Never mind this. I overlooked an erroneously wrong Attr result that
On Wed, Sep 14, 2011 at 1:52 AM, Han-Wen Nienhuys <hanwenn@...> wrote:
> Hi there,
> When running under load, I'm seeing situations where 0 is returned for
> stats in the st_ino field. It's not deterministic, so it is looking
> like a race condition of sorts.
> The symptom is that 'rm -rf' says
> rm: WARNING: Circular directory structure.
> This almost certainly means that you have a corrupted file system.
> NOTIFY YOUR SYSTEM MANAGER.
> The following directory is part of the cycle:
> I added some debug prints to the rm source, exactly here,
> the printouts say that there were two files that returned 0 for their
> inode numbers. The mtime for these files looked sane.
> I checked by running debug, and my FUSE filesystem is returning a
> normal number as nodeid, and it uses a high (30 second) timeout, so
> AFAICS, that is the number that should be returned to user-space. How
> could the kernel have returned 0 for the inode number?
> I'm running compiles on 12 Go-FUSE file systems (connected to the same
> daemon binary) in parallel on a 6 core machine. The OS is Ubuntu
> Linux oats.bhz.corp.google.com 2.6.32-gg508-generic #gg508-Ubuntu
> SMP Tue May 24 05:09:12 PDT 2011 x86_64 GNU/Linux.
> Han-Wen Nienhuys - hanwen@... - http://www.xs4all.nl/~hanwen
Han-Wen Nienhuys - hanwen@... - http://www.xs4all.nl/~hanwen