Re: [RFC] atomic open(..., O_CREAT | ...)

[Miklos Szeredi <miklos@szeredi.hu>]

> > > There is quite a bit of code out there that assumes it is free to stuff
> > > things into nd->mnt and nd->dentry. Some of it is Al Viro's code, some
> > > of it is from other people.
> > > For instance, the ESTALE handling will just save nd->mnt/nd->dentry
> > > before calling __link_path_walk(), then restore + retry if the ESTALE
> > > error comes up.
> > 
> > Yeah, but how is that relevant to the fact, that after
> > path_release_*() _nothing_ will be valid in the nameidata, not
> > nd->mnt, not nd->dentry, and not nd->intent.open.file.  So what's the
> > point in setting it to NULL if it must never be used anyway?
> 
> path_release() does _not_ invalidate the nameidata. Look for instance at
> __emul_lookup_dentry(), which clearly makes use of that fact.

Trond, wake up!  __emul_lookup_dentry() does nothing of the sort.
Neither does anything else.  In theory it could, but that's not a
reason to do a confusing thing like that.

> Firstly, the open_namei() flags field is not a "permissions" field. It
> contains open mode information. The calculation of the open permissions
> flags is done by open_namei() itself.

Based on flags.  It's just a FMODE_* -> MAY_* transformation

> Secondly, what advantage is there in allowing callers of open_namei() to
> be able to override the MAY_WRITE check when doing open(O_TRUNC)? This
> is a calculation that should be done _once_ in order to always get it
> right, and it should therefore be done in open_namei() together with the
> rest of the permissions calculation.

I think the _only_ caller of open_namei() is filp_open(), so this is
not much of an issue, but yeah, you could do it that way too.

Or you could initialize nameidata from filp_open().

Miklos
-
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/