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
47 * (See mount_nullfs(8) for more information.)
49 * The null layer duplicates a portion of the filesystem
50 * name space under a new name. In this respect, it is
51 * similar to the loopback filesystem. It differs from
52 * the loopback fs in two respects: it is implemented using
53 * a stackable layers techniques, and its "null-node"s stack above
54 * all lower-layer vnodes, not just over directory vnodes.
56 * The null layer has two purposes. First, it serves as a demonstration
57 * of layering by proving a layer which does nothing. (It actually
58 * does everything the loopback filesystem does, which is slightly
59 * more than nothing.) Second, the null layer can serve as a prototype
60 * layer. Since it provides all necessary layer framework,
61 * new filesystem layers can be created very easily be starting
64 * The remainder of this man page examines the null layer as a basis
65 * for constructing new layers.
68 * INSTANTIATING NEW NULL LAYERS
70 * New null layers are created with mount_nullfs(8).
71 * Mount_nullfs(8) takes two arguments, the pathname
72 * of the lower vfs (target-pn) and the pathname where the null
73 * layer will appear in the namespace (alias-pn). After
74 * the null layer is put into place, the contents
75 * of target-pn subtree will be aliased under alias-pn.
78 * OPERATION OF A NULL LAYER
80 * The null layer is the minimum filesystem layer,
81 * simply bypassing all possible operations to the lower layer
82 * for processing there. The majority of its activity centers
83 * on the bypass routine, through which nearly all vnode operations
86 * The bypass routine accepts arbitrary vnode operations for
87 * handling by the lower layer. It begins by examining vnode
88 * operation arguments and replacing any null-nodes by their
89 * lower-layer equivlants. It then invokes the operation
90 * on the lower layer. Finally, it replaces the null-nodes
91 * in the arguments and, if a vnode is return by the operation,
92 * stacks a null-node on top of the returned vnode.
94 * Although bypass handles most operations, vop_getattr, vop_lock,
95 * vop_unlock, vop_inactive, vop_reclaim, and vop_print are not
96 * bypassed. Vop_getattr must change the fsid being returned.
97 * Vop_lock and vop_unlock must handle any locking for the
98 * current vnode as well as pass the lock request down.
99 * Vop_inactive and vop_reclaim are not bypassed so that
100 * they can handle freeing null-layer specific data. Vop_print
101 * is not bypassed to avoid excessive debugging information.
102 * Also, certain vnode operations change the locking state within
103 * the operation (create, mknod, remove, link, rename, mkdir, rmdir,
104 * and symlink). Ideally these operations should not change the
105 * lock state, but should be changed to let the caller of the
106 * function unlock them. Otherwise all intermediate vnode layers
107 * (such as union, umapfs, etc) must catch these functions to do
108 * the necessary locking at their layer.
111 * INSTANTIATING VNODE STACKS
113 * Mounting associates the null layer with a lower layer,
114 * effect stacking two VFSes. Vnode stacks are instead
115 * created on demand as files are accessed.
117 * The initial mount creates a single vnode stack for the
118 * root of the new null layer. All other vnode stacks
119 * are created as a result of vnode operations on
120 * this or other null vnode stacks.
122 * New vnode stacks come into existence as a result of
123 * an operation which returns a vnode.
124 * The bypass routine stacks a null-node above the new
125 * vnode before returning it to the caller.
127 * For example, imagine mounting a null layer with
128 * "mount_nullfs /usr/include /dev/layer/null".
129 * Changing directory to /dev/layer/null will assign
130 * the root null-node (which was created when the null layer was mounted).
131 * Now consider opening "sys". A vop_lookup would be
132 * done on the root null-node. This operation would bypass through
133 * to the lower layer which would return a vnode representing
134 * the UFS "sys". Null_bypass then builds a null-node
135 * aliasing the UFS "sys" and returns this to the caller.
136 * Later operations on the null-node "sys" will repeat this
137 * process when constructing other vnode stacks.
140 * CREATING OTHER FILE SYSTEM LAYERS
142 * One of the easiest ways to construct new filesystem layers is to make
143 * a copy of the null layer, rename all files and variables, and
144 * then begin modifing the copy. Sed can be used to easily rename
147 * The umap layer is an example of a layer descended from the
151 * INVOKING OPERATIONS ON LOWER LAYERS
153 * There are two techniques to invoke operations on a lower layer
154 * when the operation cannot be completely bypassed. Each method
155 * is appropriate in different situations. In both cases,
156 * it is the responsibility of the aliasing layer to make
157 * the operation arguments "correct" for the lower layer
158 * by mapping a vnode arguments to the lower layer.
160 * The first approach is to call the aliasing layer's bypass routine.
161 * This method is most suitable when you wish to invoke the operation
162 * currently being handled on the lower layer. It has the advantage
163 * that the bypass routine already must do argument mapping.
164 * An example of this is null_getattrs in the null layer.
166 * A second approach is to directly invoke vnode operations on
167 * the lower layer with the VOP_OPERATIONNAME interface.
168 * The advantage of this method is that it is easy to invoke
169 * arbitrary operations on the lower layer. The disadvantage
170 * is that vnode arguments must be manualy mapped.
174 #include <sys/param.h>
175 #include <sys/systm.h>
176 #include <sys/conf.h>
177 #include <sys/kernel.h>
178 #include <sys/lock.h>
179 #include <sys/malloc.h>
180 #include <sys/mount.h>
181 #include <sys/mutex.h>
182 #include <sys/namei.h>
183 #include <sys/sysctl.h>
184 #include <sys/vnode.h>
185 #include <sys/stat.h>
187 #include <fs/nullfs/null.h>
190 #include <vm/vm_extern.h>
191 #include <vm/vm_object.h>
192 #include <vm/vnode_pager.h>
194 static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */
195 SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW,
196 &null_bug_bypass, 0, "");
199 * This is the 10-Apr-92 bypass routine.
200 * This version has been optimized for speed, throwing away some
201 * safety checks. It should still always work, but it's not as
202 * robust to programmer errors.
204 * In general, we map all vnodes going down and unmap them on the way back.
205 * As an exception to this, vnodes can be marked "unmapped" by setting
206 * the Nth bit in operation's vdesc_flags.
208 * Also, some BSD vnode operations have the side effect of vrele'ing
209 * their arguments. With stacking, the reference counts are held
210 * by the upper node, not the lower one, so we must handle these
211 * side-effects here. This is not of concern in Sun-derived systems
212 * since there are no such side-effects.
214 * This makes the following assumptions:
215 * - only one returned vpp
216 * - no INOUT vpp's (Sun's vop_open has one of these)
217 * - the vnode operation vector of the first vnode should be used
218 * to determine what implementation of the op should be invoked
219 * - all mapped vnodes are of our vnode-type (NEEDSWORK:
220 * problems on rmdir'ing mount points and renaming?)
223 null_bypass(struct vop_generic_args *ap)
225 struct vnode **this_vp_p;
226 struct vnode *old_vps[VDESC_MAX_VPS];
227 struct vnode **vps_p[VDESC_MAX_VPS];
228 struct vnode ***vppp;
230 struct vnodeop_desc *descp = ap->a_desc;
234 printf ("null_bypass: %s\n", descp->vdesc_name);
238 * We require at least one vp.
240 if (descp->vdesc_vp_offsets == NULL ||
241 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
242 panic ("null_bypass: no vp's in map");
246 * Map the vnodes going in.
247 * Later, we'll invoke the operation based on
248 * the first mapped vnode's operation vector.
250 reles = descp->vdesc_flags;
251 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
252 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
253 break; /* bail out at end of list */
254 vps_p[i] = this_vp_p = VOPARG_OFFSETTO(struct vnode **,
255 descp->vdesc_vp_offsets[i], ap);
258 * We're not guaranteed that any but the first vnode
259 * are of our type. Check for and don't map any
260 * that aren't. (We must always map first vp or vclean fails.)
262 if (i != 0 && (*this_vp_p == NULLVP ||
263 (*this_vp_p)->v_op != &null_vnodeops)) {
266 old_vps[i] = *this_vp_p;
267 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
270 * The upper vnode reference to the lower
271 * vnode is the only reference that keeps our
272 * pointer to the lower vnode alive. If lower
273 * vnode is relocked during the VOP call,
274 * upper vnode might become unlocked and
275 * reclaimed, which invalidates our reference.
276 * Add a transient hold around VOP call.
281 * XXX - Several operations have the side effect
282 * of vrele'ing their vp's. We must account for
283 * that. (This should go away in the future.)
285 if (reles & VDESC_VP0_WILLRELE)
291 * Call the operation on the lower layer
292 * with the modified argument structure.
294 if (vps_p[0] != NULL && *vps_p[0] != NULL) {
297 printf("null_bypass: no map for %s\n", descp->vdesc_name);
302 * Maintain the illusion of call-by-value
303 * by restoring vnodes in the argument structure
304 * to their original value.
306 reles = descp->vdesc_flags;
307 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
308 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
309 break; /* bail out at end of list */
310 if (old_vps[i] != NULL) {
314 * Get rid of the transient hold on lvp.
315 * If lowervp was unlocked during VOP
316 * operation, nullfs upper vnode could have
317 * been reclaimed, which changes its v_vnlock
318 * back to private v_lock. In this case we
319 * must move lock ownership from lower to
320 * upper (reclaimed) vnode.
323 if (VOP_ISLOCKED(lvp) == LK_EXCLUSIVE &&
324 old_vps[i]->v_vnlock != lvp->v_vnlock) {
326 VOP_LOCK(old_vps[i], LK_EXCLUSIVE |
332 *(vps_p[i]) = old_vps[i];
334 if (reles & VDESC_VP0_WILLUNLOCK)
335 VOP_UNLOCK(*(vps_p[i]), 0);
337 if (reles & VDESC_VP0_WILLRELE)
343 * Map the possible out-going vpp
344 * (Assumes that the lower layer always returns
345 * a VREF'ed vpp unless it gets an error.)
347 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && error == 0) {
349 * XXX - even though some ops have vpp returned vp's,
350 * several ops actually vrele this before returning.
351 * We must avoid these ops.
352 * (This should go away when these ops are regularized.)
354 vppp = VOPARG_OFFSETTO(struct vnode ***,
355 descp->vdesc_vpp_offset, ap);
357 error = null_nodeget(old_vps[0]->v_mount, **vppp,
365 null_add_writecount(struct vop_add_writecount_args *ap)
367 struct vnode *lvp, *vp;
371 lvp = NULLVPTOLOWERVP(vp);
373 /* text refs are bypassed to lowervp */
374 VNASSERT(vp->v_writecount >= 0, vp, ("wrong null writecount"));
375 VNASSERT(vp->v_writecount + ap->a_inc >= 0, vp,
376 ("wrong writecount inc %d", ap->a_inc));
377 error = VOP_ADD_WRITECOUNT(lvp, ap->a_inc);
379 vp->v_writecount += ap->a_inc;
385 * We have to carry on the locking protocol on the null layer vnodes
386 * as we progress through the tree. We also have to enforce read-only
387 * if this layer is mounted read-only.
390 null_lookup(struct vop_lookup_args *ap)
392 struct componentname *cnp = ap->a_cnp;
393 struct vnode *dvp = ap->a_dvp;
394 int flags = cnp->cn_flags;
395 struct vnode *vp, *ldvp, *lvp;
400 if ((flags & ISLASTCN) != 0 && (mp->mnt_flag & MNT_RDONLY) != 0 &&
401 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
404 * Although it is possible to call null_bypass(), we'll do
405 * a direct call to reduce overhead
407 ldvp = NULLVPTOLOWERVP(dvp);
411 * Renames in the lower mounts might create an inconsistent
412 * configuration where lower vnode is moved out of the
413 * directory tree remounted by our null mount. Do not try to
414 * handle it fancy, just avoid VOP_LOOKUP() with DOTDOT name
415 * which cannot be handled by VOP, at least passing over lower
418 if ((ldvp->v_vflag & VV_ROOT) != 0 && (flags & ISDOTDOT) != 0) {
419 KASSERT((dvp->v_vflag & VV_ROOT) == 0,
420 ("ldvp %p fl %#x dvp %p fl %#x flags %#x",
421 ldvp, ldvp->v_vflag, dvp, dvp->v_vflag, flags));
426 * Hold ldvp. The reference on it, owned by dvp, is lost in
427 * case of dvp reclamation, and we need ldvp to move our lock
432 error = VOP_LOOKUP(ldvp, &lvp, cnp);
435 * VOP_LOOKUP() on lower vnode may unlock ldvp, which allows
436 * dvp to be reclaimed due to shared v_vnlock. Check for the
437 * doomed state and return error.
439 if (VN_IS_DOOMED(dvp)) {
440 if (error == 0 || error == EJUSTRETURN) {
447 * If vgone() did reclaimed dvp before curthread
448 * relocked ldvp, the locks of dvp and ldpv are no
449 * longer shared. In this case, relock of ldvp in
450 * lower fs VOP_LOOKUP() does not restore the locking
451 * state of dvp. Compensate for this by unlocking
452 * ldvp and locking dvp, which is also correct if the
453 * locks are still shared.
456 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
460 if (error == EJUSTRETURN && (flags & ISLASTCN) != 0 &&
461 (mp->mnt_flag & MNT_RDONLY) != 0 &&
462 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
465 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
471 error = null_nodeget(mp, lvp, &vp);
480 null_open(struct vop_open_args *ap)
483 struct vnode *vp, *ldvp;
486 ldvp = NULLVPTOLOWERVP(vp);
487 retval = null_bypass(&ap->a_gen);
489 vp->v_object = ldvp->v_object;
490 if ((vn_irflag_read(ldvp) & VIRF_PGREAD) != 0) {
491 MPASS(vp->v_object != NULL);
492 if ((vn_irflag_read(vp) & VIRF_PGREAD) == 0) {
493 vn_irflag_set_cond(vp, VIRF_PGREAD);
501 * Setattr call. Disallow write attempts if the layer is mounted read-only.
504 null_setattr(struct vop_setattr_args *ap)
506 struct vnode *vp = ap->a_vp;
507 struct vattr *vap = ap->a_vap;
509 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
510 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
511 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
512 (vp->v_mount->mnt_flag & MNT_RDONLY))
514 if (vap->va_size != VNOVAL) {
515 switch (vp->v_type) {
522 if (vap->va_flags != VNOVAL)
529 * Disallow write attempts if the filesystem is
532 if (vp->v_mount->mnt_flag & MNT_RDONLY)
537 return (null_bypass((struct vop_generic_args *)ap));
541 * We handle stat and getattr only to change the fsid.
544 null_stat(struct vop_stat_args *ap)
548 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0)
551 ap->a_sb->st_dev = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
556 null_getattr(struct vop_getattr_args *ap)
560 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0)
563 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
568 * Handle to disallow write access if mounted read-only.
571 null_access(struct vop_access_args *ap)
573 struct vnode *vp = ap->a_vp;
574 accmode_t accmode = ap->a_accmode;
577 * Disallow write attempts on read-only layers;
578 * unless the file is a socket, fifo, or a block or
579 * character device resident on the filesystem.
581 if (accmode & VWRITE) {
582 switch (vp->v_type) {
586 if (vp->v_mount->mnt_flag & MNT_RDONLY)
593 return (null_bypass((struct vop_generic_args *)ap));
597 null_accessx(struct vop_accessx_args *ap)
599 struct vnode *vp = ap->a_vp;
600 accmode_t accmode = ap->a_accmode;
603 * Disallow write attempts on read-only layers;
604 * unless the file is a socket, fifo, or a block or
605 * character device resident on the filesystem.
607 if (accmode & VWRITE) {
608 switch (vp->v_type) {
612 if (vp->v_mount->mnt_flag & MNT_RDONLY)
619 return (null_bypass((struct vop_generic_args *)ap));
623 * Increasing refcount of lower vnode is needed at least for the case
624 * when lower FS is NFS to do sillyrename if the file is in use.
625 * Unfortunately v_usecount is incremented in many places in
626 * the kernel and, as such, there may be races that result in
627 * the NFS client doing an extraneous silly rename, but that seems
628 * preferable to not doing a silly rename when it is needed.
631 null_remove(struct vop_remove_args *ap)
634 struct vnode *lvp, *vp;
637 if (vrefcnt(vp) > 1) {
638 lvp = NULLVPTOLOWERVP(vp);
643 VTONULL(vp)->null_flags |= NULLV_DROP;
644 retval = null_bypass(&ap->a_gen);
651 * We handle this to eliminate null FS to lower FS
652 * file moving. Don't know why we don't allow this,
653 * possibly we should.
656 null_rename(struct vop_rename_args *ap)
658 struct vnode *fdvp, *fvp, *tdvp, *tvp;
659 struct vnode *lfdvp, *lfvp, *ltdvp, *ltvp;
660 struct null_node *fdnn, *fnn, *tdnn, *tnn;
669 /* Check for cross-device rename. */
670 if ((fvp->v_mount != tdvp->v_mount) ||
671 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
677 fdnn = VTONULL(fdvp);
678 if (fdnn == NULL) { /* fdvp is not locked, can be doomed */
683 lfdvp = fdnn->null_lowervp;
694 lfvp = fnn->null_lowervp;
698 tdnn = VTONULL(tdvp);
699 ltdvp = tdnn->null_lowervp;
704 ltvp = tnn->null_lowervp;
706 tnn->null_flags |= NULLV_DROP;
711 error = VOP_RENAME(lfdvp, lfvp, ap->a_fcnp, ltdvp, ltvp, ap->a_tcnp);
734 null_rmdir(struct vop_rmdir_args *ap)
737 VTONULL(ap->a_vp)->null_flags |= NULLV_DROP;
738 return (null_bypass(&ap->a_gen));
742 * We need to process our own vnode lock and then clear the
743 * interlock flag as it applies only to our vnode, not the
744 * vnodes below us on the stack.
747 null_lock(struct vop_lock1_args *ap)
749 struct vnode *vp = ap->a_vp;
751 struct null_node *nn;
755 if ((ap->a_flags & LK_INTERLOCK) == 0)
758 ap->a_flags &= ~LK_INTERLOCK;
762 * If we're still active we must ask the lower layer to
763 * lock as ffs has special lock considerations in its
766 if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
768 * We have to hold the vnode here to solve a potential
769 * reclaim race. If we're forcibly vgone'd while we
770 * still have refs, a thread could be sleeping inside
771 * the lowervp's vop_lock routine. When we vgone we will
772 * drop our last ref to the lowervp, which would allow it
773 * to be reclaimed. The lowervp could then be recycled,
774 * in which case it is not legal to be sleeping in its VOP.
775 * We prevent it from being recycled by holding the vnode
780 error = VOP_LOCK(lvp, flags);
783 * We might have slept to get the lock and someone might have
784 * clean our vnode already, switching vnode lock from one in
785 * lowervp to v_lock in our own vnode structure. Handle this
786 * case by reacquiring correct lock in requested mode.
788 if (VTONULL(vp) == NULL && error == 0) {
789 ap->a_flags &= ~LK_TYPE_MASK;
790 switch (flags & LK_TYPE_MASK) {
792 ap->a_flags |= LK_SHARED;
796 ap->a_flags |= LK_EXCLUSIVE;
799 panic("Unsupported lock request %d\n",
803 error = vop_stdlock(ap);
808 error = vop_stdlock(ap);
815 * We need to process our own vnode unlock and then clear the
816 * interlock flag as it applies only to our vnode, not the
817 * vnodes below us on the stack.
820 null_unlock(struct vop_unlock_args *ap)
822 struct vnode *vp = ap->a_vp;
823 struct null_node *nn;
828 if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
830 error = VOP_UNLOCK(lvp);
833 error = vop_stdunlock(ap);
840 * Do not allow the VOP_INACTIVE to be passed to the lower layer,
841 * since the reference count on the lower vnode is not related to
845 null_want_recycle(struct vnode *vp)
848 struct null_node *xp;
850 struct null_mount *xmp;
853 lvp = NULLVPTOLOWERVP(vp);
855 xmp = MOUNTTONULLMOUNT(mp);
856 if ((xmp->nullm_flags & NULLM_CACHE) == 0 ||
857 (xp->null_flags & NULLV_DROP) != 0 ||
858 (lvp->v_vflag & VV_NOSYNC) != 0) {
860 * If this is the last reference and caching of the
861 * nullfs vnodes is not enabled, or the lower vnode is
862 * deleted, then free up the vnode so as not to tie up
871 null_inactive(struct vop_inactive_args *ap)
876 if (null_want_recycle(vp)) {
884 null_need_inactive(struct vop_need_inactive_args *ap)
887 return (null_want_recycle(ap->a_vp) || vn_need_pageq_flush(ap->a_vp));
891 * Now, the nullfs vnode and, due to the sharing lock, the lower
892 * vnode, are exclusively locked, and we shall destroy the null vnode.
895 null_reclaim(struct vop_reclaim_args *ap)
898 struct null_node *xp;
899 struct vnode *lowervp;
903 lowervp = xp->null_lowervp;
905 KASSERT(lowervp != NULL && vp->v_vnlock != &vp->v_lock,
906 ("Reclaiming incomplete null vnode %p", vp));
910 * Use the interlock to protect the clearing of v_data to
911 * prevent faults in null_lock().
913 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
917 vp->v_vnlock = &vp->v_lock;
920 * If we were opened for write, we leased the write reference
921 * to the lower vnode. If this is a reclamation due to the
922 * forced unmount, undo the reference now.
924 if (vp->v_writecount > 0)
925 VOP_ADD_WRITECOUNT(lowervp, -vp->v_writecount);
926 else if (vp->v_writecount < 0)
927 vp->v_writecount = 0;
931 if ((xp->null_flags & NULLV_NOUNLOCK) != 0)
935 free(xp, M_NULLFSNODE);
941 null_print(struct vop_print_args *ap)
943 struct vnode *vp = ap->a_vp;
945 printf("\tvp=%p, lowervp=%p\n", vp, VTONULL(vp)->null_lowervp);
951 null_getwritemount(struct vop_getwritemount_args *ap)
953 struct null_node *xp;
954 struct vnode *lowervp;
960 if (xp && (lowervp = xp->null_lowervp)) {
963 VOP_GETWRITEMOUNT(lowervp, ap->a_mpp);
973 null_vptofh(struct vop_vptofh_args *ap)
977 lvp = NULLVPTOLOWERVP(ap->a_vp);
978 return VOP_VPTOFH(lvp, ap->a_fhp);
982 null_vptocnp(struct vop_vptocnp_args *ap)
984 struct vnode *vp = ap->a_vp;
985 struct vnode **dvp = ap->a_vpp;
986 struct vnode *lvp, *ldvp;
990 locked = VOP_ISLOCKED(vp);
991 lvp = NULLVPTOLOWERVP(vp);
993 error = vfs_busy(mp, MBF_NOWAIT);
997 VOP_UNLOCK(vp); /* vp is held by vn_vptocnp_locked that called us */
1000 error = vn_vptocnp(&ldvp, ap->a_buf, ap->a_buflen);
1003 vn_lock(vp, locked | LK_RETRY);
1008 error = vn_lock(ldvp, LK_SHARED);
1011 vn_lock(vp, locked | LK_RETRY);
1015 error = null_nodeget(mp, ldvp, dvp);
1018 NULLVPTOLOWERVP(*dvp);
1020 VOP_UNLOCK(*dvp); /* keep reference on *dvp */
1022 vn_lock(vp, locked | LK_RETRY);
1028 null_read_pgcache(struct vop_read_pgcache_args *ap)
1030 struct vnode *lvp, *vp;
1031 struct null_node *xp;
1039 return (EJUSTRETURN);
1041 lvp = xp->null_lowervp;
1044 error = VOP_READ_PGCACHE(lvp, ap->a_uio, ap->a_ioflag, ap->a_cred);
1050 null_advlock(struct vop_advlock_args *ap)
1052 struct vnode *lvp, *vp;
1053 struct null_node *xp;
1063 lvp = xp->null_lowervp;
1066 error = VOP_ADVLOCK(lvp, ap->a_id, ap->a_op, ap->a_fl, ap->a_flags);
1072 * Avoid standard bypass, since lower dvp and vp could be no longer
1073 * valid after vput().
1076 null_vput_pair(struct vop_vput_pair_args *ap)
1079 struct vnode *dvp, *ldvp, *lvp, *vp, *vp1, **vpp;
1083 ldvp = NULLVPTOLOWERVP(dvp);
1093 lvp = NULLVPTOLOWERVP(vp);
1095 if (!ap->a_unlock_vp) {
1103 res = VOP_VPUT_PAIR(ldvp, lvp != NULL ? &lvp : NULL, true);
1104 if (vp != NULL && ap->a_unlock_vp)
1108 if (vp == NULL || ap->a_unlock_vp)
1111 /* lvp has been unlocked and vp might be reclaimed */
1112 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1113 if (vp->v_data == NULL && vfs_busy(mp, MBF_NOWAIT) == 0) {
1115 vget(lvp, LK_EXCLUSIVE | LK_RETRY);
1116 if (VN_IS_DOOMED(lvp)) {
1118 vget(vp, LK_EXCLUSIVE | LK_RETRY);
1120 error = null_nodeget(mp, lvp, &vp1);
1124 vget(vp, LK_EXCLUSIVE | LK_RETRY);
1137 * Global vfs data structures
1139 struct vop_vector null_vnodeops = {
1140 .vop_bypass = null_bypass,
1141 .vop_access = null_access,
1142 .vop_accessx = null_accessx,
1143 .vop_advlock = null_advlock,
1144 .vop_advlockpurge = vop_stdadvlockpurge,
1145 .vop_bmap = VOP_EOPNOTSUPP,
1146 .vop_stat = null_stat,
1147 .vop_getattr = null_getattr,
1148 .vop_getwritemount = null_getwritemount,
1149 .vop_inactive = null_inactive,
1150 .vop_need_inactive = null_need_inactive,
1151 .vop_islocked = vop_stdislocked,
1152 .vop_lock1 = null_lock,
1153 .vop_lookup = null_lookup,
1154 .vop_open = null_open,
1155 .vop_print = null_print,
1156 .vop_read_pgcache = null_read_pgcache,
1157 .vop_reclaim = null_reclaim,
1158 .vop_remove = null_remove,
1159 .vop_rename = null_rename,
1160 .vop_rmdir = null_rmdir,
1161 .vop_setattr = null_setattr,
1162 .vop_strategy = VOP_EOPNOTSUPP,
1163 .vop_unlock = null_unlock,
1164 .vop_vptocnp = null_vptocnp,
1165 .vop_vptofh = null_vptofh,
1166 .vop_add_writecount = null_add_writecount,
1167 .vop_vput_pair = null_vput_pair,
1169 VFS_VOP_VECTOR_REGISTER(null_vnodeops);