cache: add empty path support

This avoids spurious drop offs as EMPTY is passed regardless of the
actual path name.

Pushign the work inside the lookup instead of just ignorign the flag
allows avoid checking for empty pathname for all other lookups.

(cherry picked from commit 7dd419cabc6bb9e019c56d15f8e6a88ee2f46859)

3 files changed, 73 insertions, 12 deletions

diff --git a/sys/kern/kern_descrip.c b/sys/kern/kern_descrip.c
index 025259dd20f3..1cf88be09a27 100644
--- a/sys/kern/kern_descrip.c
+++ b/sys/kern/kern_descrip.c
@@ -3034,7 +3034,7 @@ fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsear
 		return (EAGAIN);
 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
 	vp = fp->f_vnode;
-	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
+	if (__predict_false(vp == NULL)) {
 		return (EAGAIN);
 	}
 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
diff --git a/sys/kern/vfs_cache.c b/sys/kern/vfs_cache.c
index 7273126a008c..313879cf6d1f 100644
--- a/sys/kern/vfs_cache.c
+++ b/sys/kern/vfs_cache.c
@@ -4174,8 +4174,8 @@ cache_fpl_terminated(struct cache_fpl *fpl)
 
 #define CACHE_FPL_SUPPORTED_CN_FLAGS \
 	(NC_NOMAKEENTRY | NC_KEEPPOSENTRY | LOCKLEAF | LOCKPARENT | WANTPARENT | \
-	 FAILIFEXISTS | FOLLOW | LOCKSHARED | SAVENAME | SAVESTART | WILLBEDIR | \
-	 ISOPEN | NOMACCHECK | AUDITVNODE1 | AUDITVNODE2 | NOCAPCHECK)
+	 FAILIFEXISTS | FOLLOW | EMPTYPATH | LOCKSHARED | SAVENAME | SAVESTART | \
+	 WILLBEDIR | ISOPEN | NOMACCHECK | AUDITVNODE1 | AUDITVNODE2 | NOCAPCHECK)
 
 #define CACHE_FPL_INTERNAL_CN_FLAGS \
 	(ISDOTDOT | MAKEENTRY | ISLASTCN)
@@ -4194,6 +4194,7 @@ static bool
 cache_fpl_istrailingslash(struct cache_fpl *fpl)
 {
 
+	MPASS(fpl->nulchar > fpl->cnp->cn_pnbuf);
 	return (*(fpl->nulchar - 1) == '/');
 }
 
@@ -4765,6 +4766,54 @@ cache_fplookup_degenerate(struct cache_fpl *fpl)
 }
 
 static int __noinline
+cache_fplookup_emptypath(struct cache_fpl *fpl)
+{
+	struct nameidata *ndp;
+	struct componentname *cnp;
+	enum vgetstate tvs;
+	struct vnode *tvp;
+	seqc_t tvp_seqc;
+	int error, lkflags;
+
+	fpl->tvp = fpl->dvp;
+	fpl->tvp_seqc = fpl->dvp_seqc;
+
+	ndp = fpl->ndp;
+	cnp = fpl->cnp;
+	tvp = fpl->tvp;
+	tvp_seqc = fpl->tvp_seqc;
+
+	MPASS(*cnp->cn_pnbuf == '\0');
+	MPASS((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) == 0);
+
+	if (__predict_false((cnp->cn_flags & EMPTYPATH) == 0)) {
+		cache_fpl_smr_exit(fpl);
+		return (cache_fpl_handled_error(fpl, ENOENT));
+	}
+
+	tvs = vget_prep_smr(tvp);
+	cache_fpl_smr_exit(fpl);
+	if (__predict_false(tvs == VGET_NONE)) {
+		return (cache_fpl_aborted(fpl));
+	}
+
+	if ((cnp->cn_flags & LOCKLEAF) != 0) {
+		lkflags = LK_SHARED;
+		if ((cnp->cn_flags & LOCKSHARED) == 0)
+			lkflags = LK_EXCLUSIVE;
+		error = vget_finish(tvp, lkflags, tvs);
+		if (__predict_false(error != 0)) {
+			return (cache_fpl_aborted(fpl));
+		}
+	} else {
+		vget_finish_ref(tvp, tvs);
+	}
+
+	ndp->ni_resflags |= NIRES_EMPTYPATH;
+	return (cache_fpl_handled(fpl));
+}
+
+static int __noinline
 cache_fplookup_noentry(struct cache_fpl *fpl)
 {
 	struct nameidata *ndp;
@@ -4796,6 +4845,10 @@ cache_fplookup_noentry(struct cache_fpl *fpl)
 		return (cache_fplookup_skip_slashes(fpl));
 	}
 
+	if (cnp->cn_pnbuf[0] == '\0') {
+		return (cache_fplookup_emptypath(fpl));
+	}
+
 	if (cnp->cn_nameptr[0] == '\0') {
 		if (fpl->tvp == NULL) {
 			return (cache_fplookup_degenerate(fpl));
@@ -5483,6 +5536,7 @@ cache_fplookup_parse(struct cache_fpl *fpl)
 	 *
 	 * TODO: fix this to be word-sized.
 	 */
+	MPASS(&cnp->cn_nameptr[fpl->debug.ni_pathlen - 1] >= cnp->cn_pnbuf);
 	KASSERT(&cnp->cn_nameptr[fpl->debug.ni_pathlen - 1] == fpl->nulchar,
 	    ("%s: mismatch between pathlen (%zu) and nulchar (%p != %p), string [%s]\n",
 	    __func__, fpl->debug.ni_pathlen, &cnp->cn_nameptr[fpl->debug.ni_pathlen - 1],
@@ -5737,6 +5791,13 @@ cache_fplookup_failed_vexec(struct cache_fpl *fpl, int error)
 	dvp_seqc = fpl->dvp_seqc;
 
 	/*
+	 * Hack: delayed empty path checking.
+	 */
+	if (cnp->cn_pnbuf[0] == '\0') {
+		return (cache_fplookup_emptypath(fpl));
+	}
+
+	/*
 	 * TODO: Due to ignoring trailing slashes lookup will perform a
 	 * permission check on the last dir when it should not be doing it.  It
 	 * may fail, but said failure should be ignored. It is possible to fix
diff --git a/sys/kern/vfs_lookup.c b/sys/kern/vfs_lookup.c
index f4faf0c77210..95c5599ab232 100644
--- a/sys/kern/vfs_lookup.c
+++ b/sys/kern/vfs_lookup.c
@@ -465,13 +465,6 @@ namei_getpath(struct nameidata *ndp)
 	if (__predict_false(error != 0))
 		return (error);
 
-	/*
-	 * Don't allow empty pathnames unless EMPTYPATH is specified.
-	 * Caller checks for ENOENT as an indication for the empty path.
-	 */
-	if (__predict_false(*cnp->cn_pnbuf == '\0'))
-		return (ENOENT);
-
 	cnp->cn_nameptr = cnp->cn_pnbuf;
 	return (0);
 }
@@ -602,8 +595,6 @@ namei(struct nameidata *ndp)
 
 	error = namei_getpath(ndp);
 	if (__predict_false(error != 0)) {
-		if (error == ENOENT && (cnp->cn_flags & EMPTYPATH) != 0) 
-			return (namei_emptypath(ndp));
 		namei_cleanup_cnp(cnp);
 		SDT_PROBE4(vfs, namei, lookup, return, error, NULL,
 		    false, ndp);
@@ -646,6 +637,15 @@ namei(struct nameidata *ndp)
 	case CACHE_FPL_STATUS_ABORTED:
 		TAILQ_INIT(&ndp->ni_cap_tracker);
 		MPASS(ndp->ni_lcf == 0);
+		if (*cnp->cn_pnbuf == '\0') {
+			if ((cnp->cn_flags & EMPTYPATH) != 0) {
+				return (namei_emptypath(ndp));
+			}
+			namei_cleanup_cnp(cnp);
+			SDT_PROBE4(vfs, namei, lookup, return, ENOENT, NULL,
+			    false, ndp);
+			return (ENOENT);
+		}
 		error = namei_setup(ndp, &dp, &pwd);
 		if (error != 0) {
 			namei_cleanup_cnp(cnp);