tmpfs_vnops.c revision 233385
1202719Sgabor/*	$NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $	*/
2202719Sgabor
3202719Sgabor/*-
4202719Sgabor * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
5202719Sgabor * All rights reserved.
6202719Sgabor *
7202719Sgabor * This code is derived from software contributed to The NetBSD Foundation
8202719Sgabor * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9202719Sgabor * 2005 program.
10202719Sgabor *
11202719Sgabor * Redistribution and use in source and binary forms, with or without
12202719Sgabor * modification, are permitted provided that the following conditions
13202719Sgabor * are met:
14202719Sgabor * 1. Redistributions of source code must retain the above copyright
15202719Sgabor *    notice, this list of conditions and the following disclaimer.
16202719Sgabor * 2. Redistributions in binary form must reproduce the above copyright
17202719Sgabor *    notice, this list of conditions and the following disclaimer in the
18202719Sgabor *    documentation and/or other materials provided with the distribution.
19202719Sgabor *
20202719Sgabor * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21202719Sgabor * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22202719Sgabor * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23202719Sgabor * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24202719Sgabor * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25202719Sgabor * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26202719Sgabor * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27202719Sgabor * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28202719Sgabor * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29202719Sgabor * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30202719Sgabor * POSSIBILITY OF SUCH DAMAGE.
31202719Sgabor */
32202719Sgabor
33202719Sgabor/*
34202719Sgabor * tmpfs vnode interface.
35202719Sgabor */
36202719Sgabor#include <sys/cdefs.h>
37202719Sgabor__FBSDID("$FreeBSD: stable/9/sys/fs/tmpfs/tmpfs_vnops.c 233385 2012-03-23 18:29:09Z jhb $");
38202719Sgabor
39202719Sgabor#include <sys/param.h>
40202719Sgabor#include <sys/fcntl.h>
41202719Sgabor#include <sys/lockf.h>
42202719Sgabor#include <sys/namei.h>
43202719Sgabor#include <sys/priv.h>
44202719Sgabor#include <sys/proc.h>
45202719Sgabor#include <sys/sched.h>
46202719Sgabor#include <sys/sf_buf.h>
47202719Sgabor#include <sys/stat.h>
48202719Sgabor#include <sys/systm.h>
49202719Sgabor#include <sys/unistd.h>
50202719Sgabor#include <sys/vnode.h>
51202719Sgabor
52202719Sgabor#include <vm/vm.h>
53202719Sgabor#include <vm/vm_object.h>
54202719Sgabor#include <vm/vm_page.h>
55202719Sgabor#include <vm/vm_pager.h>
56202719Sgabor
57202719Sgabor#include <machine/_inttypes.h>
58202719Sgabor
59202719Sgabor#include <fs/fifofs/fifo.h>
60202719Sgabor#include <fs/tmpfs/tmpfs_vnops.h>
61202719Sgabor#include <fs/tmpfs/tmpfs.h>
62202719Sgabor
63202719Sgabor/* --------------------------------------------------------------------- */
64202719Sgabor
65202719Sgaborstatic int
66202719Sgabortmpfs_lookup(struct vop_cachedlookup_args *v)
67202719Sgabor{
68202719Sgabor	struct vnode *dvp = v->a_dvp;
69202719Sgabor	struct vnode **vpp = v->a_vpp;
70202719Sgabor	struct componentname *cnp = v->a_cnp;
71202719Sgabor
72202719Sgabor	int error;
73202719Sgabor	struct tmpfs_dirent *de;
74202719Sgabor	struct tmpfs_node *dnode;
75202719Sgabor
76202719Sgabor	dnode = VP_TO_TMPFS_DIR(dvp);
77202719Sgabor	*vpp = NULLVP;
78202719Sgabor
79202719Sgabor	/* Check accessibility of requested node as a first step. */
80202719Sgabor	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
81202719Sgabor	if (error != 0)
82202719Sgabor		goto out;
83202719Sgabor
84202719Sgabor	/* We cannot be requesting the parent directory of the root node. */
85202719Sgabor	MPASS(IMPLIES(dnode->tn_type == VDIR &&
86202719Sgabor	    dnode->tn_dir.tn_parent == dnode,
87202719Sgabor	    !(cnp->cn_flags & ISDOTDOT)));
88202719Sgabor
89202719Sgabor	TMPFS_ASSERT_LOCKED(dnode);
90202719Sgabor	if (dnode->tn_dir.tn_parent == NULL) {
91202719Sgabor		error = ENOENT;
92202719Sgabor		goto out;
93202719Sgabor	}
94202719Sgabor	if (cnp->cn_flags & ISDOTDOT) {
95202719Sgabor		int ltype = 0;
96202719Sgabor
97202719Sgabor		ltype = VOP_ISLOCKED(dvp);
98202719Sgabor		vhold(dvp);
99202719Sgabor		VOP_UNLOCK(dvp, 0);
100202719Sgabor		/* Allocate a new vnode on the matching entry. */
101202719Sgabor		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
102202719Sgabor		    cnp->cn_lkflags, vpp);
103202719Sgabor
104202719Sgabor		vn_lock(dvp, ltype | LK_RETRY);
105202719Sgabor		vdrop(dvp);
106202719Sgabor	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
107202719Sgabor		VREF(dvp);
108202719Sgabor		*vpp = dvp;
109202719Sgabor		error = 0;
110202719Sgabor	} else {
111202719Sgabor		de = tmpfs_dir_lookup(dnode, NULL, cnp);
112202719Sgabor		if (de != NULL && de->td_node == NULL)
113202719Sgabor			cnp->cn_flags |= ISWHITEOUT;
114202719Sgabor		if (de == NULL || de->td_node == NULL) {
115202719Sgabor			/* The entry was not found in the directory.
116202719Sgabor			 * This is OK if we are creating or renaming an
117202719Sgabor			 * entry and are working on the last component of
118202719Sgabor			 * the path name. */
119202719Sgabor			if ((cnp->cn_flags & ISLASTCN) &&
120202719Sgabor			    (cnp->cn_nameiop == CREATE || \
121202719Sgabor			    cnp->cn_nameiop == RENAME ||
122202719Sgabor			    (cnp->cn_nameiop == DELETE &&
123202719Sgabor			    cnp->cn_flags & DOWHITEOUT &&
124202719Sgabor			    cnp->cn_flags & ISWHITEOUT))) {
125202719Sgabor				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
126202719Sgabor				    cnp->cn_thread);
127202719Sgabor				if (error != 0)
128202719Sgabor					goto out;
129202719Sgabor
130202719Sgabor				/* Keep the component name in the buffer for
131202719Sgabor				 * future uses. */
132202719Sgabor				cnp->cn_flags |= SAVENAME;
133202719Sgabor
134202719Sgabor				error = EJUSTRETURN;
135202719Sgabor			} else
136202719Sgabor				error = ENOENT;
137202719Sgabor		} else {
138202719Sgabor			struct tmpfs_node *tnode;
139202719Sgabor
140202719Sgabor			/* The entry was found, so get its associated
141202719Sgabor			 * tmpfs_node. */
142202719Sgabor			tnode = de->td_node;
143202719Sgabor
144202719Sgabor			/* If we are not at the last path component and
145202719Sgabor			 * found a non-directory or non-link entry (which
146202719Sgabor			 * may itself be pointing to a directory), raise
147202719Sgabor			 * an error. */
148202719Sgabor			if ((tnode->tn_type != VDIR &&
149202719Sgabor			    tnode->tn_type != VLNK) &&
150202719Sgabor			    !(cnp->cn_flags & ISLASTCN)) {
151202719Sgabor				error = ENOTDIR;
152202719Sgabor				goto out;
153202719Sgabor			}
154202719Sgabor
155202719Sgabor			/* If we are deleting or renaming the entry, keep
156202719Sgabor			 * track of its tmpfs_dirent so that it can be
157202719Sgabor			 * easily deleted later. */
158202719Sgabor			if ((cnp->cn_flags & ISLASTCN) &&
159202719Sgabor			    (cnp->cn_nameiop == DELETE ||
160202719Sgabor			    cnp->cn_nameiop == RENAME)) {
161202719Sgabor				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
162202719Sgabor				    cnp->cn_thread);
163202719Sgabor				if (error != 0)
164202719Sgabor					goto out;
165202719Sgabor
166202719Sgabor				/* Allocate a new vnode on the matching entry. */
167202719Sgabor				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
168202719Sgabor						cnp->cn_lkflags, vpp);
169202719Sgabor				if (error != 0)
170202719Sgabor					goto out;
171202719Sgabor
172202719Sgabor				if ((dnode->tn_mode & S_ISTXT) &&
173202719Sgabor				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) &&
174202719Sgabor				  VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) {
175202719Sgabor					error = EPERM;
176202719Sgabor					vput(*vpp);
177202719Sgabor					*vpp = NULL;
178202719Sgabor					goto out;
179202719Sgabor				}
180202719Sgabor				cnp->cn_flags |= SAVENAME;
181202719Sgabor			} else {
182202719Sgabor				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
183202719Sgabor						cnp->cn_lkflags, vpp);
184202719Sgabor			}
185202719Sgabor		}
186202719Sgabor	}
187202719Sgabor
188202719Sgabor	/* Store the result of this lookup in the cache.  Avoid this if the
189202719Sgabor	 * request was for creation, as it does not improve timings on
190202719Sgabor	 * emprical tests. */
191202719Sgabor	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
192202719Sgabor		cache_enter(dvp, *vpp, cnp);
193202719Sgabor
194202719Sgaborout:
195202719Sgabor	/* If there were no errors, *vpp cannot be null and it must be
196202719Sgabor	 * locked. */
197202719Sgabor	MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
198202719Sgabor
199202719Sgabor	return error;
200202719Sgabor}
201202719Sgabor
202202719Sgabor/* --------------------------------------------------------------------- */
203202719Sgabor
204202719Sgaborstatic int
205202719Sgabortmpfs_create(struct vop_create_args *v)
206202719Sgabor{
207202719Sgabor	struct vnode *dvp = v->a_dvp;
208202719Sgabor	struct vnode **vpp = v->a_vpp;
209202719Sgabor	struct componentname *cnp = v->a_cnp;
210202719Sgabor	struct vattr *vap = v->a_vap;
211202719Sgabor
212202719Sgabor	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
213202719Sgabor
214202719Sgabor	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
215202719Sgabor}
216202719Sgabor/* --------------------------------------------------------------------- */
217202719Sgabor
218202719Sgaborstatic int
219202719Sgabortmpfs_mknod(struct vop_mknod_args *v)
220202719Sgabor{
221202719Sgabor	struct vnode *dvp = v->a_dvp;
222202719Sgabor	struct vnode **vpp = v->a_vpp;
223202719Sgabor	struct componentname *cnp = v->a_cnp;
224202719Sgabor	struct vattr *vap = v->a_vap;
225202719Sgabor
226202719Sgabor	if (vap->va_type != VBLK && vap->va_type != VCHR &&
227202719Sgabor	    vap->va_type != VFIFO)
228202719Sgabor		return EINVAL;
229202719Sgabor
230202719Sgabor	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
231202719Sgabor}
232202719Sgabor
233202719Sgabor/* --------------------------------------------------------------------- */
234202719Sgabor
235202719Sgaborstatic int
236202719Sgabortmpfs_open(struct vop_open_args *v)
237202719Sgabor{
238202719Sgabor	struct vnode *vp = v->a_vp;
239202719Sgabor	int mode = v->a_mode;
240202719Sgabor
241202719Sgabor	int error;
242202719Sgabor	struct tmpfs_node *node;
243202719Sgabor
244202719Sgabor	MPASS(VOP_ISLOCKED(vp));
245202719Sgabor
246202719Sgabor	node = VP_TO_TMPFS_NODE(vp);
247202719Sgabor
248202719Sgabor	/* The file is still active but all its names have been removed
249202719Sgabor	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
250202719Sgabor	 * it is about to die. */
251202719Sgabor	if (node->tn_links < 1)
252202719Sgabor		return (ENOENT);
253202719Sgabor
254202719Sgabor	/* If the file is marked append-only, deny write requests. */
255202719Sgabor	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
256202719Sgabor		error = EPERM;
257202719Sgabor	else {
258202719Sgabor		error = 0;
259202719Sgabor		vnode_create_vobject(vp, node->tn_size, v->a_td);
260202719Sgabor	}
261202719Sgabor
262202719Sgabor	MPASS(VOP_ISLOCKED(vp));
263202719Sgabor	return error;
264202719Sgabor}
265202719Sgabor
266202719Sgabor/* --------------------------------------------------------------------- */
267202719Sgabor
268202719Sgaborstatic int
269202719Sgabortmpfs_close(struct vop_close_args *v)
270202719Sgabor{
271202719Sgabor	struct vnode *vp = v->a_vp;
272202719Sgabor
273202719Sgabor	MPASS(VOP_ISLOCKED(vp));
274202719Sgabor
275202719Sgabor	/* Update node times. */
276202719Sgabor	tmpfs_update(vp);
277202719Sgabor
278202719Sgabor	return (0);
279202719Sgabor}
280202719Sgabor
281202719Sgabor/* --------------------------------------------------------------------- */
282202719Sgabor
283202719Sgaborint
284202719Sgabortmpfs_access(struct vop_access_args *v)
285202719Sgabor{
286202719Sgabor	struct vnode *vp = v->a_vp;
287202719Sgabor	accmode_t accmode = v->a_accmode;
288202719Sgabor	struct ucred *cred = v->a_cred;
289202719Sgabor
290202719Sgabor	int error;
291202719Sgabor	struct tmpfs_node *node;
292202719Sgabor
293202719Sgabor	MPASS(VOP_ISLOCKED(vp));
294202719Sgabor
295202719Sgabor	node = VP_TO_TMPFS_NODE(vp);
296202719Sgabor
297202719Sgabor	switch (vp->v_type) {
298202719Sgabor	case VDIR:
299202719Sgabor		/* FALLTHROUGH */
300202719Sgabor	case VLNK:
301202719Sgabor		/* FALLTHROUGH */
302202719Sgabor	case VREG:
303202719Sgabor		if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
304202719Sgabor			error = EROFS;
305202719Sgabor			goto out;
306202719Sgabor		}
307202719Sgabor		break;
308202719Sgabor
309202719Sgabor	case VBLK:
310202719Sgabor		/* FALLTHROUGH */
311202719Sgabor	case VCHR:
312202719Sgabor		/* FALLTHROUGH */
313202719Sgabor	case VSOCK:
314202719Sgabor		/* FALLTHROUGH */
315202719Sgabor	case VFIFO:
316202719Sgabor		break;
317202719Sgabor
318202719Sgabor	default:
319202719Sgabor		error = EINVAL;
320202719Sgabor		goto out;
321202719Sgabor	}
322202719Sgabor
323202719Sgabor	if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
324202719Sgabor		error = EPERM;
325202719Sgabor		goto out;
326202719Sgabor	}
327202719Sgabor
328202719Sgabor	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
329202719Sgabor	    node->tn_gid, accmode, cred, NULL);
330202719Sgabor
331202719Sgaborout:
332202719Sgabor	MPASS(VOP_ISLOCKED(vp));
333202719Sgabor
334202719Sgabor	return error;
335202719Sgabor}
336202719Sgabor
337202719Sgabor/* --------------------------------------------------------------------- */
338202719Sgabor
339202719Sgaborint
340202719Sgabortmpfs_getattr(struct vop_getattr_args *v)
341202719Sgabor{
342202719Sgabor	struct vnode *vp = v->a_vp;
343202719Sgabor	struct vattr *vap = v->a_vap;
344202719Sgabor
345202719Sgabor	struct tmpfs_node *node;
346202719Sgabor
347202719Sgabor	node = VP_TO_TMPFS_NODE(vp);
348202719Sgabor
349202719Sgabor	tmpfs_update(vp);
350202719Sgabor
351202719Sgabor	vap->va_type = vp->v_type;
352202719Sgabor	vap->va_mode = node->tn_mode;
353202719Sgabor	vap->va_nlink = node->tn_links;
354202719Sgabor	vap->va_uid = node->tn_uid;
355202719Sgabor	vap->va_gid = node->tn_gid;
356202719Sgabor	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
357202719Sgabor	vap->va_fileid = node->tn_id;
358202719Sgabor	vap->va_size = node->tn_size;
359202719Sgabor	vap->va_blocksize = PAGE_SIZE;
360202719Sgabor	vap->va_atime = node->tn_atime;
361202719Sgabor	vap->va_mtime = node->tn_mtime;
362202719Sgabor	vap->va_ctime = node->tn_ctime;
363202719Sgabor	vap->va_birthtime = node->tn_birthtime;
364202719Sgabor	vap->va_gen = node->tn_gen;
365202719Sgabor	vap->va_flags = node->tn_flags;
366202719Sgabor	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
367202719Sgabor		node->tn_rdev : NODEV;
368202719Sgabor	vap->va_bytes = round_page(node->tn_size);
369202719Sgabor	vap->va_filerev = 0;
370202719Sgabor
371202719Sgabor	return 0;
372202719Sgabor}
373202719Sgabor
374202719Sgabor/* --------------------------------------------------------------------- */
375202719Sgabor
376202719Sgabor/* XXX Should this operation be atomic?  I think it should, but code in
377202719Sgabor * XXX other places (e.g., ufs) doesn't seem to be... */
378202719Sgaborint
379202719Sgabortmpfs_setattr(struct vop_setattr_args *v)
380202719Sgabor{
381202719Sgabor	struct vnode *vp = v->a_vp;
382202719Sgabor	struct vattr *vap = v->a_vap;
383202719Sgabor	struct ucred *cred = v->a_cred;
384202719Sgabor	struct thread *td = curthread;
385202719Sgabor
386202719Sgabor	int error;
387202719Sgabor
388202719Sgabor	MPASS(VOP_ISLOCKED(vp));
389202719Sgabor
390202719Sgabor	error = 0;
391202719Sgabor
392202719Sgabor	/* Abort if any unsettable attribute is given. */
393202719Sgabor	if (vap->va_type != VNON ||
394202719Sgabor	    vap->va_nlink != VNOVAL ||
395202719Sgabor	    vap->va_fsid != VNOVAL ||
396202719Sgabor	    vap->va_fileid != VNOVAL ||
397202719Sgabor	    vap->va_blocksize != VNOVAL ||
398202719Sgabor	    vap->va_gen != VNOVAL ||
399202719Sgabor	    vap->va_rdev != VNOVAL ||
400202719Sgabor	    vap->va_bytes != VNOVAL)
401202719Sgabor		error = EINVAL;
402202719Sgabor
403202719Sgabor	if (error == 0 && (vap->va_flags != VNOVAL))
404202719Sgabor		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
405202719Sgabor
406202719Sgabor	if (error == 0 && (vap->va_size != VNOVAL))
407202719Sgabor		error = tmpfs_chsize(vp, vap->va_size, cred, td);
408202719Sgabor
409202719Sgabor	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
410202719Sgabor		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
411202719Sgabor
412202719Sgabor	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
413202719Sgabor		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
414202719Sgabor
415202719Sgabor	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
416202719Sgabor	    vap->va_atime.tv_nsec != VNOVAL) ||
417202719Sgabor	    (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;
441
442	VM_OBJECT_LOCK(tobj);
443	vm_object_pip_add(tobj, 1);
444	m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
445	    VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
446	if (m->valid != VM_PAGE_BITS_ALL) {
447		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
448			error = vm_pager_get_pages(tobj, &m, 1, 0);
449			if (error != 0) {
450				printf("tmpfs get pages from pager error [read]\n");
451				goto out;
452			}
453		} else
454			vm_page_zero_invalid(m, TRUE);
455	}
456	VM_OBJECT_UNLOCK(tobj);
457	error = uiomove_fromphys(&m, offset, tlen, uio);
458	VM_OBJECT_LOCK(tobj);
459out:
460	vm_page_lock(m);
461	vm_page_unwire(m, TRUE);
462	vm_page_unlock(m);
463	vm_page_wakeup(m);
464	vm_object_pip_subtract(tobj, 1);
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;
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	vm_object_pip_add(tobj, 1);
668	tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
669	    VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
670	if (tpg->valid != VM_PAGE_BITS_ALL) {
671		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
672			error = vm_pager_get_pages(tobj, &tpg, 1, 0);
673			if (error != 0) {
674				printf("tmpfs get pages from pager error [write]\n");
675				goto out;
676			}
677		} else
678			vm_page_zero_invalid(tpg, TRUE);
679	}
680	VM_OBJECT_UNLOCK(tobj);
681	if (vpg == NULL)
682		error = uiomove_fromphys(&tpg, offset, tlen, uio);
683	else {
684		KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid"));
685		pmap_copy_page(vpg, tpg);
686	}
687	VM_OBJECT_LOCK(tobj);
688out:
689	if (vobj != NULL)
690		VM_OBJECT_LOCK(vobj);
691	if (error == 0) {
692		KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
693		    ("parts of tpg invalid"));
694		vm_page_dirty(tpg);
695	}
696	vm_page_lock(tpg);
697	vm_page_unwire(tpg, TRUE);
698	vm_page_unlock(tpg);
699	vm_page_wakeup(tpg);
700	if (vpg != NULL)
701		vm_page_wakeup(vpg);
702	if (vobj != NULL)
703		VM_OBJECT_UNLOCK(vobj);
704	vm_object_pip_subtract(tobj, 1);
705	VM_OBJECT_UNLOCK(tobj);
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		if (error != 0)
752			goto out;
753	}
754
755	uobj = node->tn_reg.tn_aobj;
756	while ((resid = uio->uio_resid) > 0) {
757		if (node->tn_size <= uio->uio_offset)
758			break;
759		len = MIN(node->tn_size - uio->uio_offset, resid);
760		if (len == 0)
761			break;
762		error = tmpfs_mappedwrite(vp->v_object, uobj, len, uio);
763		if ((error != 0) || (resid == uio->uio_resid))
764			break;
765	}
766
767	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
768	    (extended ? TMPFS_NODE_CHANGED : 0);
769
770	if (node->tn_mode & (S_ISUID | S_ISGID)) {
771		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
772			node->tn_mode &= ~(S_ISUID | S_ISGID);
773	}
774
775	if (error != 0)
776		(void)tmpfs_reg_resize(vp, oldsize);
777
778out:
779	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
780	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
781
782	return error;
783}
784
785/* --------------------------------------------------------------------- */
786
787static int
788tmpfs_fsync(struct vop_fsync_args *v)
789{
790	struct vnode *vp = v->a_vp;
791
792	MPASS(VOP_ISLOCKED(vp));
793
794	tmpfs_update(vp);
795
796	return 0;
797}
798
799/* --------------------------------------------------------------------- */
800
801static int
802tmpfs_remove(struct vop_remove_args *v)
803{
804	struct vnode *dvp = v->a_dvp;
805	struct vnode *vp = v->a_vp;
806
807	int error;
808	struct tmpfs_dirent *de;
809	struct tmpfs_mount *tmp;
810	struct tmpfs_node *dnode;
811	struct tmpfs_node *node;
812
813	MPASS(VOP_ISLOCKED(dvp));
814	MPASS(VOP_ISLOCKED(vp));
815
816	if (vp->v_type == VDIR) {
817		error = EISDIR;
818		goto out;
819	}
820
821	dnode = VP_TO_TMPFS_DIR(dvp);
822	node = VP_TO_TMPFS_NODE(vp);
823	tmp = VFS_TO_TMPFS(vp->v_mount);
824	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
825	MPASS(de != NULL);
826
827	/* Files marked as immutable or append-only cannot be deleted. */
828	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
829	    (dnode->tn_flags & APPEND)) {
830		error = EPERM;
831		goto out;
832	}
833
834	/* Remove the entry from the directory; as it is a file, we do not
835	 * have to change the number of hard links of the directory. */
836	tmpfs_dir_detach(dvp, de);
837	if (v->a_cnp->cn_flags & DOWHITEOUT)
838		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
839
840	/* Free the directory entry we just deleted.  Note that the node
841	 * referred by it will not be removed until the vnode is really
842	 * reclaimed. */
843	tmpfs_free_dirent(tmp, de, TRUE);
844
845	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
846	error = 0;
847
848out:
849
850	return error;
851}
852
853/* --------------------------------------------------------------------- */
854
855static int
856tmpfs_link(struct vop_link_args *v)
857{
858	struct vnode *dvp = v->a_tdvp;
859	struct vnode *vp = v->a_vp;
860	struct componentname *cnp = v->a_cnp;
861
862	int error;
863	struct tmpfs_dirent *de;
864	struct tmpfs_node *node;
865
866	MPASS(VOP_ISLOCKED(dvp));
867	MPASS(cnp->cn_flags & HASBUF);
868	MPASS(dvp != vp); /* XXX When can this be false? */
869
870	node = VP_TO_TMPFS_NODE(vp);
871
872	/* XXX: Why aren't the following two tests done by the caller? */
873
874	/* Hard links of directories are forbidden. */
875	if (vp->v_type == VDIR) {
876		error = EPERM;
877		goto out;
878	}
879
880	/* Cannot create cross-device links. */
881	if (dvp->v_mount != vp->v_mount) {
882		error = EXDEV;
883		goto out;
884	}
885
886	/* Ensure that we do not overflow the maximum number of links imposed
887	 * by the system. */
888	MPASS(node->tn_links <= LINK_MAX);
889	if (node->tn_links == LINK_MAX) {
890		error = EMLINK;
891		goto out;
892	}
893
894	/* We cannot create links of files marked immutable or append-only. */
895	if (node->tn_flags & (IMMUTABLE | APPEND)) {
896		error = EPERM;
897		goto out;
898	}
899
900	/* Allocate a new directory entry to represent the node. */
901	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
902	    cnp->cn_nameptr, cnp->cn_namelen, &de);
903	if (error != 0)
904		goto out;
905
906	/* Insert the new directory entry into the appropriate directory. */
907	if (cnp->cn_flags & ISWHITEOUT)
908		tmpfs_dir_whiteout_remove(dvp, cnp);
909	tmpfs_dir_attach(dvp, de);
910
911	/* vp link count has changed, so update node times. */
912	node->tn_status |= TMPFS_NODE_CHANGED;
913	tmpfs_update(vp);
914
915	error = 0;
916
917out:
918	return error;
919}
920
921/* --------------------------------------------------------------------- */
922
923static int
924tmpfs_rename(struct vop_rename_args *v)
925{
926	struct vnode *fdvp = v->a_fdvp;
927	struct vnode *fvp = v->a_fvp;
928	struct componentname *fcnp = v->a_fcnp;
929	struct vnode *tdvp = v->a_tdvp;
930	struct vnode *tvp = v->a_tvp;
931	struct componentname *tcnp = v->a_tcnp;
932
933	char *newname;
934	int error;
935	struct tmpfs_dirent *de;
936	struct tmpfs_mount *tmp;
937	struct tmpfs_node *fdnode;
938	struct tmpfs_node *fnode;
939	struct tmpfs_node *tnode;
940	struct tmpfs_node *tdnode;
941
942	MPASS(VOP_ISLOCKED(tdvp));
943	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
944	MPASS(fcnp->cn_flags & HASBUF);
945	MPASS(tcnp->cn_flags & HASBUF);
946
947  	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
948
949	/* Disallow cross-device renames.
950	 * XXX Why isn't this done by the caller? */
951	if (fvp->v_mount != tdvp->v_mount ||
952	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
953		error = EXDEV;
954		goto out;
955	}
956
957	tmp = VFS_TO_TMPFS(tdvp->v_mount);
958	tdnode = VP_TO_TMPFS_DIR(tdvp);
959
960	/* If source and target are the same file, there is nothing to do. */
961	if (fvp == tvp) {
962		error = 0;
963		goto out;
964	}
965
966	/* If we need to move the directory between entries, lock the
967	 * source so that we can safely operate on it. */
968	if (fdvp != tdvp && fdvp != tvp)
969		vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
970	fdnode = VP_TO_TMPFS_DIR(fdvp);
971	fnode = VP_TO_TMPFS_NODE(fvp);
972	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
973
974	/* Entry can disappear before we lock fdvp,
975	 * also avoid manipulating '.' and '..' entries. */
976	if (de == NULL) {
977		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
978		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
979			error = EINVAL;
980		else
981			error = ENOENT;
982		goto out_locked;
983	}
984	MPASS(de->td_node == fnode);
985
986	/* If re-naming a directory to another preexisting directory
987	 * ensure that the target directory is empty so that its
988	 * removal causes no side effects.
989	 * Kern_rename gurantees the destination to be a directory
990	 * if the source is one. */
991	if (tvp != NULL) {
992		MPASS(tnode != NULL);
993
994		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
995		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
996			error = EPERM;
997			goto out_locked;
998		}
999
1000		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1001			if (tnode->tn_size > 0) {
1002				error = ENOTEMPTY;
1003				goto out_locked;
1004			}
1005		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1006			error = ENOTDIR;
1007			goto out_locked;
1008		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1009			error = EISDIR;
1010			goto out_locked;
1011		} else {
1012			MPASS(fnode->tn_type != VDIR &&
1013				tnode->tn_type != VDIR);
1014		}
1015	}
1016
1017	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1018	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1019		error = EPERM;
1020		goto out_locked;
1021	}
1022
1023	/* Ensure that we have enough memory to hold the new name, if it
1024	 * has to be changed. */
1025	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1026	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1027		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1028	} else
1029		newname = NULL;
1030
1031	/* If the node is being moved to another directory, we have to do
1032	 * the move. */
1033	if (fdnode != tdnode) {
1034		/* In case we are moving a directory, we have to adjust its
1035		 * parent to point to the new parent. */
1036		if (de->td_node->tn_type == VDIR) {
1037			struct tmpfs_node *n;
1038
1039			/* Ensure the target directory is not a child of the
1040			 * directory being moved.  Otherwise, we'd end up
1041			 * with stale nodes. */
1042			n = tdnode;
1043			/* TMPFS_LOCK garanties that no nodes are freed while
1044			 * traversing the list. Nodes can only be marked as
1045			 * removed: tn_parent == NULL. */
1046			TMPFS_LOCK(tmp);
1047			TMPFS_NODE_LOCK(n);
1048			while (n != n->tn_dir.tn_parent) {
1049				struct tmpfs_node *parent;
1050
1051				if (n == fnode) {
1052					TMPFS_NODE_UNLOCK(n);
1053					TMPFS_UNLOCK(tmp);
1054					error = EINVAL;
1055					if (newname != NULL)
1056						    free(newname, M_TMPFSNAME);
1057					goto out_locked;
1058				}
1059				parent = n->tn_dir.tn_parent;
1060				TMPFS_NODE_UNLOCK(n);
1061				if (parent == NULL) {
1062					n = NULL;
1063					break;
1064				}
1065				TMPFS_NODE_LOCK(parent);
1066				if (parent->tn_dir.tn_parent == NULL) {
1067					TMPFS_NODE_UNLOCK(parent);
1068					n = NULL;
1069					break;
1070				}
1071				n = parent;
1072			}
1073			TMPFS_UNLOCK(tmp);
1074			if (n == NULL) {
1075				error = EINVAL;
1076				if (newname != NULL)
1077					    free(newname, M_TMPFSNAME);
1078				goto out_locked;
1079			}
1080			TMPFS_NODE_UNLOCK(n);
1081
1082			/* Adjust the parent pointer. */
1083			TMPFS_VALIDATE_DIR(fnode);
1084			TMPFS_NODE_LOCK(de->td_node);
1085			de->td_node->tn_dir.tn_parent = tdnode;
1086			TMPFS_NODE_UNLOCK(de->td_node);
1087
1088			/* As a result of changing the target of the '..'
1089			 * entry, the link count of the source and target
1090			 * directories has to be adjusted. */
1091			TMPFS_NODE_LOCK(tdnode);
1092			TMPFS_ASSERT_LOCKED(tdnode);
1093			tdnode->tn_links++;
1094			TMPFS_NODE_UNLOCK(tdnode);
1095
1096			TMPFS_NODE_LOCK(fdnode);
1097			TMPFS_ASSERT_LOCKED(fdnode);
1098			fdnode->tn_links--;
1099			TMPFS_NODE_UNLOCK(fdnode);
1100		}
1101
1102		/* Do the move: just remove the entry from the source directory
1103		 * and insert it into the target one. */
1104		tmpfs_dir_detach(fdvp, de);
1105		if (fcnp->cn_flags & DOWHITEOUT)
1106			tmpfs_dir_whiteout_add(fdvp, fcnp);
1107		if (tcnp->cn_flags & ISWHITEOUT)
1108			tmpfs_dir_whiteout_remove(tdvp, tcnp);
1109		tmpfs_dir_attach(tdvp, de);
1110	}
1111
1112	/* If the name has changed, we need to make it effective by changing
1113	 * it in the directory entry. */
1114	if (newname != NULL) {
1115		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1116
1117		free(de->td_name, M_TMPFSNAME);
1118		de->td_namelen = (uint16_t)tcnp->cn_namelen;
1119		memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen);
1120		de->td_name = newname;
1121
1122		fnode->tn_status |= TMPFS_NODE_CHANGED;
1123		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1124	}
1125
1126	/* If we are overwriting an entry, we have to remove the old one
1127	 * from the target directory. */
1128	if (tvp != NULL) {
1129		/* Remove the old entry from the target directory. */
1130		de = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1131		tmpfs_dir_detach(tdvp, de);
1132
1133		/* Free the directory entry we just deleted.  Note that the
1134		 * node referred by it will not be removed until the vnode is
1135		 * really reclaimed. */
1136		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE);
1137	}
1138	cache_purge(fvp);
1139	if (tvp != NULL)
1140		cache_purge(tvp);
1141
1142	error = 0;
1143
1144out_locked:
1145	if (fdvp != tdvp && fdvp != tvp)
1146		VOP_UNLOCK(fdvp, 0);
1147
1148out:
1149	/* Release target nodes. */
1150	/* XXX: I don't understand when tdvp can be the same as tvp, but
1151	 * other code takes care of this... */
1152	if (tdvp == tvp)
1153		vrele(tdvp);
1154	else
1155		vput(tdvp);
1156	if (tvp != NULL)
1157		vput(tvp);
1158
1159	/* Release source nodes. */
1160	vrele(fdvp);
1161	vrele(fvp);
1162
1163	return error;
1164}
1165
1166/* --------------------------------------------------------------------- */
1167
1168static int
1169tmpfs_mkdir(struct vop_mkdir_args *v)
1170{
1171	struct vnode *dvp = v->a_dvp;
1172	struct vnode **vpp = v->a_vpp;
1173	struct componentname *cnp = v->a_cnp;
1174	struct vattr *vap = v->a_vap;
1175
1176	MPASS(vap->va_type == VDIR);
1177
1178	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1179}
1180
1181/* --------------------------------------------------------------------- */
1182
1183static int
1184tmpfs_rmdir(struct vop_rmdir_args *v)
1185{
1186	struct vnode *dvp = v->a_dvp;
1187	struct vnode *vp = v->a_vp;
1188
1189	int error;
1190	struct tmpfs_dirent *de;
1191	struct tmpfs_mount *tmp;
1192	struct tmpfs_node *dnode;
1193	struct tmpfs_node *node;
1194
1195	MPASS(VOP_ISLOCKED(dvp));
1196	MPASS(VOP_ISLOCKED(vp));
1197
1198	tmp = VFS_TO_TMPFS(dvp->v_mount);
1199	dnode = VP_TO_TMPFS_DIR(dvp);
1200	node = VP_TO_TMPFS_DIR(vp);
1201
1202	/* Directories with more than two entries ('.' and '..') cannot be
1203	 * removed. */
1204	 if (node->tn_size > 0) {
1205		 error = ENOTEMPTY;
1206		 goto out;
1207	 }
1208
1209	if ((dnode->tn_flags & APPEND)
1210	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1211		error = EPERM;
1212		goto out;
1213	}
1214
1215	/* This invariant holds only if we are not trying to remove "..".
1216	  * We checked for that above so this is safe now. */
1217	MPASS(node->tn_dir.tn_parent == dnode);
1218
1219	/* Get the directory entry associated with node (vp).  This was
1220	 * filled by tmpfs_lookup while looking up the entry. */
1221	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1222	MPASS(TMPFS_DIRENT_MATCHES(de,
1223	    v->a_cnp->cn_nameptr,
1224	    v->a_cnp->cn_namelen));
1225
1226	/* Check flags to see if we are allowed to remove the directory. */
1227	if (dnode->tn_flags & APPEND
1228		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1229		error = EPERM;
1230		goto out;
1231	}
1232
1233
1234	/* Detach the directory entry from the directory (dnode). */
1235	tmpfs_dir_detach(dvp, de);
1236	if (v->a_cnp->cn_flags & DOWHITEOUT)
1237		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1238
1239	/* No vnode should be allocated for this entry from this point */
1240	TMPFS_NODE_LOCK(node);
1241	TMPFS_ASSERT_ELOCKED(node);
1242	node->tn_links--;
1243	node->tn_dir.tn_parent = NULL;
1244	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1245	    TMPFS_NODE_MODIFIED;
1246
1247	TMPFS_NODE_UNLOCK(node);
1248
1249	TMPFS_NODE_LOCK(dnode);
1250	TMPFS_ASSERT_ELOCKED(dnode);
1251	dnode->tn_links--;
1252	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1253	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1254	TMPFS_NODE_UNLOCK(dnode);
1255
1256	cache_purge(dvp);
1257	cache_purge(vp);
1258
1259	/* Free the directory entry we just deleted.  Note that the node
1260	 * referred by it will not be removed until the vnode is really
1261	 * reclaimed. */
1262	tmpfs_free_dirent(tmp, de, TRUE);
1263
1264	/* Release the deleted vnode (will destroy the node, notify
1265	 * interested parties and clean it from the cache). */
1266
1267	dnode->tn_status |= TMPFS_NODE_CHANGED;
1268	tmpfs_update(dvp);
1269
1270	error = 0;
1271
1272out:
1273	return error;
1274}
1275
1276/* --------------------------------------------------------------------- */
1277
1278static int
1279tmpfs_symlink(struct vop_symlink_args *v)
1280{
1281	struct vnode *dvp = v->a_dvp;
1282	struct vnode **vpp = v->a_vpp;
1283	struct componentname *cnp = v->a_cnp;
1284	struct vattr *vap = v->a_vap;
1285	char *target = v->a_target;
1286
1287#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1288	MPASS(vap->va_type == VLNK);
1289#else
1290	vap->va_type = VLNK;
1291#endif
1292
1293	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1294}
1295
1296/* --------------------------------------------------------------------- */
1297
1298static int
1299tmpfs_readdir(struct vop_readdir_args *v)
1300{
1301	struct vnode *vp = v->a_vp;
1302	struct uio *uio = v->a_uio;
1303	int *eofflag = v->a_eofflag;
1304	u_long **cookies = v->a_cookies;
1305	int *ncookies = v->a_ncookies;
1306
1307	int error;
1308	off_t startoff;
1309	off_t cnt = 0;
1310	struct tmpfs_node *node;
1311
1312	/* This operation only makes sense on directory nodes. */
1313	if (vp->v_type != VDIR)
1314		return ENOTDIR;
1315
1316	node = VP_TO_TMPFS_DIR(vp);
1317
1318	startoff = uio->uio_offset;
1319
1320	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1321		error = tmpfs_dir_getdotdent(node, uio);
1322		if (error != 0)
1323			goto outok;
1324		cnt++;
1325	}
1326
1327	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1328		error = tmpfs_dir_getdotdotdent(node, uio);
1329		if (error != 0)
1330			goto outok;
1331		cnt++;
1332	}
1333
1334	error = tmpfs_dir_getdents(node, uio, &cnt);
1335
1336outok:
1337	MPASS(error >= -1);
1338
1339	if (error == -1)
1340		error = (cnt != 0) ? 0 : EINVAL;
1341
1342	if (eofflag != NULL)
1343		*eofflag =
1344		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1345
1346	/* Update NFS-related variables. */
1347	if (error == 0 && cookies != NULL && ncookies != NULL) {
1348		off_t i;
1349		off_t off = startoff;
1350		struct tmpfs_dirent *de = NULL;
1351
1352		*ncookies = cnt;
1353		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1354
1355		for (i = 0; i < cnt; i++) {
1356			MPASS(off != TMPFS_DIRCOOKIE_EOF);
1357			if (off == TMPFS_DIRCOOKIE_DOT) {
1358				off = TMPFS_DIRCOOKIE_DOTDOT;
1359			} else {
1360				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1361					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1362				} else if (de != NULL) {
1363					de = TAILQ_NEXT(de, td_entries);
1364				} else {
1365					de = tmpfs_dir_lookupbycookie(node,
1366					    off);
1367					MPASS(de != NULL);
1368					de = TAILQ_NEXT(de, td_entries);
1369				}
1370				if (de == NULL)
1371					off = TMPFS_DIRCOOKIE_EOF;
1372				else
1373					off = tmpfs_dircookie(de);
1374			}
1375
1376			(*cookies)[i] = off;
1377		}
1378		MPASS(uio->uio_offset == off);
1379	}
1380
1381	return error;
1382}
1383
1384/* --------------------------------------------------------------------- */
1385
1386static int
1387tmpfs_readlink(struct vop_readlink_args *v)
1388{
1389	struct vnode *vp = v->a_vp;
1390	struct uio *uio = v->a_uio;
1391
1392	int error;
1393	struct tmpfs_node *node;
1394
1395	MPASS(uio->uio_offset == 0);
1396	MPASS(vp->v_type == VLNK);
1397
1398	node = VP_TO_TMPFS_NODE(vp);
1399
1400	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1401	    uio);
1402	node->tn_status |= TMPFS_NODE_ACCESSED;
1403
1404	return error;
1405}
1406
1407/* --------------------------------------------------------------------- */
1408
1409static int
1410tmpfs_inactive(struct vop_inactive_args *v)
1411{
1412	struct vnode *vp = v->a_vp;
1413	struct thread *l = v->a_td;
1414
1415	struct tmpfs_node *node;
1416
1417	MPASS(VOP_ISLOCKED(vp));
1418
1419	node = VP_TO_TMPFS_NODE(vp);
1420
1421	if (node->tn_links == 0)
1422		vrecycle(vp, l);
1423
1424	return 0;
1425}
1426
1427/* --------------------------------------------------------------------- */
1428
1429int
1430tmpfs_reclaim(struct vop_reclaim_args *v)
1431{
1432	struct vnode *vp = v->a_vp;
1433
1434	struct tmpfs_mount *tmp;
1435	struct tmpfs_node *node;
1436
1437	node = VP_TO_TMPFS_NODE(vp);
1438	tmp = VFS_TO_TMPFS(vp->v_mount);
1439
1440	vnode_destroy_vobject(vp);
1441	cache_purge(vp);
1442
1443	TMPFS_NODE_LOCK(node);
1444	TMPFS_ASSERT_ELOCKED(node);
1445	tmpfs_free_vp(vp);
1446
1447	/* If the node referenced by this vnode was deleted by the user,
1448	 * we must free its associated data structures (now that the vnode
1449	 * is being reclaimed). */
1450	if (node->tn_links == 0 &&
1451	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1452		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1453		TMPFS_NODE_UNLOCK(node);
1454		tmpfs_free_node(tmp, node);
1455	} else
1456		TMPFS_NODE_UNLOCK(node);
1457
1458	MPASS(vp->v_data == NULL);
1459	return 0;
1460}
1461
1462/* --------------------------------------------------------------------- */
1463
1464static int
1465tmpfs_print(struct vop_print_args *v)
1466{
1467	struct vnode *vp = v->a_vp;
1468
1469	struct tmpfs_node *node;
1470
1471	node = VP_TO_TMPFS_NODE(vp);
1472
1473	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1474	    node, node->tn_flags, node->tn_links);
1475	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1476	    ", status 0x%x\n",
1477	    node->tn_mode, node->tn_uid, node->tn_gid,
1478	    (uintmax_t)node->tn_size, node->tn_status);
1479
1480	if (vp->v_type == VFIFO)
1481		fifo_printinfo(vp);
1482
1483	printf("\n");
1484
1485	return 0;
1486}
1487
1488/* --------------------------------------------------------------------- */
1489
1490static int
1491tmpfs_pathconf(struct vop_pathconf_args *v)
1492{
1493	int name = v->a_name;
1494	register_t *retval = v->a_retval;
1495
1496	int error;
1497
1498	error = 0;
1499
1500	switch (name) {
1501	case _PC_LINK_MAX:
1502		*retval = LINK_MAX;
1503		break;
1504
1505	case _PC_NAME_MAX:
1506		*retval = NAME_MAX;
1507		break;
1508
1509	case _PC_PATH_MAX:
1510		*retval = PATH_MAX;
1511		break;
1512
1513	case _PC_PIPE_BUF:
1514		*retval = PIPE_BUF;
1515		break;
1516
1517	case _PC_CHOWN_RESTRICTED:
1518		*retval = 1;
1519		break;
1520
1521	case _PC_NO_TRUNC:
1522		*retval = 1;
1523		break;
1524
1525	case _PC_SYNC_IO:
1526		*retval = 1;
1527		break;
1528
1529	case _PC_FILESIZEBITS:
1530		*retval = 0; /* XXX Don't know which value should I return. */
1531		break;
1532
1533	default:
1534		error = EINVAL;
1535	}
1536
1537	return error;
1538}
1539
1540static int
1541tmpfs_vptofh(struct vop_vptofh_args *ap)
1542{
1543	struct tmpfs_fid *tfhp;
1544	struct tmpfs_node *node;
1545
1546	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1547	node = VP_TO_TMPFS_NODE(ap->a_vp);
1548
1549	tfhp->tf_len = sizeof(struct tmpfs_fid);
1550	tfhp->tf_id = node->tn_id;
1551	tfhp->tf_gen = node->tn_gen;
1552
1553	return (0);
1554}
1555
1556static int
1557tmpfs_whiteout(struct vop_whiteout_args *ap)
1558{
1559	struct vnode *dvp = ap->a_dvp;
1560	struct componentname *cnp = ap->a_cnp;
1561	struct tmpfs_dirent *de;
1562
1563	switch (ap->a_flags) {
1564	case LOOKUP:
1565		return (0);
1566	case CREATE:
1567		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1568		if (de != NULL)
1569			return (de->td_node == NULL ? 0 : EEXIST);
1570		return (tmpfs_dir_whiteout_add(dvp, cnp));
1571	case DELETE:
1572		tmpfs_dir_whiteout_remove(dvp, cnp);
1573		return (0);
1574	default:
1575		panic("tmpfs_whiteout: unknown op");
1576	}
1577}
1578
1579/* --------------------------------------------------------------------- */
1580
1581/*
1582 * vnode operations vector used for files stored in a tmpfs file system.
1583 */
1584struct vop_vector tmpfs_vnodeop_entries = {
1585	.vop_default =			&default_vnodeops,
1586	.vop_lookup =			vfs_cache_lookup,
1587	.vop_cachedlookup =		tmpfs_lookup,
1588	.vop_create =			tmpfs_create,
1589	.vop_mknod =			tmpfs_mknod,
1590	.vop_open =			tmpfs_open,
1591	.vop_close =			tmpfs_close,
1592	.vop_access =			tmpfs_access,
1593	.vop_getattr =			tmpfs_getattr,
1594	.vop_setattr =			tmpfs_setattr,
1595	.vop_read =			tmpfs_read,
1596	.vop_write =			tmpfs_write,
1597	.vop_fsync =			tmpfs_fsync,
1598	.vop_remove =			tmpfs_remove,
1599	.vop_link =			tmpfs_link,
1600	.vop_rename =			tmpfs_rename,
1601	.vop_mkdir =			tmpfs_mkdir,
1602	.vop_rmdir =			tmpfs_rmdir,
1603	.vop_symlink =			tmpfs_symlink,
1604	.vop_readdir =			tmpfs_readdir,
1605	.vop_readlink =			tmpfs_readlink,
1606	.vop_inactive =			tmpfs_inactive,
1607	.vop_reclaim =			tmpfs_reclaim,
1608	.vop_print =			tmpfs_print,
1609	.vop_pathconf =			tmpfs_pathconf,
1610	.vop_vptofh =			tmpfs_vptofh,
1611	.vop_whiteout =			tmpfs_whiteout,
1612	.vop_bmap =			VOP_EOPNOTSUPP,
1613};
1614
1615