xref: /dflybsd-src/sys/vfs/tmpfs/tmpfs_vnops.c (revision d76863bd1dbf9c1e4c04a977e3a7271a23cdb718)
1 /*-
2  * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
3  * All rights reserved.
4  *
5  * This code is derived from software contributed to The NetBSD Foundation
6  * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
7  * 2005 program.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
22  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28  * POSSIBILITY OF SUCH DAMAGE.
29  *
30  * $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $
31  */
32 
33 /*
34  * tmpfs vnode interface.
35  */
36 
37 #include <sys/kernel.h>
38 #include <sys/kern_syscall.h>
39 #include <sys/param.h>
40 #include <sys/fcntl.h>
41 #include <sys/lockf.h>
42 #include <sys/priv.h>
43 #include <sys/proc.h>
44 #include <sys/resourcevar.h>
45 #include <sys/sched.h>
46 #include <sys/stat.h>
47 #include <sys/systm.h>
48 #include <sys/unistd.h>
49 #include <sys/vfsops.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 #include <vm/swap_pager.h>
57 
58 #include <vfs/fifofs/fifo.h>
59 #include <vfs/tmpfs/tmpfs_vnops.h>
60 #include <vfs/tmpfs/tmpfs.h>
61 
62 MALLOC_DECLARE(M_TMPFS);
63 
64 static __inline
65 void
66 tmpfs_knote(struct vnode *vp, int flags)
67 {
68 	if (flags)
69 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
70 }
71 
72 
73 /* --------------------------------------------------------------------- */
74 
75 static int
76 tmpfs_nresolve(struct vop_nresolve_args *v)
77 {
78 	struct vnode *dvp = v->a_dvp;
79 	struct vnode *vp = NULL;
80 	struct namecache *ncp = v->a_nch->ncp;
81 	struct tmpfs_node *tnode;
82 
83 	int error;
84 	struct tmpfs_dirent *de;
85 	struct tmpfs_node *dnode;
86 
87 	dnode = VP_TO_TMPFS_DIR(dvp);
88 
89 	de = tmpfs_dir_lookup(dnode, NULL, ncp);
90 	if (de == NULL) {
91 		error = ENOENT;
92 	} else {
93 		/*
94 		 * Allocate a vnode for the node we found.
95 		 */
96 		tnode = de->td_node;
97 		error = tmpfs_alloc_vp(dvp->v_mount, tnode,
98 				       LK_EXCLUSIVE | LK_RETRY, &vp);
99 		if (error)
100 			goto out;
101 		KKASSERT(vp);
102 	}
103 
104 out:
105 	/*
106 	 * Store the result of this lookup in the cache.  Avoid this if the
107 	 * request was for creation, as it does not improve timings on
108 	 * emprical tests.
109 	 */
110 	if (vp) {
111 		vn_unlock(vp);
112 		cache_setvp(v->a_nch, vp);
113 		vrele(vp);
114 	} else if (error == ENOENT) {
115 		cache_setvp(v->a_nch, NULL);
116 	}
117 	return error;
118 }
119 
120 static int
121 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
122 {
123 	struct vnode *dvp = v->a_dvp;
124 	struct vnode **vpp = v->a_vpp;
125 	struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
126 	struct ucred *cred = v->a_cred;
127 	int error;
128 
129 	*vpp = NULL;
130 	/* Check accessibility of requested node as a first step. */
131 	error = VOP_ACCESS(dvp, VEXEC, cred);
132 	if (error != 0)
133 		return error;
134 
135 	if (dnode->tn_dir.tn_parent != NULL) {
136 		/* Allocate a new vnode on the matching entry. */
137 		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
138 		    LK_EXCLUSIVE | LK_RETRY, vpp);
139 
140 		if (*vpp)
141 			vn_unlock(*vpp);
142 	}
143 
144 	return (*vpp == NULL) ? ENOENT : 0;
145 }
146 
147 /* --------------------------------------------------------------------- */
148 
149 static int
150 tmpfs_ncreate(struct vop_ncreate_args *v)
151 {
152 	struct vnode *dvp = v->a_dvp;
153 	struct vnode **vpp = v->a_vpp;
154 	struct namecache *ncp = v->a_nch->ncp;
155 	struct vattr *vap = v->a_vap;
156 	struct ucred *cred = v->a_cred;
157 	int error;
158 
159 	KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
160 
161 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
162 	if (error == 0) {
163 		cache_setunresolved(v->a_nch);
164 		cache_setvp(v->a_nch, *vpp);
165 		tmpfs_knote(dvp, NOTE_WRITE);
166 	}
167 
168 	return error;
169 }
170 /* --------------------------------------------------------------------- */
171 
172 static int
173 tmpfs_nmknod(struct vop_nmknod_args *v)
174 {
175 	struct vnode *dvp = v->a_dvp;
176 	struct vnode **vpp = v->a_vpp;
177 	struct namecache *ncp = v->a_nch->ncp;
178 	struct vattr *vap = v->a_vap;
179 	struct ucred *cred = v->a_cred;
180 	int error;
181 
182 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
183 	    vap->va_type != VFIFO)
184 		return EINVAL;
185 
186 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
187 	if (error == 0) {
188 		cache_setunresolved(v->a_nch);
189 		cache_setvp(v->a_nch, *vpp);
190 		tmpfs_knote(dvp, NOTE_WRITE);
191 	}
192 
193 	return error;
194 }
195 
196 /* --------------------------------------------------------------------- */
197 
198 static int
199 tmpfs_open(struct vop_open_args *v)
200 {
201 	struct vnode *vp = v->a_vp;
202 	int mode = v->a_mode;
203 
204 	int error;
205 	struct tmpfs_node *node;
206 
207 	node = VP_TO_TMPFS_NODE(vp);
208 
209 	/* The file is still active but all its names have been removed
210 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
211 	 * it is about to die. */
212 	if (node->tn_links < 1)
213 		return (ENOENT);
214 
215 	/* If the file is marked append-only, deny write requests. */
216 	if ((node->tn_flags & APPEND) &&
217 	    (mode & (FWRITE | O_APPEND)) == FWRITE) {
218 		error = EPERM;
219 	} else {
220 		return (vop_stdopen(v));
221 	}
222 	return error;
223 }
224 
225 /* --------------------------------------------------------------------- */
226 
227 static int
228 tmpfs_close(struct vop_close_args *v)
229 {
230 	struct vnode *vp = v->a_vp;
231 	struct tmpfs_node *node;
232 
233 	node = VP_TO_TMPFS_NODE(vp);
234 
235 	if (node->tn_links > 0) {
236 		/* Update node times.  No need to do it if the node has
237 		 * been deleted, because it will vanish after we return. */
238 		tmpfs_update(vp);
239 	}
240 
241 	return vop_stdclose(v);
242 }
243 
244 /* --------------------------------------------------------------------- */
245 
246 int
247 tmpfs_access(struct vop_access_args *v)
248 {
249 	struct vnode *vp = v->a_vp;
250 	int error;
251 	struct tmpfs_node *node;
252 
253 	node = VP_TO_TMPFS_NODE(vp);
254 
255 	switch (vp->v_type) {
256 	case VDIR:
257 		/* FALLTHROUGH */
258 	case VLNK:
259 		/* FALLTHROUGH */
260 	case VREG:
261 		if ((v->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
262 			error = EROFS;
263 			goto out;
264 		}
265 		break;
266 
267 	case VBLK:
268 		/* FALLTHROUGH */
269 	case VCHR:
270 		/* FALLTHROUGH */
271 	case VSOCK:
272 		/* FALLTHROUGH */
273 	case VFIFO:
274 		break;
275 
276 	default:
277 		error = EINVAL;
278 		goto out;
279 	}
280 
281 	if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
282 		error = EPERM;
283 		goto out;
284 	}
285 
286 	error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
287 
288 out:
289 
290 	return error;
291 }
292 
293 /* --------------------------------------------------------------------- */
294 
295 int
296 tmpfs_getattr(struct vop_getattr_args *v)
297 {
298 	struct vnode *vp = v->a_vp;
299 	struct vattr *vap = v->a_vap;
300 	struct tmpfs_node *node;
301 
302 	node = VP_TO_TMPFS_NODE(vp);
303 
304 	lwkt_gettoken(&vp->v_mount->mnt_token);
305 	tmpfs_update(vp);
306 
307 	vap->va_type = vp->v_type;
308 	vap->va_mode = node->tn_mode;
309 	vap->va_nlink = node->tn_links;
310 	vap->va_uid = node->tn_uid;
311 	vap->va_gid = node->tn_gid;
312 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
313 	vap->va_fileid = node->tn_id;
314 	vap->va_size = node->tn_size;
315 	vap->va_blocksize = PAGE_SIZE;
316 	vap->va_atime.tv_sec = node->tn_atime;
317 	vap->va_atime.tv_nsec = node->tn_atimensec;
318 	vap->va_mtime.tv_sec = node->tn_mtime;
319 	vap->va_mtime.tv_nsec = node->tn_mtimensec;
320 	vap->va_ctime.tv_sec = node->tn_ctime;
321 	vap->va_ctime.tv_nsec = node->tn_ctimensec;
322 	vap->va_gen = node->tn_gen;
323 	vap->va_flags = node->tn_flags;
324 	if (vp->v_type == VBLK || vp->v_type == VCHR)
325 	{
326 		vap->va_rmajor = umajor(node->tn_rdev);
327 		vap->va_rminor = uminor(node->tn_rdev);
328 	}
329 	vap->va_bytes = round_page(node->tn_size);
330 	vap->va_filerev = 0;
331 
332 	lwkt_reltoken(&vp->v_mount->mnt_token);
333 
334 	return 0;
335 }
336 
337 /* --------------------------------------------------------------------- */
338 
339 int
340 tmpfs_setattr(struct vop_setattr_args *v)
341 {
342 	struct vnode *vp = v->a_vp;
343 	struct vattr *vap = v->a_vap;
344 	struct ucred *cred = v->a_cred;
345 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
346 	int error = 0;
347 	int kflags = 0;
348 
349 	if (error == 0 && (vap->va_flags != VNOVAL)) {
350 		error = tmpfs_chflags(vp, vap->va_flags, cred);
351 		kflags |= NOTE_ATTRIB;
352 	}
353 
354 	if (error == 0 && (vap->va_size != VNOVAL)) {
355 		if (vap->va_size > node->tn_size)
356 			kflags |= NOTE_WRITE | NOTE_EXTEND;
357 		else
358 			kflags |= NOTE_WRITE;
359 		error = tmpfs_chsize(vp, vap->va_size, cred);
360 	}
361 
362 	if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
363 			   vap->va_gid != (gid_t)VNOVAL)) {
364 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
365 		kflags |= NOTE_ATTRIB;
366 	}
367 
368 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
369 		error = tmpfs_chmod(vp, vap->va_mode, cred);
370 		kflags |= NOTE_ATTRIB;
371 	}
372 
373 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
374 	    vap->va_atime.tv_nsec != VNOVAL) ||
375 	    (vap->va_mtime.tv_sec != VNOVAL &&
376 	    vap->va_mtime.tv_nsec != VNOVAL) )) {
377 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
378 				      vap->va_vaflags, cred);
379 		kflags |= NOTE_ATTRIB;
380 	}
381 
382 	/* Update the node times.  We give preference to the error codes
383 	 * generated by this function rather than the ones that may arise
384 	 * from tmpfs_update. */
385 	tmpfs_update(vp);
386 	tmpfs_knote(vp, kflags);
387 
388 	return error;
389 }
390 
391 /* --------------------------------------------------------------------- */
392 
393 /*
394  * fsync is usually a NOP, but we must take action when unmounting or
395  * when recycling.
396  */
397 static int
398 tmpfs_fsync(struct vop_fsync_args *v)
399 {
400 	struct tmpfs_mount *tmp;
401 	struct tmpfs_node *node;
402 	struct vnode *vp = v->a_vp;
403 
404 	tmp = VFS_TO_TMPFS(vp->v_mount);
405 	node = VP_TO_TMPFS_NODE(vp);
406 
407 	tmpfs_update(vp);
408 	if (vp->v_type == VREG) {
409 		if (vp->v_flag & VRECLAIMED) {
410 			if (node->tn_links == 0)
411 				tmpfs_truncate(vp, 0);
412 			else
413 				vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
414 		}
415 	}
416 	return 0;
417 }
418 
419 /* --------------------------------------------------------------------- */
420 
421 static int
422 tmpfs_read (struct vop_read_args *ap)
423 {
424 	struct buf *bp;
425 	struct vnode *vp = ap->a_vp;
426 	struct uio *uio = ap->a_uio;
427 	struct tmpfs_node *node;
428 	off_t base_offset;
429 	size_t offset;
430 	size_t len;
431 	int error;
432 
433 	error = 0;
434 	if (uio->uio_resid == 0) {
435 		return error;
436 	}
437 
438 	node = VP_TO_TMPFS_NODE(vp);
439 
440 	if (uio->uio_offset < 0)
441 		return (EINVAL);
442 	if (vp->v_type != VREG)
443 		return (EINVAL);
444 
445 	while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
446 		/*
447 		 * Use buffer cache I/O (via tmpfs_strategy)
448 		 */
449 		offset = (size_t)uio->uio_offset & BMASK;
450 		base_offset = (off_t)uio->uio_offset - offset;
451 		bp = getcacheblk(vp, base_offset, BSIZE);
452 		if (bp == NULL)
453 		{
454 			lwkt_gettoken(&vp->v_mount->mnt_token);
455 			error = bread(vp, base_offset, BSIZE, &bp);
456 			if (error) {
457 				brelse(bp);
458 				lwkt_reltoken(&vp->v_mount->mnt_token);
459 				kprintf("tmpfs_read bread error %d\n", error);
460 				break;
461 			}
462 			lwkt_reltoken(&vp->v_mount->mnt_token);
463 		}
464 
465 		/*
466 		 * Figure out how many bytes we can actually copy this loop.
467 		 */
468 		len = BSIZE - offset;
469 		if (len > uio->uio_resid)
470 			len = uio->uio_resid;
471 		if (len > node->tn_size - uio->uio_offset)
472 			len = (size_t)(node->tn_size - uio->uio_offset);
473 
474 		error = uiomove((char *)bp->b_data + offset, len, uio);
475 		bqrelse(bp);
476 		if (error) {
477 			kprintf("tmpfs_read uiomove error %d\n", error);
478 			break;
479 		}
480 	}
481 
482 	TMPFS_NODE_LOCK(node);
483 	node->tn_status |= TMPFS_NODE_ACCESSED;
484 	TMPFS_NODE_UNLOCK(node);
485 
486 	return(error);
487 }
488 
489 static int
490 tmpfs_write (struct vop_write_args *ap)
491 {
492 	struct buf *bp;
493 	struct vnode *vp = ap->a_vp;
494 	struct uio *uio = ap->a_uio;
495 	struct thread *td = uio->uio_td;
496 	struct tmpfs_node *node;
497 	boolean_t extended;
498 	off_t oldsize;
499 	int error;
500 	off_t base_offset;
501 	size_t offset;
502 	size_t len;
503 	struct rlimit limit;
504 	int trivial = 0;
505 	int kflags = 0;
506 
507 	error = 0;
508 	if (uio->uio_resid == 0) {
509 		return error;
510 	}
511 
512 	node = VP_TO_TMPFS_NODE(vp);
513 
514 	if (vp->v_type != VREG)
515 		return (EINVAL);
516 
517 	lwkt_gettoken(&vp->v_mount->mnt_token);
518 
519 	oldsize = node->tn_size;
520 	if (ap->a_ioflag & IO_APPEND)
521 		uio->uio_offset = node->tn_size;
522 
523 	/*
524 	 * Check for illegal write offsets.
525 	 */
526 	if (uio->uio_offset + uio->uio_resid >
527 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
528 		lwkt_reltoken(&vp->v_mount->mnt_token);
529 		return (EFBIG);
530 	}
531 
532 	if (vp->v_type == VREG && td != NULL) {
533 		error = kern_getrlimit(RLIMIT_FSIZE, &limit);
534 		if (error != 0) {
535 			lwkt_reltoken(&vp->v_mount->mnt_token);
536 			return error;
537 		}
538 		if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
539 			ksignal(td->td_proc, SIGXFSZ);
540 			lwkt_reltoken(&vp->v_mount->mnt_token);
541 			return (EFBIG);
542 		}
543 	}
544 
545 
546 	/*
547 	 * Extend the file's size if necessary
548 	 */
549 	extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
550 
551 	while (uio->uio_resid > 0) {
552 		/*
553 		 * Use buffer cache I/O (via tmpfs_strategy)
554 		 */
555 		offset = (size_t)uio->uio_offset & BMASK;
556 		base_offset = (off_t)uio->uio_offset - offset;
557 		len = BSIZE - offset;
558 		if (len > uio->uio_resid)
559 			len = uio->uio_resid;
560 
561 		if ((uio->uio_offset + len) > node->tn_size) {
562 			trivial = (uio->uio_offset <= node->tn_size);
563 			error = tmpfs_reg_resize(vp, uio->uio_offset + len,  trivial);
564 			if (error)
565 				break;
566 		}
567 
568 		/*
569 		 * Read to fill in any gaps.  Theoretically we could
570 		 * optimize this if the write covers the entire buffer
571 		 * and is not a UIO_NOCOPY write, however this can lead
572 		 * to a security violation exposing random kernel memory
573 		 * (whatever junk was in the backing VM pages before).
574 		 *
575 		 * So just use bread() to do the right thing.
576 		 */
577 		error = bread(vp, base_offset, BSIZE, &bp);
578 		error = uiomove((char *)bp->b_data + offset, len, uio);
579 		if (error) {
580 			kprintf("tmpfs_write uiomove error %d\n", error);
581 			brelse(bp);
582 			break;
583 		}
584 
585 		if (uio->uio_offset > node->tn_size) {
586 			node->tn_size = uio->uio_offset;
587 			kflags |= NOTE_EXTEND;
588 		}
589 		kflags |= NOTE_WRITE;
590 
591 		/*
592 		 * The data has been loaded into the buffer, write it out.
593 		 *
594 		 * We want tmpfs to be able to use all available ram, not
595 		 * just the buffer cache, so if not explicitly paging we
596 		 * use buwrite() to leave the buffer clean but mark all the
597 		 * VM pages valid+dirty.
598 		 *
599 		 * When the kernel is paging, either via normal pageout
600 		 * operation or when cleaning the object during a recycle,
601 		 * the underlying VM pages are going to get thrown away
602 		 * so we MUST write them to swap.
603 		 *
604 		 * XXX unfortunately this catches msync() system calls too
605 		 * for the moment.
606 		 */
607 		if (vm_swap_size == 0) {
608 			/*
609 			 * if swap isn't configured yet, force a buwrite() to
610 			 * avoid problems further down the line, due to flushing
611 			 * to swap.
612 			 */
613 			buwrite(bp);
614 		} else {
615 			if (ap->a_ioflag & IO_SYNC) {
616 				bwrite(bp);
617 			} else if ((ap->a_ioflag & IO_ASYNC) ||
618 				 (uio->uio_segflg == UIO_NOCOPY)) {
619 				bawrite(bp);
620 			} else {
621 				buwrite(bp);
622 			}
623 		}
624 
625 		if (bp->b_error) {
626 			kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
627 			break;
628 		}
629 	}
630 
631 	if (error) {
632 		if (extended) {
633 			(void)tmpfs_reg_resize(vp, oldsize, trivial);
634 			kflags &= ~NOTE_EXTEND;
635 		}
636 		goto done;
637 	}
638 
639 	TMPFS_NODE_LOCK(node);
640 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
641 	    (extended? TMPFS_NODE_CHANGED : 0);
642 
643 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
644 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
645 			node->tn_mode &= ~(S_ISUID | S_ISGID);
646 	}
647 	TMPFS_NODE_UNLOCK(node);
648 done:
649 
650 	tmpfs_knote(vp, kflags);
651 
652 
653 	lwkt_reltoken(&vp->v_mount->mnt_token);
654 	return(error);
655 }
656 
657 static int
658 tmpfs_advlock (struct vop_advlock_args *ap)
659 {
660 	struct tmpfs_node *node;
661 	struct vnode *vp = ap->a_vp;
662 
663 	node = VP_TO_TMPFS_NODE(vp);
664 
665 	return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
666 }
667 
668 static int
669 tmpfs_strategy(struct vop_strategy_args *ap)
670 {
671 	struct bio *bio = ap->a_bio;
672 	struct buf *bp = bio->bio_buf;
673 	struct vnode *vp = ap->a_vp;
674 	struct tmpfs_node *node;
675 	vm_object_t uobj;
676 
677 	if (vp->v_type != VREG) {
678 		bp->b_resid = bp->b_bcount;
679 		bp->b_flags |= B_ERROR | B_INVAL;
680 		bp->b_error = EINVAL;
681 		biodone(bio);
682 		return(0);
683 	}
684 
685 	lwkt_gettoken(&vp->v_mount->mnt_token);
686 	node = VP_TO_TMPFS_NODE(vp);
687 
688 	uobj = node->tn_reg.tn_aobj;
689 
690 	/*
691 	 * Call swap_pager_strategy to read or write between the VM
692 	 * object and the buffer cache.
693 	 */
694 	swap_pager_strategy(uobj, bio);
695 
696 	lwkt_reltoken(&vp->v_mount->mnt_token);
697 	return 0;
698 }
699 
700 static int
701 tmpfs_bmap(struct vop_bmap_args *ap)
702 {
703 	if (ap->a_doffsetp != NULL)
704 		*ap->a_doffsetp = ap->a_loffset;
705 	if (ap->a_runp != NULL)
706 		*ap->a_runp = 0;
707 	if (ap->a_runb != NULL)
708 		*ap->a_runb = 0;
709 
710 	return 0;
711 }
712 
713 /* --------------------------------------------------------------------- */
714 
715 static int
716 tmpfs_nremove(struct vop_nremove_args *v)
717 {
718 	struct vnode *dvp = v->a_dvp;
719 	struct namecache *ncp = v->a_nch->ncp;
720 	struct vnode *vp;
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 	/*
728 	 * We have to acquire the vp from v->a_nch because we will likely
729 	 * unresolve the namecache entry, and a vrele/vput is needed to
730 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
731 	 *
732 	 * We have to use vget to clear any inactive state on the vnode,
733 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
734 	 * will not get called when we release it.
735 	 */
736 	error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
737 	KKASSERT(error == 0);
738 	vn_unlock(vp);
739 
740 	if (vp->v_type == VDIR) {
741 		error = EISDIR;
742 		goto out;
743 	}
744 
745 	dnode = VP_TO_TMPFS_DIR(dvp);
746 	node = VP_TO_TMPFS_NODE(vp);
747 	tmp = VFS_TO_TMPFS(vp->v_mount);
748 	de = tmpfs_dir_lookup(dnode, node, ncp);
749 	if (de == NULL) {
750 		error = ENOENT;
751 		goto out;
752 	}
753 
754 	/* Files marked as immutable or append-only cannot be deleted. */
755 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
756 	    (dnode->tn_flags & APPEND)) {
757 		error = EPERM;
758 		goto out;
759 	}
760 
761 	/* Remove the entry from the directory; as it is a file, we do not
762 	 * have to change the number of hard links of the directory. */
763 	tmpfs_dir_detach(dnode, de);
764 
765 	/* Free the directory entry we just deleted.  Note that the node
766 	 * referred by it will not be removed until the vnode is really
767 	 * reclaimed. */
768 	tmpfs_free_dirent(tmp, de);
769 
770 	if (node->tn_links > 0) {
771 	        TMPFS_NODE_LOCK(node);
772 		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
773 	                TMPFS_NODE_MODIFIED;
774 	        TMPFS_NODE_UNLOCK(node);
775 	}
776 
777 	cache_setunresolved(v->a_nch);
778 	cache_setvp(v->a_nch, NULL);
779 	tmpfs_knote(vp, NOTE_DELETE);
780 	/*cache_inval_vp(vp, CINV_DESTROY);*/
781 	tmpfs_knote(dvp, NOTE_WRITE);
782 	error = 0;
783 
784 out:
785 	vrele(vp);
786 
787 	return error;
788 }
789 
790 /* --------------------------------------------------------------------- */
791 
792 static int
793 tmpfs_nlink(struct vop_nlink_args *v)
794 {
795 	struct vnode *dvp = v->a_dvp;
796 	struct vnode *vp = v->a_vp;
797 	struct namecache *ncp = v->a_nch->ncp;
798 	struct tmpfs_dirent *de;
799 	struct tmpfs_node *node;
800 	struct tmpfs_node *dnode;
801 	int error;
802 
803 	KKASSERT(dvp != vp); /* XXX When can this be false? */
804 
805 	node = VP_TO_TMPFS_NODE(vp);
806 	dnode = VP_TO_TMPFS_NODE(dvp);
807 
808 	/* XXX: Why aren't the following two tests done by the caller? */
809 
810 	/* Hard links of directories are forbidden. */
811 	if (vp->v_type == VDIR) {
812 		error = EPERM;
813 		goto out;
814 	}
815 
816 	/* Cannot create cross-device links. */
817 	if (dvp->v_mount != vp->v_mount) {
818 		error = EXDEV;
819 		goto out;
820 	}
821 
822 	/* Ensure that we do not overflow the maximum number of links imposed
823 	 * by the system. */
824 	KKASSERT(node->tn_links <= LINK_MAX);
825 	if (node->tn_links == LINK_MAX) {
826 		error = EMLINK;
827 		goto out;
828 	}
829 
830 	/* We cannot create links of files marked immutable or append-only. */
831 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
832 		error = EPERM;
833 		goto out;
834 	}
835 
836 	/* Allocate a new directory entry to represent the node. */
837 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
838 	    ncp->nc_name, ncp->nc_nlen, &de);
839 	if (error != 0)
840 		goto out;
841 
842 	/* Insert the new directory entry into the appropriate directory. */
843 	tmpfs_dir_attach(dnode, de);
844 
845 	/* vp link count has changed, so update node times. */
846 
847 	TMPFS_NODE_LOCK(node);
848 	node->tn_status |= TMPFS_NODE_CHANGED;
849 	TMPFS_NODE_UNLOCK(node);
850 	tmpfs_update(vp);
851 
852 	tmpfs_knote(vp, NOTE_LINK);
853 	cache_setunresolved(v->a_nch);
854 	cache_setvp(v->a_nch, vp);
855 	tmpfs_knote(dvp, NOTE_WRITE);
856 	error = 0;
857 
858 out:
859 	return error;
860 }
861 
862 /* --------------------------------------------------------------------- */
863 
864 static int
865 tmpfs_nrename(struct vop_nrename_args *v)
866 {
867 	struct vnode *fdvp = v->a_fdvp;
868 	struct namecache *fncp = v->a_fnch->ncp;
869 	struct vnode *fvp = fncp->nc_vp;
870 	struct vnode *tdvp = v->a_tdvp;
871 	struct namecache *tncp = v->a_tnch->ncp;
872 	struct vnode *tvp;
873 	struct tmpfs_dirent *de;
874 	struct tmpfs_mount *tmp;
875 	struct tmpfs_node *fdnode;
876 	struct tmpfs_node *fnode;
877 	struct tmpfs_node *tnode;
878 	struct tmpfs_node *tdnode;
879 	char *newname;
880 	char *oldname;
881 	int error;
882 
883 	/*
884 	 * Because tvp can get overwritten we have to vget it instead of
885 	 * just vref or use it, otherwise it's VINACTIVE flag may not get
886 	 * cleared and the node won't get destroyed.
887 	 */
888 	error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp);
889 	if (error == 0) {
890 		tnode = VP_TO_TMPFS_NODE(tvp);
891 		vn_unlock(tvp);
892 	} else {
893 		tnode = NULL;
894 	}
895 
896 	/* Disallow cross-device renames.
897 	 * XXX Why isn't this done by the caller? */
898 	if (fvp->v_mount != tdvp->v_mount ||
899 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
900 		error = EXDEV;
901 		goto out;
902 	}
903 
904 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
905 	tdnode = VP_TO_TMPFS_DIR(tdvp);
906 
907 	/* If source and target are the same file, there is nothing to do. */
908 	if (fvp == tvp) {
909 		error = 0;
910 		goto out;
911 	}
912 
913 	fdnode = VP_TO_TMPFS_DIR(fdvp);
914 	fnode = VP_TO_TMPFS_NODE(fvp);
915 	de = tmpfs_dir_lookup(fdnode, fnode, fncp);
916 
917 	/* Avoid manipulating '.' and '..' entries. */
918 	if (de == NULL) {
919 		error = ENOENT;
920 		goto out_locked;
921 	}
922 	KKASSERT(de->td_node == fnode);
923 
924 	/*
925 	 * If replacing an entry in the target directory and that entry
926 	 * is a directory, it must be empty.
927 	 *
928 	 * Kern_rename gurantees the destination to be a directory
929 	 * if the source is one (it does?).
930 	 */
931 	if (tvp != NULL) {
932 		KKASSERT(tnode != NULL);
933 
934 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
935 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
936 			error = EPERM;
937 			goto out_locked;
938 		}
939 
940 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
941 			if (tnode->tn_size > 0) {
942 				error = ENOTEMPTY;
943 				goto out_locked;
944 			}
945 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
946 			error = ENOTDIR;
947 			goto out_locked;
948 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
949 			error = EISDIR;
950 			goto out_locked;
951 		} else {
952 			KKASSERT(fnode->tn_type != VDIR &&
953 				tnode->tn_type != VDIR);
954 		}
955 	}
956 
957 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
958 	    (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
959 		error = EPERM;
960 		goto out_locked;
961 	}
962 
963 	/*
964 	 * Ensure that we have enough memory to hold the new name, if it
965 	 * has to be changed.
966 	 */
967 	if (fncp->nc_nlen != tncp->nc_nlen ||
968 	    bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
969 		newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
970 				  M_WAITOK | M_NULLOK);
971 		if (newname == NULL) {
972 			error = ENOSPC;
973 			goto out_locked;
974 		}
975 		bcopy(tncp->nc_name, newname, tncp->nc_nlen);
976 		newname[tncp->nc_nlen] = '\0';
977 	} else {
978 		newname = NULL;
979 	}
980 
981 	/*
982 	 * Unlink entry from source directory.  Note that the kernel has
983 	 * already checked for illegal recursion cases (renaming a directory
984 	 * into a subdirectory of itself).
985 	 */
986 	if (fdnode != tdnode)
987 		tmpfs_dir_detach(fdnode, de);
988 
989 	/*
990 	 * Handle any name change.  Swap with newname, we will
991 	 * deallocate it at the end.
992 	 */
993 	if (newname != NULL) {
994 #if 0
995 		TMPFS_NODE_LOCK(fnode);
996 		fnode->tn_status |= TMPFS_NODE_CHANGED;
997 		TMPFS_NODE_UNLOCK(fnode);
998 #endif
999 		oldname = de->td_name;
1000 		de->td_name = newname;
1001 		de->td_namelen = (uint16_t)tncp->nc_nlen;
1002 		newname = oldname;
1003 	}
1004 
1005 	/*
1006 	 * Link entry to target directory.  If the entry
1007 	 * represents a directory move the parent linkage
1008 	 * as well.
1009 	 */
1010 	if (fdnode != tdnode) {
1011 		if (de->td_node->tn_type == VDIR) {
1012 			TMPFS_VALIDATE_DIR(fnode);
1013 
1014 			TMPFS_NODE_LOCK(tdnode);
1015 			tdnode->tn_links++;
1016 			tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1017 			TMPFS_NODE_UNLOCK(tdnode);
1018 
1019 			TMPFS_NODE_LOCK(fnode);
1020 			fnode->tn_dir.tn_parent = tdnode;
1021 			fnode->tn_status |= TMPFS_NODE_CHANGED;
1022 			TMPFS_NODE_UNLOCK(fnode);
1023 
1024 			TMPFS_NODE_LOCK(fdnode);
1025 			fdnode->tn_links--;
1026 			fdnode->tn_status |= TMPFS_NODE_MODIFIED;
1027 			TMPFS_NODE_UNLOCK(fdnode);
1028 		}
1029 		tmpfs_dir_attach(tdnode, de);
1030 	} else {
1031 		TMPFS_NODE_LOCK(tdnode);
1032 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1033 		TMPFS_NODE_UNLOCK(tdnode);
1034 	}
1035 
1036 	/*
1037 	 * If we are overwriting an entry, we have to remove the old one
1038 	 * from the target directory.
1039 	 */
1040 	if (tvp != NULL) {
1041 		/* Remove the old entry from the target directory. */
1042 		de = tmpfs_dir_lookup(tdnode, tnode, tncp);
1043 		tmpfs_dir_detach(tdnode, de);
1044 		tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1045 
1046 		/*
1047 		 * Free the directory entry we just deleted.  Note that the
1048 		 * node referred by it will not be removed until the vnode is
1049 		 * really reclaimed.
1050 		 */
1051 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de);
1052 		/*cache_inval_vp(tvp, CINV_DESTROY);*/
1053 	}
1054 
1055 	/*
1056 	 * Finish up
1057 	 */
1058 	if (newname) {
1059 		kfree(newname, tmp->tm_name_zone);
1060 		newname = NULL;
1061 	}
1062 	cache_rename(v->a_fnch, v->a_tnch);
1063 	tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1064 	tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1065 	if (fnode->tn_vnode)
1066 		tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1067 	error = 0;
1068 
1069 out_locked:
1070 	;
1071 
1072 out:
1073 	if (tvp)
1074 		vrele(tvp);
1075 
1076 	return error;
1077 }
1078 
1079 /* --------------------------------------------------------------------- */
1080 
1081 static int
1082 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1083 {
1084 	struct vnode *dvp = v->a_dvp;
1085 	struct vnode **vpp = v->a_vpp;
1086 	struct namecache *ncp = v->a_nch->ncp;
1087 	struct vattr *vap = v->a_vap;
1088 	struct ucred *cred = v->a_cred;
1089 	int error;
1090 
1091 	KKASSERT(vap->va_type == VDIR);
1092 
1093 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1094 	if (error == 0) {
1095 		cache_setunresolved(v->a_nch);
1096 		cache_setvp(v->a_nch, *vpp);
1097 		tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1098 	}
1099 
1100 	return error;
1101 }
1102 
1103 /* --------------------------------------------------------------------- */
1104 
1105 static int
1106 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1107 {
1108 	struct vnode *dvp = v->a_dvp;
1109 	struct namecache *ncp = v->a_nch->ncp;
1110 	struct vnode *vp;
1111 	struct tmpfs_dirent *de;
1112 	struct tmpfs_mount *tmp;
1113 	struct tmpfs_node *dnode;
1114 	struct tmpfs_node *node;
1115 	int error;
1116 
1117 	/*
1118 	 * We have to acquire the vp from v->a_nch because we will likely
1119 	 * unresolve the namecache entry, and a vrele/vput is needed to
1120 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1121 	 *
1122 	 * We have to use vget to clear any inactive state on the vnode,
1123 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1124 	 * will not get called when we release it.
1125 	 */
1126 	error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
1127 	KKASSERT(error == 0);
1128 	vn_unlock(vp);
1129 
1130 	/*
1131 	 * Prevalidate so we don't hit an assertion later
1132 	 */
1133 	if (vp->v_type != VDIR) {
1134 		error = ENOTDIR;
1135 		goto out;
1136 	}
1137 
1138 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1139 	dnode = VP_TO_TMPFS_DIR(dvp);
1140 	node = VP_TO_TMPFS_DIR(vp);
1141 
1142 	/* Directories with more than two entries ('.' and '..') cannot be
1143 	 * removed. */
1144 	 if (node->tn_size > 0) {
1145 		 error = ENOTEMPTY;
1146 		 goto out;
1147 	 }
1148 
1149 	if ((dnode->tn_flags & APPEND)
1150 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1151 		error = EPERM;
1152 		goto out;
1153 	}
1154 
1155 	/* This invariant holds only if we are not trying to remove "..".
1156 	  * We checked for that above so this is safe now. */
1157 	KKASSERT(node->tn_dir.tn_parent == dnode);
1158 
1159 	/* Get the directory entry associated with node (vp).  This was
1160 	 * filled by tmpfs_lookup while looking up the entry. */
1161 	de = tmpfs_dir_lookup(dnode, node, ncp);
1162 	KKASSERT(TMPFS_DIRENT_MATCHES(de,
1163 	    ncp->nc_name,
1164 	    ncp->nc_nlen));
1165 
1166 	/* Check flags to see if we are allowed to remove the directory. */
1167 	if ((dnode->tn_flags & APPEND) ||
1168 	    node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1169 		error = EPERM;
1170 		goto out;
1171 	}
1172 
1173 
1174 	/* Detach the directory entry from the directory (dnode). */
1175 	tmpfs_dir_detach(dnode, de);
1176 
1177 	/* No vnode should be allocated for this entry from this point */
1178 	TMPFS_NODE_LOCK(node);
1179 	TMPFS_ASSERT_ELOCKED(node);
1180 	TMPFS_NODE_LOCK(dnode);
1181 	TMPFS_ASSERT_ELOCKED(dnode);
1182 
1183 #if 0
1184 	/* handled by tmpfs_free_node */
1185 	KKASSERT(node->tn_links > 0);
1186 	node->tn_links--;
1187 	node->tn_dir.tn_parent = NULL;
1188 #endif
1189 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1190 	    TMPFS_NODE_MODIFIED;
1191 
1192 #if 0
1193 	/* handled by tmpfs_free_node */
1194 	KKASSERT(dnode->tn_links > 0);
1195 	dnode->tn_links--;
1196 #endif
1197 	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1198 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1199 
1200 	TMPFS_NODE_UNLOCK(dnode);
1201 	TMPFS_NODE_UNLOCK(node);
1202 
1203 	/* Free the directory entry we just deleted.  Note that the node
1204 	 * referred by it will not be removed until the vnode is really
1205 	 * reclaimed. */
1206 	tmpfs_free_dirent(tmp, de);
1207 
1208 	/* Release the deleted vnode (will destroy the node, notify
1209 	 * interested parties and clean it from the cache). */
1210 
1211 	TMPFS_NODE_LOCK(dnode);
1212 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1213 	TMPFS_NODE_UNLOCK(dnode);
1214 	tmpfs_update(dvp);
1215 
1216 	cache_setunresolved(v->a_nch);
1217 	cache_setvp(v->a_nch, NULL);
1218 	/*cache_inval_vp(vp, CINV_DESTROY);*/
1219 	tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1220 	error = 0;
1221 
1222 out:
1223 	vrele(vp);
1224 
1225 	return error;
1226 }
1227 
1228 /* --------------------------------------------------------------------- */
1229 
1230 static int
1231 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1232 {
1233 	struct vnode *dvp = v->a_dvp;
1234 	struct vnode **vpp = v->a_vpp;
1235 	struct namecache *ncp = v->a_nch->ncp;
1236 	struct vattr *vap = v->a_vap;
1237 	struct ucred *cred = v->a_cred;
1238 	char *target = v->a_target;
1239 	int error;
1240 
1241 	vap->va_type = VLNK;
1242 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1243 	if (error == 0) {
1244 		tmpfs_knote(*vpp, NOTE_WRITE);
1245 		cache_setunresolved(v->a_nch);
1246 		cache_setvp(v->a_nch, *vpp);
1247 	}
1248 
1249 	return error;
1250 }
1251 
1252 /* --------------------------------------------------------------------- */
1253 
1254 static int
1255 tmpfs_readdir(struct vop_readdir_args *v)
1256 {
1257 	struct vnode *vp = v->a_vp;
1258 	struct uio *uio = v->a_uio;
1259 	int *eofflag = v->a_eofflag;
1260 	off_t **cookies = v->a_cookies;
1261 	int *ncookies = v->a_ncookies;
1262 	struct tmpfs_mount *tmp;
1263 	int error;
1264 	off_t startoff;
1265 	off_t cnt = 0;
1266 	struct tmpfs_node *node;
1267 
1268 	/* This operation only makes sense on directory nodes. */
1269 	if (vp->v_type != VDIR)
1270 		return ENOTDIR;
1271 
1272 	tmp = VFS_TO_TMPFS(vp->v_mount);
1273 	node = VP_TO_TMPFS_DIR(vp);
1274 	startoff = uio->uio_offset;
1275 
1276 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1277 		error = tmpfs_dir_getdotdent(node, uio);
1278 		if (error != 0)
1279 			goto outok;
1280 		cnt++;
1281 	}
1282 
1283 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1284 		error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1285 		if (error != 0)
1286 			goto outok;
1287 		cnt++;
1288 	}
1289 
1290 	error = tmpfs_dir_getdents(node, uio, &cnt);
1291 
1292 outok:
1293 	KKASSERT(error >= -1);
1294 
1295 	if (error == -1)
1296 		error = 0;
1297 
1298 	if (eofflag != NULL)
1299 		*eofflag =
1300 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1301 
1302 	/* Update NFS-related variables. */
1303 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1304 		off_t i;
1305 		off_t off = startoff;
1306 		struct tmpfs_dirent *de = NULL;
1307 
1308 		*ncookies = cnt;
1309 		*cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1310 
1311 		for (i = 0; i < cnt; i++) {
1312 			KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1313 			if (off == TMPFS_DIRCOOKIE_DOT) {
1314 				off = TMPFS_DIRCOOKIE_DOTDOT;
1315 			} else {
1316 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1317 					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1318 				} else if (de != NULL) {
1319 					de = TAILQ_NEXT(de, td_entries);
1320 				} else {
1321 					de = tmpfs_dir_lookupbycookie(node,
1322 					    off);
1323 					KKASSERT(de != NULL);
1324 					de = TAILQ_NEXT(de, td_entries);
1325 				}
1326 				if (de == NULL)
1327 					off = TMPFS_DIRCOOKIE_EOF;
1328 				else
1329 					off = tmpfs_dircookie(de);
1330 			}
1331 
1332 			(*cookies)[i] = off;
1333 		}
1334 		KKASSERT(uio->uio_offset == off);
1335 	}
1336 
1337 	return error;
1338 }
1339 
1340 /* --------------------------------------------------------------------- */
1341 
1342 static int
1343 tmpfs_readlink(struct vop_readlink_args *v)
1344 {
1345 	struct vnode *vp = v->a_vp;
1346 	struct uio *uio = v->a_uio;
1347 
1348 	int error;
1349 	struct tmpfs_node *node;
1350 
1351 	KKASSERT(uio->uio_offset == 0);
1352 	KKASSERT(vp->v_type == VLNK);
1353 
1354 	node = VP_TO_TMPFS_NODE(vp);
1355 
1356 	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1357 	    uio);
1358 	TMPFS_NODE_LOCK(node);
1359 	node->tn_status |= TMPFS_NODE_ACCESSED;
1360 	TMPFS_NODE_UNLOCK(node);
1361 
1362 	return error;
1363 }
1364 
1365 /* --------------------------------------------------------------------- */
1366 
1367 static int
1368 tmpfs_inactive(struct vop_inactive_args *v)
1369 {
1370 	struct vnode *vp = v->a_vp;
1371 	struct tmpfs_node *node;
1372 
1373 	node = VP_TO_TMPFS_NODE(vp);
1374 
1375 	/*
1376 	 * Degenerate case
1377 	 */
1378 	if (node == NULL) {
1379 		vrecycle(vp);
1380 		return(0);
1381 	}
1382 
1383 	/*
1384 	 * Get rid of unreferenced deleted vnodes sooner rather than
1385 	 * later so the data memory can be recovered immediately.
1386 	 *
1387 	 * We must truncate the vnode to prevent the normal reclamation
1388 	 * path from flushing the data for the removed file to disk.
1389 	 */
1390 	TMPFS_NODE_LOCK(node);
1391 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1392 	    (node->tn_links == 0 ||
1393 	     (node->tn_links == 1 && node->tn_type == VDIR &&
1394 	      node->tn_dir.tn_parent)))
1395 	{
1396 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1397 		TMPFS_NODE_UNLOCK(node);
1398 		if (node->tn_type == VREG)
1399 			tmpfs_truncate(vp, 0);
1400 		vrecycle(vp);
1401 	} else {
1402 		TMPFS_NODE_UNLOCK(node);
1403 	}
1404 
1405 	return 0;
1406 }
1407 
1408 /* --------------------------------------------------------------------- */
1409 
1410 int
1411 tmpfs_reclaim(struct vop_reclaim_args *v)
1412 {
1413 	struct vnode *vp = v->a_vp;
1414 	struct tmpfs_mount *tmp;
1415 	struct tmpfs_node *node;
1416 
1417 	node = VP_TO_TMPFS_NODE(vp);
1418 	tmp = VFS_TO_TMPFS(vp->v_mount);
1419 
1420 	tmpfs_free_vp(vp);
1421 
1422 	/*
1423 	 * If the node referenced by this vnode was deleted by the
1424 	 * user, we must free its associated data structures now that
1425 	 * the vnode is being reclaimed.
1426 	 *
1427 	 * Directories have an extra link ref.
1428 	 */
1429 	TMPFS_NODE_LOCK(node);
1430 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1431 	    (node->tn_links == 0 ||
1432 	     (node->tn_links == 1 && node->tn_type == VDIR &&
1433 	      node->tn_dir.tn_parent)))
1434 	{
1435 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1436 		tmpfs_free_node(tmp, node);
1437 		/* eats the lock */
1438 	} else {
1439 		TMPFS_NODE_UNLOCK(node);
1440 	}
1441 
1442 	KKASSERT(vp->v_data == NULL);
1443 	return 0;
1444 }
1445 
1446 /* --------------------------------------------------------------------- */
1447 
1448 static int
1449 tmpfs_print(struct vop_print_args *v)
1450 {
1451 	struct vnode *vp = v->a_vp;
1452 
1453 	struct tmpfs_node *node;
1454 
1455 	node = VP_TO_TMPFS_NODE(vp);
1456 
1457 	kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1458 	    node, node->tn_flags, node->tn_links);
1459 	kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1460 	    node->tn_mode, node->tn_uid, node->tn_gid,
1461 	    (uintmax_t)node->tn_size, node->tn_status);
1462 
1463 	if (vp->v_type == VFIFO)
1464 		fifo_printinfo(vp);
1465 
1466 	kprintf("\n");
1467 
1468 	return 0;
1469 }
1470 
1471 /* --------------------------------------------------------------------- */
1472 
1473 static int
1474 tmpfs_pathconf(struct vop_pathconf_args *v)
1475 {
1476 	int name = v->a_name;
1477 	register_t *retval = v->a_retval;
1478 
1479 	int error;
1480 
1481 	error = 0;
1482 
1483 	switch (name) {
1484 	case _PC_LINK_MAX:
1485 		*retval = LINK_MAX;
1486 		break;
1487 
1488 	case _PC_NAME_MAX:
1489 		*retval = NAME_MAX;
1490 		break;
1491 
1492 	case _PC_PATH_MAX:
1493 		*retval = PATH_MAX;
1494 		break;
1495 
1496 	case _PC_PIPE_BUF:
1497 		*retval = PIPE_BUF;
1498 		break;
1499 
1500 	case _PC_CHOWN_RESTRICTED:
1501 		*retval = 1;
1502 		break;
1503 
1504 	case _PC_NO_TRUNC:
1505 		*retval = 1;
1506 		break;
1507 
1508 	case _PC_SYNC_IO:
1509 		*retval = 1;
1510 		break;
1511 
1512 	case _PC_FILESIZEBITS:
1513 		*retval = 0; /* XXX Don't know which value should I return. */
1514 		break;
1515 
1516 	default:
1517 		error = EINVAL;
1518 	}
1519 
1520 	return error;
1521 }
1522 
1523 /************************************************************************
1524  *                          KQFILTER OPS                                *
1525  ************************************************************************/
1526 
1527 static void filt_tmpfsdetach(struct knote *kn);
1528 static int filt_tmpfsread(struct knote *kn, long hint);
1529 static int filt_tmpfswrite(struct knote *kn, long hint);
1530 static int filt_tmpfsvnode(struct knote *kn, long hint);
1531 
1532 static struct filterops tmpfsread_filtops =
1533 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1534 static struct filterops tmpfswrite_filtops =
1535 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1536 static struct filterops tmpfsvnode_filtops =
1537 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1538 
1539 static int
1540 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1541 {
1542 	struct vnode *vp = ap->a_vp;
1543 	struct knote *kn = ap->a_kn;
1544 
1545 	switch (kn->kn_filter) {
1546 	case EVFILT_READ:
1547 		kn->kn_fop = &tmpfsread_filtops;
1548 		break;
1549 	case EVFILT_WRITE:
1550 		kn->kn_fop = &tmpfswrite_filtops;
1551 		break;
1552 	case EVFILT_VNODE:
1553 		kn->kn_fop = &tmpfsvnode_filtops;
1554 		break;
1555 	default:
1556 		return (EOPNOTSUPP);
1557 	}
1558 
1559 	kn->kn_hook = (caddr_t)vp;
1560 
1561 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1562 
1563 	return(0);
1564 }
1565 
1566 static void
1567 filt_tmpfsdetach(struct knote *kn)
1568 {
1569 	struct vnode *vp = (void *)kn->kn_hook;
1570 
1571 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1572 }
1573 
1574 static int
1575 filt_tmpfsread(struct knote *kn, long hint)
1576 {
1577 	struct vnode *vp = (void *)kn->kn_hook;
1578 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1579 	off_t off;
1580 
1581 	if (hint == NOTE_REVOKE) {
1582 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1583 		return(1);
1584 	}
1585 
1586 	/*
1587 	 * Interlock against MP races when performing this function.
1588 	 */
1589 	lwkt_gettoken(&vp->v_mount->mnt_token);
1590 	off = node->tn_size - kn->kn_fp->f_offset;
1591 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1592 	if (kn->kn_sfflags & NOTE_OLDAPI) {
1593 		lwkt_reltoken(&vp->v_mount->mnt_token);
1594 		return(1);
1595 	}
1596 
1597 	if (kn->kn_data == 0) {
1598 		kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1599 	}
1600 	lwkt_reltoken(&vp->v_mount->mnt_token);
1601 	return (kn->kn_data != 0);
1602 }
1603 
1604 static int
1605 filt_tmpfswrite(struct knote *kn, long hint)
1606 {
1607 	if (hint == NOTE_REVOKE)
1608 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1609 	kn->kn_data = 0;
1610 	return (1);
1611 }
1612 
1613 static int
1614 filt_tmpfsvnode(struct knote *kn, long hint)
1615 {
1616 	if (kn->kn_sfflags & hint)
1617 		kn->kn_fflags |= hint;
1618 	if (hint == NOTE_REVOKE) {
1619 		kn->kn_flags |= (EV_EOF | EV_NODATA);
1620 		return (1);
1621 	}
1622 	return (kn->kn_fflags != 0);
1623 }
1624 
1625 
1626 /* --------------------------------------------------------------------- */
1627 
1628 /*
1629  * vnode operations vector used for files stored in a tmpfs file system.
1630  */
1631 struct vop_ops tmpfs_vnode_vops = {
1632 	.vop_default =			vop_defaultop,
1633 	.vop_getpages = 		vop_stdgetpages,
1634 	.vop_putpages = 		vop_stdputpages,
1635 	.vop_ncreate =			tmpfs_ncreate,
1636 	.vop_nresolve =			tmpfs_nresolve,
1637 	.vop_nlookupdotdot =		tmpfs_nlookupdotdot,
1638 	.vop_nmknod =			tmpfs_nmknod,
1639 	.vop_open =			tmpfs_open,
1640 	.vop_close =			tmpfs_close,
1641 	.vop_access =			tmpfs_access,
1642 	.vop_getattr =			tmpfs_getattr,
1643 	.vop_setattr =			tmpfs_setattr,
1644 	.vop_read =			tmpfs_read,
1645 	.vop_write =			tmpfs_write,
1646 	.vop_fsync =			tmpfs_fsync,
1647 	.vop_nremove =			tmpfs_nremove,
1648 	.vop_nlink =			tmpfs_nlink,
1649 	.vop_nrename =			tmpfs_nrename,
1650 	.vop_nmkdir =			tmpfs_nmkdir,
1651 	.vop_nrmdir =			tmpfs_nrmdir,
1652 	.vop_nsymlink =			tmpfs_nsymlink,
1653 	.vop_readdir =			tmpfs_readdir,
1654 	.vop_readlink =			tmpfs_readlink,
1655 	.vop_inactive =			tmpfs_inactive,
1656 	.vop_reclaim =			tmpfs_reclaim,
1657 	.vop_print =			tmpfs_print,
1658 	.vop_pathconf =			tmpfs_pathconf,
1659 	.vop_bmap =			tmpfs_bmap,
1660 	.vop_strategy =			tmpfs_strategy,
1661 	.vop_advlock =			tmpfs_advlock,
1662 	.vop_kqfilter =			tmpfs_kqfilter
1663 };
1664