1/*-
2 * Copyright (c) 1992, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * John Heidemann of the UCLA Ficus project.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
19 *
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
30 * SUCH DAMAGE.
31 *
32 * @(#)null_vnops.c 8.6 (Berkeley) 5/27/95
33 *
34 * Ancestors:
35 * @(#)lofs_vnops.c 1.2 (Berkeley) 6/18/92
36 * ...and...
37 * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project
38 *
39 * $FreeBSD: head/sys/fs/nullfs/null_vnops.c 252714 2013-07-04 19:01:18Z kib $
40 */
41
42/*
43 * Null Layer
44 *
45 * (See mount_nullfs(8) for more information.)
46 *
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.
53 *
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
60 * with a null layer.
61 *
62 * The remainder of this man page examines the null layer as a basis
63 * for constructing new layers.
64 *
65 *
66 * INSTANTIATING NEW NULL LAYERS
67 *
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.
74 *
75 *
76 * OPERATION OF A NULL LAYER
77 *
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
82 * pass.
83 *
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.
91 *
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.
107 *
108 *
109 * INSTANTIATING VNODE STACKS
110 *
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.
114 *
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.
119 *
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.
124 *
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.
136 *
137 *
138 * CREATING OTHER FILE SYSTEM LAYERS
139 *
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
143 * all variables.
144 *
145 * The umap layer is an example of a layer descended from the
146 * null layer.
147 *
148 *
149 * INVOKING OPERATIONS ON LOWER LAYERS
150 *
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.
157 *
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.
163 *
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.
169 *
170 */
171
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
184#include <fs/nullfs/null.h>
185
186#include <vm/vm.h>
187#include <vm/vm_extern.h>
188#include <vm/vm_object.h>
189#include <vm/vnode_pager.h>
190
191static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */
192SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW,
193 &null_bug_bypass, 0, "");
194
195/*
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.
200 *
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.
204 *
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.
210 *
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?)
218 */
219int
220null_bypass(struct vop_generic_args *ap)
221{
222 struct vnode **this_vp_p;
223 int error;
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;
228 int reles, i;
229
230 if (null_bug_bypass)
231 printf ("null_bypass: %s\n", descp->vdesc_name);
232
233#ifdef DIAGNOSTIC
234 /*
235 * We require at least one vp.
236 */
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");
240#endif
241
242 /*
243 * Map the vnodes going in.
244 * Later, we'll invoke the operation based on
245 * the first mapped vnode's operation vector.
246 */
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);
253 /*
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.)
257 */
258 if (i && (*this_vp_p == NULLVP ||
259 (*this_vp_p)->v_op != &null_vnodeops)) {
260 old_vps[i] = NULLVP;
261 } else {
262 old_vps[i] = *this_vp_p;
263 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
264 /*
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.)
268 */
269 if (reles & VDESC_VP0_WILLRELE)
270 VREF(*this_vp_p);
271 }
272
273 }
274
275 /*
276 * Call the operation on the lower layer
277 * with the modified argument structure.
278 */
279 if (vps_p[0] && *vps_p[0])
280 error = VCALL(ap);
281 else {
282 printf("null_bypass: no map for %s\n", descp->vdesc_name);
283 error = EINVAL;
284 }
285
286 /*
287 * Maintain the illusion of call-by-value
288 * by restoring vnodes in the argument structure
289 * to their original value.
290 */
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 */
295 if (old_vps[i]) {
296 *(vps_p[i]) = old_vps[i];
297#if 0
298 if (reles & VDESC_VP0_WILLUNLOCK)
299 VOP_UNLOCK(*(vps_p[i]), 0);
300#endif
301 if (reles & VDESC_VP0_WILLRELE)
302 vrele(*(vps_p[i]));
303 }
304 }
305
306 /*
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.)
310 */
311 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
312 !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
313 !error) {
314 /*
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.)
319 */
320 if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
321 goto out;
322 vppp = VOPARG_OFFSETTO(struct vnode***,
323 descp->vdesc_vpp_offset,ap);
324 if (*vppp)
325 error = null_nodeget(old_vps[0]->v_mount, **vppp, *vppp);
326 }
327
328 out:
329 return (error);
330}
331
332static int
333null_add_writecount(struct vop_add_writecount_args *ap)
334{
335 struct vnode *lvp, *vp;
336 int error;
337
338 vp = ap->a_vp;
339 lvp = NULLVPTOLOWERVP(vp);
340 KASSERT(vp->v_writecount + ap->a_inc >= 0, ("wrong writecount inc"));
341 if (vp->v_writecount > 0 && vp->v_writecount + ap->a_inc == 0)
342 error = VOP_ADD_WRITECOUNT(lvp, -1);
343 else if (vp->v_writecount == 0 && vp->v_writecount + ap->a_inc > 0)
344 error = VOP_ADD_WRITECOUNT(lvp, 1);
345 else
346 error = 0;
347 if (error == 0)
348 vp->v_writecount += ap->a_inc;
349 return (error);
350}
351
352/*
353 * We have to carry on the locking protocol on the null layer vnodes
354 * as we progress through the tree. We also have to enforce read-only
355 * if this layer is mounted read-only.
356 */
357static int
358null_lookup(struct vop_lookup_args *ap)
359{
360 struct componentname *cnp = ap->a_cnp;
361 struct vnode *dvp = ap->a_dvp;
362 int flags = cnp->cn_flags;
363 struct vnode *vp, *ldvp, *lvp;
364 int error;
365
366 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
367 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
368 return (EROFS);
369 /*
370 * Although it is possible to call null_bypass(), we'll do
371 * a direct call to reduce overhead
372 */
373 ldvp = NULLVPTOLOWERVP(dvp);
374 vp = lvp = NULL;
375 error = VOP_LOOKUP(ldvp, &lvp, cnp);
376 if (error == EJUSTRETURN && (flags & ISLASTCN) &&
377 (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
378 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
379 error = EROFS;
380
381 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
382 if (ldvp == lvp) {
383 *ap->a_vpp = dvp;
384 VREF(dvp);
385 vrele(lvp);
386 } else {
387 error = null_nodeget(dvp->v_mount, lvp, &vp);
388 if (error == 0)
389 *ap->a_vpp = vp;
390 }
391 }
392 return (error);
393}
394
395static int
396null_open(struct vop_open_args *ap)
397{
398 int retval;
399 struct vnode *vp, *ldvp;
400
401 vp = ap->a_vp;
402 ldvp = NULLVPTOLOWERVP(vp);
403 retval = null_bypass(&ap->a_gen);
404 if (retval == 0)
405 vp->v_object = ldvp->v_object;
406 return (retval);
407}
408
409/*
410 * Setattr call. Disallow write attempts if the layer is mounted read-only.
411 */
412static int
413null_setattr(struct vop_setattr_args *ap)
414{
415 struct vnode *vp = ap->a_vp;
416 struct vattr *vap = ap->a_vap;
417
418 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
419 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
420 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
421 (vp->v_mount->mnt_flag & MNT_RDONLY))
422 return (EROFS);
423 if (vap->va_size != VNOVAL) {
424 switch (vp->v_type) {
425 case VDIR:
426 return (EISDIR);
427 case VCHR:
428 case VBLK:
429 case VSOCK:
430 case VFIFO:
431 if (vap->va_flags != VNOVAL)
432 return (EOPNOTSUPP);
433 return (0);
434 case VREG:
435 case VLNK:
436 default:
437 /*
438 * Disallow write attempts if the filesystem is
439 * mounted read-only.
440 */
441 if (vp->v_mount->mnt_flag & MNT_RDONLY)
442 return (EROFS);
443 }
444 }
445
446 return (null_bypass((struct vop_generic_args *)ap));
447}
448
449/*
450 * We handle getattr only to change the fsid.
451 */
452static int
453null_getattr(struct vop_getattr_args *ap)
454{
455 int error;
456
457 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0)
458 return (error);
459
460 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
461 return (0);
462}
463
464/*
465 * Handle to disallow write access if mounted read-only.
466 */
467static int
468null_access(struct vop_access_args *ap)
469{
470 struct vnode *vp = ap->a_vp;
471 accmode_t accmode = ap->a_accmode;
472
473 /*
474 * Disallow write attempts on read-only layers;
475 * unless the file is a socket, fifo, or a block or
476 * character device resident on the filesystem.
477 */
478 if (accmode & VWRITE) {
479 switch (vp->v_type) {
480 case VDIR:
481 case VLNK:
482 case VREG:
483 if (vp->v_mount->mnt_flag & MNT_RDONLY)
484 return (EROFS);
485 break;
486 default:
487 break;
488 }
489 }
490 return (null_bypass((struct vop_generic_args *)ap));
491}
492
493static int
494null_accessx(struct vop_accessx_args *ap)
495{
496 struct vnode *vp = ap->a_vp;
497 accmode_t accmode = ap->a_accmode;
498
499 /*
500 * Disallow write attempts on read-only layers;
501 * unless the file is a socket, fifo, or a block or
502 * character device resident on the filesystem.
503 */
504 if (accmode & VWRITE) {
505 switch (vp->v_type) {
506 case VDIR:
507 case VLNK:
508 case VREG:
509 if (vp->v_mount->mnt_flag & MNT_RDONLY)
510 return (EROFS);
511 break;
512 default:
513 break;
514 }
515 }
516 return (null_bypass((struct vop_generic_args *)ap));
517}
518
519/*
520 * Increasing refcount of lower vnode is needed at least for the case
521 * when lower FS is NFS to do sillyrename if the file is in use.
522 * Unfortunately v_usecount is incremented in many places in
523 * the kernel and, as such, there may be races that result in
524 * the NFS client doing an extraneous silly rename, but that seems
525 * preferable to not doing a silly rename when it is needed.
526 */
527static int
528null_remove(struct vop_remove_args *ap)
529{
530 int retval, vreleit;
531 struct vnode *lvp;
532
533 if (vrefcnt(ap->a_vp) > 1) {
534 lvp = NULLVPTOLOWERVP(ap->a_vp);
535 VREF(lvp);
536 vreleit = 1;
537 } else
538 vreleit = 0;
539 retval = null_bypass(&ap->a_gen);
540 if (vreleit != 0)
541 vrele(lvp);
542 return (retval);
543}
544
545/*
546 * We handle this to eliminate null FS to lower FS
547 * file moving. Don't know why we don't allow this,
548 * possibly we should.
549 */
550static int
551null_rename(struct vop_rename_args *ap)
552{
553 struct vnode *tdvp = ap->a_tdvp;
554 struct vnode *fvp = ap->a_fvp;
555 struct vnode *fdvp = ap->a_fdvp;
556 struct vnode *tvp = ap->a_tvp;
557 struct null_node *tnn;
558
559 /* Check for cross-device rename. */
560 if ((fvp->v_mount != tdvp->v_mount) ||
561 (tvp && (fvp->v_mount != tvp->v_mount))) {
562 if (tdvp == tvp)
563 vrele(tdvp);
564 else
565 vput(tdvp);
566 if (tvp)
567 vput(tvp);
568 vrele(fdvp);
569 vrele(fvp);
570 return (EXDEV);
571 }
572
573 if (tvp != NULL) {
574 tnn = VTONULL(tvp);
575 tnn->null_flags |= NULLV_DROP;
576 }
577 return (null_bypass((struct vop_generic_args *)ap));
578}
579
580/*
581 * We need to process our own vnode lock and then clear the
582 * interlock flag as it applies only to our vnode, not the
583 * vnodes below us on the stack.
584 */
585static int
586null_lock(struct vop_lock1_args *ap)
587{
588 struct vnode *vp = ap->a_vp;
589 int flags = ap->a_flags;
590 struct null_node *nn;
591 struct vnode *lvp;
592 int error;
593
594
595 if ((flags & LK_INTERLOCK) == 0) {
596 VI_LOCK(vp);
597 ap->a_flags = flags |= LK_INTERLOCK;
598 }
599 nn = VTONULL(vp);
600 /*
601 * If we're still active we must ask the lower layer to
602 * lock as ffs has special lock considerations in it's
603 * vop lock.
604 */
605 if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
606 VI_LOCK_FLAGS(lvp, MTX_DUPOK);
607 VI_UNLOCK(vp);
608 /*
609 * We have to hold the vnode here to solve a potential
610 * reclaim race. If we're forcibly vgone'd while we
611 * still have refs, a thread could be sleeping inside
612 * the lowervp's vop_lock routine. When we vgone we will
613 * drop our last ref to the lowervp, which would allow it
614 * to be reclaimed. The lowervp could then be recycled,
615 * in which case it is not legal to be sleeping in it's VOP.
616 * We prevent it from being recycled by holding the vnode
617 * here.
618 */
619 vholdl(lvp);
620 error = VOP_LOCK(lvp, flags);
621
622 /*
623 * We might have slept to get the lock and someone might have
624 * clean our vnode already, switching vnode lock from one in
625 * lowervp to v_lock in our own vnode structure. Handle this
626 * case by reacquiring correct lock in requested mode.
627 */
628 if (VTONULL(vp) == NULL && error == 0) {
629 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
630 switch (flags & LK_TYPE_MASK) {
631 case LK_SHARED:
632 ap->a_flags |= LK_SHARED;
633 break;
634 case LK_UPGRADE:
635 case LK_EXCLUSIVE:
636 ap->a_flags |= LK_EXCLUSIVE;
637 break;
638 default:
639 panic("Unsupported lock request %d\n",
640 ap->a_flags);
641 }
642 VOP_UNLOCK(lvp, 0);
643 error = vop_stdlock(ap);
644 }
645 vdrop(lvp);
646 } else
647 error = vop_stdlock(ap);
648
649 return (error);
650}
651
652/*
653 * We need to process our own vnode unlock and then clear the
654 * interlock flag as it applies only to our vnode, not the
655 * vnodes below us on the stack.
656 */
657static int
658null_unlock(struct vop_unlock_args *ap)
659{
660 struct vnode *vp = ap->a_vp;
661 int flags = ap->a_flags;
662 int mtxlkflag = 0;
663 struct null_node *nn;
664 struct vnode *lvp;
665 int error;
666
667 if ((flags & LK_INTERLOCK) != 0)
668 mtxlkflag = 1;
669 else if (mtx_owned(VI_MTX(vp)) == 0) {
670 VI_LOCK(vp);
671 mtxlkflag = 2;
672 }
673 nn = VTONULL(vp);
674 if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
675 VI_LOCK_FLAGS(lvp, MTX_DUPOK);
676 flags |= LK_INTERLOCK;
677 vholdl(lvp);
678 VI_UNLOCK(vp);
679 error = VOP_UNLOCK(lvp, flags);
680 vdrop(lvp);
681 if (mtxlkflag == 0)
682 VI_LOCK(vp);
683 } else {
684 if (mtxlkflag == 2)
685 VI_UNLOCK(vp);
686 error = vop_stdunlock(ap);
687 }
688
689 return (error);
690}
691
692/*
693 * Do not allow the VOP_INACTIVE to be passed to the lower layer,
694 * since the reference count on the lower vnode is not related to
695 * ours.
696 */
697static int
698null_inactive(struct vop_inactive_args *ap __unused)
699{
700 struct vnode *vp, *lvp;
701 struct null_node *xp;
702 struct mount *mp;
703 struct null_mount *xmp;
704
705 vp = ap->a_vp;
706 xp = VTONULL(vp);
707 lvp = NULLVPTOLOWERVP(vp);
708 mp = vp->v_mount;
709 xmp = MOUNTTONULLMOUNT(mp);
710 if ((xmp->nullm_flags & NULLM_CACHE) == 0 ||
711 (xp->null_flags & NULLV_DROP) != 0 ||
712 (lvp->v_vflag & VV_NOSYNC) != 0) {
713 /*
714 * If this is the last reference and caching of the
715 * nullfs vnodes is not enabled, or the lower vnode is
716 * deleted, then free up the vnode so as not to tie up
717 * the lower vnodes.
718 */
719 vp->v_object = NULL;
720 vrecycle(vp);
721 }
722 return (0);
723}
724
725/*
726 * Now, the nullfs vnode and, due to the sharing lock, the lower
727 * vnode, are exclusively locked, and we shall destroy the null vnode.
728 */
729static int
730null_reclaim(struct vop_reclaim_args *ap)
731{
732 struct vnode *vp;
733 struct null_node *xp;
734 struct vnode *lowervp;
735
736 vp = ap->a_vp;
737 xp = VTONULL(vp);
738 lowervp = xp->null_lowervp;
739
740 KASSERT(lowervp != NULL && vp->v_vnlock != &vp->v_lock,
741 ("Reclaiming inclomplete null vnode %p", vp));
742
743 null_hashrem(xp);
744 /*
745 * Use the interlock to protect the clearing of v_data to
746 * prevent faults in null_lock().
747 */
748 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
749 VI_LOCK(vp);
750 vp->v_data = NULL;
751 vp->v_object = NULL;
752 vp->v_vnlock = &vp->v_lock;
753 VI_UNLOCK(vp);
754
755 /*
756 * If we were opened for write, we leased one write reference
757 * to the lower vnode. If this is a reclamation due to the
758 * forced unmount, undo the reference now.
759 */
760 if (vp->v_writecount > 0)
761 VOP_ADD_WRITECOUNT(lowervp, -1);
762 if ((xp->null_flags & NULLV_NOUNLOCK) != 0)
763 vunref(lowervp);
764 else
765 vput(lowervp);
766 free(xp, M_NULLFSNODE);
767
768 return (0);
769}
770
771static int
772null_print(struct vop_print_args *ap)
773{
774 struct vnode *vp = ap->a_vp;
775
776 printf("\tvp=%p, lowervp=%p\n", vp, VTONULL(vp)->null_lowervp);
777 return (0);
778}
779
780/* ARGSUSED */
781static int
782null_getwritemount(struct vop_getwritemount_args *ap)
783{
784 struct null_node *xp;
785 struct vnode *lowervp;
786 struct vnode *vp;
787
788 vp = ap->a_vp;
789 VI_LOCK(vp);
790 xp = VTONULL(vp);
791 if (xp && (lowervp = xp->null_lowervp)) {
792 VI_LOCK_FLAGS(lowervp, MTX_DUPOK);
793 VI_UNLOCK(vp);
794 vholdl(lowervp);
795 VI_UNLOCK(lowervp);
796 VOP_GETWRITEMOUNT(lowervp, ap->a_mpp);
797 vdrop(lowervp);
798 } else {
799 VI_UNLOCK(vp);
800 *(ap->a_mpp) = NULL;
801 }
802 return (0);
803}
804
805static int
806null_vptofh(struct vop_vptofh_args *ap)
807{
808 struct vnode *lvp;
809
810 lvp = NULLVPTOLOWERVP(ap->a_vp);
811 return VOP_VPTOFH(lvp, ap->a_fhp);
812}
813
814static int
815null_vptocnp(struct vop_vptocnp_args *ap)
816{
817 struct vnode *vp = ap->a_vp;
818 struct vnode **dvp = ap->a_vpp;
819 struct vnode *lvp, *ldvp;
820 struct ucred *cred = ap->a_cred;
821 int error, locked;
822
823 if (vp->v_type == VDIR)
824 return (vop_stdvptocnp(ap));
825
826 locked = VOP_ISLOCKED(vp);
827 lvp = NULLVPTOLOWERVP(vp);
828 vhold(lvp);
829 VOP_UNLOCK(vp, 0); /* vp is held by vn_vptocnp_locked that called us */
830 ldvp = lvp;
831 vref(lvp);
832 error = vn_vptocnp(&ldvp, cred, ap->a_buf, ap->a_buflen);
833 vdrop(lvp);
834 if (error != 0) {
835 vn_lock(vp, locked | LK_RETRY);
836 return (ENOENT);
837 }
838
839 /*
840 * Exclusive lock is required by insmntque1 call in
841 * null_nodeget()
842 */
843 error = vn_lock(ldvp, LK_EXCLUSIVE);
844 if (error != 0) {
845 vrele(ldvp);
846 vn_lock(vp, locked | LK_RETRY);
847 return (ENOENT);
848 }
849 vref(ldvp);
850 error = null_nodeget(vp->v_mount, ldvp, dvp);
851 if (error == 0) {
852#ifdef DIAGNOSTIC
853 NULLVPTOLOWERVP(*dvp);
854#endif
855 VOP_UNLOCK(*dvp, 0); /* keep reference on *dvp */
856 }
857 vn_lock(vp, locked | LK_RETRY);
858 return (error);
859}
860
861/*
862 * Global vfs data structures
863 */
864struct vop_vector null_vnodeops = {
865 .vop_bypass = null_bypass,
866 .vop_access = null_access,
867 .vop_accessx = null_accessx,
868 .vop_advlockpurge = vop_stdadvlockpurge,
869 .vop_bmap = VOP_EOPNOTSUPP,
870 .vop_getattr = null_getattr,
871 .vop_getwritemount = null_getwritemount,
872 .vop_inactive = null_inactive,
873 .vop_islocked = vop_stdislocked,
874 .vop_lock1 = null_lock,
875 .vop_lookup = null_lookup,
876 .vop_open = null_open,
877 .vop_print = null_print,
878 .vop_reclaim = null_reclaim,
879 .vop_remove = null_remove,
880 .vop_rename = null_rename,
881 .vop_setattr = null_setattr,
882 .vop_strategy = VOP_EOPNOTSUPP,
883 .vop_unlock = null_unlock,
884 .vop_vptocnp = null_vptocnp,
885 .vop_vptofh = null_vptofh,
886 .vop_add_writecount = null_add_writecount,
887};
2 * Copyright (c) 1992, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * John Heidemann of the UCLA Ficus project.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
19 *
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
30 * SUCH DAMAGE.
31 *
32 * @(#)null_vnops.c 8.6 (Berkeley) 5/27/95
33 *
34 * Ancestors:
35 * @(#)lofs_vnops.c 1.2 (Berkeley) 6/18/92
36 * ...and...
37 * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project
38 *
39 * $FreeBSD: head/sys/fs/nullfs/null_vnops.c 252714 2013-07-04 19:01:18Z kib $
40 */
41
42/*
43 * Null Layer
44 *
45 * (See mount_nullfs(8) for more information.)
46 *
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.
53 *
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
60 * with a null layer.
61 *
62 * The remainder of this man page examines the null layer as a basis
63 * for constructing new layers.
64 *
65 *
66 * INSTANTIATING NEW NULL LAYERS
67 *
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.
74 *
75 *
76 * OPERATION OF A NULL LAYER
77 *
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
82 * pass.
83 *
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.
91 *
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.
107 *
108 *
109 * INSTANTIATING VNODE STACKS
110 *
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.
114 *
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.
119 *
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.
124 *
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.
136 *
137 *
138 * CREATING OTHER FILE SYSTEM LAYERS
139 *
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
143 * all variables.
144 *
145 * The umap layer is an example of a layer descended from the
146 * null layer.
147 *
148 *
149 * INVOKING OPERATIONS ON LOWER LAYERS
150 *
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.
157 *
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.
163 *
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.
169 *
170 */
171
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
184#include <fs/nullfs/null.h>
185
186#include <vm/vm.h>
187#include <vm/vm_extern.h>
188#include <vm/vm_object.h>
189#include <vm/vnode_pager.h>
190
191static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */
192SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW,
193 &null_bug_bypass, 0, "");
194
195/*
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.
200 *
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.
204 *
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.
210 *
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?)
218 */
219int
220null_bypass(struct vop_generic_args *ap)
221{
222 struct vnode **this_vp_p;
223 int error;
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;
228 int reles, i;
229
230 if (null_bug_bypass)
231 printf ("null_bypass: %s\n", descp->vdesc_name);
232
233#ifdef DIAGNOSTIC
234 /*
235 * We require at least one vp.
236 */
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");
240#endif
241
242 /*
243 * Map the vnodes going in.
244 * Later, we'll invoke the operation based on
245 * the first mapped vnode's operation vector.
246 */
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);
253 /*
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.)
257 */
258 if (i && (*this_vp_p == NULLVP ||
259 (*this_vp_p)->v_op != &null_vnodeops)) {
260 old_vps[i] = NULLVP;
261 } else {
262 old_vps[i] = *this_vp_p;
263 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
264 /*
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.)
268 */
269 if (reles & VDESC_VP0_WILLRELE)
270 VREF(*this_vp_p);
271 }
272
273 }
274
275 /*
276 * Call the operation on the lower layer
277 * with the modified argument structure.
278 */
279 if (vps_p[0] && *vps_p[0])
280 error = VCALL(ap);
281 else {
282 printf("null_bypass: no map for %s\n", descp->vdesc_name);
283 error = EINVAL;
284 }
285
286 /*
287 * Maintain the illusion of call-by-value
288 * by restoring vnodes in the argument structure
289 * to their original value.
290 */
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 */
295 if (old_vps[i]) {
296 *(vps_p[i]) = old_vps[i];
297#if 0
298 if (reles & VDESC_VP0_WILLUNLOCK)
299 VOP_UNLOCK(*(vps_p[i]), 0);
300#endif
301 if (reles & VDESC_VP0_WILLRELE)
302 vrele(*(vps_p[i]));
303 }
304 }
305
306 /*
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.)
310 */
311 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
312 !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
313 !error) {
314 /*
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.)
319 */
320 if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
321 goto out;
322 vppp = VOPARG_OFFSETTO(struct vnode***,
323 descp->vdesc_vpp_offset,ap);
324 if (*vppp)
325 error = null_nodeget(old_vps[0]->v_mount, **vppp, *vppp);
326 }
327
328 out:
329 return (error);
330}
331
332static int
333null_add_writecount(struct vop_add_writecount_args *ap)
334{
335 struct vnode *lvp, *vp;
336 int error;
337
338 vp = ap->a_vp;
339 lvp = NULLVPTOLOWERVP(vp);
340 KASSERT(vp->v_writecount + ap->a_inc >= 0, ("wrong writecount inc"));
341 if (vp->v_writecount > 0 && vp->v_writecount + ap->a_inc == 0)
342 error = VOP_ADD_WRITECOUNT(lvp, -1);
343 else if (vp->v_writecount == 0 && vp->v_writecount + ap->a_inc > 0)
344 error = VOP_ADD_WRITECOUNT(lvp, 1);
345 else
346 error = 0;
347 if (error == 0)
348 vp->v_writecount += ap->a_inc;
349 return (error);
350}
351
352/*
353 * We have to carry on the locking protocol on the null layer vnodes
354 * as we progress through the tree. We also have to enforce read-only
355 * if this layer is mounted read-only.
356 */
357static int
358null_lookup(struct vop_lookup_args *ap)
359{
360 struct componentname *cnp = ap->a_cnp;
361 struct vnode *dvp = ap->a_dvp;
362 int flags = cnp->cn_flags;
363 struct vnode *vp, *ldvp, *lvp;
364 int error;
365
366 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
367 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
368 return (EROFS);
369 /*
370 * Although it is possible to call null_bypass(), we'll do
371 * a direct call to reduce overhead
372 */
373 ldvp = NULLVPTOLOWERVP(dvp);
374 vp = lvp = NULL;
375 error = VOP_LOOKUP(ldvp, &lvp, cnp);
376 if (error == EJUSTRETURN && (flags & ISLASTCN) &&
377 (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
378 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
379 error = EROFS;
380
381 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
382 if (ldvp == lvp) {
383 *ap->a_vpp = dvp;
384 VREF(dvp);
385 vrele(lvp);
386 } else {
387 error = null_nodeget(dvp->v_mount, lvp, &vp);
388 if (error == 0)
389 *ap->a_vpp = vp;
390 }
391 }
392 return (error);
393}
394
395static int
396null_open(struct vop_open_args *ap)
397{
398 int retval;
399 struct vnode *vp, *ldvp;
400
401 vp = ap->a_vp;
402 ldvp = NULLVPTOLOWERVP(vp);
403 retval = null_bypass(&ap->a_gen);
404 if (retval == 0)
405 vp->v_object = ldvp->v_object;
406 return (retval);
407}
408
409/*
410 * Setattr call. Disallow write attempts if the layer is mounted read-only.
411 */
412static int
413null_setattr(struct vop_setattr_args *ap)
414{
415 struct vnode *vp = ap->a_vp;
416 struct vattr *vap = ap->a_vap;
417
418 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
419 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
420 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
421 (vp->v_mount->mnt_flag & MNT_RDONLY))
422 return (EROFS);
423 if (vap->va_size != VNOVAL) {
424 switch (vp->v_type) {
425 case VDIR:
426 return (EISDIR);
427 case VCHR:
428 case VBLK:
429 case VSOCK:
430 case VFIFO:
431 if (vap->va_flags != VNOVAL)
432 return (EOPNOTSUPP);
433 return (0);
434 case VREG:
435 case VLNK:
436 default:
437 /*
438 * Disallow write attempts if the filesystem is
439 * mounted read-only.
440 */
441 if (vp->v_mount->mnt_flag & MNT_RDONLY)
442 return (EROFS);
443 }
444 }
445
446 return (null_bypass((struct vop_generic_args *)ap));
447}
448
449/*
450 * We handle getattr only to change the fsid.
451 */
452static int
453null_getattr(struct vop_getattr_args *ap)
454{
455 int error;
456
457 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0)
458 return (error);
459
460 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
461 return (0);
462}
463
464/*
465 * Handle to disallow write access if mounted read-only.
466 */
467static int
468null_access(struct vop_access_args *ap)
469{
470 struct vnode *vp = ap->a_vp;
471 accmode_t accmode = ap->a_accmode;
472
473 /*
474 * Disallow write attempts on read-only layers;
475 * unless the file is a socket, fifo, or a block or
476 * character device resident on the filesystem.
477 */
478 if (accmode & VWRITE) {
479 switch (vp->v_type) {
480 case VDIR:
481 case VLNK:
482 case VREG:
483 if (vp->v_mount->mnt_flag & MNT_RDONLY)
484 return (EROFS);
485 break;
486 default:
487 break;
488 }
489 }
490 return (null_bypass((struct vop_generic_args *)ap));
491}
492
493static int
494null_accessx(struct vop_accessx_args *ap)
495{
496 struct vnode *vp = ap->a_vp;
497 accmode_t accmode = ap->a_accmode;
498
499 /*
500 * Disallow write attempts on read-only layers;
501 * unless the file is a socket, fifo, or a block or
502 * character device resident on the filesystem.
503 */
504 if (accmode & VWRITE) {
505 switch (vp->v_type) {
506 case VDIR:
507 case VLNK:
508 case VREG:
509 if (vp->v_mount->mnt_flag & MNT_RDONLY)
510 return (EROFS);
511 break;
512 default:
513 break;
514 }
515 }
516 return (null_bypass((struct vop_generic_args *)ap));
517}
518
519/*
520 * Increasing refcount of lower vnode is needed at least for the case
521 * when lower FS is NFS to do sillyrename if the file is in use.
522 * Unfortunately v_usecount is incremented in many places in
523 * the kernel and, as such, there may be races that result in
524 * the NFS client doing an extraneous silly rename, but that seems
525 * preferable to not doing a silly rename when it is needed.
526 */
527static int
528null_remove(struct vop_remove_args *ap)
529{
530 int retval, vreleit;
531 struct vnode *lvp;
532
533 if (vrefcnt(ap->a_vp) > 1) {
534 lvp = NULLVPTOLOWERVP(ap->a_vp);
535 VREF(lvp);
536 vreleit = 1;
537 } else
538 vreleit = 0;
539 retval = null_bypass(&ap->a_gen);
540 if (vreleit != 0)
541 vrele(lvp);
542 return (retval);
543}
544
545/*
546 * We handle this to eliminate null FS to lower FS
547 * file moving. Don't know why we don't allow this,
548 * possibly we should.
549 */
550static int
551null_rename(struct vop_rename_args *ap)
552{
553 struct vnode *tdvp = ap->a_tdvp;
554 struct vnode *fvp = ap->a_fvp;
555 struct vnode *fdvp = ap->a_fdvp;
556 struct vnode *tvp = ap->a_tvp;
557 struct null_node *tnn;
558
559 /* Check for cross-device rename. */
560 if ((fvp->v_mount != tdvp->v_mount) ||
561 (tvp && (fvp->v_mount != tvp->v_mount))) {
562 if (tdvp == tvp)
563 vrele(tdvp);
564 else
565 vput(tdvp);
566 if (tvp)
567 vput(tvp);
568 vrele(fdvp);
569 vrele(fvp);
570 return (EXDEV);
571 }
572
573 if (tvp != NULL) {
574 tnn = VTONULL(tvp);
575 tnn->null_flags |= NULLV_DROP;
576 }
577 return (null_bypass((struct vop_generic_args *)ap));
578}
579
580/*
581 * We need to process our own vnode lock and then clear the
582 * interlock flag as it applies only to our vnode, not the
583 * vnodes below us on the stack.
584 */
585static int
586null_lock(struct vop_lock1_args *ap)
587{
588 struct vnode *vp = ap->a_vp;
589 int flags = ap->a_flags;
590 struct null_node *nn;
591 struct vnode *lvp;
592 int error;
593
594
595 if ((flags & LK_INTERLOCK) == 0) {
596 VI_LOCK(vp);
597 ap->a_flags = flags |= LK_INTERLOCK;
598 }
599 nn = VTONULL(vp);
600 /*
601 * If we're still active we must ask the lower layer to
602 * lock as ffs has special lock considerations in it's
603 * vop lock.
604 */
605 if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
606 VI_LOCK_FLAGS(lvp, MTX_DUPOK);
607 VI_UNLOCK(vp);
608 /*
609 * We have to hold the vnode here to solve a potential
610 * reclaim race. If we're forcibly vgone'd while we
611 * still have refs, a thread could be sleeping inside
612 * the lowervp's vop_lock routine. When we vgone we will
613 * drop our last ref to the lowervp, which would allow it
614 * to be reclaimed. The lowervp could then be recycled,
615 * in which case it is not legal to be sleeping in it's VOP.
616 * We prevent it from being recycled by holding the vnode
617 * here.
618 */
619 vholdl(lvp);
620 error = VOP_LOCK(lvp, flags);
621
622 /*
623 * We might have slept to get the lock and someone might have
624 * clean our vnode already, switching vnode lock from one in
625 * lowervp to v_lock in our own vnode structure. Handle this
626 * case by reacquiring correct lock in requested mode.
627 */
628 if (VTONULL(vp) == NULL && error == 0) {
629 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
630 switch (flags & LK_TYPE_MASK) {
631 case LK_SHARED:
632 ap->a_flags |= LK_SHARED;
633 break;
634 case LK_UPGRADE:
635 case LK_EXCLUSIVE:
636 ap->a_flags |= LK_EXCLUSIVE;
637 break;
638 default:
639 panic("Unsupported lock request %d\n",
640 ap->a_flags);
641 }
642 VOP_UNLOCK(lvp, 0);
643 error = vop_stdlock(ap);
644 }
645 vdrop(lvp);
646 } else
647 error = vop_stdlock(ap);
648
649 return (error);
650}
651
652/*
653 * We need to process our own vnode unlock and then clear the
654 * interlock flag as it applies only to our vnode, not the
655 * vnodes below us on the stack.
656 */
657static int
658null_unlock(struct vop_unlock_args *ap)
659{
660 struct vnode *vp = ap->a_vp;
661 int flags = ap->a_flags;
662 int mtxlkflag = 0;
663 struct null_node *nn;
664 struct vnode *lvp;
665 int error;
666
667 if ((flags & LK_INTERLOCK) != 0)
668 mtxlkflag = 1;
669 else if (mtx_owned(VI_MTX(vp)) == 0) {
670 VI_LOCK(vp);
671 mtxlkflag = 2;
672 }
673 nn = VTONULL(vp);
674 if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
675 VI_LOCK_FLAGS(lvp, MTX_DUPOK);
676 flags |= LK_INTERLOCK;
677 vholdl(lvp);
678 VI_UNLOCK(vp);
679 error = VOP_UNLOCK(lvp, flags);
680 vdrop(lvp);
681 if (mtxlkflag == 0)
682 VI_LOCK(vp);
683 } else {
684 if (mtxlkflag == 2)
685 VI_UNLOCK(vp);
686 error = vop_stdunlock(ap);
687 }
688
689 return (error);
690}
691
692/*
693 * Do not allow the VOP_INACTIVE to be passed to the lower layer,
694 * since the reference count on the lower vnode is not related to
695 * ours.
696 */
697static int
698null_inactive(struct vop_inactive_args *ap __unused)
699{
700 struct vnode *vp, *lvp;
701 struct null_node *xp;
702 struct mount *mp;
703 struct null_mount *xmp;
704
705 vp = ap->a_vp;
706 xp = VTONULL(vp);
707 lvp = NULLVPTOLOWERVP(vp);
708 mp = vp->v_mount;
709 xmp = MOUNTTONULLMOUNT(mp);
710 if ((xmp->nullm_flags & NULLM_CACHE) == 0 ||
711 (xp->null_flags & NULLV_DROP) != 0 ||
712 (lvp->v_vflag & VV_NOSYNC) != 0) {
713 /*
714 * If this is the last reference and caching of the
715 * nullfs vnodes is not enabled, or the lower vnode is
716 * deleted, then free up the vnode so as not to tie up
717 * the lower vnodes.
718 */
719 vp->v_object = NULL;
720 vrecycle(vp);
721 }
722 return (0);
723}
724
725/*
726 * Now, the nullfs vnode and, due to the sharing lock, the lower
727 * vnode, are exclusively locked, and we shall destroy the null vnode.
728 */
729static int
730null_reclaim(struct vop_reclaim_args *ap)
731{
732 struct vnode *vp;
733 struct null_node *xp;
734 struct vnode *lowervp;
735
736 vp = ap->a_vp;
737 xp = VTONULL(vp);
738 lowervp = xp->null_lowervp;
739
740 KASSERT(lowervp != NULL && vp->v_vnlock != &vp->v_lock,
741 ("Reclaiming inclomplete null vnode %p", vp));
742
743 null_hashrem(xp);
744 /*
745 * Use the interlock to protect the clearing of v_data to
746 * prevent faults in null_lock().
747 */
748 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
749 VI_LOCK(vp);
750 vp->v_data = NULL;
751 vp->v_object = NULL;
752 vp->v_vnlock = &vp->v_lock;
753 VI_UNLOCK(vp);
754
755 /*
756 * If we were opened for write, we leased one write reference
757 * to the lower vnode. If this is a reclamation due to the
758 * forced unmount, undo the reference now.
759 */
760 if (vp->v_writecount > 0)
761 VOP_ADD_WRITECOUNT(lowervp, -1);
762 if ((xp->null_flags & NULLV_NOUNLOCK) != 0)
763 vunref(lowervp);
764 else
765 vput(lowervp);
766 free(xp, M_NULLFSNODE);
767
768 return (0);
769}
770
771static int
772null_print(struct vop_print_args *ap)
773{
774 struct vnode *vp = ap->a_vp;
775
776 printf("\tvp=%p, lowervp=%p\n", vp, VTONULL(vp)->null_lowervp);
777 return (0);
778}
779
780/* ARGSUSED */
781static int
782null_getwritemount(struct vop_getwritemount_args *ap)
783{
784 struct null_node *xp;
785 struct vnode *lowervp;
786 struct vnode *vp;
787
788 vp = ap->a_vp;
789 VI_LOCK(vp);
790 xp = VTONULL(vp);
791 if (xp && (lowervp = xp->null_lowervp)) {
792 VI_LOCK_FLAGS(lowervp, MTX_DUPOK);
793 VI_UNLOCK(vp);
794 vholdl(lowervp);
795 VI_UNLOCK(lowervp);
796 VOP_GETWRITEMOUNT(lowervp, ap->a_mpp);
797 vdrop(lowervp);
798 } else {
799 VI_UNLOCK(vp);
800 *(ap->a_mpp) = NULL;
801 }
802 return (0);
803}
804
805static int
806null_vptofh(struct vop_vptofh_args *ap)
807{
808 struct vnode *lvp;
809
810 lvp = NULLVPTOLOWERVP(ap->a_vp);
811 return VOP_VPTOFH(lvp, ap->a_fhp);
812}
813
814static int
815null_vptocnp(struct vop_vptocnp_args *ap)
816{
817 struct vnode *vp = ap->a_vp;
818 struct vnode **dvp = ap->a_vpp;
819 struct vnode *lvp, *ldvp;
820 struct ucred *cred = ap->a_cred;
821 int error, locked;
822
823 if (vp->v_type == VDIR)
824 return (vop_stdvptocnp(ap));
825
826 locked = VOP_ISLOCKED(vp);
827 lvp = NULLVPTOLOWERVP(vp);
828 vhold(lvp);
829 VOP_UNLOCK(vp, 0); /* vp is held by vn_vptocnp_locked that called us */
830 ldvp = lvp;
831 vref(lvp);
832 error = vn_vptocnp(&ldvp, cred, ap->a_buf, ap->a_buflen);
833 vdrop(lvp);
834 if (error != 0) {
835 vn_lock(vp, locked | LK_RETRY);
836 return (ENOENT);
837 }
838
839 /*
840 * Exclusive lock is required by insmntque1 call in
841 * null_nodeget()
842 */
843 error = vn_lock(ldvp, LK_EXCLUSIVE);
844 if (error != 0) {
845 vrele(ldvp);
846 vn_lock(vp, locked | LK_RETRY);
847 return (ENOENT);
848 }
849 vref(ldvp);
850 error = null_nodeget(vp->v_mount, ldvp, dvp);
851 if (error == 0) {
852#ifdef DIAGNOSTIC
853 NULLVPTOLOWERVP(*dvp);
854#endif
855 VOP_UNLOCK(*dvp, 0); /* keep reference on *dvp */
856 }
857 vn_lock(vp, locked | LK_RETRY);
858 return (error);
859}
860
861/*
862 * Global vfs data structures
863 */
864struct vop_vector null_vnodeops = {
865 .vop_bypass = null_bypass,
866 .vop_access = null_access,
867 .vop_accessx = null_accessx,
868 .vop_advlockpurge = vop_stdadvlockpurge,
869 .vop_bmap = VOP_EOPNOTSUPP,
870 .vop_getattr = null_getattr,
871 .vop_getwritemount = null_getwritemount,
872 .vop_inactive = null_inactive,
873 .vop_islocked = vop_stdislocked,
874 .vop_lock1 = null_lock,
875 .vop_lookup = null_lookup,
876 .vop_open = null_open,
877 .vop_print = null_print,
878 .vop_reclaim = null_reclaim,
879 .vop_remove = null_remove,
880 .vop_rename = null_rename,
881 .vop_setattr = null_setattr,
882 .vop_strategy = VOP_EOPNOTSUPP,
883 .vop_unlock = null_unlock,
884 .vop_vptocnp = null_vptocnp,
885 .vop_vptofh = null_vptofh,
886 .vop_add_writecount = null_add_writecount,
887};