tmpfs_vnops.c revision 230180
1/*	$NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $	*/
2
3/*-
4 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9 * 2005 program.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 *    notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 *    notice, this list of conditions and the following disclaimer in the
18 *    documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
31 */
32
33/*
34 * tmpfs vnode interface.
35 */
36#include <sys/cdefs.h>
37__FBSDID("$FreeBSD: head/sys/fs/tmpfs/tmpfs_vnops.c 230180 2012-01-16 00:26:49Z alc $");
38
39#include <sys/param.h>
40#include <sys/fcntl.h>
41#include <sys/lockf.h>
42#include <sys/namei.h>
43#include <sys/priv.h>
44#include <sys/proc.h>
45#include <sys/sched.h>
46#include <sys/sf_buf.h>
47#include <sys/stat.h>
48#include <sys/systm.h>
49#include <sys/unistd.h>
50#include <sys/vnode.h>
51
52#include <vm/vm.h>
53#include <vm/vm_object.h>
54#include <vm/vm_page.h>
55#include <vm/vm_pager.h>
56
57#include <machine/_inttypes.h>
58
59#include <fs/fifofs/fifo.h>
60#include <fs/tmpfs/tmpfs_vnops.h>
61#include <fs/tmpfs/tmpfs.h>
62
63/* --------------------------------------------------------------------- */
64
65static int
66tmpfs_lookup(struct vop_cachedlookup_args *v)
67{
68	struct vnode *dvp = v->a_dvp;
69	struct vnode **vpp = v->a_vpp;
70	struct componentname *cnp = v->a_cnp;
71
72	int error;
73	struct tmpfs_dirent *de;
74	struct tmpfs_node *dnode;
75
76	dnode = VP_TO_TMPFS_DIR(dvp);
77	*vpp = NULLVP;
78
79	/* Check accessibility of requested node as a first step. */
80	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
81	if (error != 0)
82		goto out;
83
84	/* We cannot be requesting the parent directory of the root node. */
85	MPASS(IMPLIES(dnode->tn_type == VDIR &&
86	    dnode->tn_dir.tn_parent == dnode,
87	    !(cnp->cn_flags & ISDOTDOT)));
88
89	TMPFS_ASSERT_LOCKED(dnode);
90	if (dnode->tn_dir.tn_parent == NULL) {
91		error = ENOENT;
92		goto out;
93	}
94	if (cnp->cn_flags & ISDOTDOT) {
95		int ltype = 0;
96
97		ltype = VOP_ISLOCKED(dvp);
98		vhold(dvp);
99		VOP_UNLOCK(dvp, 0);
100		/* Allocate a new vnode on the matching entry. */
101		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
102		    cnp->cn_lkflags, vpp);
103
104		vn_lock(dvp, ltype | LK_RETRY);
105		vdrop(dvp);
106	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
107		VREF(dvp);
108		*vpp = dvp;
109		error = 0;
110	} else {
111		de = tmpfs_dir_lookup(dnode, NULL, cnp);
112		if (de != NULL && de->td_node == NULL)
113			cnp->cn_flags |= ISWHITEOUT;
114		if (de == NULL || de->td_node == NULL) {
115			/* The entry was not found in the directory.
116			 * This is OK if we are creating or renaming an
117			 * entry and are working on the last component of
118			 * the path name. */
119			if ((cnp->cn_flags & ISLASTCN) &&
120			    (cnp->cn_nameiop == CREATE || \
121			    cnp->cn_nameiop == RENAME ||
122			    (cnp->cn_nameiop == DELETE &&
123			    cnp->cn_flags & DOWHITEOUT &&
124			    cnp->cn_flags & ISWHITEOUT))) {
125				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
126				    cnp->cn_thread);
127				if (error != 0)
128					goto out;
129
130				/* Keep the component name in the buffer for
131				 * future uses. */
132				cnp->cn_flags |= SAVENAME;
133
134				error = EJUSTRETURN;
135			} else
136				error = ENOENT;
137		} else {
138			struct tmpfs_node *tnode;
139
140			/* The entry was found, so get its associated
141			 * tmpfs_node. */
142			tnode = de->td_node;
143
144			/* If we are not at the last path component and
145			 * found a non-directory or non-link entry (which
146			 * may itself be pointing to a directory), raise
147			 * an error. */
148			if ((tnode->tn_type != VDIR &&
149			    tnode->tn_type != VLNK) &&
150			    !(cnp->cn_flags & ISLASTCN)) {
151				error = ENOTDIR;
152				goto out;
153			}
154
155			/* If we are deleting or renaming the entry, keep
156			 * track of its tmpfs_dirent so that it can be
157			 * easily deleted later. */
158			if ((cnp->cn_flags & ISLASTCN) &&
159			    (cnp->cn_nameiop == DELETE ||
160			    cnp->cn_nameiop == RENAME)) {
161				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
162				    cnp->cn_thread);
163				if (error != 0)
164					goto out;
165
166				/* Allocate a new vnode on the matching entry. */
167				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
168						cnp->cn_lkflags, vpp);
169				if (error != 0)
170					goto out;
171
172				if ((dnode->tn_mode & S_ISTXT) &&
173				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) &&
174				  VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) {
175					error = EPERM;
176					vput(*vpp);
177					*vpp = NULL;
178					goto out;
179				}
180				cnp->cn_flags |= SAVENAME;
181			} else {
182				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
183						cnp->cn_lkflags, vpp);
184			}
185		}
186	}
187
188	/* Store the result of this lookup in the cache.  Avoid this if the
189	 * request was for creation, as it does not improve timings on
190	 * emprical tests. */
191	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
192		cache_enter(dvp, *vpp, cnp);
193
194out:
195	/* If there were no errors, *vpp cannot be null and it must be
196	 * locked. */
197	MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
198
199	return error;
200}
201
202/* --------------------------------------------------------------------- */
203
204static int
205tmpfs_create(struct vop_create_args *v)
206{
207	struct vnode *dvp = v->a_dvp;
208	struct vnode **vpp = v->a_vpp;
209	struct componentname *cnp = v->a_cnp;
210	struct vattr *vap = v->a_vap;
211
212	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
213
214	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
215}
216/* --------------------------------------------------------------------- */
217
218static int
219tmpfs_mknod(struct vop_mknod_args *v)
220{
221	struct vnode *dvp = v->a_dvp;
222	struct vnode **vpp = v->a_vpp;
223	struct componentname *cnp = v->a_cnp;
224	struct vattr *vap = v->a_vap;
225
226	if (vap->va_type != VBLK && vap->va_type != VCHR &&
227	    vap->va_type != VFIFO)
228		return EINVAL;
229
230	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
231}
232
233/* --------------------------------------------------------------------- */
234
235static int
236tmpfs_open(struct vop_open_args *v)
237{
238	struct vnode *vp = v->a_vp;
239	int mode = v->a_mode;
240
241	int error;
242	struct tmpfs_node *node;
243
244	MPASS(VOP_ISLOCKED(vp));
245
246	node = VP_TO_TMPFS_NODE(vp);
247
248	/* The file is still active but all its names have been removed
249	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
250	 * it is about to die. */
251	if (node->tn_links < 1)
252		return (ENOENT);
253
254	/* If the file is marked append-only, deny write requests. */
255	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
256		error = EPERM;
257	else {
258		error = 0;
259		vnode_create_vobject(vp, node->tn_size, v->a_td);
260	}
261
262	MPASS(VOP_ISLOCKED(vp));
263	return error;
264}
265
266/* --------------------------------------------------------------------- */
267
268static int
269tmpfs_close(struct vop_close_args *v)
270{
271	struct vnode *vp = v->a_vp;
272
273	MPASS(VOP_ISLOCKED(vp));
274
275	/* Update node times. */
276	tmpfs_update(vp);
277
278	return (0);
279}
280
281/* --------------------------------------------------------------------- */
282
283int
284tmpfs_access(struct vop_access_args *v)
285{
286	struct vnode *vp = v->a_vp;
287	accmode_t accmode = v->a_accmode;
288	struct ucred *cred = v->a_cred;
289
290	int error;
291	struct tmpfs_node *node;
292
293	MPASS(VOP_ISLOCKED(vp));
294
295	node = VP_TO_TMPFS_NODE(vp);
296
297	switch (vp->v_type) {
298	case VDIR:
299		/* FALLTHROUGH */
300	case VLNK:
301		/* FALLTHROUGH */
302	case VREG:
303		if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
304			error = EROFS;
305			goto out;
306		}
307		break;
308
309	case VBLK:
310		/* FALLTHROUGH */
311	case VCHR:
312		/* FALLTHROUGH */
313	case VSOCK:
314		/* FALLTHROUGH */
315	case VFIFO:
316		break;
317
318	default:
319		error = EINVAL;
320		goto out;
321	}
322
323	if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
324		error = EPERM;
325		goto out;
326	}
327
328	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
329	    node->tn_gid, accmode, cred, NULL);
330
331out:
332	MPASS(VOP_ISLOCKED(vp));
333
334	return error;
335}
336
337/* --------------------------------------------------------------------- */
338
339int
340tmpfs_getattr(struct vop_getattr_args *v)
341{
342	struct vnode *vp = v->a_vp;
343	struct vattr *vap = v->a_vap;
344
345	struct tmpfs_node *node;
346
347	node = VP_TO_TMPFS_NODE(vp);
348
349	tmpfs_update(vp);
350
351	vap->va_type = vp->v_type;
352	vap->va_mode = node->tn_mode;
353	vap->va_nlink = node->tn_links;
354	vap->va_uid = node->tn_uid;
355	vap->va_gid = node->tn_gid;
356	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
357	vap->va_fileid = node->tn_id;
358	vap->va_size = node->tn_size;
359	vap->va_blocksize = PAGE_SIZE;
360	vap->va_atime = node->tn_atime;
361	vap->va_mtime = node->tn_mtime;
362	vap->va_ctime = node->tn_ctime;
363	vap->va_birthtime = node->tn_birthtime;
364	vap->va_gen = node->tn_gen;
365	vap->va_flags = node->tn_flags;
366	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
367		node->tn_rdev : NODEV;
368	vap->va_bytes = round_page(node->tn_size);
369	vap->va_filerev = 0;
370
371	return 0;
372}
373
374/* --------------------------------------------------------------------- */
375
376/* XXX Should this operation be atomic?  I think it should, but code in
377 * XXX other places (e.g., ufs) doesn't seem to be... */
378int
379tmpfs_setattr(struct vop_setattr_args *v)
380{
381	struct vnode *vp = v->a_vp;
382	struct vattr *vap = v->a_vap;
383	struct ucred *cred = v->a_cred;
384	struct thread *td = curthread;
385
386	int error;
387
388	MPASS(VOP_ISLOCKED(vp));
389
390	error = 0;
391
392	/* Abort if any unsettable attribute is given. */
393	if (vap->va_type != VNON ||
394	    vap->va_nlink != VNOVAL ||
395	    vap->va_fsid != VNOVAL ||
396	    vap->va_fileid != VNOVAL ||
397	    vap->va_blocksize != VNOVAL ||
398	    vap->va_gen != VNOVAL ||
399	    vap->va_rdev != VNOVAL ||
400	    vap->va_bytes != VNOVAL)
401		error = EINVAL;
402
403	if (error == 0 && (vap->va_flags != VNOVAL))
404		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
405
406	if (error == 0 && (vap->va_size != VNOVAL))
407		error = tmpfs_chsize(vp, vap->va_size, cred, td);
408
409	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
410		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
411
412	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
413		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
414
415	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
416	    vap->va_atime.tv_nsec != VNOVAL) ||
417	    (vap->va_mtime.tv_sec != VNOVAL &&
418	    vap->va_mtime.tv_nsec != VNOVAL) ||
419	    (vap->va_birthtime.tv_sec != VNOVAL &&
420	    vap->va_birthtime.tv_nsec != VNOVAL)))
421		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
422			&vap->va_birthtime, vap->va_vaflags, cred, td);
423
424	/* Update the node times.  We give preference to the error codes
425	 * generated by this function rather than the ones that may arise
426	 * from tmpfs_update. */
427	tmpfs_update(vp);
428
429	MPASS(VOP_ISLOCKED(vp));
430
431	return error;
432}
433
434/* --------------------------------------------------------------------- */
435static int
436tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
437    vm_offset_t offset, size_t tlen, struct uio *uio)
438{
439	vm_page_t	m;
440	int		error, rv;
441
442	VM_OBJECT_LOCK(tobj);
443	m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
444	    VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
445	if (m->valid != VM_PAGE_BITS_ALL) {
446		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
447			rv = vm_pager_get_pages(tobj, &m, 1, 0);
448			if (rv != VM_PAGER_OK) {
449				vm_page_lock(m);
450				vm_page_free(m);
451				vm_page_unlock(m);
452				VM_OBJECT_UNLOCK(tobj);
453				return (EIO);
454			}
455		} else
456			vm_page_zero_invalid(m, TRUE);
457	}
458	VM_OBJECT_UNLOCK(tobj);
459	error = uiomove_fromphys(&m, offset, tlen, uio);
460	VM_OBJECT_LOCK(tobj);
461	vm_page_lock(m);
462	vm_page_unwire(m, TRUE);
463	vm_page_unlock(m);
464	vm_page_wakeup(m);
465	VM_OBJECT_UNLOCK(tobj);
466
467	return (error);
468}
469
470static __inline int
471tmpfs_nocacheread_buf(vm_object_t tobj, vm_pindex_t idx,
472    vm_offset_t offset, size_t tlen, void *buf)
473{
474	struct uio uio;
475	struct iovec iov;
476
477	uio.uio_iovcnt = 1;
478	uio.uio_iov = &iov;
479	iov.iov_base = buf;
480	iov.iov_len = tlen;
481
482	uio.uio_offset = 0;
483	uio.uio_resid = tlen;
484	uio.uio_rw = UIO_READ;
485	uio.uio_segflg = UIO_SYSSPACE;
486	uio.uio_td = curthread;
487
488	return (tmpfs_nocacheread(tobj, idx, offset, tlen, &uio));
489}
490
491static int
492tmpfs_mappedread(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
493{
494	struct sf_buf	*sf;
495	vm_pindex_t	idx;
496	vm_page_t	m;
497	vm_offset_t	offset;
498	off_t		addr;
499	size_t		tlen;
500	char		*ma;
501	int		error;
502
503	addr = uio->uio_offset;
504	idx = OFF_TO_IDX(addr);
505	offset = addr & PAGE_MASK;
506	tlen = MIN(PAGE_SIZE - offset, len);
507
508	if ((vobj == NULL) ||
509	    (vobj->resident_page_count == 0 && vobj->cache == NULL))
510		goto nocache;
511
512	VM_OBJECT_LOCK(vobj);
513lookupvpg:
514	if (((m = vm_page_lookup(vobj, idx)) != NULL) &&
515	    vm_page_is_valid(m, offset, tlen)) {
516		if ((m->oflags & VPO_BUSY) != 0) {
517			/*
518			 * Reference the page before unlocking and sleeping so
519			 * that the page daemon is less likely to reclaim it.
520			 */
521			vm_page_reference(m);
522			vm_page_sleep(m, "tmfsmr");
523			goto lookupvpg;
524		}
525		vm_page_busy(m);
526		VM_OBJECT_UNLOCK(vobj);
527		error = uiomove_fromphys(&m, offset, tlen, uio);
528		VM_OBJECT_LOCK(vobj);
529		vm_page_wakeup(m);
530		VM_OBJECT_UNLOCK(vobj);
531		return	(error);
532	} else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) {
533		KASSERT(offset == 0,
534		    ("unexpected offset in tmpfs_mappedread for sendfile"));
535		if ((m->oflags & VPO_BUSY) != 0) {
536			/*
537			 * Reference the page before unlocking and sleeping so
538			 * that the page daemon is less likely to reclaim it.
539			 */
540			vm_page_reference(m);
541			vm_page_sleep(m, "tmfsmr");
542			goto lookupvpg;
543		}
544		vm_page_busy(m);
545		VM_OBJECT_UNLOCK(vobj);
546		sched_pin();
547		sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
548		ma = (char *)sf_buf_kva(sf);
549		error = tmpfs_nocacheread_buf(tobj, idx, 0, tlen, ma);
550		if (error == 0) {
551			if (tlen != PAGE_SIZE)
552				bzero(ma + tlen, PAGE_SIZE - tlen);
553			uio->uio_offset += tlen;
554			uio->uio_resid -= tlen;
555		}
556		sf_buf_free(sf);
557		sched_unpin();
558		VM_OBJECT_LOCK(vobj);
559		if (error == 0)
560			m->valid = VM_PAGE_BITS_ALL;
561		vm_page_wakeup(m);
562		VM_OBJECT_UNLOCK(vobj);
563		return	(error);
564	}
565	VM_OBJECT_UNLOCK(vobj);
566nocache:
567	error = tmpfs_nocacheread(tobj, idx, offset, tlen, uio);
568
569	return	(error);
570}
571
572static int
573tmpfs_read(struct vop_read_args *v)
574{
575	struct vnode *vp = v->a_vp;
576	struct uio *uio = v->a_uio;
577
578	struct tmpfs_node *node;
579	vm_object_t uobj;
580	size_t len;
581	int resid;
582
583	int error = 0;
584
585	node = VP_TO_TMPFS_NODE(vp);
586
587	if (vp->v_type != VREG) {
588		error = EISDIR;
589		goto out;
590	}
591
592	if (uio->uio_offset < 0) {
593		error = EINVAL;
594		goto out;
595	}
596
597	node->tn_status |= TMPFS_NODE_ACCESSED;
598
599	uobj = node->tn_reg.tn_aobj;
600	while ((resid = uio->uio_resid) > 0) {
601		error = 0;
602		if (node->tn_size <= uio->uio_offset)
603			break;
604		len = MIN(node->tn_size - uio->uio_offset, resid);
605		if (len == 0)
606			break;
607		error = tmpfs_mappedread(vp->v_object, uobj, len, uio);
608		if ((error != 0) || (resid == uio->uio_resid))
609			break;
610	}
611
612out:
613
614	return error;
615}
616
617/* --------------------------------------------------------------------- */
618
619static int
620tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
621{
622	vm_pindex_t	idx;
623	vm_page_t	vpg, tpg;
624	vm_offset_t	offset;
625	off_t		addr;
626	size_t		tlen;
627	int		error, rv;
628
629	error = 0;
630
631	addr = uio->uio_offset;
632	idx = OFF_TO_IDX(addr);
633	offset = addr & PAGE_MASK;
634	tlen = MIN(PAGE_SIZE - offset, len);
635
636	if ((vobj == NULL) ||
637	    (vobj->resident_page_count == 0 && vobj->cache == NULL)) {
638		vpg = NULL;
639		goto nocache;
640	}
641
642	VM_OBJECT_LOCK(vobj);
643lookupvpg:
644	if (((vpg = vm_page_lookup(vobj, idx)) != NULL) &&
645	    vm_page_is_valid(vpg, offset, tlen)) {
646		if ((vpg->oflags & VPO_BUSY) != 0) {
647			/*
648			 * Reference the page before unlocking and sleeping so
649			 * that the page daemon is less likely to reclaim it.
650			 */
651			vm_page_reference(vpg);
652			vm_page_sleep(vpg, "tmfsmw");
653			goto lookupvpg;
654		}
655		vm_page_busy(vpg);
656		vm_page_undirty(vpg);
657		VM_OBJECT_UNLOCK(vobj);
658		error = uiomove_fromphys(&vpg, offset, tlen, uio);
659	} else {
660		if (__predict_false(vobj->cache != NULL))
661			vm_page_cache_free(vobj, idx, idx + 1);
662		VM_OBJECT_UNLOCK(vobj);
663		vpg = NULL;
664	}
665nocache:
666	VM_OBJECT_LOCK(tobj);
667	tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
668	    VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
669	if (tpg->valid != VM_PAGE_BITS_ALL) {
670		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
671			rv = vm_pager_get_pages(tobj, &tpg, 1, 0);
672			if (rv != VM_PAGER_OK) {
673				vm_page_lock(tpg);
674				vm_page_free(tpg);
675				vm_page_unlock(tpg);
676				error = EIO;
677				goto out;
678			}
679		} else
680			vm_page_zero_invalid(tpg, TRUE);
681	}
682	VM_OBJECT_UNLOCK(tobj);
683	if (vpg == NULL)
684		error = uiomove_fromphys(&tpg, offset, tlen, uio);
685	else {
686		KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid"));
687		pmap_copy_page(vpg, tpg);
688	}
689	VM_OBJECT_LOCK(tobj);
690	if (error == 0) {
691		KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
692		    ("parts of tpg invalid"));
693		vm_page_dirty(tpg);
694	}
695	vm_page_lock(tpg);
696	vm_page_unwire(tpg, TRUE);
697	vm_page_unlock(tpg);
698	vm_page_wakeup(tpg);
699out:
700	VM_OBJECT_UNLOCK(tobj);
701	if (vpg != NULL) {
702		VM_OBJECT_LOCK(vobj);
703		vm_page_wakeup(vpg);
704		VM_OBJECT_UNLOCK(vobj);
705	}
706
707	return	(error);
708}
709
710static int
711tmpfs_write(struct vop_write_args *v)
712{
713	struct vnode *vp = v->a_vp;
714	struct uio *uio = v->a_uio;
715	int ioflag = v->a_ioflag;
716
717	boolean_t extended;
718	int error = 0;
719	off_t oldsize;
720	struct tmpfs_node *node;
721	vm_object_t uobj;
722	size_t len;
723	int resid;
724
725	node = VP_TO_TMPFS_NODE(vp);
726	oldsize = node->tn_size;
727
728	if (uio->uio_offset < 0 || vp->v_type != VREG) {
729		error = EINVAL;
730		goto out;
731	}
732
733	if (uio->uio_resid == 0) {
734		error = 0;
735		goto out;
736	}
737
738	if (ioflag & IO_APPEND)
739		uio->uio_offset = node->tn_size;
740
741	if (uio->uio_offset + uio->uio_resid >
742	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
743		return (EFBIG);
744
745	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
746		return (EFBIG);
747
748	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
749	if (extended) {
750		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid,
751		    FALSE);
752		if (error != 0)
753			goto out;
754	}
755
756	uobj = node->tn_reg.tn_aobj;
757	while ((resid = uio->uio_resid) > 0) {
758		if (node->tn_size <= uio->uio_offset)
759			break;
760		len = MIN(node->tn_size - uio->uio_offset, resid);
761		if (len == 0)
762			break;
763		error = tmpfs_mappedwrite(vp->v_object, uobj, len, uio);
764		if ((error != 0) || (resid == uio->uio_resid))
765			break;
766	}
767
768	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
769	    (extended ? TMPFS_NODE_CHANGED : 0);
770
771	if (node->tn_mode & (S_ISUID | S_ISGID)) {
772		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
773			node->tn_mode &= ~(S_ISUID | S_ISGID);
774	}
775
776	if (error != 0)
777		(void)tmpfs_reg_resize(vp, oldsize, TRUE);
778
779out:
780	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
781	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
782
783	return error;
784}
785
786/* --------------------------------------------------------------------- */
787
788static int
789tmpfs_fsync(struct vop_fsync_args *v)
790{
791	struct vnode *vp = v->a_vp;
792
793	MPASS(VOP_ISLOCKED(vp));
794
795	tmpfs_update(vp);
796
797	return 0;
798}
799
800/* --------------------------------------------------------------------- */
801
802static int
803tmpfs_remove(struct vop_remove_args *v)
804{
805	struct vnode *dvp = v->a_dvp;
806	struct vnode *vp = v->a_vp;
807
808	int error;
809	struct tmpfs_dirent *de;
810	struct tmpfs_mount *tmp;
811	struct tmpfs_node *dnode;
812	struct tmpfs_node *node;
813
814	MPASS(VOP_ISLOCKED(dvp));
815	MPASS(VOP_ISLOCKED(vp));
816
817	if (vp->v_type == VDIR) {
818		error = EISDIR;
819		goto out;
820	}
821
822	dnode = VP_TO_TMPFS_DIR(dvp);
823	node = VP_TO_TMPFS_NODE(vp);
824	tmp = VFS_TO_TMPFS(vp->v_mount);
825	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
826	MPASS(de != NULL);
827
828	/* Files marked as immutable or append-only cannot be deleted. */
829	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
830	    (dnode->tn_flags & APPEND)) {
831		error = EPERM;
832		goto out;
833	}
834
835	/* Remove the entry from the directory; as it is a file, we do not
836	 * have to change the number of hard links of the directory. */
837	tmpfs_dir_detach(dvp, de);
838	if (v->a_cnp->cn_flags & DOWHITEOUT)
839		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
840
841	/* Free the directory entry we just deleted.  Note that the node
842	 * referred by it will not be removed until the vnode is really
843	 * reclaimed. */
844	tmpfs_free_dirent(tmp, de, TRUE);
845
846	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
847	error = 0;
848
849out:
850
851	return error;
852}
853
854/* --------------------------------------------------------------------- */
855
856static int
857tmpfs_link(struct vop_link_args *v)
858{
859	struct vnode *dvp = v->a_tdvp;
860	struct vnode *vp = v->a_vp;
861	struct componentname *cnp = v->a_cnp;
862
863	int error;
864	struct tmpfs_dirent *de;
865	struct tmpfs_node *node;
866
867	MPASS(VOP_ISLOCKED(dvp));
868	MPASS(cnp->cn_flags & HASBUF);
869	MPASS(dvp != vp); /* XXX When can this be false? */
870
871	node = VP_TO_TMPFS_NODE(vp);
872
873	/* XXX: Why aren't the following two tests done by the caller? */
874
875	/* Hard links of directories are forbidden. */
876	if (vp->v_type == VDIR) {
877		error = EPERM;
878		goto out;
879	}
880
881	/* Cannot create cross-device links. */
882	if (dvp->v_mount != vp->v_mount) {
883		error = EXDEV;
884		goto out;
885	}
886
887	/* Ensure that we do not overflow the maximum number of links imposed
888	 * by the system. */
889	MPASS(node->tn_links <= LINK_MAX);
890	if (node->tn_links == LINK_MAX) {
891		error = EMLINK;
892		goto out;
893	}
894
895	/* We cannot create links of files marked immutable or append-only. */
896	if (node->tn_flags & (IMMUTABLE | APPEND)) {
897		error = EPERM;
898		goto out;
899	}
900
901	/* Allocate a new directory entry to represent the node. */
902	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
903	    cnp->cn_nameptr, cnp->cn_namelen, &de);
904	if (error != 0)
905		goto out;
906
907	/* Insert the new directory entry into the appropriate directory. */
908	if (cnp->cn_flags & ISWHITEOUT)
909		tmpfs_dir_whiteout_remove(dvp, cnp);
910	tmpfs_dir_attach(dvp, de);
911
912	/* vp link count has changed, so update node times. */
913	node->tn_status |= TMPFS_NODE_CHANGED;
914	tmpfs_update(vp);
915
916	error = 0;
917
918out:
919	return error;
920}
921
922/* --------------------------------------------------------------------- */
923
924static int
925tmpfs_rename(struct vop_rename_args *v)
926{
927	struct vnode *fdvp = v->a_fdvp;
928	struct vnode *fvp = v->a_fvp;
929	struct componentname *fcnp = v->a_fcnp;
930	struct vnode *tdvp = v->a_tdvp;
931	struct vnode *tvp = v->a_tvp;
932	struct componentname *tcnp = v->a_tcnp;
933
934	char *newname;
935	int error;
936	struct tmpfs_dirent *de;
937	struct tmpfs_mount *tmp;
938	struct tmpfs_node *fdnode;
939	struct tmpfs_node *fnode;
940	struct tmpfs_node *tnode;
941	struct tmpfs_node *tdnode;
942
943	MPASS(VOP_ISLOCKED(tdvp));
944	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
945	MPASS(fcnp->cn_flags & HASBUF);
946	MPASS(tcnp->cn_flags & HASBUF);
947
948  	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
949
950	/* Disallow cross-device renames.
951	 * XXX Why isn't this done by the caller? */
952	if (fvp->v_mount != tdvp->v_mount ||
953	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
954		error = EXDEV;
955		goto out;
956	}
957
958	tmp = VFS_TO_TMPFS(tdvp->v_mount);
959	tdnode = VP_TO_TMPFS_DIR(tdvp);
960
961	/* If source and target are the same file, there is nothing to do. */
962	if (fvp == tvp) {
963		error = 0;
964		goto out;
965	}
966
967	/* If we need to move the directory between entries, lock the
968	 * source so that we can safely operate on it. */
969	if (fdvp != tdvp && fdvp != tvp)
970		vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
971	fdnode = VP_TO_TMPFS_DIR(fdvp);
972	fnode = VP_TO_TMPFS_NODE(fvp);
973	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
974
975	/* Entry can disappear before we lock fdvp,
976	 * also avoid manipulating '.' and '..' entries. */
977	if (de == NULL) {
978		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
979		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
980			error = EINVAL;
981		else
982			error = ENOENT;
983		goto out_locked;
984	}
985	MPASS(de->td_node == fnode);
986
987	/* If re-naming a directory to another preexisting directory
988	 * ensure that the target directory is empty so that its
989	 * removal causes no side effects.
990	 * Kern_rename gurantees the destination to be a directory
991	 * if the source is one. */
992	if (tvp != NULL) {
993		MPASS(tnode != NULL);
994
995		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
996		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
997			error = EPERM;
998			goto out_locked;
999		}
1000
1001		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1002			if (tnode->tn_size > 0) {
1003				error = ENOTEMPTY;
1004				goto out_locked;
1005			}
1006		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1007			error = ENOTDIR;
1008			goto out_locked;
1009		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1010			error = EISDIR;
1011			goto out_locked;
1012		} else {
1013			MPASS(fnode->tn_type != VDIR &&
1014				tnode->tn_type != VDIR);
1015		}
1016	}
1017
1018	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1019	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1020		error = EPERM;
1021		goto out_locked;
1022	}
1023
1024	/* Ensure that we have enough memory to hold the new name, if it
1025	 * has to be changed. */
1026	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1027	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1028		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1029	} else
1030		newname = NULL;
1031
1032	/* If the node is being moved to another directory, we have to do
1033	 * the move. */
1034	if (fdnode != tdnode) {
1035		/* In case we are moving a directory, we have to adjust its
1036		 * parent to point to the new parent. */
1037		if (de->td_node->tn_type == VDIR) {
1038			struct tmpfs_node *n;
1039
1040			/* Ensure the target directory is not a child of the
1041			 * directory being moved.  Otherwise, we'd end up
1042			 * with stale nodes. */
1043			n = tdnode;
1044			/* TMPFS_LOCK garanties that no nodes are freed while
1045			 * traversing the list. Nodes can only be marked as
1046			 * removed: tn_parent == NULL. */
1047			TMPFS_LOCK(tmp);
1048			TMPFS_NODE_LOCK(n);
1049			while (n != n->tn_dir.tn_parent) {
1050				struct tmpfs_node *parent;
1051
1052				if (n == fnode) {
1053					TMPFS_NODE_UNLOCK(n);
1054					TMPFS_UNLOCK(tmp);
1055					error = EINVAL;
1056					if (newname != NULL)
1057						    free(newname, M_TMPFSNAME);
1058					goto out_locked;
1059				}
1060				parent = n->tn_dir.tn_parent;
1061				TMPFS_NODE_UNLOCK(n);
1062				if (parent == NULL) {
1063					n = NULL;
1064					break;
1065				}
1066				TMPFS_NODE_LOCK(parent);
1067				if (parent->tn_dir.tn_parent == NULL) {
1068					TMPFS_NODE_UNLOCK(parent);
1069					n = NULL;
1070					break;
1071				}
1072				n = parent;
1073			}
1074			TMPFS_UNLOCK(tmp);
1075			if (n == NULL) {
1076				error = EINVAL;
1077				if (newname != NULL)
1078					    free(newname, M_TMPFSNAME);
1079				goto out_locked;
1080			}
1081			TMPFS_NODE_UNLOCK(n);
1082
1083			/* Adjust the parent pointer. */
1084			TMPFS_VALIDATE_DIR(fnode);
1085			TMPFS_NODE_LOCK(de->td_node);
1086			de->td_node->tn_dir.tn_parent = tdnode;
1087			TMPFS_NODE_UNLOCK(de->td_node);
1088
1089			/* As a result of changing the target of the '..'
1090			 * entry, the link count of the source and target
1091			 * directories has to be adjusted. */
1092			TMPFS_NODE_LOCK(tdnode);
1093			TMPFS_ASSERT_LOCKED(tdnode);
1094			tdnode->tn_links++;
1095			TMPFS_NODE_UNLOCK(tdnode);
1096
1097			TMPFS_NODE_LOCK(fdnode);
1098			TMPFS_ASSERT_LOCKED(fdnode);
1099			fdnode->tn_links--;
1100			TMPFS_NODE_UNLOCK(fdnode);
1101		}
1102
1103		/* Do the move: just remove the entry from the source directory
1104		 * and insert it into the target one. */
1105		tmpfs_dir_detach(fdvp, de);
1106		if (fcnp->cn_flags & DOWHITEOUT)
1107			tmpfs_dir_whiteout_add(fdvp, fcnp);
1108		if (tcnp->cn_flags & ISWHITEOUT)
1109			tmpfs_dir_whiteout_remove(tdvp, tcnp);
1110		tmpfs_dir_attach(tdvp, de);
1111	}
1112
1113	/* If the name has changed, we need to make it effective by changing
1114	 * it in the directory entry. */
1115	if (newname != NULL) {
1116		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1117
1118		free(de->td_name, M_TMPFSNAME);
1119		de->td_namelen = (uint16_t)tcnp->cn_namelen;
1120		memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen);
1121		de->td_name = newname;
1122
1123		fnode->tn_status |= TMPFS_NODE_CHANGED;
1124		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1125	}
1126
1127	/* If we are overwriting an entry, we have to remove the old one
1128	 * from the target directory. */
1129	if (tvp != NULL) {
1130		/* Remove the old entry from the target directory. */
1131		de = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1132		tmpfs_dir_detach(tdvp, de);
1133
1134		/* Free the directory entry we just deleted.  Note that the
1135		 * node referred by it will not be removed until the vnode is
1136		 * really reclaimed. */
1137		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE);
1138	}
1139	cache_purge(fvp);
1140
1141	error = 0;
1142
1143out_locked:
1144	if (fdvp != tdvp && fdvp != tvp)
1145		VOP_UNLOCK(fdvp, 0);
1146
1147out:
1148	/* Release target nodes. */
1149	/* XXX: I don't understand when tdvp can be the same as tvp, but
1150	 * other code takes care of this... */
1151	if (tdvp == tvp)
1152		vrele(tdvp);
1153	else
1154		vput(tdvp);
1155	if (tvp != NULL)
1156		vput(tvp);
1157
1158	/* Release source nodes. */
1159	vrele(fdvp);
1160	vrele(fvp);
1161
1162	return error;
1163}
1164
1165/* --------------------------------------------------------------------- */
1166
1167static int
1168tmpfs_mkdir(struct vop_mkdir_args *v)
1169{
1170	struct vnode *dvp = v->a_dvp;
1171	struct vnode **vpp = v->a_vpp;
1172	struct componentname *cnp = v->a_cnp;
1173	struct vattr *vap = v->a_vap;
1174
1175	MPASS(vap->va_type == VDIR);
1176
1177	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1178}
1179
1180/* --------------------------------------------------------------------- */
1181
1182static int
1183tmpfs_rmdir(struct vop_rmdir_args *v)
1184{
1185	struct vnode *dvp = v->a_dvp;
1186	struct vnode *vp = v->a_vp;
1187
1188	int error;
1189	struct tmpfs_dirent *de;
1190	struct tmpfs_mount *tmp;
1191	struct tmpfs_node *dnode;
1192	struct tmpfs_node *node;
1193
1194	MPASS(VOP_ISLOCKED(dvp));
1195	MPASS(VOP_ISLOCKED(vp));
1196
1197	tmp = VFS_TO_TMPFS(dvp->v_mount);
1198	dnode = VP_TO_TMPFS_DIR(dvp);
1199	node = VP_TO_TMPFS_DIR(vp);
1200
1201	/* Directories with more than two entries ('.' and '..') cannot be
1202	 * removed. */
1203	 if (node->tn_size > 0) {
1204		 error = ENOTEMPTY;
1205		 goto out;
1206	 }
1207
1208	if ((dnode->tn_flags & APPEND)
1209	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1210		error = EPERM;
1211		goto out;
1212	}
1213
1214	/* This invariant holds only if we are not trying to remove "..".
1215	  * We checked for that above so this is safe now. */
1216	MPASS(node->tn_dir.tn_parent == dnode);
1217
1218	/* Get the directory entry associated with node (vp).  This was
1219	 * filled by tmpfs_lookup while looking up the entry. */
1220	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1221	MPASS(TMPFS_DIRENT_MATCHES(de,
1222	    v->a_cnp->cn_nameptr,
1223	    v->a_cnp->cn_namelen));
1224
1225	/* Check flags to see if we are allowed to remove the directory. */
1226	if (dnode->tn_flags & APPEND
1227		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1228		error = EPERM;
1229		goto out;
1230	}
1231
1232
1233	/* Detach the directory entry from the directory (dnode). */
1234	tmpfs_dir_detach(dvp, de);
1235	if (v->a_cnp->cn_flags & DOWHITEOUT)
1236		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1237
1238	/* No vnode should be allocated for this entry from this point */
1239	TMPFS_NODE_LOCK(node);
1240	TMPFS_ASSERT_ELOCKED(node);
1241	node->tn_links--;
1242	node->tn_dir.tn_parent = NULL;
1243	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1244	    TMPFS_NODE_MODIFIED;
1245
1246	TMPFS_NODE_UNLOCK(node);
1247
1248	TMPFS_NODE_LOCK(dnode);
1249	TMPFS_ASSERT_ELOCKED(dnode);
1250	dnode->tn_links--;
1251	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1252	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1253	TMPFS_NODE_UNLOCK(dnode);
1254
1255	cache_purge(dvp);
1256	cache_purge(vp);
1257
1258	/* Free the directory entry we just deleted.  Note that the node
1259	 * referred by it will not be removed until the vnode is really
1260	 * reclaimed. */
1261	tmpfs_free_dirent(tmp, de, TRUE);
1262
1263	/* Release the deleted vnode (will destroy the node, notify
1264	 * interested parties and clean it from the cache). */
1265
1266	dnode->tn_status |= TMPFS_NODE_CHANGED;
1267	tmpfs_update(dvp);
1268
1269	error = 0;
1270
1271out:
1272	return error;
1273}
1274
1275/* --------------------------------------------------------------------- */
1276
1277static int
1278tmpfs_symlink(struct vop_symlink_args *v)
1279{
1280	struct vnode *dvp = v->a_dvp;
1281	struct vnode **vpp = v->a_vpp;
1282	struct componentname *cnp = v->a_cnp;
1283	struct vattr *vap = v->a_vap;
1284	char *target = v->a_target;
1285
1286#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1287	MPASS(vap->va_type == VLNK);
1288#else
1289	vap->va_type = VLNK;
1290#endif
1291
1292	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1293}
1294
1295/* --------------------------------------------------------------------- */
1296
1297static int
1298tmpfs_readdir(struct vop_readdir_args *v)
1299{
1300	struct vnode *vp = v->a_vp;
1301	struct uio *uio = v->a_uio;
1302	int *eofflag = v->a_eofflag;
1303	u_long **cookies = v->a_cookies;
1304	int *ncookies = v->a_ncookies;
1305
1306	int error;
1307	off_t startoff;
1308	off_t cnt = 0;
1309	struct tmpfs_node *node;
1310
1311	/* This operation only makes sense on directory nodes. */
1312	if (vp->v_type != VDIR)
1313		return ENOTDIR;
1314
1315	node = VP_TO_TMPFS_DIR(vp);
1316
1317	startoff = uio->uio_offset;
1318
1319	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1320		error = tmpfs_dir_getdotdent(node, uio);
1321		if (error != 0)
1322			goto outok;
1323		cnt++;
1324	}
1325
1326	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1327		error = tmpfs_dir_getdotdotdent(node, uio);
1328		if (error != 0)
1329			goto outok;
1330		cnt++;
1331	}
1332
1333	error = tmpfs_dir_getdents(node, uio, &cnt);
1334
1335outok:
1336	MPASS(error >= -1);
1337
1338	if (error == -1)
1339		error = (cnt != 0) ? 0 : EINVAL;
1340
1341	if (eofflag != NULL)
1342		*eofflag =
1343		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1344
1345	/* Update NFS-related variables. */
1346	if (error == 0 && cookies != NULL && ncookies != NULL) {
1347		off_t i;
1348		off_t off = startoff;
1349		struct tmpfs_dirent *de = NULL;
1350
1351		*ncookies = cnt;
1352		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1353
1354		for (i = 0; i < cnt; i++) {
1355			MPASS(off != TMPFS_DIRCOOKIE_EOF);
1356			if (off == TMPFS_DIRCOOKIE_DOT) {
1357				off = TMPFS_DIRCOOKIE_DOTDOT;
1358			} else {
1359				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1360					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1361				} else if (de != NULL) {
1362					de = TAILQ_NEXT(de, td_entries);
1363				} else {
1364					de = tmpfs_dir_lookupbycookie(node,
1365					    off);
1366					MPASS(de != NULL);
1367					de = TAILQ_NEXT(de, td_entries);
1368				}
1369				if (de == NULL)
1370					off = TMPFS_DIRCOOKIE_EOF;
1371				else
1372					off = tmpfs_dircookie(de);
1373			}
1374
1375			(*cookies)[i] = off;
1376		}
1377		MPASS(uio->uio_offset == off);
1378	}
1379
1380	return error;
1381}
1382
1383/* --------------------------------------------------------------------- */
1384
1385static int
1386tmpfs_readlink(struct vop_readlink_args *v)
1387{
1388	struct vnode *vp = v->a_vp;
1389	struct uio *uio = v->a_uio;
1390
1391	int error;
1392	struct tmpfs_node *node;
1393
1394	MPASS(uio->uio_offset == 0);
1395	MPASS(vp->v_type == VLNK);
1396
1397	node = VP_TO_TMPFS_NODE(vp);
1398
1399	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1400	    uio);
1401	node->tn_status |= TMPFS_NODE_ACCESSED;
1402
1403	return error;
1404}
1405
1406/* --------------------------------------------------------------------- */
1407
1408static int
1409tmpfs_inactive(struct vop_inactive_args *v)
1410{
1411	struct vnode *vp = v->a_vp;
1412	struct thread *l = v->a_td;
1413
1414	struct tmpfs_node *node;
1415
1416	MPASS(VOP_ISLOCKED(vp));
1417
1418	node = VP_TO_TMPFS_NODE(vp);
1419
1420	if (node->tn_links == 0)
1421		vrecycle(vp, l);
1422
1423	return 0;
1424}
1425
1426/* --------------------------------------------------------------------- */
1427
1428int
1429tmpfs_reclaim(struct vop_reclaim_args *v)
1430{
1431	struct vnode *vp = v->a_vp;
1432
1433	struct tmpfs_mount *tmp;
1434	struct tmpfs_node *node;
1435
1436	node = VP_TO_TMPFS_NODE(vp);
1437	tmp = VFS_TO_TMPFS(vp->v_mount);
1438
1439	vnode_destroy_vobject(vp);
1440	cache_purge(vp);
1441
1442	TMPFS_NODE_LOCK(node);
1443	TMPFS_ASSERT_ELOCKED(node);
1444	tmpfs_free_vp(vp);
1445
1446	/* If the node referenced by this vnode was deleted by the user,
1447	 * we must free its associated data structures (now that the vnode
1448	 * is being reclaimed). */
1449	if (node->tn_links == 0 &&
1450	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1451		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1452		TMPFS_NODE_UNLOCK(node);
1453		tmpfs_free_node(tmp, node);
1454	} else
1455		TMPFS_NODE_UNLOCK(node);
1456
1457	MPASS(vp->v_data == NULL);
1458	return 0;
1459}
1460
1461/* --------------------------------------------------------------------- */
1462
1463static int
1464tmpfs_print(struct vop_print_args *v)
1465{
1466	struct vnode *vp = v->a_vp;
1467
1468	struct tmpfs_node *node;
1469
1470	node = VP_TO_TMPFS_NODE(vp);
1471
1472	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1473	    node, node->tn_flags, node->tn_links);
1474	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1475	    ", status 0x%x\n",
1476	    node->tn_mode, node->tn_uid, node->tn_gid,
1477	    (uintmax_t)node->tn_size, node->tn_status);
1478
1479	if (vp->v_type == VFIFO)
1480		fifo_printinfo(vp);
1481
1482	printf("\n");
1483
1484	return 0;
1485}
1486
1487/* --------------------------------------------------------------------- */
1488
1489static int
1490tmpfs_pathconf(struct vop_pathconf_args *v)
1491{
1492	int name = v->a_name;
1493	register_t *retval = v->a_retval;
1494
1495	int error;
1496
1497	error = 0;
1498
1499	switch (name) {
1500	case _PC_LINK_MAX:
1501		*retval = LINK_MAX;
1502		break;
1503
1504	case _PC_NAME_MAX:
1505		*retval = NAME_MAX;
1506		break;
1507
1508	case _PC_PATH_MAX:
1509		*retval = PATH_MAX;
1510		break;
1511
1512	case _PC_PIPE_BUF:
1513		*retval = PIPE_BUF;
1514		break;
1515
1516	case _PC_CHOWN_RESTRICTED:
1517		*retval = 1;
1518		break;
1519
1520	case _PC_NO_TRUNC:
1521		*retval = 1;
1522		break;
1523
1524	case _PC_SYNC_IO:
1525		*retval = 1;
1526		break;
1527
1528	case _PC_FILESIZEBITS:
1529		*retval = 0; /* XXX Don't know which value should I return. */
1530		break;
1531
1532	default:
1533		error = EINVAL;
1534	}
1535
1536	return error;
1537}
1538
1539static int
1540tmpfs_vptofh(struct vop_vptofh_args *ap)
1541{
1542	struct tmpfs_fid *tfhp;
1543	struct tmpfs_node *node;
1544
1545	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1546	node = VP_TO_TMPFS_NODE(ap->a_vp);
1547
1548	tfhp->tf_len = sizeof(struct tmpfs_fid);
1549	tfhp->tf_id = node->tn_id;
1550	tfhp->tf_gen = node->tn_gen;
1551
1552	return (0);
1553}
1554
1555static int
1556tmpfs_whiteout(struct vop_whiteout_args *ap)
1557{
1558	struct vnode *dvp = ap->a_dvp;
1559	struct componentname *cnp = ap->a_cnp;
1560	struct tmpfs_dirent *de;
1561
1562	switch (ap->a_flags) {
1563	case LOOKUP:
1564		return (0);
1565	case CREATE:
1566		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1567		if (de != NULL)
1568			return (de->td_node == NULL ? 0 : EEXIST);
1569		return (tmpfs_dir_whiteout_add(dvp, cnp));
1570	case DELETE:
1571		tmpfs_dir_whiteout_remove(dvp, cnp);
1572		return (0);
1573	default:
1574		panic("tmpfs_whiteout: unknown op");
1575	}
1576}
1577
1578/* --------------------------------------------------------------------- */
1579
1580/*
1581 * vnode operations vector used for files stored in a tmpfs file system.
1582 */
1583struct vop_vector tmpfs_vnodeop_entries = {
1584	.vop_default =			&default_vnodeops,
1585	.vop_lookup =			vfs_cache_lookup,
1586	.vop_cachedlookup =		tmpfs_lookup,
1587	.vop_create =			tmpfs_create,
1588	.vop_mknod =			tmpfs_mknod,
1589	.vop_open =			tmpfs_open,
1590	.vop_close =			tmpfs_close,
1591	.vop_access =			tmpfs_access,
1592	.vop_getattr =			tmpfs_getattr,
1593	.vop_setattr =			tmpfs_setattr,
1594	.vop_read =			tmpfs_read,
1595	.vop_write =			tmpfs_write,
1596	.vop_fsync =			tmpfs_fsync,
1597	.vop_remove =			tmpfs_remove,
1598	.vop_link =			tmpfs_link,
1599	.vop_rename =			tmpfs_rename,
1600	.vop_mkdir =			tmpfs_mkdir,
1601	.vop_rmdir =			tmpfs_rmdir,
1602	.vop_symlink =			tmpfs_symlink,
1603	.vop_readdir =			tmpfs_readdir,
1604	.vop_readlink =			tmpfs_readlink,
1605	.vop_inactive =			tmpfs_inactive,
1606	.vop_reclaim =			tmpfs_reclaim,
1607	.vop_print =			tmpfs_print,
1608	.vop_pathconf =			tmpfs_pathconf,
1609	.vop_vptofh =			tmpfs_vptofh,
1610	.vop_whiteout =			tmpfs_whiteout,
1611	.vop_bmap =			VOP_EOPNOTSUPP,
1612};
1613
1614