When looking for identical xattr blocks to share, we were not comparing
the name_index fields.  This could lead to false sharing when two xattr
blocks ended up with identical attribute names and values, and the only
difference was in name_index.  Specifically this could trigger with
default acls.  Because acls are cached, the bug was hidden until the
next reload of the affected inode.

  $ mkdir -m 700 a b
  $ setfacl -m u:bin:rwx a
        < acl of a goes in the mbcache

  $ setfacl -dm u:bin:rwx b
        < acl of b differs only in name_index, so a's acl is reused

  $ getfacl b
        < shows the result from the inode cache

  < empty inode cache (remount, etc.)

  $ getfacl b
        < shows an access acl instead of a default acl.

Signed-off-by: Andreas Gruenbacher <[EMAIL PROTECTED]>

Index: linux-2.6.11-rc3/fs/ext2/xattr.c
===================================================================
--- linux-2.6.11-rc3.orig/fs/ext2/xattr.c
+++ linux-2.6.11-rc3/fs/ext2/xattr.c
@@ -881,6 +881,7 @@ ext2_xattr_cmp(struct ext2_xattr_header 
                if (IS_LAST_ENTRY(entry2))
                        return 1;
                if (entry1->e_hash != entry2->e_hash ||
+                   entry1->e_name_index != entry2->e_name_index ||
                    entry1->e_name_len != entry2->e_name_len ||
                    entry1->e_value_size != entry2->e_value_size ||
                    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
Index: linux-2.6.11-rc3/fs/ext3/xattr.c
===================================================================
--- linux-2.6.11-rc3.orig/fs/ext3/xattr.c
+++ linux-2.6.11-rc3/fs/ext3/xattr.c
@@ -1162,6 +1162,7 @@ ext3_xattr_cmp(struct ext3_xattr_header 
                if (IS_LAST_ENTRY(entry2))
                        return 1;
                if (entry1->e_hash != entry2->e_hash ||
+                   entry1->e_name_index != entry2->e_name_index ||
                    entry1->e_name_len != entry2->e_name_len ||
                    entry1->e_value_size != entry2->e_value_size ||
                    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))


Regards,
-- 
Andreas Gruenbacher <[EMAIL PROTECTED]>
SUSE Labs, SUSE LINUX GMBH

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

Reply via email to