xref: /freebsd/sys/fs/tmpfs/tmpfs_vnops.c (revision 0cdfe2ae89834a1a4b3468bfac870941ee17f2d5)
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("$FreeBSD$");
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/sf_buf.h>
49 #include <sys/stat.h>
50 #include <sys/systm.h>
51 #include <sys/sysctl.h>
52 #include <sys/unistd.h>
53 #include <sys/vnode.h>
54 
55 #include <vm/vm.h>
56 #include <vm/vm_param.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_pager.h>
60 
61 #include <fs/tmpfs/tmpfs_vnops.h>
62 #include <fs/tmpfs/tmpfs.h>
63 
64 SYSCTL_DECL(_vfs_tmpfs);
65 
66 static volatile int tmpfs_rename_restarts;
67 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD,
68     __DEVOLATILE(int *, &tmpfs_rename_restarts), 0,
69     "Times rename had to restart due to lock contention");
70 
71 /* --------------------------------------------------------------------- */
72 
73 static int
74 tmpfs_lookup(struct vop_cachedlookup_args *v)
75 {
76 	struct vnode *dvp = v->a_dvp;
77 	struct vnode **vpp = v->a_vpp;
78 	struct componentname *cnp = v->a_cnp;
79 
80 	int error;
81 	struct tmpfs_dirent *de;
82 	struct tmpfs_node *dnode;
83 
84 	dnode = VP_TO_TMPFS_DIR(dvp);
85 	*vpp = NULLVP;
86 
87 	/* Check accessibility of requested node as a first step. */
88 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
89 	if (error != 0)
90 		goto out;
91 
92 	/* We cannot be requesting the parent directory of the root node. */
93 	MPASS(IMPLIES(dnode->tn_type == VDIR &&
94 	    dnode->tn_dir.tn_parent == dnode,
95 	    !(cnp->cn_flags & ISDOTDOT)));
96 
97 	TMPFS_ASSERT_LOCKED(dnode);
98 	if (dnode->tn_dir.tn_parent == NULL) {
99 		error = ENOENT;
100 		goto out;
101 	}
102 	if (cnp->cn_flags & ISDOTDOT) {
103 		int ltype = 0;
104 
105 		ltype = VOP_ISLOCKED(dvp);
106 		vhold(dvp);
107 		VOP_UNLOCK(dvp, 0);
108 		/* Allocate a new vnode on the matching entry. */
109 		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
110 		    cnp->cn_lkflags, vpp);
111 
112 		vn_lock(dvp, ltype | LK_RETRY);
113 		vdrop(dvp);
114 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
115 		VREF(dvp);
116 		*vpp = dvp;
117 		error = 0;
118 	} else {
119 		de = tmpfs_dir_lookup(dnode, NULL, cnp);
120 		if (de != NULL && de->td_node == NULL)
121 			cnp->cn_flags |= ISWHITEOUT;
122 		if (de == NULL || de->td_node == NULL) {
123 			/* The entry was not found in the directory.
124 			 * This is OK if we are creating or renaming an
125 			 * entry and are working on the last component of
126 			 * the path name. */
127 			if ((cnp->cn_flags & ISLASTCN) &&
128 			    (cnp->cn_nameiop == CREATE || \
129 			    cnp->cn_nameiop == RENAME ||
130 			    (cnp->cn_nameiop == DELETE &&
131 			    cnp->cn_flags & DOWHITEOUT &&
132 			    cnp->cn_flags & ISWHITEOUT))) {
133 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
134 				    cnp->cn_thread);
135 				if (error != 0)
136 					goto out;
137 
138 				/* Keep the component name in the buffer for
139 				 * future uses. */
140 				cnp->cn_flags |= SAVENAME;
141 
142 				error = EJUSTRETURN;
143 			} else
144 				error = ENOENT;
145 		} else {
146 			struct tmpfs_node *tnode;
147 
148 			/* The entry was found, so get its associated
149 			 * tmpfs_node. */
150 			tnode = de->td_node;
151 
152 			/* If we are not at the last path component and
153 			 * found a non-directory or non-link entry (which
154 			 * may itself be pointing to a directory), raise
155 			 * an error. */
156 			if ((tnode->tn_type != VDIR &&
157 			    tnode->tn_type != VLNK) &&
158 			    !(cnp->cn_flags & ISLASTCN)) {
159 				error = ENOTDIR;
160 				goto out;
161 			}
162 
163 			/* If we are deleting or renaming the entry, keep
164 			 * track of its tmpfs_dirent so that it can be
165 			 * easily deleted later. */
166 			if ((cnp->cn_flags & ISLASTCN) &&
167 			    (cnp->cn_nameiop == DELETE ||
168 			    cnp->cn_nameiop == RENAME)) {
169 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
170 				    cnp->cn_thread);
171 				if (error != 0)
172 					goto out;
173 
174 				/* Allocate a new vnode on the matching entry. */
175 				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
176 						cnp->cn_lkflags, vpp);
177 				if (error != 0)
178 					goto out;
179 
180 				if ((dnode->tn_mode & S_ISTXT) &&
181 				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) &&
182 				  VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) {
183 					error = EPERM;
184 					vput(*vpp);
185 					*vpp = NULL;
186 					goto out;
187 				}
188 				cnp->cn_flags |= SAVENAME;
189 			} else {
190 				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
191 						cnp->cn_lkflags, vpp);
192 			}
193 		}
194 	}
195 
196 	/* Store the result of this lookup in the cache.  Avoid this if the
197 	 * request was for creation, as it does not improve timings on
198 	 * emprical tests. */
199 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
200 		cache_enter(dvp, *vpp, cnp);
201 
202 out:
203 	/* If there were no errors, *vpp cannot be null and it must be
204 	 * locked. */
205 	MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
206 
207 	return error;
208 }
209 
210 /* --------------------------------------------------------------------- */
211 
212 static int
213 tmpfs_create(struct vop_create_args *v)
214 {
215 	struct vnode *dvp = v->a_dvp;
216 	struct vnode **vpp = v->a_vpp;
217 	struct componentname *cnp = v->a_cnp;
218 	struct vattr *vap = v->a_vap;
219 
220 	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
221 
222 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
223 }
224 /* --------------------------------------------------------------------- */
225 
226 static int
227 tmpfs_mknod(struct vop_mknod_args *v)
228 {
229 	struct vnode *dvp = v->a_dvp;
230 	struct vnode **vpp = v->a_vpp;
231 	struct componentname *cnp = v->a_cnp;
232 	struct vattr *vap = v->a_vap;
233 
234 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
235 	    vap->va_type != VFIFO)
236 		return EINVAL;
237 
238 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
239 }
240 
241 /* --------------------------------------------------------------------- */
242 
243 static int
244 tmpfs_open(struct vop_open_args *v)
245 {
246 	struct vnode *vp = v->a_vp;
247 	int mode = v->a_mode;
248 
249 	int error;
250 	struct tmpfs_node *node;
251 
252 	MPASS(VOP_ISLOCKED(vp));
253 
254 	node = VP_TO_TMPFS_NODE(vp);
255 
256 	/* The file is still active but all its names have been removed
257 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
258 	 * it is about to die. */
259 	if (node->tn_links < 1)
260 		return (ENOENT);
261 
262 	/* If the file is marked append-only, deny write requests. */
263 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
264 		error = EPERM;
265 	else {
266 		error = 0;
267 		/* For regular files, the call below is nop. */
268 		vnode_create_vobject(vp, node->tn_size, v->a_td);
269 	}
270 
271 	MPASS(VOP_ISLOCKED(vp));
272 	return error;
273 }
274 
275 /* --------------------------------------------------------------------- */
276 
277 static int
278 tmpfs_close(struct vop_close_args *v)
279 {
280 	struct vnode *vp = v->a_vp;
281 
282 	/* Update node times. */
283 	tmpfs_update(vp);
284 
285 	return (0);
286 }
287 
288 /* --------------------------------------------------------------------- */
289 
290 int
291 tmpfs_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 
338 out:
339 	MPASS(VOP_ISLOCKED(vp));
340 
341 	return error;
342 }
343 
344 /* --------------------------------------------------------------------- */
345 
346 int
347 tmpfs_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... */
385 int
386 tmpfs_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 static int
442 tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
443     vm_offset_t offset, size_t tlen, struct uio *uio)
444 {
445 	vm_page_t	m;
446 	int		error, rv;
447 
448 	VM_OBJECT_WLOCK(tobj);
449 
450 	/*
451 	 * Parallel reads of the page content from disk are prevented
452 	 * by exclusive busy.
453 	 *
454 	 * Although the tmpfs vnode lock is held here, it is
455 	 * nonetheless safe to sleep waiting for a free page.  The
456 	 * pageout daemon does not need to acquire the tmpfs vnode
457 	 * lock to page out tobj's pages because tobj is a OBJT_SWAP
458 	 * type object.
459 	 */
460 	m = vm_page_grab(tobj, idx, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
461 	if (m->valid != VM_PAGE_BITS_ALL) {
462 		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
463 			rv = vm_pager_get_pages(tobj, &m, 1, 0);
464 			m = vm_page_lookup(tobj, idx);
465 			if (m == NULL) {
466 				printf(
467 		    "tmpfs: vm_obj %p idx %jd null lookup rv %d\n",
468 				    tobj, idx, rv);
469 				VM_OBJECT_WUNLOCK(tobj);
470 				return (EIO);
471 			}
472 			if (rv != VM_PAGER_OK) {
473 				printf(
474 		    "tmpfs: vm_obj %p idx %jd valid %x pager error %d\n",
475 				    tobj, idx, m->valid, rv);
476 				vm_page_lock(m);
477 				vm_page_free(m);
478 				vm_page_unlock(m);
479 				VM_OBJECT_WUNLOCK(tobj);
480 				return (EIO);
481 			}
482 		} else
483 			vm_page_zero_invalid(m, TRUE);
484 	}
485 	vm_page_xunbusy(m);
486 	vm_page_lock(m);
487 	vm_page_hold(m);
488 	vm_page_unlock(m);
489 	VM_OBJECT_WUNLOCK(tobj);
490 	error = uiomove_fromphys(&m, offset, tlen, uio);
491 	vm_page_lock(m);
492 	vm_page_unhold(m);
493 	if (m->queue == PQ_NONE) {
494 		vm_page_deactivate(m);
495 	} else {
496 		/* Requeue to maintain LRU ordering. */
497 		vm_page_requeue(m);
498 	}
499 	vm_page_unlock(m);
500 
501 	return (error);
502 }
503 
504 static int
505 tmpfs_read(struct vop_read_args *v)
506 {
507 	struct vnode *vp = v->a_vp;
508 	struct uio *uio = v->a_uio;
509 	struct tmpfs_node *node;
510 	vm_object_t uobj;
511 	size_t len;
512 	int resid;
513 	int error = 0;
514 	vm_pindex_t	idx;
515 	vm_offset_t	offset;
516 	off_t		addr;
517 	size_t		tlen;
518 
519 	node = VP_TO_TMPFS_NODE(vp);
520 
521 	if (vp->v_type != VREG) {
522 		error = EISDIR;
523 		goto out;
524 	}
525 
526 	if (uio->uio_offset < 0) {
527 		error = EINVAL;
528 		goto out;
529 	}
530 
531 	node->tn_status |= TMPFS_NODE_ACCESSED;
532 
533 	uobj = node->tn_reg.tn_aobj;
534 	while ((resid = uio->uio_resid) > 0) {
535 		error = 0;
536 		if (node->tn_size <= uio->uio_offset)
537 			break;
538 		len = MIN(node->tn_size - uio->uio_offset, resid);
539 		if (len == 0)
540 			break;
541 		addr = uio->uio_offset;
542 		idx = OFF_TO_IDX(addr);
543 		offset = addr & PAGE_MASK;
544 		tlen = MIN(PAGE_SIZE - offset, len);
545 		error = tmpfs_nocacheread(uobj, idx, offset, tlen, uio);
546 		if ((error != 0) || (resid == uio->uio_resid))
547 			break;
548 	}
549 
550 out:
551 
552 	return error;
553 }
554 
555 /* --------------------------------------------------------------------- */
556 
557 static int
558 tmpfs_mappedwrite(vm_object_t tobj, size_t len, struct uio *uio)
559 {
560 	vm_pindex_t	idx;
561 	vm_page_t	tpg;
562 	vm_offset_t	offset;
563 	off_t		addr;
564 	size_t		tlen;
565 	int		error, rv;
566 
567 	error = 0;
568 
569 	addr = uio->uio_offset;
570 	idx = OFF_TO_IDX(addr);
571 	offset = addr & PAGE_MASK;
572 	tlen = MIN(PAGE_SIZE - offset, len);
573 
574 	VM_OBJECT_WLOCK(tobj);
575 	tpg = vm_page_grab(tobj, idx, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
576 	if (tpg->valid != VM_PAGE_BITS_ALL) {
577 		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
578 			rv = vm_pager_get_pages(tobj, &tpg, 1, 0);
579 			tpg = vm_page_lookup(tobj, idx);
580 			if (tpg == NULL) {
581 				printf(
582 		    "tmpfs: vm_obj %p idx %jd null lookup rv %d\n",
583 				    tobj, idx, rv);
584 				VM_OBJECT_WUNLOCK(tobj);
585 				return (EIO);
586 			}
587 			if (rv != VM_PAGER_OK) {
588 				printf(
589 		    "tmpfs: vm_obj %p idx %jd valid %x pager error %d\n",
590 				    tobj, idx, tpg->valid, rv);
591 				vm_page_lock(tpg);
592 				vm_page_free(tpg);
593 				vm_page_unlock(tpg);
594 				VM_OBJECT_WUNLOCK(tobj);
595 				return (EIO);
596 			}
597 		} else
598 			vm_page_zero_invalid(tpg, TRUE);
599 	}
600 	vm_page_xunbusy(tpg);
601 	vm_page_lock(tpg);
602 	vm_page_hold(tpg);
603 	vm_page_unlock(tpg);
604 	VM_OBJECT_WUNLOCK(tobj);
605 	error = uiomove_fromphys(&tpg, offset, tlen, uio);
606 	VM_OBJECT_WLOCK(tobj);
607 	if (error == 0)
608 		vm_page_dirty(tpg);
609 	vm_page_lock(tpg);
610 	vm_page_unhold(tpg);
611 	if (tpg->queue == PQ_NONE) {
612 		vm_page_deactivate(tpg);
613 	} else {
614 		/* Requeue to maintain LRU ordering. */
615 		vm_page_requeue(tpg);
616 	}
617 	vm_page_unlock(tpg);
618 	VM_OBJECT_WUNLOCK(tobj);
619 
620 	return	(error);
621 }
622 
623 static int
624 tmpfs_write(struct vop_write_args *v)
625 {
626 	struct vnode *vp = v->a_vp;
627 	struct uio *uio = v->a_uio;
628 	int ioflag = v->a_ioflag;
629 
630 	boolean_t extended;
631 	int error = 0;
632 	off_t oldsize;
633 	struct tmpfs_node *node;
634 	vm_object_t uobj;
635 	size_t len;
636 	int resid;
637 
638 	node = VP_TO_TMPFS_NODE(vp);
639 	oldsize = node->tn_size;
640 
641 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
642 		error = EINVAL;
643 		goto out;
644 	}
645 
646 	if (uio->uio_resid == 0) {
647 		error = 0;
648 		goto out;
649 	}
650 
651 	if (ioflag & IO_APPEND)
652 		uio->uio_offset = node->tn_size;
653 
654 	if (uio->uio_offset + uio->uio_resid >
655 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
656 		return (EFBIG);
657 
658 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
659 		return (EFBIG);
660 
661 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
662 	if (extended) {
663 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid,
664 		    FALSE);
665 		if (error != 0)
666 			goto out;
667 	}
668 
669 	uobj = node->tn_reg.tn_aobj;
670 	while ((resid = uio->uio_resid) > 0) {
671 		if (node->tn_size <= uio->uio_offset)
672 			break;
673 		len = MIN(node->tn_size - uio->uio_offset, resid);
674 		if (len == 0)
675 			break;
676 		error = tmpfs_mappedwrite(uobj, len, uio);
677 		if ((error != 0) || (resid == uio->uio_resid))
678 			break;
679 	}
680 
681 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
682 	    (extended ? TMPFS_NODE_CHANGED : 0);
683 
684 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
685 		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
686 			node->tn_mode &= ~(S_ISUID | S_ISGID);
687 	}
688 
689 	if (error != 0)
690 		(void)tmpfs_reg_resize(vp, oldsize, TRUE);
691 
692 out:
693 	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
694 	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
695 
696 	return error;
697 }
698 
699 /* --------------------------------------------------------------------- */
700 
701 static int
702 tmpfs_fsync(struct vop_fsync_args *v)
703 {
704 	struct vnode *vp = v->a_vp;
705 
706 	MPASS(VOP_ISLOCKED(vp));
707 
708 	tmpfs_update(vp);
709 
710 	return 0;
711 }
712 
713 /* --------------------------------------------------------------------- */
714 
715 static int
716 tmpfs_remove(struct vop_remove_args *v)
717 {
718 	struct vnode *dvp = v->a_dvp;
719 	struct vnode *vp = v->a_vp;
720 
721 	int error;
722 	struct tmpfs_dirent *de;
723 	struct tmpfs_mount *tmp;
724 	struct tmpfs_node *dnode;
725 	struct tmpfs_node *node;
726 
727 	MPASS(VOP_ISLOCKED(dvp));
728 	MPASS(VOP_ISLOCKED(vp));
729 
730 	if (vp->v_type == VDIR) {
731 		error = EISDIR;
732 		goto out;
733 	}
734 
735 	dnode = VP_TO_TMPFS_DIR(dvp);
736 	node = VP_TO_TMPFS_NODE(vp);
737 	tmp = VFS_TO_TMPFS(vp->v_mount);
738 	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
739 	MPASS(de != NULL);
740 
741 	/* Files marked as immutable or append-only cannot be deleted. */
742 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
743 	    (dnode->tn_flags & APPEND)) {
744 		error = EPERM;
745 		goto out;
746 	}
747 
748 	/* Remove the entry from the directory; as it is a file, we do not
749 	 * have to change the number of hard links of the directory. */
750 	tmpfs_dir_detach(dvp, de);
751 	if (v->a_cnp->cn_flags & DOWHITEOUT)
752 		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
753 
754 	/* Free the directory entry we just deleted.  Note that the node
755 	 * referred by it will not be removed until the vnode is really
756 	 * reclaimed. */
757 	tmpfs_free_dirent(tmp, de);
758 
759 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
760 	error = 0;
761 
762 out:
763 
764 	return error;
765 }
766 
767 /* --------------------------------------------------------------------- */
768 
769 static int
770 tmpfs_link(struct vop_link_args *v)
771 {
772 	struct vnode *dvp = v->a_tdvp;
773 	struct vnode *vp = v->a_vp;
774 	struct componentname *cnp = v->a_cnp;
775 
776 	int error;
777 	struct tmpfs_dirent *de;
778 	struct tmpfs_node *node;
779 
780 	MPASS(VOP_ISLOCKED(dvp));
781 	MPASS(cnp->cn_flags & HASBUF);
782 	MPASS(dvp != vp); /* XXX When can this be false? */
783 
784 	node = VP_TO_TMPFS_NODE(vp);
785 
786 	/* XXX: Why aren't the following two tests done by the caller? */
787 
788 	/* Hard links of directories are forbidden. */
789 	if (vp->v_type == VDIR) {
790 		error = EPERM;
791 		goto out;
792 	}
793 
794 	/* Cannot create cross-device links. */
795 	if (dvp->v_mount != vp->v_mount) {
796 		error = EXDEV;
797 		goto out;
798 	}
799 
800 	/* Ensure that we do not overflow the maximum number of links imposed
801 	 * by the system. */
802 	MPASS(node->tn_links <= LINK_MAX);
803 	if (node->tn_links == LINK_MAX) {
804 		error = EMLINK;
805 		goto out;
806 	}
807 
808 	/* We cannot create links of files marked immutable or append-only. */
809 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
810 		error = EPERM;
811 		goto out;
812 	}
813 
814 	/* Allocate a new directory entry to represent the node. */
815 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
816 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
817 	if (error != 0)
818 		goto out;
819 
820 	/* Insert the new directory entry into the appropriate directory. */
821 	if (cnp->cn_flags & ISWHITEOUT)
822 		tmpfs_dir_whiteout_remove(dvp, cnp);
823 	tmpfs_dir_attach(dvp, de);
824 
825 	/* vp link count has changed, so update node times. */
826 	node->tn_status |= TMPFS_NODE_CHANGED;
827 	tmpfs_update(vp);
828 
829 	error = 0;
830 
831 out:
832 	return error;
833 }
834 
835 /* --------------------------------------------------------------------- */
836 
837 /*
838  * We acquire all but fdvp locks using non-blocking acquisitions.  If we
839  * fail to acquire any lock in the path we will drop all held locks,
840  * acquire the new lock in a blocking fashion, and then release it and
841  * restart the rename.  This acquire/release step ensures that we do not
842  * spin on a lock waiting for release.  On error release all vnode locks
843  * and decrement references the way tmpfs_rename() would do.
844  */
845 static int
846 tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp,
847     struct vnode *tdvp, struct vnode **tvpp,
848     struct componentname *fcnp, struct componentname *tcnp)
849 {
850 	struct vnode *nvp;
851 	struct mount *mp;
852 	struct tmpfs_dirent *de;
853 	int error, restarts = 0;
854 
855 	VOP_UNLOCK(tdvp, 0);
856 	if (*tvpp != NULL && *tvpp != tdvp)
857 		VOP_UNLOCK(*tvpp, 0);
858 	mp = fdvp->v_mount;
859 
860 relock:
861 	restarts += 1;
862 	error = vn_lock(fdvp, LK_EXCLUSIVE);
863 	if (error)
864 		goto releout;
865 	if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
866 		VOP_UNLOCK(fdvp, 0);
867 		error = vn_lock(tdvp, LK_EXCLUSIVE);
868 		if (error)
869 			goto releout;
870 		VOP_UNLOCK(tdvp, 0);
871 		goto relock;
872 	}
873 	/*
874 	 * Re-resolve fvp to be certain it still exists and fetch the
875 	 * correct vnode.
876 	 */
877 	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp);
878 	if (de == NULL) {
879 		VOP_UNLOCK(fdvp, 0);
880 		VOP_UNLOCK(tdvp, 0);
881 		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
882 		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
883 			error = EINVAL;
884 		else
885 			error = ENOENT;
886 		goto releout;
887 	}
888 	error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
889 	if (error != 0) {
890 		VOP_UNLOCK(fdvp, 0);
891 		VOP_UNLOCK(tdvp, 0);
892 		if (error != EBUSY)
893 			goto releout;
894 		error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp);
895 		if (error != 0)
896 			goto releout;
897 		VOP_UNLOCK(nvp, 0);
898 		/*
899 		 * Concurrent rename race.
900 		 */
901 		if (nvp == tdvp) {
902 			vrele(nvp);
903 			error = EINVAL;
904 			goto releout;
905 		}
906 		vrele(*fvpp);
907 		*fvpp = nvp;
908 		goto relock;
909 	}
910 	vrele(*fvpp);
911 	*fvpp = nvp;
912 	VOP_UNLOCK(*fvpp, 0);
913 	/*
914 	 * Re-resolve tvp and acquire the vnode lock if present.
915 	 */
916 	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp);
917 	/*
918 	 * If tvp disappeared we just carry on.
919 	 */
920 	if (de == NULL && *tvpp != NULL) {
921 		vrele(*tvpp);
922 		*tvpp = NULL;
923 	}
924 	/*
925 	 * Get the tvp ino if the lookup succeeded.  We may have to restart
926 	 * if the non-blocking acquire fails.
927 	 */
928 	if (de != NULL) {
929 		nvp = NULL;
930 		error = tmpfs_alloc_vp(mp, de->td_node,
931 		    LK_EXCLUSIVE | LK_NOWAIT, &nvp);
932 		if (*tvpp != NULL)
933 			vrele(*tvpp);
934 		*tvpp = nvp;
935 		if (error != 0) {
936 			VOP_UNLOCK(fdvp, 0);
937 			VOP_UNLOCK(tdvp, 0);
938 			if (error != EBUSY)
939 				goto releout;
940 			error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE,
941 			    &nvp);
942 			if (error != 0)
943 				goto releout;
944 			VOP_UNLOCK(nvp, 0);
945 			/*
946 			 * fdvp contains fvp, thus tvp (=fdvp) is not empty.
947 			 */
948 			if (nvp == fdvp) {
949 				error = ENOTEMPTY;
950 				goto releout;
951 			}
952 			goto relock;
953 		}
954 	}
955 	tmpfs_rename_restarts += restarts;
956 
957 	return (0);
958 
959 releout:
960 	vrele(fdvp);
961 	vrele(*fvpp);
962 	vrele(tdvp);
963 	if (*tvpp != NULL)
964 		vrele(*tvpp);
965 	tmpfs_rename_restarts += restarts;
966 
967 	return (error);
968 }
969 
970 static int
971 tmpfs_rename(struct vop_rename_args *v)
972 {
973 	struct vnode *fdvp = v->a_fdvp;
974 	struct vnode *fvp = v->a_fvp;
975 	struct componentname *fcnp = v->a_fcnp;
976 	struct vnode *tdvp = v->a_tdvp;
977 	struct vnode *tvp = v->a_tvp;
978 	struct componentname *tcnp = v->a_tcnp;
979 	struct mount *mp = NULL;
980 
981 	char *newname;
982 	int error;
983 	struct tmpfs_dirent *de;
984 	struct tmpfs_mount *tmp;
985 	struct tmpfs_node *fdnode;
986 	struct tmpfs_node *fnode;
987 	struct tmpfs_node *tnode;
988 	struct tmpfs_node *tdnode;
989 
990 	MPASS(VOP_ISLOCKED(tdvp));
991 	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
992 	MPASS(fcnp->cn_flags & HASBUF);
993 	MPASS(tcnp->cn_flags & HASBUF);
994 
995 	/* Disallow cross-device renames.
996 	 * XXX Why isn't this done by the caller? */
997 	if (fvp->v_mount != tdvp->v_mount ||
998 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
999 		error = EXDEV;
1000 		goto out;
1001 	}
1002 
1003 	/* If source and target are the same file, there is nothing to do. */
1004 	if (fvp == tvp) {
1005 		error = 0;
1006 		goto out;
1007 	}
1008 
1009 	/* If we need to move the directory between entries, lock the
1010 	 * source so that we can safely operate on it. */
1011 	if (fdvp != tdvp && fdvp != tvp) {
1012 		if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1013 			mp = tdvp->v_mount;
1014 			error = vfs_busy(mp, 0);
1015 			if (error != 0) {
1016 				mp = NULL;
1017 				goto out;
1018 			}
1019 			error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp,
1020 			    fcnp, tcnp);
1021 			if (error != 0) {
1022 				vfs_unbusy(mp);
1023 				return (error);
1024 			}
1025 			ASSERT_VOP_ELOCKED(fdvp,
1026 			    "tmpfs_rename: fdvp not locked");
1027 			ASSERT_VOP_ELOCKED(tdvp,
1028 			    "tmpfs_rename: tdvp not locked");
1029 			if (tvp != NULL)
1030 				ASSERT_VOP_ELOCKED(tvp,
1031 				    "tmpfs_rename: tvp not locked");
1032 			if (fvp == tvp) {
1033 				error = 0;
1034 				goto out_locked;
1035 			}
1036 		}
1037 	}
1038 
1039 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
1040 	tdnode = VP_TO_TMPFS_DIR(tdvp);
1041 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
1042 	fdnode = VP_TO_TMPFS_DIR(fdvp);
1043 	fnode = VP_TO_TMPFS_NODE(fvp);
1044 	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
1045 
1046 	/* Entry can disappear before we lock fdvp,
1047 	 * also avoid manipulating '.' and '..' entries. */
1048 	if (de == NULL) {
1049 		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
1050 		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
1051 			error = EINVAL;
1052 		else
1053 			error = ENOENT;
1054 		goto out_locked;
1055 	}
1056 	MPASS(de->td_node == fnode);
1057 
1058 	/* If re-naming a directory to another preexisting directory
1059 	 * ensure that the target directory is empty so that its
1060 	 * removal causes no side effects.
1061 	 * Kern_rename gurantees the destination to be a directory
1062 	 * if the source is one. */
1063 	if (tvp != NULL) {
1064 		MPASS(tnode != NULL);
1065 
1066 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1067 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1068 			error = EPERM;
1069 			goto out_locked;
1070 		}
1071 
1072 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1073 			if (tnode->tn_size > 0) {
1074 				error = ENOTEMPTY;
1075 				goto out_locked;
1076 			}
1077 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1078 			error = ENOTDIR;
1079 			goto out_locked;
1080 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1081 			error = EISDIR;
1082 			goto out_locked;
1083 		} else {
1084 			MPASS(fnode->tn_type != VDIR &&
1085 				tnode->tn_type != VDIR);
1086 		}
1087 	}
1088 
1089 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1090 	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1091 		error = EPERM;
1092 		goto out_locked;
1093 	}
1094 
1095 	/* Ensure that we have enough memory to hold the new name, if it
1096 	 * has to be changed. */
1097 	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1098 	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1099 		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1100 	} else
1101 		newname = NULL;
1102 
1103 	/* If the node is being moved to another directory, we have to do
1104 	 * the move. */
1105 	if (fdnode != tdnode) {
1106 		/* In case we are moving a directory, we have to adjust its
1107 		 * parent to point to the new parent. */
1108 		if (de->td_node->tn_type == VDIR) {
1109 			struct tmpfs_node *n;
1110 
1111 			/* Ensure the target directory is not a child of the
1112 			 * directory being moved.  Otherwise, we'd end up
1113 			 * with stale nodes. */
1114 			n = tdnode;
1115 			/* TMPFS_LOCK garanties that no nodes are freed while
1116 			 * traversing the list. Nodes can only be marked as
1117 			 * removed: tn_parent == NULL. */
1118 			TMPFS_LOCK(tmp);
1119 			TMPFS_NODE_LOCK(n);
1120 			while (n != n->tn_dir.tn_parent) {
1121 				struct tmpfs_node *parent;
1122 
1123 				if (n == fnode) {
1124 					TMPFS_NODE_UNLOCK(n);
1125 					TMPFS_UNLOCK(tmp);
1126 					error = EINVAL;
1127 					if (newname != NULL)
1128 						    free(newname, M_TMPFSNAME);
1129 					goto out_locked;
1130 				}
1131 				parent = n->tn_dir.tn_parent;
1132 				TMPFS_NODE_UNLOCK(n);
1133 				if (parent == NULL) {
1134 					n = NULL;
1135 					break;
1136 				}
1137 				TMPFS_NODE_LOCK(parent);
1138 				if (parent->tn_dir.tn_parent == NULL) {
1139 					TMPFS_NODE_UNLOCK(parent);
1140 					n = NULL;
1141 					break;
1142 				}
1143 				n = parent;
1144 			}
1145 			TMPFS_UNLOCK(tmp);
1146 			if (n == NULL) {
1147 				error = EINVAL;
1148 				if (newname != NULL)
1149 					    free(newname, M_TMPFSNAME);
1150 				goto out_locked;
1151 			}
1152 			TMPFS_NODE_UNLOCK(n);
1153 
1154 			/* Adjust the parent pointer. */
1155 			TMPFS_VALIDATE_DIR(fnode);
1156 			TMPFS_NODE_LOCK(de->td_node);
1157 			de->td_node->tn_dir.tn_parent = tdnode;
1158 			TMPFS_NODE_UNLOCK(de->td_node);
1159 
1160 			/* As a result of changing the target of the '..'
1161 			 * entry, the link count of the source and target
1162 			 * directories has to be adjusted. */
1163 			TMPFS_NODE_LOCK(tdnode);
1164 			TMPFS_ASSERT_LOCKED(tdnode);
1165 			tdnode->tn_links++;
1166 			TMPFS_NODE_UNLOCK(tdnode);
1167 
1168 			TMPFS_NODE_LOCK(fdnode);
1169 			TMPFS_ASSERT_LOCKED(fdnode);
1170 			fdnode->tn_links--;
1171 			TMPFS_NODE_UNLOCK(fdnode);
1172 		}
1173 	}
1174 
1175 	/* Do the move: just remove the entry from the source directory
1176 	 * and insert it into the target one. */
1177 	tmpfs_dir_detach(fdvp, de);
1178 
1179 	if (fcnp->cn_flags & DOWHITEOUT)
1180 		tmpfs_dir_whiteout_add(fdvp, fcnp);
1181 	if (tcnp->cn_flags & ISWHITEOUT)
1182 		tmpfs_dir_whiteout_remove(tdvp, tcnp);
1183 
1184 	/* If the name has changed, we need to make it effective by changing
1185 	 * it in the directory entry. */
1186 	if (newname != NULL) {
1187 		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1188 
1189 		free(de->ud.td_name, M_TMPFSNAME);
1190 		de->ud.td_name = newname;
1191 		tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen);
1192 
1193 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1194 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1195 	}
1196 
1197 	/* If we are overwriting an entry, we have to remove the old one
1198 	 * from the target directory. */
1199 	if (tvp != NULL) {
1200 		struct tmpfs_dirent *tde;
1201 
1202 		/* Remove the old entry from the target directory. */
1203 		tde = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1204 		tmpfs_dir_detach(tdvp, tde);
1205 
1206 		/* Free the directory entry we just deleted.  Note that the
1207 		 * node referred by it will not be removed until the vnode is
1208 		 * really reclaimed. */
1209 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1210 	}
1211 
1212 	tmpfs_dir_attach(tdvp, de);
1213 
1214 	cache_purge(fvp);
1215 	if (tvp != NULL)
1216 		cache_purge(tvp);
1217 	cache_purge_negative(tdvp);
1218 
1219 	error = 0;
1220 
1221 out_locked:
1222 	if (fdvp != tdvp && fdvp != tvp)
1223 		VOP_UNLOCK(fdvp, 0);
1224 
1225 out:
1226 	/* Release target nodes. */
1227 	/* XXX: I don't understand when tdvp can be the same as tvp, but
1228 	 * other code takes care of this... */
1229 	if (tdvp == tvp)
1230 		vrele(tdvp);
1231 	else
1232 		vput(tdvp);
1233 	if (tvp != NULL)
1234 		vput(tvp);
1235 
1236 	/* Release source nodes. */
1237 	vrele(fdvp);
1238 	vrele(fvp);
1239 
1240 	if (mp != NULL)
1241 		vfs_unbusy(mp);
1242 
1243 	return error;
1244 }
1245 
1246 /* --------------------------------------------------------------------- */
1247 
1248 static int
1249 tmpfs_mkdir(struct vop_mkdir_args *v)
1250 {
1251 	struct vnode *dvp = v->a_dvp;
1252 	struct vnode **vpp = v->a_vpp;
1253 	struct componentname *cnp = v->a_cnp;
1254 	struct vattr *vap = v->a_vap;
1255 
1256 	MPASS(vap->va_type == VDIR);
1257 
1258 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1259 }
1260 
1261 /* --------------------------------------------------------------------- */
1262 
1263 static int
1264 tmpfs_rmdir(struct vop_rmdir_args *v)
1265 {
1266 	struct vnode *dvp = v->a_dvp;
1267 	struct vnode *vp = v->a_vp;
1268 
1269 	int error;
1270 	struct tmpfs_dirent *de;
1271 	struct tmpfs_mount *tmp;
1272 	struct tmpfs_node *dnode;
1273 	struct tmpfs_node *node;
1274 
1275 	MPASS(VOP_ISLOCKED(dvp));
1276 	MPASS(VOP_ISLOCKED(vp));
1277 
1278 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1279 	dnode = VP_TO_TMPFS_DIR(dvp);
1280 	node = VP_TO_TMPFS_DIR(vp);
1281 
1282 	/* Directories with more than two entries ('.' and '..') cannot be
1283 	 * removed. */
1284 	 if (node->tn_size > 0) {
1285 		 error = ENOTEMPTY;
1286 		 goto out;
1287 	 }
1288 
1289 	if ((dnode->tn_flags & APPEND)
1290 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1291 		error = EPERM;
1292 		goto out;
1293 	}
1294 
1295 	/* This invariant holds only if we are not trying to remove "..".
1296 	  * We checked for that above so this is safe now. */
1297 	MPASS(node->tn_dir.tn_parent == dnode);
1298 
1299 	/* Get the directory entry associated with node (vp).  This was
1300 	 * filled by tmpfs_lookup while looking up the entry. */
1301 	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1302 	MPASS(TMPFS_DIRENT_MATCHES(de,
1303 	    v->a_cnp->cn_nameptr,
1304 	    v->a_cnp->cn_namelen));
1305 
1306 	/* Check flags to see if we are allowed to remove the directory. */
1307 	if (dnode->tn_flags & APPEND
1308 		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1309 		error = EPERM;
1310 		goto out;
1311 	}
1312 
1313 
1314 	/* Detach the directory entry from the directory (dnode). */
1315 	tmpfs_dir_detach(dvp, de);
1316 	if (v->a_cnp->cn_flags & DOWHITEOUT)
1317 		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1318 
1319 	/* No vnode should be allocated for this entry from this point */
1320 	TMPFS_NODE_LOCK(node);
1321 	TMPFS_ASSERT_ELOCKED(node);
1322 	node->tn_links--;
1323 	node->tn_dir.tn_parent = NULL;
1324 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1325 	    TMPFS_NODE_MODIFIED;
1326 
1327 	TMPFS_NODE_UNLOCK(node);
1328 
1329 	TMPFS_NODE_LOCK(dnode);
1330 	TMPFS_ASSERT_ELOCKED(dnode);
1331 	dnode->tn_links--;
1332 	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1333 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1334 	TMPFS_NODE_UNLOCK(dnode);
1335 
1336 	cache_purge(dvp);
1337 	cache_purge(vp);
1338 
1339 	/* Free the directory entry we just deleted.  Note that the node
1340 	 * referred by it will not be removed until the vnode is really
1341 	 * reclaimed. */
1342 	tmpfs_free_dirent(tmp, de);
1343 
1344 	/* Release the deleted vnode (will destroy the node, notify
1345 	 * interested parties and clean it from the cache). */
1346 
1347 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1348 	tmpfs_update(dvp);
1349 
1350 	error = 0;
1351 
1352 out:
1353 	return error;
1354 }
1355 
1356 /* --------------------------------------------------------------------- */
1357 
1358 static int
1359 tmpfs_symlink(struct vop_symlink_args *v)
1360 {
1361 	struct vnode *dvp = v->a_dvp;
1362 	struct vnode **vpp = v->a_vpp;
1363 	struct componentname *cnp = v->a_cnp;
1364 	struct vattr *vap = v->a_vap;
1365 	char *target = v->a_target;
1366 
1367 #ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1368 	MPASS(vap->va_type == VLNK);
1369 #else
1370 	vap->va_type = VLNK;
1371 #endif
1372 
1373 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1374 }
1375 
1376 /* --------------------------------------------------------------------- */
1377 
1378 static int
1379 tmpfs_readdir(struct vop_readdir_args *v)
1380 {
1381 	struct vnode *vp = v->a_vp;
1382 	struct uio *uio = v->a_uio;
1383 	int *eofflag = v->a_eofflag;
1384 	u_long **cookies = v->a_cookies;
1385 	int *ncookies = v->a_ncookies;
1386 
1387 	int error;
1388 	ssize_t startresid;
1389 	int cnt = 0;
1390 	struct tmpfs_node *node;
1391 
1392 	/* This operation only makes sense on directory nodes. */
1393 	if (vp->v_type != VDIR)
1394 		return ENOTDIR;
1395 
1396 	node = VP_TO_TMPFS_DIR(vp);
1397 
1398 	startresid = uio->uio_resid;
1399 
1400 	if (cookies != NULL && ncookies != NULL) {
1401 		cnt = howmany(node->tn_size, sizeof(struct tmpfs_dirent)) + 2;
1402 		*cookies = malloc(cnt * sizeof(**cookies), M_TEMP, M_WAITOK);
1403 		*ncookies = 0;
1404 	}
1405 
1406 	if (cnt == 0)
1407 		error = tmpfs_dir_getdents(node, uio, 0, NULL, NULL);
1408 	else
1409 		error = tmpfs_dir_getdents(node, uio, cnt, *cookies, ncookies);
1410 
1411 	if (error == EJUSTRETURN)
1412 		error = (uio->uio_resid != startresid) ? 0 : EINVAL;
1413 
1414 	if (error != 0 && cnt != 0)
1415 		free(*cookies, M_TEMP);
1416 
1417 	if (eofflag != NULL)
1418 		*eofflag =
1419 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1420 
1421 	return error;
1422 }
1423 
1424 /* --------------------------------------------------------------------- */
1425 
1426 static int
1427 tmpfs_readlink(struct vop_readlink_args *v)
1428 {
1429 	struct vnode *vp = v->a_vp;
1430 	struct uio *uio = v->a_uio;
1431 
1432 	int error;
1433 	struct tmpfs_node *node;
1434 
1435 	MPASS(uio->uio_offset == 0);
1436 	MPASS(vp->v_type == VLNK);
1437 
1438 	node = VP_TO_TMPFS_NODE(vp);
1439 
1440 	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1441 	    uio);
1442 	node->tn_status |= TMPFS_NODE_ACCESSED;
1443 
1444 	return error;
1445 }
1446 
1447 /* --------------------------------------------------------------------- */
1448 
1449 static int
1450 tmpfs_inactive(struct vop_inactive_args *v)
1451 {
1452 	struct vnode *vp = v->a_vp;
1453 
1454 	struct tmpfs_node *node;
1455 
1456 	node = VP_TO_TMPFS_NODE(vp);
1457 
1458 	if (node->tn_links == 0)
1459 		vrecycle(vp);
1460 
1461 	return 0;
1462 }
1463 
1464 /* --------------------------------------------------------------------- */
1465 
1466 int
1467 tmpfs_reclaim(struct vop_reclaim_args *v)
1468 {
1469 	struct vnode *vp = v->a_vp;
1470 
1471 	struct tmpfs_mount *tmp;
1472 	struct tmpfs_node *node;
1473 
1474 	node = VP_TO_TMPFS_NODE(vp);
1475 	tmp = VFS_TO_TMPFS(vp->v_mount);
1476 
1477 	if (vp->v_type == VREG)
1478 		tmpfs_destroy_vobject(vp, node->tn_reg.tn_aobj);
1479 	else
1480 		vnode_destroy_vobject(vp);
1481 	vp->v_object = NULL;
1482 	cache_purge(vp);
1483 
1484 	TMPFS_NODE_LOCK(node);
1485 	TMPFS_ASSERT_ELOCKED(node);
1486 	tmpfs_free_vp(vp);
1487 
1488 	/* If the node referenced by this vnode was deleted by the user,
1489 	 * we must free its associated data structures (now that the vnode
1490 	 * is being reclaimed). */
1491 	if (node->tn_links == 0 &&
1492 	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1493 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1494 		TMPFS_NODE_UNLOCK(node);
1495 		tmpfs_free_node(tmp, node);
1496 	} else
1497 		TMPFS_NODE_UNLOCK(node);
1498 
1499 	MPASS(vp->v_data == NULL);
1500 	return 0;
1501 }
1502 
1503 /* --------------------------------------------------------------------- */
1504 
1505 static int
1506 tmpfs_print(struct vop_print_args *v)
1507 {
1508 	struct vnode *vp = v->a_vp;
1509 
1510 	struct tmpfs_node *node;
1511 
1512 	node = VP_TO_TMPFS_NODE(vp);
1513 
1514 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%lx, links %d\n",
1515 	    node, node->tn_flags, node->tn_links);
1516 	printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n",
1517 	    node->tn_mode, node->tn_uid, node->tn_gid,
1518 	    (intmax_t)node->tn_size, node->tn_status);
1519 
1520 	if (vp->v_type == VFIFO)
1521 		fifo_printinfo(vp);
1522 
1523 	printf("\n");
1524 
1525 	return 0;
1526 }
1527 
1528 /* --------------------------------------------------------------------- */
1529 
1530 static int
1531 tmpfs_pathconf(struct vop_pathconf_args *v)
1532 {
1533 	int name = v->a_name;
1534 	register_t *retval = v->a_retval;
1535 
1536 	int error;
1537 
1538 	error = 0;
1539 
1540 	switch (name) {
1541 	case _PC_LINK_MAX:
1542 		*retval = LINK_MAX;
1543 		break;
1544 
1545 	case _PC_NAME_MAX:
1546 		*retval = NAME_MAX;
1547 		break;
1548 
1549 	case _PC_PATH_MAX:
1550 		*retval = PATH_MAX;
1551 		break;
1552 
1553 	case _PC_PIPE_BUF:
1554 		*retval = PIPE_BUF;
1555 		break;
1556 
1557 	case _PC_CHOWN_RESTRICTED:
1558 		*retval = 1;
1559 		break;
1560 
1561 	case _PC_NO_TRUNC:
1562 		*retval = 1;
1563 		break;
1564 
1565 	case _PC_SYNC_IO:
1566 		*retval = 1;
1567 		break;
1568 
1569 	case _PC_FILESIZEBITS:
1570 		*retval = 0; /* XXX Don't know which value should I return. */
1571 		break;
1572 
1573 	default:
1574 		error = EINVAL;
1575 	}
1576 
1577 	return error;
1578 }
1579 
1580 static int
1581 tmpfs_vptofh(struct vop_vptofh_args *ap)
1582 {
1583 	struct tmpfs_fid *tfhp;
1584 	struct tmpfs_node *node;
1585 
1586 	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1587 	node = VP_TO_TMPFS_NODE(ap->a_vp);
1588 
1589 	tfhp->tf_len = sizeof(struct tmpfs_fid);
1590 	tfhp->tf_id = node->tn_id;
1591 	tfhp->tf_gen = node->tn_gen;
1592 
1593 	return (0);
1594 }
1595 
1596 static int
1597 tmpfs_whiteout(struct vop_whiteout_args *ap)
1598 {
1599 	struct vnode *dvp = ap->a_dvp;
1600 	struct componentname *cnp = ap->a_cnp;
1601 	struct tmpfs_dirent *de;
1602 
1603 	switch (ap->a_flags) {
1604 	case LOOKUP:
1605 		return (0);
1606 	case CREATE:
1607 		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1608 		if (de != NULL)
1609 			return (de->td_node == NULL ? 0 : EEXIST);
1610 		return (tmpfs_dir_whiteout_add(dvp, cnp));
1611 	case DELETE:
1612 		tmpfs_dir_whiteout_remove(dvp, cnp);
1613 		return (0);
1614 	default:
1615 		panic("tmpfs_whiteout: unknown op");
1616 	}
1617 }
1618 
1619 /* --------------------------------------------------------------------- */
1620 
1621 /*
1622  * vnode operations vector used for files stored in a tmpfs file system.
1623  */
1624 struct vop_vector tmpfs_vnodeop_entries = {
1625 	.vop_default =			&default_vnodeops,
1626 	.vop_lookup =			vfs_cache_lookup,
1627 	.vop_cachedlookup =		tmpfs_lookup,
1628 	.vop_create =			tmpfs_create,
1629 	.vop_mknod =			tmpfs_mknod,
1630 	.vop_open =			tmpfs_open,
1631 	.vop_close =			tmpfs_close,
1632 	.vop_access =			tmpfs_access,
1633 	.vop_getattr =			tmpfs_getattr,
1634 	.vop_setattr =			tmpfs_setattr,
1635 	.vop_read =			tmpfs_read,
1636 	.vop_write =			tmpfs_write,
1637 	.vop_fsync =			tmpfs_fsync,
1638 	.vop_remove =			tmpfs_remove,
1639 	.vop_link =			tmpfs_link,
1640 	.vop_rename =			tmpfs_rename,
1641 	.vop_mkdir =			tmpfs_mkdir,
1642 	.vop_rmdir =			tmpfs_rmdir,
1643 	.vop_symlink =			tmpfs_symlink,
1644 	.vop_readdir =			tmpfs_readdir,
1645 	.vop_readlink =			tmpfs_readlink,
1646 	.vop_inactive =			tmpfs_inactive,
1647 	.vop_reclaim =			tmpfs_reclaim,
1648 	.vop_print =			tmpfs_print,
1649 	.vop_pathconf =			tmpfs_pathconf,
1650 	.vop_vptofh =			tmpfs_vptofh,
1651 	.vop_whiteout =			tmpfs_whiteout,
1652 	.vop_bmap =			VOP_EOPNOTSUPP,
1653 };
1654 
1655