Allow user created symbolic links to cover device files and directories

if the device file appears during or after the link creation.

User created symbolic links are now inserted at the head of the
directory entry list after the "." and ".." entries. A new directory
entry flag DE_COVERED indicates that an entry is covered by a symbolic
link.

PR:		kern/114057
Reviewed by:	kib
Idea from:	kib
Discussed on:	freebsd-current (mostly silence)

3 files changed, 56 insertions, 15 deletions

diff --git a/sys/fs/devfs/devfs.h b/sys/fs/devfs/devfs.h
index 2c3a79c83100..ee5eafb31c8d 100644
--- a/sys/fs/devfs/devfs.h
+++ b/sys/fs/devfs/devfs.h
@@ -126,10 +126,11 @@ struct devfs_dirent {
 	struct cdev_priv	*de_cdp;
 	int			de_inode;
 	int			de_flags;
-#define	DE_WHITEOUT	0x1
-#define	DE_DOT		0x2
-#define	DE_DOTDOT	0x4
-#define DE_DOOMED	0x8
+#define	DE_WHITEOUT	0x01
+#define	DE_DOT		0x02
+#define	DE_DOTDOT	0x04
+#define	DE_DOOMED	0x08
+#define	DE_COVERED	0x10
 	int			de_holdcnt;
 	struct dirent 		*de_dirent;
 	TAILQ_ENTRY(devfs_dirent) de_list;
@@ -182,7 +183,7 @@ void devfs_unmount_final(struct devfs_mount *mp);
 struct devfs_dirent *devfs_newdirent (char *name, int namelen);
 struct devfs_dirent *devfs_parent_dirent(struct devfs_dirent *de);
 struct devfs_dirent *devfs_vmkdir (struct devfs_mount *, char *name, int namelen, struct devfs_dirent *dotdot, u_int inode);
-struct devfs_dirent *devfs_find (struct devfs_dirent *dd, const char *name, int namelen);
+struct devfs_dirent *devfs_find(struct devfs_dirent *dd, const char *name, int namelen, int type);
 
 #endif /* _KERNEL */
 
diff --git a/sys/fs/devfs/devfs_devs.c b/sys/fs/devfs/devfs_devs.c
index b14a41fdc5b9..a11544f80031 100644
--- a/sys/fs/devfs/devfs_devs.c
+++ b/sys/fs/devfs/devfs_devs.c
@@ -158,13 +158,15 @@ devfs_free(struct cdev *cdev)
 }
 
 struct devfs_dirent *
-devfs_find(struct devfs_dirent *dd, const char *name, int namelen)
+devfs_find(struct devfs_dirent *dd, const char *name, int namelen, int type)
 {
 	struct devfs_dirent *de;
 
 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
 		if (namelen != de->de_dirent->d_namlen)
 			continue;
+		if (type != 0 && type != de->de_dirent->d_type)
+			continue;
 		if (bcmp(name, de->de_dirent->d_name, namelen) != 0)
 			continue;
 		break;
@@ -235,14 +237,19 @@ devfs_vmkdir(struct devfs_mount *dmp, char *name, int namelen, struct devfs_dire
 	else
 		dd->de_inode = alloc_unr(devfs_inos);
 
-	/* Create the "." entry in the new directory */
+	/*
+	 * "." and ".." are always the two first entries in the
+	 * de_dlist list.
+	 *
+	 * Create the "." entry in the new directory.
+	 */
 	de = devfs_newdirent(".", 1);
 	de->de_dirent->d_type = DT_DIR;
 	de->de_flags |= DE_DOT;
 	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
 	de->de_dir = dd;
 
-	/* Create the ".." entry in the new directory */
+	/* Create the ".." entry in the new directory. */
 	de = devfs_newdirent("..", 2);
 	de->de_dirent->d_type = DT_DIR;
 	de->de_flags |= DE_DOTDOT;
@@ -382,7 +389,7 @@ devfs_populate_loop(struct devfs_mount *dm, int cleanup)
 	struct devfs_dirent *de;
 	struct devfs_dirent *dd;
 	struct cdev *pdev;
-	int j;
+	int de_flags, j;
 	char *q, *s;
 
 	sx_assert(&dm->dm_lock, SX_XLOCKED);
@@ -454,12 +461,27 @@ devfs_populate_loop(struct devfs_mount *dm, int cleanup)
 				continue;
 			if (*q != '/')
 				break;
-			de = devfs_find(dd, s, q - s);
+			de = devfs_find(dd, s, q - s, 0);
 			if (de == NULL)
 				de = devfs_vmkdir(dm, s, q - s, dd, 0);
+			else if (de->de_dirent->d_type == DT_LNK) {
+				de = devfs_find(dd, s, q - s, DT_DIR);
+				if (de == NULL)
+					de = devfs_vmkdir(dm, s, q - s, dd, 0);
+				de->de_flags |= DE_COVERED;
+			}
 			s = q + 1;
 			dd = de;
+			KASSERT(dd->de_dirent->d_type == DT_DIR &&
+			    (dd->de_flags & (DE_DOT | DE_DOTDOT)) == 0,
+			    ("%s: invalid directory (si_name=%s)",
+			    __func__, cdp->cdp_c.si_name));
+
 		}
+		de_flags = 0;
+		de = devfs_find(dd, s, q - s, DT_LNK);
+		if (de != NULL)
+			de_flags |= DE_COVERED;
 
 		de = devfs_newdirent(s, q - s);
 		if (cdp->cdp_c.si_flags & SI_ALIAS) {
@@ -477,6 +499,7 @@ devfs_populate_loop(struct devfs_mount *dm, int cleanup)
 			de->de_mode = cdp->cdp_c.si_mode;
 			de->de_dirent->d_type = DT_CHR;
 		}
+		de->de_flags |= de_flags;
 		de->de_inode = cdp->cdp_inode;
 		de->de_cdp = cdp;
 #ifdef MAC
diff --git a/sys/fs/devfs/devfs_vnops.c b/sys/fs/devfs/devfs_vnops.c
index 3e9172906295..542ecaed35e2 100644
--- a/sys/fs/devfs/devfs_vnops.c
+++ b/sys/fs/devfs/devfs_vnops.c
@@ -822,7 +822,7 @@ devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
 		return (ENOENT);
 	}
 	dd = dvp->v_data;
-	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen);
+	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
 	while (de == NULL) {	/* While(...) so we can use break */
 
 		if (nameiop == DELETE)
@@ -1151,7 +1151,7 @@ devfs_readdir(struct vop_readdir_args *ap)
 	off = 0;
 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
-		if (dd->de_flags & DE_WHITEOUT)
+		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
 			continue;
 		if (devfs_prison_check(dd, uio->uio_td))
 			continue;
@@ -1232,7 +1232,7 @@ devfs_remove(struct vop_remove_args *ap)
 {
 	struct vnode *vp = ap->a_vp;
 	struct devfs_dirent *dd;
-	struct devfs_dirent *de;
+	struct devfs_dirent *de, *de_covered;
 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
 
 	sx_xlock(&dmp->dm_lock);
@@ -1240,6 +1240,12 @@ devfs_remove(struct vop_remove_args *ap)
 	de = vp->v_data;
 	if (de->de_cdp == NULL) {
 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
+		if (de->de_dirent->d_type == DT_LNK) {
+			de_covered = devfs_find(dd, de->de_dirent->d_name,
+			    de->de_dirent->d_namlen, 0);
+			if (de_covered != NULL)
+				de_covered->de_flags &= ~DE_COVERED;
+		}
 		devfs_delete(dmp, de, 1);
 	} else {
 		de->de_flags |= DE_WHITEOUT;
@@ -1479,7 +1485,7 @@ devfs_symlink(struct vop_symlink_args *ap)
 {
 	int i, error;
 	struct devfs_dirent *dd;
-	struct devfs_dirent *de;
+	struct devfs_dirent *de, *de_covered, *de_dotdot;
 	struct devfs_mount *dmp;
 
 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
@@ -1500,7 +1506,18 @@ devfs_symlink(struct vop_symlink_args *ap)
 #ifdef MAC
 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
 #endif
-	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
+	de_covered = devfs_find(dd, de->de_dirent->d_name,
+	    de->de_dirent->d_namlen, 0);
+	if (de_covered != NULL) {
+		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
+		    ("devfs_symlink: entry %p already covered", de_covered));
+		de_covered->de_flags |= DE_COVERED;
+	}
+
+	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
+	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
+	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
+
 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
 }