xref: /dflybsd-src/sys/vfs/tmpfs/tmpfs_vnops.c (revision cfc9634056ea0cd181df744a3d243a686e85bab3)
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 #include <sys/mountctl.h>
52 
53 #include <vm/vm.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pager.h>
57 #include <vm/swap_pager.h>
58 
59 #include <sys/buf2.h>
60 
61 #include <vfs/fifofs/fifo.h>
62 #include <vfs/tmpfs/tmpfs_vnops.h>
63 #include <vfs/tmpfs/tmpfs.h>
64 
65 MALLOC_DECLARE(M_TMPFS);
66 
67 static void tmpfs_strategy_done(struct bio *bio);
68 
69 static __inline
70 void
71 tmpfs_knote(struct vnode *vp, int flags)
72 {
73 	if (flags)
74 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
75 }
76 
77 
78 /* --------------------------------------------------------------------- */
79 
80 static int
81 tmpfs_nresolve(struct vop_nresolve_args *v)
82 {
83 	struct vnode *dvp = v->a_dvp;
84 	struct vnode *vp = NULL;
85 	struct namecache *ncp = v->a_nch->ncp;
86 	struct tmpfs_node *tnode;
87 
88 	int error;
89 	struct tmpfs_dirent *de;
90 	struct tmpfs_node *dnode;
91 
92 	dnode = VP_TO_TMPFS_DIR(dvp);
93 
94 	de = tmpfs_dir_lookup(dnode, NULL, ncp);
95 	if (de == NULL) {
96 		error = ENOENT;
97 	} else {
98 		/*
99 		 * Allocate a vnode for the node we found.
100 		 */
101 		tnode = de->td_node;
102 		error = tmpfs_alloc_vp(dvp->v_mount, tnode,
103 				       LK_EXCLUSIVE | LK_RETRY, &vp);
104 		if (error)
105 			goto out;
106 		KKASSERT(vp);
107 	}
108 
109 out:
110 	/*
111 	 * Store the result of this lookup in the cache.  Avoid this if the
112 	 * request was for creation, as it does not improve timings on
113 	 * emprical tests.
114 	 */
115 	if (vp) {
116 		vn_unlock(vp);
117 		cache_setvp(v->a_nch, vp);
118 		vrele(vp);
119 	} else if (error == ENOENT) {
120 		cache_setvp(v->a_nch, NULL);
121 	}
122 	return error;
123 }
124 
125 static int
126 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
127 {
128 	struct vnode *dvp = v->a_dvp;
129 	struct vnode **vpp = v->a_vpp;
130 	struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
131 	struct ucred *cred = v->a_cred;
132 	int error;
133 
134 	*vpp = NULL;
135 	/* Check accessibility of requested node as a first step. */
136 	error = VOP_ACCESS(dvp, VEXEC, cred);
137 	if (error != 0)
138 		return error;
139 
140 	if (dnode->tn_dir.tn_parent != NULL) {
141 		/* Allocate a new vnode on the matching entry. */
142 		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
143 		    LK_EXCLUSIVE | LK_RETRY, vpp);
144 
145 		if (*vpp)
146 			vn_unlock(*vpp);
147 	}
148 
149 	return (*vpp == NULL) ? ENOENT : 0;
150 }
151 
152 /* --------------------------------------------------------------------- */
153 
154 static int
155 tmpfs_ncreate(struct vop_ncreate_args *v)
156 {
157 	struct vnode *dvp = v->a_dvp;
158 	struct vnode **vpp = v->a_vpp;
159 	struct namecache *ncp = v->a_nch->ncp;
160 	struct vattr *vap = v->a_vap;
161 	struct ucred *cred = v->a_cred;
162 	int error;
163 
164 	KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
165 
166 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
167 	if (error == 0) {
168 		cache_setunresolved(v->a_nch);
169 		cache_setvp(v->a_nch, *vpp);
170 		tmpfs_knote(dvp, NOTE_WRITE);
171 	}
172 
173 	return error;
174 }
175 /* --------------------------------------------------------------------- */
176 
177 static int
178 tmpfs_nmknod(struct vop_nmknod_args *v)
179 {
180 	struct vnode *dvp = v->a_dvp;
181 	struct vnode **vpp = v->a_vpp;
182 	struct namecache *ncp = v->a_nch->ncp;
183 	struct vattr *vap = v->a_vap;
184 	struct ucred *cred = v->a_cred;
185 	int error;
186 
187 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
188 	    vap->va_type != VFIFO)
189 		return EINVAL;
190 
191 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
192 	if (error == 0) {
193 		cache_setunresolved(v->a_nch);
194 		cache_setvp(v->a_nch, *vpp);
195 		tmpfs_knote(dvp, NOTE_WRITE);
196 	}
197 
198 	return error;
199 }
200 
201 /* --------------------------------------------------------------------- */
202 
203 static int
204 tmpfs_open(struct vop_open_args *v)
205 {
206 	struct vnode *vp = v->a_vp;
207 	int mode = v->a_mode;
208 
209 	int error;
210 	struct tmpfs_node *node;
211 
212 	node = VP_TO_TMPFS_NODE(vp);
213 
214 #if 0
215 	/* The file is still active but all its names have been removed
216 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
217 	 * it is about to die. */
218 	if (node->tn_links < 1)
219 		return (ENOENT);
220 #endif
221 
222 	/* If the file is marked append-only, deny write requests. */
223 	if ((node->tn_flags & APPEND) &&
224 	    (mode & (FWRITE | O_APPEND)) == FWRITE) {
225 		error = EPERM;
226 	} else {
227 		return (vop_stdopen(v));
228 	}
229 	return error;
230 }
231 
232 /* --------------------------------------------------------------------- */
233 
234 static int
235 tmpfs_close(struct vop_close_args *v)
236 {
237 	struct vnode *vp = v->a_vp;
238 	struct tmpfs_node *node;
239 
240 	node = VP_TO_TMPFS_NODE(vp);
241 
242 	if (node->tn_links > 0) {
243 		/*
244 		 * Update node times.  No need to do it if the node has
245 		 * been deleted, because it will vanish after we return.
246 		 */
247 		tmpfs_update(vp);
248 	}
249 
250 	return vop_stdclose(v);
251 }
252 
253 /* --------------------------------------------------------------------- */
254 
255 int
256 tmpfs_access(struct vop_access_args *v)
257 {
258 	struct vnode *vp = v->a_vp;
259 	int error;
260 	struct tmpfs_node *node;
261 
262 	node = VP_TO_TMPFS_NODE(vp);
263 
264 	switch (vp->v_type) {
265 	case VDIR:
266 		/* FALLTHROUGH */
267 	case VLNK:
268 		/* FALLTHROUGH */
269 	case VREG:
270 		if ((v->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
271 			error = EROFS;
272 			goto out;
273 		}
274 		break;
275 
276 	case VBLK:
277 		/* FALLTHROUGH */
278 	case VCHR:
279 		/* FALLTHROUGH */
280 	case VSOCK:
281 		/* FALLTHROUGH */
282 	case VFIFO:
283 		break;
284 
285 	default:
286 		error = EINVAL;
287 		goto out;
288 	}
289 
290 	if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
291 		error = EPERM;
292 		goto out;
293 	}
294 
295 	error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
296 
297 out:
298 
299 	return error;
300 }
301 
302 /* --------------------------------------------------------------------- */
303 
304 int
305 tmpfs_getattr(struct vop_getattr_args *v)
306 {
307 	struct vnode *vp = v->a_vp;
308 	struct vattr *vap = v->a_vap;
309 	struct tmpfs_node *node;
310 
311 	node = VP_TO_TMPFS_NODE(vp);
312 
313 	lwkt_gettoken(&vp->v_mount->mnt_token);
314 	tmpfs_update(vp);
315 
316 	vap->va_type = vp->v_type;
317 	vap->va_mode = node->tn_mode;
318 	vap->va_nlink = node->tn_links;
319 	vap->va_uid = node->tn_uid;
320 	vap->va_gid = node->tn_gid;
321 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
322 	vap->va_fileid = node->tn_id;
323 	vap->va_size = node->tn_size;
324 	vap->va_blocksize = PAGE_SIZE;
325 	vap->va_atime.tv_sec = node->tn_atime;
326 	vap->va_atime.tv_nsec = node->tn_atimensec;
327 	vap->va_mtime.tv_sec = node->tn_mtime;
328 	vap->va_mtime.tv_nsec = node->tn_mtimensec;
329 	vap->va_ctime.tv_sec = node->tn_ctime;
330 	vap->va_ctime.tv_nsec = node->tn_ctimensec;
331 	vap->va_gen = node->tn_gen;
332 	vap->va_flags = node->tn_flags;
333 	if (vp->v_type == VBLK || vp->v_type == VCHR)
334 	{
335 		vap->va_rmajor = umajor(node->tn_rdev);
336 		vap->va_rminor = uminor(node->tn_rdev);
337 	}
338 	vap->va_bytes = round_page(node->tn_size);
339 	vap->va_filerev = 0;
340 
341 	lwkt_reltoken(&vp->v_mount->mnt_token);
342 
343 	return 0;
344 }
345 
346 /* --------------------------------------------------------------------- */
347 
348 int
349 tmpfs_setattr(struct vop_setattr_args *v)
350 {
351 	struct vnode *vp = v->a_vp;
352 	struct vattr *vap = v->a_vap;
353 	struct ucred *cred = v->a_cred;
354 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
355 	int error = 0;
356 	int kflags = 0;
357 
358 	if (error == 0 && (vap->va_flags != VNOVAL)) {
359 		error = tmpfs_chflags(vp, vap->va_flags, cred);
360 		kflags |= NOTE_ATTRIB;
361 	}
362 
363 	if (error == 0 && (vap->va_size != VNOVAL)) {
364 		if (vap->va_size > node->tn_size)
365 			kflags |= NOTE_WRITE | NOTE_EXTEND;
366 		else
367 			kflags |= NOTE_WRITE;
368 		error = tmpfs_chsize(vp, vap->va_size, cred);
369 	}
370 
371 	if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
372 			   vap->va_gid != (gid_t)VNOVAL)) {
373 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
374 		kflags |= NOTE_ATTRIB;
375 	}
376 
377 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
378 		error = tmpfs_chmod(vp, vap->va_mode, cred);
379 		kflags |= NOTE_ATTRIB;
380 	}
381 
382 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
383 	    vap->va_atime.tv_nsec != VNOVAL) ||
384 	    (vap->va_mtime.tv_sec != VNOVAL &&
385 	    vap->va_mtime.tv_nsec != VNOVAL) )) {
386 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
387 				      vap->va_vaflags, cred);
388 		kflags |= NOTE_ATTRIB;
389 	}
390 
391 	/* Update the node times.  We give preference to the error codes
392 	 * generated by this function rather than the ones that may arise
393 	 * from tmpfs_update. */
394 	tmpfs_update(vp);
395 	tmpfs_knote(vp, kflags);
396 
397 	return error;
398 }
399 
400 /* --------------------------------------------------------------------- */
401 
402 /*
403  * fsync is usually a NOP, but we must take action when unmounting or
404  * when recycling.
405  */
406 static int
407 tmpfs_fsync(struct vop_fsync_args *v)
408 {
409 	struct tmpfs_node *node;
410 	struct vnode *vp = v->a_vp;
411 
412 	node = VP_TO_TMPFS_NODE(vp);
413 
414 	tmpfs_update(vp);
415 	if (vp->v_type == VREG) {
416 		if (vp->v_flag & VRECLAIMED) {
417 			if (node->tn_links == 0)
418 				tmpfs_truncate(vp, 0);
419 			else
420 				vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
421 		}
422 	}
423 	return 0;
424 }
425 
426 /* --------------------------------------------------------------------- */
427 
428 static int
429 tmpfs_read (struct vop_read_args *ap)
430 {
431 	struct buf *bp;
432 	struct vnode *vp = ap->a_vp;
433 	struct uio *uio = ap->a_uio;
434 	struct tmpfs_node *node;
435 	off_t base_offset;
436 	size_t offset;
437 	size_t len;
438 	int error;
439 
440 	error = 0;
441 	if (uio->uio_resid == 0) {
442 		return error;
443 	}
444 
445 	node = VP_TO_TMPFS_NODE(vp);
446 
447 	if (uio->uio_offset < 0)
448 		return (EINVAL);
449 	if (vp->v_type != VREG)
450 		return (EINVAL);
451 
452 	while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
453 		/*
454 		 * Use buffer cache I/O (via tmpfs_strategy)
455 		 */
456 		offset = (size_t)uio->uio_offset & BMASK;
457 		base_offset = (off_t)uio->uio_offset - offset;
458 		bp = getcacheblk(vp, base_offset, BSIZE, 0);
459 		if (bp == NULL) {
460 			lwkt_gettoken(&vp->v_mount->mnt_token);
461 			error = bread(vp, base_offset, BSIZE, &bp);
462 			if (error) {
463 				brelse(bp);
464 				lwkt_reltoken(&vp->v_mount->mnt_token);
465 				kprintf("tmpfs_read bread error %d\n", error);
466 				break;
467 			}
468 			lwkt_reltoken(&vp->v_mount->mnt_token);
469 		}
470 
471 		/*
472 		 * Figure out how many bytes we can actually copy this loop.
473 		 */
474 		len = BSIZE - offset;
475 		if (len > uio->uio_resid)
476 			len = uio->uio_resid;
477 		if (len > node->tn_size - uio->uio_offset)
478 			len = (size_t)(node->tn_size - uio->uio_offset);
479 
480 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
481 		bqrelse(bp);
482 		if (error) {
483 			kprintf("tmpfs_read uiomove error %d\n", error);
484 			break;
485 		}
486 	}
487 
488 	TMPFS_NODE_LOCK(node);
489 	node->tn_status |= TMPFS_NODE_ACCESSED;
490 	TMPFS_NODE_UNLOCK(node);
491 
492 	return(error);
493 }
494 
495 static int
496 tmpfs_write (struct vop_write_args *ap)
497 {
498 	struct buf *bp;
499 	struct vnode *vp = ap->a_vp;
500 	struct uio *uio = ap->a_uio;
501 	struct thread *td = uio->uio_td;
502 	struct tmpfs_node *node;
503 	boolean_t extended;
504 	off_t oldsize;
505 	int error;
506 	off_t base_offset;
507 	size_t offset;
508 	size_t len;
509 	struct rlimit limit;
510 	int trivial = 0;
511 	int kflags = 0;
512 
513 	error = 0;
514 	if (uio->uio_resid == 0) {
515 		return error;
516 	}
517 
518 	node = VP_TO_TMPFS_NODE(vp);
519 
520 	if (vp->v_type != VREG)
521 		return (EINVAL);
522 
523 	lwkt_gettoken(&vp->v_mount->mnt_token);
524 
525 	oldsize = node->tn_size;
526 	if (ap->a_ioflag & IO_APPEND)
527 		uio->uio_offset = node->tn_size;
528 
529 	/*
530 	 * Check for illegal write offsets.
531 	 */
532 	if (uio->uio_offset + uio->uio_resid >
533 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
534 		lwkt_reltoken(&vp->v_mount->mnt_token);
535 		return (EFBIG);
536 	}
537 
538 	/*
539 	 * NOTE: Ignore if UIO does not come from a user thread (e.g. VN).
540 	 */
541 	if (vp->v_type == VREG && td != NULL && td->td_lwp != NULL) {
542 		error = kern_getrlimit(RLIMIT_FSIZE, &limit);
543 		if (error != 0) {
544 			lwkt_reltoken(&vp->v_mount->mnt_token);
545 			return error;
546 		}
547 		if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
548 			ksignal(td->td_proc, SIGXFSZ);
549 			lwkt_reltoken(&vp->v_mount->mnt_token);
550 			return (EFBIG);
551 		}
552 	}
553 
554 
555 	/*
556 	 * Extend the file's size if necessary
557 	 */
558 	extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
559 
560 	while (uio->uio_resid > 0) {
561 		/*
562 		 * Use buffer cache I/O (via tmpfs_strategy)
563 		 */
564 		offset = (size_t)uio->uio_offset & BMASK;
565 		base_offset = (off_t)uio->uio_offset - offset;
566 		len = BSIZE - offset;
567 		if (len > uio->uio_resid)
568 			len = uio->uio_resid;
569 
570 		if ((uio->uio_offset + len) > node->tn_size) {
571 			trivial = (uio->uio_offset <= node->tn_size);
572 			error = tmpfs_reg_resize(vp, uio->uio_offset + len,  trivial);
573 			if (error)
574 				break;
575 		}
576 
577 		/*
578 		 * Read to fill in any gaps.  Theoretically we could
579 		 * optimize this if the write covers the entire buffer
580 		 * and is not a UIO_NOCOPY write, however this can lead
581 		 * to a security violation exposing random kernel memory
582 		 * (whatever junk was in the backing VM pages before).
583 		 *
584 		 * So just use bread() to do the right thing.
585 		 */
586 		error = bread(vp, base_offset, BSIZE, &bp);
587 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
588 		if (error) {
589 			kprintf("tmpfs_write uiomove error %d\n", error);
590 			brelse(bp);
591 			break;
592 		}
593 
594 		if (uio->uio_offset > node->tn_size) {
595 			node->tn_size = uio->uio_offset;
596 			kflags |= NOTE_EXTEND;
597 		}
598 		kflags |= NOTE_WRITE;
599 
600 		/*
601 		 * Always try to flush the page in the UIO_NOCOPY case.  This
602 		 * can come from the pageout daemon or during vnode eviction.
603 		 * It is not necessarily going to be marked IO_ASYNC/IO_SYNC.
604 		 *
605 		 * For the normal case we buwrite(), dirtying the underlying
606 		 * VM pages instead of dirtying the buffer and releasing the
607 		 * buffer as a clean buffer.  This allows tmpfs to use
608 		 * essentially all available memory to cache file data.
609 		 * If we used bdwrite() the buffer cache would wind up
610 		 * flushing the data to swap too quickly.
611 		 */
612 		bp->b_flags |= B_AGE;
613 		if (uio->uio_segflg == UIO_NOCOPY) {
614 			bawrite(bp);
615 		} else {
616 			buwrite(bp);
617 		}
618 
619 		if (bp->b_error) {
620 			kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
621 			break;
622 		}
623 	}
624 
625 	if (error) {
626 		if (extended) {
627 			(void)tmpfs_reg_resize(vp, oldsize, trivial);
628 			kflags &= ~NOTE_EXTEND;
629 		}
630 		goto done;
631 	}
632 
633 	/*
634 	 * Currently we don't set the mtime on files modified via mmap()
635 	 * because we can't tell the difference between those modifications
636 	 * and an attempt by the pageout daemon to flush tmpfs pages to
637 	 * swap.
638 	 *
639 	 * This is because in order to defer flushes as long as possible
640 	 * buwrite() works by marking the underlying VM pages dirty in
641 	 * order to be able to dispose of the buffer cache buffer without
642 	 * flushing it.
643 	 */
644 	TMPFS_NODE_LOCK(node);
645 	if (uio->uio_segflg != UIO_NOCOPY)
646 		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED;
647 	if (extended)
648 		node->tn_status |= TMPFS_NODE_CHANGED;
649 
650 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
651 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
652 			node->tn_mode &= ~(S_ISUID | S_ISGID);
653 	}
654 	TMPFS_NODE_UNLOCK(node);
655 done:
656 
657 	tmpfs_knote(vp, kflags);
658 
659 
660 	lwkt_reltoken(&vp->v_mount->mnt_token);
661 	return(error);
662 }
663 
664 static int
665 tmpfs_advlock (struct vop_advlock_args *ap)
666 {
667 	struct tmpfs_node *node;
668 	struct vnode *vp = ap->a_vp;
669 
670 	node = VP_TO_TMPFS_NODE(vp);
671 
672 	return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
673 }
674 
675 /*
676  * The strategy function is typically only called when memory pressure
677  * forces the system to attempt to pageout pages.  It can also be called
678  * by [n]vtruncbuf() when a truncation cuts a page in half.  Normal write
679  * operations
680  */
681 static int
682 tmpfs_strategy(struct vop_strategy_args *ap)
683 {
684 	struct bio *bio = ap->a_bio;
685 	struct bio *nbio;
686 	struct buf *bp = bio->bio_buf;
687 	struct vnode *vp = ap->a_vp;
688 	struct tmpfs_node *node;
689 	vm_object_t uobj;
690 	vm_page_t m;
691 	int i;
692 
693 	if (vp->v_type != VREG) {
694 		bp->b_resid = bp->b_bcount;
695 		bp->b_flags |= B_ERROR | B_INVAL;
696 		bp->b_error = EINVAL;
697 		biodone(bio);
698 		return(0);
699 	}
700 
701 	lwkt_gettoken(&vp->v_mount->mnt_token);
702 	node = VP_TO_TMPFS_NODE(vp);
703 
704 	uobj = node->tn_reg.tn_aobj;
705 
706 	/*
707 	 * Don't bother flushing to swap if there is no swap, just
708 	 * ensure that the pages are marked as needing a commit (still).
709 	 */
710 	if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
711 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
712 			m = bp->b_xio.xio_pages[i];
713 			vm_page_need_commit(m);
714 		}
715 		bp->b_resid = 0;
716 		bp->b_error = 0;
717 		biodone(bio);
718 	} else {
719 		nbio = push_bio(bio);
720 		nbio->bio_done = tmpfs_strategy_done;
721 		nbio->bio_offset = bio->bio_offset;
722 		swap_pager_strategy(uobj, nbio);
723 	}
724 
725 	lwkt_reltoken(&vp->v_mount->mnt_token);
726 	return 0;
727 }
728 
729 /*
730  * If we were unable to commit the pages to swap make sure they are marked
731  * as needing a commit (again).  If we were, clear the flag to allow the
732  * pages to be freed.
733  */
734 static void
735 tmpfs_strategy_done(struct bio *bio)
736 {
737 	struct buf *bp;
738 	vm_page_t m;
739 	int i;
740 
741 	bp = bio->bio_buf;
742 
743 	if (bp->b_flags & B_ERROR) {
744 		bp->b_flags &= ~B_ERROR;
745 		bp->b_error = 0;
746 		bp->b_resid = 0;
747 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
748 			m = bp->b_xio.xio_pages[i];
749 			vm_page_need_commit(m);
750 		}
751 	} else {
752 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
753 			m = bp->b_xio.xio_pages[i];
754 			vm_page_clear_commit(m);
755 		}
756 	}
757 	bio = pop_bio(bio);
758 	biodone(bio);
759 }
760 
761 static int
762 tmpfs_bmap(struct vop_bmap_args *ap)
763 {
764 	if (ap->a_doffsetp != NULL)
765 		*ap->a_doffsetp = ap->a_loffset;
766 	if (ap->a_runp != NULL)
767 		*ap->a_runp = 0;
768 	if (ap->a_runb != NULL)
769 		*ap->a_runb = 0;
770 
771 	return 0;
772 }
773 
774 /* --------------------------------------------------------------------- */
775 
776 static int
777 tmpfs_nremove(struct vop_nremove_args *v)
778 {
779 	struct vnode *dvp = v->a_dvp;
780 	struct namecache *ncp = v->a_nch->ncp;
781 	struct vnode *vp;
782 	int error;
783 	struct tmpfs_dirent *de;
784 	struct tmpfs_mount *tmp;
785 	struct tmpfs_node *dnode;
786 	struct tmpfs_node *node;
787 
788 	/*
789 	 * We have to acquire the vp from v->a_nch because we will likely
790 	 * unresolve the namecache entry, and a vrele/vput is needed to
791 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
792 	 *
793 	 * We have to use vget to clear any inactive state on the vnode,
794 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
795 	 * will not get called when we release it.
796 	 */
797 	error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
798 	KKASSERT(error == 0);
799 	vn_unlock(vp);
800 
801 	if (vp->v_type == VDIR) {
802 		error = EISDIR;
803 		goto out;
804 	}
805 
806 	dnode = VP_TO_TMPFS_DIR(dvp);
807 	node = VP_TO_TMPFS_NODE(vp);
808 	tmp = VFS_TO_TMPFS(vp->v_mount);
809 	de = tmpfs_dir_lookup(dnode, node, ncp);
810 	if (de == NULL) {
811 		error = ENOENT;
812 		goto out;
813 	}
814 
815 	/* Files marked as immutable or append-only cannot be deleted. */
816 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
817 	    (dnode->tn_flags & APPEND)) {
818 		error = EPERM;
819 		goto out;
820 	}
821 
822 	/* Remove the entry from the directory; as it is a file, we do not
823 	 * have to change the number of hard links of the directory. */
824 	tmpfs_dir_detach(dnode, de);
825 
826 	/* Free the directory entry we just deleted.  Note that the node
827 	 * referred by it will not be removed until the vnode is really
828 	 * reclaimed. */
829 	tmpfs_free_dirent(tmp, de);
830 
831 	if (node->tn_links > 0) {
832 	        TMPFS_NODE_LOCK(node);
833 		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
834 	                TMPFS_NODE_MODIFIED;
835 	        TMPFS_NODE_UNLOCK(node);
836 	}
837 
838 	cache_unlink(v->a_nch);
839 	tmpfs_knote(vp, NOTE_DELETE);
840 	tmpfs_knote(dvp, NOTE_WRITE);
841 	error = 0;
842 
843 out:
844 	vrele(vp);
845 
846 	return error;
847 }
848 
849 /* --------------------------------------------------------------------- */
850 
851 static int
852 tmpfs_nlink(struct vop_nlink_args *v)
853 {
854 	struct vnode *dvp = v->a_dvp;
855 	struct vnode *vp = v->a_vp;
856 	struct namecache *ncp = v->a_nch->ncp;
857 	struct tmpfs_dirent *de;
858 	struct tmpfs_node *node;
859 	struct tmpfs_node *dnode;
860 	int error;
861 
862 	KKASSERT(dvp != vp); /* XXX When can this be false? */
863 
864 	node = VP_TO_TMPFS_NODE(vp);
865 	dnode = VP_TO_TMPFS_NODE(dvp);
866 
867 	/* XXX: Why aren't the following two tests done by the caller? */
868 
869 	/* Hard links of directories are forbidden. */
870 	if (vp->v_type == VDIR) {
871 		error = EPERM;
872 		goto out;
873 	}
874 
875 	/* Cannot create cross-device links. */
876 	if (dvp->v_mount != vp->v_mount) {
877 		error = EXDEV;
878 		goto out;
879 	}
880 
881 	/* Ensure that we do not overflow the maximum number of links imposed
882 	 * by the system. */
883 	KKASSERT(node->tn_links <= LINK_MAX);
884 	if (node->tn_links == LINK_MAX) {
885 		error = EMLINK;
886 		goto out;
887 	}
888 
889 	/* We cannot create links of files marked immutable or append-only. */
890 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
891 		error = EPERM;
892 		goto out;
893 	}
894 
895 	/* Allocate a new directory entry to represent the node. */
896 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
897 	    ncp->nc_name, ncp->nc_nlen, &de);
898 	if (error != 0)
899 		goto out;
900 
901 	/* Insert the new directory entry into the appropriate directory. */
902 	tmpfs_dir_attach(dnode, de);
903 
904 	/* vp link count has changed, so update node times. */
905 
906 	TMPFS_NODE_LOCK(node);
907 	node->tn_status |= TMPFS_NODE_CHANGED;
908 	TMPFS_NODE_UNLOCK(node);
909 	tmpfs_update(vp);
910 
911 	tmpfs_knote(vp, NOTE_LINK);
912 	cache_setunresolved(v->a_nch);
913 	cache_setvp(v->a_nch, vp);
914 	tmpfs_knote(dvp, NOTE_WRITE);
915 	error = 0;
916 
917 out:
918 	return error;
919 }
920 
921 /* --------------------------------------------------------------------- */
922 
923 static int
924 tmpfs_nrename(struct vop_nrename_args *v)
925 {
926 	struct vnode *fdvp = v->a_fdvp;
927 	struct namecache *fncp = v->a_fnch->ncp;
928 	struct vnode *fvp = fncp->nc_vp;
929 	struct vnode *tdvp = v->a_tdvp;
930 	struct namecache *tncp = v->a_tnch->ncp;
931 	struct vnode *tvp;
932 	struct tmpfs_dirent *de, *tde;
933 	struct tmpfs_mount *tmp;
934 	struct tmpfs_node *fdnode;
935 	struct tmpfs_node *fnode;
936 	struct tmpfs_node *tnode;
937 	struct tmpfs_node *tdnode;
938 	char *newname;
939 	char *oldname;
940 	int error;
941 
942 	/*
943 	 * Because tvp can get overwritten we have to vget it instead of
944 	 * just vref or use it, otherwise it's VINACTIVE flag may not get
945 	 * cleared and the node won't get destroyed.
946 	 */
947 	error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp);
948 	if (error == 0) {
949 		tnode = VP_TO_TMPFS_NODE(tvp);
950 		vn_unlock(tvp);
951 	} else {
952 		tnode = NULL;
953 	}
954 
955 	/* Disallow cross-device renames.
956 	 * XXX Why isn't this done by the caller? */
957 	if (fvp->v_mount != tdvp->v_mount ||
958 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
959 		error = EXDEV;
960 		goto out;
961 	}
962 
963 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
964 	tdnode = VP_TO_TMPFS_DIR(tdvp);
965 
966 	/* If source and target are the same file, there is nothing to do. */
967 	if (fvp == tvp) {
968 		error = 0;
969 		goto out;
970 	}
971 
972 	fdnode = VP_TO_TMPFS_DIR(fdvp);
973 	fnode = VP_TO_TMPFS_NODE(fvp);
974 	de = tmpfs_dir_lookup(fdnode, fnode, fncp);
975 
976 	/* Avoid manipulating '.' and '..' entries. */
977 	if (de == NULL) {
978 		error = ENOENT;
979 		goto out_locked;
980 	}
981 	KKASSERT(de->td_node == fnode);
982 
983 	/*
984 	 * If replacing an entry in the target directory and that entry
985 	 * is a directory, it must be empty.
986 	 *
987 	 * Kern_rename gurantees the destination to be a directory
988 	 * if the source is one (it does?).
989 	 */
990 	if (tvp != NULL) {
991 		KKASSERT(tnode != NULL);
992 
993 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
994 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
995 			error = EPERM;
996 			goto out_locked;
997 		}
998 
999 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1000 			if (tnode->tn_size > 0) {
1001 				error = ENOTEMPTY;
1002 				goto out_locked;
1003 			}
1004 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1005 			error = ENOTDIR;
1006 			goto out_locked;
1007 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1008 			error = EISDIR;
1009 			goto out_locked;
1010 		} else {
1011 			KKASSERT(fnode->tn_type != VDIR &&
1012 				tnode->tn_type != VDIR);
1013 		}
1014 	}
1015 
1016 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1017 	    (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1018 		error = EPERM;
1019 		goto out_locked;
1020 	}
1021 
1022 	/*
1023 	 * Ensure that we have enough memory to hold the new name, if it
1024 	 * has to be changed.
1025 	 */
1026 	if (fncp->nc_nlen != tncp->nc_nlen ||
1027 	    bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1028 		newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1029 				  M_WAITOK | M_NULLOK);
1030 		if (newname == NULL) {
1031 			error = ENOSPC;
1032 			goto out_locked;
1033 		}
1034 		bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1035 		newname[tncp->nc_nlen] = '\0';
1036 	} else {
1037 		newname = NULL;
1038 	}
1039 
1040 	/*
1041 	 * Unlink entry from source directory.  Note that the kernel has
1042 	 * already checked for illegal recursion cases (renaming a directory
1043 	 * into a subdirectory of itself).
1044 	 */
1045 	if (fdnode != tdnode) {
1046 		tmpfs_dir_detach(fdnode, de);
1047 	} else {
1048 		RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de);
1049 	}
1050 
1051 	/*
1052 	 * Handle any name change.  Swap with newname, we will
1053 	 * deallocate it at the end.
1054 	 */
1055 	if (newname != NULL) {
1056 #if 0
1057 		TMPFS_NODE_LOCK(fnode);
1058 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1059 		TMPFS_NODE_UNLOCK(fnode);
1060 #endif
1061 		oldname = de->td_name;
1062 		de->td_name = newname;
1063 		de->td_namelen = (uint16_t)tncp->nc_nlen;
1064 		newname = oldname;
1065 	}
1066 
1067 	/*
1068 	 * If we are overwriting an entry, we have to remove the old one
1069 	 * from the target directory.
1070 	 */
1071 	if (tvp != NULL) {
1072 		/* Remove the old entry from the target directory. */
1073 		tde = tmpfs_dir_lookup(tdnode, tnode, tncp);
1074 		tmpfs_dir_detach(tdnode, tde);
1075 		tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1076 
1077 		/*
1078 		 * Free the directory entry we just deleted.  Note that the
1079 		 * node referred by it will not be removed until the vnode is
1080 		 * really reclaimed.
1081 		 */
1082 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1083 		/*cache_inval_vp(tvp, CINV_DESTROY);*/
1084 	}
1085 
1086 	/*
1087 	 * Link entry to target directory.  If the entry
1088 	 * represents a directory move the parent linkage
1089 	 * as well.
1090 	 */
1091 	if (fdnode != tdnode) {
1092 		if (de->td_node->tn_type == VDIR) {
1093 			TMPFS_VALIDATE_DIR(fnode);
1094 		}
1095 		tmpfs_dir_attach(tdnode, de);
1096 	} else {
1097 		TMPFS_NODE_LOCK(tdnode);
1098 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1099 		RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de);
1100 		TMPFS_NODE_UNLOCK(tdnode);
1101 	}
1102 
1103 	/*
1104 	 * Finish up
1105 	 */
1106 	if (newname) {
1107 		kfree(newname, tmp->tm_name_zone);
1108 		newname = NULL;
1109 	}
1110 	cache_rename(v->a_fnch, v->a_tnch);
1111 	tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1112 	tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1113 	if (fnode->tn_vnode)
1114 		tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1115 	error = 0;
1116 
1117 out_locked:
1118 	;
1119 
1120 out:
1121 	if (tvp)
1122 		vrele(tvp);
1123 
1124 	return error;
1125 }
1126 
1127 /* --------------------------------------------------------------------- */
1128 
1129 static int
1130 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1131 {
1132 	struct vnode *dvp = v->a_dvp;
1133 	struct vnode **vpp = v->a_vpp;
1134 	struct namecache *ncp = v->a_nch->ncp;
1135 	struct vattr *vap = v->a_vap;
1136 	struct ucred *cred = v->a_cred;
1137 	int error;
1138 
1139 	KKASSERT(vap->va_type == VDIR);
1140 
1141 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1142 	if (error == 0) {
1143 		cache_setunresolved(v->a_nch);
1144 		cache_setvp(v->a_nch, *vpp);
1145 		tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1146 	}
1147 
1148 	return error;
1149 }
1150 
1151 /* --------------------------------------------------------------------- */
1152 
1153 static int
1154 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1155 {
1156 	struct vnode *dvp = v->a_dvp;
1157 	struct namecache *ncp = v->a_nch->ncp;
1158 	struct vnode *vp;
1159 	struct tmpfs_dirent *de;
1160 	struct tmpfs_mount *tmp;
1161 	struct tmpfs_node *dnode;
1162 	struct tmpfs_node *node;
1163 	int error;
1164 
1165 	/*
1166 	 * We have to acquire the vp from v->a_nch because we will likely
1167 	 * unresolve the namecache entry, and a vrele/vput is needed to
1168 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1169 	 *
1170 	 * We have to use vget to clear any inactive state on the vnode,
1171 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1172 	 * will not get called when we release it.
1173 	 */
1174 	error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
1175 	KKASSERT(error == 0);
1176 	vn_unlock(vp);
1177 
1178 	/*
1179 	 * Prevalidate so we don't hit an assertion later
1180 	 */
1181 	if (vp->v_type != VDIR) {
1182 		error = ENOTDIR;
1183 		goto out;
1184 	}
1185 
1186 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1187 	dnode = VP_TO_TMPFS_DIR(dvp);
1188 	node = VP_TO_TMPFS_DIR(vp);
1189 
1190 	/* Directories with more than two entries ('.' and '..') cannot be
1191 	 * removed. */
1192 	 if (node->tn_size > 0) {
1193 		 error = ENOTEMPTY;
1194 		 goto out;
1195 	 }
1196 
1197 	if ((dnode->tn_flags & APPEND)
1198 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1199 		error = EPERM;
1200 		goto out;
1201 	}
1202 
1203 	/* This invariant holds only if we are not trying to remove "..".
1204 	  * We checked for that above so this is safe now. */
1205 	KKASSERT(node->tn_dir.tn_parent == dnode);
1206 
1207 	/* Get the directory entry associated with node (vp).  This was
1208 	 * filled by tmpfs_lookup while looking up the entry. */
1209 	de = tmpfs_dir_lookup(dnode, node, ncp);
1210 	KKASSERT(TMPFS_DIRENT_MATCHES(de,
1211 	    ncp->nc_name,
1212 	    ncp->nc_nlen));
1213 
1214 	/* Check flags to see if we are allowed to remove the directory. */
1215 	if ((dnode->tn_flags & APPEND) ||
1216 	    node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1217 		error = EPERM;
1218 		goto out;
1219 	}
1220 
1221 
1222 	/* Detach the directory entry from the directory (dnode). */
1223 	tmpfs_dir_detach(dnode, de);
1224 
1225 	/* No vnode should be allocated for this entry from this point */
1226 	TMPFS_NODE_LOCK(node);
1227 	TMPFS_ASSERT_ELOCKED(node);
1228 	TMPFS_NODE_LOCK(dnode);
1229 	TMPFS_ASSERT_ELOCKED(dnode);
1230 
1231 	/*
1232 	 * Must set parent linkage to NULL (tested by ncreate to disallow
1233 	 * the creation of new files/dirs in a deleted directory)
1234 	 */
1235 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1236 	    TMPFS_NODE_MODIFIED;
1237 
1238 	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1239 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1240 
1241 	TMPFS_NODE_UNLOCK(dnode);
1242 	TMPFS_NODE_UNLOCK(node);
1243 
1244 	/* Free the directory entry we just deleted.  Note that the node
1245 	 * referred by it will not be removed until the vnode is really
1246 	 * reclaimed. */
1247 	tmpfs_free_dirent(tmp, de);
1248 
1249 	/* Release the deleted vnode (will destroy the node, notify
1250 	 * interested parties and clean it from the cache). */
1251 
1252 	TMPFS_NODE_LOCK(dnode);
1253 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1254 	TMPFS_NODE_UNLOCK(dnode);
1255 	tmpfs_update(dvp);
1256 
1257 	cache_unlink(v->a_nch);
1258 	tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1259 	error = 0;
1260 
1261 out:
1262 	vrele(vp);
1263 
1264 	return error;
1265 }
1266 
1267 /* --------------------------------------------------------------------- */
1268 
1269 static int
1270 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1271 {
1272 	struct vnode *dvp = v->a_dvp;
1273 	struct vnode **vpp = v->a_vpp;
1274 	struct namecache *ncp = v->a_nch->ncp;
1275 	struct vattr *vap = v->a_vap;
1276 	struct ucred *cred = v->a_cred;
1277 	char *target = v->a_target;
1278 	int error;
1279 
1280 	vap->va_type = VLNK;
1281 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1282 	if (error == 0) {
1283 		tmpfs_knote(*vpp, NOTE_WRITE);
1284 		cache_setunresolved(v->a_nch);
1285 		cache_setvp(v->a_nch, *vpp);
1286 	}
1287 
1288 	return error;
1289 }
1290 
1291 /* --------------------------------------------------------------------- */
1292 
1293 static int
1294 tmpfs_readdir(struct vop_readdir_args *v)
1295 {
1296 	struct vnode *vp = v->a_vp;
1297 	struct uio *uio = v->a_uio;
1298 	int *eofflag = v->a_eofflag;
1299 	off_t **cookies = v->a_cookies;
1300 	int *ncookies = v->a_ncookies;
1301 	struct tmpfs_mount *tmp;
1302 	int error;
1303 	off_t startoff;
1304 	off_t cnt = 0;
1305 	struct tmpfs_node *node;
1306 
1307 	/* This operation only makes sense on directory nodes. */
1308 	if (vp->v_type != VDIR)
1309 		return ENOTDIR;
1310 
1311 	tmp = VFS_TO_TMPFS(vp->v_mount);
1312 	node = VP_TO_TMPFS_DIR(vp);
1313 	startoff = uio->uio_offset;
1314 
1315 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1316 		error = tmpfs_dir_getdotdent(node, uio);
1317 		if (error != 0)
1318 			goto outok;
1319 		cnt++;
1320 	}
1321 
1322 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1323 		error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1324 		if (error != 0)
1325 			goto outok;
1326 		cnt++;
1327 	}
1328 
1329 	error = tmpfs_dir_getdents(node, uio, &cnt);
1330 
1331 outok:
1332 	KKASSERT(error >= -1);
1333 
1334 	if (error == -1)
1335 		error = 0;
1336 
1337 	if (eofflag != NULL)
1338 		*eofflag =
1339 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1340 
1341 	/* Update NFS-related variables. */
1342 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1343 		off_t i;
1344 		off_t off = startoff;
1345 		struct tmpfs_dirent *de = NULL;
1346 
1347 		*ncookies = cnt;
1348 		*cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1349 
1350 		for (i = 0; i < cnt; i++) {
1351 			KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1352 			if (off == TMPFS_DIRCOOKIE_DOT) {
1353 				off = TMPFS_DIRCOOKIE_DOTDOT;
1354 			} else {
1355 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1356 					de = RB_MIN(tmpfs_dirtree, &node->tn_dir.tn_dirtree);
1357 				} else if (de != NULL) {
1358 					de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
1359 				} else {
1360 					de = tmpfs_dir_lookupbycookie(node,
1361 					    off);
1362 					KKASSERT(de != NULL);
1363 					de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
1364 				}
1365 				if (de == NULL)
1366 					off = TMPFS_DIRCOOKIE_EOF;
1367 				else
1368 					off = tmpfs_dircookie(de);
1369 			}
1370 
1371 			(*cookies)[i] = off;
1372 		}
1373 		KKASSERT(uio->uio_offset == off);
1374 	}
1375 
1376 	return error;
1377 }
1378 
1379 /* --------------------------------------------------------------------- */
1380 
1381 static int
1382 tmpfs_readlink(struct vop_readlink_args *v)
1383 {
1384 	struct vnode *vp = v->a_vp;
1385 	struct uio *uio = v->a_uio;
1386 
1387 	int error;
1388 	struct tmpfs_node *node;
1389 
1390 	KKASSERT(uio->uio_offset == 0);
1391 	KKASSERT(vp->v_type == VLNK);
1392 
1393 	node = VP_TO_TMPFS_NODE(vp);
1394 
1395 	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1396 	    uio);
1397 	TMPFS_NODE_LOCK(node);
1398 	node->tn_status |= TMPFS_NODE_ACCESSED;
1399 	TMPFS_NODE_UNLOCK(node);
1400 
1401 	return error;
1402 }
1403 
1404 /* --------------------------------------------------------------------- */
1405 
1406 static int
1407 tmpfs_inactive(struct vop_inactive_args *v)
1408 {
1409 	struct vnode *vp = v->a_vp;
1410 	struct tmpfs_node *node;
1411 
1412 	node = VP_TO_TMPFS_NODE(vp);
1413 
1414 	/*
1415 	 * Degenerate case
1416 	 */
1417 	if (node == NULL) {
1418 		vrecycle(vp);
1419 		return(0);
1420 	}
1421 
1422 	/*
1423 	 * Get rid of unreferenced deleted vnodes sooner rather than
1424 	 * later so the data memory can be recovered immediately.
1425 	 *
1426 	 * We must truncate the vnode to prevent the normal reclamation
1427 	 * path from flushing the data for the removed file to disk.
1428 	 */
1429 	TMPFS_NODE_LOCK(node);
1430 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1431 	    node->tn_links == 0)
1432 	{
1433 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1434 		TMPFS_NODE_UNLOCK(node);
1435 		if (node->tn_type == VREG)
1436 			tmpfs_truncate(vp, 0);
1437 		vrecycle(vp);
1438 	} else {
1439 		TMPFS_NODE_UNLOCK(node);
1440 	}
1441 
1442 	return 0;
1443 }
1444 
1445 /* --------------------------------------------------------------------- */
1446 #include <vm/vm_object.h>
1447 #include <vm/vm_page.h>
1448 
1449 int
1450 tmpfs_reclaim(struct vop_reclaim_args *v)
1451 {
1452 	struct vnode *vp = v->a_vp;
1453 	struct tmpfs_mount *tmp;
1454 	struct tmpfs_node *node;
1455 
1456 	node = VP_TO_TMPFS_NODE(vp);
1457 	tmp = VFS_TO_TMPFS(vp->v_mount);
1458 
1459 	tmpfs_free_vp(vp);
1460 
1461 	/*
1462 	 * If the node referenced by this vnode was deleted by the
1463 	 * user, we must free its associated data structures now that
1464 	 * the vnode is being reclaimed.
1465 	 *
1466 	 * Directories have an extra link ref.
1467 	 */
1468 	TMPFS_NODE_LOCK(node);
1469 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1470 	    node->tn_links == 0) {
1471 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1472 		tmpfs_free_node(tmp, node);
1473 		/* eats the lock */
1474 	} else {
1475 		TMPFS_NODE_UNLOCK(node);
1476 	}
1477 
1478 	KKASSERT(vp->v_data == NULL);
1479 	return 0;
1480 }
1481 
1482 /* --------------------------------------------------------------------- */
1483 
1484 static int
1485 tmpfs_mountctl(struct vop_mountctl_args *ap)
1486 {
1487 	struct tmpfs_mount *tmp;
1488 	struct mount *mp;
1489 	int rc;
1490 
1491 	switch (ap->a_op) {
1492 	case (MOUNTCTL_SET_EXPORT):
1493 		mp = ap->a_head.a_ops->head.vv_mount;
1494 		tmp = (struct tmpfs_mount *) mp->mnt_data;
1495 
1496 		if (ap->a_ctllen != sizeof(struct export_args))
1497 			rc = (EINVAL);
1498 		else
1499 			rc = vfs_export(mp, &tmp->tm_export,
1500 					(const struct export_args *) ap->a_ctl);
1501 		break;
1502 	default:
1503 		rc = vop_stdmountctl(ap);
1504 		break;
1505 	}
1506 	return (rc);
1507 }
1508 
1509 /* --------------------------------------------------------------------- */
1510 
1511 static int
1512 tmpfs_print(struct vop_print_args *v)
1513 {
1514 	struct vnode *vp = v->a_vp;
1515 
1516 	struct tmpfs_node *node;
1517 
1518 	node = VP_TO_TMPFS_NODE(vp);
1519 
1520 	kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1521 	    node, node->tn_flags, node->tn_links);
1522 	kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1523 	    node->tn_mode, node->tn_uid, node->tn_gid,
1524 	    (uintmax_t)node->tn_size, node->tn_status);
1525 
1526 	if (vp->v_type == VFIFO)
1527 		fifo_printinfo(vp);
1528 
1529 	kprintf("\n");
1530 
1531 	return 0;
1532 }
1533 
1534 /* --------------------------------------------------------------------- */
1535 
1536 static int
1537 tmpfs_pathconf(struct vop_pathconf_args *v)
1538 {
1539 	int name = v->a_name;
1540 	register_t *retval = v->a_retval;
1541 
1542 	int error;
1543 
1544 	error = 0;
1545 
1546 	switch (name) {
1547 	case _PC_LINK_MAX:
1548 		*retval = LINK_MAX;
1549 		break;
1550 
1551 	case _PC_NAME_MAX:
1552 		*retval = NAME_MAX;
1553 		break;
1554 
1555 	case _PC_PATH_MAX:
1556 		*retval = PATH_MAX;
1557 		break;
1558 
1559 	case _PC_PIPE_BUF:
1560 		*retval = PIPE_BUF;
1561 		break;
1562 
1563 	case _PC_CHOWN_RESTRICTED:
1564 		*retval = 1;
1565 		break;
1566 
1567 	case _PC_NO_TRUNC:
1568 		*retval = 1;
1569 		break;
1570 
1571 	case _PC_SYNC_IO:
1572 		*retval = 1;
1573 		break;
1574 
1575 	case _PC_FILESIZEBITS:
1576 		*retval = 0; /* XXX Don't know which value should I return. */
1577 		break;
1578 
1579 	default:
1580 		error = EINVAL;
1581 	}
1582 
1583 	return error;
1584 }
1585 
1586 /************************************************************************
1587  *                          KQFILTER OPS                                *
1588  ************************************************************************/
1589 
1590 static void filt_tmpfsdetach(struct knote *kn);
1591 static int filt_tmpfsread(struct knote *kn, long hint);
1592 static int filt_tmpfswrite(struct knote *kn, long hint);
1593 static int filt_tmpfsvnode(struct knote *kn, long hint);
1594 
1595 static struct filterops tmpfsread_filtops =
1596 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1597 static struct filterops tmpfswrite_filtops =
1598 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1599 static struct filterops tmpfsvnode_filtops =
1600 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1601 
1602 static int
1603 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1604 {
1605 	struct vnode *vp = ap->a_vp;
1606 	struct knote *kn = ap->a_kn;
1607 
1608 	switch (kn->kn_filter) {
1609 	case EVFILT_READ:
1610 		kn->kn_fop = &tmpfsread_filtops;
1611 		break;
1612 	case EVFILT_WRITE:
1613 		kn->kn_fop = &tmpfswrite_filtops;
1614 		break;
1615 	case EVFILT_VNODE:
1616 		kn->kn_fop = &tmpfsvnode_filtops;
1617 		break;
1618 	default:
1619 		return (EOPNOTSUPP);
1620 	}
1621 
1622 	kn->kn_hook = (caddr_t)vp;
1623 
1624 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1625 
1626 	return(0);
1627 }
1628 
1629 static void
1630 filt_tmpfsdetach(struct knote *kn)
1631 {
1632 	struct vnode *vp = (void *)kn->kn_hook;
1633 
1634 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1635 }
1636 
1637 static int
1638 filt_tmpfsread(struct knote *kn, long hint)
1639 {
1640 	struct vnode *vp = (void *)kn->kn_hook;
1641 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1642 	off_t off;
1643 
1644 	if (hint == NOTE_REVOKE) {
1645 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1646 		return(1);
1647 	}
1648 
1649 	/*
1650 	 * Interlock against MP races when performing this function.
1651 	 */
1652 	lwkt_gettoken(&vp->v_mount->mnt_token);
1653 	off = node->tn_size - kn->kn_fp->f_offset;
1654 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1655 	if (kn->kn_sfflags & NOTE_OLDAPI) {
1656 		lwkt_reltoken(&vp->v_mount->mnt_token);
1657 		return(1);
1658 	}
1659 
1660 	if (kn->kn_data == 0) {
1661 		kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1662 	}
1663 	lwkt_reltoken(&vp->v_mount->mnt_token);
1664 	return (kn->kn_data != 0);
1665 }
1666 
1667 static int
1668 filt_tmpfswrite(struct knote *kn, long hint)
1669 {
1670 	if (hint == NOTE_REVOKE)
1671 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1672 	kn->kn_data = 0;
1673 	return (1);
1674 }
1675 
1676 static int
1677 filt_tmpfsvnode(struct knote *kn, long hint)
1678 {
1679 	if (kn->kn_sfflags & hint)
1680 		kn->kn_fflags |= hint;
1681 	if (hint == NOTE_REVOKE) {
1682 		kn->kn_flags |= (EV_EOF | EV_NODATA);
1683 		return (1);
1684 	}
1685 	return (kn->kn_fflags != 0);
1686 }
1687 
1688 
1689 /* --------------------------------------------------------------------- */
1690 
1691 /*
1692  * vnode operations vector used for files stored in a tmpfs file system.
1693  */
1694 struct vop_ops tmpfs_vnode_vops = {
1695 	.vop_default =			vop_defaultop,
1696 	.vop_getpages = 		vop_stdgetpages,
1697 	.vop_putpages = 		vop_stdputpages,
1698 	.vop_ncreate =			tmpfs_ncreate,
1699 	.vop_nresolve =			tmpfs_nresolve,
1700 	.vop_nlookupdotdot =		tmpfs_nlookupdotdot,
1701 	.vop_nmknod =			tmpfs_nmknod,
1702 	.vop_open =			tmpfs_open,
1703 	.vop_close =			tmpfs_close,
1704 	.vop_access =			tmpfs_access,
1705 	.vop_getattr =			tmpfs_getattr,
1706 	.vop_setattr =			tmpfs_setattr,
1707 	.vop_read =			tmpfs_read,
1708 	.vop_write =			tmpfs_write,
1709 	.vop_fsync =			tmpfs_fsync,
1710 	.vop_mountctl =			tmpfs_mountctl,
1711 	.vop_nremove =			tmpfs_nremove,
1712 	.vop_nlink =			tmpfs_nlink,
1713 	.vop_nrename =			tmpfs_nrename,
1714 	.vop_nmkdir =			tmpfs_nmkdir,
1715 	.vop_nrmdir =			tmpfs_nrmdir,
1716 	.vop_nsymlink =			tmpfs_nsymlink,
1717 	.vop_readdir =			tmpfs_readdir,
1718 	.vop_readlink =			tmpfs_readlink,
1719 	.vop_inactive =			tmpfs_inactive,
1720 	.vop_reclaim =			tmpfs_reclaim,
1721 	.vop_print =			tmpfs_print,
1722 	.vop_pathconf =			tmpfs_pathconf,
1723 	.vop_bmap =			tmpfs_bmap,
1724 	.vop_strategy =			tmpfs_strategy,
1725 	.vop_advlock =			tmpfs_advlock,
1726 	.vop_kqfilter =			tmpfs_kqfilter
1727 };
1728