tmpfs_vnops.c revision 213735
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 213735 2010-10-12 17:16:51Z avg $");
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	struct tmpfs_node *node;
274
275	MPASS(VOP_ISLOCKED(vp));
276
277	node = VP_TO_TMPFS_NODE(vp);
278
279	if (node->tn_links > 0) {
280		/* Update node times.  No need to do it if the node has
281		 * been deleted, because it will vanish after we return. */
282		tmpfs_update(vp);
283	}
284
285	return 0;
286}
287
288/* --------------------------------------------------------------------- */
289
290int
291tmpfs_access(struct vop_access_args *v)
292{
293	struct vnode *vp = v->a_vp;
294	accmode_t accmode = v->a_accmode;
295	struct ucred *cred = v->a_cred;
296
297	int error;
298	struct tmpfs_node *node;
299
300	MPASS(VOP_ISLOCKED(vp));
301
302	node = VP_TO_TMPFS_NODE(vp);
303
304	switch (vp->v_type) {
305	case VDIR:
306		/* FALLTHROUGH */
307	case VLNK:
308		/* FALLTHROUGH */
309	case VREG:
310		if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
311			error = EROFS;
312			goto out;
313		}
314		break;
315
316	case VBLK:
317		/* FALLTHROUGH */
318	case VCHR:
319		/* FALLTHROUGH */
320	case VSOCK:
321		/* FALLTHROUGH */
322	case VFIFO:
323		break;
324
325	default:
326		error = EINVAL;
327		goto out;
328	}
329
330	if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
331		error = EPERM;
332		goto out;
333	}
334
335	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
336	    node->tn_gid, accmode, cred, NULL);
337
338out:
339	MPASS(VOP_ISLOCKED(vp));
340
341	return error;
342}
343
344/* --------------------------------------------------------------------- */
345
346int
347tmpfs_getattr(struct vop_getattr_args *v)
348{
349	struct vnode *vp = v->a_vp;
350	struct vattr *vap = v->a_vap;
351
352	struct tmpfs_node *node;
353
354	node = VP_TO_TMPFS_NODE(vp);
355
356	tmpfs_update(vp);
357
358	vap->va_type = vp->v_type;
359	vap->va_mode = node->tn_mode;
360	vap->va_nlink = node->tn_links;
361	vap->va_uid = node->tn_uid;
362	vap->va_gid = node->tn_gid;
363	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
364	vap->va_fileid = node->tn_id;
365	vap->va_size = node->tn_size;
366	vap->va_blocksize = PAGE_SIZE;
367	vap->va_atime = node->tn_atime;
368	vap->va_mtime = node->tn_mtime;
369	vap->va_ctime = node->tn_ctime;
370	vap->va_birthtime = node->tn_birthtime;
371	vap->va_gen = node->tn_gen;
372	vap->va_flags = node->tn_flags;
373	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
374		node->tn_rdev : NODEV;
375	vap->va_bytes = round_page(node->tn_size);
376	vap->va_filerev = 0;
377
378	return 0;
379}
380
381/* --------------------------------------------------------------------- */
382
383/* XXX Should this operation be atomic?  I think it should, but code in
384 * XXX other places (e.g., ufs) doesn't seem to be... */
385int
386tmpfs_setattr(struct vop_setattr_args *v)
387{
388	struct vnode *vp = v->a_vp;
389	struct vattr *vap = v->a_vap;
390	struct ucred *cred = v->a_cred;
391	struct thread *td = curthread;
392
393	int error;
394
395	MPASS(VOP_ISLOCKED(vp));
396
397	error = 0;
398
399	/* Abort if any unsettable attribute is given. */
400	if (vap->va_type != VNON ||
401	    vap->va_nlink != VNOVAL ||
402	    vap->va_fsid != VNOVAL ||
403	    vap->va_fileid != VNOVAL ||
404	    vap->va_blocksize != VNOVAL ||
405	    vap->va_gen != VNOVAL ||
406	    vap->va_rdev != VNOVAL ||
407	    vap->va_bytes != VNOVAL)
408		error = EINVAL;
409
410	if (error == 0 && (vap->va_flags != VNOVAL))
411		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
412
413	if (error == 0 && (vap->va_size != VNOVAL))
414		error = tmpfs_chsize(vp, vap->va_size, cred, td);
415
416	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
417		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
418
419	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
420		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
421
422	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
423	    vap->va_atime.tv_nsec != VNOVAL) ||
424	    (vap->va_mtime.tv_sec != VNOVAL &&
425	    vap->va_mtime.tv_nsec != VNOVAL) ||
426	    (vap->va_birthtime.tv_sec != VNOVAL &&
427	    vap->va_birthtime.tv_nsec != VNOVAL)))
428		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
429			&vap->va_birthtime, vap->va_vaflags, cred, td);
430
431	/* Update the node times.  We give preference to the error codes
432	 * generated by this function rather than the ones that may arise
433	 * from tmpfs_update. */
434	tmpfs_update(vp);
435
436	MPASS(VOP_ISLOCKED(vp));
437
438	return error;
439}
440
441/* --------------------------------------------------------------------- */
442static int
443tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
444    vm_offset_t offset, size_t tlen, struct uio *uio)
445{
446	vm_page_t	m;
447	int		error;
448
449	VM_OBJECT_LOCK(tobj);
450	vm_object_pip_add(tobj, 1);
451	m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
452	    VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
453	if (m->valid != VM_PAGE_BITS_ALL) {
454		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
455			error = vm_pager_get_pages(tobj, &m, 1, 0);
456			if (error != 0) {
457				printf("tmpfs get pages from pager error [read]\n");
458				goto out;
459			}
460		} else
461			vm_page_zero_invalid(m, TRUE);
462	}
463	VM_OBJECT_UNLOCK(tobj);
464	error = uiomove_fromphys(&m, offset, tlen, uio);
465	VM_OBJECT_LOCK(tobj);
466out:
467	vm_page_lock(m);
468	vm_page_unwire(m, TRUE);
469	vm_page_unlock(m);
470	vm_page_wakeup(m);
471	vm_object_pip_subtract(tobj, 1);
472	VM_OBJECT_UNLOCK(tobj);
473
474	return (error);
475}
476
477static __inline int
478tmpfs_nocacheread_buf(vm_object_t tobj, vm_pindex_t idx,
479    vm_offset_t offset, size_t tlen, void *buf)
480{
481	struct uio uio;
482	struct iovec iov;
483
484	uio.uio_iovcnt = 1;
485	uio.uio_iov = &iov;
486	iov.iov_base = buf;
487	iov.iov_len = tlen;
488
489	uio.uio_offset = 0;
490	uio.uio_resid = tlen;
491	uio.uio_rw = UIO_READ;
492	uio.uio_segflg = UIO_SYSSPACE;
493	uio.uio_td = curthread;
494
495	return (tmpfs_nocacheread(tobj, idx, offset, tlen, &uio));
496}
497
498static int
499tmpfs_mappedread(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
500{
501	struct sf_buf	*sf;
502	vm_pindex_t	idx;
503	vm_page_t	m;
504	vm_offset_t	offset;
505	off_t		addr;
506	size_t		tlen;
507	char		*ma;
508	int		error;
509
510	addr = uio->uio_offset;
511	idx = OFF_TO_IDX(addr);
512	offset = addr & PAGE_MASK;
513	tlen = MIN(PAGE_SIZE - offset, len);
514
515	if ((vobj == NULL) ||
516	    (vobj->resident_page_count == 0 && vobj->cache == NULL))
517		goto nocache;
518
519	VM_OBJECT_LOCK(vobj);
520lookupvpg:
521	if (((m = vm_page_lookup(vobj, idx)) != NULL) &&
522	    vm_page_is_valid(m, offset, tlen)) {
523		if ((m->oflags & VPO_BUSY) != 0) {
524			/*
525			 * Reference the page before unlocking and sleeping so
526			 * that the page daemon is less likely to reclaim it.
527			 */
528			vm_page_lock_queues();
529			vm_page_flag_set(m, PG_REFERENCED);
530			vm_page_sleep(m, "tmfsmr");
531			goto lookupvpg;
532		}
533		vm_page_busy(m);
534		VM_OBJECT_UNLOCK(vobj);
535		error = uiomove_fromphys(&m, offset, tlen, uio);
536		VM_OBJECT_LOCK(vobj);
537		vm_page_wakeup(m);
538		VM_OBJECT_UNLOCK(vobj);
539		return	(error);
540	} else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) {
541		KASSERT(offset == 0,
542		    ("unexpected offset in tmpfs_mappedread for sendfile"));
543		if ((m->oflags & VPO_BUSY) != 0) {
544			/*
545			 * Reference the page before unlocking and sleeping so
546			 * that the page daemon is less likely to reclaim it.
547			 */
548			vm_page_lock_queues();
549			vm_page_flag_set(m, PG_REFERENCED);
550			vm_page_sleep(m, "tmfsmr");
551			goto lookupvpg;
552		}
553		vm_page_busy(m);
554		VM_OBJECT_UNLOCK(vobj);
555		sched_pin();
556		sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
557		ma = (char *)sf_buf_kva(sf);
558		error = tmpfs_nocacheread_buf(tobj, idx, 0, tlen, ma);
559		if (error == 0) {
560			if (tlen != PAGE_SIZE)
561				bzero(ma + tlen, PAGE_SIZE - tlen);
562			uio->uio_offset += tlen;
563			uio->uio_resid -= tlen;
564		}
565		sf_buf_free(sf);
566		sched_unpin();
567		VM_OBJECT_LOCK(vobj);
568		if (error == 0)
569			m->valid = VM_PAGE_BITS_ALL;
570		vm_page_wakeup(m);
571		VM_OBJECT_UNLOCK(vobj);
572		return	(error);
573	}
574	VM_OBJECT_UNLOCK(vobj);
575nocache:
576	error = tmpfs_nocacheread(tobj, idx, offset, tlen, uio);
577
578	return	(error);
579}
580
581static int
582tmpfs_read(struct vop_read_args *v)
583{
584	struct vnode *vp = v->a_vp;
585	struct uio *uio = v->a_uio;
586
587	struct tmpfs_node *node;
588	vm_object_t uobj;
589	size_t len;
590	int resid;
591
592	int error = 0;
593
594	node = VP_TO_TMPFS_NODE(vp);
595
596	if (vp->v_type != VREG) {
597		error = EISDIR;
598		goto out;
599	}
600
601	if (uio->uio_offset < 0) {
602		error = EINVAL;
603		goto out;
604	}
605
606	node->tn_status |= TMPFS_NODE_ACCESSED;
607
608	uobj = node->tn_reg.tn_aobj;
609	while ((resid = uio->uio_resid) > 0) {
610		error = 0;
611		if (node->tn_size <= uio->uio_offset)
612			break;
613		len = MIN(node->tn_size - uio->uio_offset, resid);
614		if (len == 0)
615			break;
616		error = tmpfs_mappedread(vp->v_object, uobj, len, uio);
617		if ((error != 0) || (resid == uio->uio_resid))
618			break;
619	}
620
621out:
622
623	return error;
624}
625
626/* --------------------------------------------------------------------- */
627
628static int
629tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
630{
631	vm_pindex_t	idx;
632	vm_page_t	vpg, tpg;
633	vm_offset_t	offset;
634	off_t		addr;
635	size_t		tlen;
636	int		error;
637
638	error = 0;
639
640	addr = uio->uio_offset;
641	idx = OFF_TO_IDX(addr);
642	offset = addr & PAGE_MASK;
643	tlen = MIN(PAGE_SIZE - offset, len);
644
645	if ((vobj == NULL) ||
646	    (vobj->resident_page_count == 0 && vobj->cache == NULL)) {
647		vpg = NULL;
648		goto nocache;
649	}
650
651	VM_OBJECT_LOCK(vobj);
652lookupvpg:
653	if (((vpg = vm_page_lookup(vobj, idx)) != NULL) &&
654	    vm_page_is_valid(vpg, offset, tlen)) {
655		if ((vpg->oflags & VPO_BUSY) != 0) {
656			/*
657			 * Reference the page before unlocking and sleeping so
658			 * that the page daemon is less likely to reclaim it.
659			 */
660			vm_page_lock_queues();
661			vm_page_flag_set(vpg, PG_REFERENCED);
662			vm_page_sleep(vpg, "tmfsmw");
663			goto lookupvpg;
664		}
665		vm_page_busy(vpg);
666		vm_page_undirty(vpg);
667		VM_OBJECT_UNLOCK(vobj);
668		error = uiomove_fromphys(&vpg, offset, tlen, uio);
669	} else {
670		if (__predict_false(vobj->cache != NULL))
671			vm_page_cache_free(vobj, idx, idx + 1);
672		VM_OBJECT_UNLOCK(vobj);
673		vpg = NULL;
674	}
675nocache:
676	VM_OBJECT_LOCK(tobj);
677	vm_object_pip_add(tobj, 1);
678	tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
679	    VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
680	if (tpg->valid != VM_PAGE_BITS_ALL) {
681		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
682			error = vm_pager_get_pages(tobj, &tpg, 1, 0);
683			if (error != 0) {
684				printf("tmpfs get pages from pager error [write]\n");
685				goto out;
686			}
687		} else
688			vm_page_zero_invalid(tpg, TRUE);
689	}
690	VM_OBJECT_UNLOCK(tobj);
691	if (vpg == NULL)
692		error = uiomove_fromphys(&tpg, offset, tlen, uio);
693	else {
694		KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid"));
695		pmap_copy_page(vpg, tpg);
696	}
697	VM_OBJECT_LOCK(tobj);
698out:
699	if (vobj != NULL)
700		VM_OBJECT_LOCK(vobj);
701	if (error == 0) {
702		KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
703		    ("parts of tpg invalid"));
704		vm_page_dirty(tpg);
705	}
706	vm_page_lock(tpg);
707	vm_page_unwire(tpg, TRUE);
708	vm_page_unlock(tpg);
709	vm_page_wakeup(tpg);
710	if (vpg != NULL)
711		vm_page_wakeup(vpg);
712	if (vobj != NULL)
713		VM_OBJECT_UNLOCK(vobj);
714	vm_object_pip_subtract(tobj, 1);
715	VM_OBJECT_UNLOCK(tobj);
716
717	return	(error);
718}
719
720static int
721tmpfs_write(struct vop_write_args *v)
722{
723	struct vnode *vp = v->a_vp;
724	struct uio *uio = v->a_uio;
725	int ioflag = v->a_ioflag;
726
727	boolean_t extended;
728	int error = 0;
729	off_t oldsize;
730	struct tmpfs_node *node;
731	vm_object_t uobj;
732	size_t len;
733	int resid;
734
735	node = VP_TO_TMPFS_NODE(vp);
736	oldsize = node->tn_size;
737
738	if (uio->uio_offset < 0 || vp->v_type != VREG) {
739		error = EINVAL;
740		goto out;
741	}
742
743	if (uio->uio_resid == 0) {
744		error = 0;
745		goto out;
746	}
747
748	if (ioflag & IO_APPEND)
749		uio->uio_offset = node->tn_size;
750
751	if (uio->uio_offset + uio->uio_resid >
752	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
753		return (EFBIG);
754
755	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
756		return (EFBIG);
757
758	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
759	if (extended) {
760		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
761		if (error != 0)
762			goto out;
763	}
764
765	uobj = node->tn_reg.tn_aobj;
766	while ((resid = uio->uio_resid) > 0) {
767		if (node->tn_size <= uio->uio_offset)
768			break;
769		len = MIN(node->tn_size - uio->uio_offset, resid);
770		if (len == 0)
771			break;
772		error = tmpfs_mappedwrite(vp->v_object, uobj, len, uio);
773		if ((error != 0) || (resid == uio->uio_resid))
774			break;
775	}
776
777	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
778	    (extended ? TMPFS_NODE_CHANGED : 0);
779
780	if (node->tn_mode & (S_ISUID | S_ISGID)) {
781		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
782			node->tn_mode &= ~(S_ISUID | S_ISGID);
783	}
784
785	if (error != 0)
786		(void)tmpfs_reg_resize(vp, oldsize);
787
788out:
789	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
790	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
791
792	return error;
793}
794
795/* --------------------------------------------------------------------- */
796
797static int
798tmpfs_fsync(struct vop_fsync_args *v)
799{
800	struct vnode *vp = v->a_vp;
801
802	MPASS(VOP_ISLOCKED(vp));
803
804	tmpfs_update(vp);
805
806	return 0;
807}
808
809/* --------------------------------------------------------------------- */
810
811static int
812tmpfs_remove(struct vop_remove_args *v)
813{
814	struct vnode *dvp = v->a_dvp;
815	struct vnode *vp = v->a_vp;
816
817	int error;
818	struct tmpfs_dirent *de;
819	struct tmpfs_mount *tmp;
820	struct tmpfs_node *dnode;
821	struct tmpfs_node *node;
822
823	MPASS(VOP_ISLOCKED(dvp));
824	MPASS(VOP_ISLOCKED(vp));
825
826	if (vp->v_type == VDIR) {
827		error = EISDIR;
828		goto out;
829	}
830
831	dnode = VP_TO_TMPFS_DIR(dvp);
832	node = VP_TO_TMPFS_NODE(vp);
833	tmp = VFS_TO_TMPFS(vp->v_mount);
834	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
835	MPASS(de != NULL);
836
837	/* Files marked as immutable or append-only cannot be deleted. */
838	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
839	    (dnode->tn_flags & APPEND)) {
840		error = EPERM;
841		goto out;
842	}
843
844	/* Remove the entry from the directory; as it is a file, we do not
845	 * have to change the number of hard links of the directory. */
846	tmpfs_dir_detach(dvp, de);
847	if (v->a_cnp->cn_flags & DOWHITEOUT)
848		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
849
850	/* Free the directory entry we just deleted.  Note that the node
851	 * referred by it will not be removed until the vnode is really
852	 * reclaimed. */
853	tmpfs_free_dirent(tmp, de, TRUE);
854
855	if (node->tn_links > 0)
856		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
857	    TMPFS_NODE_MODIFIED;
858	error = 0;
859
860out:
861
862	return error;
863}
864
865/* --------------------------------------------------------------------- */
866
867static int
868tmpfs_link(struct vop_link_args *v)
869{
870	struct vnode *dvp = v->a_tdvp;
871	struct vnode *vp = v->a_vp;
872	struct componentname *cnp = v->a_cnp;
873
874	int error;
875	struct tmpfs_dirent *de;
876	struct tmpfs_node *node;
877
878	MPASS(VOP_ISLOCKED(dvp));
879	MPASS(cnp->cn_flags & HASBUF);
880	MPASS(dvp != vp); /* XXX When can this be false? */
881
882	node = VP_TO_TMPFS_NODE(vp);
883
884	/* XXX: Why aren't the following two tests done by the caller? */
885
886	/* Hard links of directories are forbidden. */
887	if (vp->v_type == VDIR) {
888		error = EPERM;
889		goto out;
890	}
891
892	/* Cannot create cross-device links. */
893	if (dvp->v_mount != vp->v_mount) {
894		error = EXDEV;
895		goto out;
896	}
897
898	/* Ensure that we do not overflow the maximum number of links imposed
899	 * by the system. */
900	MPASS(node->tn_links <= LINK_MAX);
901	if (node->tn_links == LINK_MAX) {
902		error = EMLINK;
903		goto out;
904	}
905
906	/* We cannot create links of files marked immutable or append-only. */
907	if (node->tn_flags & (IMMUTABLE | APPEND)) {
908		error = EPERM;
909		goto out;
910	}
911
912	/* Allocate a new directory entry to represent the node. */
913	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
914	    cnp->cn_nameptr, cnp->cn_namelen, &de);
915	if (error != 0)
916		goto out;
917
918	/* Insert the new directory entry into the appropriate directory. */
919	if (cnp->cn_flags & ISWHITEOUT)
920		tmpfs_dir_whiteout_remove(dvp, cnp);
921	tmpfs_dir_attach(dvp, de);
922
923	/* vp link count has changed, so update node times. */
924	node->tn_status |= TMPFS_NODE_CHANGED;
925	tmpfs_update(vp);
926
927	error = 0;
928
929out:
930	return error;
931}
932
933/* --------------------------------------------------------------------- */
934
935static int
936tmpfs_rename(struct vop_rename_args *v)
937{
938	struct vnode *fdvp = v->a_fdvp;
939	struct vnode *fvp = v->a_fvp;
940	struct componentname *fcnp = v->a_fcnp;
941	struct vnode *tdvp = v->a_tdvp;
942	struct vnode *tvp = v->a_tvp;
943	struct componentname *tcnp = v->a_tcnp;
944
945	char *newname;
946	int error;
947	struct tmpfs_dirent *de;
948	struct tmpfs_mount *tmp;
949	struct tmpfs_node *fdnode;
950	struct tmpfs_node *fnode;
951	struct tmpfs_node *tnode;
952	struct tmpfs_node *tdnode;
953
954	MPASS(VOP_ISLOCKED(tdvp));
955	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
956	MPASS(fcnp->cn_flags & HASBUF);
957	MPASS(tcnp->cn_flags & HASBUF);
958
959  	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
960
961	/* Disallow cross-device renames.
962	 * XXX Why isn't this done by the caller? */
963	if (fvp->v_mount != tdvp->v_mount ||
964	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
965		error = EXDEV;
966		goto out;
967	}
968
969	tmp = VFS_TO_TMPFS(tdvp->v_mount);
970	tdnode = VP_TO_TMPFS_DIR(tdvp);
971
972	/* If source and target are the same file, there is nothing to do. */
973	if (fvp == tvp) {
974		error = 0;
975		goto out;
976	}
977
978	/* If we need to move the directory between entries, lock the
979	 * source so that we can safely operate on it. */
980	if (tdvp != fdvp) {
981		error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
982		if (error != 0)
983			goto out;
984	}
985	fdnode = VP_TO_TMPFS_DIR(fdvp);
986	fnode = VP_TO_TMPFS_NODE(fvp);
987	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
988
989	/* Entry can disappear before we lock fdvp,
990	 * also avoid manipulating '.' and '..' entries. */
991	if (de == NULL) {
992		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
993		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
994			error = EINVAL;
995		else
996			error = ENOENT;
997		goto out_locked;
998	}
999	MPASS(de->td_node == fnode);
1000
1001	/* If re-naming a directory to another preexisting directory
1002	 * ensure that the target directory is empty so that its
1003	 * removal causes no side effects.
1004	 * Kern_rename gurantees the destination to be a directory
1005	 * if the source is one. */
1006	if (tvp != NULL) {
1007		MPASS(tnode != NULL);
1008
1009		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1010		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1011			error = EPERM;
1012			goto out_locked;
1013		}
1014
1015		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1016			if (tnode->tn_size > 0) {
1017				error = ENOTEMPTY;
1018				goto out_locked;
1019			}
1020		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1021			error = ENOTDIR;
1022			goto out_locked;
1023		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1024			error = EISDIR;
1025			goto out_locked;
1026		} else {
1027			MPASS(fnode->tn_type != VDIR &&
1028				tnode->tn_type != VDIR);
1029		}
1030	}
1031
1032	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1033	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1034		error = EPERM;
1035		goto out_locked;
1036	}
1037
1038	/* Ensure that we have enough memory to hold the new name, if it
1039	 * has to be changed. */
1040	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1041	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1042		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1043	} else
1044		newname = NULL;
1045
1046	/* If the node is being moved to another directory, we have to do
1047	 * the move. */
1048	if (fdnode != tdnode) {
1049		/* In case we are moving a directory, we have to adjust its
1050		 * parent to point to the new parent. */
1051		if (de->td_node->tn_type == VDIR) {
1052			struct tmpfs_node *n;
1053
1054			/* Ensure the target directory is not a child of the
1055			 * directory being moved.  Otherwise, we'd end up
1056			 * with stale nodes. */
1057			n = tdnode;
1058			/* TMPFS_LOCK garanties that no nodes are freed while
1059			 * traversing the list. Nodes can only be marked as
1060			 * removed: tn_parent == NULL. */
1061			TMPFS_LOCK(tmp);
1062			TMPFS_NODE_LOCK(n);
1063			while (n != n->tn_dir.tn_parent) {
1064				struct tmpfs_node *parent;
1065
1066				if (n == fnode) {
1067					TMPFS_NODE_UNLOCK(n);
1068					TMPFS_UNLOCK(tmp);
1069					error = EINVAL;
1070					if (newname != NULL)
1071						    free(newname, M_TMPFSNAME);
1072					goto out_locked;
1073				}
1074				parent = n->tn_dir.tn_parent;
1075				TMPFS_NODE_UNLOCK(n);
1076				if (parent == NULL) {
1077					n = NULL;
1078					break;
1079				}
1080				TMPFS_NODE_LOCK(parent);
1081				if (parent->tn_dir.tn_parent == NULL) {
1082					TMPFS_NODE_UNLOCK(parent);
1083					n = NULL;
1084					break;
1085				}
1086				n = parent;
1087			}
1088			TMPFS_UNLOCK(tmp);
1089			if (n == NULL) {
1090				error = EINVAL;
1091				if (newname != NULL)
1092					    free(newname, M_TMPFSNAME);
1093				goto out_locked;
1094			}
1095			TMPFS_NODE_UNLOCK(n);
1096
1097			/* Adjust the parent pointer. */
1098			TMPFS_VALIDATE_DIR(fnode);
1099			TMPFS_NODE_LOCK(de->td_node);
1100			de->td_node->tn_dir.tn_parent = tdnode;
1101			TMPFS_NODE_UNLOCK(de->td_node);
1102
1103			/* As a result of changing the target of the '..'
1104			 * entry, the link count of the source and target
1105			 * directories has to be adjusted. */
1106			TMPFS_NODE_LOCK(tdnode);
1107			TMPFS_ASSERT_LOCKED(tdnode);
1108			tdnode->tn_links++;
1109			TMPFS_NODE_UNLOCK(tdnode);
1110
1111			TMPFS_NODE_LOCK(fdnode);
1112			TMPFS_ASSERT_LOCKED(fdnode);
1113			fdnode->tn_links--;
1114			TMPFS_NODE_UNLOCK(fdnode);
1115		}
1116
1117		/* Do the move: just remove the entry from the source directory
1118		 * and insert it into the target one. */
1119		tmpfs_dir_detach(fdvp, de);
1120		if (fcnp->cn_flags & DOWHITEOUT)
1121			tmpfs_dir_whiteout_add(fdvp, fcnp);
1122		if (tcnp->cn_flags & ISWHITEOUT)
1123			tmpfs_dir_whiteout_remove(tdvp, tcnp);
1124		tmpfs_dir_attach(tdvp, de);
1125	}
1126
1127	/* If the name has changed, we need to make it effective by changing
1128	 * it in the directory entry. */
1129	if (newname != NULL) {
1130		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1131
1132		free(de->td_name, M_TMPFSNAME);
1133		de->td_namelen = (uint16_t)tcnp->cn_namelen;
1134		memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen);
1135		de->td_name = newname;
1136
1137		fnode->tn_status |= TMPFS_NODE_CHANGED;
1138		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1139	}
1140
1141	/* If we are overwriting an entry, we have to remove the old one
1142	 * from the target directory. */
1143	if (tvp != NULL) {
1144		/* Remove the old entry from the target directory. */
1145		de = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1146		tmpfs_dir_detach(tdvp, de);
1147
1148		/* Free the directory entry we just deleted.  Note that the
1149		 * node referred by it will not be removed until the vnode is
1150		 * really reclaimed. */
1151		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE);
1152	}
1153
1154	error = 0;
1155
1156out_locked:
1157	if (fdnode != tdnode)
1158		VOP_UNLOCK(fdvp, 0);
1159
1160out:
1161	/* Release target nodes. */
1162	/* XXX: I don't understand when tdvp can be the same as tvp, but
1163	 * other code takes care of this... */
1164	if (tdvp == tvp)
1165		vrele(tdvp);
1166	else
1167		vput(tdvp);
1168	if (tvp != NULL)
1169		vput(tvp);
1170
1171	/* Release source nodes. */
1172	vrele(fdvp);
1173	vrele(fvp);
1174
1175	return error;
1176}
1177
1178/* --------------------------------------------------------------------- */
1179
1180static int
1181tmpfs_mkdir(struct vop_mkdir_args *v)
1182{
1183	struct vnode *dvp = v->a_dvp;
1184	struct vnode **vpp = v->a_vpp;
1185	struct componentname *cnp = v->a_cnp;
1186	struct vattr *vap = v->a_vap;
1187
1188	MPASS(vap->va_type == VDIR);
1189
1190	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1191}
1192
1193/* --------------------------------------------------------------------- */
1194
1195static int
1196tmpfs_rmdir(struct vop_rmdir_args *v)
1197{
1198	struct vnode *dvp = v->a_dvp;
1199	struct vnode *vp = v->a_vp;
1200
1201	int error;
1202	struct tmpfs_dirent *de;
1203	struct tmpfs_mount *tmp;
1204	struct tmpfs_node *dnode;
1205	struct tmpfs_node *node;
1206
1207	MPASS(VOP_ISLOCKED(dvp));
1208	MPASS(VOP_ISLOCKED(vp));
1209
1210	tmp = VFS_TO_TMPFS(dvp->v_mount);
1211	dnode = VP_TO_TMPFS_DIR(dvp);
1212	node = VP_TO_TMPFS_DIR(vp);
1213
1214	/* Directories with more than two entries ('.' and '..') cannot be
1215	 * removed. */
1216	 if (node->tn_size > 0) {
1217		 error = ENOTEMPTY;
1218		 goto out;
1219	 }
1220
1221	if ((dnode->tn_flags & APPEND)
1222	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1223		error = EPERM;
1224		goto out;
1225	}
1226
1227	/* This invariant holds only if we are not trying to remove "..".
1228	  * We checked for that above so this is safe now. */
1229	MPASS(node->tn_dir.tn_parent == dnode);
1230
1231	/* Get the directory entry associated with node (vp).  This was
1232	 * filled by tmpfs_lookup while looking up the entry. */
1233	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1234	MPASS(TMPFS_DIRENT_MATCHES(de,
1235	    v->a_cnp->cn_nameptr,
1236	    v->a_cnp->cn_namelen));
1237
1238	/* Check flags to see if we are allowed to remove the directory. */
1239	if (dnode->tn_flags & APPEND
1240		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1241		error = EPERM;
1242		goto out;
1243	}
1244
1245
1246	/* Detach the directory entry from the directory (dnode). */
1247	tmpfs_dir_detach(dvp, de);
1248	if (v->a_cnp->cn_flags & DOWHITEOUT)
1249		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1250
1251	/* No vnode should be allocated for this entry from this point */
1252	TMPFS_NODE_LOCK(node);
1253	TMPFS_ASSERT_ELOCKED(node);
1254	node->tn_links--;
1255	node->tn_dir.tn_parent = NULL;
1256	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1257	    TMPFS_NODE_MODIFIED;
1258
1259	TMPFS_NODE_UNLOCK(node);
1260
1261	TMPFS_NODE_LOCK(dnode);
1262	TMPFS_ASSERT_ELOCKED(dnode);
1263	dnode->tn_links--;
1264	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1265	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1266	TMPFS_NODE_UNLOCK(dnode);
1267
1268	cache_purge(dvp);
1269	cache_purge(vp);
1270
1271	/* Free the directory entry we just deleted.  Note that the node
1272	 * referred by it will not be removed until the vnode is really
1273	 * reclaimed. */
1274	tmpfs_free_dirent(tmp, de, TRUE);
1275
1276	/* Release the deleted vnode (will destroy the node, notify
1277	 * interested parties and clean it from the cache). */
1278
1279	dnode->tn_status |= TMPFS_NODE_CHANGED;
1280	tmpfs_update(dvp);
1281
1282	error = 0;
1283
1284out:
1285	return error;
1286}
1287
1288/* --------------------------------------------------------------------- */
1289
1290static int
1291tmpfs_symlink(struct vop_symlink_args *v)
1292{
1293	struct vnode *dvp = v->a_dvp;
1294	struct vnode **vpp = v->a_vpp;
1295	struct componentname *cnp = v->a_cnp;
1296	struct vattr *vap = v->a_vap;
1297	char *target = v->a_target;
1298
1299#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1300	MPASS(vap->va_type == VLNK);
1301#else
1302	vap->va_type = VLNK;
1303#endif
1304
1305	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1306}
1307
1308/* --------------------------------------------------------------------- */
1309
1310static int
1311tmpfs_readdir(struct vop_readdir_args *v)
1312{
1313	struct vnode *vp = v->a_vp;
1314	struct uio *uio = v->a_uio;
1315	int *eofflag = v->a_eofflag;
1316	u_long **cookies = v->a_cookies;
1317	int *ncookies = v->a_ncookies;
1318
1319	int error;
1320	off_t startoff;
1321	off_t cnt = 0;
1322	struct tmpfs_node *node;
1323
1324	/* This operation only makes sense on directory nodes. */
1325	if (vp->v_type != VDIR)
1326		return ENOTDIR;
1327
1328	node = VP_TO_TMPFS_DIR(vp);
1329
1330	startoff = uio->uio_offset;
1331
1332	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1333		error = tmpfs_dir_getdotdent(node, uio);
1334		if (error != 0)
1335			goto outok;
1336		cnt++;
1337	}
1338
1339	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1340		error = tmpfs_dir_getdotdotdent(node, uio);
1341		if (error != 0)
1342			goto outok;
1343		cnt++;
1344	}
1345
1346	error = tmpfs_dir_getdents(node, uio, &cnt);
1347
1348outok:
1349	MPASS(error >= -1);
1350
1351	if (error == -1)
1352		error = 0;
1353
1354	if (eofflag != NULL)
1355		*eofflag =
1356		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1357
1358	/* Update NFS-related variables. */
1359	if (error == 0 && cookies != NULL && ncookies != NULL) {
1360		off_t i;
1361		off_t off = startoff;
1362		struct tmpfs_dirent *de = NULL;
1363
1364		*ncookies = cnt;
1365		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1366
1367		for (i = 0; i < cnt; i++) {
1368			MPASS(off != TMPFS_DIRCOOKIE_EOF);
1369			if (off == TMPFS_DIRCOOKIE_DOT) {
1370				off = TMPFS_DIRCOOKIE_DOTDOT;
1371			} else {
1372				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1373					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1374				} else if (de != NULL) {
1375					de = TAILQ_NEXT(de, td_entries);
1376				} else {
1377					de = tmpfs_dir_lookupbycookie(node,
1378					    off);
1379					MPASS(de != NULL);
1380					de = TAILQ_NEXT(de, td_entries);
1381				}
1382				if (de == NULL)
1383					off = TMPFS_DIRCOOKIE_EOF;
1384				else
1385					off = tmpfs_dircookie(de);
1386			}
1387
1388			(*cookies)[i] = off;
1389		}
1390		MPASS(uio->uio_offset == off);
1391	}
1392
1393	return error;
1394}
1395
1396/* --------------------------------------------------------------------- */
1397
1398static int
1399tmpfs_readlink(struct vop_readlink_args *v)
1400{
1401	struct vnode *vp = v->a_vp;
1402	struct uio *uio = v->a_uio;
1403
1404	int error;
1405	struct tmpfs_node *node;
1406
1407	MPASS(uio->uio_offset == 0);
1408	MPASS(vp->v_type == VLNK);
1409
1410	node = VP_TO_TMPFS_NODE(vp);
1411
1412	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1413	    uio);
1414	node->tn_status |= TMPFS_NODE_ACCESSED;
1415
1416	return error;
1417}
1418
1419/* --------------------------------------------------------------------- */
1420
1421static int
1422tmpfs_inactive(struct vop_inactive_args *v)
1423{
1424	struct vnode *vp = v->a_vp;
1425	struct thread *l = v->a_td;
1426
1427	struct tmpfs_node *node;
1428
1429	MPASS(VOP_ISLOCKED(vp));
1430
1431	node = VP_TO_TMPFS_NODE(vp);
1432
1433	if (node->tn_links == 0)
1434		vrecycle(vp, l);
1435
1436	return 0;
1437}
1438
1439/* --------------------------------------------------------------------- */
1440
1441int
1442tmpfs_reclaim(struct vop_reclaim_args *v)
1443{
1444	struct vnode *vp = v->a_vp;
1445
1446	struct tmpfs_mount *tmp;
1447	struct tmpfs_node *node;
1448
1449	node = VP_TO_TMPFS_NODE(vp);
1450	tmp = VFS_TO_TMPFS(vp->v_mount);
1451
1452	vnode_destroy_vobject(vp);
1453	cache_purge(vp);
1454
1455	TMPFS_NODE_LOCK(node);
1456	TMPFS_ASSERT_ELOCKED(node);
1457	tmpfs_free_vp(vp);
1458
1459	/* If the node referenced by this vnode was deleted by the user,
1460	 * we must free its associated data structures (now that the vnode
1461	 * is being reclaimed). */
1462	if (node->tn_links == 0 &&
1463	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1464		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1465		TMPFS_NODE_UNLOCK(node);
1466		tmpfs_free_node(tmp, node);
1467	} else
1468		TMPFS_NODE_UNLOCK(node);
1469
1470	MPASS(vp->v_data == NULL);
1471	return 0;
1472}
1473
1474/* --------------------------------------------------------------------- */
1475
1476static int
1477tmpfs_print(struct vop_print_args *v)
1478{
1479	struct vnode *vp = v->a_vp;
1480
1481	struct tmpfs_node *node;
1482
1483	node = VP_TO_TMPFS_NODE(vp);
1484
1485	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1486	    node, node->tn_flags, node->tn_links);
1487	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1488	    ", status 0x%x\n",
1489	    node->tn_mode, node->tn_uid, node->tn_gid,
1490	    (uintmax_t)node->tn_size, node->tn_status);
1491
1492	if (vp->v_type == VFIFO)
1493		fifo_printinfo(vp);
1494
1495	printf("\n");
1496
1497	return 0;
1498}
1499
1500/* --------------------------------------------------------------------- */
1501
1502static int
1503tmpfs_pathconf(struct vop_pathconf_args *v)
1504{
1505	int name = v->a_name;
1506	register_t *retval = v->a_retval;
1507
1508	int error;
1509
1510	error = 0;
1511
1512	switch (name) {
1513	case _PC_LINK_MAX:
1514		*retval = LINK_MAX;
1515		break;
1516
1517	case _PC_NAME_MAX:
1518		*retval = NAME_MAX;
1519		break;
1520
1521	case _PC_PATH_MAX:
1522		*retval = PATH_MAX;
1523		break;
1524
1525	case _PC_PIPE_BUF:
1526		*retval = PIPE_BUF;
1527		break;
1528
1529	case _PC_CHOWN_RESTRICTED:
1530		*retval = 1;
1531		break;
1532
1533	case _PC_NO_TRUNC:
1534		*retval = 1;
1535		break;
1536
1537	case _PC_SYNC_IO:
1538		*retval = 1;
1539		break;
1540
1541	case _PC_FILESIZEBITS:
1542		*retval = 0; /* XXX Don't know which value should I return. */
1543		break;
1544
1545	default:
1546		error = EINVAL;
1547	}
1548
1549	return error;
1550}
1551
1552static int
1553tmpfs_vptofh(struct vop_vptofh_args *ap)
1554{
1555	struct tmpfs_fid *tfhp;
1556	struct tmpfs_node *node;
1557
1558	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1559	node = VP_TO_TMPFS_NODE(ap->a_vp);
1560
1561	tfhp->tf_len = sizeof(struct tmpfs_fid);
1562	tfhp->tf_id = node->tn_id;
1563	tfhp->tf_gen = node->tn_gen;
1564
1565	return (0);
1566}
1567
1568static int
1569tmpfs_whiteout(struct vop_whiteout_args *ap)
1570{
1571	struct vnode *dvp = ap->a_dvp;
1572	struct componentname *cnp = ap->a_cnp;
1573	struct tmpfs_dirent *de;
1574
1575	switch (ap->a_flags) {
1576	case LOOKUP:
1577		return (0);
1578	case CREATE:
1579		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1580		if (de != NULL)
1581			return (de->td_node == NULL ? 0 : EEXIST);
1582		return (tmpfs_dir_whiteout_add(dvp, cnp));
1583	case DELETE:
1584		tmpfs_dir_whiteout_remove(dvp, cnp);
1585		return (0);
1586	default:
1587		panic("tmpfs_whiteout: unknown op");
1588	}
1589}
1590
1591/* --------------------------------------------------------------------- */
1592
1593/*
1594 * vnode operations vector used for files stored in a tmpfs file system.
1595 */
1596struct vop_vector tmpfs_vnodeop_entries = {
1597	.vop_default =			&default_vnodeops,
1598	.vop_lookup =			vfs_cache_lookup,
1599	.vop_cachedlookup =		tmpfs_lookup,
1600	.vop_create =			tmpfs_create,
1601	.vop_mknod =			tmpfs_mknod,
1602	.vop_open =			tmpfs_open,
1603	.vop_close =			tmpfs_close,
1604	.vop_access =			tmpfs_access,
1605	.vop_getattr =			tmpfs_getattr,
1606	.vop_setattr =			tmpfs_setattr,
1607	.vop_read =			tmpfs_read,
1608	.vop_write =			tmpfs_write,
1609	.vop_fsync =			tmpfs_fsync,
1610	.vop_remove =			tmpfs_remove,
1611	.vop_link =			tmpfs_link,
1612	.vop_rename =			tmpfs_rename,
1613	.vop_mkdir =			tmpfs_mkdir,
1614	.vop_rmdir =			tmpfs_rmdir,
1615	.vop_symlink =			tmpfs_symlink,
1616	.vop_readdir =			tmpfs_readdir,
1617	.vop_readlink =			tmpfs_readlink,
1618	.vop_inactive =			tmpfs_inactive,
1619	.vop_reclaim =			tmpfs_reclaim,
1620	.vop_print =			tmpfs_print,
1621	.vop_pathconf =			tmpfs_pathconf,
1622	.vop_vptofh =			tmpfs_vptofh,
1623	.vop_whiteout =			tmpfs_whiteout,
1624	.vop_bmap =			VOP_EOPNOTSUPP,
1625};
1626
1627