xref: /dragonfly/sys/vfs/tmpfs/tmpfs_vnops.c (revision 81c11cd3)
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 <sys/mplock2.h>
53 
54 #include <vm/vm.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_page.h>
57 #include <vm/vm_pager.h>
58 #include <vm/swap_pager.h>
59 
60 #include <vfs/fifofs/fifo.h>
61 #include <vfs/tmpfs/tmpfs_vnops.h>
62 #include <vfs/tmpfs/tmpfs.h>
63 
64 MALLOC_DECLARE(M_TMPFS);
65 
66 static __inline
67 void
68 tmpfs_knote(struct vnode *vp, int flags)
69 {
70 	if (flags)
71 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
72 }
73 
74 
75 /* --------------------------------------------------------------------- */
76 
77 static int
78 tmpfs_nresolve(struct vop_nresolve_args *v)
79 {
80 	struct vnode *dvp = v->a_dvp;
81 	struct vnode *vp = NULL;
82 	struct namecache *ncp = v->a_nch->ncp;
83 	struct tmpfs_node *tnode;
84 
85 	int error;
86 	struct tmpfs_dirent *de;
87 	struct tmpfs_node *dnode;
88 
89 	dnode = VP_TO_TMPFS_DIR(dvp);
90 
91 	de = tmpfs_dir_lookup(dnode, NULL, ncp);
92 	if (de == NULL) {
93 		error = ENOENT;
94 	} else {
95 		/*
96 		 * Allocate a vnode for the node we found.
97 		 */
98 		tnode = de->td_node;
99 		error = tmpfs_alloc_vp(dvp->v_mount, tnode,
100 				       LK_EXCLUSIVE | LK_RETRY, &vp);
101 		if (error)
102 			goto out;
103 		KKASSERT(vp);
104 	}
105 
106 out:
107 	/*
108 	 * Store the result of this lookup in the cache.  Avoid this if the
109 	 * request was for creation, as it does not improve timings on
110 	 * emprical tests.
111 	 */
112 	if (vp) {
113 		vn_unlock(vp);
114 		cache_setvp(v->a_nch, vp);
115 		vrele(vp);
116 	} else if (error == ENOENT) {
117 		cache_setvp(v->a_nch, NULL);
118 	}
119 	return error;
120 }
121 
122 static int
123 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
124 {
125 	struct vnode *dvp = v->a_dvp;
126 	struct vnode **vpp = v->a_vpp;
127 	struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
128 	struct ucred *cred = v->a_cred;
129 	int error;
130 
131 	*vpp = NULL;
132 	/* Check accessibility of requested node as a first step. */
133 	error = VOP_ACCESS(dvp, VEXEC, cred);
134 	if (error != 0)
135 		return error;
136 
137 	if (dnode->tn_dir.tn_parent != NULL) {
138 		/* Allocate a new vnode on the matching entry. */
139 		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
140 		    LK_EXCLUSIVE | LK_RETRY, vpp);
141 
142 		if (*vpp)
143 			vn_unlock(*vpp);
144 	}
145 
146 	return (*vpp == NULL) ? ENOENT : 0;
147 }
148 
149 /* --------------------------------------------------------------------- */
150 
151 static int
152 tmpfs_ncreate(struct vop_ncreate_args *v)
153 {
154 	struct vnode *dvp = v->a_dvp;
155 	struct vnode **vpp = v->a_vpp;
156 	struct namecache *ncp = v->a_nch->ncp;
157 	struct vattr *vap = v->a_vap;
158 	struct ucred *cred = v->a_cred;
159 	int error;
160 
161 	KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
162 
163 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
164 	if (error == 0) {
165 		cache_setunresolved(v->a_nch);
166 		cache_setvp(v->a_nch, *vpp);
167 		tmpfs_knote(dvp, NOTE_WRITE);
168 	}
169 
170 	return error;
171 }
172 /* --------------------------------------------------------------------- */
173 
174 static int
175 tmpfs_nmknod(struct vop_nmknod_args *v)
176 {
177 	struct vnode *dvp = v->a_dvp;
178 	struct vnode **vpp = v->a_vpp;
179 	struct namecache *ncp = v->a_nch->ncp;
180 	struct vattr *vap = v->a_vap;
181 	struct ucred *cred = v->a_cred;
182 	int error;
183 
184 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
185 	    vap->va_type != VFIFO)
186 		return EINVAL;
187 
188 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
189 	if (error == 0) {
190 		cache_setunresolved(v->a_nch);
191 		cache_setvp(v->a_nch, *vpp);
192 		tmpfs_knote(dvp, NOTE_WRITE);
193 	}
194 
195 	return error;
196 }
197 
198 /* --------------------------------------------------------------------- */
199 
200 static int
201 tmpfs_open(struct vop_open_args *v)
202 {
203 	struct vnode *vp = v->a_vp;
204 	int mode = v->a_mode;
205 
206 	int error;
207 	struct tmpfs_node *node;
208 
209 	node = VP_TO_TMPFS_NODE(vp);
210 
211 	/* The file is still active but all its names have been removed
212 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
213 	 * it is about to die. */
214 	if (node->tn_links < 1)
215 		return (ENOENT);
216 
217 	/* If the file is marked append-only, deny write requests. */
218 	if ((node->tn_flags & APPEND) &&
219 	    (mode & (FWRITE | O_APPEND)) == FWRITE) {
220 		error = EPERM;
221 	} else {
222 		return (vop_stdopen(v));
223 	}
224 	return error;
225 }
226 
227 /* --------------------------------------------------------------------- */
228 
229 static int
230 tmpfs_close(struct vop_close_args *v)
231 {
232 	struct vnode *vp = v->a_vp;
233 	struct tmpfs_node *node;
234 
235 	node = VP_TO_TMPFS_NODE(vp);
236 
237 	if (node->tn_links > 0) {
238 		/* Update node times.  No need to do it if the node has
239 		 * been deleted, because it will vanish after we return. */
240 		tmpfs_update(vp);
241 	}
242 
243 	return vop_stdclose(v);
244 }
245 
246 /* --------------------------------------------------------------------- */
247 
248 int
249 tmpfs_access(struct vop_access_args *v)
250 {
251 	struct vnode *vp = v->a_vp;
252 	int error;
253 	struct tmpfs_node *node;
254 
255 	node = VP_TO_TMPFS_NODE(vp);
256 
257 	switch (vp->v_type) {
258 	case VDIR:
259 		/* FALLTHROUGH */
260 	case VLNK:
261 		/* FALLTHROUGH */
262 	case VREG:
263 		if ((v->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
264 			error = EROFS;
265 			goto out;
266 		}
267 		break;
268 
269 	case VBLK:
270 		/* FALLTHROUGH */
271 	case VCHR:
272 		/* FALLTHROUGH */
273 	case VSOCK:
274 		/* FALLTHROUGH */
275 	case VFIFO:
276 		break;
277 
278 	default:
279 		error = EINVAL;
280 		goto out;
281 	}
282 
283 	if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
284 		error = EPERM;
285 		goto out;
286 	}
287 
288 	error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
289 
290 out:
291 
292 	return error;
293 }
294 
295 /* --------------------------------------------------------------------- */
296 
297 int
298 tmpfs_getattr(struct vop_getattr_args *v)
299 {
300 	struct vnode *vp = v->a_vp;
301 	struct vattr *vap = v->a_vap;
302 	struct tmpfs_node *node;
303 
304 	node = VP_TO_TMPFS_NODE(vp);
305 
306 	tmpfs_update(vp);
307 
308 	vap->va_type = vp->v_type;
309 	vap->va_mode = node->tn_mode;
310 	vap->va_nlink = node->tn_links;
311 	vap->va_uid = node->tn_uid;
312 	vap->va_gid = node->tn_gid;
313 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
314 	vap->va_fileid = node->tn_id;
315 	vap->va_size = node->tn_size;
316 	vap->va_blocksize = PAGE_SIZE;
317 	vap->va_atime.tv_sec = node->tn_atime;
318 	vap->va_atime.tv_nsec = node->tn_atimensec;
319 	vap->va_mtime.tv_sec = node->tn_mtime;
320 	vap->va_mtime.tv_nsec = node->tn_mtimensec;
321 	vap->va_ctime.tv_sec = node->tn_ctime;
322 	vap->va_ctime.tv_nsec = node->tn_ctimensec;
323 	vap->va_gen = node->tn_gen;
324 	vap->va_flags = node->tn_flags;
325 	if (vp->v_type == VBLK || vp->v_type == VCHR)
326 	{
327 		vap->va_rmajor = umajor(node->tn_rdev);
328 		vap->va_rminor = uminor(node->tn_rdev);
329 	}
330 	vap->va_bytes = round_page(node->tn_size);
331 	vap->va_filerev = 0;
332 
333 	return 0;
334 }
335 
336 /* --------------------------------------------------------------------- */
337 
338 int
339 tmpfs_setattr(struct vop_setattr_args *v)
340 {
341 	struct vnode *vp = v->a_vp;
342 	struct vattr *vap = v->a_vap;
343 	struct ucred *cred = v->a_cred;
344 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
345 	int error = 0;
346 	int kflags = 0;
347 
348 	if (error == 0 && (vap->va_flags != VNOVAL)) {
349 		error = tmpfs_chflags(vp, vap->va_flags, cred);
350 		kflags |= NOTE_ATTRIB;
351 	}
352 
353 	if (error == 0 && (vap->va_size != VNOVAL)) {
354 		if (vap->va_size > node->tn_size)
355 			kflags |= NOTE_WRITE | NOTE_EXTEND;
356 		else
357 			kflags |= NOTE_WRITE;
358 		error = tmpfs_chsize(vp, vap->va_size, cred);
359 	}
360 
361 	if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
362 			   vap->va_gid != (gid_t)VNOVAL)) {
363 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
364 		kflags |= NOTE_ATTRIB;
365 	}
366 
367 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
368 		error = tmpfs_chmod(vp, vap->va_mode, cred);
369 		kflags |= NOTE_ATTRIB;
370 	}
371 
372 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
373 	    vap->va_atime.tv_nsec != VNOVAL) ||
374 	    (vap->va_mtime.tv_sec != VNOVAL &&
375 	    vap->va_mtime.tv_nsec != VNOVAL) )) {
376 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
377 				      vap->va_vaflags, cred);
378 		kflags |= NOTE_ATTRIB;
379 	}
380 
381 	/* Update the node times.  We give preference to the error codes
382 	 * generated by this function rather than the ones that may arise
383 	 * from tmpfs_update. */
384 	tmpfs_update(vp);
385 	tmpfs_knote(vp, kflags);
386 
387 	return error;
388 }
389 
390 /* --------------------------------------------------------------------- */
391 
392 /*
393  * fsync is usually a NOP, but we must take action when unmounting or
394  * when recycling.
395  */
396 static int
397 tmpfs_fsync(struct vop_fsync_args *v)
398 {
399 	struct tmpfs_mount *tmp;
400 	struct tmpfs_node *node;
401 	struct vnode *vp = v->a_vp;
402 
403 	tmp = VFS_TO_TMPFS(vp->v_mount);
404 	node = VP_TO_TMPFS_NODE(vp);
405 
406 	tmpfs_update(vp);
407 	if (vp->v_type == VREG) {
408 		if (vp->v_flag & VRECLAIMED) {
409 			if (node->tn_links == 0)
410 				tmpfs_truncate(vp, 0);
411 			else
412 				vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
413 		}
414 	}
415 	return 0;
416 }
417 
418 /* --------------------------------------------------------------------- */
419 
420 static int
421 tmpfs_read (struct vop_read_args *ap)
422 {
423 	struct buf *bp;
424 	struct vnode *vp = ap->a_vp;
425 	struct uio *uio = ap->a_uio;
426 	struct tmpfs_node *node;
427 	off_t base_offset;
428 	size_t offset;
429 	size_t len;
430 	int error;
431 
432 	error = 0;
433 	if (uio->uio_resid == 0) {
434 		return error;
435 	}
436 
437 	node = VP_TO_TMPFS_NODE(vp);
438 
439 	if (uio->uio_offset < 0)
440 		return (EINVAL);
441 	if (vp->v_type != VREG)
442 		return (EINVAL);
443 
444 	while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
445 		/*
446 		 * Use buffer cache I/O (via tmpfs_strategy)
447 		 */
448 		offset = (size_t)uio->uio_offset & BMASK;
449 		base_offset = (off_t)uio->uio_offset - offset;
450 		bp = getcacheblk(vp, base_offset, BSIZE);
451 		if (bp == NULL)
452 		{
453 			lwkt_gettoken(&vp->v_mount->mnt_token);
454 			error = bread(vp, base_offset, BSIZE, &bp);
455 			if (error) {
456 				brelse(bp);
457 				lwkt_reltoken(&vp->v_mount->mnt_token);
458 				kprintf("tmpfs_read bread error %d\n", error);
459 				break;
460 			}
461 			lwkt_reltoken(&vp->v_mount->mnt_token);
462 		}
463 
464 		/*
465 		 * Figure out how many bytes we can actually copy this loop.
466 		 */
467 		len = BSIZE - offset;
468 		if (len > uio->uio_resid)
469 			len = uio->uio_resid;
470 		if (len > node->tn_size - uio->uio_offset)
471 			len = (size_t)(node->tn_size - uio->uio_offset);
472 
473 		error = uiomove((char *)bp->b_data + offset, len, uio);
474 		bqrelse(bp);
475 		if (error) {
476 			kprintf("tmpfs_read uiomove error %d\n", error);
477 			break;
478 		}
479 	}
480 
481 	TMPFS_NODE_LOCK(node);
482 	node->tn_status |= TMPFS_NODE_ACCESSED;
483 	TMPFS_NODE_UNLOCK(node);
484 
485 	return(error);
486 }
487 
488 static int
489 tmpfs_write (struct vop_write_args *ap)
490 {
491 	struct buf *bp;
492 	struct vnode *vp = ap->a_vp;
493 	struct uio *uio = ap->a_uio;
494 	struct thread *td = uio->uio_td;
495 	struct tmpfs_node *node;
496 	boolean_t extended;
497 	off_t oldsize;
498 	int error;
499 	off_t base_offset;
500 	size_t offset;
501 	size_t len;
502 	struct rlimit limit;
503 	int trivial = 0;
504 	int kflags = 0;
505 
506 	error = 0;
507 	if (uio->uio_resid == 0) {
508 		return error;
509 	}
510 
511 	node = VP_TO_TMPFS_NODE(vp);
512 
513 	if (vp->v_type != VREG)
514 		return (EINVAL);
515 
516 	oldsize = node->tn_size;
517 	if (ap->a_ioflag & IO_APPEND)
518 		uio->uio_offset = node->tn_size;
519 
520 	/*
521 	 * Check for illegal write offsets.
522 	 */
523 	if (uio->uio_offset + uio->uio_resid >
524 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
525 		return (EFBIG);
526 
527 	if (vp->v_type == VREG && td != NULL) {
528 		error = kern_getrlimit(RLIMIT_FSIZE, &limit);
529 		if (error != 0)
530 			return error;
531 		if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
532 			ksignal(td->td_proc, SIGXFSZ);
533 			return (EFBIG);
534 		}
535 	}
536 
537 
538 	/*
539 	 * Extend the file's size if necessary
540 	 */
541 	extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
542 
543 	get_mplock();
544 
545 	while (uio->uio_resid > 0) {
546 		/*
547 		 * Use buffer cache I/O (via tmpfs_strategy)
548 		 */
549 		offset = (size_t)uio->uio_offset & BMASK;
550 		base_offset = (off_t)uio->uio_offset - offset;
551 		len = BSIZE - offset;
552 		if (len > uio->uio_resid)
553 			len = uio->uio_resid;
554 
555 		if ((uio->uio_offset + len) > node->tn_size) {
556 			trivial = (uio->uio_offset <= node->tn_size);
557 			error = tmpfs_reg_resize(vp, uio->uio_offset + len,  trivial);
558 			if (error)
559 				break;
560 		}
561 
562 		/*
563 		 * Read to fill in any gaps.  Theoretically we could
564 		 * optimize this if the write covers the entire buffer
565 		 * and is not a UIO_NOCOPY write, however this can lead
566 		 * to a security violation exposing random kernel memory
567 		 * (whatever junk was in the backing VM pages before).
568 		 *
569 		 * So just use bread() to do the right thing.
570 		 */
571 		error = bread(vp, base_offset, BSIZE, &bp);
572 		error = uiomove((char *)bp->b_data + offset, len, uio);
573 		if (error) {
574 			kprintf("tmpfs_write uiomove error %d\n", error);
575 			brelse(bp);
576 			break;
577 		}
578 
579 		if (uio->uio_offset > node->tn_size) {
580 			node->tn_size = uio->uio_offset;
581 			kflags |= NOTE_EXTEND;
582 		}
583 		kflags |= NOTE_WRITE;
584 
585 		/*
586 		 * The data has been loaded into the buffer, write it out.
587 		 *
588 		 * We want tmpfs to be able to use all available ram, not
589 		 * just the buffer cache, so if not explicitly paging we
590 		 * use buwrite() to leave the buffer clean but mark all the
591 		 * VM pages valid+dirty.
592 		 *
593 		 * When the kernel is paging, either via normal pageout
594 		 * operation or when cleaning the object during a recycle,
595 		 * the underlying VM pages are going to get thrown away
596 		 * so we MUST write them to swap.
597 		 *
598 		 * XXX unfortunately this catches msync() system calls too
599 		 * for the moment.
600 		 */
601 		if (vm_swap_size == 0) {
602 			/*
603 			 * if swap isn't configured yet, force a buwrite() to
604 			 * avoid problems further down the line, due to flushing
605 			 * to swap.
606 			 */
607 			buwrite(bp);
608 		} else {
609 			if (ap->a_ioflag & IO_SYNC) {
610 				bwrite(bp);
611 			} else if ((ap->a_ioflag & IO_ASYNC) ||
612 				 (uio->uio_segflg == UIO_NOCOPY)) {
613 				bawrite(bp);
614 			} else {
615 				buwrite(bp);
616 			}
617 		}
618 
619 		if (bp->b_error) {
620 			kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
621 			break;
622 		}
623 	}
624 
625 	rel_mplock();
626 
627 	if (error) {
628 		if (extended) {
629 			(void)tmpfs_reg_resize(vp, oldsize, trivial);
630 			kflags &= ~NOTE_EXTEND;
631 		}
632 		goto done;
633 	}
634 
635 	TMPFS_NODE_LOCK(node);
636 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
637 	    (extended? TMPFS_NODE_CHANGED : 0);
638 
639 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
640 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
641 			node->tn_mode &= ~(S_ISUID | S_ISGID);
642 	}
643 	TMPFS_NODE_UNLOCK(node);
644 done:
645 
646 	tmpfs_knote(vp, kflags);
647 	return(error);
648 }
649 
650 static int
651 tmpfs_advlock (struct vop_advlock_args *ap)
652 {
653 	struct tmpfs_node *node;
654 	struct vnode *vp = ap->a_vp;
655 
656 	node = VP_TO_TMPFS_NODE(vp);
657 
658 	return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
659 }
660 
661 static int
662 tmpfs_strategy(struct vop_strategy_args *ap)
663 {
664 	struct bio *bio = ap->a_bio;
665 	struct buf *bp = bio->bio_buf;
666 	struct vnode *vp = ap->a_vp;
667 	struct tmpfs_node *node;
668 	vm_object_t uobj;
669 
670 	if (vp->v_type != VREG) {
671 		bp->b_resid = bp->b_bcount;
672 		bp->b_flags |= B_ERROR | B_INVAL;
673 		bp->b_error = EINVAL;
674 		biodone(bio);
675 		return(0);
676 	}
677 
678 	lwkt_gettoken(&vp->v_mount->mnt_token);
679 	node = VP_TO_TMPFS_NODE(vp);
680 
681 	uobj = node->tn_reg.tn_aobj;
682 
683 	/*
684 	 * Call swap_pager_strategy to read or write between the VM
685 	 * object and the buffer cache.
686 	 */
687 	swap_pager_strategy(uobj, bio);
688 
689 	lwkt_reltoken(&vp->v_mount->mnt_token);
690 	return 0;
691 }
692 
693 static int
694 tmpfs_bmap(struct vop_bmap_args *ap)
695 {
696 	if (ap->a_doffsetp != NULL)
697 		*ap->a_doffsetp = ap->a_loffset;
698 	if (ap->a_runp != NULL)
699 		*ap->a_runp = 0;
700 	if (ap->a_runb != NULL)
701 		*ap->a_runb = 0;
702 
703 	return 0;
704 }
705 
706 /* --------------------------------------------------------------------- */
707 
708 static int
709 tmpfs_nremove(struct vop_nremove_args *v)
710 {
711 	struct vnode *dvp = v->a_dvp;
712 	struct namecache *ncp = v->a_nch->ncp;
713 	struct vnode *vp;
714 	int error;
715 	struct tmpfs_dirent *de;
716 	struct tmpfs_mount *tmp;
717 	struct tmpfs_node *dnode;
718 	struct tmpfs_node *node;
719 
720 	/*
721 	 * We have to acquire the vp from v->a_nch because
722 	 * we will likely unresolve the namecache entry, and
723 	 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
724 	 * sequence to recover space from the file.
725 	 */
726 	error = cache_vref(v->a_nch, v->a_cred, &vp);
727 	KKASSERT(error == 0);
728 
729 	if (vp->v_type == VDIR) {
730 		error = EISDIR;
731 		goto out;
732 	}
733 
734 	dnode = VP_TO_TMPFS_DIR(dvp);
735 	node = VP_TO_TMPFS_NODE(vp);
736 	tmp = VFS_TO_TMPFS(vp->v_mount);
737 	de = tmpfs_dir_lookup(dnode, node, ncp);
738 	if (de == NULL) {
739 		error = ENOENT;
740 		goto out;
741 	}
742 
743 	/* Files marked as immutable or append-only cannot be deleted. */
744 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
745 	    (dnode->tn_flags & APPEND)) {
746 		error = EPERM;
747 		goto out;
748 	}
749 
750 	/* Remove the entry from the directory; as it is a file, we do not
751 	 * have to change the number of hard links of the directory. */
752 	tmpfs_dir_detach(dnode, de);
753 
754 	/* Free the directory entry we just deleted.  Note that the node
755 	 * referred by it will not be removed until the vnode is really
756 	 * reclaimed. */
757 	tmpfs_free_dirent(tmp, de);
758 
759 	if (node->tn_links > 0) {
760 	        TMPFS_NODE_LOCK(node);
761 		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
762 	                TMPFS_NODE_MODIFIED;
763 	        TMPFS_NODE_UNLOCK(node);
764 	}
765 
766 	cache_setunresolved(v->a_nch);
767 	cache_setvp(v->a_nch, NULL);
768 	tmpfs_knote(vp, NOTE_DELETE);
769 	/*cache_inval_vp(vp, CINV_DESTROY);*/
770 	tmpfs_knote(dvp, NOTE_WRITE);
771 	error = 0;
772 
773 out:
774 	vrele(vp);
775 
776 	return error;
777 }
778 
779 /* --------------------------------------------------------------------- */
780 
781 static int
782 tmpfs_nlink(struct vop_nlink_args *v)
783 {
784 	struct vnode *dvp = v->a_dvp;
785 	struct vnode *vp = v->a_vp;
786 	struct namecache *ncp = v->a_nch->ncp;
787 	struct tmpfs_dirent *de;
788 	struct tmpfs_node *node;
789 	struct tmpfs_node *dnode;
790 	int error;
791 
792 	KKASSERT(dvp != vp); /* XXX When can this be false? */
793 
794 	node = VP_TO_TMPFS_NODE(vp);
795 	dnode = VP_TO_TMPFS_NODE(dvp);
796 
797 	/* XXX: Why aren't the following two tests done by the caller? */
798 
799 	/* Hard links of directories are forbidden. */
800 	if (vp->v_type == VDIR) {
801 		error = EPERM;
802 		goto out;
803 	}
804 
805 	/* Cannot create cross-device links. */
806 	if (dvp->v_mount != vp->v_mount) {
807 		error = EXDEV;
808 		goto out;
809 	}
810 
811 	/* Ensure that we do not overflow the maximum number of links imposed
812 	 * by the system. */
813 	KKASSERT(node->tn_links <= LINK_MAX);
814 	if (node->tn_links == LINK_MAX) {
815 		error = EMLINK;
816 		goto out;
817 	}
818 
819 	/* We cannot create links of files marked immutable or append-only. */
820 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
821 		error = EPERM;
822 		goto out;
823 	}
824 
825 	/* Allocate a new directory entry to represent the node. */
826 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
827 	    ncp->nc_name, ncp->nc_nlen, &de);
828 	if (error != 0)
829 		goto out;
830 
831 	/* Insert the new directory entry into the appropriate directory. */
832 	tmpfs_dir_attach(dnode, de);
833 
834 	/* vp link count has changed, so update node times. */
835 
836 	TMPFS_NODE_LOCK(node);
837 	node->tn_status |= TMPFS_NODE_CHANGED;
838 	TMPFS_NODE_UNLOCK(node);
839 	tmpfs_update(vp);
840 
841 	tmpfs_knote(vp, NOTE_LINK);
842 	cache_setunresolved(v->a_nch);
843 	cache_setvp(v->a_nch, vp);
844 	tmpfs_knote(dvp, NOTE_WRITE);
845 	error = 0;
846 
847 out:
848 	return error;
849 }
850 
851 /* --------------------------------------------------------------------- */
852 
853 static int
854 tmpfs_nrename(struct vop_nrename_args *v)
855 {
856 	struct vnode *fdvp = v->a_fdvp;
857 	struct namecache *fncp = v->a_fnch->ncp;
858 	struct vnode *fvp = fncp->nc_vp;
859 	struct vnode *tdvp = v->a_tdvp;
860 	struct namecache *tncp = v->a_tnch->ncp;
861 	struct vnode *tvp = tncp->nc_vp;
862 	struct tmpfs_dirent *de;
863 	struct tmpfs_mount *tmp;
864 	struct tmpfs_node *fdnode;
865 	struct tmpfs_node *fnode;
866 	struct tmpfs_node *tnode;
867 	struct tmpfs_node *tdnode;
868 	char *newname;
869 	char *oldname;
870 	int error;
871 
872 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
873 
874 	/* Disallow cross-device renames.
875 	 * XXX Why isn't this done by the caller? */
876 	if (fvp->v_mount != tdvp->v_mount ||
877 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
878 		error = EXDEV;
879 		goto out;
880 	}
881 
882 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
883 	tdnode = VP_TO_TMPFS_DIR(tdvp);
884 
885 	/* If source and target are the same file, there is nothing to do. */
886 	if (fvp == tvp) {
887 		error = 0;
888 		goto out;
889 	}
890 
891 	fdnode = VP_TO_TMPFS_DIR(fdvp);
892 	fnode = VP_TO_TMPFS_NODE(fvp);
893 	de = tmpfs_dir_lookup(fdnode, fnode, fncp);
894 
895 	/* Avoid manipulating '.' and '..' entries. */
896 	if (de == NULL) {
897 		error = ENOENT;
898 		goto out_locked;
899 	}
900 	KKASSERT(de->td_node == fnode);
901 
902 	/*
903 	 * If replacing an entry in the target directory and that entry
904 	 * is a directory, it must be empty.
905 	 *
906 	 * Kern_rename gurantees the destination to be a directory
907 	 * if the source is one (it does?).
908 	 */
909 	if (tvp != NULL) {
910 		KKASSERT(tnode != NULL);
911 
912 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
913 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
914 			error = EPERM;
915 			goto out_locked;
916 		}
917 
918 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
919 			if (tnode->tn_size > 0) {
920 				error = ENOTEMPTY;
921 				goto out_locked;
922 			}
923 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
924 			error = ENOTDIR;
925 			goto out_locked;
926 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
927 			error = EISDIR;
928 			goto out_locked;
929 		} else {
930 			KKASSERT(fnode->tn_type != VDIR &&
931 				tnode->tn_type != VDIR);
932 		}
933 	}
934 
935 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
936 	    (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
937 		error = EPERM;
938 		goto out_locked;
939 	}
940 
941 	/*
942 	 * Ensure that we have enough memory to hold the new name, if it
943 	 * has to be changed.
944 	 */
945 	if (fncp->nc_nlen != tncp->nc_nlen ||
946 	    bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
947 		newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
948 				  M_WAITOK | M_NULLOK);
949 		if (newname == NULL) {
950 			error = ENOSPC;
951 			goto out_locked;
952 		}
953 		bcopy(tncp->nc_name, newname, tncp->nc_nlen);
954 		newname[tncp->nc_nlen] = '\0';
955 	} else {
956 		newname = NULL;
957 	}
958 
959 	/*
960 	 * Unlink entry from source directory.  Note that the kernel has
961 	 * already checked for illegal recursion cases (renaming a directory
962 	 * into a subdirectory of itself).
963 	 */
964 	if (fdnode != tdnode)
965 		tmpfs_dir_detach(fdnode, de);
966 
967 	/*
968 	 * Handle any name change.  Swap with newname, we will
969 	 * deallocate it at the end.
970 	 */
971 	if (newname != NULL) {
972 #if 0
973 		TMPFS_NODE_LOCK(fnode);
974 		fnode->tn_status |= TMPFS_NODE_CHANGED;
975 		TMPFS_NODE_UNLOCK(fnode);
976 #endif
977 		oldname = de->td_name;
978 		de->td_name = newname;
979 		de->td_namelen = (uint16_t)tncp->nc_nlen;
980 		newname = oldname;
981 	}
982 
983 	/*
984 	 * Link entry to target directory.  If the entry
985 	 * represents a directory move the parent linkage
986 	 * as well.
987 	 */
988 	if (fdnode != tdnode) {
989 		if (de->td_node->tn_type == VDIR) {
990 			TMPFS_VALIDATE_DIR(fnode);
991 
992 			TMPFS_NODE_LOCK(tdnode);
993 			tdnode->tn_links++;
994 			tdnode->tn_status |= TMPFS_NODE_MODIFIED;
995 			TMPFS_NODE_UNLOCK(tdnode);
996 
997 			TMPFS_NODE_LOCK(fnode);
998 			fnode->tn_dir.tn_parent = tdnode;
999 			fnode->tn_status |= TMPFS_NODE_CHANGED;
1000 			TMPFS_NODE_UNLOCK(fnode);
1001 
1002 			TMPFS_NODE_LOCK(fdnode);
1003 			fdnode->tn_links--;
1004 			fdnode->tn_status |= TMPFS_NODE_MODIFIED;
1005 			TMPFS_NODE_UNLOCK(fdnode);
1006 		}
1007 		tmpfs_dir_attach(tdnode, de);
1008 	} else {
1009 		TMPFS_NODE_LOCK(tdnode);
1010 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1011 		TMPFS_NODE_UNLOCK(tdnode);
1012 	}
1013 
1014 	/*
1015 	 * If we are overwriting an entry, we have to remove the old one
1016 	 * from the target directory.
1017 	 */
1018 	if (tvp != NULL) {
1019 		/* Remove the old entry from the target directory. */
1020 		de = tmpfs_dir_lookup(tdnode, tnode, tncp);
1021 		tmpfs_dir_detach(tdnode, de);
1022 		tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1023 
1024 		/*
1025 		 * Free the directory entry we just deleted.  Note that the
1026 		 * node referred by it will not be removed until the vnode is
1027 		 * really reclaimed.
1028 		 */
1029 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de);
1030 		/*cache_inval_vp(tvp, CINV_DESTROY);*/
1031 	}
1032 
1033 	/*
1034 	 * Finish up
1035 	 */
1036 	if (newname) {
1037 		kfree(newname, tmp->tm_name_zone);
1038 		newname = NULL;
1039 	}
1040 	cache_rename(v->a_fnch, v->a_tnch);
1041 	tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1042 	tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1043 	if (fnode->tn_vnode)
1044 		tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1045 	error = 0;
1046 
1047 out_locked:
1048 	;
1049 
1050 out:
1051 	/* Release target nodes. */
1052 	/* XXX: I don't understand when tdvp can be the same as tvp, but
1053 	 * other code takes care of this... */
1054 	if (tdvp == tvp)
1055 		vrele(tdvp);
1056 
1057 	return error;
1058 }
1059 
1060 /* --------------------------------------------------------------------- */
1061 
1062 static int
1063 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1064 {
1065 	struct vnode *dvp = v->a_dvp;
1066 	struct vnode **vpp = v->a_vpp;
1067 	struct namecache *ncp = v->a_nch->ncp;
1068 	struct vattr *vap = v->a_vap;
1069 	struct ucred *cred = v->a_cred;
1070 	int error;
1071 
1072 	KKASSERT(vap->va_type == VDIR);
1073 
1074 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1075 	if (error == 0) {
1076 		cache_setunresolved(v->a_nch);
1077 		cache_setvp(v->a_nch, *vpp);
1078 		tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1079 	}
1080 
1081 	return error;
1082 }
1083 
1084 /* --------------------------------------------------------------------- */
1085 
1086 static int
1087 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1088 {
1089 	struct vnode *dvp = v->a_dvp;
1090 	struct namecache *ncp = v->a_nch->ncp;
1091 	struct vnode *vp;
1092 	struct tmpfs_dirent *de;
1093 	struct tmpfs_mount *tmp;
1094 	struct tmpfs_node *dnode;
1095 	struct tmpfs_node *node;
1096 	int error;
1097 
1098 	/*
1099 	 * We have to acquire the vp from v->a_nch because
1100 	 * we will likely unresolve the namecache entry, and
1101 	 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
1102 	 * sequence.
1103 	 */
1104 	error = cache_vref(v->a_nch, v->a_cred, &vp);
1105 	KKASSERT(error == 0);
1106 
1107 	/*
1108 	 * Prevalidate so we don't hit an assertion later
1109 	 */
1110 	if (vp->v_type != VDIR) {
1111 		error = ENOTDIR;
1112 		goto out;
1113 	}
1114 
1115 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1116 	dnode = VP_TO_TMPFS_DIR(dvp);
1117 	node = VP_TO_TMPFS_DIR(vp);
1118 
1119 	/* Directories with more than two entries ('.' and '..') cannot be
1120 	 * removed. */
1121 	 if (node->tn_size > 0) {
1122 		 error = ENOTEMPTY;
1123 		 goto out;
1124 	 }
1125 
1126 	if ((dnode->tn_flags & APPEND)
1127 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1128 		error = EPERM;
1129 		goto out;
1130 	}
1131 
1132 	/* This invariant holds only if we are not trying to remove "..".
1133 	  * We checked for that above so this is safe now. */
1134 	KKASSERT(node->tn_dir.tn_parent == dnode);
1135 
1136 	/* Get the directory entry associated with node (vp).  This was
1137 	 * filled by tmpfs_lookup while looking up the entry. */
1138 	de = tmpfs_dir_lookup(dnode, node, ncp);
1139 	KKASSERT(TMPFS_DIRENT_MATCHES(de,
1140 	    ncp->nc_name,
1141 	    ncp->nc_nlen));
1142 
1143 	/* Check flags to see if we are allowed to remove the directory. */
1144 	if ((dnode->tn_flags & APPEND) ||
1145 	    node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1146 		error = EPERM;
1147 		goto out;
1148 	}
1149 
1150 
1151 	/* Detach the directory entry from the directory (dnode). */
1152 	tmpfs_dir_detach(dnode, de);
1153 
1154 	/* No vnode should be allocated for this entry from this point */
1155 	TMPFS_NODE_LOCK(node);
1156 	TMPFS_ASSERT_ELOCKED(node);
1157 	TMPFS_NODE_LOCK(dnode);
1158 	TMPFS_ASSERT_ELOCKED(dnode);
1159 
1160 #if 0
1161 	/* handled by tmpfs_free_node */
1162 	KKASSERT(node->tn_links > 0);
1163 	node->tn_links--;
1164 	node->tn_dir.tn_parent = NULL;
1165 #endif
1166 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1167 	    TMPFS_NODE_MODIFIED;
1168 
1169 #if 0
1170 	/* handled by tmpfs_free_node */
1171 	KKASSERT(dnode->tn_links > 0);
1172 	dnode->tn_links--;
1173 #endif
1174 	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1175 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1176 
1177 	TMPFS_NODE_UNLOCK(dnode);
1178 	TMPFS_NODE_UNLOCK(node);
1179 
1180 	/* Free the directory entry we just deleted.  Note that the node
1181 	 * referred by it will not be removed until the vnode is really
1182 	 * reclaimed. */
1183 	tmpfs_free_dirent(tmp, de);
1184 
1185 	/* Release the deleted vnode (will destroy the node, notify
1186 	 * interested parties and clean it from the cache). */
1187 
1188 	TMPFS_NODE_LOCK(dnode);
1189 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1190 	TMPFS_NODE_UNLOCK(dnode);
1191 	tmpfs_update(dvp);
1192 
1193 	cache_setunresolved(v->a_nch);
1194 	cache_setvp(v->a_nch, NULL);
1195 	/*cache_inval_vp(vp, CINV_DESTROY);*/
1196 	tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1197 	error = 0;
1198 
1199 out:
1200 	vrele(vp);
1201 
1202 	return error;
1203 }
1204 
1205 /* --------------------------------------------------------------------- */
1206 
1207 static int
1208 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1209 {
1210 	struct vnode *dvp = v->a_dvp;
1211 	struct vnode **vpp = v->a_vpp;
1212 	struct namecache *ncp = v->a_nch->ncp;
1213 	struct vattr *vap = v->a_vap;
1214 	struct ucred *cred = v->a_cred;
1215 	char *target = v->a_target;
1216 	int error;
1217 
1218 	vap->va_type = VLNK;
1219 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1220 	if (error == 0) {
1221 		tmpfs_knote(*vpp, NOTE_WRITE);
1222 		cache_setunresolved(v->a_nch);
1223 		cache_setvp(v->a_nch, *vpp);
1224 	}
1225 
1226 	return error;
1227 }
1228 
1229 /* --------------------------------------------------------------------- */
1230 
1231 static int
1232 tmpfs_readdir(struct vop_readdir_args *v)
1233 {
1234 	struct vnode *vp = v->a_vp;
1235 	struct uio *uio = v->a_uio;
1236 	int *eofflag = v->a_eofflag;
1237 	off_t **cookies = v->a_cookies;
1238 	int *ncookies = v->a_ncookies;
1239 	struct tmpfs_mount *tmp;
1240 	int error;
1241 	off_t startoff;
1242 	off_t cnt = 0;
1243 	struct tmpfs_node *node;
1244 
1245 	/* This operation only makes sense on directory nodes. */
1246 	if (vp->v_type != VDIR)
1247 		return ENOTDIR;
1248 
1249 	tmp = VFS_TO_TMPFS(vp->v_mount);
1250 	node = VP_TO_TMPFS_DIR(vp);
1251 	startoff = uio->uio_offset;
1252 
1253 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1254 		error = tmpfs_dir_getdotdent(node, uio);
1255 		if (error != 0)
1256 			goto outok;
1257 		cnt++;
1258 	}
1259 
1260 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1261 		error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1262 		if (error != 0)
1263 			goto outok;
1264 		cnt++;
1265 	}
1266 
1267 	error = tmpfs_dir_getdents(node, uio, &cnt);
1268 
1269 outok:
1270 	KKASSERT(error >= -1);
1271 
1272 	if (error == -1)
1273 		error = 0;
1274 
1275 	if (eofflag != NULL)
1276 		*eofflag =
1277 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1278 
1279 	/* Update NFS-related variables. */
1280 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1281 		off_t i;
1282 		off_t off = startoff;
1283 		struct tmpfs_dirent *de = NULL;
1284 
1285 		*ncookies = cnt;
1286 		*cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1287 
1288 		for (i = 0; i < cnt; i++) {
1289 			KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1290 			if (off == TMPFS_DIRCOOKIE_DOT) {
1291 				off = TMPFS_DIRCOOKIE_DOTDOT;
1292 			} else {
1293 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1294 					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1295 				} else if (de != NULL) {
1296 					de = TAILQ_NEXT(de, td_entries);
1297 				} else {
1298 					de = tmpfs_dir_lookupbycookie(node,
1299 					    off);
1300 					KKASSERT(de != NULL);
1301 					de = TAILQ_NEXT(de, td_entries);
1302 				}
1303 				if (de == NULL)
1304 					off = TMPFS_DIRCOOKIE_EOF;
1305 				else
1306 					off = tmpfs_dircookie(de);
1307 			}
1308 
1309 			(*cookies)[i] = off;
1310 		}
1311 		KKASSERT(uio->uio_offset == off);
1312 	}
1313 
1314 	return error;
1315 }
1316 
1317 /* --------------------------------------------------------------------- */
1318 
1319 static int
1320 tmpfs_readlink(struct vop_readlink_args *v)
1321 {
1322 	struct vnode *vp = v->a_vp;
1323 	struct uio *uio = v->a_uio;
1324 
1325 	int error;
1326 	struct tmpfs_node *node;
1327 
1328 	KKASSERT(uio->uio_offset == 0);
1329 	KKASSERT(vp->v_type == VLNK);
1330 
1331 	node = VP_TO_TMPFS_NODE(vp);
1332 
1333 	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1334 	    uio);
1335 	TMPFS_NODE_LOCK(node);
1336 	node->tn_status |= TMPFS_NODE_ACCESSED;
1337 	TMPFS_NODE_UNLOCK(node);
1338 
1339 	return error;
1340 }
1341 
1342 /* --------------------------------------------------------------------- */
1343 
1344 static int
1345 tmpfs_inactive(struct vop_inactive_args *v)
1346 {
1347 	struct vnode *vp = v->a_vp;
1348 
1349 	struct tmpfs_node *node;
1350 
1351 	node = VP_TO_TMPFS_NODE(vp);
1352 
1353 	/*
1354 	 * Get rid of unreferenced deleted vnodes sooner rather than
1355 	 * later so the data memory can be recovered immediately.
1356 	 *
1357 	 * We must truncate the vnode to prevent the normal reclamation
1358 	 * path from flushing the data for the removed file to disk.
1359 	 */
1360 	TMPFS_NODE_LOCK(node);
1361 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1362 	    (node->tn_links == 0 ||
1363 	     (node->tn_links == 1 && node->tn_type == VDIR &&
1364 	      node->tn_dir.tn_parent)))
1365 	{
1366 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1367 		TMPFS_NODE_UNLOCK(node);
1368 		if (node->tn_type == VREG)
1369 			tmpfs_truncate(vp, 0);
1370 		vrecycle(vp);
1371 	} else {
1372 		TMPFS_NODE_UNLOCK(node);
1373 	}
1374 
1375 	return 0;
1376 }
1377 
1378 /* --------------------------------------------------------------------- */
1379 
1380 int
1381 tmpfs_reclaim(struct vop_reclaim_args *v)
1382 {
1383 	struct vnode *vp = v->a_vp;
1384 	struct tmpfs_mount *tmp;
1385 	struct tmpfs_node *node;
1386 
1387 	node = VP_TO_TMPFS_NODE(vp);
1388 	tmp = VFS_TO_TMPFS(vp->v_mount);
1389 
1390 	tmpfs_free_vp(vp);
1391 
1392 	/*
1393 	 * If the node referenced by this vnode was deleted by the
1394 	 * user, we must free its associated data structures now that
1395 	 * the vnode is being reclaimed.
1396 	 *
1397 	 * Directories have an extra link ref.
1398 	 */
1399 	TMPFS_NODE_LOCK(node);
1400 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1401 	    (node->tn_links == 0 ||
1402 	     (node->tn_links == 1 && node->tn_type == VDIR &&
1403 	      node->tn_dir.tn_parent)))
1404 	{
1405 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1406 		tmpfs_free_node(tmp, node);
1407 		/* eats the lock */
1408 	} else {
1409 		TMPFS_NODE_UNLOCK(node);
1410 	}
1411 
1412 	KKASSERT(vp->v_data == NULL);
1413 	return 0;
1414 }
1415 
1416 /* --------------------------------------------------------------------- */
1417 
1418 static int
1419 tmpfs_print(struct vop_print_args *v)
1420 {
1421 	struct vnode *vp = v->a_vp;
1422 
1423 	struct tmpfs_node *node;
1424 
1425 	node = VP_TO_TMPFS_NODE(vp);
1426 
1427 	kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1428 	    node, node->tn_flags, node->tn_links);
1429 	kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1430 	    node->tn_mode, node->tn_uid, node->tn_gid,
1431 	    (uintmax_t)node->tn_size, node->tn_status);
1432 
1433 	if (vp->v_type == VFIFO)
1434 		fifo_printinfo(vp);
1435 
1436 	kprintf("\n");
1437 
1438 	return 0;
1439 }
1440 
1441 /* --------------------------------------------------------------------- */
1442 
1443 static int
1444 tmpfs_pathconf(struct vop_pathconf_args *v)
1445 {
1446 	int name = v->a_name;
1447 	register_t *retval = v->a_retval;
1448 
1449 	int error;
1450 
1451 	error = 0;
1452 
1453 	switch (name) {
1454 	case _PC_LINK_MAX:
1455 		*retval = LINK_MAX;
1456 		break;
1457 
1458 	case _PC_NAME_MAX:
1459 		*retval = NAME_MAX;
1460 		break;
1461 
1462 	case _PC_PATH_MAX:
1463 		*retval = PATH_MAX;
1464 		break;
1465 
1466 	case _PC_PIPE_BUF:
1467 		*retval = PIPE_BUF;
1468 		break;
1469 
1470 	case _PC_CHOWN_RESTRICTED:
1471 		*retval = 1;
1472 		break;
1473 
1474 	case _PC_NO_TRUNC:
1475 		*retval = 1;
1476 		break;
1477 
1478 	case _PC_SYNC_IO:
1479 		*retval = 1;
1480 		break;
1481 
1482 	case _PC_FILESIZEBITS:
1483 		*retval = 0; /* XXX Don't know which value should I return. */
1484 		break;
1485 
1486 	default:
1487 		error = EINVAL;
1488 	}
1489 
1490 	return error;
1491 }
1492 
1493 /************************************************************************
1494  *                          KQFILTER OPS                                *
1495  ************************************************************************/
1496 
1497 static void filt_tmpfsdetach(struct knote *kn);
1498 static int filt_tmpfsread(struct knote *kn, long hint);
1499 static int filt_tmpfswrite(struct knote *kn, long hint);
1500 static int filt_tmpfsvnode(struct knote *kn, long hint);
1501 
1502 static struct filterops tmpfsread_filtops =
1503 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1504 static struct filterops tmpfswrite_filtops =
1505 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1506 static struct filterops tmpfsvnode_filtops =
1507 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1508 
1509 static int
1510 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1511 {
1512 	struct vnode *vp = ap->a_vp;
1513 	struct knote *kn = ap->a_kn;
1514 
1515 	switch (kn->kn_filter) {
1516 	case EVFILT_READ:
1517 		kn->kn_fop = &tmpfsread_filtops;
1518 		break;
1519 	case EVFILT_WRITE:
1520 		kn->kn_fop = &tmpfswrite_filtops;
1521 		break;
1522 	case EVFILT_VNODE:
1523 		kn->kn_fop = &tmpfsvnode_filtops;
1524 		break;
1525 	default:
1526 		return (EOPNOTSUPP);
1527 	}
1528 
1529 	kn->kn_hook = (caddr_t)vp;
1530 
1531 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1532 
1533 	return(0);
1534 }
1535 
1536 static void
1537 filt_tmpfsdetach(struct knote *kn)
1538 {
1539 	struct vnode *vp = (void *)kn->kn_hook;
1540 
1541 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1542 }
1543 
1544 static int
1545 filt_tmpfsread(struct knote *kn, long hint)
1546 {
1547 	struct vnode *vp = (void *)kn->kn_hook;
1548 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1549 	off_t off;
1550 
1551 	if (hint == NOTE_REVOKE) {
1552 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1553 		return(1);
1554 	}
1555 	off = node->tn_size - kn->kn_fp->f_offset;
1556 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1557 	if (kn->kn_sfflags & NOTE_OLDAPI)
1558 		return(1);
1559 
1560 	/*
1561 	 * Handle possible MP race interlock on filter check/write
1562 	 */
1563 	if (kn->kn_data == 0) {
1564 		get_mplock();
1565 		kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1566 		rel_mplock();
1567 	}
1568 	return (kn->kn_data != 0);
1569 }
1570 
1571 static int
1572 filt_tmpfswrite(struct knote *kn, long hint)
1573 {
1574 	if (hint == NOTE_REVOKE)
1575 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1576 	kn->kn_data = 0;
1577 	return (1);
1578 }
1579 
1580 static int
1581 filt_tmpfsvnode(struct knote *kn, long hint)
1582 {
1583 	if (kn->kn_sfflags & hint)
1584 		kn->kn_fflags |= hint;
1585 	if (hint == NOTE_REVOKE) {
1586 		kn->kn_flags |= EV_EOF;
1587 		return (1);
1588 	}
1589 	return (kn->kn_fflags != 0);
1590 }
1591 
1592 
1593 /* --------------------------------------------------------------------- */
1594 
1595 /*
1596  * vnode operations vector used for files stored in a tmpfs file system.
1597  */
1598 struct vop_ops tmpfs_vnode_vops = {
1599 	.vop_default =			vop_defaultop,
1600 	.vop_getpages = 		vop_stdgetpages,
1601 	.vop_putpages = 		vop_stdputpages,
1602 	.vop_ncreate =			tmpfs_ncreate,
1603 	.vop_nresolve =			tmpfs_nresolve,
1604 	.vop_nlookupdotdot =		tmpfs_nlookupdotdot,
1605 	.vop_nmknod =			tmpfs_nmknod,
1606 	.vop_open =			tmpfs_open,
1607 	.vop_close =			tmpfs_close,
1608 	.vop_access =			tmpfs_access,
1609 	.vop_getattr =			tmpfs_getattr,
1610 	.vop_setattr =			tmpfs_setattr,
1611 	.vop_read =			tmpfs_read,
1612 	.vop_write =			tmpfs_write,
1613 	.vop_fsync =			tmpfs_fsync,
1614 	.vop_nremove =			tmpfs_nremove,
1615 	.vop_nlink =			tmpfs_nlink,
1616 	.vop_nrename =			tmpfs_nrename,
1617 	.vop_nmkdir =			tmpfs_nmkdir,
1618 	.vop_nrmdir =			tmpfs_nrmdir,
1619 	.vop_nsymlink =			tmpfs_nsymlink,
1620 	.vop_readdir =			tmpfs_readdir,
1621 	.vop_readlink =			tmpfs_readlink,
1622 	.vop_inactive =			tmpfs_inactive,
1623 	.vop_reclaim =			tmpfs_reclaim,
1624 	.vop_print =			tmpfs_print,
1625 	.vop_pathconf =			tmpfs_pathconf,
1626 	.vop_bmap =			tmpfs_bmap,
1627 	.vop_strategy =			tmpfs_strategy,
1628 	.vop_advlock =			tmpfs_advlock,
1629 	.vop_kqfilter =			tmpfs_kqfilter
1630 };
1631