Created attachment 45722 [details] screen snapshot I got a kernel dump at umount time of a huge reiserfs partition. It said Null pointer dereference at 0000000000000010 IP: [ffffffffa01dc818] open_xa_dir+0x2d/0x1b0 [reiserfs] : The config file and a photo with a complete stack trace are attached. reiserfsck of this partition showed a lot of files to sync due to the missing umount, but no other problems. Kernel is 2.6.37 (amd64). Please mail if I can help to track this down.
Created attachment 45732 [details] config file
Created attachment 46052 [details] reiserfs: Fix crash during umount There is a conflict between shrink_dcache_for_umount_subtree and how xattrs are cleaned up for deleted files at umount. shrink_dcache_for_umount_subtree wants to ensure that all dentries for a file system have been evicted and it walks the dentry tree from the root of the file system to do this. It will BUG if there are any dentries left with elevated counts. The reiserfs xattr infrastructure caches two dentries. One is for .reiserfs_priv and the other is for .reiserfs_priv/xattrs. When shrink_dcache_for_umount_subtree goes through, it will BUG on these dentries if they aren't freed. OTOH, it will Oops in reiserfs_delete_xattrs if they are due to the xattr code needing to walk the list of xattrs for a file undergoing delayed deletion. This will end up loading up other dentries and possibly queue up more delayed deletions. This patch detaches the tree under .reiserfs_priv during ->kill_sb. This allows shrink_dcache_for_umount_subtree to complete successfully as well as keeps around the .reiserfs_priv tree for deletion handling. After the regular cycle is completed, it will call shrink_dcache_for_umount_subtree again itself to ensure that the xattrs are cleaned up.