tmpfs_vnops.c revision 231669
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 231669 2012-02-14 11:24:24Z tijl $");
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 <fs/fifofs/fifo.h>
58#include <fs/tmpfs/tmpfs_vnops.h>
59#include <fs/tmpfs/tmpfs.h>
60
61/* --------------------------------------------------------------------- */
62
63static int
64tmpfs_lookup(struct vop_cachedlookup_args *v)
65{
66	struct vnode *dvp = v->a_dvp;
67	struct vnode **vpp = v->a_vpp;
68	struct componentname *cnp = v->a_cnp;
69
70	int error;
71	struct tmpfs_dirent *de;
72	struct tmpfs_node *dnode;
73
74	dnode = VP_TO_TMPFS_DIR(dvp);
75	*vpp = NULLVP;
76
77	/* Check accessibility of requested node as a first step. */
78	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
79	if (error != 0)
80		goto out;
81
82	/* We cannot be requesting the parent directory of the root node. */
83	MPASS(IMPLIES(dnode->tn_type == VDIR &&
84	    dnode->tn_dir.tn_parent == dnode,
85	    !(cnp->cn_flags & ISDOTDOT)));
86
87	TMPFS_ASSERT_LOCKED(dnode);
88	if (dnode->tn_dir.tn_parent == NULL) {
89		error = ENOENT;
90		goto out;
91	}
92	if (cnp->cn_flags & ISDOTDOT) {
93		int ltype = 0;
94
95		ltype = VOP_ISLOCKED(dvp);
96		vhold(dvp);
97		VOP_UNLOCK(dvp, 0);
98		/* Allocate a new vnode on the matching entry. */
99		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
100		    cnp->cn_lkflags, vpp);
101
102		vn_lock(dvp, ltype | LK_RETRY);
103		vdrop(dvp);
104	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
105		VREF(dvp);
106		*vpp = dvp;
107		error = 0;
108	} else {
109		de = tmpfs_dir_lookup(dnode, NULL, cnp);
110		if (de != NULL && de->td_node == NULL)
111			cnp->cn_flags |= ISWHITEOUT;
112		if (de == NULL || de->td_node == NULL) {
113			/* The entry was not found in the directory.
114			 * This is OK if we are creating or renaming an
115			 * entry and are working on the last component of
116			 * the path name. */
117			if ((cnp->cn_flags & ISLASTCN) &&
118			    (cnp->cn_nameiop == CREATE || \
119			    cnp->cn_nameiop == RENAME ||
120			    (cnp->cn_nameiop == DELETE &&
121			    cnp->cn_flags & DOWHITEOUT &&
122			    cnp->cn_flags & ISWHITEOUT))) {
123				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
124				    cnp->cn_thread);
125				if (error != 0)
126					goto out;
127
128				/* Keep the component name in the buffer for
129				 * future uses. */
130				cnp->cn_flags |= SAVENAME;
131
132				error = EJUSTRETURN;
133			} else
134				error = ENOENT;
135		} else {
136			struct tmpfs_node *tnode;
137
138			/* The entry was found, so get its associated
139			 * tmpfs_node. */
140			tnode = de->td_node;
141
142			/* If we are not at the last path component and
143			 * found a non-directory or non-link entry (which
144			 * may itself be pointing to a directory), raise
145			 * an error. */
146			if ((tnode->tn_type != VDIR &&
147			    tnode->tn_type != VLNK) &&
148			    !(cnp->cn_flags & ISLASTCN)) {
149				error = ENOTDIR;
150				goto out;
151			}
152
153			/* If we are deleting or renaming the entry, keep
154			 * track of its tmpfs_dirent so that it can be
155			 * easily deleted later. */
156			if ((cnp->cn_flags & ISLASTCN) &&
157			    (cnp->cn_nameiop == DELETE ||
158			    cnp->cn_nameiop == RENAME)) {
159				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
160				    cnp->cn_thread);
161				if (error != 0)
162					goto out;
163
164				/* Allocate a new vnode on the matching entry. */
165				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
166						cnp->cn_lkflags, vpp);
167				if (error != 0)
168					goto out;
169
170				if ((dnode->tn_mode & S_ISTXT) &&
171				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) &&
172				  VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) {
173					error = EPERM;
174					vput(*vpp);
175					*vpp = NULL;
176					goto out;
177				}
178				cnp->cn_flags |= SAVENAME;
179			} else {
180				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
181						cnp->cn_lkflags, vpp);
182			}
183		}
184	}
185
186	/* Store the result of this lookup in the cache.  Avoid this if the
187	 * request was for creation, as it does not improve timings on
188	 * emprical tests. */
189	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
190		cache_enter(dvp, *vpp, cnp);
191
192out:
193	/* If there were no errors, *vpp cannot be null and it must be
194	 * locked. */
195	MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
196
197	return error;
198}
199
200/* --------------------------------------------------------------------- */
201
202static int
203tmpfs_create(struct vop_create_args *v)
204{
205	struct vnode *dvp = v->a_dvp;
206	struct vnode **vpp = v->a_vpp;
207	struct componentname *cnp = v->a_cnp;
208	struct vattr *vap = v->a_vap;
209
210	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
211
212	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
213}
214/* --------------------------------------------------------------------- */
215
216static int
217tmpfs_mknod(struct vop_mknod_args *v)
218{
219	struct vnode *dvp = v->a_dvp;
220	struct vnode **vpp = v->a_vpp;
221	struct componentname *cnp = v->a_cnp;
222	struct vattr *vap = v->a_vap;
223
224	if (vap->va_type != VBLK && vap->va_type != VCHR &&
225	    vap->va_type != VFIFO)
226		return EINVAL;
227
228	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
229}
230
231/* --------------------------------------------------------------------- */
232
233static int
234tmpfs_open(struct vop_open_args *v)
235{
236	struct vnode *vp = v->a_vp;
237	int mode = v->a_mode;
238
239	int error;
240	struct tmpfs_node *node;
241
242	MPASS(VOP_ISLOCKED(vp));
243
244	node = VP_TO_TMPFS_NODE(vp);
245
246	/* The file is still active but all its names have been removed
247	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
248	 * it is about to die. */
249	if (node->tn_links < 1)
250		return (ENOENT);
251
252	/* If the file is marked append-only, deny write requests. */
253	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
254		error = EPERM;
255	else {
256		error = 0;
257		vnode_create_vobject(vp, node->tn_size, v->a_td);
258	}
259
260	MPASS(VOP_ISLOCKED(vp));
261	return error;
262}
263
264/* --------------------------------------------------------------------- */
265
266static int
267tmpfs_close(struct vop_close_args *v)
268{
269	struct vnode *vp = v->a_vp;
270
271	MPASS(VOP_ISLOCKED(vp));
272
273	/* Update node times. */
274	tmpfs_update(vp);
275
276	return (0);
277}
278
279/* --------------------------------------------------------------------- */
280
281int
282tmpfs_access(struct vop_access_args *v)
283{
284	struct vnode *vp = v->a_vp;
285	accmode_t accmode = v->a_accmode;
286	struct ucred *cred = v->a_cred;
287
288	int error;
289	struct tmpfs_node *node;
290
291	MPASS(VOP_ISLOCKED(vp));
292
293	node = VP_TO_TMPFS_NODE(vp);
294
295	switch (vp->v_type) {
296	case VDIR:
297		/* FALLTHROUGH */
298	case VLNK:
299		/* FALLTHROUGH */
300	case VREG:
301		if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
302			error = EROFS;
303			goto out;
304		}
305		break;
306
307	case VBLK:
308		/* FALLTHROUGH */
309	case VCHR:
310		/* FALLTHROUGH */
311	case VSOCK:
312		/* FALLTHROUGH */
313	case VFIFO:
314		break;
315
316	default:
317		error = EINVAL;
318		goto out;
319	}
320
321	if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
322		error = EPERM;
323		goto out;
324	}
325
326	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
327	    node->tn_gid, accmode, cred, NULL);
328
329out:
330	MPASS(VOP_ISLOCKED(vp));
331
332	return error;
333}
334
335/* --------------------------------------------------------------------- */
336
337int
338tmpfs_getattr(struct vop_getattr_args *v)
339{
340	struct vnode *vp = v->a_vp;
341	struct vattr *vap = v->a_vap;
342
343	struct tmpfs_node *node;
344
345	node = VP_TO_TMPFS_NODE(vp);
346
347	tmpfs_update(vp);
348
349	vap->va_type = vp->v_type;
350	vap->va_mode = node->tn_mode;
351	vap->va_nlink = node->tn_links;
352	vap->va_uid = node->tn_uid;
353	vap->va_gid = node->tn_gid;
354	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
355	vap->va_fileid = node->tn_id;
356	vap->va_size = node->tn_size;
357	vap->va_blocksize = PAGE_SIZE;
358	vap->va_atime = node->tn_atime;
359	vap->va_mtime = node->tn_mtime;
360	vap->va_ctime = node->tn_ctime;
361	vap->va_birthtime = node->tn_birthtime;
362	vap->va_gen = node->tn_gen;
363	vap->va_flags = node->tn_flags;
364	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
365		node->tn_rdev : NODEV;
366	vap->va_bytes = round_page(node->tn_size);
367	vap->va_filerev = 0;
368
369	return 0;
370}
371
372/* --------------------------------------------------------------------- */
373
374/* XXX Should this operation be atomic?  I think it should, but code in
375 * XXX other places (e.g., ufs) doesn't seem to be... */
376int
377tmpfs_setattr(struct vop_setattr_args *v)
378{
379	struct vnode *vp = v->a_vp;
380	struct vattr *vap = v->a_vap;
381	struct ucred *cred = v->a_cred;
382	struct thread *td = curthread;
383
384	int error;
385
386	MPASS(VOP_ISLOCKED(vp));
387
388	error = 0;
389
390	/* Abort if any unsettable attribute is given. */
391	if (vap->va_type != VNON ||
392	    vap->va_nlink != VNOVAL ||
393	    vap->va_fsid != VNOVAL ||
394	    vap->va_fileid != VNOVAL ||
395	    vap->va_blocksize != VNOVAL ||
396	    vap->va_gen != VNOVAL ||
397	    vap->va_rdev != VNOVAL ||
398	    vap->va_bytes != VNOVAL)
399		error = EINVAL;
400
401	if (error == 0 && (vap->va_flags != VNOVAL))
402		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
403
404	if (error == 0 && (vap->va_size != VNOVAL))
405		error = tmpfs_chsize(vp, vap->va_size, cred, td);
406
407	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
408		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
409
410	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
411		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
412
413	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
414	    vap->va_atime.tv_nsec != VNOVAL) ||
415	    (vap->va_mtime.tv_sec != VNOVAL &&
416	    vap->va_mtime.tv_nsec != VNOVAL) ||
417	    (vap->va_birthtime.tv_sec != VNOVAL &&
418	    vap->va_birthtime.tv_nsec != VNOVAL)))
419		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
420			&vap->va_birthtime, vap->va_vaflags, cred, td);
421
422	/* Update the node times.  We give preference to the error codes
423	 * generated by this function rather than the ones that may arise
424	 * from tmpfs_update. */
425	tmpfs_update(vp);
426
427	MPASS(VOP_ISLOCKED(vp));
428
429	return error;
430}
431
432/* --------------------------------------------------------------------- */
433static int
434tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
435    vm_offset_t offset, size_t tlen, struct uio *uio)
436{
437	vm_page_t	m;
438	int		error, rv;
439
440	VM_OBJECT_LOCK(tobj);
441	m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
442	    VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
443	if (m->valid != VM_PAGE_BITS_ALL) {
444		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
445			rv = vm_pager_get_pages(tobj, &m, 1, 0);
446			if (rv != VM_PAGER_OK) {
447				vm_page_lock(m);
448				vm_page_free(m);
449				vm_page_unlock(m);
450				VM_OBJECT_UNLOCK(tobj);
451				return (EIO);
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);
459	vm_page_lock(m);
460	vm_page_unwire(m, TRUE);
461	vm_page_unlock(m);
462	vm_page_wakeup(m);
463	VM_OBJECT_UNLOCK(tobj);
464
465	return (error);
466}
467
468static __inline int
469tmpfs_nocacheread_buf(vm_object_t tobj, vm_pindex_t idx,
470    vm_offset_t offset, size_t tlen, void *buf)
471{
472	struct uio uio;
473	struct iovec iov;
474
475	uio.uio_iovcnt = 1;
476	uio.uio_iov = &iov;
477	iov.iov_base = buf;
478	iov.iov_len = tlen;
479
480	uio.uio_offset = 0;
481	uio.uio_resid = tlen;
482	uio.uio_rw = UIO_READ;
483	uio.uio_segflg = UIO_SYSSPACE;
484	uio.uio_td = curthread;
485
486	return (tmpfs_nocacheread(tobj, idx, offset, tlen, &uio));
487}
488
489static int
490tmpfs_mappedread(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
491{
492	struct sf_buf	*sf;
493	vm_pindex_t	idx;
494	vm_page_t	m;
495	vm_offset_t	offset;
496	off_t		addr;
497	size_t		tlen;
498	char		*ma;
499	int		error;
500
501	addr = uio->uio_offset;
502	idx = OFF_TO_IDX(addr);
503	offset = addr & PAGE_MASK;
504	tlen = MIN(PAGE_SIZE - offset, len);
505
506	if ((vobj == NULL) ||
507	    (vobj->resident_page_count == 0 && vobj->cache == NULL))
508		goto nocache;
509
510	VM_OBJECT_LOCK(vobj);
511lookupvpg:
512	if (((m = vm_page_lookup(vobj, idx)) != NULL) &&
513	    vm_page_is_valid(m, offset, tlen)) {
514		if ((m->oflags & VPO_BUSY) != 0) {
515			/*
516			 * Reference the page before unlocking and sleeping so
517			 * that the page daemon is less likely to reclaim it.
518			 */
519			vm_page_reference(m);
520			vm_page_sleep(m, "tmfsmr");
521			goto lookupvpg;
522		}
523		vm_page_busy(m);
524		VM_OBJECT_UNLOCK(vobj);
525		error = uiomove_fromphys(&m, offset, tlen, uio);
526		VM_OBJECT_LOCK(vobj);
527		vm_page_wakeup(m);
528		VM_OBJECT_UNLOCK(vobj);
529		return	(error);
530	} else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) {
531		KASSERT(offset == 0,
532		    ("unexpected offset in tmpfs_mappedread for sendfile"));
533		if ((m->oflags & VPO_BUSY) != 0) {
534			/*
535			 * Reference the page before unlocking and sleeping so
536			 * that the page daemon is less likely to reclaim it.
537			 */
538			vm_page_reference(m);
539			vm_page_sleep(m, "tmfsmr");
540			goto lookupvpg;
541		}
542		vm_page_busy(m);
543		VM_OBJECT_UNLOCK(vobj);
544		sched_pin();
545		sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
546		ma = (char *)sf_buf_kva(sf);
547		error = tmpfs_nocacheread_buf(tobj, idx, 0, tlen, ma);
548		if (error == 0) {
549			if (tlen != PAGE_SIZE)
550				bzero(ma + tlen, PAGE_SIZE - tlen);
551			uio->uio_offset += tlen;
552			uio->uio_resid -= tlen;
553		}
554		sf_buf_free(sf);
555		sched_unpin();
556		VM_OBJECT_LOCK(vobj);
557		if (error == 0)
558			m->valid = VM_PAGE_BITS_ALL;
559		vm_page_wakeup(m);
560		VM_OBJECT_UNLOCK(vobj);
561		return	(error);
562	}
563	VM_OBJECT_UNLOCK(vobj);
564nocache:
565	error = tmpfs_nocacheread(tobj, idx, offset, tlen, uio);
566
567	return	(error);
568}
569
570static int
571tmpfs_read(struct vop_read_args *v)
572{
573	struct vnode *vp = v->a_vp;
574	struct uio *uio = v->a_uio;
575
576	struct tmpfs_node *node;
577	vm_object_t uobj;
578	size_t len;
579	int resid;
580
581	int error = 0;
582
583	node = VP_TO_TMPFS_NODE(vp);
584
585	if (vp->v_type != VREG) {
586		error = EISDIR;
587		goto out;
588	}
589
590	if (uio->uio_offset < 0) {
591		error = EINVAL;
592		goto out;
593	}
594
595	node->tn_status |= TMPFS_NODE_ACCESSED;
596
597	uobj = node->tn_reg.tn_aobj;
598	while ((resid = uio->uio_resid) > 0) {
599		error = 0;
600		if (node->tn_size <= uio->uio_offset)
601			break;
602		len = MIN(node->tn_size - uio->uio_offset, resid);
603		if (len == 0)
604			break;
605		error = tmpfs_mappedread(vp->v_object, uobj, len, uio);
606		if ((error != 0) || (resid == uio->uio_resid))
607			break;
608	}
609
610out:
611
612	return error;
613}
614
615/* --------------------------------------------------------------------- */
616
617static int
618tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
619{
620	vm_pindex_t	idx;
621	vm_page_t	vpg, tpg;
622	vm_offset_t	offset;
623	off_t		addr;
624	size_t		tlen;
625	int		error, rv;
626
627	error = 0;
628
629	addr = uio->uio_offset;
630	idx = OFF_TO_IDX(addr);
631	offset = addr & PAGE_MASK;
632	tlen = MIN(PAGE_SIZE - offset, len);
633
634	if ((vobj == NULL) ||
635	    (vobj->resident_page_count == 0 && vobj->cache == NULL)) {
636		vpg = NULL;
637		goto nocache;
638	}
639
640	VM_OBJECT_LOCK(vobj);
641lookupvpg:
642	if (((vpg = vm_page_lookup(vobj, idx)) != NULL) &&
643	    vm_page_is_valid(vpg, offset, tlen)) {
644		if ((vpg->oflags & VPO_BUSY) != 0) {
645			/*
646			 * Reference the page before unlocking and sleeping so
647			 * that the page daemon is less likely to reclaim it.
648			 */
649			vm_page_reference(vpg);
650			vm_page_sleep(vpg, "tmfsmw");
651			goto lookupvpg;
652		}
653		vm_page_busy(vpg);
654		vm_page_undirty(vpg);
655		VM_OBJECT_UNLOCK(vobj);
656		error = uiomove_fromphys(&vpg, offset, tlen, uio);
657	} else {
658		if (__predict_false(vobj->cache != NULL))
659			vm_page_cache_free(vobj, idx, idx + 1);
660		VM_OBJECT_UNLOCK(vobj);
661		vpg = NULL;
662	}
663nocache:
664	VM_OBJECT_LOCK(tobj);
665	tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
666	    VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
667	if (tpg->valid != VM_PAGE_BITS_ALL) {
668		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
669			rv = vm_pager_get_pages(tobj, &tpg, 1, 0);
670			if (rv != VM_PAGER_OK) {
671				vm_page_lock(tpg);
672				vm_page_free(tpg);
673				vm_page_unlock(tpg);
674				error = EIO;
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);
688	if (error == 0) {
689		KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
690		    ("parts of tpg invalid"));
691		vm_page_dirty(tpg);
692	}
693	vm_page_lock(tpg);
694	vm_page_unwire(tpg, TRUE);
695	vm_page_unlock(tpg);
696	vm_page_wakeup(tpg);
697out:
698	VM_OBJECT_UNLOCK(tobj);
699	if (vpg != NULL) {
700		VM_OBJECT_LOCK(vobj);
701		vm_page_wakeup(vpg);
702		VM_OBJECT_UNLOCK(vobj);
703	}
704
705	return	(error);
706}
707
708static int
709tmpfs_write(struct vop_write_args *v)
710{
711	struct vnode *vp = v->a_vp;
712	struct uio *uio = v->a_uio;
713	int ioflag = v->a_ioflag;
714
715	boolean_t extended;
716	int error = 0;
717	off_t oldsize;
718	struct tmpfs_node *node;
719	vm_object_t uobj;
720	size_t len;
721	int resid;
722
723	node = VP_TO_TMPFS_NODE(vp);
724	oldsize = node->tn_size;
725
726	if (uio->uio_offset < 0 || vp->v_type != VREG) {
727		error = EINVAL;
728		goto out;
729	}
730
731	if (uio->uio_resid == 0) {
732		error = 0;
733		goto out;
734	}
735
736	if (ioflag & IO_APPEND)
737		uio->uio_offset = node->tn_size;
738
739	if (uio->uio_offset + uio->uio_resid >
740	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
741		return (EFBIG);
742
743	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
744		return (EFBIG);
745
746	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
747	if (extended) {
748		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid,
749		    FALSE);
750		if (error != 0)
751			goto out;
752	}
753
754	uobj = node->tn_reg.tn_aobj;
755	while ((resid = uio->uio_resid) > 0) {
756		if (node->tn_size <= uio->uio_offset)
757			break;
758		len = MIN(node->tn_size - uio->uio_offset, resid);
759		if (len == 0)
760			break;
761		error = tmpfs_mappedwrite(vp->v_object, uobj, len, uio);
762		if ((error != 0) || (resid == uio->uio_resid))
763			break;
764	}
765
766	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
767	    (extended ? TMPFS_NODE_CHANGED : 0);
768
769	if (node->tn_mode & (S_ISUID | S_ISGID)) {
770		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
771			node->tn_mode &= ~(S_ISUID | S_ISGID);
772	}
773
774	if (error != 0)
775		(void)tmpfs_reg_resize(vp, oldsize, TRUE);
776
777out:
778	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
779	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
780
781	return error;
782}
783
784/* --------------------------------------------------------------------- */
785
786static int
787tmpfs_fsync(struct vop_fsync_args *v)
788{
789	struct vnode *vp = v->a_vp;
790
791	MPASS(VOP_ISLOCKED(vp));
792
793	tmpfs_update(vp);
794
795	return 0;
796}
797
798/* --------------------------------------------------------------------- */
799
800static int
801tmpfs_remove(struct vop_remove_args *v)
802{
803	struct vnode *dvp = v->a_dvp;
804	struct vnode *vp = v->a_vp;
805
806	int error;
807	struct tmpfs_dirent *de;
808	struct tmpfs_mount *tmp;
809	struct tmpfs_node *dnode;
810	struct tmpfs_node *node;
811
812	MPASS(VOP_ISLOCKED(dvp));
813	MPASS(VOP_ISLOCKED(vp));
814
815	if (vp->v_type == VDIR) {
816		error = EISDIR;
817		goto out;
818	}
819
820	dnode = VP_TO_TMPFS_DIR(dvp);
821	node = VP_TO_TMPFS_NODE(vp);
822	tmp = VFS_TO_TMPFS(vp->v_mount);
823	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
824	MPASS(de != NULL);
825
826	/* Files marked as immutable or append-only cannot be deleted. */
827	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
828	    (dnode->tn_flags & APPEND)) {
829		error = EPERM;
830		goto out;
831	}
832
833	/* Remove the entry from the directory; as it is a file, we do not
834	 * have to change the number of hard links of the directory. */
835	tmpfs_dir_detach(dvp, de);
836	if (v->a_cnp->cn_flags & DOWHITEOUT)
837		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
838
839	/* Free the directory entry we just deleted.  Note that the node
840	 * referred by it will not be removed until the vnode is really
841	 * reclaimed. */
842	tmpfs_free_dirent(tmp, de, TRUE);
843
844	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
845	error = 0;
846
847out:
848
849	return error;
850}
851
852/* --------------------------------------------------------------------- */
853
854static int
855tmpfs_link(struct vop_link_args *v)
856{
857	struct vnode *dvp = v->a_tdvp;
858	struct vnode *vp = v->a_vp;
859	struct componentname *cnp = v->a_cnp;
860
861	int error;
862	struct tmpfs_dirent *de;
863	struct tmpfs_node *node;
864
865	MPASS(VOP_ISLOCKED(dvp));
866	MPASS(cnp->cn_flags & HASBUF);
867	MPASS(dvp != vp); /* XXX When can this be false? */
868
869	node = VP_TO_TMPFS_NODE(vp);
870
871	/* XXX: Why aren't the following two tests done by the caller? */
872
873	/* Hard links of directories are forbidden. */
874	if (vp->v_type == VDIR) {
875		error = EPERM;
876		goto out;
877	}
878
879	/* Cannot create cross-device links. */
880	if (dvp->v_mount != vp->v_mount) {
881		error = EXDEV;
882		goto out;
883	}
884
885	/* Ensure that we do not overflow the maximum number of links imposed
886	 * by the system. */
887	MPASS(node->tn_links <= LINK_MAX);
888	if (node->tn_links == LINK_MAX) {
889		error = EMLINK;
890		goto out;
891	}
892
893	/* We cannot create links of files marked immutable or append-only. */
894	if (node->tn_flags & (IMMUTABLE | APPEND)) {
895		error = EPERM;
896		goto out;
897	}
898
899	/* Allocate a new directory entry to represent the node. */
900	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
901	    cnp->cn_nameptr, cnp->cn_namelen, &de);
902	if (error != 0)
903		goto out;
904
905	/* Insert the new directory entry into the appropriate directory. */
906	if (cnp->cn_flags & ISWHITEOUT)
907		tmpfs_dir_whiteout_remove(dvp, cnp);
908	tmpfs_dir_attach(dvp, de);
909
910	/* vp link count has changed, so update node times. */
911	node->tn_status |= TMPFS_NODE_CHANGED;
912	tmpfs_update(vp);
913
914	error = 0;
915
916out:
917	return error;
918}
919
920/* --------------------------------------------------------------------- */
921
922static int
923tmpfs_rename(struct vop_rename_args *v)
924{
925	struct vnode *fdvp = v->a_fdvp;
926	struct vnode *fvp = v->a_fvp;
927	struct componentname *fcnp = v->a_fcnp;
928	struct vnode *tdvp = v->a_tdvp;
929	struct vnode *tvp = v->a_tvp;
930	struct componentname *tcnp = v->a_tcnp;
931
932	char *newname;
933	int error;
934	struct tmpfs_dirent *de;
935	struct tmpfs_mount *tmp;
936	struct tmpfs_node *fdnode;
937	struct tmpfs_node *fnode;
938	struct tmpfs_node *tnode;
939	struct tmpfs_node *tdnode;
940
941	MPASS(VOP_ISLOCKED(tdvp));
942	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
943	MPASS(fcnp->cn_flags & HASBUF);
944	MPASS(tcnp->cn_flags & HASBUF);
945
946  	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
947
948	/* Disallow cross-device renames.
949	 * XXX Why isn't this done by the caller? */
950	if (fvp->v_mount != tdvp->v_mount ||
951	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
952		error = EXDEV;
953		goto out;
954	}
955
956	tmp = VFS_TO_TMPFS(tdvp->v_mount);
957	tdnode = VP_TO_TMPFS_DIR(tdvp);
958
959	/* If source and target are the same file, there is nothing to do. */
960	if (fvp == tvp) {
961		error = 0;
962		goto out;
963	}
964
965	/* If we need to move the directory between entries, lock the
966	 * source so that we can safely operate on it. */
967	if (fdvp != tdvp && fdvp != tvp)
968		vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
969	fdnode = VP_TO_TMPFS_DIR(fdvp);
970	fnode = VP_TO_TMPFS_NODE(fvp);
971	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
972
973	/* Entry can disappear before we lock fdvp,
974	 * also avoid manipulating '.' and '..' entries. */
975	if (de == NULL) {
976		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
977		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
978			error = EINVAL;
979		else
980			error = ENOENT;
981		goto out_locked;
982	}
983	MPASS(de->td_node == fnode);
984
985	/* If re-naming a directory to another preexisting directory
986	 * ensure that the target directory is empty so that its
987	 * removal causes no side effects.
988	 * Kern_rename gurantees the destination to be a directory
989	 * if the source is one. */
990	if (tvp != NULL) {
991		MPASS(tnode != NULL);
992
993		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
994		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
995			error = EPERM;
996			goto out_locked;
997		}
998
999		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1000			if (tnode->tn_size > 0) {
1001				error = ENOTEMPTY;
1002				goto out_locked;
1003			}
1004		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1005			error = ENOTDIR;
1006			goto out_locked;
1007		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1008			error = EISDIR;
1009			goto out_locked;
1010		} else {
1011			MPASS(fnode->tn_type != VDIR &&
1012				tnode->tn_type != VDIR);
1013		}
1014	}
1015
1016	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1017	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1018		error = EPERM;
1019		goto out_locked;
1020	}
1021
1022	/* Ensure that we have enough memory to hold the new name, if it
1023	 * has to be changed. */
1024	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1025	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1026		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1027	} else
1028		newname = NULL;
1029
1030	/* If the node is being moved to another directory, we have to do
1031	 * the move. */
1032	if (fdnode != tdnode) {
1033		/* In case we are moving a directory, we have to adjust its
1034		 * parent to point to the new parent. */
1035		if (de->td_node->tn_type == VDIR) {
1036			struct tmpfs_node *n;
1037
1038			/* Ensure the target directory is not a child of the
1039			 * directory being moved.  Otherwise, we'd end up
1040			 * with stale nodes. */
1041			n = tdnode;
1042			/* TMPFS_LOCK garanties that no nodes are freed while
1043			 * traversing the list. Nodes can only be marked as
1044			 * removed: tn_parent == NULL. */
1045			TMPFS_LOCK(tmp);
1046			TMPFS_NODE_LOCK(n);
1047			while (n != n->tn_dir.tn_parent) {
1048				struct tmpfs_node *parent;
1049
1050				if (n == fnode) {
1051					TMPFS_NODE_UNLOCK(n);
1052					TMPFS_UNLOCK(tmp);
1053					error = EINVAL;
1054					if (newname != NULL)
1055						    free(newname, M_TMPFSNAME);
1056					goto out_locked;
1057				}
1058				parent = n->tn_dir.tn_parent;
1059				TMPFS_NODE_UNLOCK(n);
1060				if (parent == NULL) {
1061					n = NULL;
1062					break;
1063				}
1064				TMPFS_NODE_LOCK(parent);
1065				if (parent->tn_dir.tn_parent == NULL) {
1066					TMPFS_NODE_UNLOCK(parent);
1067					n = NULL;
1068					break;
1069				}
1070				n = parent;
1071			}
1072			TMPFS_UNLOCK(tmp);
1073			if (n == NULL) {
1074				error = EINVAL;
1075				if (newname != NULL)
1076					    free(newname, M_TMPFSNAME);
1077				goto out_locked;
1078			}
1079			TMPFS_NODE_UNLOCK(n);
1080
1081			/* Adjust the parent pointer. */
1082			TMPFS_VALIDATE_DIR(fnode);
1083			TMPFS_NODE_LOCK(de->td_node);
1084			de->td_node->tn_dir.tn_parent = tdnode;
1085			TMPFS_NODE_UNLOCK(de->td_node);
1086
1087			/* As a result of changing the target of the '..'
1088			 * entry, the link count of the source and target
1089			 * directories has to be adjusted. */
1090			TMPFS_NODE_LOCK(tdnode);
1091			TMPFS_ASSERT_LOCKED(tdnode);
1092			tdnode->tn_links++;
1093			TMPFS_NODE_UNLOCK(tdnode);
1094
1095			TMPFS_NODE_LOCK(fdnode);
1096			TMPFS_ASSERT_LOCKED(fdnode);
1097			fdnode->tn_links--;
1098			TMPFS_NODE_UNLOCK(fdnode);
1099		}
1100
1101		/* Do the move: just remove the entry from the source directory
1102		 * and insert it into the target one. */
1103		tmpfs_dir_detach(fdvp, de);
1104		if (fcnp->cn_flags & DOWHITEOUT)
1105			tmpfs_dir_whiteout_add(fdvp, fcnp);
1106		if (tcnp->cn_flags & ISWHITEOUT)
1107			tmpfs_dir_whiteout_remove(tdvp, tcnp);
1108		tmpfs_dir_attach(tdvp, de);
1109	}
1110
1111	/* If the name has changed, we need to make it effective by changing
1112	 * it in the directory entry. */
1113	if (newname != NULL) {
1114		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1115
1116		free(de->td_name, M_TMPFSNAME);
1117		de->td_namelen = (uint16_t)tcnp->cn_namelen;
1118		memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen);
1119		de->td_name = newname;
1120
1121		fnode->tn_status |= TMPFS_NODE_CHANGED;
1122		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1123	}
1124
1125	/* If we are overwriting an entry, we have to remove the old one
1126	 * from the target directory. */
1127	if (tvp != NULL) {
1128		/* Remove the old entry from the target directory. */
1129		de = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1130		tmpfs_dir_detach(tdvp, de);
1131
1132		/* Free the directory entry we just deleted.  Note that the
1133		 * node referred by it will not be removed until the vnode is
1134		 * really reclaimed. */
1135		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE);
1136	}
1137	cache_purge(fvp);
1138
1139	error = 0;
1140
1141out_locked:
1142	if (fdvp != tdvp && fdvp != tvp)
1143		VOP_UNLOCK(fdvp, 0);
1144
1145out:
1146	/* Release target nodes. */
1147	/* XXX: I don't understand when tdvp can be the same as tvp, but
1148	 * other code takes care of this... */
1149	if (tdvp == tvp)
1150		vrele(tdvp);
1151	else
1152		vput(tdvp);
1153	if (tvp != NULL)
1154		vput(tvp);
1155
1156	/* Release source nodes. */
1157	vrele(fdvp);
1158	vrele(fvp);
1159
1160	return error;
1161}
1162
1163/* --------------------------------------------------------------------- */
1164
1165static int
1166tmpfs_mkdir(struct vop_mkdir_args *v)
1167{
1168	struct vnode *dvp = v->a_dvp;
1169	struct vnode **vpp = v->a_vpp;
1170	struct componentname *cnp = v->a_cnp;
1171	struct vattr *vap = v->a_vap;
1172
1173	MPASS(vap->va_type == VDIR);
1174
1175	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1176}
1177
1178/* --------------------------------------------------------------------- */
1179
1180static int
1181tmpfs_rmdir(struct vop_rmdir_args *v)
1182{
1183	struct vnode *dvp = v->a_dvp;
1184	struct vnode *vp = v->a_vp;
1185
1186	int error;
1187	struct tmpfs_dirent *de;
1188	struct tmpfs_mount *tmp;
1189	struct tmpfs_node *dnode;
1190	struct tmpfs_node *node;
1191
1192	MPASS(VOP_ISLOCKED(dvp));
1193	MPASS(VOP_ISLOCKED(vp));
1194
1195	tmp = VFS_TO_TMPFS(dvp->v_mount);
1196	dnode = VP_TO_TMPFS_DIR(dvp);
1197	node = VP_TO_TMPFS_DIR(vp);
1198
1199	/* Directories with more than two entries ('.' and '..') cannot be
1200	 * removed. */
1201	 if (node->tn_size > 0) {
1202		 error = ENOTEMPTY;
1203		 goto out;
1204	 }
1205
1206	if ((dnode->tn_flags & APPEND)
1207	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1208		error = EPERM;
1209		goto out;
1210	}
1211
1212	/* This invariant holds only if we are not trying to remove "..".
1213	  * We checked for that above so this is safe now. */
1214	MPASS(node->tn_dir.tn_parent == dnode);
1215
1216	/* Get the directory entry associated with node (vp).  This was
1217	 * filled by tmpfs_lookup while looking up the entry. */
1218	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1219	MPASS(TMPFS_DIRENT_MATCHES(de,
1220	    v->a_cnp->cn_nameptr,
1221	    v->a_cnp->cn_namelen));
1222
1223	/* Check flags to see if we are allowed to remove the directory. */
1224	if (dnode->tn_flags & APPEND
1225		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1226		error = EPERM;
1227		goto out;
1228	}
1229
1230
1231	/* Detach the directory entry from the directory (dnode). */
1232	tmpfs_dir_detach(dvp, de);
1233	if (v->a_cnp->cn_flags & DOWHITEOUT)
1234		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1235
1236	/* No vnode should be allocated for this entry from this point */
1237	TMPFS_NODE_LOCK(node);
1238	TMPFS_ASSERT_ELOCKED(node);
1239	node->tn_links--;
1240	node->tn_dir.tn_parent = NULL;
1241	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1242	    TMPFS_NODE_MODIFIED;
1243
1244	TMPFS_NODE_UNLOCK(node);
1245
1246	TMPFS_NODE_LOCK(dnode);
1247	TMPFS_ASSERT_ELOCKED(dnode);
1248	dnode->tn_links--;
1249	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1250	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1251	TMPFS_NODE_UNLOCK(dnode);
1252
1253	cache_purge(dvp);
1254	cache_purge(vp);
1255
1256	/* Free the directory entry we just deleted.  Note that the node
1257	 * referred by it will not be removed until the vnode is really
1258	 * reclaimed. */
1259	tmpfs_free_dirent(tmp, de, TRUE);
1260
1261	/* Release the deleted vnode (will destroy the node, notify
1262	 * interested parties and clean it from the cache). */
1263
1264	dnode->tn_status |= TMPFS_NODE_CHANGED;
1265	tmpfs_update(dvp);
1266
1267	error = 0;
1268
1269out:
1270	return error;
1271}
1272
1273/* --------------------------------------------------------------------- */
1274
1275static int
1276tmpfs_symlink(struct vop_symlink_args *v)
1277{
1278	struct vnode *dvp = v->a_dvp;
1279	struct vnode **vpp = v->a_vpp;
1280	struct componentname *cnp = v->a_cnp;
1281	struct vattr *vap = v->a_vap;
1282	char *target = v->a_target;
1283
1284#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1285	MPASS(vap->va_type == VLNK);
1286#else
1287	vap->va_type = VLNK;
1288#endif
1289
1290	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1291}
1292
1293/* --------------------------------------------------------------------- */
1294
1295static int
1296tmpfs_readdir(struct vop_readdir_args *v)
1297{
1298	struct vnode *vp = v->a_vp;
1299	struct uio *uio = v->a_uio;
1300	int *eofflag = v->a_eofflag;
1301	u_long **cookies = v->a_cookies;
1302	int *ncookies = v->a_ncookies;
1303
1304	int error;
1305	off_t startoff;
1306	off_t cnt = 0;
1307	struct tmpfs_node *node;
1308
1309	/* This operation only makes sense on directory nodes. */
1310	if (vp->v_type != VDIR)
1311		return ENOTDIR;
1312
1313	node = VP_TO_TMPFS_DIR(vp);
1314
1315	startoff = uio->uio_offset;
1316
1317	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1318		error = tmpfs_dir_getdotdent(node, uio);
1319		if (error != 0)
1320			goto outok;
1321		cnt++;
1322	}
1323
1324	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1325		error = tmpfs_dir_getdotdotdent(node, uio);
1326		if (error != 0)
1327			goto outok;
1328		cnt++;
1329	}
1330
1331	error = tmpfs_dir_getdents(node, uio, &cnt);
1332
1333outok:
1334	MPASS(error >= -1);
1335
1336	if (error == -1)
1337		error = (cnt != 0) ? 0 : EINVAL;
1338
1339	if (eofflag != NULL)
1340		*eofflag =
1341		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1342
1343	/* Update NFS-related variables. */
1344	if (error == 0 && cookies != NULL && ncookies != NULL) {
1345		off_t i;
1346		off_t off = startoff;
1347		struct tmpfs_dirent *de = NULL;
1348
1349		*ncookies = cnt;
1350		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1351
1352		for (i = 0; i < cnt; i++) {
1353			MPASS(off != TMPFS_DIRCOOKIE_EOF);
1354			if (off == TMPFS_DIRCOOKIE_DOT) {
1355				off = TMPFS_DIRCOOKIE_DOTDOT;
1356			} else {
1357				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1358					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1359				} else if (de != NULL) {
1360					de = TAILQ_NEXT(de, td_entries);
1361				} else {
1362					de = tmpfs_dir_lookupbycookie(node,
1363					    off);
1364					MPASS(de != NULL);
1365					de = TAILQ_NEXT(de, td_entries);
1366				}
1367				if (de == NULL)
1368					off = TMPFS_DIRCOOKIE_EOF;
1369				else
1370					off = tmpfs_dircookie(de);
1371			}
1372
1373			(*cookies)[i] = off;
1374		}
1375		MPASS(uio->uio_offset == off);
1376	}
1377
1378	return error;
1379}
1380
1381/* --------------------------------------------------------------------- */
1382
1383static int
1384tmpfs_readlink(struct vop_readlink_args *v)
1385{
1386	struct vnode *vp = v->a_vp;
1387	struct uio *uio = v->a_uio;
1388
1389	int error;
1390	struct tmpfs_node *node;
1391
1392	MPASS(uio->uio_offset == 0);
1393	MPASS(vp->v_type == VLNK);
1394
1395	node = VP_TO_TMPFS_NODE(vp);
1396
1397	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1398	    uio);
1399	node->tn_status |= TMPFS_NODE_ACCESSED;
1400
1401	return error;
1402}
1403
1404/* --------------------------------------------------------------------- */
1405
1406static int
1407tmpfs_inactive(struct vop_inactive_args *v)
1408{
1409	struct vnode *vp = v->a_vp;
1410	struct thread *l = v->a_td;
1411
1412	struct tmpfs_node *node;
1413
1414	MPASS(VOP_ISLOCKED(vp));
1415
1416	node = VP_TO_TMPFS_NODE(vp);
1417
1418	if (node->tn_links == 0)
1419		vrecycle(vp, l);
1420
1421	return 0;
1422}
1423
1424/* --------------------------------------------------------------------- */
1425
1426int
1427tmpfs_reclaim(struct vop_reclaim_args *v)
1428{
1429	struct vnode *vp = v->a_vp;
1430
1431	struct tmpfs_mount *tmp;
1432	struct tmpfs_node *node;
1433
1434	node = VP_TO_TMPFS_NODE(vp);
1435	tmp = VFS_TO_TMPFS(vp->v_mount);
1436
1437	vnode_destroy_vobject(vp);
1438	cache_purge(vp);
1439
1440	TMPFS_NODE_LOCK(node);
1441	TMPFS_ASSERT_ELOCKED(node);
1442	tmpfs_free_vp(vp);
1443
1444	/* If the node referenced by this vnode was deleted by the user,
1445	 * we must free its associated data structures (now that the vnode
1446	 * is being reclaimed). */
1447	if (node->tn_links == 0 &&
1448	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1449		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1450		TMPFS_NODE_UNLOCK(node);
1451		tmpfs_free_node(tmp, node);
1452	} else
1453		TMPFS_NODE_UNLOCK(node);
1454
1455	MPASS(vp->v_data == NULL);
1456	return 0;
1457}
1458
1459/* --------------------------------------------------------------------- */
1460
1461static int
1462tmpfs_print(struct vop_print_args *v)
1463{
1464	struct vnode *vp = v->a_vp;
1465
1466	struct tmpfs_node *node;
1467
1468	node = VP_TO_TMPFS_NODE(vp);
1469
1470	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1471	    node, node->tn_flags, node->tn_links);
1472	printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n",
1473	    node->tn_mode, node->tn_uid, node->tn_gid,
1474	    (intmax_t)node->tn_size, node->tn_status);
1475
1476	if (vp->v_type == VFIFO)
1477		fifo_printinfo(vp);
1478
1479	printf("\n");
1480
1481	return 0;
1482}
1483
1484/* --------------------------------------------------------------------- */
1485
1486static int
1487tmpfs_pathconf(struct vop_pathconf_args *v)
1488{
1489	int name = v->a_name;
1490	register_t *retval = v->a_retval;
1491
1492	int error;
1493
1494	error = 0;
1495
1496	switch (name) {
1497	case _PC_LINK_MAX:
1498		*retval = LINK_MAX;
1499		break;
1500
1501	case _PC_NAME_MAX:
1502		*retval = NAME_MAX;
1503		break;
1504
1505	case _PC_PATH_MAX:
1506		*retval = PATH_MAX;
1507		break;
1508
1509	case _PC_PIPE_BUF:
1510		*retval = PIPE_BUF;
1511		break;
1512
1513	case _PC_CHOWN_RESTRICTED:
1514		*retval = 1;
1515		break;
1516
1517	case _PC_NO_TRUNC:
1518		*retval = 1;
1519		break;
1520
1521	case _PC_SYNC_IO:
1522		*retval = 1;
1523		break;
1524
1525	case _PC_FILESIZEBITS:
1526		*retval = 0; /* XXX Don't know which value should I return. */
1527		break;
1528
1529	default:
1530		error = EINVAL;
1531	}
1532
1533	return error;
1534}
1535
1536static int
1537tmpfs_vptofh(struct vop_vptofh_args *ap)
1538{
1539	struct tmpfs_fid *tfhp;
1540	struct tmpfs_node *node;
1541
1542	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1543	node = VP_TO_TMPFS_NODE(ap->a_vp);
1544
1545	tfhp->tf_len = sizeof(struct tmpfs_fid);
1546	tfhp->tf_id = node->tn_id;
1547	tfhp->tf_gen = node->tn_gen;
1548
1549	return (0);
1550}
1551
1552static int
1553tmpfs_whiteout(struct vop_whiteout_args *ap)
1554{
1555	struct vnode *dvp = ap->a_dvp;
1556	struct componentname *cnp = ap->a_cnp;
1557	struct tmpfs_dirent *de;
1558
1559	switch (ap->a_flags) {
1560	case LOOKUP:
1561		return (0);
1562	case CREATE:
1563		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1564		if (de != NULL)
1565			return (de->td_node == NULL ? 0 : EEXIST);
1566		return (tmpfs_dir_whiteout_add(dvp, cnp));
1567	case DELETE:
1568		tmpfs_dir_whiteout_remove(dvp, cnp);
1569		return (0);
1570	default:
1571		panic("tmpfs_whiteout: unknown op");
1572	}
1573}
1574
1575/* --------------------------------------------------------------------- */
1576
1577/*
1578 * vnode operations vector used for files stored in a tmpfs file system.
1579 */
1580struct vop_vector tmpfs_vnodeop_entries = {
1581	.vop_default =			&default_vnodeops,
1582	.vop_lookup =			vfs_cache_lookup,
1583	.vop_cachedlookup =		tmpfs_lookup,
1584	.vop_create =			tmpfs_create,
1585	.vop_mknod =			tmpfs_mknod,
1586	.vop_open =			tmpfs_open,
1587	.vop_close =			tmpfs_close,
1588	.vop_access =			tmpfs_access,
1589	.vop_getattr =			tmpfs_getattr,
1590	.vop_setattr =			tmpfs_setattr,
1591	.vop_read =			tmpfs_read,
1592	.vop_write =			tmpfs_write,
1593	.vop_fsync =			tmpfs_fsync,
1594	.vop_remove =			tmpfs_remove,
1595	.vop_link =			tmpfs_link,
1596	.vop_rename =			tmpfs_rename,
1597	.vop_mkdir =			tmpfs_mkdir,
1598	.vop_rmdir =			tmpfs_rmdir,
1599	.vop_symlink =			tmpfs_symlink,
1600	.vop_readdir =			tmpfs_readdir,
1601	.vop_readlink =			tmpfs_readlink,
1602	.vop_inactive =			tmpfs_inactive,
1603	.vop_reclaim =			tmpfs_reclaim,
1604	.vop_print =			tmpfs_print,
1605	.vop_pathconf =			tmpfs_pathconf,
1606	.vop_vptofh =			tmpfs_vptofh,
1607	.vop_whiteout =			tmpfs_whiteout,
1608	.vop_bmap =			VOP_EOPNOTSUPP,
1609};
1610
1611