XFS SoftAssert when doing hard links in AFS?

[Christopher Allen Wing]

Love:

Thanks for writing back.

> We redid the installnode scheme in before 0.30, this way its more sane (and
> shorter). But there is a leek somewhere in there, not serious, since we
> find (and correct) the misstake.
> 
> I tried to do a small c-program that trigged the bug (like we almost always
> do, see inte tests/), but I can't reproduce it. I have also tried to run
> pine, still no problem.

I tried to write a c program that reproduces the same system calls that
pine did, but the problem didn't occur.

Maybe it's not related to the hard links directly, but to the actual
mailbox write itself. The mailbox is about 2 megabytes and gets written to
and then flushed to AFS immediately after the sillylocking.

> Can you try to do a shortest complete strace that triggers the bug and put
> in on a ftp/web-site ? It there is a some program that triggers the bug, it
> would be even better.

The darn thing doesn't want to produce the assert now. When it starts
acting up again I'll capture the trace.

I'll try and figure out exactly where the panic occurs if
possible.

> Another way to find the buf is to turn on xfs-debugging and looking 
> at the logs.


How would I go about doing this?


Thanks,

Chris Wing
wingc at engin.umich.edu



p.s what is the plan to fix libroken? are you planning to just toss in the
roken from kth-krb and clean it up from there?