b2648d512e
Switch all users of lock_parent() to the approach used by ->unlink() and ->rmdir() - instead of playing with dget_parent() of underlying dentry of child, * start with ecryptfs dentry of child. * find underlying dentries for that dentry and its parent (which is stable, since the parent directory in upper layer is held at least shared). No need to pin them, they are already pinned by ecryptfs dentries. * lock the inode of undelying directory of parent * check if it's the parent of underlying dentry of child. ->d_parent of underlying dentry of child might be unstable. However, result of its comparison with underlying dentry of parent *is* stable now. Turn that into replacement of lock_parent(), convert the existing callers of lock_parent() to that, along with ecryptfs_unlink() and ecryptfs_rmdir(). Callers need only the underlying dentry of child and inode of underlying dentry of parent, so lock_parent() passes those to the caller now. Note that underlying directory is locked in any case, success or failure. That approach does not need a primitive for unlocking - we hadn't grabbed any dentry references, so all we need is to unlock the underlying directory inode. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> |
||
---|---|---|
.. | ||
crypto.c | ||
debug.c | ||
dentry.c | ||
ecryptfs_kernel.h | ||
file.c | ||
inode.c | ||
Kconfig | ||
keystore.c | ||
kthread.c | ||
main.c | ||
Makefile | ||
messaging.c | ||
miscdev.c | ||
mmap.c | ||
read_write.c | ||
super.c |