xref: /freebsd/sys/fs/tmpfs/tmpfs_vnops.c (revision 94e3ee44c3581ff37c5e01b5ffe5eb16d30079a7)
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/namei.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/sched.h>
46 #include <sys/sf_buf.h>
47 #include <sys/stat.h>
48 #include <sys/systm.h>
49 #include <sys/unistd.h>
50 #include <sys/vnode.h>
51 
52 #include <vm/vm.h>
53 #include <vm/vm_object.h>
54 #include <vm/vm_page.h>
55 #include <vm/vm_pager.h>
56 
57 #include <fs/tmpfs/tmpfs_vnops.h>
58 #include <fs/tmpfs/tmpfs.h>
59 
60 /* --------------------------------------------------------------------- */
61 
62 static int
63 tmpfs_lookup(struct vop_cachedlookup_args *v)
64 {
65 	struct vnode *dvp = v->a_dvp;
66 	struct vnode **vpp = v->a_vpp;
67 	struct componentname *cnp = v->a_cnp;
68 
69 	int error;
70 	struct tmpfs_dirent *de;
71 	struct tmpfs_node *dnode;
72 
73 	dnode = VP_TO_TMPFS_DIR(dvp);
74 	*vpp = NULLVP;
75 
76 	/* Check accessibility of requested node as a first step. */
77 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
78 	if (error != 0)
79 		goto out;
80 
81 	/* We cannot be requesting the parent directory of the root node. */
82 	MPASS(IMPLIES(dnode->tn_type == VDIR &&
83 	    dnode->tn_dir.tn_parent == dnode,
84 	    !(cnp->cn_flags & ISDOTDOT)));
85 
86 	TMPFS_ASSERT_LOCKED(dnode);
87 	if (dnode->tn_dir.tn_parent == NULL) {
88 		error = ENOENT;
89 		goto out;
90 	}
91 	if (cnp->cn_flags & ISDOTDOT) {
92 		int ltype = 0;
93 
94 		ltype = VOP_ISLOCKED(dvp);
95 		vhold(dvp);
96 		VOP_UNLOCK(dvp, 0);
97 		/* Allocate a new vnode on the matching entry. */
98 		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
99 		    cnp->cn_lkflags, vpp);
100 
101 		vn_lock(dvp, ltype | LK_RETRY);
102 		vdrop(dvp);
103 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
104 		VREF(dvp);
105 		*vpp = dvp;
106 		error = 0;
107 	} else {
108 		de = tmpfs_dir_lookup(dnode, NULL, cnp);
109 		if (de != NULL && de->td_node == NULL)
110 			cnp->cn_flags |= ISWHITEOUT;
111 		if (de == NULL || de->td_node == NULL) {
112 			/* The entry was not found in the directory.
113 			 * This is OK if we are creating or renaming an
114 			 * entry and are working on the last component of
115 			 * the path name. */
116 			if ((cnp->cn_flags & ISLASTCN) &&
117 			    (cnp->cn_nameiop == CREATE || \
118 			    cnp->cn_nameiop == RENAME ||
119 			    (cnp->cn_nameiop == DELETE &&
120 			    cnp->cn_flags & DOWHITEOUT &&
121 			    cnp->cn_flags & ISWHITEOUT))) {
122 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
123 				    cnp->cn_thread);
124 				if (error != 0)
125 					goto out;
126 
127 				/* Keep the component name in the buffer for
128 				 * future uses. */
129 				cnp->cn_flags |= SAVENAME;
130 
131 				error = EJUSTRETURN;
132 			} else
133 				error = ENOENT;
134 		} else {
135 			struct tmpfs_node *tnode;
136 
137 			/* The entry was found, so get its associated
138 			 * tmpfs_node. */
139 			tnode = de->td_node;
140 
141 			/* If we are not at the last path component and
142 			 * found a non-directory or non-link entry (which
143 			 * may itself be pointing to a directory), raise
144 			 * an error. */
145 			if ((tnode->tn_type != VDIR &&
146 			    tnode->tn_type != VLNK) &&
147 			    !(cnp->cn_flags & ISLASTCN)) {
148 				error = ENOTDIR;
149 				goto out;
150 			}
151 
152 			/* If we are deleting or renaming the entry, keep
153 			 * track of its tmpfs_dirent so that it can be
154 			 * easily deleted later. */
155 			if ((cnp->cn_flags & ISLASTCN) &&
156 			    (cnp->cn_nameiop == DELETE ||
157 			    cnp->cn_nameiop == RENAME)) {
158 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
159 				    cnp->cn_thread);
160 				if (error != 0)
161 					goto out;
162 
163 				/* Allocate a new vnode on the matching entry. */
164 				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
165 						cnp->cn_lkflags, vpp);
166 				if (error != 0)
167 					goto out;
168 
169 				if ((dnode->tn_mode & S_ISTXT) &&
170 				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) &&
171 				  VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) {
172 					error = EPERM;
173 					vput(*vpp);
174 					*vpp = NULL;
175 					goto out;
176 				}
177 				cnp->cn_flags |= SAVENAME;
178 			} else {
179 				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
180 						cnp->cn_lkflags, vpp);
181 			}
182 		}
183 	}
184 
185 	/* Store the result of this lookup in the cache.  Avoid this if the
186 	 * request was for creation, as it does not improve timings on
187 	 * emprical tests. */
188 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
189 		cache_enter(dvp, *vpp, cnp);
190 
191 out:
192 	/* If there were no errors, *vpp cannot be null and it must be
193 	 * locked. */
194 	MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
195 
196 	return error;
197 }
198 
199 /* --------------------------------------------------------------------- */
200 
201 static int
202 tmpfs_create(struct vop_create_args *v)
203 {
204 	struct vnode *dvp = v->a_dvp;
205 	struct vnode **vpp = v->a_vpp;
206 	struct componentname *cnp = v->a_cnp;
207 	struct vattr *vap = v->a_vap;
208 
209 	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
210 
211 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
212 }
213 /* --------------------------------------------------------------------- */
214 
215 static int
216 tmpfs_mknod(struct vop_mknod_args *v)
217 {
218 	struct vnode *dvp = v->a_dvp;
219 	struct vnode **vpp = v->a_vpp;
220 	struct componentname *cnp = v->a_cnp;
221 	struct vattr *vap = v->a_vap;
222 
223 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
224 	    vap->va_type != VFIFO)
225 		return EINVAL;
226 
227 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
228 }
229 
230 /* --------------------------------------------------------------------- */
231 
232 static int
233 tmpfs_open(struct vop_open_args *v)
234 {
235 	struct vnode *vp = v->a_vp;
236 	int mode = v->a_mode;
237 
238 	int error;
239 	struct tmpfs_node *node;
240 
241 	MPASS(VOP_ISLOCKED(vp));
242 
243 	node = VP_TO_TMPFS_NODE(vp);
244 
245 	/* The file is still active but all its names have been removed
246 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
247 	 * it is about to die. */
248 	if (node->tn_links < 1)
249 		return (ENOENT);
250 
251 	/* If the file is marked append-only, deny write requests. */
252 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
253 		error = EPERM;
254 	else {
255 		error = 0;
256 		vnode_create_vobject(vp, node->tn_size, v->a_td);
257 	}
258 
259 	MPASS(VOP_ISLOCKED(vp));
260 	return error;
261 }
262 
263 /* --------------------------------------------------------------------- */
264 
265 static int
266 tmpfs_close(struct vop_close_args *v)
267 {
268 	struct vnode *vp = v->a_vp;
269 
270 	MPASS(VOP_ISLOCKED(vp));
271 
272 	/* Update node times. */
273 	tmpfs_update(vp);
274 
275 	return (0);
276 }
277 
278 /* --------------------------------------------------------------------- */
279 
280 int
281 tmpfs_access(struct vop_access_args *v)
282 {
283 	struct vnode *vp = v->a_vp;
284 	accmode_t accmode = v->a_accmode;
285 	struct ucred *cred = v->a_cred;
286 
287 	int error;
288 	struct tmpfs_node *node;
289 
290 	MPASS(VOP_ISLOCKED(vp));
291 
292 	node = VP_TO_TMPFS_NODE(vp);
293 
294 	switch (vp->v_type) {
295 	case VDIR:
296 		/* FALLTHROUGH */
297 	case VLNK:
298 		/* FALLTHROUGH */
299 	case VREG:
300 		if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
301 			error = EROFS;
302 			goto out;
303 		}
304 		break;
305 
306 	case VBLK:
307 		/* FALLTHROUGH */
308 	case VCHR:
309 		/* FALLTHROUGH */
310 	case VSOCK:
311 		/* FALLTHROUGH */
312 	case VFIFO:
313 		break;
314 
315 	default:
316 		error = EINVAL;
317 		goto out;
318 	}
319 
320 	if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
321 		error = EPERM;
322 		goto out;
323 	}
324 
325 	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
326 	    node->tn_gid, accmode, cred, NULL);
327 
328 out:
329 	MPASS(VOP_ISLOCKED(vp));
330 
331 	return error;
332 }
333 
334 /* --------------------------------------------------------------------- */
335 
336 int
337 tmpfs_getattr(struct vop_getattr_args *v)
338 {
339 	struct vnode *vp = v->a_vp;
340 	struct vattr *vap = v->a_vap;
341 
342 	struct tmpfs_node *node;
343 
344 	node = VP_TO_TMPFS_NODE(vp);
345 
346 	tmpfs_update(vp);
347 
348 	vap->va_type = vp->v_type;
349 	vap->va_mode = node->tn_mode;
350 	vap->va_nlink = node->tn_links;
351 	vap->va_uid = node->tn_uid;
352 	vap->va_gid = node->tn_gid;
353 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
354 	vap->va_fileid = node->tn_id;
355 	vap->va_size = node->tn_size;
356 	vap->va_blocksize = PAGE_SIZE;
357 	vap->va_atime = node->tn_atime;
358 	vap->va_mtime = node->tn_mtime;
359 	vap->va_ctime = node->tn_ctime;
360 	vap->va_birthtime = node->tn_birthtime;
361 	vap->va_gen = node->tn_gen;
362 	vap->va_flags = node->tn_flags;
363 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
364 		node->tn_rdev : NODEV;
365 	vap->va_bytes = round_page(node->tn_size);
366 	vap->va_filerev = 0;
367 
368 	return 0;
369 }
370 
371 /* --------------------------------------------------------------------- */
372 
373 /* XXX Should this operation be atomic?  I think it should, but code in
374  * XXX other places (e.g., ufs) doesn't seem to be... */
375 int
376 tmpfs_setattr(struct vop_setattr_args *v)
377 {
378 	struct vnode *vp = v->a_vp;
379 	struct vattr *vap = v->a_vap;
380 	struct ucred *cred = v->a_cred;
381 	struct thread *td = curthread;
382 
383 	int error;
384 
385 	MPASS(VOP_ISLOCKED(vp));
386 
387 	error = 0;
388 
389 	/* Abort if any unsettable attribute is given. */
390 	if (vap->va_type != VNON ||
391 	    vap->va_nlink != VNOVAL ||
392 	    vap->va_fsid != VNOVAL ||
393 	    vap->va_fileid != VNOVAL ||
394 	    vap->va_blocksize != VNOVAL ||
395 	    vap->va_gen != VNOVAL ||
396 	    vap->va_rdev != VNOVAL ||
397 	    vap->va_bytes != VNOVAL)
398 		error = EINVAL;
399 
400 	if (error == 0 && (vap->va_flags != VNOVAL))
401 		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
402 
403 	if (error == 0 && (vap->va_size != VNOVAL))
404 		error = tmpfs_chsize(vp, vap->va_size, cred, td);
405 
406 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
407 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
408 
409 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
410 		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
411 
412 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
413 	    vap->va_atime.tv_nsec != VNOVAL) ||
414 	    (vap->va_mtime.tv_sec != VNOVAL &&
415 	    vap->va_mtime.tv_nsec != VNOVAL) ||
416 	    (vap->va_birthtime.tv_sec != VNOVAL &&
417 	    vap->va_birthtime.tv_nsec != VNOVAL)))
418 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
419 			&vap->va_birthtime, vap->va_vaflags, cred, td);
420 
421 	/* Update the node times.  We give preference to the error codes
422 	 * generated by this function rather than the ones that may arise
423 	 * from tmpfs_update. */
424 	tmpfs_update(vp);
425 
426 	MPASS(VOP_ISLOCKED(vp));
427 
428 	return error;
429 }
430 
431 /* --------------------------------------------------------------------- */
432 static int
433 tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
434     vm_offset_t offset, size_t tlen, struct uio *uio)
435 {
436 	vm_page_t	m;
437 	int		error, rv;
438 
439 	VM_OBJECT_LOCK(tobj);
440 	m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
441 	    VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
442 	if (m->valid != VM_PAGE_BITS_ALL) {
443 		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
444 			rv = vm_pager_get_pages(tobj, &m, 1, 0);
445 			if (rv != VM_PAGER_OK) {
446 				vm_page_lock(m);
447 				vm_page_free(m);
448 				vm_page_unlock(m);
449 				VM_OBJECT_UNLOCK(tobj);
450 				return (EIO);
451 			}
452 		} else
453 			vm_page_zero_invalid(m, TRUE);
454 	}
455 	VM_OBJECT_UNLOCK(tobj);
456 	error = uiomove_fromphys(&m, offset, tlen, uio);
457 	VM_OBJECT_LOCK(tobj);
458 	vm_page_lock(m);
459 	vm_page_unwire(m, TRUE);
460 	vm_page_unlock(m);
461 	vm_page_wakeup(m);
462 	VM_OBJECT_UNLOCK(tobj);
463 
464 	return (error);
465 }
466 
467 static __inline int
468 tmpfs_nocacheread_buf(vm_object_t tobj, vm_pindex_t idx,
469     vm_offset_t offset, size_t tlen, void *buf)
470 {
471 	struct uio uio;
472 	struct iovec iov;
473 
474 	uio.uio_iovcnt = 1;
475 	uio.uio_iov = &iov;
476 	iov.iov_base = buf;
477 	iov.iov_len = tlen;
478 
479 	uio.uio_offset = 0;
480 	uio.uio_resid = tlen;
481 	uio.uio_rw = UIO_READ;
482 	uio.uio_segflg = UIO_SYSSPACE;
483 	uio.uio_td = curthread;
484 
485 	return (tmpfs_nocacheread(tobj, idx, offset, tlen, &uio));
486 }
487 
488 static int
489 tmpfs_mappedread(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
490 {
491 	struct sf_buf	*sf;
492 	vm_pindex_t	idx;
493 	vm_page_t	m;
494 	vm_offset_t	offset;
495 	off_t		addr;
496 	size_t		tlen;
497 	char		*ma;
498 	int		error;
499 
500 	addr = uio->uio_offset;
501 	idx = OFF_TO_IDX(addr);
502 	offset = addr & PAGE_MASK;
503 	tlen = MIN(PAGE_SIZE - offset, len);
504 
505 	if ((vobj == NULL) ||
506 	    (vobj->resident_page_count == 0 && vobj->cache == NULL))
507 		goto nocache;
508 
509 	VM_OBJECT_LOCK(vobj);
510 lookupvpg:
511 	if (((m = vm_page_lookup(vobj, idx)) != NULL) &&
512 	    vm_page_is_valid(m, offset, tlen)) {
513 		if ((m->oflags & VPO_BUSY) != 0) {
514 			/*
515 			 * Reference the page before unlocking and sleeping so
516 			 * that the page daemon is less likely to reclaim it.
517 			 */
518 			vm_page_reference(m);
519 			vm_page_sleep(m, "tmfsmr");
520 			goto lookupvpg;
521 		}
522 		vm_page_busy(m);
523 		VM_OBJECT_UNLOCK(vobj);
524 		error = uiomove_fromphys(&m, offset, tlen, uio);
525 		VM_OBJECT_LOCK(vobj);
526 		vm_page_wakeup(m);
527 		VM_OBJECT_UNLOCK(vobj);
528 		return	(error);
529 	} else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) {
530 		KASSERT(offset == 0,
531 		    ("unexpected offset in tmpfs_mappedread for sendfile"));
532 		if ((m->oflags & VPO_BUSY) != 0) {
533 			/*
534 			 * Reference the page before unlocking and sleeping so
535 			 * that the page daemon is less likely to reclaim it.
536 			 */
537 			vm_page_reference(m);
538 			vm_page_sleep(m, "tmfsmr");
539 			goto lookupvpg;
540 		}
541 		vm_page_busy(m);
542 		VM_OBJECT_UNLOCK(vobj);
543 		sched_pin();
544 		sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
545 		ma = (char *)sf_buf_kva(sf);
546 		error = tmpfs_nocacheread_buf(tobj, idx, 0, tlen, ma);
547 		if (error == 0) {
548 			if (tlen != PAGE_SIZE)
549 				bzero(ma + tlen, PAGE_SIZE - tlen);
550 			uio->uio_offset += tlen;
551 			uio->uio_resid -= tlen;
552 		}
553 		sf_buf_free(sf);
554 		sched_unpin();
555 		VM_OBJECT_LOCK(vobj);
556 		if (error == 0)
557 			m->valid = VM_PAGE_BITS_ALL;
558 		vm_page_wakeup(m);
559 		VM_OBJECT_UNLOCK(vobj);
560 		return	(error);
561 	}
562 	VM_OBJECT_UNLOCK(vobj);
563 nocache:
564 	error = tmpfs_nocacheread(tobj, idx, offset, tlen, uio);
565 
566 	return	(error);
567 }
568 
569 static int
570 tmpfs_read(struct vop_read_args *v)
571 {
572 	struct vnode *vp = v->a_vp;
573 	struct uio *uio = v->a_uio;
574 
575 	struct tmpfs_node *node;
576 	vm_object_t uobj;
577 	size_t len;
578 	int resid;
579 
580 	int error = 0;
581 
582 	node = VP_TO_TMPFS_NODE(vp);
583 
584 	if (vp->v_type != VREG) {
585 		error = EISDIR;
586 		goto out;
587 	}
588 
589 	if (uio->uio_offset < 0) {
590 		error = EINVAL;
591 		goto out;
592 	}
593 
594 	node->tn_status |= TMPFS_NODE_ACCESSED;
595 
596 	uobj = node->tn_reg.tn_aobj;
597 	while ((resid = uio->uio_resid) > 0) {
598 		error = 0;
599 		if (node->tn_size <= uio->uio_offset)
600 			break;
601 		len = MIN(node->tn_size - uio->uio_offset, resid);
602 		if (len == 0)
603 			break;
604 		error = tmpfs_mappedread(vp->v_object, uobj, len, uio);
605 		if ((error != 0) || (resid == uio->uio_resid))
606 			break;
607 	}
608 
609 out:
610 
611 	return error;
612 }
613 
614 /* --------------------------------------------------------------------- */
615 
616 static int
617 tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
618 {
619 	vm_pindex_t	idx;
620 	vm_page_t	vpg, tpg;
621 	vm_offset_t	offset;
622 	off_t		addr;
623 	size_t		tlen;
624 	int		error, rv;
625 
626 	error = 0;
627 
628 	addr = uio->uio_offset;
629 	idx = OFF_TO_IDX(addr);
630 	offset = addr & PAGE_MASK;
631 	tlen = MIN(PAGE_SIZE - offset, len);
632 
633 	if ((vobj == NULL) ||
634 	    (vobj->resident_page_count == 0 && vobj->cache == NULL)) {
635 		vpg = NULL;
636 		goto nocache;
637 	}
638 
639 	VM_OBJECT_LOCK(vobj);
640 lookupvpg:
641 	if (((vpg = vm_page_lookup(vobj, idx)) != NULL) &&
642 	    vm_page_is_valid(vpg, offset, tlen)) {
643 		if ((vpg->oflags & VPO_BUSY) != 0) {
644 			/*
645 			 * Reference the page before unlocking and sleeping so
646 			 * that the page daemon is less likely to reclaim it.
647 			 */
648 			vm_page_reference(vpg);
649 			vm_page_sleep(vpg, "tmfsmw");
650 			goto lookupvpg;
651 		}
652 		vm_page_busy(vpg);
653 		vm_page_undirty(vpg);
654 		VM_OBJECT_UNLOCK(vobj);
655 		error = uiomove_fromphys(&vpg, offset, tlen, uio);
656 	} else {
657 		if (__predict_false(vobj->cache != NULL))
658 			vm_page_cache_free(vobj, idx, idx + 1);
659 		VM_OBJECT_UNLOCK(vobj);
660 		vpg = NULL;
661 	}
662 nocache:
663 	VM_OBJECT_LOCK(tobj);
664 	tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
665 	    VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
666 	if (tpg->valid != VM_PAGE_BITS_ALL) {
667 		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
668 			rv = vm_pager_get_pages(tobj, &tpg, 1, 0);
669 			if (rv != VM_PAGER_OK) {
670 				vm_page_lock(tpg);
671 				vm_page_free(tpg);
672 				vm_page_unlock(tpg);
673 				error = EIO;
674 				goto out;
675 			}
676 		} else
677 			vm_page_zero_invalid(tpg, TRUE);
678 	}
679 	VM_OBJECT_UNLOCK(tobj);
680 	if (vpg == NULL)
681 		error = uiomove_fromphys(&tpg, offset, tlen, uio);
682 	else {
683 		KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid"));
684 		pmap_copy_page(vpg, tpg);
685 	}
686 	VM_OBJECT_LOCK(tobj);
687 	if (error == 0) {
688 		KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
689 		    ("parts of tpg invalid"));
690 		vm_page_dirty(tpg);
691 	}
692 	vm_page_lock(tpg);
693 	vm_page_unwire(tpg, TRUE);
694 	vm_page_unlock(tpg);
695 	vm_page_wakeup(tpg);
696 out:
697 	VM_OBJECT_UNLOCK(tobj);
698 	if (vpg != NULL) {
699 		VM_OBJECT_LOCK(vobj);
700 		vm_page_wakeup(vpg);
701 		VM_OBJECT_UNLOCK(vobj);
702 	}
703 
704 	return	(error);
705 }
706 
707 static int
708 tmpfs_write(struct vop_write_args *v)
709 {
710 	struct vnode *vp = v->a_vp;
711 	struct uio *uio = v->a_uio;
712 	int ioflag = v->a_ioflag;
713 
714 	boolean_t extended;
715 	int error = 0;
716 	off_t oldsize;
717 	struct tmpfs_node *node;
718 	vm_object_t uobj;
719 	size_t len;
720 	int resid;
721 
722 	node = VP_TO_TMPFS_NODE(vp);
723 	oldsize = node->tn_size;
724 
725 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
726 		error = EINVAL;
727 		goto out;
728 	}
729 
730 	if (uio->uio_resid == 0) {
731 		error = 0;
732 		goto out;
733 	}
734 
735 	if (ioflag & IO_APPEND)
736 		uio->uio_offset = node->tn_size;
737 
738 	if (uio->uio_offset + uio->uio_resid >
739 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
740 		return (EFBIG);
741 
742 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
743 		return (EFBIG);
744 
745 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
746 	if (extended) {
747 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid,
748 		    FALSE);
749 		if (error != 0)
750 			goto out;
751 	}
752 
753 	uobj = node->tn_reg.tn_aobj;
754 	while ((resid = uio->uio_resid) > 0) {
755 		if (node->tn_size <= uio->uio_offset)
756 			break;
757 		len = MIN(node->tn_size - uio->uio_offset, resid);
758 		if (len == 0)
759 			break;
760 		error = tmpfs_mappedwrite(vp->v_object, uobj, len, uio);
761 		if ((error != 0) || (resid == uio->uio_resid))
762 			break;
763 	}
764 
765 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
766 	    (extended ? TMPFS_NODE_CHANGED : 0);
767 
768 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
769 		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
770 			node->tn_mode &= ~(S_ISUID | S_ISGID);
771 	}
772 
773 	if (error != 0)
774 		(void)tmpfs_reg_resize(vp, oldsize, TRUE);
775 
776 out:
777 	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
778 	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
779 
780 	return error;
781 }
782 
783 /* --------------------------------------------------------------------- */
784 
785 static int
786 tmpfs_fsync(struct vop_fsync_args *v)
787 {
788 	struct vnode *vp = v->a_vp;
789 
790 	MPASS(VOP_ISLOCKED(vp));
791 
792 	tmpfs_update(vp);
793 
794 	return 0;
795 }
796 
797 /* --------------------------------------------------------------------- */
798 
799 static int
800 tmpfs_remove(struct vop_remove_args *v)
801 {
802 	struct vnode *dvp = v->a_dvp;
803 	struct vnode *vp = v->a_vp;
804 
805 	int error;
806 	struct tmpfs_dirent *de;
807 	struct tmpfs_mount *tmp;
808 	struct tmpfs_node *dnode;
809 	struct tmpfs_node *node;
810 
811 	MPASS(VOP_ISLOCKED(dvp));
812 	MPASS(VOP_ISLOCKED(vp));
813 
814 	if (vp->v_type == VDIR) {
815 		error = EISDIR;
816 		goto out;
817 	}
818 
819 	dnode = VP_TO_TMPFS_DIR(dvp);
820 	node = VP_TO_TMPFS_NODE(vp);
821 	tmp = VFS_TO_TMPFS(vp->v_mount);
822 	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
823 	MPASS(de != NULL);
824 
825 	/* Files marked as immutable or append-only cannot be deleted. */
826 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
827 	    (dnode->tn_flags & APPEND)) {
828 		error = EPERM;
829 		goto out;
830 	}
831 
832 	/* Remove the entry from the directory; as it is a file, we do not
833 	 * have to change the number of hard links of the directory. */
834 	tmpfs_dir_detach(dvp, de);
835 	if (v->a_cnp->cn_flags & DOWHITEOUT)
836 		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
837 
838 	/* Free the directory entry we just deleted.  Note that the node
839 	 * referred by it will not be removed until the vnode is really
840 	 * reclaimed. */
841 	tmpfs_free_dirent(tmp, de, TRUE);
842 
843 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
844 	error = 0;
845 
846 out:
847 
848 	return error;
849 }
850 
851 /* --------------------------------------------------------------------- */
852 
853 static int
854 tmpfs_link(struct vop_link_args *v)
855 {
856 	struct vnode *dvp = v->a_tdvp;
857 	struct vnode *vp = v->a_vp;
858 	struct componentname *cnp = v->a_cnp;
859 
860 	int error;
861 	struct tmpfs_dirent *de;
862 	struct tmpfs_node *node;
863 
864 	MPASS(VOP_ISLOCKED(dvp));
865 	MPASS(cnp->cn_flags & HASBUF);
866 	MPASS(dvp != vp); /* XXX When can this be false? */
867 
868 	node = VP_TO_TMPFS_NODE(vp);
869 
870 	/* XXX: Why aren't the following two tests done by the caller? */
871 
872 	/* Hard links of directories are forbidden. */
873 	if (vp->v_type == VDIR) {
874 		error = EPERM;
875 		goto out;
876 	}
877 
878 	/* Cannot create cross-device links. */
879 	if (dvp->v_mount != vp->v_mount) {
880 		error = EXDEV;
881 		goto out;
882 	}
883 
884 	/* Ensure that we do not overflow the maximum number of links imposed
885 	 * by the system. */
886 	MPASS(node->tn_links <= LINK_MAX);
887 	if (node->tn_links == LINK_MAX) {
888 		error = EMLINK;
889 		goto out;
890 	}
891 
892 	/* We cannot create links of files marked immutable or append-only. */
893 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
894 		error = EPERM;
895 		goto out;
896 	}
897 
898 	/* Allocate a new directory entry to represent the node. */
899 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
900 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
901 	if (error != 0)
902 		goto out;
903 
904 	/* Insert the new directory entry into the appropriate directory. */
905 	if (cnp->cn_flags & ISWHITEOUT)
906 		tmpfs_dir_whiteout_remove(dvp, cnp);
907 	tmpfs_dir_attach(dvp, de);
908 
909 	/* vp link count has changed, so update node times. */
910 	node->tn_status |= TMPFS_NODE_CHANGED;
911 	tmpfs_update(vp);
912 
913 	error = 0;
914 
915 out:
916 	return error;
917 }
918 
919 /* --------------------------------------------------------------------- */
920 
921 static int
922 tmpfs_rename(struct vop_rename_args *v)
923 {
924 	struct vnode *fdvp = v->a_fdvp;
925 	struct vnode *fvp = v->a_fvp;
926 	struct componentname *fcnp = v->a_fcnp;
927 	struct vnode *tdvp = v->a_tdvp;
928 	struct vnode *tvp = v->a_tvp;
929 	struct componentname *tcnp = v->a_tcnp;
930 
931 	char *newname;
932 	int error;
933 	struct tmpfs_dirent *de;
934 	struct tmpfs_mount *tmp;
935 	struct tmpfs_node *fdnode;
936 	struct tmpfs_node *fnode;
937 	struct tmpfs_node *tnode;
938 	struct tmpfs_node *tdnode;
939 
940 	MPASS(VOP_ISLOCKED(tdvp));
941 	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
942 	MPASS(fcnp->cn_flags & HASBUF);
943 	MPASS(tcnp->cn_flags & HASBUF);
944 
945   	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
946 
947 	/* Disallow cross-device renames.
948 	 * XXX Why isn't this done by the caller? */
949 	if (fvp->v_mount != tdvp->v_mount ||
950 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
951 		error = EXDEV;
952 		goto out;
953 	}
954 
955 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
956 	tdnode = VP_TO_TMPFS_DIR(tdvp);
957 
958 	/* If source and target are the same file, there is nothing to do. */
959 	if (fvp == tvp) {
960 		error = 0;
961 		goto out;
962 	}
963 
964 	/* If we need to move the directory between entries, lock the
965 	 * source so that we can safely operate on it. */
966 	if (fdvp != tdvp && fdvp != tvp)
967 		vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
968 	fdnode = VP_TO_TMPFS_DIR(fdvp);
969 	fnode = VP_TO_TMPFS_NODE(fvp);
970 	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
971 
972 	/* Entry can disappear before we lock fdvp,
973 	 * also avoid manipulating '.' and '..' entries. */
974 	if (de == NULL) {
975 		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
976 		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
977 			error = EINVAL;
978 		else
979 			error = ENOENT;
980 		goto out_locked;
981 	}
982 	MPASS(de->td_node == fnode);
983 
984 	/* If re-naming a directory to another preexisting directory
985 	 * ensure that the target directory is empty so that its
986 	 * removal causes no side effects.
987 	 * Kern_rename gurantees the destination to be a directory
988 	 * if the source is one. */
989 	if (tvp != NULL) {
990 		MPASS(tnode != NULL);
991 
992 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
993 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
994 			error = EPERM;
995 			goto out_locked;
996 		}
997 
998 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
999 			if (tnode->tn_size > 0) {
1000 				error = ENOTEMPTY;
1001 				goto out_locked;
1002 			}
1003 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1004 			error = ENOTDIR;
1005 			goto out_locked;
1006 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1007 			error = EISDIR;
1008 			goto out_locked;
1009 		} else {
1010 			MPASS(fnode->tn_type != VDIR &&
1011 				tnode->tn_type != VDIR);
1012 		}
1013 	}
1014 
1015 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1016 	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1017 		error = EPERM;
1018 		goto out_locked;
1019 	}
1020 
1021 	/* Ensure that we have enough memory to hold the new name, if it
1022 	 * has to be changed. */
1023 	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1024 	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1025 		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1026 	} else
1027 		newname = NULL;
1028 
1029 	/* If the node is being moved to another directory, we have to do
1030 	 * the move. */
1031 	if (fdnode != tdnode) {
1032 		/* In case we are moving a directory, we have to adjust its
1033 		 * parent to point to the new parent. */
1034 		if (de->td_node->tn_type == VDIR) {
1035 			struct tmpfs_node *n;
1036 
1037 			/* Ensure the target directory is not a child of the
1038 			 * directory being moved.  Otherwise, we'd end up
1039 			 * with stale nodes. */
1040 			n = tdnode;
1041 			/* TMPFS_LOCK garanties that no nodes are freed while
1042 			 * traversing the list. Nodes can only be marked as
1043 			 * removed: tn_parent == NULL. */
1044 			TMPFS_LOCK(tmp);
1045 			TMPFS_NODE_LOCK(n);
1046 			while (n != n->tn_dir.tn_parent) {
1047 				struct tmpfs_node *parent;
1048 
1049 				if (n == fnode) {
1050 					TMPFS_NODE_UNLOCK(n);
1051 					TMPFS_UNLOCK(tmp);
1052 					error = EINVAL;
1053 					if (newname != NULL)
1054 						    free(newname, M_TMPFSNAME);
1055 					goto out_locked;
1056 				}
1057 				parent = n->tn_dir.tn_parent;
1058 				TMPFS_NODE_UNLOCK(n);
1059 				if (parent == NULL) {
1060 					n = NULL;
1061 					break;
1062 				}
1063 				TMPFS_NODE_LOCK(parent);
1064 				if (parent->tn_dir.tn_parent == NULL) {
1065 					TMPFS_NODE_UNLOCK(parent);
1066 					n = NULL;
1067 					break;
1068 				}
1069 				n = parent;
1070 			}
1071 			TMPFS_UNLOCK(tmp);
1072 			if (n == NULL) {
1073 				error = EINVAL;
1074 				if (newname != NULL)
1075 					    free(newname, M_TMPFSNAME);
1076 				goto out_locked;
1077 			}
1078 			TMPFS_NODE_UNLOCK(n);
1079 
1080 			/* Adjust the parent pointer. */
1081 			TMPFS_VALIDATE_DIR(fnode);
1082 			TMPFS_NODE_LOCK(de->td_node);
1083 			de->td_node->tn_dir.tn_parent = tdnode;
1084 			TMPFS_NODE_UNLOCK(de->td_node);
1085 
1086 			/* As a result of changing the target of the '..'
1087 			 * entry, the link count of the source and target
1088 			 * directories has to be adjusted. */
1089 			TMPFS_NODE_LOCK(tdnode);
1090 			TMPFS_ASSERT_LOCKED(tdnode);
1091 			tdnode->tn_links++;
1092 			TMPFS_NODE_UNLOCK(tdnode);
1093 
1094 			TMPFS_NODE_LOCK(fdnode);
1095 			TMPFS_ASSERT_LOCKED(fdnode);
1096 			fdnode->tn_links--;
1097 			TMPFS_NODE_UNLOCK(fdnode);
1098 		}
1099 
1100 		/* Do the move: just remove the entry from the source directory
1101 		 * and insert it into the target one. */
1102 		tmpfs_dir_detach(fdvp, de);
1103 		if (fcnp->cn_flags & DOWHITEOUT)
1104 			tmpfs_dir_whiteout_add(fdvp, fcnp);
1105 		if (tcnp->cn_flags & ISWHITEOUT)
1106 			tmpfs_dir_whiteout_remove(tdvp, tcnp);
1107 		tmpfs_dir_attach(tdvp, de);
1108 	}
1109 
1110 	/* If the name has changed, we need to make it effective by changing
1111 	 * it in the directory entry. */
1112 	if (newname != NULL) {
1113 		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1114 
1115 		free(de->td_name, M_TMPFSNAME);
1116 		de->td_namelen = (uint16_t)tcnp->cn_namelen;
1117 		memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen);
1118 		de->td_name = newname;
1119 
1120 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1121 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1122 	}
1123 
1124 	/* If we are overwriting an entry, we have to remove the old one
1125 	 * from the target directory. */
1126 	if (tvp != NULL) {
1127 		/* Remove the old entry from the target directory. */
1128 		de = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1129 		tmpfs_dir_detach(tdvp, de);
1130 
1131 		/* Free the directory entry we just deleted.  Note that the
1132 		 * node referred by it will not be removed until the vnode is
1133 		 * really reclaimed. */
1134 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE);
1135 	}
1136 	cache_purge(fvp);
1137 	if (tvp != NULL)
1138 		cache_purge(tvp);
1139 
1140 	error = 0;
1141 
1142 out_locked:
1143 	if (fdvp != tdvp && fdvp != tvp)
1144 		VOP_UNLOCK(fdvp, 0);
1145 
1146 out:
1147 	/* Release target nodes. */
1148 	/* XXX: I don't understand when tdvp can be the same as tvp, but
1149 	 * other code takes care of this... */
1150 	if (tdvp == tvp)
1151 		vrele(tdvp);
1152 	else
1153 		vput(tdvp);
1154 	if (tvp != NULL)
1155 		vput(tvp);
1156 
1157 	/* Release source nodes. */
1158 	vrele(fdvp);
1159 	vrele(fvp);
1160 
1161 	return error;
1162 }
1163 
1164 /* --------------------------------------------------------------------- */
1165 
1166 static int
1167 tmpfs_mkdir(struct vop_mkdir_args *v)
1168 {
1169 	struct vnode *dvp = v->a_dvp;
1170 	struct vnode **vpp = v->a_vpp;
1171 	struct componentname *cnp = v->a_cnp;
1172 	struct vattr *vap = v->a_vap;
1173 
1174 	MPASS(vap->va_type == VDIR);
1175 
1176 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1177 }
1178 
1179 /* --------------------------------------------------------------------- */
1180 
1181 static int
1182 tmpfs_rmdir(struct vop_rmdir_args *v)
1183 {
1184 	struct vnode *dvp = v->a_dvp;
1185 	struct vnode *vp = v->a_vp;
1186 
1187 	int error;
1188 	struct tmpfs_dirent *de;
1189 	struct tmpfs_mount *tmp;
1190 	struct tmpfs_node *dnode;
1191 	struct tmpfs_node *node;
1192 
1193 	MPASS(VOP_ISLOCKED(dvp));
1194 	MPASS(VOP_ISLOCKED(vp));
1195 
1196 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1197 	dnode = VP_TO_TMPFS_DIR(dvp);
1198 	node = VP_TO_TMPFS_DIR(vp);
1199 
1200 	/* Directories with more than two entries ('.' and '..') cannot be
1201 	 * removed. */
1202 	 if (node->tn_size > 0) {
1203 		 error = ENOTEMPTY;
1204 		 goto out;
1205 	 }
1206 
1207 	if ((dnode->tn_flags & APPEND)
1208 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1209 		error = EPERM;
1210 		goto out;
1211 	}
1212 
1213 	/* This invariant holds only if we are not trying to remove "..".
1214 	  * We checked for that above so this is safe now. */
1215 	MPASS(node->tn_dir.tn_parent == dnode);
1216 
1217 	/* Get the directory entry associated with node (vp).  This was
1218 	 * filled by tmpfs_lookup while looking up the entry. */
1219 	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1220 	MPASS(TMPFS_DIRENT_MATCHES(de,
1221 	    v->a_cnp->cn_nameptr,
1222 	    v->a_cnp->cn_namelen));
1223 
1224 	/* Check flags to see if we are allowed to remove the directory. */
1225 	if (dnode->tn_flags & APPEND
1226 		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1227 		error = EPERM;
1228 		goto out;
1229 	}
1230 
1231 
1232 	/* Detach the directory entry from the directory (dnode). */
1233 	tmpfs_dir_detach(dvp, de);
1234 	if (v->a_cnp->cn_flags & DOWHITEOUT)
1235 		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1236 
1237 	/* No vnode should be allocated for this entry from this point */
1238 	TMPFS_NODE_LOCK(node);
1239 	TMPFS_ASSERT_ELOCKED(node);
1240 	node->tn_links--;
1241 	node->tn_dir.tn_parent = NULL;
1242 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1243 	    TMPFS_NODE_MODIFIED;
1244 
1245 	TMPFS_NODE_UNLOCK(node);
1246 
1247 	TMPFS_NODE_LOCK(dnode);
1248 	TMPFS_ASSERT_ELOCKED(dnode);
1249 	dnode->tn_links--;
1250 	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1251 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1252 	TMPFS_NODE_UNLOCK(dnode);
1253 
1254 	cache_purge(dvp);
1255 	cache_purge(vp);
1256 
1257 	/* Free the directory entry we just deleted.  Note that the node
1258 	 * referred by it will not be removed until the vnode is really
1259 	 * reclaimed. */
1260 	tmpfs_free_dirent(tmp, de, TRUE);
1261 
1262 	/* Release the deleted vnode (will destroy the node, notify
1263 	 * interested parties and clean it from the cache). */
1264 
1265 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1266 	tmpfs_update(dvp);
1267 
1268 	error = 0;
1269 
1270 out:
1271 	return error;
1272 }
1273 
1274 /* --------------------------------------------------------------------- */
1275 
1276 static int
1277 tmpfs_symlink(struct vop_symlink_args *v)
1278 {
1279 	struct vnode *dvp = v->a_dvp;
1280 	struct vnode **vpp = v->a_vpp;
1281 	struct componentname *cnp = v->a_cnp;
1282 	struct vattr *vap = v->a_vap;
1283 	char *target = v->a_target;
1284 
1285 #ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1286 	MPASS(vap->va_type == VLNK);
1287 #else
1288 	vap->va_type = VLNK;
1289 #endif
1290 
1291 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1292 }
1293 
1294 /* --------------------------------------------------------------------- */
1295 
1296 static int
1297 tmpfs_readdir(struct vop_readdir_args *v)
1298 {
1299 	struct vnode *vp = v->a_vp;
1300 	struct uio *uio = v->a_uio;
1301 	int *eofflag = v->a_eofflag;
1302 	u_long **cookies = v->a_cookies;
1303 	int *ncookies = v->a_ncookies;
1304 
1305 	int error;
1306 	off_t startoff;
1307 	off_t cnt = 0;
1308 	struct tmpfs_node *node;
1309 
1310 	/* This operation only makes sense on directory nodes. */
1311 	if (vp->v_type != VDIR)
1312 		return ENOTDIR;
1313 
1314 	node = VP_TO_TMPFS_DIR(vp);
1315 
1316 	startoff = uio->uio_offset;
1317 
1318 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1319 		error = tmpfs_dir_getdotdent(node, uio);
1320 		if (error != 0)
1321 			goto outok;
1322 		cnt++;
1323 	}
1324 
1325 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1326 		error = tmpfs_dir_getdotdotdent(node, uio);
1327 		if (error != 0)
1328 			goto outok;
1329 		cnt++;
1330 	}
1331 
1332 	error = tmpfs_dir_getdents(node, uio, &cnt);
1333 
1334 outok:
1335 	MPASS(error >= -1);
1336 
1337 	if (error == -1)
1338 		error = (cnt != 0) ? 0 : EINVAL;
1339 
1340 	if (eofflag != NULL)
1341 		*eofflag =
1342 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1343 
1344 	/* Update NFS-related variables. */
1345 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1346 		off_t i;
1347 		off_t off = startoff;
1348 		struct tmpfs_dirent *de = NULL;
1349 
1350 		*ncookies = cnt;
1351 		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1352 
1353 		for (i = 0; i < cnt; i++) {
1354 			MPASS(off != TMPFS_DIRCOOKIE_EOF);
1355 			if (off == TMPFS_DIRCOOKIE_DOT) {
1356 				off = TMPFS_DIRCOOKIE_DOTDOT;
1357 			} else {
1358 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1359 					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1360 				} else if (de != NULL) {
1361 					de = TAILQ_NEXT(de, td_entries);
1362 				} else {
1363 					de = tmpfs_dir_lookupbycookie(node,
1364 					    off);
1365 					MPASS(de != NULL);
1366 					de = TAILQ_NEXT(de, td_entries);
1367 				}
1368 				if (de == NULL)
1369 					off = TMPFS_DIRCOOKIE_EOF;
1370 				else
1371 					off = tmpfs_dircookie(de);
1372 			}
1373 
1374 			(*cookies)[i] = off;
1375 		}
1376 		MPASS(uio->uio_offset == off);
1377 	}
1378 
1379 	return error;
1380 }
1381 
1382 /* --------------------------------------------------------------------- */
1383 
1384 static int
1385 tmpfs_readlink(struct vop_readlink_args *v)
1386 {
1387 	struct vnode *vp = v->a_vp;
1388 	struct uio *uio = v->a_uio;
1389 
1390 	int error;
1391 	struct tmpfs_node *node;
1392 
1393 	MPASS(uio->uio_offset == 0);
1394 	MPASS(vp->v_type == VLNK);
1395 
1396 	node = VP_TO_TMPFS_NODE(vp);
1397 
1398 	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1399 	    uio);
1400 	node->tn_status |= TMPFS_NODE_ACCESSED;
1401 
1402 	return error;
1403 }
1404 
1405 /* --------------------------------------------------------------------- */
1406 
1407 static int
1408 tmpfs_inactive(struct vop_inactive_args *v)
1409 {
1410 	struct vnode *vp = v->a_vp;
1411 	struct thread *l = v->a_td;
1412 
1413 	struct tmpfs_node *node;
1414 
1415 	MPASS(VOP_ISLOCKED(vp));
1416 
1417 	node = VP_TO_TMPFS_NODE(vp);
1418 
1419 	if (node->tn_links == 0)
1420 		vrecycle(vp, l);
1421 
1422 	return 0;
1423 }
1424 
1425 /* --------------------------------------------------------------------- */
1426 
1427 int
1428 tmpfs_reclaim(struct vop_reclaim_args *v)
1429 {
1430 	struct vnode *vp = v->a_vp;
1431 
1432 	struct tmpfs_mount *tmp;
1433 	struct tmpfs_node *node;
1434 
1435 	node = VP_TO_TMPFS_NODE(vp);
1436 	tmp = VFS_TO_TMPFS(vp->v_mount);
1437 
1438 	vnode_destroy_vobject(vp);
1439 	cache_purge(vp);
1440 
1441 	TMPFS_NODE_LOCK(node);
1442 	TMPFS_ASSERT_ELOCKED(node);
1443 	tmpfs_free_vp(vp);
1444 
1445 	/* If the node referenced by this vnode was deleted by the user,
1446 	 * we must free its associated data structures (now that the vnode
1447 	 * is being reclaimed). */
1448 	if (node->tn_links == 0 &&
1449 	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1450 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1451 		TMPFS_NODE_UNLOCK(node);
1452 		tmpfs_free_node(tmp, node);
1453 	} else
1454 		TMPFS_NODE_UNLOCK(node);
1455 
1456 	MPASS(vp->v_data == NULL);
1457 	return 0;
1458 }
1459 
1460 /* --------------------------------------------------------------------- */
1461 
1462 static int
1463 tmpfs_print(struct vop_print_args *v)
1464 {
1465 	struct vnode *vp = v->a_vp;
1466 
1467 	struct tmpfs_node *node;
1468 
1469 	node = VP_TO_TMPFS_NODE(vp);
1470 
1471 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1472 	    node, node->tn_flags, node->tn_links);
1473 	printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n",
1474 	    node->tn_mode, node->tn_uid, node->tn_gid,
1475 	    (intmax_t)node->tn_size, node->tn_status);
1476 
1477 	if (vp->v_type == VFIFO)
1478 		fifo_printinfo(vp);
1479 
1480 	printf("\n");
1481 
1482 	return 0;
1483 }
1484 
1485 /* --------------------------------------------------------------------- */
1486 
1487 static int
1488 tmpfs_pathconf(struct vop_pathconf_args *v)
1489 {
1490 	int name = v->a_name;
1491 	register_t *retval = v->a_retval;
1492 
1493 	int error;
1494 
1495 	error = 0;
1496 
1497 	switch (name) {
1498 	case _PC_LINK_MAX:
1499 		*retval = LINK_MAX;
1500 		break;
1501 
1502 	case _PC_NAME_MAX:
1503 		*retval = NAME_MAX;
1504 		break;
1505 
1506 	case _PC_PATH_MAX:
1507 		*retval = PATH_MAX;
1508 		break;
1509 
1510 	case _PC_PIPE_BUF:
1511 		*retval = PIPE_BUF;
1512 		break;
1513 
1514 	case _PC_CHOWN_RESTRICTED:
1515 		*retval = 1;
1516 		break;
1517 
1518 	case _PC_NO_TRUNC:
1519 		*retval = 1;
1520 		break;
1521 
1522 	case _PC_SYNC_IO:
1523 		*retval = 1;
1524 		break;
1525 
1526 	case _PC_FILESIZEBITS:
1527 		*retval = 0; /* XXX Don't know which value should I return. */
1528 		break;
1529 
1530 	default:
1531 		error = EINVAL;
1532 	}
1533 
1534 	return error;
1535 }
1536 
1537 static int
1538 tmpfs_vptofh(struct vop_vptofh_args *ap)
1539 {
1540 	struct tmpfs_fid *tfhp;
1541 	struct tmpfs_node *node;
1542 
1543 	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1544 	node = VP_TO_TMPFS_NODE(ap->a_vp);
1545 
1546 	tfhp->tf_len = sizeof(struct tmpfs_fid);
1547 	tfhp->tf_id = node->tn_id;
1548 	tfhp->tf_gen = node->tn_gen;
1549 
1550 	return (0);
1551 }
1552 
1553 static int
1554 tmpfs_whiteout(struct vop_whiteout_args *ap)
1555 {
1556 	struct vnode *dvp = ap->a_dvp;
1557 	struct componentname *cnp = ap->a_cnp;
1558 	struct tmpfs_dirent *de;
1559 
1560 	switch (ap->a_flags) {
1561 	case LOOKUP:
1562 		return (0);
1563 	case CREATE:
1564 		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1565 		if (de != NULL)
1566 			return (de->td_node == NULL ? 0 : EEXIST);
1567 		return (tmpfs_dir_whiteout_add(dvp, cnp));
1568 	case DELETE:
1569 		tmpfs_dir_whiteout_remove(dvp, cnp);
1570 		return (0);
1571 	default:
1572 		panic("tmpfs_whiteout: unknown op");
1573 	}
1574 }
1575 
1576 /* --------------------------------------------------------------------- */
1577 
1578 /*
1579  * vnode operations vector used for files stored in a tmpfs file system.
1580  */
1581 struct vop_vector tmpfs_vnodeop_entries = {
1582 	.vop_default =			&default_vnodeops,
1583 	.vop_lookup =			vfs_cache_lookup,
1584 	.vop_cachedlookup =		tmpfs_lookup,
1585 	.vop_create =			tmpfs_create,
1586 	.vop_mknod =			tmpfs_mknod,
1587 	.vop_open =			tmpfs_open,
1588 	.vop_close =			tmpfs_close,
1589 	.vop_access =			tmpfs_access,
1590 	.vop_getattr =			tmpfs_getattr,
1591 	.vop_setattr =			tmpfs_setattr,
1592 	.vop_read =			tmpfs_read,
1593 	.vop_write =			tmpfs_write,
1594 	.vop_fsync =			tmpfs_fsync,
1595 	.vop_remove =			tmpfs_remove,
1596 	.vop_link =			tmpfs_link,
1597 	.vop_rename =			tmpfs_rename,
1598 	.vop_mkdir =			tmpfs_mkdir,
1599 	.vop_rmdir =			tmpfs_rmdir,
1600 	.vop_symlink =			tmpfs_symlink,
1601 	.vop_readdir =			tmpfs_readdir,
1602 	.vop_readlink =			tmpfs_readlink,
1603 	.vop_inactive =			tmpfs_inactive,
1604 	.vop_reclaim =			tmpfs_reclaim,
1605 	.vop_print =			tmpfs_print,
1606 	.vop_pathconf =			tmpfs_pathconf,
1607 	.vop_vptofh =			tmpfs_vptofh,
1608 	.vop_whiteout =			tmpfs_whiteout,
1609 	.vop_bmap =			VOP_EOPNOTSUPP,
1610 };
1611 
1612