[Planetlab-devel] why FUSE should be safe, and why it isn't

[Russ Cox]

Miklos has found and fixed the bug that was causing
kernel Oops messages in 2.6.18.2.

The bug that wedges the kernel in 2.6.17 and 2.6.18.2
is a general Linux problem in the dentry cache.
Specifically, flushing the dentry cache seems to take
time quadratic in the depth of the directory entries.
If you create a completely linear directory structure
hundreds of thousands of directories deep, it will take
a very long time to clear that cache.

Allowing arbitrary NFS mounts should have the same
security problems as allowing arbitrary FUSE mounts
as far as this bug is concerned.

I do not know whether LKML has been told about the bug.

Russ