On Tue, May 10, 2016 at 04:27:05PM +0100, Al Viro wrote:
> NAK on lookup_hash(). If nothing else, you need either
> lookup_one_len_unlocked() or its "we already know hash" variant (assuming it's
> worth bothering with). That locking the parent is potentially a lot costlier
> than recalculating the (default) hash function. Especially in -next, where
> lookup_one_len_unlocked() will take the lock shared if it has to take it at
> all, but the mainline one also has a good chance to avoid taking the lock.
Can't use lookup_one_len_unlocked() because point of the fix is to lose the
permission check.
So how about this?
| struct dentry *lookup_hash(const struct qstr *name, struct dentry *base)
| {
| struct dentry *ret;
|
| ret = lookup_dcache(name, base, 0);
| if (!ret)
| ret = lookup_slow(name, base, 0);
|
| return ret;
| }
Updated pull request:
git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/vfs.git overlayfs-linus
---
Miklos Szeredi (3):
vfs: rename: check backing inode being equal
vfs: export lookup_hash() to modules
ovl: ignore permissions on underlying lookup
---
fs/namei.c | 40 +++++++++++++++++++++++++++++++++++++++-
fs/overlayfs/super.c | 4 +---
include/linux/namei.h | 2 ++
3 files changed, 42 insertions(+), 4 deletions(-)
