git: 592de68e328b - stable/13 - namei: Make stackable filesystems check harder for jail roots

From: Mark Johnston <markj_at_FreeBSD.org>
Date: Fri, 21 Nov 2025 15:19:23 UTC
The branch stable/13 has been updated by markj:

URL: https://cgit.FreeBSD.org/src/commit/?id=592de68e328b4edf8acdb2a0b0b66a813747b440

commit 592de68e328b4edf8acdb2a0b0b66a813747b440
Author:     Mark Johnston <markj@FreeBSD.org>
AuthorDate: 2025-05-23 12:52:24 +0000
Commit:     Mark Johnston <markj@FreeBSD.org>
CommitDate: 2025-11-19 20:47:50 +0000

    namei: Make stackable filesystems check harder for jail roots
    
    Suppose a process has its cwd pointing to a nullfs directory, where the
    lower directory is also visible in the jail's filesystem namespace.
    Suppose that the lower directory vnode is moved out from under the
    nullfs mount.  The nullfs vnode still shadows the lower vnode, and
    dotdot lookups relative to that directory will instantiate new nullfs
    vnodes outside of the nullfs mountpoint, effectively shadowing the lower
    filesystem.
    
    This phenomenon can be abused to escape a chroot, since the nullfs
    vnodes instantiated by these dotdot lookups defeat the root vnode check
    in vfs_lookup(), which uses vnode pointer equality to test for the
    process root.
    
    Fix this by extending nullfs and unionfs to perform the same check,
    exploiting the fact that the passed componentname is embedded in a
    nameidata structure to avoid changing the VOP_LOOKUP interface.  That
    is, add a flag to indicate that containerof can be used to get the full
    nameidata structure, and perform the root vnode check on the lower vnode
    when performing a dotdot lookup.
    
    PR:             262180
    Reviewed by:    olce, kib
    MFC after:      2 weeks
    Differential Revision:  https://reviews.freebsd.org/D50418
    
    (cherry picked from commit 7587f6d4840f8d363e457cddc14c184cf1fe7cc1)
---
 sys/fs/nullfs/null_vnops.c   | 28 +++++++++++++++++----------
 sys/fs/unionfs/union_vnops.c | 19 ++++++++++++++++++
 sys/kern/vfs_cache.c         | 11 +----------
 sys/kern/vfs_lookup.c        | 46 +++++++++++++++++++++++++++++++-------------
 sys/sys/namei.h              |  5 ++++-
 5 files changed, 75 insertions(+), 34 deletions(-)

diff --git a/sys/fs/nullfs/null_vnops.c b/sys/fs/nullfs/null_vnops.c
index f2c426d41f61..90fe043f8b33 100644
--- a/sys/fs/nullfs/null_vnops.c
+++ b/sys/fs/nullfs/null_vnops.c
@@ -407,17 +407,25 @@ null_lookup(struct vop_lookup_args *ap)
 
 	/*
 	 * Renames in the lower mounts might create an inconsistent
-	 * configuration where lower vnode is moved out of the
-	 * directory tree remounted by our null mount.  Do not try to
-	 * handle it fancy, just avoid VOP_LOOKUP() with DOTDOT name
-	 * which cannot be handled by VOP, at least passing over lower
-	 * root.
+	 * configuration where lower vnode is moved out of the directory tree
+	 * remounted by our null mount.
+	 *
+	 * Do not try to handle it fancy, just avoid VOP_LOOKUP() with DOTDOT
+	 * name which cannot be handled by the VOP.
 	 */
-	if ((ldvp->v_vflag & VV_ROOT) != 0 && (flags & ISDOTDOT) != 0) {
-		KASSERT((dvp->v_vflag & VV_ROOT) == 0,
-		    ("ldvp %p fl %#x dvp %p fl %#x flags %#jx",
-		    ldvp, ldvp->v_vflag, dvp, dvp->v_vflag, (uintmax_t)flags));
-		return (ENOENT);
+	if ((flags & ISDOTDOT) != 0) {
+		struct nameidata *ndp;
+
+		if ((ldvp->v_vflag & VV_ROOT) != 0) {
+			KASSERT((dvp->v_vflag & VV_ROOT) == 0,
+			    ("ldvp %p fl %#x dvp %p fl %#x flags %#jx",
+			    ldvp, ldvp->v_vflag, dvp, dvp->v_vflag,
+			    (uintmax_t)flags));
+			return (ENOENT);
+		}
+		ndp = lookup_nameidata(cnp);
+		if (ndp != NULL && lookup_isroot(ndp, ldvp))
+			return (ENOENT);
 	}
 
 	/*
diff --git a/sys/fs/unionfs/union_vnops.c b/sys/fs/unionfs/union_vnops.c
index cd57a5cae459..153a5db323d9 100644
--- a/sys/fs/unionfs/union_vnops.c
+++ b/sys/fs/unionfs/union_vnops.c
@@ -76,6 +76,21 @@
 	KASSERT(((vp)->v_op == &unionfs_vnodeops), \
 	    ("unionfs: it is not unionfs-vnode"))
 
+static bool
+unionfs_lookup_isroot(struct componentname *cnp, struct vnode *dvp)
+{
+	struct nameidata *ndp;
+
+	if (dvp == NULL)
+		return (false);
+	if ((dvp->v_vflag & VV_ROOT) != 0)
+		return (true);
+	ndp = lookup_nameidata(cnp);
+	if (ndp == NULL)
+		return (false);
+	return (lookup_isroot(ndp, dvp));
+}
+
 static int
 unionfs_lookup(struct vop_cachedlookup_args *ap)
 {
@@ -124,6 +139,10 @@ unionfs_lookup(struct vop_cachedlookup_args *ap)
 		if (LOOKUP != nameiop && udvp == NULLVP)
 			return (EROFS);
 
+		if (unionfs_lookup_isroot(cnp, udvp) ||
+		    unionfs_lookup_isroot(cnp, ldvp))
+			return (ENOENT);
+
 		if (udvp != NULLVP) {
 			dtmpvp = udvp;
 			if (ldvp != NULLVP)
diff --git a/sys/kern/vfs_cache.c b/sys/kern/vfs_cache.c
index aacbd43403e1..7f3a3c9532f1 100644
--- a/sys/kern/vfs_cache.c
+++ b/sys/kern/vfs_cache.c
@@ -5129,7 +5129,6 @@ cache_fplookup_dotdot(struct cache_fpl *fpl)
 	struct componentname *cnp;
 	struct namecache *ncp;
 	struct vnode *dvp;
-	struct prison *pr;
 	u_char nc_flag;
 
 	ndp = fpl->ndp;
@@ -5141,15 +5140,7 @@ cache_fplookup_dotdot(struct cache_fpl *fpl)
 	/*
 	 * XXX this is racy the same way regular lookup is
 	 */
-	for (pr = cnp->cn_cred->cr_prison; pr != NULL;
-	    pr = pr->pr_parent)
-		if (dvp == pr->pr_root)
-			break;
-
-	if (dvp == ndp->ni_rootdir ||
-	    dvp == ndp->ni_topdir ||
-	    dvp == rootvnode ||
-	    pr != NULL) {
+	if (lookup_isroot(ndp, dvp)) {
 		fpl->tvp = dvp;
 		fpl->tvp_seqc = vn_seqc_read_any(dvp);
 		if (seqc_in_modify(fpl->tvp_seqc)) {
diff --git a/sys/kern/vfs_lookup.c b/sys/kern/vfs_lookup.c
index cb013eb7ff83..dc8b7b92ccd4 100644
--- a/sys/kern/vfs_lookup.c
+++ b/sys/kern/vfs_lookup.c
@@ -530,12 +530,12 @@ namei(struct nameidata *ndp)
 	cnp->cn_origflags = cnp->cn_flags;
 #endif
 	ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
-	KASSERT(ndp->ni_resflags == 0, ("%s: garbage in ni_resflags: %x\n",
+	KASSERT(ndp->ni_resflags == 0, ("%s: garbage in ni_resflags: %x",
 	    __func__, ndp->ni_resflags));
 	KASSERT(cnp->cn_cred && td->td_proc, ("namei: bad cred/proc"));
 	KASSERT((cnp->cn_flags & NAMEI_INTERNAL_FLAGS) == 0,
-	    ("namei: unexpected flags: %" PRIx64 "\n",
-	    cnp->cn_flags & NAMEI_INTERNAL_FLAGS));
+	    ("namei: unexpected flags: %#jx",
+	    (uintmax_t)(cnp->cn_flags & NAMEI_INTERNAL_FLAGS)));
 	if (cnp->cn_flags & NOCACHE)
 		KASSERT(cnp->cn_nameiop != LOOKUP,
 		    ("%s: NOCACHE passed with LOOKUP", __func__));
@@ -761,6 +761,31 @@ needs_exclusive_leaf(struct mount *mp, int flags)
 _Static_assert(MAXNAMLEN == NAME_MAX,
     "MAXNAMLEN and NAME_MAX have different values");
 
+
+struct nameidata *
+lookup_nameidata(struct componentname *cnp)
+{
+	if ((cnp->cn_flags & NAMEILOOKUP) == 0)
+		return (NULL);
+	return (__containerof(cnp, struct nameidata, ni_cnd));
+}
+
+/*
+ * Would a dotdot lookup relative to dvp cause this lookup to cross a jail or
+ * chroot boundary?
+ */
+bool
+lookup_isroot(struct nameidata *ndp, struct vnode *dvp)
+{
+	for (struct prison *pr = ndp->ni_cnd.cn_cred->cr_prison; pr != NULL;
+	    pr = pr->pr_parent) {
+		if (dvp == pr->pr_root)
+			return (true);
+	}
+	return (dvp == ndp->ni_rootdir || dvp == ndp->ni_topdir ||
+	    dvp == rootvnode);
+}
+
 /*
  * Search a pathname.
  * This is a very central and rather complicated routine.
@@ -808,7 +833,6 @@ lookup(struct nameidata *ndp)
 	struct vnode *dp = NULL;	/* the directory we are searching */
 	struct vnode *tdp;		/* saved dp */
 	struct mount *mp;		/* mount table entry */
-	struct prison *pr;
 	size_t prev_ni_pathlen;		/* saved ndp->ni_pathlen */
 	int docache;			/* == 0 do not cache last component */
 	int wantparent;			/* 1 => wantparent or lockparent flag */
@@ -1008,15 +1032,11 @@ dirloop:
 			goto bad;
 		}
 		for (;;) {
-			for (pr = cnp->cn_cred->cr_prison; pr != NULL;
-			     pr = pr->pr_parent)
-				if (dp == pr->pr_root)
-					break;
-			bool isroot = dp == ndp->ni_rootdir ||
-			    dp == ndp->ni_topdir || dp == rootvnode ||
-			    pr != NULL;
-			if (isroot && (ndp->ni_lcf &
-			    NI_LCF_STRICTRELATIVE) != 0) {
+			bool isroot;
+
+			isroot = lookup_isroot(ndp, dp);
+			if (__predict_false(isroot && (ndp->ni_lcf &
+			    NI_LCF_STRICTRELATIVE) != 0)) {
 				error = ENOTCAPABLE;
 				goto capdotdot;
 			}
diff --git a/sys/sys/namei.h b/sys/sys/namei.h
index 1d0fd1c96449..053a64367ff9 100644
--- a/sys/sys/namei.h
+++ b/sys/sys/namei.h
@@ -154,6 +154,7 @@ int	cache_fplookup(struct nameidata *ndp, enum cache_fpl_status *status,
 #define	LOCKSHARED	0x0100	/* Shared lock leaf */
 #define	NOFOLLOW	0x0000	/* do not follow symbolic links (pseudo) */
 #define	RBENEATH	0x100000000ULL /* No escape, even tmp, from start dir */
+#define	NAMEILOOKUP	0x200000000ULL /* cnp is embedded in nameidata */
 #define	MODMASK		0xf000001ffULL	/* mask of operational modifiers */
 
 /*
@@ -254,7 +255,7 @@ do {										\
 	NDINIT_PREFILL(_ndp);							\
 	NDINIT_DBG(_ndp);							\
 	_ndp->ni_cnd.cn_nameiop = op;						\
-	_ndp->ni_cnd.cn_flags = flags;						\
+	_ndp->ni_cnd.cn_flags = (flags) | NAMEILOOKUP;				\
 	_ndp->ni_segflg = segflg;						\
 	_ndp->ni_dirp = namep;							\
 	_ndp->ni_dirfd = dirfd;							\
@@ -312,6 +313,8 @@ void NDVALIDATE(struct nameidata *);
 
 int	namei(struct nameidata *ndp);
 int	lookup(struct nameidata *ndp);
+bool	lookup_isroot(struct nameidata *ndp, struct vnode *dvp);
+struct nameidata *lookup_nameidata(struct componentname *cnp);
 int	relookup(struct vnode *dvp, struct vnode **vpp,
 	    struct componentname *cnp);
 #endif