On Fri, Sep 12, 2014 at 3:08 PM, Josef Bacik <jbacik@xxxxxx> wrote:
>
> "New hash" is the number you want. Not as bad as originally, not as good as
> replacing fold_hash with hash_64 and not changing d_hash. Thanks,
Ok, so that hash folding really does hurt. And now that I look at it,
I understand quite well. It's extra horrible, because it does the
addition of bits on a byte boundary. So it very easily causes bits to
basically cancel out when the first four bytes have a high commonality
with the next four bytes (like your numeric names do - lots of ascii
decimal numbers). So yeah, we need to fix fold_hash(), no question
about it.
I get the feeling that "hash_long()" is a bit excessive and we end up
mixing bits twice, but I guess it's the simple approach.
So with that, the kernel patch would look something like the attached.
Totally untested. Does it fix the regression you saw in addition to
looking more reasonable in your test-program?
Linus
fs/dcache.c | 3 +--
fs/namei.c | 4 ++--
2 files changed, 3 insertions(+), 4 deletions(-)
diff --git a/fs/dcache.c b/fs/dcache.c
index d30ce699ae4b..4023e77b800e 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -106,8 +106,7 @@ static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
unsigned int hash)
{
hash += (unsigned long) parent / L1_CACHE_BYTES;
- hash = hash + (hash >> d_hash_shift);
- return dentry_hashtable + (hash & d_hash_mask);
+ return dentry_hashtable + hash_32(hash, d_hash_shift);
}
/* Statistics gathering. */
diff --git a/fs/namei.c b/fs/namei.c
index a996bb48dfab..229235862e50 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -34,6 +34,7 @@
#include <linux/device_cgroup.h>
#include <linux/fs_struct.h>
#include <linux/posix_acl.h>
+#include <linux/hash.h>
#include <asm/uaccess.h>
#include "internal.h"
@@ -1634,8 +1635,7 @@ static inline int nested_symlink(struct path *path, struct nameidata *nd)
static inline unsigned int fold_hash(unsigned long hash)
{
- hash += hash >> (8*sizeof(int));
- return hash;
+ return hash_64(hash, 32);
}
#else /* 32-bit case */
