2 * Copyright (c) 1992, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * John Heidemann of the UCLA Ficus project.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * @(#)null_vnops.c 8.6 (Berkeley) 5/27/95
35 * @(#)lofs_vnops.c 1.2 (Berkeley) 6/18/92
37 * @(#)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 examing 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 existance 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>
184 #include <fs/nullfs/null.h>
187 #include <vm/vm_extern.h>
188 #include <vm/vm_object.h>
189 #include <vm/vnode_pager.h>
191 static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */
192 SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW,
193 &null_bug_bypass, 0, "");
196 * This is the 10-Apr-92 bypass routine.
197 * This version has been optimized for speed, throwing away some
198 * safety checks. It should still always work, but it's not as
199 * robust to programmer errors.
201 * In general, we map all vnodes going down and unmap them on the way back.
202 * As an exception to this, vnodes can be marked "unmapped" by setting
203 * the Nth bit in operation's vdesc_flags.
205 * Also, some BSD vnode operations have the side effect of vrele'ing
206 * their arguments. With stacking, the reference counts are held
207 * by the upper node, not the lower one, so we must handle these
208 * side-effects here. This is not of concern in Sun-derived systems
209 * since there are no such side-effects.
211 * This makes the following assumptions:
212 * - only one returned vpp
213 * - no INOUT vpp's (Sun's vop_open has one of these)
214 * - the vnode operation vector of the first vnode should be used
215 * to determine what implementation of the op should be invoked
216 * - all mapped vnodes are of our vnode-type (NEEDSWORK:
217 * problems on rmdir'ing mount points and renaming?)
220 null_bypass(struct vop_generic_args *ap)
222 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;
227 struct vnodeop_desc *descp = ap->a_desc;
231 printf ("null_bypass: %s\n", descp->vdesc_name);
235 * We require at least one vp.
237 if (descp->vdesc_vp_offsets == NULL ||
238 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
239 panic ("null_bypass: no vp's in map");
243 * Map the vnodes going in.
244 * Later, we'll invoke the operation based on
245 * the first mapped vnode's operation vector.
247 reles = descp->vdesc_flags;
248 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
249 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
250 break; /* bail out at end of list */
251 vps_p[i] = this_vp_p =
252 VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap);
254 * We're not guaranteed that any but the first vnode
255 * are of our type. Check for and don't map any
256 * that aren't. (We must always map first vp or vclean fails.)
258 if (i && (*this_vp_p == NULLVP ||
259 (*this_vp_p)->v_op != &null_vnodeops)) {
262 old_vps[i] = *this_vp_p;
263 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
265 * XXX - Several operations have the side effect
266 * of vrele'ing their vp's. We must account for
267 * that. (This should go away in the future.)
269 if (reles & VDESC_VP0_WILLRELE)
276 * Call the operation on the lower layer
277 * with the modified argument structure.
279 if (vps_p[0] && *vps_p[0])
282 printf("null_bypass: no map for %s\n", descp->vdesc_name);
287 * Maintain the illusion of call-by-value
288 * by restoring vnodes in the argument structure
289 * to their original value.
291 reles = descp->vdesc_flags;
292 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
293 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
294 break; /* bail out at end of list */
296 *(vps_p[i]) = old_vps[i];
298 if (reles & VDESC_VP0_WILLUNLOCK)
299 VOP_UNLOCK(*(vps_p[i]), 0, curthread);
301 if (reles & VDESC_VP0_WILLRELE)
307 * Map the possible out-going vpp
308 * (Assumes that the lower layer always returns
309 * a VREF'ed vpp unless it gets an error.)
311 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
312 !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
315 * XXX - even though some ops have vpp returned vp's,
316 * several ops actually vrele this before returning.
317 * We must avoid these ops.
318 * (This should go away when these ops are regularized.)
320 if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
322 vppp = VOPARG_OFFSETTO(struct vnode***,
323 descp->vdesc_vpp_offset,ap);
325 error = null_nodeget(old_vps[0]->v_mount, **vppp, *vppp);
333 * We have to carry on the locking protocol on the null layer vnodes
334 * as we progress through the tree. We also have to enforce read-only
335 * if this layer is mounted read-only.
338 null_lookup(struct vop_lookup_args *ap)
340 struct componentname *cnp = ap->a_cnp;
341 struct vnode *dvp = ap->a_dvp;
342 int flags = cnp->cn_flags;
343 struct vnode *vp, *ldvp, *lvp;
346 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
347 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
350 * Although it is possible to call null_bypass(), we'll do
351 * a direct call to reduce overhead
353 ldvp = NULLVPTOLOWERVP(dvp);
355 error = VOP_LOOKUP(ldvp, &lvp, cnp);
356 if (error == EJUSTRETURN && (flags & ISLASTCN) &&
357 (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
358 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
361 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
367 error = null_nodeget(dvp->v_mount, lvp, &vp);
369 /* XXX Cleanup needed... */
370 panic("null_nodeget failed");
379 null_open(struct vop_open_args *ap)
382 struct vnode *vp, *ldvp;
385 ldvp = NULLVPTOLOWERVP(vp);
386 retval = null_bypass(&ap->a_gen);
388 vp->v_object = ldvp->v_object;
393 * Setattr call. Disallow write attempts if the layer is mounted read-only.
396 null_setattr(struct vop_setattr_args *ap)
398 struct vnode *vp = ap->a_vp;
399 struct vattr *vap = ap->a_vap;
401 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
402 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
403 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
404 (vp->v_mount->mnt_flag & MNT_RDONLY))
406 if (vap->va_size != VNOVAL) {
407 switch (vp->v_type) {
414 if (vap->va_flags != VNOVAL)
421 * Disallow write attempts if the filesystem is
424 if (vp->v_mount->mnt_flag & MNT_RDONLY)
429 return (null_bypass((struct vop_generic_args *)ap));
433 * We handle getattr only to change the fsid.
436 null_getattr(struct vop_getattr_args *ap)
440 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0)
443 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
448 * Handle to disallow write access if mounted read-only.
451 null_access(struct vop_access_args *ap)
453 struct vnode *vp = ap->a_vp;
454 mode_t mode = ap->a_mode;
457 * Disallow write attempts on read-only layers;
458 * unless the file is a socket, fifo, or a block or
459 * character device resident on the filesystem.
462 switch (vp->v_type) {
466 if (vp->v_mount->mnt_flag & MNT_RDONLY)
473 return (null_bypass((struct vop_generic_args *)ap));
477 * We handle this to eliminate null FS to lower FS
478 * file moving. Don't know why we don't allow this,
479 * possibly we should.
482 null_rename(struct vop_rename_args *ap)
484 struct vnode *tdvp = ap->a_tdvp;
485 struct vnode *fvp = ap->a_fvp;
486 struct vnode *fdvp = ap->a_fdvp;
487 struct vnode *tvp = ap->a_tvp;
489 /* Check for cross-device rename. */
490 if ((fvp->v_mount != tdvp->v_mount) ||
491 (tvp && (fvp->v_mount != tvp->v_mount))) {
503 return (null_bypass((struct vop_generic_args *)ap));
507 * We need to process our own vnode lock and then clear the
508 * interlock flag as it applies only to our vnode, not the
509 * vnodes below us on the stack.
512 null_lock(struct vop_lock1_args *ap)
514 struct vnode *vp = ap->a_vp;
515 int flags = ap->a_flags;
516 struct thread *td = ap->a_td;
517 struct null_node *nn;
522 if ((flags & LK_INTERLOCK) == 0) {
524 ap->a_flags = flags |= LK_INTERLOCK;
528 * If we're still active we must ask the lower layer to
529 * lock as ffs has special lock considerations in it's
532 if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
533 VI_LOCK_FLAGS(lvp, MTX_DUPOK);
536 * We have to hold the vnode here to solve a potential
537 * reclaim race. If we're forcibly vgone'd while we
538 * still have refs, a thread could be sleeping inside
539 * the lowervp's vop_lock routine. When we vgone we will
540 * drop our last ref to the lowervp, which would allow it
541 * to be reclaimed. The lowervp could then be recycled,
542 * in which case it is not legal to be sleeping in it's VOP.
543 * We prevent it from being recycled by holding the vnode
547 error = VOP_LOCK(lvp, flags, td);
550 * We might have slept to get the lock and someone might have
551 * clean our vnode already, switching vnode lock from one in
552 * lowervp to v_lock in our own vnode structure. Handle this
553 * case by reacquiring correct lock in requested mode.
555 if (VTONULL(vp) == NULL && error == 0) {
556 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
557 switch (flags & LK_TYPE_MASK) {
559 ap->a_flags |= LK_SHARED;
563 ap->a_flags |= LK_EXCLUSIVE;
566 panic("Unsupported lock request %d\n",
569 VOP_UNLOCK(lvp, 0, td);
570 error = vop_stdlock(ap);
574 error = vop_stdlock(ap);
580 * We need to process our own vnode unlock and then clear the
581 * interlock flag as it applies only to our vnode, not the
582 * vnodes below us on the stack.
585 null_unlock(struct vop_unlock_args *ap)
587 struct vnode *vp = ap->a_vp;
588 int flags = ap->a_flags;
589 struct thread *td = ap->a_td;
590 struct null_node *nn;
594 if ((flags & LK_INTERLOCK) != 0) {
596 ap->a_flags = flags &= ~LK_INTERLOCK;
599 if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL)
600 error = VOP_UNLOCK(lvp, flags, td);
602 error = vop_stdunlock(ap);
608 null_islocked(struct vop_islocked_args *ap)
610 struct vnode *vp = ap->a_vp;
611 struct thread *td = ap->a_td;
613 return (lockstatus(vp->v_vnlock, td));
617 * There is no way to tell that someone issued remove/rmdir operation
618 * on the underlying filesystem. For now we just have to release lowevrp
619 * as soon as possible.
621 * Note, we can't release any resources nor remove vnode from hash before
622 * appropriate VXLOCK stuff is is done because other process can find this
623 * vnode in hash during inactivation and may be sitting in vget() and waiting
624 * for null_inactive to unlock vnode. Thus we will do all those in VOP_RECLAIM.
627 null_inactive(struct vop_inactive_args *ap)
629 struct vnode *vp = ap->a_vp;
630 struct thread *td = ap->a_td;
635 * If this is the last reference, then free up the vnode
636 * so as not to tie up the lower vnodes.
644 * Now, the VXLOCK is in force and we're free to destroy the null vnode.
647 null_reclaim(struct vop_reclaim_args *ap)
649 struct vnode *vp = ap->a_vp;
650 struct null_node *xp = VTONULL(vp);
651 struct vnode *lowervp = xp->null_lowervp;
657 * Use the interlock to protect the clearing of v_data to
658 * prevent faults in null_lock().
663 vnlock = vp->v_vnlock;
664 vp->v_vnlock = &vp->v_lock;
666 lockmgr(vp->v_vnlock,
667 LK_EXCLUSIVE|LK_INTERLOCK, VI_MTX(vp), curthread);
670 panic("null_reclaim: reclaiming an node with now lowervp");
671 FREE(xp, M_NULLFSNODE);
677 null_print(struct vop_print_args *ap)
679 struct vnode *vp = ap->a_vp;
681 printf("\tvp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp));
687 null_getwritemount(struct vop_getwritemount_args *ap)
689 struct null_node *xp;
690 struct vnode *lowervp;
696 if (xp && (lowervp = xp->null_lowervp)) {
697 VI_LOCK_FLAGS(lowervp, MTX_DUPOK);
701 VOP_GETWRITEMOUNT(lowervp, ap->a_mpp);
711 null_vptofh(struct vop_vptofh_args *ap)
715 lvp = NULLVPTOLOWERVP(ap->a_vp);
716 return VOP_VPTOFH(lvp, ap->a_fhp);
720 * Global vfs data structures
722 struct vop_vector null_vnodeops = {
723 .vop_bypass = null_bypass,
724 .vop_access = null_access,
725 .vop_bmap = VOP_EOPNOTSUPP,
726 .vop_getattr = null_getattr,
727 .vop_getwritemount = null_getwritemount,
728 .vop_inactive = null_inactive,
729 .vop_islocked = null_islocked,
730 .vop_lock1 = null_lock,
731 .vop_lookup = null_lookup,
732 .vop_open = null_open,
733 .vop_print = null_print,
734 .vop_reclaim = null_reclaim,
735 .vop_rename = null_rename,
736 .vop_setattr = null_setattr,
737 .vop_strategy = VOP_EOPNOTSUPP,
738 .vop_unlock = null_unlock,
739 .vop_vptofh = null_vptofh,