation of all the reference-counting rules and any other invariants I won't claim that any of these are easy. Greg Troxel <gdt_at_ir.bbn.com>Received on 2000-11-22 08:18:45