tmpfs_vnops.c revision 212650
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 212650 2010-09-15 10:31:27Z 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		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		if (error == 0)
566			vm_page_set_valid(m, offset, tlen);
567		vm_page_wakeup(m);
568		VM_OBJECT_UNLOCK(vobj);
569		return	(error);
570	}
571	VM_OBJECT_UNLOCK(vobj);
572nocache:
573	error = tmpfs_nocacheread(tobj, idx, offset, tlen, uio);
574
575	return	(error);
576}
577
578static int
579tmpfs_read(struct vop_read_args *v)
580{
581	struct vnode *vp = v->a_vp;
582	struct uio *uio = v->a_uio;
583
584	struct tmpfs_node *node;
585	vm_object_t uobj;
586	size_t len;
587	int resid;
588
589	int error = 0;
590
591	node = VP_TO_TMPFS_NODE(vp);
592
593	if (vp->v_type != VREG) {
594		error = EISDIR;
595		goto out;
596	}
597
598	if (uio->uio_offset < 0) {
599		error = EINVAL;
600		goto out;
601	}
602
603	node->tn_status |= TMPFS_NODE_ACCESSED;
604
605	uobj = node->tn_reg.tn_aobj;
606	while ((resid = uio->uio_resid) > 0) {
607		error = 0;
608		if (node->tn_size <= uio->uio_offset)
609			break;
610		len = MIN(node->tn_size - uio->uio_offset, resid);
611		if (len == 0)
612			break;
613		error = tmpfs_mappedread(vp->v_object, uobj, len, uio);
614		if ((error != 0) || (resid == uio->uio_resid))
615			break;
616	}
617
618out:
619
620	return error;
621}
622
623/* --------------------------------------------------------------------- */
624
625static int
626tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
627{
628	vm_pindex_t	idx;
629	vm_page_t	vpg, tpg;
630	vm_offset_t	offset;
631	off_t		addr;
632	size_t		tlen;
633	int		error;
634
635	error = 0;
636
637	addr = uio->uio_offset;
638	idx = OFF_TO_IDX(addr);
639	offset = addr & PAGE_MASK;
640	tlen = MIN(PAGE_SIZE - offset, len);
641
642	if ((vobj == NULL) ||
643	    (vobj->resident_page_count == 0 && vobj->cache == NULL)) {
644		vpg = NULL;
645		goto nocache;
646	}
647
648	VM_OBJECT_LOCK(vobj);
649lookupvpg:
650	if (((vpg = vm_page_lookup(vobj, idx)) != NULL) &&
651	    vm_page_is_valid(vpg, offset, tlen)) {
652		if ((vpg->oflags & VPO_BUSY) != 0) {
653			/*
654			 * Reference the page before unlocking and sleeping so
655			 * that the page daemon is less likely to reclaim it.
656			 */
657			vm_page_lock_queues();
658			vm_page_flag_set(vpg, PG_REFERENCED);
659			vm_page_sleep(vpg, "tmfsmw");
660			goto lookupvpg;
661		}
662		vm_page_busy(vpg);
663		vm_page_undirty(vpg);
664		VM_OBJECT_UNLOCK(vobj);
665		error = uiomove_fromphys(&vpg, offset, tlen, uio);
666	} else {
667		if (__predict_false(vobj->cache != NULL))
668			vm_page_cache_free(vobj, idx, idx + 1);
669		VM_OBJECT_UNLOCK(vobj);
670		vpg = NULL;
671	}
672nocache:
673	VM_OBJECT_LOCK(tobj);
674	vm_object_pip_add(tobj, 1);
675	tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
676	    VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
677	if (tpg->valid != VM_PAGE_BITS_ALL) {
678		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
679			error = vm_pager_get_pages(tobj, &tpg, 1, 0);
680			if (error != 0) {
681				printf("tmpfs get pages from pager error [write]\n");
682				goto out;
683			}
684		} else
685			vm_page_zero_invalid(tpg, TRUE);
686	}
687	VM_OBJECT_UNLOCK(tobj);
688	if (vpg == NULL)
689		error = uiomove_fromphys(&tpg, offset, tlen, uio);
690	else {
691		KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid"));
692		pmap_copy_page(vpg, tpg);
693	}
694	VM_OBJECT_LOCK(tobj);
695out:
696	if (vobj != NULL)
697		VM_OBJECT_LOCK(vobj);
698	if (error == 0) {
699		KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
700		    ("parts of tpg invalid"));
701		vm_page_dirty(tpg);
702	}
703	vm_page_lock(tpg);
704	vm_page_unwire(tpg, TRUE);
705	vm_page_unlock(tpg);
706	vm_page_wakeup(tpg);
707	if (vpg != NULL)
708		vm_page_wakeup(vpg);
709	if (vobj != NULL)
710		VM_OBJECT_UNLOCK(vobj);
711	vm_object_pip_subtract(tobj, 1);
712	VM_OBJECT_UNLOCK(tobj);
713
714	return	(error);
715}
716
717static int
718tmpfs_write(struct vop_write_args *v)
719{
720	struct vnode *vp = v->a_vp;
721	struct uio *uio = v->a_uio;
722	int ioflag = v->a_ioflag;
723
724	boolean_t extended;
725	int error = 0;
726	off_t oldsize;
727	struct tmpfs_node *node;
728	vm_object_t uobj;
729	size_t len;
730	int resid;
731
732	node = VP_TO_TMPFS_NODE(vp);
733	oldsize = node->tn_size;
734
735	if (uio->uio_offset < 0 || vp->v_type != VREG) {
736		error = EINVAL;
737		goto out;
738	}
739
740	if (uio->uio_resid == 0) {
741		error = 0;
742		goto out;
743	}
744
745	if (ioflag & IO_APPEND)
746		uio->uio_offset = node->tn_size;
747
748	if (uio->uio_offset + uio->uio_resid >
749	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
750		return (EFBIG);
751
752	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
753		return (EFBIG);
754
755	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
756	if (extended) {
757		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
758		if (error != 0)
759			goto out;
760	}
761
762	uobj = node->tn_reg.tn_aobj;
763	while ((resid = uio->uio_resid) > 0) {
764		if (node->tn_size <= uio->uio_offset)
765			break;
766		len = MIN(node->tn_size - uio->uio_offset, resid);
767		if (len == 0)
768			break;
769		error = tmpfs_mappedwrite(vp->v_object, uobj, len, uio);
770		if ((error != 0) || (resid == uio->uio_resid))
771			break;
772	}
773
774	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
775	    (extended ? TMPFS_NODE_CHANGED : 0);
776
777	if (node->tn_mode & (S_ISUID | S_ISGID)) {
778		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
779			node->tn_mode &= ~(S_ISUID | S_ISGID);
780	}
781
782	if (error != 0)
783		(void)tmpfs_reg_resize(vp, oldsize);
784
785out:
786	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
787	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
788
789	return error;
790}
791
792/* --------------------------------------------------------------------- */
793
794static int
795tmpfs_fsync(struct vop_fsync_args *v)
796{
797	struct vnode *vp = v->a_vp;
798
799	MPASS(VOP_ISLOCKED(vp));
800
801	tmpfs_update(vp);
802
803	return 0;
804}
805
806/* --------------------------------------------------------------------- */
807
808static int
809tmpfs_remove(struct vop_remove_args *v)
810{
811	struct vnode *dvp = v->a_dvp;
812	struct vnode *vp = v->a_vp;
813
814	int error;
815	struct tmpfs_dirent *de;
816	struct tmpfs_mount *tmp;
817	struct tmpfs_node *dnode;
818	struct tmpfs_node *node;
819
820	MPASS(VOP_ISLOCKED(dvp));
821	MPASS(VOP_ISLOCKED(vp));
822
823	if (vp->v_type == VDIR) {
824		error = EISDIR;
825		goto out;
826	}
827
828	dnode = VP_TO_TMPFS_DIR(dvp);
829	node = VP_TO_TMPFS_NODE(vp);
830	tmp = VFS_TO_TMPFS(vp->v_mount);
831	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
832	MPASS(de != NULL);
833
834	/* Files marked as immutable or append-only cannot be deleted. */
835	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
836	    (dnode->tn_flags & APPEND)) {
837		error = EPERM;
838		goto out;
839	}
840
841	/* Remove the entry from the directory; as it is a file, we do not
842	 * have to change the number of hard links of the directory. */
843	tmpfs_dir_detach(dvp, de);
844	if (v->a_cnp->cn_flags & DOWHITEOUT)
845		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
846
847	/* Free the directory entry we just deleted.  Note that the node
848	 * referred by it will not be removed until the vnode is really
849	 * reclaimed. */
850	tmpfs_free_dirent(tmp, de, TRUE);
851
852	if (node->tn_links > 0)
853		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
854	    TMPFS_NODE_MODIFIED;
855	error = 0;
856
857out:
858
859	return error;
860}
861
862/* --------------------------------------------------------------------- */
863
864static int
865tmpfs_link(struct vop_link_args *v)
866{
867	struct vnode *dvp = v->a_tdvp;
868	struct vnode *vp = v->a_vp;
869	struct componentname *cnp = v->a_cnp;
870
871	int error;
872	struct tmpfs_dirent *de;
873	struct tmpfs_node *node;
874
875	MPASS(VOP_ISLOCKED(dvp));
876	MPASS(cnp->cn_flags & HASBUF);
877	MPASS(dvp != vp); /* XXX When can this be false? */
878
879	node = VP_TO_TMPFS_NODE(vp);
880
881	/* XXX: Why aren't the following two tests done by the caller? */
882
883	/* Hard links of directories are forbidden. */
884	if (vp->v_type == VDIR) {
885		error = EPERM;
886		goto out;
887	}
888
889	/* Cannot create cross-device links. */
890	if (dvp->v_mount != vp->v_mount) {
891		error = EXDEV;
892		goto out;
893	}
894
895	/* Ensure that we do not overflow the maximum number of links imposed
896	 * by the system. */
897	MPASS(node->tn_links <= LINK_MAX);
898	if (node->tn_links == LINK_MAX) {
899		error = EMLINK;
900		goto out;
901	}
902
903	/* We cannot create links of files marked immutable or append-only. */
904	if (node->tn_flags & (IMMUTABLE | APPEND)) {
905		error = EPERM;
906		goto out;
907	}
908
909	/* Allocate a new directory entry to represent the node. */
910	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
911	    cnp->cn_nameptr, cnp->cn_namelen, &de);
912	if (error != 0)
913		goto out;
914
915	/* Insert the new directory entry into the appropriate directory. */
916	if (cnp->cn_flags & ISWHITEOUT)
917		tmpfs_dir_whiteout_remove(dvp, cnp);
918	tmpfs_dir_attach(dvp, de);
919
920	/* vp link count has changed, so update node times. */
921	node->tn_status |= TMPFS_NODE_CHANGED;
922	tmpfs_update(vp);
923
924	error = 0;
925
926out:
927	return error;
928}
929
930/* --------------------------------------------------------------------- */
931
932static int
933tmpfs_rename(struct vop_rename_args *v)
934{
935	struct vnode *fdvp = v->a_fdvp;
936	struct vnode *fvp = v->a_fvp;
937	struct componentname *fcnp = v->a_fcnp;
938	struct vnode *tdvp = v->a_tdvp;
939	struct vnode *tvp = v->a_tvp;
940	struct componentname *tcnp = v->a_tcnp;
941
942	char *newname;
943	int error;
944	struct tmpfs_dirent *de;
945	struct tmpfs_mount *tmp;
946	struct tmpfs_node *fdnode;
947	struct tmpfs_node *fnode;
948	struct tmpfs_node *tnode;
949	struct tmpfs_node *tdnode;
950
951	MPASS(VOP_ISLOCKED(tdvp));
952	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
953	MPASS(fcnp->cn_flags & HASBUF);
954	MPASS(tcnp->cn_flags & HASBUF);
955
956  	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
957
958	/* Disallow cross-device renames.
959	 * XXX Why isn't this done by the caller? */
960	if (fvp->v_mount != tdvp->v_mount ||
961	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
962		error = EXDEV;
963		goto out;
964	}
965
966	tmp = VFS_TO_TMPFS(tdvp->v_mount);
967	tdnode = VP_TO_TMPFS_DIR(tdvp);
968
969	/* If source and target are the same file, there is nothing to do. */
970	if (fvp == tvp) {
971		error = 0;
972		goto out;
973	}
974
975	/* If we need to move the directory between entries, lock the
976	 * source so that we can safely operate on it. */
977	if (tdvp != fdvp) {
978		error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
979		if (error != 0)
980			goto out;
981	}
982	fdnode = VP_TO_TMPFS_DIR(fdvp);
983	fnode = VP_TO_TMPFS_NODE(fvp);
984	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
985
986	/* Entry can disappear before we lock fdvp,
987	 * also avoid manipulating '.' and '..' entries. */
988	if (de == NULL) {
989		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
990		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
991			error = EINVAL;
992		else
993			error = ENOENT;
994		goto out_locked;
995	}
996	MPASS(de->td_node == fnode);
997
998	/* If re-naming a directory to another preexisting directory
999	 * ensure that the target directory is empty so that its
1000	 * removal causes no side effects.
1001	 * Kern_rename gurantees the destination to be a directory
1002	 * if the source is one. */
1003	if (tvp != NULL) {
1004		MPASS(tnode != NULL);
1005
1006		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1007		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1008			error = EPERM;
1009			goto out_locked;
1010		}
1011
1012		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1013			if (tnode->tn_size > 0) {
1014				error = ENOTEMPTY;
1015				goto out_locked;
1016			}
1017		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1018			error = ENOTDIR;
1019			goto out_locked;
1020		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1021			error = EISDIR;
1022			goto out_locked;
1023		} else {
1024			MPASS(fnode->tn_type != VDIR &&
1025				tnode->tn_type != VDIR);
1026		}
1027	}
1028
1029	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1030	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1031		error = EPERM;
1032		goto out_locked;
1033	}
1034
1035	/* Ensure that we have enough memory to hold the new name, if it
1036	 * has to be changed. */
1037	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1038	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1039		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1040	} else
1041		newname = NULL;
1042
1043	/* If the node is being moved to another directory, we have to do
1044	 * the move. */
1045	if (fdnode != tdnode) {
1046		/* In case we are moving a directory, we have to adjust its
1047		 * parent to point to the new parent. */
1048		if (de->td_node->tn_type == VDIR) {
1049			struct tmpfs_node *n;
1050
1051			/* Ensure the target directory is not a child of the
1052			 * directory being moved.  Otherwise, we'd end up
1053			 * with stale nodes. */
1054			n = tdnode;
1055			/* TMPFS_LOCK garanties that no nodes are freed while
1056			 * traversing the list. Nodes can only be marked as
1057			 * removed: tn_parent == NULL. */
1058			TMPFS_LOCK(tmp);
1059			TMPFS_NODE_LOCK(n);
1060			while (n != n->tn_dir.tn_parent) {
1061				struct tmpfs_node *parent;
1062
1063				if (n == fnode) {
1064					TMPFS_NODE_UNLOCK(n);
1065					TMPFS_UNLOCK(tmp);
1066					error = EINVAL;
1067					if (newname != NULL)
1068						    free(newname, M_TMPFSNAME);
1069					goto out_locked;
1070				}
1071				parent = n->tn_dir.tn_parent;
1072				TMPFS_NODE_UNLOCK(n);
1073				if (parent == NULL) {
1074					n = NULL;
1075					break;
1076				}
1077				TMPFS_NODE_LOCK(parent);
1078				if (parent->tn_dir.tn_parent == NULL) {
1079					TMPFS_NODE_UNLOCK(parent);
1080					n = NULL;
1081					break;
1082				}
1083				n = parent;
1084			}
1085			TMPFS_UNLOCK(tmp);
1086			if (n == NULL) {
1087				error = EINVAL;
1088				if (newname != NULL)
1089					    free(newname, M_TMPFSNAME);
1090				goto out_locked;
1091			}
1092			TMPFS_NODE_UNLOCK(n);
1093
1094			/* Adjust the parent pointer. */
1095			TMPFS_VALIDATE_DIR(fnode);
1096			TMPFS_NODE_LOCK(de->td_node);
1097			de->td_node->tn_dir.tn_parent = tdnode;
1098			TMPFS_NODE_UNLOCK(de->td_node);
1099
1100			/* As a result of changing the target of the '..'
1101			 * entry, the link count of the source and target
1102			 * directories has to be adjusted. */
1103			TMPFS_NODE_LOCK(tdnode);
1104			TMPFS_ASSERT_LOCKED(tdnode);
1105			tdnode->tn_links++;
1106			TMPFS_NODE_UNLOCK(tdnode);
1107
1108			TMPFS_NODE_LOCK(fdnode);
1109			TMPFS_ASSERT_LOCKED(fdnode);
1110			fdnode->tn_links--;
1111			TMPFS_NODE_UNLOCK(fdnode);
1112		}
1113
1114		/* Do the move: just remove the entry from the source directory
1115		 * and insert it into the target one. */
1116		tmpfs_dir_detach(fdvp, de);
1117		if (fcnp->cn_flags & DOWHITEOUT)
1118			tmpfs_dir_whiteout_add(fdvp, fcnp);
1119		if (tcnp->cn_flags & ISWHITEOUT)
1120			tmpfs_dir_whiteout_remove(tdvp, tcnp);
1121		tmpfs_dir_attach(tdvp, de);
1122	}
1123
1124	/* If the name has changed, we need to make it effective by changing
1125	 * it in the directory entry. */
1126	if (newname != NULL) {
1127		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1128
1129		free(de->td_name, M_TMPFSNAME);
1130		de->td_namelen = (uint16_t)tcnp->cn_namelen;
1131		memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen);
1132		de->td_name = newname;
1133
1134		fnode->tn_status |= TMPFS_NODE_CHANGED;
1135		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1136	}
1137
1138	/* If we are overwriting an entry, we have to remove the old one
1139	 * from the target directory. */
1140	if (tvp != NULL) {
1141		/* Remove the old entry from the target directory. */
1142		de = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1143		tmpfs_dir_detach(tdvp, de);
1144
1145		/* Free the directory entry we just deleted.  Note that the
1146		 * node referred by it will not be removed until the vnode is
1147		 * really reclaimed. */
1148		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE);
1149	}
1150
1151	error = 0;
1152
1153out_locked:
1154	if (fdnode != tdnode)
1155		VOP_UNLOCK(fdvp, 0);
1156
1157out:
1158	/* Release target nodes. */
1159	/* XXX: I don't understand when tdvp can be the same as tvp, but
1160	 * other code takes care of this... */
1161	if (tdvp == tvp)
1162		vrele(tdvp);
1163	else
1164		vput(tdvp);
1165	if (tvp != NULL)
1166		vput(tvp);
1167
1168	/* Release source nodes. */
1169	vrele(fdvp);
1170	vrele(fvp);
1171
1172	return error;
1173}
1174
1175/* --------------------------------------------------------------------- */
1176
1177static int
1178tmpfs_mkdir(struct vop_mkdir_args *v)
1179{
1180	struct vnode *dvp = v->a_dvp;
1181	struct vnode **vpp = v->a_vpp;
1182	struct componentname *cnp = v->a_cnp;
1183	struct vattr *vap = v->a_vap;
1184
1185	MPASS(vap->va_type == VDIR);
1186
1187	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1188}
1189
1190/* --------------------------------------------------------------------- */
1191
1192static int
1193tmpfs_rmdir(struct vop_rmdir_args *v)
1194{
1195	struct vnode *dvp = v->a_dvp;
1196	struct vnode *vp = v->a_vp;
1197
1198	int error;
1199	struct tmpfs_dirent *de;
1200	struct tmpfs_mount *tmp;
1201	struct tmpfs_node *dnode;
1202	struct tmpfs_node *node;
1203
1204	MPASS(VOP_ISLOCKED(dvp));
1205	MPASS(VOP_ISLOCKED(vp));
1206
1207	tmp = VFS_TO_TMPFS(dvp->v_mount);
1208	dnode = VP_TO_TMPFS_DIR(dvp);
1209	node = VP_TO_TMPFS_DIR(vp);
1210
1211	/* Directories with more than two entries ('.' and '..') cannot be
1212	 * removed. */
1213	 if (node->tn_size > 0) {
1214		 error = ENOTEMPTY;
1215		 goto out;
1216	 }
1217
1218	if ((dnode->tn_flags & APPEND)
1219	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1220		error = EPERM;
1221		goto out;
1222	}
1223
1224	/* This invariant holds only if we are not trying to remove "..".
1225	  * We checked for that above so this is safe now. */
1226	MPASS(node->tn_dir.tn_parent == dnode);
1227
1228	/* Get the directory entry associated with node (vp).  This was
1229	 * filled by tmpfs_lookup while looking up the entry. */
1230	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1231	MPASS(TMPFS_DIRENT_MATCHES(de,
1232	    v->a_cnp->cn_nameptr,
1233	    v->a_cnp->cn_namelen));
1234
1235	/* Check flags to see if we are allowed to remove the directory. */
1236	if (dnode->tn_flags & APPEND
1237		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1238		error = EPERM;
1239		goto out;
1240	}
1241
1242
1243	/* Detach the directory entry from the directory (dnode). */
1244	tmpfs_dir_detach(dvp, de);
1245	if (v->a_cnp->cn_flags & DOWHITEOUT)
1246		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1247
1248	/* No vnode should be allocated for this entry from this point */
1249	TMPFS_NODE_LOCK(node);
1250	TMPFS_ASSERT_ELOCKED(node);
1251	node->tn_links--;
1252	node->tn_dir.tn_parent = NULL;
1253	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1254	    TMPFS_NODE_MODIFIED;
1255
1256	TMPFS_NODE_UNLOCK(node);
1257
1258	TMPFS_NODE_LOCK(dnode);
1259	TMPFS_ASSERT_ELOCKED(dnode);
1260	dnode->tn_links--;
1261	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1262	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1263	TMPFS_NODE_UNLOCK(dnode);
1264
1265	cache_purge(dvp);
1266	cache_purge(vp);
1267
1268	/* Free the directory entry we just deleted.  Note that the node
1269	 * referred by it will not be removed until the vnode is really
1270	 * reclaimed. */
1271	tmpfs_free_dirent(tmp, de, TRUE);
1272
1273	/* Release the deleted vnode (will destroy the node, notify
1274	 * interested parties and clean it from the cache). */
1275
1276	dnode->tn_status |= TMPFS_NODE_CHANGED;
1277	tmpfs_update(dvp);
1278
1279	error = 0;
1280
1281out:
1282	return error;
1283}
1284
1285/* --------------------------------------------------------------------- */
1286
1287static int
1288tmpfs_symlink(struct vop_symlink_args *v)
1289{
1290	struct vnode *dvp = v->a_dvp;
1291	struct vnode **vpp = v->a_vpp;
1292	struct componentname *cnp = v->a_cnp;
1293	struct vattr *vap = v->a_vap;
1294	char *target = v->a_target;
1295
1296#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1297	MPASS(vap->va_type == VLNK);
1298#else
1299	vap->va_type = VLNK;
1300#endif
1301
1302	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1303}
1304
1305/* --------------------------------------------------------------------- */
1306
1307static int
1308tmpfs_readdir(struct vop_readdir_args *v)
1309{
1310	struct vnode *vp = v->a_vp;
1311	struct uio *uio = v->a_uio;
1312	int *eofflag = v->a_eofflag;
1313	u_long **cookies = v->a_cookies;
1314	int *ncookies = v->a_ncookies;
1315
1316	int error;
1317	off_t startoff;
1318	off_t cnt = 0;
1319	struct tmpfs_node *node;
1320
1321	/* This operation only makes sense on directory nodes. */
1322	if (vp->v_type != VDIR)
1323		return ENOTDIR;
1324
1325	node = VP_TO_TMPFS_DIR(vp);
1326
1327	startoff = uio->uio_offset;
1328
1329	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1330		error = tmpfs_dir_getdotdent(node, uio);
1331		if (error != 0)
1332			goto outok;
1333		cnt++;
1334	}
1335
1336	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1337		error = tmpfs_dir_getdotdotdent(node, uio);
1338		if (error != 0)
1339			goto outok;
1340		cnt++;
1341	}
1342
1343	error = tmpfs_dir_getdents(node, uio, &cnt);
1344
1345outok:
1346	MPASS(error >= -1);
1347
1348	if (error == -1)
1349		error = 0;
1350
1351	if (eofflag != NULL)
1352		*eofflag =
1353		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1354
1355	/* Update NFS-related variables. */
1356	if (error == 0 && cookies != NULL && ncookies != NULL) {
1357		off_t i;
1358		off_t off = startoff;
1359		struct tmpfs_dirent *de = NULL;
1360
1361		*ncookies = cnt;
1362		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1363
1364		for (i = 0; i < cnt; i++) {
1365			MPASS(off != TMPFS_DIRCOOKIE_EOF);
1366			if (off == TMPFS_DIRCOOKIE_DOT) {
1367				off = TMPFS_DIRCOOKIE_DOTDOT;
1368			} else {
1369				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1370					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1371				} else if (de != NULL) {
1372					de = TAILQ_NEXT(de, td_entries);
1373				} else {
1374					de = tmpfs_dir_lookupbycookie(node,
1375					    off);
1376					MPASS(de != NULL);
1377					de = TAILQ_NEXT(de, td_entries);
1378				}
1379				if (de == NULL)
1380					off = TMPFS_DIRCOOKIE_EOF;
1381				else
1382					off = tmpfs_dircookie(de);
1383			}
1384
1385			(*cookies)[i] = off;
1386		}
1387		MPASS(uio->uio_offset == off);
1388	}
1389
1390	return error;
1391}
1392
1393/* --------------------------------------------------------------------- */
1394
1395static int
1396tmpfs_readlink(struct vop_readlink_args *v)
1397{
1398	struct vnode *vp = v->a_vp;
1399	struct uio *uio = v->a_uio;
1400
1401	int error;
1402	struct tmpfs_node *node;
1403
1404	MPASS(uio->uio_offset == 0);
1405	MPASS(vp->v_type == VLNK);
1406
1407	node = VP_TO_TMPFS_NODE(vp);
1408
1409	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1410	    uio);
1411	node->tn_status |= TMPFS_NODE_ACCESSED;
1412
1413	return error;
1414}
1415
1416/* --------------------------------------------------------------------- */
1417
1418static int
1419tmpfs_inactive(struct vop_inactive_args *v)
1420{
1421	struct vnode *vp = v->a_vp;
1422	struct thread *l = v->a_td;
1423
1424	struct tmpfs_node *node;
1425
1426	MPASS(VOP_ISLOCKED(vp));
1427
1428	node = VP_TO_TMPFS_NODE(vp);
1429
1430	if (node->tn_links == 0)
1431		vrecycle(vp, l);
1432
1433	return 0;
1434}
1435
1436/* --------------------------------------------------------------------- */
1437
1438int
1439tmpfs_reclaim(struct vop_reclaim_args *v)
1440{
1441	struct vnode *vp = v->a_vp;
1442
1443	struct tmpfs_mount *tmp;
1444	struct tmpfs_node *node;
1445
1446	node = VP_TO_TMPFS_NODE(vp);
1447	tmp = VFS_TO_TMPFS(vp->v_mount);
1448
1449	vnode_destroy_vobject(vp);
1450	cache_purge(vp);
1451
1452	TMPFS_NODE_LOCK(node);
1453	TMPFS_ASSERT_ELOCKED(node);
1454	tmpfs_free_vp(vp);
1455
1456	/* If the node referenced by this vnode was deleted by the user,
1457	 * we must free its associated data structures (now that the vnode
1458	 * is being reclaimed). */
1459	if (node->tn_links == 0 &&
1460	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1461		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1462		TMPFS_NODE_UNLOCK(node);
1463		tmpfs_free_node(tmp, node);
1464	} else
1465		TMPFS_NODE_UNLOCK(node);
1466
1467	MPASS(vp->v_data == NULL);
1468	return 0;
1469}
1470
1471/* --------------------------------------------------------------------- */
1472
1473static int
1474tmpfs_print(struct vop_print_args *v)
1475{
1476	struct vnode *vp = v->a_vp;
1477
1478	struct tmpfs_node *node;
1479
1480	node = VP_TO_TMPFS_NODE(vp);
1481
1482	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1483	    node, node->tn_flags, node->tn_links);
1484	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1485	    ", status 0x%x\n",
1486	    node->tn_mode, node->tn_uid, node->tn_gid,
1487	    (uintmax_t)node->tn_size, node->tn_status);
1488
1489	if (vp->v_type == VFIFO)
1490		fifo_printinfo(vp);
1491
1492	printf("\n");
1493
1494	return 0;
1495}
1496
1497/* --------------------------------------------------------------------- */
1498
1499static int
1500tmpfs_pathconf(struct vop_pathconf_args *v)
1501{
1502	int name = v->a_name;
1503	register_t *retval = v->a_retval;
1504
1505	int error;
1506
1507	error = 0;
1508
1509	switch (name) {
1510	case _PC_LINK_MAX:
1511		*retval = LINK_MAX;
1512		break;
1513
1514	case _PC_NAME_MAX:
1515		*retval = NAME_MAX;
1516		break;
1517
1518	case _PC_PATH_MAX:
1519		*retval = PATH_MAX;
1520		break;
1521
1522	case _PC_PIPE_BUF:
1523		*retval = PIPE_BUF;
1524		break;
1525
1526	case _PC_CHOWN_RESTRICTED:
1527		*retval = 1;
1528		break;
1529
1530	case _PC_NO_TRUNC:
1531		*retval = 1;
1532		break;
1533
1534	case _PC_SYNC_IO:
1535		*retval = 1;
1536		break;
1537
1538	case _PC_FILESIZEBITS:
1539		*retval = 0; /* XXX Don't know which value should I return. */
1540		break;
1541
1542	default:
1543		error = EINVAL;
1544	}
1545
1546	return error;
1547}
1548
1549static int
1550tmpfs_vptofh(struct vop_vptofh_args *ap)
1551{
1552	struct tmpfs_fid *tfhp;
1553	struct tmpfs_node *node;
1554
1555	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1556	node = VP_TO_TMPFS_NODE(ap->a_vp);
1557
1558	tfhp->tf_len = sizeof(struct tmpfs_fid);
1559	tfhp->tf_id = node->tn_id;
1560	tfhp->tf_gen = node->tn_gen;
1561
1562	return (0);
1563}
1564
1565static int
1566tmpfs_whiteout(struct vop_whiteout_args *ap)
1567{
1568	struct vnode *dvp = ap->a_dvp;
1569	struct componentname *cnp = ap->a_cnp;
1570	struct tmpfs_dirent *de;
1571
1572	switch (ap->a_flags) {
1573	case LOOKUP:
1574		return (0);
1575	case CREATE:
1576		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1577		if (de != NULL)
1578			return (de->td_node == NULL ? 0 : EEXIST);
1579		return (tmpfs_dir_whiteout_add(dvp, cnp));
1580	case DELETE:
1581		tmpfs_dir_whiteout_remove(dvp, cnp);
1582		return (0);
1583	default:
1584		panic("tmpfs_whiteout: unknown op");
1585	}
1586}
1587
1588/* --------------------------------------------------------------------- */
1589
1590/*
1591 * vnode operations vector used for files stored in a tmpfs file system.
1592 */
1593struct vop_vector tmpfs_vnodeop_entries = {
1594	.vop_default =			&default_vnodeops,
1595	.vop_lookup =			vfs_cache_lookup,
1596	.vop_cachedlookup =		tmpfs_lookup,
1597	.vop_create =			tmpfs_create,
1598	.vop_mknod =			tmpfs_mknod,
1599	.vop_open =			tmpfs_open,
1600	.vop_close =			tmpfs_close,
1601	.vop_access =			tmpfs_access,
1602	.vop_getattr =			tmpfs_getattr,
1603	.vop_setattr =			tmpfs_setattr,
1604	.vop_read =			tmpfs_read,
1605	.vop_write =			tmpfs_write,
1606	.vop_fsync =			tmpfs_fsync,
1607	.vop_remove =			tmpfs_remove,
1608	.vop_link =			tmpfs_link,
1609	.vop_rename =			tmpfs_rename,
1610	.vop_mkdir =			tmpfs_mkdir,
1611	.vop_rmdir =			tmpfs_rmdir,
1612	.vop_symlink =			tmpfs_symlink,
1613	.vop_readdir =			tmpfs_readdir,
1614	.vop_readlink =			tmpfs_readlink,
1615	.vop_inactive =			tmpfs_inactive,
1616	.vop_reclaim =			tmpfs_reclaim,
1617	.vop_print =			tmpfs_print,
1618	.vop_pathconf =			tmpfs_pathconf,
1619	.vop_vptofh =			tmpfs_vptofh,
1620	.vop_whiteout =			tmpfs_whiteout,
1621	.vop_bmap =			VOP_EOPNOTSUPP,
1622};
1623
1624