2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1992, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * John Heidemann of the UCLA Ficus project.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)null_vnops.c 8.6 (Berkeley) 5/27/95
37 * @(#)lofs_vnops.c 1.2 (Berkeley) 6/18/92
39 * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project
45 * (See mount_nullfs(8) for more information.)
47 * The null layer duplicates a portion of the filesystem
48 * name space under a new name. In this respect, it is
49 * similar to the loopback filesystem. It differs from
50 * the loopback fs in two respects: it is implemented using
51 * a stackable layers techniques, and its "null-node"s stack above
52 * all lower-layer vnodes, not just over directory vnodes.
54 * The null layer has two purposes. First, it serves as a demonstration
55 * of layering by proving a layer which does nothing. (It actually
56 * does everything the loopback filesystem does, which is slightly
57 * more than nothing.) Second, the null layer can serve as a prototype
58 * layer. Since it provides all necessary layer framework,
59 * new filesystem layers can be created very easily be starting
62 * The remainder of this man page examines the null layer as a basis
63 * for constructing new layers.
66 * INSTANTIATING NEW NULL LAYERS
68 * New null layers are created with mount_nullfs(8).
69 * Mount_nullfs(8) takes two arguments, the pathname
70 * of the lower vfs (target-pn) and the pathname where the null
71 * layer will appear in the namespace (alias-pn). After
72 * the null layer is put into place, the contents
73 * of target-pn subtree will be aliased under alias-pn.
76 * OPERATION OF A NULL LAYER
78 * The null layer is the minimum filesystem layer,
79 * simply bypassing all possible operations to the lower layer
80 * for processing there. The majority of its activity centers
81 * on the bypass routine, through which nearly all vnode operations
84 * The bypass routine accepts arbitrary vnode operations for
85 * handling by the lower layer. It begins by examining vnode
86 * operation arguments and replacing any null-nodes by their
87 * lower-layer equivlants. It then invokes the operation
88 * on the lower layer. Finally, it replaces the null-nodes
89 * in the arguments and, if a vnode is return by the operation,
90 * stacks a null-node on top of the returned vnode.
92 * Although bypass handles most operations, vop_getattr, vop_lock,
93 * vop_unlock, vop_inactive, vop_reclaim, and vop_print are not
94 * bypassed. Vop_getattr must change the fsid being returned.
95 * Vop_lock and vop_unlock must handle any locking for the
96 * current vnode as well as pass the lock request down.
97 * Vop_inactive and vop_reclaim are not bypassed so that
98 * they can handle freeing null-layer specific data. Vop_print
99 * is not bypassed to avoid excessive debugging information.
100 * Also, certain vnode operations change the locking state within
101 * the operation (create, mknod, remove, link, rename, mkdir, rmdir,
102 * and symlink). Ideally these operations should not change the
103 * lock state, but should be changed to let the caller of the
104 * function unlock them. Otherwise all intermediate vnode layers
105 * (such as union, umapfs, etc) must catch these functions to do
106 * the necessary locking at their layer.
109 * INSTANTIATING VNODE STACKS
111 * Mounting associates the null layer with a lower layer,
112 * effect stacking two VFSes. Vnode stacks are instead
113 * created on demand as files are accessed.
115 * The initial mount creates a single vnode stack for the
116 * root of the new null layer. All other vnode stacks
117 * are created as a result of vnode operations on
118 * this or other null vnode stacks.
120 * New vnode stacks come into existence as a result of
121 * an operation which returns a vnode.
122 * The bypass routine stacks a null-node above the new
123 * vnode before returning it to the caller.
125 * For example, imagine mounting a null layer with
126 * "mount_nullfs /usr/include /dev/layer/null".
127 * Changing directory to /dev/layer/null will assign
128 * the root null-node (which was created when the null layer was mounted).
129 * Now consider opening "sys". A vop_lookup would be
130 * done on the root null-node. This operation would bypass through
131 * to the lower layer which would return a vnode representing
132 * the UFS "sys". Null_bypass then builds a null-node
133 * aliasing the UFS "sys" and returns this to the caller.
134 * Later operations on the null-node "sys" will repeat this
135 * process when constructing other vnode stacks.
138 * CREATING OTHER FILE SYSTEM LAYERS
140 * One of the easiest ways to construct new filesystem layers is to make
141 * a copy of the null layer, rename all files and variables, and
142 * then begin modifing the copy. Sed can be used to easily rename
145 * The umap layer is an example of a layer descended from the
149 * INVOKING OPERATIONS ON LOWER LAYERS
151 * There are two techniques to invoke operations on a lower layer
152 * when the operation cannot be completely bypassed. Each method
153 * is appropriate in different situations. In both cases,
154 * it is the responsibility of the aliasing layer to make
155 * the operation arguments "correct" for the lower layer
156 * by mapping a vnode arguments to the lower layer.
158 * The first approach is to call the aliasing layer's bypass routine.
159 * This method is most suitable when you wish to invoke the operation
160 * currently being handled on the lower layer. It has the advantage
161 * that the bypass routine already must do argument mapping.
162 * An example of this is null_getattrs in the null layer.
164 * A second approach is to directly invoke vnode operations on
165 * the lower layer with the VOP_OPERATIONNAME interface.
166 * The advantage of this method is that it is easy to invoke
167 * arbitrary operations on the lower layer. The disadvantage
168 * is that vnode arguments must be manualy mapped.
172 #include <sys/param.h>
173 #include <sys/systm.h>
174 #include <sys/conf.h>
175 #include <sys/kernel.h>
176 #include <sys/lock.h>
177 #include <sys/malloc.h>
178 #include <sys/mount.h>
179 #include <sys/mutex.h>
180 #include <sys/namei.h>
181 #include <sys/sysctl.h>
182 #include <sys/vnode.h>
183 #include <sys/stat.h>
185 #include <fs/nullfs/null.h>
188 #include <vm/vm_extern.h>
189 #include <vm/vm_object.h>
190 #include <vm/vnode_pager.h>
192 static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */
193 SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW,
194 &null_bug_bypass, 0, "");
197 * This is the 10-Apr-92 bypass routine.
198 * This version has been optimized for speed, throwing away some
199 * safety checks. It should still always work, but it's not as
200 * robust to programmer errors.
202 * In general, we map all vnodes going down and unmap them on the way back.
203 * As an exception to this, vnodes can be marked "unmapped" by setting
204 * the Nth bit in operation's vdesc_flags.
206 * Also, some BSD vnode operations have the side effect of vrele'ing
207 * their arguments. With stacking, the reference counts are held
208 * by the upper node, not the lower one, so we must handle these
209 * side-effects here. This is not of concern in Sun-derived systems
210 * since there are no such side-effects.
212 * This makes the following assumptions:
213 * - only one returned vpp
214 * - no INOUT vpp's (Sun's vop_open has one of these)
215 * - the vnode operation vector of the first vnode should be used
216 * to determine what implementation of the op should be invoked
217 * - all mapped vnodes are of our vnode-type (NEEDSWORK:
218 * problems on rmdir'ing mount points and renaming?)
221 null_bypass(struct vop_generic_args *ap)
223 struct vnode **this_vp_p;
224 struct vnode *old_vps[VDESC_MAX_VPS];
225 struct vnode **vps_p[VDESC_MAX_VPS];
226 struct vnode ***vppp;
228 struct vnodeop_desc *descp = ap->a_desc;
232 printf ("null_bypass: %s\n", descp->vdesc_name);
236 * We require at least one vp.
238 if (descp->vdesc_vp_offsets == NULL ||
239 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
240 panic ("null_bypass: no vp's in map");
244 * Map the vnodes going in.
245 * Later, we'll invoke the operation based on
246 * the first mapped vnode's operation vector.
248 reles = descp->vdesc_flags;
249 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
250 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
251 break; /* bail out at end of list */
252 vps_p[i] = this_vp_p = VOPARG_OFFSETTO(struct vnode **,
253 descp->vdesc_vp_offsets[i], ap);
256 * We're not guaranteed that any but the first vnode
257 * are of our type. Check for and don't map any
258 * that aren't. (We must always map first vp or vclean fails.)
260 if (i != 0 && (*this_vp_p == NULLVP ||
261 (*this_vp_p)->v_op != &null_vnodeops)) {
264 old_vps[i] = *this_vp_p;
265 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
268 * The upper vnode reference to the lower
269 * vnode is the only reference that keeps our
270 * pointer to the lower vnode alive. If lower
271 * vnode is relocked during the VOP call,
272 * upper vnode might become unlocked and
273 * reclaimed, which invalidates our reference.
274 * Add a transient hold around VOP call.
279 * XXX - Several operations have the side effect
280 * of vrele'ing their vp's. We must account for
281 * that. (This should go away in the future.)
283 if (reles & VDESC_VP0_WILLRELE)
289 * Call the operation on the lower layer
290 * with the modified argument structure.
292 if (vps_p[0] != NULL && *vps_p[0] != NULL) {
295 printf("null_bypass: no map for %s\n", descp->vdesc_name);
300 * Maintain the illusion of call-by-value
301 * by restoring vnodes in the argument structure
302 * to their original value.
304 reles = descp->vdesc_flags;
305 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
306 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
307 break; /* bail out at end of list */
308 if (old_vps[i] != NULL) {
312 * Get rid of the transient hold on lvp.
313 * If lowervp was unlocked during VOP
314 * operation, nullfs upper vnode could have
315 * been reclaimed, which changes its v_vnlock
316 * back to private v_lock. In this case we
317 * must move lock ownership from lower to
318 * upper (reclaimed) vnode.
321 if (VOP_ISLOCKED(lvp) == LK_EXCLUSIVE &&
322 old_vps[i]->v_vnlock != lvp->v_vnlock) {
324 VOP_LOCK(old_vps[i], LK_EXCLUSIVE |
330 *(vps_p[i]) = old_vps[i];
332 if (reles & VDESC_VP0_WILLUNLOCK)
333 VOP_UNLOCK(*(vps_p[i]), 0);
335 if (reles & VDESC_VP0_WILLRELE)
341 * Map the possible out-going vpp
342 * (Assumes that the lower layer always returns
343 * a VREF'ed vpp unless it gets an error.)
345 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && error == 0) {
347 * XXX - even though some ops have vpp returned vp's,
348 * several ops actually vrele this before returning.
349 * We must avoid these ops.
350 * (This should go away when these ops are regularized.)
352 vppp = VOPARG_OFFSETTO(struct vnode ***,
353 descp->vdesc_vpp_offset, ap);
355 error = null_nodeget(old_vps[0]->v_mount, **vppp,
363 null_add_writecount(struct vop_add_writecount_args *ap)
365 struct vnode *lvp, *vp;
369 lvp = NULLVPTOLOWERVP(vp);
371 /* text refs are bypassed to lowervp */
372 VNASSERT(vp->v_writecount >= 0, vp, ("wrong null writecount"));
373 VNASSERT(vp->v_writecount + ap->a_inc >= 0, vp,
374 ("wrong writecount inc %d", ap->a_inc));
375 error = VOP_ADD_WRITECOUNT(lvp, ap->a_inc);
377 vp->v_writecount += ap->a_inc;
383 * We have to carry on the locking protocol on the null layer vnodes
384 * as we progress through the tree. We also have to enforce read-only
385 * if this layer is mounted read-only.
388 null_lookup(struct vop_lookup_args *ap)
390 struct componentname *cnp = ap->a_cnp;
391 struct vnode *dvp = ap->a_dvp;
392 int flags = cnp->cn_flags;
393 struct vnode *vp, *ldvp, *lvp;
398 if ((flags & ISLASTCN) != 0 && (mp->mnt_flag & MNT_RDONLY) != 0 &&
399 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
402 * Although it is possible to call null_bypass(), we'll do
403 * a direct call to reduce overhead
405 ldvp = NULLVPTOLOWERVP(dvp);
409 * Renames in the lower mounts might create an inconsistent
410 * configuration where lower vnode is moved out of the
411 * directory tree remounted by our null mount. Do not try to
412 * handle it fancy, just avoid VOP_LOOKUP() with DOTDOT name
413 * which cannot be handled by VOP, at least passing over lower
416 if ((ldvp->v_vflag & VV_ROOT) != 0 && (flags & ISDOTDOT) != 0) {
417 KASSERT((dvp->v_vflag & VV_ROOT) == 0,
418 ("ldvp %p fl %#x dvp %p fl %#x flags %#x",
419 ldvp, ldvp->v_vflag, dvp, dvp->v_vflag, flags));
424 * Hold ldvp. The reference on it, owned by dvp, is lost in
425 * case of dvp reclamation, and we need ldvp to move our lock
430 error = VOP_LOOKUP(ldvp, &lvp, cnp);
433 * VOP_LOOKUP() on lower vnode may unlock ldvp, which allows
434 * dvp to be reclaimed due to shared v_vnlock. Check for the
435 * doomed state and return error.
437 if (VN_IS_DOOMED(dvp)) {
438 if (error == 0 || error == EJUSTRETURN) {
445 * If vgone() did reclaimed dvp before curthread
446 * relocked ldvp, the locks of dvp and ldpv are no
447 * longer shared. In this case, relock of ldvp in
448 * lower fs VOP_LOOKUP() does not restore the locking
449 * state of dvp. Compensate for this by unlocking
450 * ldvp and locking dvp, which is also correct if the
451 * locks are still shared.
454 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
458 if (error == EJUSTRETURN && (flags & ISLASTCN) != 0 &&
459 (mp->mnt_flag & MNT_RDONLY) != 0 &&
460 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
463 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
469 error = null_nodeget(mp, lvp, &vp);
478 null_open(struct vop_open_args *ap)
481 struct vnode *vp, *ldvp;
484 ldvp = NULLVPTOLOWERVP(vp);
485 retval = null_bypass(&ap->a_gen);
487 vp->v_object = ldvp->v_object;
488 if ((vn_irflag_read(ldvp) & VIRF_PGREAD) != 0) {
489 MPASS(vp->v_object != NULL);
490 if ((vn_irflag_read(vp) & VIRF_PGREAD) == 0) {
491 vn_irflag_set_cond(vp, VIRF_PGREAD);
499 * Setattr call. Disallow write attempts if the layer is mounted read-only.
502 null_setattr(struct vop_setattr_args *ap)
504 struct vnode *vp = ap->a_vp;
505 struct vattr *vap = ap->a_vap;
507 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
508 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
509 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
510 (vp->v_mount->mnt_flag & MNT_RDONLY))
512 if (vap->va_size != VNOVAL) {
513 switch (vp->v_type) {
520 if (vap->va_flags != VNOVAL)
527 * Disallow write attempts if the filesystem is
530 if (vp->v_mount->mnt_flag & MNT_RDONLY)
535 return (null_bypass((struct vop_generic_args *)ap));
539 * We handle stat and getattr only to change the fsid.
542 null_stat(struct vop_stat_args *ap)
546 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0)
549 ap->a_sb->st_dev = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
554 null_getattr(struct vop_getattr_args *ap)
558 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0)
561 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
566 * Handle to disallow write access if mounted read-only.
569 null_access(struct vop_access_args *ap)
571 struct vnode *vp = ap->a_vp;
572 accmode_t accmode = ap->a_accmode;
575 * Disallow write attempts on read-only layers;
576 * unless the file is a socket, fifo, or a block or
577 * character device resident on the filesystem.
579 if (accmode & VWRITE) {
580 switch (vp->v_type) {
584 if (vp->v_mount->mnt_flag & MNT_RDONLY)
591 return (null_bypass((struct vop_generic_args *)ap));
595 null_accessx(struct vop_accessx_args *ap)
597 struct vnode *vp = ap->a_vp;
598 accmode_t accmode = ap->a_accmode;
601 * Disallow write attempts on read-only layers;
602 * unless the file is a socket, fifo, or a block or
603 * character device resident on the filesystem.
605 if (accmode & VWRITE) {
606 switch (vp->v_type) {
610 if (vp->v_mount->mnt_flag & MNT_RDONLY)
617 return (null_bypass((struct vop_generic_args *)ap));
621 * Increasing refcount of lower vnode is needed at least for the case
622 * when lower FS is NFS to do sillyrename if the file is in use.
623 * Unfortunately v_usecount is incremented in many places in
624 * the kernel and, as such, there may be races that result in
625 * the NFS client doing an extraneous silly rename, but that seems
626 * preferable to not doing a silly rename when it is needed.
629 null_remove(struct vop_remove_args *ap)
632 struct vnode *lvp, *vp;
635 if (vrefcnt(vp) > 1) {
636 lvp = NULLVPTOLOWERVP(vp);
641 VTONULL(vp)->null_flags |= NULLV_DROP;
642 retval = null_bypass(&ap->a_gen);
649 * We handle this to eliminate null FS to lower FS
650 * file moving. Don't know why we don't allow this,
651 * possibly we should.
654 null_rename(struct vop_rename_args *ap)
656 struct vnode *fdvp, *fvp, *tdvp, *tvp;
657 struct vnode *lfdvp, *lfvp, *ltdvp, *ltvp;
658 struct null_node *fdnn, *fnn, *tdnn, *tnn;
667 /* Check for cross-device rename. */
668 if ((fvp->v_mount != tdvp->v_mount) ||
669 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
675 fdnn = VTONULL(fdvp);
676 if (fdnn == NULL) { /* fdvp is not locked, can be doomed */
681 lfdvp = fdnn->null_lowervp;
692 lfvp = fnn->null_lowervp;
696 tdnn = VTONULL(tdvp);
697 ltdvp = tdnn->null_lowervp;
702 ltvp = tnn->null_lowervp;
704 tnn->null_flags |= NULLV_DROP;
709 error = VOP_RENAME(lfdvp, lfvp, ap->a_fcnp, ltdvp, ltvp, ap->a_tcnp);
732 null_rmdir(struct vop_rmdir_args *ap)
735 VTONULL(ap->a_vp)->null_flags |= NULLV_DROP;
736 return (null_bypass(&ap->a_gen));
740 * We need to process our own vnode lock and then clear the
741 * interlock flag as it applies only to our vnode, not the
742 * vnodes below us on the stack.
745 null_lock(struct vop_lock1_args *ap)
747 struct vnode *vp = ap->a_vp;
749 struct null_node *nn;
753 if ((ap->a_flags & LK_INTERLOCK) == 0)
756 ap->a_flags &= ~LK_INTERLOCK;
760 * If we're still active we must ask the lower layer to
761 * lock as ffs has special lock considerations in its
764 if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
766 * We have to hold the vnode here to solve a potential
767 * reclaim race. If we're forcibly vgone'd while we
768 * still have refs, a thread could be sleeping inside
769 * the lowervp's vop_lock routine. When we vgone we will
770 * drop our last ref to the lowervp, which would allow it
771 * to be reclaimed. The lowervp could then be recycled,
772 * in which case it is not legal to be sleeping in its VOP.
773 * We prevent it from being recycled by holding the vnode
778 error = VOP_LOCK(lvp, flags);
781 * We might have slept to get the lock and someone might have
782 * clean our vnode already, switching vnode lock from one in
783 * lowervp to v_lock in our own vnode structure. Handle this
784 * case by reacquiring correct lock in requested mode.
786 if (VTONULL(vp) == NULL && error == 0) {
787 ap->a_flags &= ~LK_TYPE_MASK;
788 switch (flags & LK_TYPE_MASK) {
790 ap->a_flags |= LK_SHARED;
794 ap->a_flags |= LK_EXCLUSIVE;
797 panic("Unsupported lock request %d\n",
801 error = vop_stdlock(ap);
806 error = vop_stdlock(ap);
813 * We need to process our own vnode unlock and then clear the
814 * interlock flag as it applies only to our vnode, not the
815 * vnodes below us on the stack.
818 null_unlock(struct vop_unlock_args *ap)
820 struct vnode *vp = ap->a_vp;
821 struct null_node *nn;
826 if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
828 error = VOP_UNLOCK(lvp);
831 error = vop_stdunlock(ap);
838 * Do not allow the VOP_INACTIVE to be passed to the lower layer,
839 * since the reference count on the lower vnode is not related to
843 null_want_recycle(struct vnode *vp)
846 struct null_node *xp;
848 struct null_mount *xmp;
851 lvp = NULLVPTOLOWERVP(vp);
853 xmp = MOUNTTONULLMOUNT(mp);
854 if ((xmp->nullm_flags & NULLM_CACHE) == 0 ||
855 (xp->null_flags & NULLV_DROP) != 0 ||
856 (lvp->v_vflag & VV_NOSYNC) != 0) {
858 * If this is the last reference and caching of the
859 * nullfs vnodes is not enabled, or the lower vnode is
860 * deleted, then free up the vnode so as not to tie up
869 null_inactive(struct vop_inactive_args *ap)
874 if (null_want_recycle(vp)) {
882 null_need_inactive(struct vop_need_inactive_args *ap)
885 return (null_want_recycle(ap->a_vp) || vn_need_pageq_flush(ap->a_vp));
889 * Now, the nullfs vnode and, due to the sharing lock, the lower
890 * vnode, are exclusively locked, and we shall destroy the null vnode.
893 null_reclaim(struct vop_reclaim_args *ap)
896 struct null_node *xp;
897 struct vnode *lowervp;
901 lowervp = xp->null_lowervp;
903 KASSERT(lowervp != NULL && vp->v_vnlock != &vp->v_lock,
904 ("Reclaiming incomplete null vnode %p", vp));
908 * Use the interlock to protect the clearing of v_data to
909 * prevent faults in null_lock().
911 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
915 vp->v_vnlock = &vp->v_lock;
918 * If we were opened for write, we leased the write reference
919 * to the lower vnode. If this is a reclamation due to the
920 * forced unmount, undo the reference now.
922 if (vp->v_writecount > 0)
923 VOP_ADD_WRITECOUNT(lowervp, -vp->v_writecount);
924 else if (vp->v_writecount < 0)
925 vp->v_writecount = 0;
929 if ((xp->null_flags & NULLV_NOUNLOCK) != 0)
933 free(xp, M_NULLFSNODE);
939 null_print(struct vop_print_args *ap)
941 struct vnode *vp = ap->a_vp;
943 printf("\tvp=%p, lowervp=%p\n", vp, VTONULL(vp)->null_lowervp);
949 null_getwritemount(struct vop_getwritemount_args *ap)
951 struct null_node *xp;
952 struct vnode *lowervp;
958 if (xp && (lowervp = xp->null_lowervp)) {
961 VOP_GETWRITEMOUNT(lowervp, ap->a_mpp);
971 null_vptofh(struct vop_vptofh_args *ap)
975 lvp = NULLVPTOLOWERVP(ap->a_vp);
976 return VOP_VPTOFH(lvp, ap->a_fhp);
980 null_vptocnp(struct vop_vptocnp_args *ap)
982 struct vnode *vp = ap->a_vp;
983 struct vnode **dvp = ap->a_vpp;
984 struct vnode *lvp, *ldvp;
988 locked = VOP_ISLOCKED(vp);
989 lvp = NULLVPTOLOWERVP(vp);
991 error = vfs_busy(mp, MBF_NOWAIT);
995 VOP_UNLOCK(vp); /* vp is held by vn_vptocnp_locked that called us */
998 error = vn_vptocnp(&ldvp, ap->a_buf, ap->a_buflen);
1001 vn_lock(vp, locked | LK_RETRY);
1006 error = vn_lock(ldvp, LK_SHARED);
1009 vn_lock(vp, locked | LK_RETRY);
1013 error = null_nodeget(mp, ldvp, dvp);
1016 NULLVPTOLOWERVP(*dvp);
1018 VOP_UNLOCK(*dvp); /* keep reference on *dvp */
1020 vn_lock(vp, locked | LK_RETRY);
1026 null_read_pgcache(struct vop_read_pgcache_args *ap)
1028 struct vnode *lvp, *vp;
1029 struct null_node *xp;
1037 return (EJUSTRETURN);
1039 lvp = xp->null_lowervp;
1042 error = VOP_READ_PGCACHE(lvp, ap->a_uio, ap->a_ioflag, ap->a_cred);
1048 null_advlock(struct vop_advlock_args *ap)
1050 struct vnode *lvp, *vp;
1051 struct null_node *xp;
1061 lvp = xp->null_lowervp;
1064 error = VOP_ADVLOCK(lvp, ap->a_id, ap->a_op, ap->a_fl, ap->a_flags);
1070 * Avoid standard bypass, since lower dvp and vp could be no longer
1071 * valid after vput().
1074 null_vput_pair(struct vop_vput_pair_args *ap)
1077 struct vnode *dvp, *ldvp, *lvp, *vp, *vp1, **vpp;
1081 ldvp = NULLVPTOLOWERVP(dvp);
1091 lvp = NULLVPTOLOWERVP(vp);
1093 if (!ap->a_unlock_vp) {
1101 res = VOP_VPUT_PAIR(ldvp, lvp != NULL ? &lvp : NULL, true);
1102 if (vp != NULL && ap->a_unlock_vp)
1106 if (vp == NULL || ap->a_unlock_vp)
1109 /* lvp has been unlocked and vp might be reclaimed */
1110 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1111 if (vp->v_data == NULL && vfs_busy(mp, MBF_NOWAIT) == 0) {
1113 vget(lvp, LK_EXCLUSIVE | LK_RETRY);
1114 if (VN_IS_DOOMED(lvp)) {
1116 vget(vp, LK_EXCLUSIVE | LK_RETRY);
1118 error = null_nodeget(mp, lvp, &vp1);
1122 vget(vp, LK_EXCLUSIVE | LK_RETRY);
1135 * Global vfs data structures
1137 struct vop_vector null_vnodeops = {
1138 .vop_bypass = null_bypass,
1139 .vop_access = null_access,
1140 .vop_accessx = null_accessx,
1141 .vop_advlock = null_advlock,
1142 .vop_advlockpurge = vop_stdadvlockpurge,
1143 .vop_bmap = VOP_EOPNOTSUPP,
1144 .vop_stat = null_stat,
1145 .vop_getattr = null_getattr,
1146 .vop_getwritemount = null_getwritemount,
1147 .vop_inactive = null_inactive,
1148 .vop_need_inactive = null_need_inactive,
1149 .vop_islocked = vop_stdislocked,
1150 .vop_lock1 = null_lock,
1151 .vop_lookup = null_lookup,
1152 .vop_open = null_open,
1153 .vop_print = null_print,
1154 .vop_read_pgcache = null_read_pgcache,
1155 .vop_reclaim = null_reclaim,
1156 .vop_remove = null_remove,
1157 .vop_rename = null_rename,
1158 .vop_rmdir = null_rmdir,
1159 .vop_setattr = null_setattr,
1160 .vop_strategy = VOP_EOPNOTSUPP,
1161 .vop_unlock = null_unlock,
1162 .vop_vptocnp = null_vptocnp,
1163 .vop_vptofh = null_vptofh,
1164 .vop_add_writecount = null_add_writecount,
1165 .vop_vput_pair = null_vput_pair,
1167 VFS_VOP_VECTOR_REGISTER(null_vnodeops);