svn commit: r364478 - head/sys/kern

[Mateusz Guzik]

Author: mjg
Date: Sat Aug 22 06:56:04 2020
New Revision: 364478
URL: https://svnweb.freebsd.org/changeset/base/364478

Log:
  vfs: add a work around for vp_crossmp bug to realpath
  
  The actual bug is not yet addressed as it will get much easier after other
  problems are addressed (most notably rename contract).
  
  The only affected in-tree consumer is realpath. Everyone else happens to be
  performing lookups within a mount point, having a side effect of ni_dvp being
  set to mount point's root vnode in the worst case.
  
  Reported by:	pho

Modified:
  head/sys/kern/vfs_cache.c
  head/sys/kern/vfs_lookup.c

Modified: head/sys/kern/vfs_cache.c
==============================================================================
--- head/sys/kern/vfs_cache.c	Sat Aug 22 04:07:44 2020	(r364477)
+++ head/sys/kern/vfs_cache.c	Sat Aug 22 06:56:04 2020	(r364478)
@@ -2811,6 +2811,7 @@ vn_fullpath_hardlink(struct thread *td, struct nameida
 	size_t addend;
 	int error;
 	bool slash_prefixed;
+	enum vtype type;
 
 	if (*buflen < 2)
 		return (EINVAL);
@@ -2824,7 +2825,31 @@ vn_fullpath_hardlink(struct thread *td, struct nameida
 
 	addend = 0;
 	vp = ndp->ni_vp;
-	if (vp->v_type != VDIR) {
+	/*
+	 * Check for VBAD to work around the vp_crossmp bug in lookup().
+	 *
+	 * For example consider tmpfs on /tmp and realpath /tmp. ni_vp will be
+	 * set to mount point's root vnode while ni_dvp will be vp_crossmp.
+	 * If the type is VDIR (like in this very case) we can skip looking
+	 * at ni_dvp in the first place. However, since vnodes get passed here
+	 * unlocked the target may transition to doomed state (type == VBAD)
+	 * before we get to evaluate the condition. If this happens, we will
+	 * populate part of the buffer and descend to vn_fullpath_dir with
+	 * vp == vp_crossmp. Prevent the problem by checking for VBAD.
+	 *
+	 * This should be atomic_load(&vp->v_type) but it is ilegal to take
+	 * an address of a bit field, even if said field is sized to char.
+	 * Work around the problem by reading the value into a full-sized enum
+	 * and then re-reading it with atomic_load which will still prevent
+	 * the compiler from re-reading down the road.
+	 */
+	type = vp->v_type;
+	type = atomic_load_int(&type);
+	if (type == VBAD) {
+		error = ENOENT;
+		goto out_bad;
+	}
+	if (type != VDIR) {
 		cnp = &ndp->ni_cnd;
 		addend = cnp->cn_namelen + 2;
 		if (*buflen < addend) {

Modified: head/sys/kern/vfs_lookup.c
==============================================================================
--- head/sys/kern/vfs_lookup.c	Sat Aug 22 04:07:44 2020	(r364477)
+++ head/sys/kern/vfs_lookup.c	Sat Aug 22 06:56:04 2020	(r364478)
@@ -1209,6 +1209,11 @@ nextname:
 		VOP_UNLOCK(dp);
 success:
 	/*
+	 * FIXME: for lookups which only cross a mount point to fetch the
+	 * root vnode, ni_dvp will be set to vp_crossmp. This can be a problem
+	 * if either WANTPARENT or LOCKPARENT is set.
+	 */
+	/*
 	 * Because of shared lookup we may have the vnode shared locked, but
 	 * the caller may want it to be exclusively locked.
 	 */