Re: vfs_unlink() >=2.5.5-pre1 question

Alexander Viro (viro@math.psu.edu)
Sat, 30 Mar 2002 13:28:13 -0500 (EST)


On Sat, 30 Mar 2002, Mike Galbraith wrote:

> On Sat, 30 Mar 2002, Mike Galbraith wrote:
>
> > Hi,
> >
> > d_delete() doesn't appear to ever create negative dentries when
> > called via vfs_unlink() due to the extra reference on the dentry.
> > In fact, a printk() in the d_delete() spot never ever triggers...
>
> Well shoot. I guess I've chased this about as far as I can, and
> hope this thread wasn't a total waste. I found a better way to
> get my rm -r to work as before fwiw. Rewinding the directory on
> seek failure (yeah, could do in three lines, but not the point)
> works, but is kinda b0rken. I think the only interesting thing
> in the below is the FIXME :)) but I'll post it anyway.

Your patch is broken. FWIW, there are several real issues:
a) d_delete() being called too early in vfs_unlink(). Not a big
deal, it's easy to move outside of dget()/dput(). However, you _can't_
expect unlink() to make dentry negative. It's always possible that it
will be left positive and unhashed - that's what we have to do if file
we are unlinking is opened.
b) rm -rf expecting offsets in directory to stay stable after
unlink(). B0rken, complain to GNU folks. Sorry, I'm not touching that
code - GNU fileutils source is too yucky.
c) dcache_readdir() behaviour. There was an old patch that makes
it slightly more forgiving; I'll dig it out.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/