1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
47  * a stackable layers techniques, and its "null-node"s stack above
48  * all lower-layer vnodes, not just over directory vnodes.
66  * of the lower vfs (target-pn) and the pathname where the null
67  * layer will appear in the namespace (alias-pn).  After
69  * of target-pn subtree will be aliased under alias-pn.
75  * simply bypassing all possible operations to the lower layer
81  * handling by the lower layer.  It begins by examining vnode
82  * operation arguments and replacing any null-nodes by their
83  * lower-layer equivlants.  It then invokes the operation
84  * on the lower layer.  Finally, it replaces the null-nodes
86  * stacks a null-node on top of the returned vnode.
94  * they can handle freeing null-layer specific data. Vop_print
107  * Mounting associates the null layer with a lower layer,
118  * The bypass routine stacks a null-node above the new
124  * the root null-node (which was created when the null layer was mounted).
126  * done on the root null-node.  This operation would bypass through
127  * to the lower layer which would return a vnode representing
128  * the UFS "sys".  Null_bypass then builds a null-node
130  * Later operations on the null-node "sys" will repeat this
145  * INVOKING OPERATIONS ON LOWER LAYERS
147  * There are two techniques to invoke operations on a lower layer
151  * the operation arguments "correct" for the lower layer
152  * by mapping a vnode arguments to the lower layer.
156  * currently being handled on the lower layer.  It has the advantage
161  * the lower layer with the VOP_OPERATIONNAME interface.
163  * arbitrary operations on the lower layer.  The disadvantage
197  * Synchronize inotify flags with the lower vnode:
198  * - If the upper vnode has the flag set and the lower does not, then the lower
201  * - If the lower vnode is watched, then the upper vnode should go through
217  * This is the 10-Apr-92 bypass routine.
228  * by the upper node, not the lower one, so we must handle these
229  * side-effects here.  This is not of concern in Sun-derived systems
230  * since there are no such side-effects.
233  * - only one returned vpp
234  * - no INOUT vpp's (Sun's vop_open has one of these)
235  * - the vnode operation vector of the first vnode should be used
237  * - all mapped vnodes are of our vnode-type (NEEDSWORK:
248 	struct vnodeop_desc *descp = ap->a_desc;  in null_bypass()
252 		printf ("null_bypass: %s\n", descp->vdesc_name);  in null_bypass()
258 	if (descp->vdesc_vp_offsets == NULL ||  in null_bypass()
259 	    descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)  in null_bypass()
268 	reles = descp->vdesc_flags;  in null_bypass()
270 		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)  in null_bypass()
273 		    descp->vdesc_vp_offsets[i], ap);  in null_bypass()
288 			 * The upper vnode reference to the lower  in null_bypass()
290 			 * pointer to the lower vnode alive.  If lower  in null_bypass()
299 			 * XXX - Several operations have the side effect  in null_bypass()
309 	 * Call the operation on the lower layer  in null_bypass()
313 		error = ap->a_desc->vdesc_call(ap);  in null_bypass()
315 		printf("null_bypass: no map for %s\n", descp->vdesc_name);  in null_bypass()
320 	 * Maintain the illusion of call-by-value  in null_bypass()
324 	reles = descp->vdesc_flags;  in null_bypass()
326 		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)  in null_bypass()
333 			 * flags up in case something is watching the lower  in null_bypass()
339 			 * back to private v_lock.  In this case we  in null_bypass()
340 			 * must move lock ownership from lower to  in null_bypass()
349 				    old_vps[i]->v_vnlock != lvp->v_vnlock) {  in null_bypass()
368 	 * Map the possible out-going vpp  in null_bypass()
369 	 * (Assumes that the lower layer always returns  in null_bypass()
372 	if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && error == 0) {  in null_bypass()
374 		 * XXX - even though some ops have vpp returned vp's,  in null_bypass()
380 		    descp->vdesc_vpp_offset, ap);  in null_bypass()
382 			error = null_nodeget(old_vps[0]->v_mount, **vppp,  in null_bypass()
395 	vp = ap->a_vp;  in null_add_writecount()
399 	VNASSERT(vp->v_writecount >= 0, vp, ("wrong null writecount"));  in null_add_writecount()
400 	VNASSERT(vp->v_writecount + ap->a_inc >= 0, vp,  in null_add_writecount()
401 	    ("wrong writecount inc %d", ap->a_inc));  in null_add_writecount()
402 	error = VOP_ADD_WRITECOUNT(lvp, ap->a_inc);  in null_add_writecount()
404 		vp->v_writecount += ap->a_inc;  in null_add_writecount()
411  * as we progress through the tree. We also have to enforce read-only
412  * if this layer is mounted read-only.
417 	struct componentname *cnp = ap->a_cnp;  in null_lookup()
418 	struct vnode *dvp = ap->a_dvp;  in null_lookup()
419 	uint64_t flags = cnp->cn_flags;  in null_lookup()
424 	mp = dvp->v_mount;  in null_lookup()
425 	if ((flags & ISLASTCN) != 0 && (mp->mnt_flag & MNT_RDONLY) != 0 &&  in null_lookup()
426 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))  in null_lookup()
436 	 * Renames in the lower mounts might create an inconsistent  in null_lookup()
437 	 * configuration where lower vnode is moved out of the directory tree  in null_lookup()
446 		if ((ldvp->v_vflag & VV_ROOT) != 0) {  in null_lookup()
447 			KASSERT((dvp->v_vflag & VV_ROOT) == 0,  in null_lookup()
449 			    ldvp, ldvp->v_vflag, dvp, dvp->v_vflag,  in null_lookup()
460 	 * case of dvp reclamation, and we need ldvp to move our lock  in null_lookup()
468 	 * VOP_LOOKUP() on lower vnode may unlock ldvp, which allows  in null_lookup()
482 		 * longer shared.  In this case, relock of ldvp in  in null_lookup()
483 		 * lower fs VOP_LOOKUP() does not restore the locking  in null_lookup()
494 	    (mp->mnt_flag & MNT_RDONLY) != 0 &&  in null_lookup()
495 	    (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))  in null_lookup()
500 			*ap->a_vpp = dvp;  in null_lookup()
506 				*ap->a_vpp = vp;  in null_lookup()
518 	vp = ap->a_vp;  in null_open()
520 	retval = null_bypass(&ap->a_gen);  in null_open()
522 		vp->v_object = ldvp->v_object;  in null_open()
524 			MPASS(vp->v_object != NULL);  in null_open()
534  * Setattr call. Disallow write attempts if the layer is mounted read-only.
539 	struct vnode *vp = ap->a_vp;  in null_setattr()
540 	struct vattr *vap = ap->a_vap;  in null_setattr()
542   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||  in null_setattr()
543 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||  in null_setattr()
544 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&  in null_setattr()
545 	    (vp->v_mount->mnt_flag & MNT_RDONLY))  in null_setattr()
547 	if (vap->va_size != VNOVAL) {  in null_setattr()
548  		switch (vp->v_type) {  in null_setattr()
549  		case VDIR:  in null_setattr()
551  		case VCHR:  in null_setattr()
552  		case VBLK:  in null_setattr()
553  		case VSOCK:  in null_setattr()
554  		case VFIFO:  in null_setattr()
555 			if (vap->va_flags != VNOVAL)  in null_setattr()
558 		case VREG:  in null_setattr()
559 		case VLNK:  in null_setattr()
563 			 * mounted read-only.  in null_setattr()
565 			if (vp->v_mount->mnt_flag & MNT_RDONLY)  in null_setattr()
570 	return (null_bypass(&ap->a_gen));  in null_setattr()
581 	if ((error = null_bypass(&ap->a_gen)) != 0)  in null_stat()
584 	ap->a_sb->st_dev = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];  in null_stat()
593 	if ((error = null_bypass(&ap->a_gen)) != 0)  in null_getattr()
596 	ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];  in null_getattr()
601  * Handle to disallow write access if mounted read-only.
606 	struct vnode *vp = ap->a_vp;  in null_access()
607 	accmode_t accmode = ap->a_accmode;  in null_access()
610 	 * Disallow write attempts on read-only layers;  in null_access()
615 		switch (vp->v_type) {  in null_access()
616 		case VDIR:  in null_access()
617 		case VLNK:  in null_access()
618 		case VREG:  in null_access()
619 			if (vp->v_mount->mnt_flag & MNT_RDONLY)  in null_access()
626 	return (null_bypass(&ap->a_gen));  in null_access()
632 	struct vnode *vp = ap->a_vp;  in null_accessx()
633 	accmode_t accmode = ap->a_accmode;  in null_accessx()
636 	 * Disallow write attempts on read-only layers;  in null_accessx()
641 		switch (vp->v_type) {  in null_accessx()
642 		case VDIR:  in null_accessx()
643 		case VLNK:  in null_accessx()
644 		case VREG:  in null_accessx()
645 			if (vp->v_mount->mnt_flag & MNT_RDONLY)  in null_accessx()
652 	return (null_bypass(&ap->a_gen));  in null_accessx()
656  * Increasing refcount of lower vnode is needed at least for the case
657  * when lower FS is NFS to do sillyrename if the file is in use.
669 	vp = ap->a_vp;  in null_remove()
676 	VTONULL(vp)->null_flags |= NULLV_DROP;  in null_remove()
677 	retval = null_bypass(&ap->a_gen);  in null_remove()
684  * We handle this to eliminate null FS to lower FS
696 	tdvp = ap->a_tdvp;  in null_rename()
697 	fvp = ap->a_fvp;  in null_rename()
698 	fdvp = ap->a_fdvp;  in null_rename()
699 	tvp = ap->a_tvp;  in null_rename()
702 	/* Check for cross-device rename. */  in null_rename()
703 	if ((fvp->v_mount != tdvp->v_mount) ||  in null_rename()
704 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {  in null_rename()
716 	lfdvp = fdnn->null_lowervp;  in null_rename()
727 	lfvp = fnn->null_lowervp;  in null_rename()
732 	ltdvp = tdnn->null_lowervp;  in null_rename()
737 		ltvp = tnn->null_lowervp;  in null_rename()
739 		tnn->null_flags |= NULLV_DROP;  in null_rename()
744 	error = VOP_RENAME(lfdvp, lfvp, ap->a_fcnp, ltdvp, ltvp, ap->a_tcnp);  in null_rename()
770 	VTONULL(ap->a_vp)->null_flags |= NULLV_DROP;  in null_rmdir()
771 	return (null_bypass(&ap->a_gen));  in null_rmdir()
782  * be recycled, in which case it is not legal to be sleeping in its VOP.  We
795 	nn = VTONULL_SMR(ap->a_vp);  in null_lock_prep_with_smr()
797 		lvp = nn->null_lowervp;  in null_lock_prep_with_smr()
812 	ASSERT_VI_LOCKED(ap->a_vp, __func__);  in null_lock_prep_with_interlock()
814 	ap->a_flags &= ~LK_INTERLOCK;  in null_lock_prep_with_interlock()
818 	nn = VTONULL(ap->a_vp);  in null_lock_prep_with_interlock()
820 		lvp = nn->null_lowervp;  in null_lock_prep_with_interlock()
824 	VI_UNLOCK(ap->a_vp);  in null_lock_prep_with_interlock()
834 	if (__predict_true((ap->a_flags & LK_INTERLOCK) == 0)) {  in null_lock()
837 			VI_LOCK(ap->a_vp);  in null_lock()
844 	ASSERT_VI_UNLOCKED(ap->a_vp, __func__);  in null_lock()
849 	VNPASS(lvp->v_holdcnt > 0, lvp);  in null_lock()
850 	error = VOP_LOCK(lvp, ap->a_flags);  in null_lock()
855 	 * case by reacquiring correct lock in requested mode.  in null_lock()
857 	if (VTONULL(ap->a_vp) == NULL && error == 0) {  in null_lock()
860 		flags = ap->a_flags;  in null_lock()
861 		ap->a_flags &= ~LK_TYPE_MASK;  in null_lock()
863 		case LK_SHARED:  in null_lock()
864 			ap->a_flags |= LK_SHARED;  in null_lock()
866 		case LK_UPGRADE:  in null_lock()
867 		case LK_EXCLUSIVE:  in null_lock()
868 			ap->a_flags |= LK_EXCLUSIVE;  in null_lock()
883 	struct vnode *vp = ap->a_vp;  in null_unlock()
910  * Do not allow the VOP_INACTIVE to be passed to the lower layer,
911  * since the reference count on the lower vnode is not related to
924 	mp = vp->v_mount;  in null_want_recycle()
926 	if ((xmp->nullm_flags & NULLM_CACHE) == 0 ||  in null_want_recycle()
927 	    (xp->null_flags & NULLV_DROP) != 0 ||  in null_want_recycle()
928 	    (lvp->v_vflag & VV_NOSYNC) != 0) {  in null_want_recycle()
931 		 * nullfs vnodes is not enabled, or the lower vnode is  in null_want_recycle()
933 		 * the lower vnodes.  in null_want_recycle()
945 	vp = ap->a_vp;  in null_inactive()
947 		vp->v_object = NULL;  in null_inactive()
957 	return (null_want_recycle(ap->a_vp) || vn_need_pageq_flush(ap->a_vp));  in null_need_inactive()
961  * Now, the nullfs vnode and, due to the sharing lock, the lower
971 	vp = ap->a_vp;  in null_reclaim()
973 	lowervp = xp->null_lowervp;  in null_reclaim()
975 	KASSERT(lowervp != NULL && vp->v_vnlock != &vp->v_lock,  in null_reclaim()
983 	lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);  in null_reclaim()
985 	vp->v_data = NULL;  in null_reclaim()
986 	vp->v_object = NULL;  in null_reclaim()
987 	vp->v_vnlock = &vp->v_lock;  in null_reclaim()
991 	 * to the lower vnode.  If this is a reclamation due to the  in null_reclaim()
994 	if (vp->v_writecount > 0)  in null_reclaim()
995 		VOP_ADD_WRITECOUNT(lowervp, -vp->v_writecount);  in null_reclaim()
996 	else if (vp->v_writecount < 0)  in null_reclaim()
997 		vp->v_writecount = 0;  in null_reclaim()
1001 	if ((xp->null_flags & NULLV_NOUNLOCK) != 0)  in null_reclaim()
1013 	struct vnode *vp = ap->a_vp;  in null_print()
1015 	printf("\tvp=%p, lowervp=%p\n", vp, VTONULL(vp)->null_lowervp);  in null_print()
1027 	vp = ap->a_vp;  in null_getwritemount()
1030 	if (xp && (lowervp = xp->null_lowervp)) {  in null_getwritemount()
1033 		VOP_GETWRITEMOUNT(lowervp, ap->a_mpp);  in null_getwritemount()
1037 		*(ap->a_mpp) = NULL;  in null_getwritemount()
1047 	lvp = NULLVPTOLOWERVP(ap->a_vp);  in null_vptofh()
1048 	return VOP_VPTOFH(lvp, ap->a_fhp);  in null_vptofh()
1054 	struct vnode *vp = ap->a_vp;  in null_vptocnp()
1055 	struct vnode **dvp = ap->a_vpp;  in null_vptocnp()
1062 	mp = vp->v_mount;  in null_vptocnp()
1070 	error = vn_vptocnp(&ldvp, ap->a_buf, ap->a_buflen);  in null_vptocnp()
1104 	vp = ap->a_vp;  in null_read_pgcache()
1111 	lvp = xp->null_lowervp;  in null_read_pgcache()
1114 	error = VOP_READ_PGCACHE(lvp, ap->a_uio, ap->a_ioflag, ap->a_cred);  in null_read_pgcache()
1126 	vp = ap->a_vp;  in null_advlock()
1133 	lvp = xp->null_lowervp;  in null_advlock()
1136 	error = VOP_ADVLOCK(lvp, ap->a_id, ap->a_op, ap->a_fl, ap->a_flags);  in null_advlock()
1142  * Avoid standard bypass, since lower dvp and vp could be no longer
1152 	dvp = ap->a_dvp;  in null_vput_pair()
1156 	vpp = ap->a_vpp;  in null_vput_pair()
1165 		if (!ap->a_unlock_vp) {  in null_vput_pair()
1168 			mp = vp->v_mount;  in null_vput_pair()
1174 	if (vp != NULL && ap->a_unlock_vp)  in null_vput_pair()
1178 	if (vp == NULL || ap->a_unlock_vp)  in null_vput_pair()
1183 	if (vp->v_data == NULL && vfs_busy(mp, MBF_NOWAIT) == 0) {  in null_vput_pair()
1211 	vp = ap->a_vp;  in null_getlowvnode()
1218 	VOP_GETLOWVNODE(vpl, ap->a_vplp, ap->a_flags);  in null_getlowvnode()