tmpfs_vnops.c revision 212305
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 212305 2010-09-07 22:40:45Z ivoras $");
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		if ((m->oflags & VPO_BUSY) != 0) {
542			/*
543			 * Reference the page before unlocking and sleeping so
544			 * that the page daemon is less likely to reclaim it.
545			 */
546			vm_page_lock_queues();
547			vm_page_flag_set(m, PG_REFERENCED);
548			vm_page_sleep(m, "tmfsmr");
549			goto lookupvpg;
550		}
551		vm_page_busy(m);
552		VM_OBJECT_UNLOCK(vobj);
553		sched_pin();
554		sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
555		ma = (char *)sf_buf_kva(sf);
556		error = tmpfs_nocacheread_buf(tobj, idx, offset, tlen,
557		    ma + offset);
558		if (error == 0) {
559			uio->uio_offset += tlen;
560			uio->uio_resid -= tlen;
561		}
562		sf_buf_free(sf);
563		sched_unpin();
564		VM_OBJECT_LOCK(vobj);
565		vm_page_wakeup(m);
566		VM_OBJECT_UNLOCK(vobj);
567		return	(error);
568	}
569	VM_OBJECT_UNLOCK(vobj);
570nocache:
571	error = tmpfs_nocacheread(tobj, idx, offset, tlen, uio);
572
573	return	(error);
574}
575
576static int
577tmpfs_read(struct vop_read_args *v)
578{
579	struct vnode *vp = v->a_vp;
580	struct uio *uio = v->a_uio;
581
582	struct tmpfs_node *node;
583	vm_object_t uobj;
584	size_t len;
585	int resid;
586
587	int error = 0;
588
589	node = VP_TO_TMPFS_NODE(vp);
590
591	if (vp->v_type != VREG) {
592		error = EISDIR;
593		goto out;
594	}
595
596	if (uio->uio_offset < 0) {
597		error = EINVAL;
598		goto out;
599	}
600
601	node->tn_status |= TMPFS_NODE_ACCESSED;
602
603	uobj = node->tn_reg.tn_aobj;
604	while ((resid = uio->uio_resid) > 0) {
605		error = 0;
606		if (node->tn_size <= uio->uio_offset)
607			break;
608		len = MIN(node->tn_size - uio->uio_offset, resid);
609		if (len == 0)
610			break;
611		error = tmpfs_mappedread(vp->v_object, uobj, len, uio);
612		if ((error != 0) || (resid == uio->uio_resid))
613			break;
614	}
615
616out:
617
618	return error;
619}
620
621/* --------------------------------------------------------------------- */
622
623static int
624tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
625{
626	vm_pindex_t	idx;
627	vm_page_t	vpg, tpg;
628	vm_offset_t	offset;
629	off_t		addr;
630	size_t		tlen;
631	int		error;
632
633	error = 0;
634
635	addr = uio->uio_offset;
636	idx = OFF_TO_IDX(addr);
637	offset = addr & PAGE_MASK;
638	tlen = MIN(PAGE_SIZE - offset, len);
639
640	if ((vobj == NULL) ||
641	    (vobj->resident_page_count == 0 && vobj->cache == NULL)) {
642		vpg = NULL;
643		goto nocache;
644	}
645
646	VM_OBJECT_LOCK(vobj);
647lookupvpg:
648	if (((vpg = vm_page_lookup(vobj, idx)) != NULL) &&
649	    vm_page_is_valid(vpg, offset, tlen)) {
650		if ((vpg->oflags & VPO_BUSY) != 0) {
651			/*
652			 * Reference the page before unlocking and sleeping so
653			 * that the page daemon is less likely to reclaim it.
654			 */
655			vm_page_lock_queues();
656			vm_page_flag_set(vpg, PG_REFERENCED);
657			vm_page_sleep(vpg, "tmfsmw");
658			goto lookupvpg;
659		}
660		vm_page_busy(vpg);
661		vm_page_undirty(vpg);
662		VM_OBJECT_UNLOCK(vobj);
663		error = uiomove_fromphys(&vpg, offset, tlen, uio);
664	} else {
665		if (__predict_false(vobj->cache != NULL))
666			vm_page_cache_free(vobj, idx, idx + 1);
667		VM_OBJECT_UNLOCK(vobj);
668		vpg = NULL;
669	}
670nocache:
671	VM_OBJECT_LOCK(tobj);
672	vm_object_pip_add(tobj, 1);
673	tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
674	    VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
675	if (tpg->valid != VM_PAGE_BITS_ALL) {
676		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
677			error = vm_pager_get_pages(tobj, &tpg, 1, 0);
678			if (error != 0) {
679				printf("tmpfs get pages from pager error [write]\n");
680				goto out;
681			}
682		} else
683			vm_page_zero_invalid(tpg, TRUE);
684	}
685	VM_OBJECT_UNLOCK(tobj);
686	if (vpg == NULL)
687		error = uiomove_fromphys(&tpg, offset, tlen, uio);
688	else {
689		KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid"));
690		pmap_copy_page(vpg, tpg);
691	}
692	VM_OBJECT_LOCK(tobj);
693out:
694	if (vobj != NULL)
695		VM_OBJECT_LOCK(vobj);
696	if (error == 0) {
697		KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
698		    ("parts of tpg invalid"));
699		vm_page_dirty(tpg);
700	}
701	vm_page_lock(tpg);
702	vm_page_unwire(tpg, TRUE);
703	vm_page_unlock(tpg);
704	vm_page_wakeup(tpg);
705	if (vpg != NULL)
706		vm_page_wakeup(vpg);
707	if (vobj != NULL)
708		VM_OBJECT_UNLOCK(vobj);
709	vm_object_pip_subtract(tobj, 1);
710	VM_OBJECT_UNLOCK(tobj);
711
712	return	(error);
713}
714
715static int
716tmpfs_write(struct vop_write_args *v)
717{
718	struct vnode *vp = v->a_vp;
719	struct uio *uio = v->a_uio;
720	int ioflag = v->a_ioflag;
721
722	boolean_t extended;
723	int error = 0;
724	off_t oldsize;
725	struct tmpfs_node *node;
726	vm_object_t uobj;
727	size_t len;
728	int resid;
729
730	node = VP_TO_TMPFS_NODE(vp);
731	oldsize = node->tn_size;
732
733	if (uio->uio_offset < 0 || vp->v_type != VREG) {
734		error = EINVAL;
735		goto out;
736	}
737
738	if (uio->uio_resid == 0) {
739		error = 0;
740		goto out;
741	}
742
743	if (ioflag & IO_APPEND)
744		uio->uio_offset = node->tn_size;
745
746	if (uio->uio_offset + uio->uio_resid >
747	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
748		return (EFBIG);
749
750	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
751		return (EFBIG);
752
753	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
754	if (extended) {
755		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
756		if (error != 0)
757			goto out;
758	}
759
760	uobj = node->tn_reg.tn_aobj;
761	while ((resid = uio->uio_resid) > 0) {
762		if (node->tn_size <= uio->uio_offset)
763			break;
764		len = MIN(node->tn_size - uio->uio_offset, resid);
765		if (len == 0)
766			break;
767		error = tmpfs_mappedwrite(vp->v_object, uobj, len, uio);
768		if ((error != 0) || (resid == uio->uio_resid))
769			break;
770	}
771
772	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
773	    (extended ? TMPFS_NODE_CHANGED : 0);
774
775	if (node->tn_mode & (S_ISUID | S_ISGID)) {
776		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
777			node->tn_mode &= ~(S_ISUID | S_ISGID);
778	}
779
780	if (error != 0)
781		(void)tmpfs_reg_resize(vp, oldsize);
782
783out:
784	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
785	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
786
787	return error;
788}
789
790/* --------------------------------------------------------------------- */
791
792static int
793tmpfs_fsync(struct vop_fsync_args *v)
794{
795	struct vnode *vp = v->a_vp;
796
797	MPASS(VOP_ISLOCKED(vp));
798
799	tmpfs_update(vp);
800
801	return 0;
802}
803
804/* --------------------------------------------------------------------- */
805
806static int
807tmpfs_remove(struct vop_remove_args *v)
808{
809	struct vnode *dvp = v->a_dvp;
810	struct vnode *vp = v->a_vp;
811
812	int error;
813	struct tmpfs_dirent *de;
814	struct tmpfs_mount *tmp;
815	struct tmpfs_node *dnode;
816	struct tmpfs_node *node;
817
818	MPASS(VOP_ISLOCKED(dvp));
819	MPASS(VOP_ISLOCKED(vp));
820
821	if (vp->v_type == VDIR) {
822		error = EISDIR;
823		goto out;
824	}
825
826	dnode = VP_TO_TMPFS_DIR(dvp);
827	node = VP_TO_TMPFS_NODE(vp);
828	tmp = VFS_TO_TMPFS(vp->v_mount);
829	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
830	MPASS(de != NULL);
831
832	/* Files marked as immutable or append-only cannot be deleted. */
833	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
834	    (dnode->tn_flags & APPEND)) {
835		error = EPERM;
836		goto out;
837	}
838
839	/* Remove the entry from the directory; as it is a file, we do not
840	 * have to change the number of hard links of the directory. */
841	tmpfs_dir_detach(dvp, de);
842	if (v->a_cnp->cn_flags & DOWHITEOUT)
843		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
844
845	/* Free the directory entry we just deleted.  Note that the node
846	 * referred by it will not be removed until the vnode is really
847	 * reclaimed. */
848	tmpfs_free_dirent(tmp, de, TRUE);
849
850	if (node->tn_links > 0)
851		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
852	    TMPFS_NODE_MODIFIED;
853	error = 0;
854
855out:
856
857	return error;
858}
859
860/* --------------------------------------------------------------------- */
861
862static int
863tmpfs_link(struct vop_link_args *v)
864{
865	struct vnode *dvp = v->a_tdvp;
866	struct vnode *vp = v->a_vp;
867	struct componentname *cnp = v->a_cnp;
868
869	int error;
870	struct tmpfs_dirent *de;
871	struct tmpfs_node *node;
872
873	MPASS(VOP_ISLOCKED(dvp));
874	MPASS(cnp->cn_flags & HASBUF);
875	MPASS(dvp != vp); /* XXX When can this be false? */
876
877	node = VP_TO_TMPFS_NODE(vp);
878
879	/* XXX: Why aren't the following two tests done by the caller? */
880
881	/* Hard links of directories are forbidden. */
882	if (vp->v_type == VDIR) {
883		error = EPERM;
884		goto out;
885	}
886
887	/* Cannot create cross-device links. */
888	if (dvp->v_mount != vp->v_mount) {
889		error = EXDEV;
890		goto out;
891	}
892
893	/* Ensure that we do not overflow the maximum number of links imposed
894	 * by the system. */
895	MPASS(node->tn_links <= LINK_MAX);
896	if (node->tn_links == LINK_MAX) {
897		error = EMLINK;
898		goto out;
899	}
900
901	/* We cannot create links of files marked immutable or append-only. */
902	if (node->tn_flags & (IMMUTABLE | APPEND)) {
903		error = EPERM;
904		goto out;
905	}
906
907	/* Allocate a new directory entry to represent the node. */
908	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
909	    cnp->cn_nameptr, cnp->cn_namelen, &de);
910	if (error != 0)
911		goto out;
912
913	/* Insert the new directory entry into the appropriate directory. */
914	if (cnp->cn_flags & ISWHITEOUT)
915		tmpfs_dir_whiteout_remove(dvp, cnp);
916	tmpfs_dir_attach(dvp, de);
917
918	/* vp link count has changed, so update node times. */
919	node->tn_status |= TMPFS_NODE_CHANGED;
920	tmpfs_update(vp);
921
922	error = 0;
923
924out:
925	return error;
926}
927
928/* --------------------------------------------------------------------- */
929
930static int
931tmpfs_rename(struct vop_rename_args *v)
932{
933	struct vnode *fdvp = v->a_fdvp;
934	struct vnode *fvp = v->a_fvp;
935	struct componentname *fcnp = v->a_fcnp;
936	struct vnode *tdvp = v->a_tdvp;
937	struct vnode *tvp = v->a_tvp;
938	struct componentname *tcnp = v->a_tcnp;
939
940	char *newname;
941	int error;
942	struct tmpfs_dirent *de;
943	struct tmpfs_mount *tmp;
944	struct tmpfs_node *fdnode;
945	struct tmpfs_node *fnode;
946	struct tmpfs_node *tnode;
947	struct tmpfs_node *tdnode;
948
949	MPASS(VOP_ISLOCKED(tdvp));
950	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
951	MPASS(fcnp->cn_flags & HASBUF);
952	MPASS(tcnp->cn_flags & HASBUF);
953
954  	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
955
956	/* Disallow cross-device renames.
957	 * XXX Why isn't this done by the caller? */
958	if (fvp->v_mount != tdvp->v_mount ||
959	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
960		error = EXDEV;
961		goto out;
962	}
963
964	tmp = VFS_TO_TMPFS(tdvp->v_mount);
965	tdnode = VP_TO_TMPFS_DIR(tdvp);
966
967	/* If source and target are the same file, there is nothing to do. */
968	if (fvp == tvp) {
969		error = 0;
970		goto out;
971	}
972
973	/* If we need to move the directory between entries, lock the
974	 * source so that we can safely operate on it. */
975	if (tdvp != fdvp) {
976		error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
977		if (error != 0)
978			goto out;
979	}
980	fdnode = VP_TO_TMPFS_DIR(fdvp);
981	fnode = VP_TO_TMPFS_NODE(fvp);
982	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
983
984	/* Entry can disappear before we lock fdvp,
985	 * also avoid manipulating '.' and '..' entries. */
986	if (de == NULL) {
987		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
988		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
989			error = EINVAL;
990		else
991			error = ENOENT;
992		goto out_locked;
993	}
994	MPASS(de->td_node == fnode);
995
996	/* If re-naming a directory to another preexisting directory
997	 * ensure that the target directory is empty so that its
998	 * removal causes no side effects.
999	 * Kern_rename gurantees the destination to be a directory
1000	 * if the source is one. */
1001	if (tvp != NULL) {
1002		MPASS(tnode != NULL);
1003
1004		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1005		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1006			error = EPERM;
1007			goto out_locked;
1008		}
1009
1010		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1011			if (tnode->tn_size > 0) {
1012				error = ENOTEMPTY;
1013				goto out_locked;
1014			}
1015		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1016			error = ENOTDIR;
1017			goto out_locked;
1018		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1019			error = EISDIR;
1020			goto out_locked;
1021		} else {
1022			MPASS(fnode->tn_type != VDIR &&
1023				tnode->tn_type != VDIR);
1024		}
1025	}
1026
1027	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1028	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1029		error = EPERM;
1030		goto out_locked;
1031	}
1032
1033	/* Ensure that we have enough memory to hold the new name, if it
1034	 * has to be changed. */
1035	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1036	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1037		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1038	} else
1039		newname = NULL;
1040
1041	/* If the node is being moved to another directory, we have to do
1042	 * the move. */
1043	if (fdnode != tdnode) {
1044		/* In case we are moving a directory, we have to adjust its
1045		 * parent to point to the new parent. */
1046		if (de->td_node->tn_type == VDIR) {
1047			struct tmpfs_node *n;
1048
1049			/* Ensure the target directory is not a child of the
1050			 * directory being moved.  Otherwise, we'd end up
1051			 * with stale nodes. */
1052			n = tdnode;
1053			/* TMPFS_LOCK garanties that no nodes are freed while
1054			 * traversing the list. Nodes can only be marked as
1055			 * removed: tn_parent == NULL. */
1056			TMPFS_LOCK(tmp);
1057			TMPFS_NODE_LOCK(n);
1058			while (n != n->tn_dir.tn_parent) {
1059				struct tmpfs_node *parent;
1060
1061				if (n == fnode) {
1062					TMPFS_NODE_UNLOCK(n);
1063					TMPFS_UNLOCK(tmp);
1064					error = EINVAL;
1065					if (newname != NULL)
1066						    free(newname, M_TMPFSNAME);
1067					goto out_locked;
1068				}
1069				parent = n->tn_dir.tn_parent;
1070				TMPFS_NODE_UNLOCK(n);
1071				if (parent == NULL) {
1072					n = NULL;
1073					break;
1074				}
1075				TMPFS_NODE_LOCK(parent);
1076				if (parent->tn_dir.tn_parent == NULL) {
1077					TMPFS_NODE_UNLOCK(parent);
1078					n = NULL;
1079					break;
1080				}
1081				n = parent;
1082			}
1083			TMPFS_UNLOCK(tmp);
1084			if (n == NULL) {
1085				error = EINVAL;
1086				if (newname != NULL)
1087					    free(newname, M_TMPFSNAME);
1088				goto out_locked;
1089			}
1090			TMPFS_NODE_UNLOCK(n);
1091
1092			/* Adjust the parent pointer. */
1093			TMPFS_VALIDATE_DIR(fnode);
1094			TMPFS_NODE_LOCK(de->td_node);
1095			de->td_node->tn_dir.tn_parent = tdnode;
1096			TMPFS_NODE_UNLOCK(de->td_node);
1097
1098			/* As a result of changing the target of the '..'
1099			 * entry, the link count of the source and target
1100			 * directories has to be adjusted. */
1101			TMPFS_NODE_LOCK(tdnode);
1102			TMPFS_ASSERT_LOCKED(tdnode);
1103			tdnode->tn_links++;
1104			TMPFS_NODE_UNLOCK(tdnode);
1105
1106			TMPFS_NODE_LOCK(fdnode);
1107			TMPFS_ASSERT_LOCKED(fdnode);
1108			fdnode->tn_links--;
1109			TMPFS_NODE_UNLOCK(fdnode);
1110		}
1111
1112		/* Do the move: just remove the entry from the source directory
1113		 * and insert it into the target one. */
1114		tmpfs_dir_detach(fdvp, de);
1115		if (fcnp->cn_flags & DOWHITEOUT)
1116			tmpfs_dir_whiteout_add(fdvp, fcnp);
1117		if (tcnp->cn_flags & ISWHITEOUT)
1118			tmpfs_dir_whiteout_remove(tdvp, tcnp);
1119		tmpfs_dir_attach(tdvp, de);
1120	}
1121
1122	/* If the name has changed, we need to make it effective by changing
1123	 * it in the directory entry. */
1124	if (newname != NULL) {
1125		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1126
1127		free(de->td_name, M_TMPFSNAME);
1128		de->td_namelen = (uint16_t)tcnp->cn_namelen;
1129		memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen);
1130		de->td_name = newname;
1131
1132		fnode->tn_status |= TMPFS_NODE_CHANGED;
1133		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1134	}
1135
1136	/* If we are overwriting an entry, we have to remove the old one
1137	 * from the target directory. */
1138	if (tvp != NULL) {
1139		/* Remove the old entry from the target directory. */
1140		de = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1141		tmpfs_dir_detach(tdvp, de);
1142
1143		/* Free the directory entry we just deleted.  Note that the
1144		 * node referred by it will not be removed until the vnode is
1145		 * really reclaimed. */
1146		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE);
1147	}
1148
1149	error = 0;
1150
1151out_locked:
1152	if (fdnode != tdnode)
1153		VOP_UNLOCK(fdvp, 0);
1154
1155out:
1156	/* Release target nodes. */
1157	/* XXX: I don't understand when tdvp can be the same as tvp, but
1158	 * other code takes care of this... */
1159	if (tdvp == tvp)
1160		vrele(tdvp);
1161	else
1162		vput(tdvp);
1163	if (tvp != NULL)
1164		vput(tvp);
1165
1166	/* Release source nodes. */
1167	vrele(fdvp);
1168	vrele(fvp);
1169
1170	return error;
1171}
1172
1173/* --------------------------------------------------------------------- */
1174
1175static int
1176tmpfs_mkdir(struct vop_mkdir_args *v)
1177{
1178	struct vnode *dvp = v->a_dvp;
1179	struct vnode **vpp = v->a_vpp;
1180	struct componentname *cnp = v->a_cnp;
1181	struct vattr *vap = v->a_vap;
1182
1183	MPASS(vap->va_type == VDIR);
1184
1185	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1186}
1187
1188/* --------------------------------------------------------------------- */
1189
1190static int
1191tmpfs_rmdir(struct vop_rmdir_args *v)
1192{
1193	struct vnode *dvp = v->a_dvp;
1194	struct vnode *vp = v->a_vp;
1195
1196	int error;
1197	struct tmpfs_dirent *de;
1198	struct tmpfs_mount *tmp;
1199	struct tmpfs_node *dnode;
1200	struct tmpfs_node *node;
1201
1202	MPASS(VOP_ISLOCKED(dvp));
1203	MPASS(VOP_ISLOCKED(vp));
1204
1205	tmp = VFS_TO_TMPFS(dvp->v_mount);
1206	dnode = VP_TO_TMPFS_DIR(dvp);
1207	node = VP_TO_TMPFS_DIR(vp);
1208
1209	/* Directories with more than two entries ('.' and '..') cannot be
1210	 * removed. */
1211	 if (node->tn_size > 0) {
1212		 error = ENOTEMPTY;
1213		 goto out;
1214	 }
1215
1216	if ((dnode->tn_flags & APPEND)
1217	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1218		error = EPERM;
1219		goto out;
1220	}
1221
1222	/* This invariant holds only if we are not trying to remove "..".
1223	  * We checked for that above so this is safe now. */
1224	MPASS(node->tn_dir.tn_parent == dnode);
1225
1226	/* Get the directory entry associated with node (vp).  This was
1227	 * filled by tmpfs_lookup while looking up the entry. */
1228	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1229	MPASS(TMPFS_DIRENT_MATCHES(de,
1230	    v->a_cnp->cn_nameptr,
1231	    v->a_cnp->cn_namelen));
1232
1233	/* Check flags to see if we are allowed to remove the directory. */
1234	if (dnode->tn_flags & APPEND
1235		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1236		error = EPERM;
1237		goto out;
1238	}
1239
1240
1241	/* Detach the directory entry from the directory (dnode). */
1242	tmpfs_dir_detach(dvp, de);
1243	if (v->a_cnp->cn_flags & DOWHITEOUT)
1244		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1245
1246	/* No vnode should be allocated for this entry from this point */
1247	TMPFS_NODE_LOCK(node);
1248	TMPFS_ASSERT_ELOCKED(node);
1249	node->tn_links--;
1250	node->tn_dir.tn_parent = NULL;
1251	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1252	    TMPFS_NODE_MODIFIED;
1253
1254	TMPFS_NODE_UNLOCK(node);
1255
1256	TMPFS_NODE_LOCK(dnode);
1257	TMPFS_ASSERT_ELOCKED(dnode);
1258	dnode->tn_links--;
1259	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1260	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1261	TMPFS_NODE_UNLOCK(dnode);
1262
1263	cache_purge(dvp);
1264	cache_purge(vp);
1265
1266	/* Free the directory entry we just deleted.  Note that the node
1267	 * referred by it will not be removed until the vnode is really
1268	 * reclaimed. */
1269	tmpfs_free_dirent(tmp, de, TRUE);
1270
1271	/* Release the deleted vnode (will destroy the node, notify
1272	 * interested parties and clean it from the cache). */
1273
1274	dnode->tn_status |= TMPFS_NODE_CHANGED;
1275	tmpfs_update(dvp);
1276
1277	error = 0;
1278
1279out:
1280	return error;
1281}
1282
1283/* --------------------------------------------------------------------- */
1284
1285static int
1286tmpfs_symlink(struct vop_symlink_args *v)
1287{
1288	struct vnode *dvp = v->a_dvp;
1289	struct vnode **vpp = v->a_vpp;
1290	struct componentname *cnp = v->a_cnp;
1291	struct vattr *vap = v->a_vap;
1292	char *target = v->a_target;
1293
1294#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1295	MPASS(vap->va_type == VLNK);
1296#else
1297	vap->va_type = VLNK;
1298#endif
1299
1300	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1301}
1302
1303/* --------------------------------------------------------------------- */
1304
1305static int
1306tmpfs_readdir(struct vop_readdir_args *v)
1307{
1308	struct vnode *vp = v->a_vp;
1309	struct uio *uio = v->a_uio;
1310	int *eofflag = v->a_eofflag;
1311	u_long **cookies = v->a_cookies;
1312	int *ncookies = v->a_ncookies;
1313
1314	int error;
1315	off_t startoff;
1316	off_t cnt = 0;
1317	struct tmpfs_node *node;
1318
1319	/* This operation only makes sense on directory nodes. */
1320	if (vp->v_type != VDIR)
1321		return ENOTDIR;
1322
1323	node = VP_TO_TMPFS_DIR(vp);
1324
1325	startoff = uio->uio_offset;
1326
1327	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1328		error = tmpfs_dir_getdotdent(node, uio);
1329		if (error != 0)
1330			goto outok;
1331		cnt++;
1332	}
1333
1334	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1335		error = tmpfs_dir_getdotdotdent(node, uio);
1336		if (error != 0)
1337			goto outok;
1338		cnt++;
1339	}
1340
1341	error = tmpfs_dir_getdents(node, uio, &cnt);
1342
1343outok:
1344	MPASS(error >= -1);
1345
1346	if (error == -1)
1347		error = 0;
1348
1349	if (eofflag != NULL)
1350		*eofflag =
1351		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1352
1353	/* Update NFS-related variables. */
1354	if (error == 0 && cookies != NULL && ncookies != NULL) {
1355		off_t i;
1356		off_t off = startoff;
1357		struct tmpfs_dirent *de = NULL;
1358
1359		*ncookies = cnt;
1360		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1361
1362		for (i = 0; i < cnt; i++) {
1363			MPASS(off != TMPFS_DIRCOOKIE_EOF);
1364			if (off == TMPFS_DIRCOOKIE_DOT) {
1365				off = TMPFS_DIRCOOKIE_DOTDOT;
1366			} else {
1367				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1368					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1369				} else if (de != NULL) {
1370					de = TAILQ_NEXT(de, td_entries);
1371				} else {
1372					de = tmpfs_dir_lookupbycookie(node,
1373					    off);
1374					MPASS(de != NULL);
1375					de = TAILQ_NEXT(de, td_entries);
1376				}
1377				if (de == NULL)
1378					off = TMPFS_DIRCOOKIE_EOF;
1379				else
1380					off = tmpfs_dircookie(de);
1381			}
1382
1383			(*cookies)[i] = off;
1384		}
1385		MPASS(uio->uio_offset == off);
1386	}
1387
1388	return error;
1389}
1390
1391/* --------------------------------------------------------------------- */
1392
1393static int
1394tmpfs_readlink(struct vop_readlink_args *v)
1395{
1396	struct vnode *vp = v->a_vp;
1397	struct uio *uio = v->a_uio;
1398
1399	int error;
1400	struct tmpfs_node *node;
1401
1402	MPASS(uio->uio_offset == 0);
1403	MPASS(vp->v_type == VLNK);
1404
1405	node = VP_TO_TMPFS_NODE(vp);
1406
1407	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1408	    uio);
1409	node->tn_status |= TMPFS_NODE_ACCESSED;
1410
1411	return error;
1412}
1413
1414/* --------------------------------------------------------------------- */
1415
1416static int
1417tmpfs_inactive(struct vop_inactive_args *v)
1418{
1419	struct vnode *vp = v->a_vp;
1420	struct thread *l = v->a_td;
1421
1422	struct tmpfs_node *node;
1423
1424	MPASS(VOP_ISLOCKED(vp));
1425
1426	node = VP_TO_TMPFS_NODE(vp);
1427
1428	if (node->tn_links == 0)
1429		vrecycle(vp, l);
1430
1431	return 0;
1432}
1433
1434/* --------------------------------------------------------------------- */
1435
1436int
1437tmpfs_reclaim(struct vop_reclaim_args *v)
1438{
1439	struct vnode *vp = v->a_vp;
1440
1441	struct tmpfs_mount *tmp;
1442	struct tmpfs_node *node;
1443
1444	node = VP_TO_TMPFS_NODE(vp);
1445	tmp = VFS_TO_TMPFS(vp->v_mount);
1446
1447	vnode_destroy_vobject(vp);
1448	cache_purge(vp);
1449
1450	TMPFS_NODE_LOCK(node);
1451	TMPFS_ASSERT_ELOCKED(node);
1452	tmpfs_free_vp(vp);
1453
1454	/* If the node referenced by this vnode was deleted by the user,
1455	 * we must free its associated data structures (now that the vnode
1456	 * is being reclaimed). */
1457	if (node->tn_links == 0 &&
1458	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1459		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1460		TMPFS_NODE_UNLOCK(node);
1461		tmpfs_free_node(tmp, node);
1462	} else
1463		TMPFS_NODE_UNLOCK(node);
1464
1465	MPASS(vp->v_data == NULL);
1466	return 0;
1467}
1468
1469/* --------------------------------------------------------------------- */
1470
1471static int
1472tmpfs_print(struct vop_print_args *v)
1473{
1474	struct vnode *vp = v->a_vp;
1475
1476	struct tmpfs_node *node;
1477
1478	node = VP_TO_TMPFS_NODE(vp);
1479
1480	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1481	    node, node->tn_flags, node->tn_links);
1482	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1483	    ", status 0x%x\n",
1484	    node->tn_mode, node->tn_uid, node->tn_gid,
1485	    (uintmax_t)node->tn_size, node->tn_status);
1486
1487	if (vp->v_type == VFIFO)
1488		fifo_printinfo(vp);
1489
1490	printf("\n");
1491
1492	return 0;
1493}
1494
1495/* --------------------------------------------------------------------- */
1496
1497static int
1498tmpfs_pathconf(struct vop_pathconf_args *v)
1499{
1500	int name = v->a_name;
1501	register_t *retval = v->a_retval;
1502
1503	int error;
1504
1505	error = 0;
1506
1507	switch (name) {
1508	case _PC_LINK_MAX:
1509		*retval = LINK_MAX;
1510		break;
1511
1512	case _PC_NAME_MAX:
1513		*retval = NAME_MAX;
1514		break;
1515
1516	case _PC_PATH_MAX:
1517		*retval = PATH_MAX;
1518		break;
1519
1520	case _PC_PIPE_BUF:
1521		*retval = PIPE_BUF;
1522		break;
1523
1524	case _PC_CHOWN_RESTRICTED:
1525		*retval = 1;
1526		break;
1527
1528	case _PC_NO_TRUNC:
1529		*retval = 1;
1530		break;
1531
1532	case _PC_SYNC_IO:
1533		*retval = 1;
1534		break;
1535
1536	case _PC_FILESIZEBITS:
1537		*retval = 0; /* XXX Don't know which value should I return. */
1538		break;
1539
1540	default:
1541		error = EINVAL;
1542	}
1543
1544	return error;
1545}
1546
1547static int
1548tmpfs_vptofh(struct vop_vptofh_args *ap)
1549{
1550	struct tmpfs_fid *tfhp;
1551	struct tmpfs_node *node;
1552
1553	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1554	node = VP_TO_TMPFS_NODE(ap->a_vp);
1555
1556	tfhp->tf_len = sizeof(struct tmpfs_fid);
1557	tfhp->tf_id = node->tn_id;
1558	tfhp->tf_gen = node->tn_gen;
1559
1560	return (0);
1561}
1562
1563static int
1564tmpfs_whiteout(struct vop_whiteout_args *ap)
1565{
1566	struct vnode *dvp = ap->a_dvp;
1567	struct componentname *cnp = ap->a_cnp;
1568	struct tmpfs_dirent *de;
1569
1570	switch (ap->a_flags) {
1571	case LOOKUP:
1572		return (0);
1573	case CREATE:
1574		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1575		if (de != NULL)
1576			return (de->td_node == NULL ? 0 : EEXIST);
1577		return (tmpfs_dir_whiteout_add(dvp, cnp));
1578	case DELETE:
1579		tmpfs_dir_whiteout_remove(dvp, cnp);
1580		return (0);
1581	default:
1582		panic("tmpfs_whiteout: unknown op");
1583	}
1584}
1585
1586/* --------------------------------------------------------------------- */
1587
1588/*
1589 * vnode operations vector used for files stored in a tmpfs file system.
1590 */
1591struct vop_vector tmpfs_vnodeop_entries = {
1592	.vop_default =			&default_vnodeops,
1593	.vop_lookup =			vfs_cache_lookup,
1594	.vop_cachedlookup =		tmpfs_lookup,
1595	.vop_create =			tmpfs_create,
1596	.vop_mknod =			tmpfs_mknod,
1597	.vop_open =			tmpfs_open,
1598	.vop_close =			tmpfs_close,
1599	.vop_access =			tmpfs_access,
1600	.vop_getattr =			tmpfs_getattr,
1601	.vop_setattr =			tmpfs_setattr,
1602	.vop_read =			tmpfs_read,
1603	.vop_write =			tmpfs_write,
1604	.vop_fsync =			tmpfs_fsync,
1605	.vop_remove =			tmpfs_remove,
1606	.vop_link =			tmpfs_link,
1607	.vop_rename =			tmpfs_rename,
1608	.vop_mkdir =			tmpfs_mkdir,
1609	.vop_rmdir =			tmpfs_rmdir,
1610	.vop_symlink =			tmpfs_symlink,
1611	.vop_readdir =			tmpfs_readdir,
1612	.vop_readlink =			tmpfs_readlink,
1613	.vop_inactive =			tmpfs_inactive,
1614	.vop_reclaim =			tmpfs_reclaim,
1615	.vop_print =			tmpfs_print,
1616	.vop_pathconf =			tmpfs_pathconf,
1617	.vop_vptofh =			tmpfs_vptofh,
1618	.vop_whiteout =			tmpfs_whiteout,
1619	.vop_bmap =			VOP_EOPNOTSUPP,
1620};
1621
1622