fs: fix NULL dereference due to data race in prepend_path()

[ Upstream commit 09cad07 ]

Fix data race in prepend_path() with re-reading mnt->mnt_ns twice
without holding the lock.

is_mounted() does check for NULL, but is_anon_ns(mnt->mnt_ns) might
re-read the pointer again which could be NULL already, if in between
reads one of kern_unmount()/kern_unmount_array()/umount_tree() sets
mnt->mnt_ns to NULL.

This is seen in production with the following stack trace:

  BUG: kernel NULL pointer dereference, address: 0000000000000048
  ...
  RIP: 0010:prepend_path.isra.4+0x1ce/0x2e0
  Call Trace:
    d_path+0xe6/0x150
    proc_pid_readlink+0x8f/0x100
    vfs_readlink+0xf8/0x110
    do_readlinkat+0xfd/0x120
    __x64_sys_readlinkat+0x1a/0x20
    do_syscall_64+0x42/0x110
    entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes: f2683bd ("[PATCH] fix d_absolute_path() interplay with fsmount()")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>

fs/d_path.c

@@ -102,6 +102,8 @@ static int prepend_path(const struct path *path,
 
 		if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
 			struct mount *parent = READ_ONCE(mnt->mnt_parent);
+			struct mnt_namespace *mnt_ns;
+
 			/* Escaped? */
 			if (dentry != vfsmnt->mnt_root) {
 				bptr = *buffer;
@@ -116,7 +118,9 @@ static int prepend_path(const struct path *path,
 				vfsmnt = &mnt->mnt;
 				continue;
 			}
-			if (is_mounted(vfsmnt) && !is_anon_ns(mnt->mnt_ns))
+			mnt_ns = READ_ONCE(mnt->mnt_ns);
+			/* open-coded is_mounted() to use local mnt_ns */
+			if (!IS_ERR_OR_NULL(mnt_ns) && !is_anon_ns(mnt_ns))
 				error = 1;	// absolute root
 			else
 				error = 2;	// detached or not attached yet