tmpfs_vnops.c revision 313094
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: stable/10/sys/fs/tmpfs/tmpfs_vnops.c 313094 2017-02-02 13:38:01Z kib $");
38
39#include <sys/param.h>
40#include <sys/fcntl.h>
41#include <sys/lockf.h>
42#include <sys/lock.h>
43#include <sys/namei.h>
44#include <sys/priv.h>
45#include <sys/proc.h>
46#include <sys/rwlock.h>
47#include <sys/sched.h>
48#include <sys/stat.h>
49#include <sys/systm.h>
50#include <sys/sysctl.h>
51#include <sys/unistd.h>
52#include <sys/vnode.h>
53
54#include <vm/vm.h>
55#include <vm/vm_param.h>
56#include <vm/vm_object.h>
57#include <vm/vm_page.h>
58#include <vm/vm_pager.h>
59
60#include <fs/tmpfs/tmpfs_vnops.h>
61#include <fs/tmpfs/tmpfs.h>
62
63SYSCTL_DECL(_vfs_tmpfs);
64
65static volatile int tmpfs_rename_restarts;
66SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD,
67    __DEVOLATILE(int *, &tmpfs_rename_restarts), 0,
68    "Times rename had to restart due to lock contention");
69
70static int
71tmpfs_vn_get_ino_alloc(struct mount *mp, void *arg, int lkflags,
72    struct vnode **rvp)
73{
74
75	return (tmpfs_alloc_vp(mp, arg, lkflags, rvp));
76}
77
78static int
79tmpfs_lookup(struct vop_cachedlookup_args *v)
80{
81	struct vnode *dvp = v->a_dvp;
82	struct vnode **vpp = v->a_vpp;
83	struct componentname *cnp = v->a_cnp;
84	struct tmpfs_dirent *de;
85	struct tmpfs_node *dnode, *pnode;
86	struct tmpfs_mount *tm;
87	int error;
88
89	dnode = VP_TO_TMPFS_DIR(dvp);
90	*vpp = NULLVP;
91
92	/* Check accessibility of requested node as a first step. */
93	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
94	if (error != 0)
95		goto out;
96
97	/* We cannot be requesting the parent directory of the root node. */
98	MPASS(IMPLIES(dnode->tn_type == VDIR &&
99	    dnode->tn_dir.tn_parent == dnode,
100	    !(cnp->cn_flags & ISDOTDOT)));
101
102	TMPFS_ASSERT_LOCKED(dnode);
103	if (dnode->tn_dir.tn_parent == NULL) {
104		error = ENOENT;
105		goto out;
106	}
107	if (cnp->cn_flags & ISDOTDOT) {
108		tm = VFS_TO_TMPFS(dvp->v_mount);
109		pnode = dnode->tn_dir.tn_parent;
110		tmpfs_ref_node(pnode);
111		error = vn_vget_ino_gen(dvp, tmpfs_vn_get_ino_alloc,
112		    pnode, cnp->cn_lkflags, vpp);
113		tmpfs_free_node(tm, pnode);
114		if (error != 0)
115			goto out;
116	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
117		VREF(dvp);
118		*vpp = dvp;
119		error = 0;
120	} else {
121		de = tmpfs_dir_lookup(dnode, NULL, cnp);
122		if (de != NULL && de->td_node == NULL)
123			cnp->cn_flags |= ISWHITEOUT;
124		if (de == NULL || de->td_node == NULL) {
125			/*
126			 * The entry was not found in the directory.
127			 * This is OK if we are creating or renaming an
128			 * entry and are working on the last component of
129			 * the path name.
130			 */
131			if ((cnp->cn_flags & ISLASTCN) &&
132			    (cnp->cn_nameiop == CREATE || \
133			    cnp->cn_nameiop == RENAME ||
134			    (cnp->cn_nameiop == DELETE &&
135			    cnp->cn_flags & DOWHITEOUT &&
136			    cnp->cn_flags & ISWHITEOUT))) {
137				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
138				    cnp->cn_thread);
139				if (error != 0)
140					goto out;
141
142				/*
143				 * Keep the component name in the buffer for
144				 * future uses.
145				 */
146				cnp->cn_flags |= SAVENAME;
147
148				error = EJUSTRETURN;
149			} else
150				error = ENOENT;
151		} else {
152			struct tmpfs_node *tnode;
153
154			/*
155			 * The entry was found, so get its associated
156			 * tmpfs_node.
157			 */
158			tnode = de->td_node;
159
160			/*
161			 * If we are not at the last path component and
162			 * found a non-directory or non-link entry (which
163			 * may itself be pointing to a directory), raise
164			 * an error.
165			 */
166			if ((tnode->tn_type != VDIR &&
167			    tnode->tn_type != VLNK) &&
168			    !(cnp->cn_flags & ISLASTCN)) {
169				error = ENOTDIR;
170				goto out;
171			}
172
173			/*
174			 * If we are deleting or renaming the entry, keep
175			 * track of its tmpfs_dirent so that it can be
176			 * easily deleted later.
177			 */
178			if ((cnp->cn_flags & ISLASTCN) &&
179			    (cnp->cn_nameiop == DELETE ||
180			    cnp->cn_nameiop == RENAME)) {
181				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
182				    cnp->cn_thread);
183				if (error != 0)
184					goto out;
185
186				/* Allocate a new vnode on the matching entry. */
187				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
188				    cnp->cn_lkflags, vpp);
189				if (error != 0)
190					goto out;
191
192				if ((dnode->tn_mode & S_ISTXT) &&
193				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred,
194				  cnp->cn_thread) && VOP_ACCESS(*vpp, VADMIN,
195				  cnp->cn_cred, cnp->cn_thread)) {
196					error = EPERM;
197					vput(*vpp);
198					*vpp = NULL;
199					goto out;
200				}
201				cnp->cn_flags |= SAVENAME;
202			} else {
203				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
204				    cnp->cn_lkflags, vpp);
205				if (error != 0)
206					goto out;
207			}
208		}
209	}
210
211	/*
212	 * Store the result of this lookup in the cache.  Avoid this if the
213	 * request was for creation, as it does not improve timings on
214	 * emprical tests.
215	 */
216	if ((cnp->cn_flags & MAKEENTRY) != 0)
217		cache_enter(dvp, *vpp, cnp);
218
219out:
220	/*
221	 * If there were no errors, *vpp cannot be null and it must be
222	 * locked.
223	 */
224	MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
225
226	return (error);
227}
228
229static int
230tmpfs_create(struct vop_create_args *v)
231{
232	struct vnode *dvp = v->a_dvp;
233	struct vnode **vpp = v->a_vpp;
234	struct componentname *cnp = v->a_cnp;
235	struct vattr *vap = v->a_vap;
236	int error;
237
238	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
239
240	error = tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
241	if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0)
242		cache_enter(dvp, *vpp, cnp);
243	return (error);
244}
245
246static int
247tmpfs_mknod(struct vop_mknod_args *v)
248{
249	struct vnode *dvp = v->a_dvp;
250	struct vnode **vpp = v->a_vpp;
251	struct componentname *cnp = v->a_cnp;
252	struct vattr *vap = v->a_vap;
253
254	if (vap->va_type != VBLK && vap->va_type != VCHR &&
255	    vap->va_type != VFIFO)
256		return EINVAL;
257
258	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
259}
260
261static int
262tmpfs_open(struct vop_open_args *v)
263{
264	struct vnode *vp = v->a_vp;
265	int mode = v->a_mode;
266
267	int error;
268	struct tmpfs_node *node;
269
270	MPASS(VOP_ISLOCKED(vp));
271
272	node = VP_TO_TMPFS_NODE(vp);
273
274	/* The file is still active but all its names have been removed
275	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
276	 * it is about to die. */
277	if (node->tn_links < 1)
278		return (ENOENT);
279
280	/* If the file is marked append-only, deny write requests. */
281	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
282		error = EPERM;
283	else {
284		error = 0;
285		/* For regular files, the call below is nop. */
286		KASSERT(vp->v_type != VREG || (node->tn_reg.tn_aobj->flags &
287		    OBJ_DEAD) == 0, ("dead object"));
288		vnode_create_vobject(vp, node->tn_size, v->a_td);
289	}
290
291	MPASS(VOP_ISLOCKED(vp));
292	return error;
293}
294
295static int
296tmpfs_close(struct vop_close_args *v)
297{
298	struct vnode *vp = v->a_vp;
299
300	/* Update node times. */
301	tmpfs_update(vp);
302
303	return (0);
304}
305
306int
307tmpfs_access(struct vop_access_args *v)
308{
309	struct vnode *vp = v->a_vp;
310	accmode_t accmode = v->a_accmode;
311	struct ucred *cred = v->a_cred;
312
313	int error;
314	struct tmpfs_node *node;
315
316	MPASS(VOP_ISLOCKED(vp));
317
318	node = VP_TO_TMPFS_NODE(vp);
319
320	switch (vp->v_type) {
321	case VDIR:
322		/* FALLTHROUGH */
323	case VLNK:
324		/* FALLTHROUGH */
325	case VREG:
326		if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
327			error = EROFS;
328			goto out;
329		}
330		break;
331
332	case VBLK:
333		/* FALLTHROUGH */
334	case VCHR:
335		/* FALLTHROUGH */
336	case VSOCK:
337		/* FALLTHROUGH */
338	case VFIFO:
339		break;
340
341	default:
342		error = EINVAL;
343		goto out;
344	}
345
346	if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
347		error = EPERM;
348		goto out;
349	}
350
351	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
352	    node->tn_gid, accmode, cred, NULL);
353
354out:
355	MPASS(VOP_ISLOCKED(vp));
356
357	return error;
358}
359
360int
361tmpfs_getattr(struct vop_getattr_args *v)
362{
363	struct vnode *vp = v->a_vp;
364	struct vattr *vap = v->a_vap;
365
366	struct tmpfs_node *node;
367
368	node = VP_TO_TMPFS_NODE(vp);
369
370	tmpfs_update(vp);
371
372	vap->va_type = vp->v_type;
373	vap->va_mode = node->tn_mode;
374	vap->va_nlink = node->tn_links;
375	vap->va_uid = node->tn_uid;
376	vap->va_gid = node->tn_gid;
377	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
378	vap->va_fileid = node->tn_id;
379	vap->va_size = node->tn_size;
380	vap->va_blocksize = PAGE_SIZE;
381	vap->va_atime = node->tn_atime;
382	vap->va_mtime = node->tn_mtime;
383	vap->va_ctime = node->tn_ctime;
384	vap->va_birthtime = node->tn_birthtime;
385	vap->va_gen = node->tn_gen;
386	vap->va_flags = node->tn_flags;
387	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
388		node->tn_rdev : NODEV;
389	vap->va_bytes = round_page(node->tn_size);
390	vap->va_filerev = 0;
391
392	return 0;
393}
394
395int
396tmpfs_setattr(struct vop_setattr_args *v)
397{
398	struct vnode *vp = v->a_vp;
399	struct vattr *vap = v->a_vap;
400	struct ucred *cred = v->a_cred;
401	struct thread *td = curthread;
402
403	int error;
404
405	MPASS(VOP_ISLOCKED(vp));
406
407	error = 0;
408
409	/* Abort if any unsettable attribute is given. */
410	if (vap->va_type != VNON ||
411	    vap->va_nlink != VNOVAL ||
412	    vap->va_fsid != VNOVAL ||
413	    vap->va_fileid != VNOVAL ||
414	    vap->va_blocksize != VNOVAL ||
415	    vap->va_gen != VNOVAL ||
416	    vap->va_rdev != VNOVAL ||
417	    vap->va_bytes != VNOVAL)
418		error = EINVAL;
419
420	if (error == 0 && (vap->va_flags != VNOVAL))
421		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
422
423	if (error == 0 && (vap->va_size != VNOVAL))
424		error = tmpfs_chsize(vp, vap->va_size, cred, td);
425
426	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
427		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
428
429	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
430		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
431
432	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
433	    vap->va_atime.tv_nsec != VNOVAL) ||
434	    (vap->va_mtime.tv_sec != VNOVAL &&
435	    vap->va_mtime.tv_nsec != VNOVAL) ||
436	    (vap->va_birthtime.tv_sec != VNOVAL &&
437	    vap->va_birthtime.tv_nsec != VNOVAL)))
438		error = tmpfs_chtimes(vp, vap, cred, td);
439
440	/* Update the node times.  We give preference to the error codes
441	 * generated by this function rather than the ones that may arise
442	 * from tmpfs_update. */
443	tmpfs_update(vp);
444
445	MPASS(VOP_ISLOCKED(vp));
446
447	return error;
448}
449
450static int
451tmpfs_read(struct vop_read_args *v)
452{
453	struct vnode *vp;
454	struct uio *uio;
455	struct tmpfs_node *node;
456
457	vp = v->a_vp;
458	if (vp->v_type != VREG)
459		return (EISDIR);
460	uio = v->a_uio;
461	if (uio->uio_offset < 0)
462		return (EINVAL);
463	node = VP_TO_TMPFS_NODE(vp);
464	tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
465	return (uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio));
466}
467
468static int
469tmpfs_write(struct vop_write_args *v)
470{
471	struct vnode *vp;
472	struct uio *uio;
473	struct tmpfs_node *node;
474	off_t oldsize;
475	int error, ioflag;
476
477	vp = v->a_vp;
478	uio = v->a_uio;
479	ioflag = v->a_ioflag;
480	error = 0;
481	node = VP_TO_TMPFS_NODE(vp);
482	oldsize = node->tn_size;
483
484	if (uio->uio_offset < 0 || vp->v_type != VREG)
485		return (EINVAL);
486	if (uio->uio_resid == 0)
487		return (0);
488	if (ioflag & IO_APPEND)
489		uio->uio_offset = node->tn_size;
490	if (uio->uio_offset + uio->uio_resid >
491	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
492		return (EFBIG);
493	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
494		return (EFBIG);
495	if (uio->uio_offset + uio->uio_resid > node->tn_size) {
496		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid,
497		    FALSE);
498		if (error != 0)
499			goto out;
500	}
501
502	error = uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio);
503	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
504	    TMPFS_NODE_CHANGED;
505	if (node->tn_mode & (S_ISUID | S_ISGID)) {
506		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
507			node->tn_mode &= ~(S_ISUID | S_ISGID);
508	}
509	if (error != 0)
510		(void)tmpfs_reg_resize(vp, oldsize, TRUE);
511
512out:
513	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
514	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
515
516	return (error);
517}
518
519static int
520tmpfs_fsync(struct vop_fsync_args *v)
521{
522	struct vnode *vp = v->a_vp;
523
524	MPASS(VOP_ISLOCKED(vp));
525
526	tmpfs_check_mtime(vp);
527	tmpfs_update(vp);
528
529	return 0;
530}
531
532static int
533tmpfs_remove(struct vop_remove_args *v)
534{
535	struct vnode *dvp = v->a_dvp;
536	struct vnode *vp = v->a_vp;
537
538	int error;
539	struct tmpfs_dirent *de;
540	struct tmpfs_mount *tmp;
541	struct tmpfs_node *dnode;
542	struct tmpfs_node *node;
543
544	MPASS(VOP_ISLOCKED(dvp));
545	MPASS(VOP_ISLOCKED(vp));
546
547	if (vp->v_type == VDIR) {
548		error = EISDIR;
549		goto out;
550	}
551
552	dnode = VP_TO_TMPFS_DIR(dvp);
553	node = VP_TO_TMPFS_NODE(vp);
554	tmp = VFS_TO_TMPFS(vp->v_mount);
555	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
556	MPASS(de != NULL);
557
558	/* Files marked as immutable or append-only cannot be deleted. */
559	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
560	    (dnode->tn_flags & APPEND)) {
561		error = EPERM;
562		goto out;
563	}
564
565	/* Remove the entry from the directory; as it is a file, we do not
566	 * have to change the number of hard links of the directory. */
567	tmpfs_dir_detach(dvp, de);
568	if (v->a_cnp->cn_flags & DOWHITEOUT)
569		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
570
571	/* Free the directory entry we just deleted.  Note that the node
572	 * referred by it will not be removed until the vnode is really
573	 * reclaimed. */
574	tmpfs_free_dirent(tmp, de);
575
576	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
577	error = 0;
578
579out:
580
581	return error;
582}
583
584static int
585tmpfs_link(struct vop_link_args *v)
586{
587	struct vnode *dvp = v->a_tdvp;
588	struct vnode *vp = v->a_vp;
589	struct componentname *cnp = v->a_cnp;
590
591	int error;
592	struct tmpfs_dirent *de;
593	struct tmpfs_node *node;
594
595	MPASS(VOP_ISLOCKED(dvp));
596	MPASS(cnp->cn_flags & HASBUF);
597	MPASS(dvp != vp); /* XXX When can this be false? */
598	node = VP_TO_TMPFS_NODE(vp);
599
600	/* Ensure that we do not overflow the maximum number of links imposed
601	 * by the system. */
602	MPASS(node->tn_links <= LINK_MAX);
603	if (node->tn_links == LINK_MAX) {
604		error = EMLINK;
605		goto out;
606	}
607
608	/* We cannot create links of files marked immutable or append-only. */
609	if (node->tn_flags & (IMMUTABLE | APPEND)) {
610		error = EPERM;
611		goto out;
612	}
613
614	/* Allocate a new directory entry to represent the node. */
615	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
616	    cnp->cn_nameptr, cnp->cn_namelen, &de);
617	if (error != 0)
618		goto out;
619
620	/* Insert the new directory entry into the appropriate directory. */
621	if (cnp->cn_flags & ISWHITEOUT)
622		tmpfs_dir_whiteout_remove(dvp, cnp);
623	tmpfs_dir_attach(dvp, de);
624
625	/* vp link count has changed, so update node times. */
626	node->tn_status |= TMPFS_NODE_CHANGED;
627	tmpfs_update(vp);
628
629	error = 0;
630
631out:
632	return error;
633}
634
635/*
636 * We acquire all but fdvp locks using non-blocking acquisitions.  If we
637 * fail to acquire any lock in the path we will drop all held locks,
638 * acquire the new lock in a blocking fashion, and then release it and
639 * restart the rename.  This acquire/release step ensures that we do not
640 * spin on a lock waiting for release.  On error release all vnode locks
641 * and decrement references the way tmpfs_rename() would do.
642 */
643static int
644tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp,
645    struct vnode *tdvp, struct vnode **tvpp,
646    struct componentname *fcnp, struct componentname *tcnp)
647{
648	struct vnode *nvp;
649	struct mount *mp;
650	struct tmpfs_dirent *de;
651	int error, restarts = 0;
652
653	VOP_UNLOCK(tdvp, 0);
654	if (*tvpp != NULL && *tvpp != tdvp)
655		VOP_UNLOCK(*tvpp, 0);
656	mp = fdvp->v_mount;
657
658relock:
659	restarts += 1;
660	error = vn_lock(fdvp, LK_EXCLUSIVE);
661	if (error)
662		goto releout;
663	if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
664		VOP_UNLOCK(fdvp, 0);
665		error = vn_lock(tdvp, LK_EXCLUSIVE);
666		if (error)
667			goto releout;
668		VOP_UNLOCK(tdvp, 0);
669		goto relock;
670	}
671	/*
672	 * Re-resolve fvp to be certain it still exists and fetch the
673	 * correct vnode.
674	 */
675	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp);
676	if (de == NULL) {
677		VOP_UNLOCK(fdvp, 0);
678		VOP_UNLOCK(tdvp, 0);
679		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
680		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
681			error = EINVAL;
682		else
683			error = ENOENT;
684		goto releout;
685	}
686	error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
687	if (error != 0) {
688		VOP_UNLOCK(fdvp, 0);
689		VOP_UNLOCK(tdvp, 0);
690		if (error != EBUSY)
691			goto releout;
692		error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp);
693		if (error != 0)
694			goto releout;
695		VOP_UNLOCK(nvp, 0);
696		/*
697		 * Concurrent rename race.
698		 */
699		if (nvp == tdvp) {
700			vrele(nvp);
701			error = EINVAL;
702			goto releout;
703		}
704		vrele(*fvpp);
705		*fvpp = nvp;
706		goto relock;
707	}
708	vrele(*fvpp);
709	*fvpp = nvp;
710	VOP_UNLOCK(*fvpp, 0);
711	/*
712	 * Re-resolve tvp and acquire the vnode lock if present.
713	 */
714	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp);
715	/*
716	 * If tvp disappeared we just carry on.
717	 */
718	if (de == NULL && *tvpp != NULL) {
719		vrele(*tvpp);
720		*tvpp = NULL;
721	}
722	/*
723	 * Get the tvp ino if the lookup succeeded.  We may have to restart
724	 * if the non-blocking acquire fails.
725	 */
726	if (de != NULL) {
727		nvp = NULL;
728		error = tmpfs_alloc_vp(mp, de->td_node,
729		    LK_EXCLUSIVE | LK_NOWAIT, &nvp);
730		if (*tvpp != NULL)
731			vrele(*tvpp);
732		*tvpp = nvp;
733		if (error != 0) {
734			VOP_UNLOCK(fdvp, 0);
735			VOP_UNLOCK(tdvp, 0);
736			if (error != EBUSY)
737				goto releout;
738			error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE,
739			    &nvp);
740			if (error != 0)
741				goto releout;
742			VOP_UNLOCK(nvp, 0);
743			/*
744			 * fdvp contains fvp, thus tvp (=fdvp) is not empty.
745			 */
746			if (nvp == fdvp) {
747				error = ENOTEMPTY;
748				goto releout;
749			}
750			goto relock;
751		}
752	}
753	tmpfs_rename_restarts += restarts;
754
755	return (0);
756
757releout:
758	vrele(fdvp);
759	vrele(*fvpp);
760	vrele(tdvp);
761	if (*tvpp != NULL)
762		vrele(*tvpp);
763	tmpfs_rename_restarts += restarts;
764
765	return (error);
766}
767
768static int
769tmpfs_rename(struct vop_rename_args *v)
770{
771	struct vnode *fdvp = v->a_fdvp;
772	struct vnode *fvp = v->a_fvp;
773	struct componentname *fcnp = v->a_fcnp;
774	struct vnode *tdvp = v->a_tdvp;
775	struct vnode *tvp = v->a_tvp;
776	struct componentname *tcnp = v->a_tcnp;
777	struct mount *mp = NULL;
778
779	char *newname;
780	int error;
781	struct tmpfs_dirent *de;
782	struct tmpfs_mount *tmp;
783	struct tmpfs_node *fdnode;
784	struct tmpfs_node *fnode;
785	struct tmpfs_node *tnode;
786	struct tmpfs_node *tdnode;
787
788	MPASS(VOP_ISLOCKED(tdvp));
789	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
790	MPASS(fcnp->cn_flags & HASBUF);
791	MPASS(tcnp->cn_flags & HASBUF);
792
793	/* Disallow cross-device renames.
794	 * XXX Why isn't this done by the caller? */
795	if (fvp->v_mount != tdvp->v_mount ||
796	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
797		error = EXDEV;
798		goto out;
799	}
800
801	/* If source and target are the same file, there is nothing to do. */
802	if (fvp == tvp) {
803		error = 0;
804		goto out;
805	}
806
807	/* If we need to move the directory between entries, lock the
808	 * source so that we can safely operate on it. */
809	if (fdvp != tdvp && fdvp != tvp) {
810		if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
811			mp = tdvp->v_mount;
812			error = vfs_busy(mp, 0);
813			if (error != 0) {
814				mp = NULL;
815				goto out;
816			}
817			error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp,
818			    fcnp, tcnp);
819			if (error != 0) {
820				vfs_unbusy(mp);
821				return (error);
822			}
823			ASSERT_VOP_ELOCKED(fdvp,
824			    "tmpfs_rename: fdvp not locked");
825			ASSERT_VOP_ELOCKED(tdvp,
826			    "tmpfs_rename: tdvp not locked");
827			if (tvp != NULL)
828				ASSERT_VOP_ELOCKED(tvp,
829				    "tmpfs_rename: tvp not locked");
830			if (fvp == tvp) {
831				error = 0;
832				goto out_locked;
833			}
834		}
835	}
836
837	tmp = VFS_TO_TMPFS(tdvp->v_mount);
838	tdnode = VP_TO_TMPFS_DIR(tdvp);
839	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
840	fdnode = VP_TO_TMPFS_DIR(fdvp);
841	fnode = VP_TO_TMPFS_NODE(fvp);
842	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
843
844	/* Entry can disappear before we lock fdvp,
845	 * also avoid manipulating '.' and '..' entries. */
846	if (de == NULL) {
847		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
848		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
849			error = EINVAL;
850		else
851			error = ENOENT;
852		goto out_locked;
853	}
854	MPASS(de->td_node == fnode);
855
856	/* If re-naming a directory to another preexisting directory
857	 * ensure that the target directory is empty so that its
858	 * removal causes no side effects.
859	 * Kern_rename gurantees the destination to be a directory
860	 * if the source is one. */
861	if (tvp != NULL) {
862		MPASS(tnode != NULL);
863
864		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
865		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
866			error = EPERM;
867			goto out_locked;
868		}
869
870		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
871			if (tnode->tn_size > 0) {
872				error = ENOTEMPTY;
873				goto out_locked;
874			}
875		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
876			error = ENOTDIR;
877			goto out_locked;
878		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
879			error = EISDIR;
880			goto out_locked;
881		} else {
882			MPASS(fnode->tn_type != VDIR &&
883				tnode->tn_type != VDIR);
884		}
885	}
886
887	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
888	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
889		error = EPERM;
890		goto out_locked;
891	}
892
893	/* Ensure that we have enough memory to hold the new name, if it
894	 * has to be changed. */
895	if (fcnp->cn_namelen != tcnp->cn_namelen ||
896	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
897		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
898	} else
899		newname = NULL;
900
901	/* If the node is being moved to another directory, we have to do
902	 * the move. */
903	if (fdnode != tdnode) {
904		/* In case we are moving a directory, we have to adjust its
905		 * parent to point to the new parent. */
906		if (de->td_node->tn_type == VDIR) {
907			struct tmpfs_node *n;
908
909			/* Ensure the target directory is not a child of the
910			 * directory being moved.  Otherwise, we'd end up
911			 * with stale nodes. */
912			n = tdnode;
913			/* TMPFS_LOCK garanties that no nodes are freed while
914			 * traversing the list. Nodes can only be marked as
915			 * removed: tn_parent == NULL. */
916			TMPFS_LOCK(tmp);
917			TMPFS_NODE_LOCK(n);
918			while (n != n->tn_dir.tn_parent) {
919				struct tmpfs_node *parent;
920
921				if (n == fnode) {
922					TMPFS_NODE_UNLOCK(n);
923					TMPFS_UNLOCK(tmp);
924					error = EINVAL;
925					if (newname != NULL)
926						    free(newname, M_TMPFSNAME);
927					goto out_locked;
928				}
929				parent = n->tn_dir.tn_parent;
930				TMPFS_NODE_UNLOCK(n);
931				if (parent == NULL) {
932					n = NULL;
933					break;
934				}
935				TMPFS_NODE_LOCK(parent);
936				if (parent->tn_dir.tn_parent == NULL) {
937					TMPFS_NODE_UNLOCK(parent);
938					n = NULL;
939					break;
940				}
941				n = parent;
942			}
943			TMPFS_UNLOCK(tmp);
944			if (n == NULL) {
945				error = EINVAL;
946				if (newname != NULL)
947					    free(newname, M_TMPFSNAME);
948				goto out_locked;
949			}
950			TMPFS_NODE_UNLOCK(n);
951
952			/* Adjust the parent pointer. */
953			TMPFS_VALIDATE_DIR(fnode);
954			TMPFS_NODE_LOCK(de->td_node);
955			de->td_node->tn_dir.tn_parent = tdnode;
956			TMPFS_NODE_UNLOCK(de->td_node);
957
958			/* As a result of changing the target of the '..'
959			 * entry, the link count of the source and target
960			 * directories has to be adjusted. */
961			TMPFS_NODE_LOCK(tdnode);
962			TMPFS_ASSERT_LOCKED(tdnode);
963			tdnode->tn_links++;
964			TMPFS_NODE_UNLOCK(tdnode);
965
966			TMPFS_NODE_LOCK(fdnode);
967			TMPFS_ASSERT_LOCKED(fdnode);
968			fdnode->tn_links--;
969			TMPFS_NODE_UNLOCK(fdnode);
970		}
971	}
972
973	/* Do the move: just remove the entry from the source directory
974	 * and insert it into the target one. */
975	tmpfs_dir_detach(fdvp, de);
976
977	if (fcnp->cn_flags & DOWHITEOUT)
978		tmpfs_dir_whiteout_add(fdvp, fcnp);
979	if (tcnp->cn_flags & ISWHITEOUT)
980		tmpfs_dir_whiteout_remove(tdvp, tcnp);
981
982	/* If the name has changed, we need to make it effective by changing
983	 * it in the directory entry. */
984	if (newname != NULL) {
985		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
986
987		free(de->ud.td_name, M_TMPFSNAME);
988		de->ud.td_name = newname;
989		tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen);
990
991		fnode->tn_status |= TMPFS_NODE_CHANGED;
992		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
993	}
994
995	/* If we are overwriting an entry, we have to remove the old one
996	 * from the target directory. */
997	if (tvp != NULL) {
998		struct tmpfs_dirent *tde;
999
1000		/* Remove the old entry from the target directory. */
1001		tde = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1002		tmpfs_dir_detach(tdvp, tde);
1003
1004		/* Free the directory entry we just deleted.  Note that the
1005		 * node referred by it will not be removed until the vnode is
1006		 * really reclaimed. */
1007		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1008	}
1009
1010	tmpfs_dir_attach(tdvp, de);
1011
1012	cache_purge(fvp);
1013	if (tvp != NULL)
1014		cache_purge(tvp);
1015	cache_purge_negative(tdvp);
1016
1017	error = 0;
1018
1019out_locked:
1020	if (fdvp != tdvp && fdvp != tvp)
1021		VOP_UNLOCK(fdvp, 0);
1022
1023out:
1024	/* Release target nodes. */
1025	/* XXX: I don't understand when tdvp can be the same as tvp, but
1026	 * other code takes care of this... */
1027	if (tdvp == tvp)
1028		vrele(tdvp);
1029	else
1030		vput(tdvp);
1031	if (tvp != NULL)
1032		vput(tvp);
1033
1034	/* Release source nodes. */
1035	vrele(fdvp);
1036	vrele(fvp);
1037
1038	if (mp != NULL)
1039		vfs_unbusy(mp);
1040
1041	return error;
1042}
1043
1044static int
1045tmpfs_mkdir(struct vop_mkdir_args *v)
1046{
1047	struct vnode *dvp = v->a_dvp;
1048	struct vnode **vpp = v->a_vpp;
1049	struct componentname *cnp = v->a_cnp;
1050	struct vattr *vap = v->a_vap;
1051
1052	MPASS(vap->va_type == VDIR);
1053
1054	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1055}
1056
1057static int
1058tmpfs_rmdir(struct vop_rmdir_args *v)
1059{
1060	struct vnode *dvp = v->a_dvp;
1061	struct vnode *vp = v->a_vp;
1062
1063	int error;
1064	struct tmpfs_dirent *de;
1065	struct tmpfs_mount *tmp;
1066	struct tmpfs_node *dnode;
1067	struct tmpfs_node *node;
1068
1069	MPASS(VOP_ISLOCKED(dvp));
1070	MPASS(VOP_ISLOCKED(vp));
1071
1072	tmp = VFS_TO_TMPFS(dvp->v_mount);
1073	dnode = VP_TO_TMPFS_DIR(dvp);
1074	node = VP_TO_TMPFS_DIR(vp);
1075
1076	/* Directories with more than two entries ('.' and '..') cannot be
1077	 * removed. */
1078	 if (node->tn_size > 0) {
1079		 error = ENOTEMPTY;
1080		 goto out;
1081	 }
1082
1083	if ((dnode->tn_flags & APPEND)
1084	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1085		error = EPERM;
1086		goto out;
1087	}
1088
1089	/* This invariant holds only if we are not trying to remove "..".
1090	  * We checked for that above so this is safe now. */
1091	MPASS(node->tn_dir.tn_parent == dnode);
1092
1093	/* Get the directory entry associated with node (vp).  This was
1094	 * filled by tmpfs_lookup while looking up the entry. */
1095	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1096	MPASS(TMPFS_DIRENT_MATCHES(de,
1097	    v->a_cnp->cn_nameptr,
1098	    v->a_cnp->cn_namelen));
1099
1100	/* Check flags to see if we are allowed to remove the directory. */
1101	if ((dnode->tn_flags & APPEND) != 0 ||
1102	    (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) != 0) {
1103		error = EPERM;
1104		goto out;
1105	}
1106
1107
1108	/* Detach the directory entry from the directory (dnode). */
1109	tmpfs_dir_detach(dvp, de);
1110	if (v->a_cnp->cn_flags & DOWHITEOUT)
1111		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1112
1113	/* No vnode should be allocated for this entry from this point */
1114	TMPFS_NODE_LOCK(node);
1115	node->tn_links--;
1116	node->tn_dir.tn_parent = NULL;
1117	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1118	    TMPFS_NODE_MODIFIED;
1119
1120	TMPFS_NODE_UNLOCK(node);
1121
1122	TMPFS_NODE_LOCK(dnode);
1123	dnode->tn_links--;
1124	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1125	    TMPFS_NODE_MODIFIED;
1126	TMPFS_NODE_UNLOCK(dnode);
1127
1128	cache_purge(dvp);
1129	cache_purge(vp);
1130
1131	/* Free the directory entry we just deleted.  Note that the node
1132	 * referred by it will not be removed until the vnode is really
1133	 * reclaimed. */
1134	tmpfs_free_dirent(tmp, de);
1135
1136	/* Release the deleted vnode (will destroy the node, notify
1137	 * interested parties and clean it from the cache). */
1138
1139	dnode->tn_status |= TMPFS_NODE_CHANGED;
1140	tmpfs_update(dvp);
1141
1142	error = 0;
1143
1144out:
1145	return error;
1146}
1147
1148static int
1149tmpfs_symlink(struct vop_symlink_args *v)
1150{
1151	struct vnode *dvp = v->a_dvp;
1152	struct vnode **vpp = v->a_vpp;
1153	struct componentname *cnp = v->a_cnp;
1154	struct vattr *vap = v->a_vap;
1155	char *target = v->a_target;
1156
1157#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1158	MPASS(vap->va_type == VLNK);
1159#else
1160	vap->va_type = VLNK;
1161#endif
1162
1163	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1164}
1165
1166static int
1167tmpfs_readdir(struct vop_readdir_args *v)
1168{
1169	struct vnode *vp = v->a_vp;
1170	struct uio *uio = v->a_uio;
1171	int *eofflag = v->a_eofflag;
1172	u_long **cookies = v->a_cookies;
1173	int *ncookies = v->a_ncookies;
1174
1175	int error;
1176	ssize_t startresid;
1177	int maxcookies;
1178	struct tmpfs_node *node;
1179
1180	/* This operation only makes sense on directory nodes. */
1181	if (vp->v_type != VDIR)
1182		return ENOTDIR;
1183
1184	maxcookies = 0;
1185	node = VP_TO_TMPFS_DIR(vp);
1186
1187	startresid = uio->uio_resid;
1188
1189	/* Allocate cookies for NFS and compat modules. */
1190	if (cookies != NULL && ncookies != NULL) {
1191		maxcookies = howmany(node->tn_size,
1192		    sizeof(struct tmpfs_dirent)) + 2;
1193		*cookies = malloc(maxcookies * sizeof(**cookies), M_TEMP,
1194		    M_WAITOK);
1195		*ncookies = 0;
1196	}
1197
1198	if (cookies == NULL)
1199		error = tmpfs_dir_getdents(node, uio, 0, NULL, NULL);
1200	else
1201		error = tmpfs_dir_getdents(node, uio, maxcookies, *cookies,
1202		    ncookies);
1203
1204	/* Buffer was filled without hitting EOF. */
1205	if (error == EJUSTRETURN)
1206		error = (uio->uio_resid != startresid) ? 0 : EINVAL;
1207
1208	if (error != 0 && cookies != NULL && ncookies != NULL) {
1209		free(*cookies, M_TEMP);
1210		*cookies = NULL;
1211		*ncookies = 0;
1212	}
1213
1214	if (eofflag != NULL)
1215		*eofflag =
1216		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1217
1218	return error;
1219}
1220
1221static int
1222tmpfs_readlink(struct vop_readlink_args *v)
1223{
1224	struct vnode *vp = v->a_vp;
1225	struct uio *uio = v->a_uio;
1226
1227	int error;
1228	struct tmpfs_node *node;
1229
1230	MPASS(uio->uio_offset == 0);
1231	MPASS(vp->v_type == VLNK);
1232
1233	node = VP_TO_TMPFS_NODE(vp);
1234
1235	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1236	    uio);
1237	tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1238
1239	return (error);
1240}
1241
1242static int
1243tmpfs_inactive(struct vop_inactive_args *v)
1244{
1245	struct vnode *vp;
1246	struct tmpfs_node *node;
1247
1248	vp = v->a_vp;
1249	node = VP_TO_TMPFS_NODE(vp);
1250	if (node->tn_links == 0)
1251		vrecycle(vp);
1252	else
1253		tmpfs_check_mtime(vp);
1254	return (0);
1255}
1256
1257int
1258tmpfs_reclaim(struct vop_reclaim_args *v)
1259{
1260	struct vnode *vp = v->a_vp;
1261
1262	struct tmpfs_mount *tmp;
1263	struct tmpfs_node *node;
1264
1265	node = VP_TO_TMPFS_NODE(vp);
1266	tmp = VFS_TO_TMPFS(vp->v_mount);
1267
1268	if (vp->v_type == VREG)
1269		tmpfs_destroy_vobject(vp, node->tn_reg.tn_aobj);
1270	else
1271		vnode_destroy_vobject(vp);
1272	vp->v_object = NULL;
1273	cache_purge(vp);
1274
1275	TMPFS_NODE_LOCK(node);
1276	tmpfs_free_vp(vp);
1277
1278	/* If the node referenced by this vnode was deleted by the user,
1279	 * we must free its associated data structures (now that the vnode
1280	 * is being reclaimed). */
1281	if (node->tn_links == 0 &&
1282	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1283		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1284		TMPFS_NODE_UNLOCK(node);
1285		tmpfs_free_node(tmp, node);
1286	} else
1287		TMPFS_NODE_UNLOCK(node);
1288
1289	MPASS(vp->v_data == NULL);
1290	return 0;
1291}
1292
1293static int
1294tmpfs_print(struct vop_print_args *v)
1295{
1296	struct vnode *vp = v->a_vp;
1297
1298	struct tmpfs_node *node;
1299
1300	node = VP_TO_TMPFS_NODE(vp);
1301
1302	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%lx, links %d\n",
1303	    node, node->tn_flags, node->tn_links);
1304	printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n",
1305	    node->tn_mode, node->tn_uid, node->tn_gid,
1306	    (intmax_t)node->tn_size, node->tn_status);
1307
1308	if (vp->v_type == VFIFO)
1309		fifo_printinfo(vp);
1310
1311	printf("\n");
1312
1313	return 0;
1314}
1315
1316static int
1317tmpfs_pathconf(struct vop_pathconf_args *v)
1318{
1319	int name = v->a_name;
1320	register_t *retval = v->a_retval;
1321
1322	int error;
1323
1324	error = 0;
1325
1326	switch (name) {
1327	case _PC_LINK_MAX:
1328		*retval = LINK_MAX;
1329		break;
1330
1331	case _PC_NAME_MAX:
1332		*retval = NAME_MAX;
1333		break;
1334
1335	case _PC_PATH_MAX:
1336		*retval = PATH_MAX;
1337		break;
1338
1339	case _PC_PIPE_BUF:
1340		*retval = PIPE_BUF;
1341		break;
1342
1343	case _PC_CHOWN_RESTRICTED:
1344		*retval = 1;
1345		break;
1346
1347	case _PC_NO_TRUNC:
1348		*retval = 1;
1349		break;
1350
1351	case _PC_SYNC_IO:
1352		*retval = 1;
1353		break;
1354
1355	case _PC_FILESIZEBITS:
1356		*retval = 0; /* XXX Don't know which value should I return. */
1357		break;
1358
1359	default:
1360		error = EINVAL;
1361	}
1362
1363	return error;
1364}
1365
1366static int
1367tmpfs_vptofh(struct vop_vptofh_args *ap)
1368{
1369	struct tmpfs_fid *tfhp;
1370	struct tmpfs_node *node;
1371
1372	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1373	node = VP_TO_TMPFS_NODE(ap->a_vp);
1374
1375	tfhp->tf_len = sizeof(struct tmpfs_fid);
1376	tfhp->tf_id = node->tn_id;
1377	tfhp->tf_gen = node->tn_gen;
1378
1379	return (0);
1380}
1381
1382static int
1383tmpfs_whiteout(struct vop_whiteout_args *ap)
1384{
1385	struct vnode *dvp = ap->a_dvp;
1386	struct componentname *cnp = ap->a_cnp;
1387	struct tmpfs_dirent *de;
1388
1389	switch (ap->a_flags) {
1390	case LOOKUP:
1391		return (0);
1392	case CREATE:
1393		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1394		if (de != NULL)
1395			return (de->td_node == NULL ? 0 : EEXIST);
1396		return (tmpfs_dir_whiteout_add(dvp, cnp));
1397	case DELETE:
1398		tmpfs_dir_whiteout_remove(dvp, cnp);
1399		return (0);
1400	default:
1401		panic("tmpfs_whiteout: unknown op");
1402	}
1403}
1404
1405static int
1406tmpfs_vptocnp_dir(struct tmpfs_node *tn, struct tmpfs_node *tnp,
1407    struct tmpfs_dirent **pde)
1408{
1409	struct tmpfs_dir_cursor dc;
1410	struct tmpfs_dirent *de;
1411
1412	for (de = tmpfs_dir_first(tnp, &dc); de != NULL;
1413	     de = tmpfs_dir_next(tnp, &dc)) {
1414		if (de->td_node == tn) {
1415			*pde = de;
1416			return (0);
1417		}
1418	}
1419	return (ENOENT);
1420}
1421
1422static int
1423tmpfs_vptocnp_fill(struct vnode *vp, struct tmpfs_node *tn,
1424    struct tmpfs_node *tnp, char *buf, int *buflen, struct vnode **dvp)
1425{
1426	struct tmpfs_dirent *de;
1427	int error, i;
1428
1429	error = vn_vget_ino_gen(vp, tmpfs_vn_get_ino_alloc, tnp, LK_SHARED,
1430	    dvp);
1431	if (error != 0)
1432		return (error);
1433	error = tmpfs_vptocnp_dir(tn, tnp, &de);
1434	if (error == 0) {
1435		i = *buflen;
1436		i -= de->td_namelen;
1437		if (i < 0) {
1438			error = ENOMEM;
1439		} else {
1440			bcopy(de->ud.td_name, buf + i, de->td_namelen);
1441			*buflen = i;
1442		}
1443	}
1444	if (error == 0) {
1445		if (vp != *dvp)
1446			VOP_UNLOCK(*dvp, 0);
1447	} else {
1448		if (vp != *dvp)
1449			vput(*dvp);
1450		else
1451			vrele(vp);
1452	}
1453	return (error);
1454}
1455
1456static int
1457tmpfs_vptocnp(struct vop_vptocnp_args *ap)
1458{
1459	struct vnode *vp, **dvp;
1460	struct tmpfs_node *tn, *tnp, *tnp1;
1461	struct tmpfs_dirent *de;
1462	struct tmpfs_mount *tm;
1463	char *buf;
1464	int *buflen;
1465	int error;
1466
1467	vp = ap->a_vp;
1468	dvp = ap->a_vpp;
1469	buf = ap->a_buf;
1470	buflen = ap->a_buflen;
1471
1472	tm = VFS_TO_TMPFS(vp->v_mount);
1473	tn = VP_TO_TMPFS_NODE(vp);
1474	if (tn->tn_type == VDIR) {
1475		tnp = tn->tn_dir.tn_parent;
1476		if (tnp == NULL)
1477			return (ENOENT);
1478		tmpfs_ref_node(tnp);
1479		error = tmpfs_vptocnp_fill(vp, tn, tn->tn_dir.tn_parent, buf,
1480		    buflen, dvp);
1481		tmpfs_free_node(tm, tnp);
1482		return (error);
1483	}
1484restart:
1485	TMPFS_LOCK(tm);
1486	LIST_FOREACH_SAFE(tnp, &tm->tm_nodes_used, tn_entries, tnp1) {
1487		if (tnp->tn_type != VDIR)
1488			continue;
1489		TMPFS_NODE_LOCK(tnp);
1490		tmpfs_ref_node_locked(tnp);
1491
1492		/*
1493		 * tn_vnode cannot be instantiated while we hold the
1494		 * node lock, so the directory cannot be changed while
1495		 * we iterate over it.  Do this to avoid instantiating
1496		 * vnode for directories which cannot point to our
1497		 * node.
1498		 */
1499		error = tnp->tn_vnode == NULL ? tmpfs_vptocnp_dir(tn, tnp,
1500		    &de) : 0;
1501
1502		if (error == 0) {
1503			TMPFS_NODE_UNLOCK(tnp);
1504			TMPFS_UNLOCK(tm);
1505			error = tmpfs_vptocnp_fill(vp, tn, tnp, buf, buflen,
1506			    dvp);
1507			if (error == 0) {
1508				tmpfs_free_node(tm, tnp);
1509				return (0);
1510			}
1511			if ((vp->v_iflag & VI_DOOMED) != 0) {
1512				tmpfs_free_node(tm, tnp);
1513				return (ENOENT);
1514			}
1515			TMPFS_LOCK(tm);
1516			TMPFS_NODE_LOCK(tnp);
1517		}
1518		if (tmpfs_free_node_locked(tm, tnp, false)) {
1519			goto restart;
1520		} else {
1521			KASSERT(tnp->tn_refcount > 0,
1522			    ("node %p refcount zero", tnp));
1523			tnp1 = LIST_NEXT(tnp, tn_entries);
1524			TMPFS_NODE_UNLOCK(tnp);
1525		}
1526	}
1527	TMPFS_UNLOCK(tm);
1528	return (ENOENT);
1529}
1530
1531/*
1532 * Vnode operations vector used for files stored in a tmpfs file system.
1533 */
1534struct vop_vector tmpfs_vnodeop_entries = {
1535	.vop_default =			&default_vnodeops,
1536	.vop_lookup =			vfs_cache_lookup,
1537	.vop_cachedlookup =		tmpfs_lookup,
1538	.vop_create =			tmpfs_create,
1539	.vop_mknod =			tmpfs_mknod,
1540	.vop_open =			tmpfs_open,
1541	.vop_close =			tmpfs_close,
1542	.vop_access =			tmpfs_access,
1543	.vop_getattr =			tmpfs_getattr,
1544	.vop_setattr =			tmpfs_setattr,
1545	.vop_read =			tmpfs_read,
1546	.vop_write =			tmpfs_write,
1547	.vop_fsync =			tmpfs_fsync,
1548	.vop_remove =			tmpfs_remove,
1549	.vop_link =			tmpfs_link,
1550	.vop_rename =			tmpfs_rename,
1551	.vop_mkdir =			tmpfs_mkdir,
1552	.vop_rmdir =			tmpfs_rmdir,
1553	.vop_symlink =			tmpfs_symlink,
1554	.vop_readdir =			tmpfs_readdir,
1555	.vop_readlink =			tmpfs_readlink,
1556	.vop_inactive =			tmpfs_inactive,
1557	.vop_reclaim =			tmpfs_reclaim,
1558	.vop_print =			tmpfs_print,
1559	.vop_pathconf =			tmpfs_pathconf,
1560	.vop_vptofh =			tmpfs_vptofh,
1561	.vop_whiteout =			tmpfs_whiteout,
1562	.vop_bmap =			VOP_EOPNOTSUPP,
1563	.vop_vptocnp =			tmpfs_vptocnp,
1564};
1565
1566