xref: /dflybsd-src/sys/vfs/tmpfs/tmpfs_vnops.c (revision 872a09d51adf63b4bdae6adb1d96a53f76e161e2)
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/uio.h>
41 #include <sys/fcntl.h>
42 #include <sys/lockf.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/resourcevar.h>
46 #include <sys/sched.h>
47 #include <sys/stat.h>
48 #include <sys/systm.h>
49 #include <sys/sysctl.h>
50 #include <sys/unistd.h>
51 #include <sys/vfsops.h>
52 #include <sys/vnode.h>
53 #include <sys/mountctl.h>
54 
55 #include <vm/vm.h>
56 #include <vm/vm_extern.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_pageout.h>
60 #include <vm/vm_pager.h>
61 #include <vm/swap_pager.h>
62 
63 #include <sys/buf2.h>
64 #include <vm/vm_page2.h>
65 
66 #include <vfs/fifofs/fifo.h>
67 #include <vfs/tmpfs/tmpfs_vnops.h>
68 #include "tmpfs.h"
69 
70 static void tmpfs_strategy_done(struct bio *bio);
71 static void tmpfs_move_pages(vm_object_t src, vm_object_t dst, int movflags);
72 
73 /*
74  * bufcache_mode:
75  *	0	Normal page queue operation on flush.  Try to keep in memory.
76  *	1	Try to cache on flush to swap (default).
77  *	2	Always page to swap (not recommended).
78  */
79 __read_mostly static int tmpfs_cluster_rd_enable = 1;
80 __read_mostly static int tmpfs_cluster_wr_enable = 1;
81 __read_mostly int tmpfs_bufcache_mode = 1;
82 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "TMPFS filesystem");
83 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, cluster_rd_enable, CTLFLAG_RW,
84 		&tmpfs_cluster_rd_enable, 0, "");
85 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, cluster_wr_enable, CTLFLAG_RW,
86 		&tmpfs_cluster_wr_enable, 0, "");
87 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, bufcache_mode, CTLFLAG_RW,
88 		&tmpfs_bufcache_mode, 0, "");
89 
90 #define TMPFS_MOVF_FROMBACKING	0x0001
91 #define TMPFS_MOVF_DEACTIVATE	0x0002
92 
93 
94 static __inline
95 void
96 tmpfs_knote(struct vnode *vp, int flags)
97 {
98 	if (flags)
99 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
100 }
101 
102 
103 /* --------------------------------------------------------------------- */
104 
105 static int
106 tmpfs_nresolve(struct vop_nresolve_args *ap)
107 {
108 	struct vnode *dvp = ap->a_dvp;
109 	struct vnode *vp = NULL;
110 	struct namecache *ncp = ap->a_nch->ncp;
111 	struct tmpfs_node *tnode;
112 	struct tmpfs_dirent *de;
113 	struct tmpfs_node *dnode;
114 	int error;
115 
116 	dnode = VP_TO_TMPFS_DIR(dvp);
117 
118 	TMPFS_NODE_LOCK_SH(dnode);
119 loop:
120 	de = tmpfs_dir_lookup(dnode, NULL, ncp);
121 	if (de == NULL) {
122 		error = ENOENT;
123 	} else {
124 		/*
125 		 * Allocate a vnode for the node we found.  Use
126 		 * tmpfs_alloc_vp()'s deadlock handling mode.
127 		 */
128 		tnode = de->td_node;
129 		error = tmpfs_alloc_vp(dvp->v_mount, dnode, tnode,
130 				       LK_EXCLUSIVE | LK_RETRY, &vp);
131 		if (error == EAGAIN)
132 			goto loop;
133 		if (error)
134 			goto out;
135 		KKASSERT(vp);
136 	}
137 
138 out:
139 	TMPFS_NODE_UNLOCK(dnode);
140 
141 	if ((dnode->tn_status & TMPFS_NODE_ACCESSED) == 0) {
142 		TMPFS_NODE_LOCK(dnode);
143 		dnode->tn_status |= TMPFS_NODE_ACCESSED;
144 		TMPFS_NODE_UNLOCK(dnode);
145 	}
146 
147 	/*
148 	 * Store the result of this lookup in the cache.  Avoid this if the
149 	 * request was for creation, as it does not improve timings on
150 	 * emprical tests.
151 	 */
152 	if (vp) {
153 		vn_unlock(vp);
154 		cache_setvp(ap->a_nch, vp);
155 		vrele(vp);
156 	} else if (error == ENOENT) {
157 		cache_setvp(ap->a_nch, NULL);
158 	}
159 	return (error);
160 }
161 
162 static int
163 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
164 {
165 	struct vnode *dvp = ap->a_dvp;
166 	struct vnode **vpp = ap->a_vpp;
167 	struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
168 	struct ucred *cred = ap->a_cred;
169 	int error;
170 
171 	*vpp = NULL;
172 
173 	/* Check accessibility of requested node as a first step. */
174 	error = VOP_ACCESS(dvp, VEXEC, cred);
175 	if (error != 0)
176 		return error;
177 
178 	if (dnode->tn_dir.tn_parent != NULL) {
179 		/* Allocate a new vnode on the matching entry. */
180 		error = tmpfs_alloc_vp(dvp->v_mount,
181 				       NULL, dnode->tn_dir.tn_parent,
182 				       LK_EXCLUSIVE | LK_RETRY, vpp);
183 
184 		if (*vpp)
185 			vn_unlock(*vpp);
186 	}
187 	return (*vpp == NULL) ? ENOENT : 0;
188 }
189 
190 /* --------------------------------------------------------------------- */
191 
192 static int
193 tmpfs_ncreate(struct vop_ncreate_args *ap)
194 {
195 	struct vnode *dvp = ap->a_dvp;
196 	struct vnode **vpp = ap->a_vpp;
197 	struct namecache *ncp = ap->a_nch->ncp;
198 	struct vattr *vap = ap->a_vap;
199 	struct ucred *cred = ap->a_cred;
200 	int error;
201 
202 	KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
203 
204 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
205 	if (error == 0) {
206 		cache_setunresolved(ap->a_nch);
207 		cache_setvp(ap->a_nch, *vpp);
208 		tmpfs_knote(dvp, NOTE_WRITE);
209 	}
210 	return (error);
211 }
212 /* --------------------------------------------------------------------- */
213 
214 static int
215 tmpfs_nmknod(struct vop_nmknod_args *ap)
216 {
217 	struct vnode *dvp = ap->a_dvp;
218 	struct vnode **vpp = ap->a_vpp;
219 	struct namecache *ncp = ap->a_nch->ncp;
220 	struct vattr *vap = ap->a_vap;
221 	struct ucred *cred = ap->a_cred;
222 	int error;
223 
224 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
225 	    vap->va_type != VFIFO) {
226 		return (EINVAL);
227 	}
228 
229 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
230 	if (error == 0) {
231 		cache_setunresolved(ap->a_nch);
232 		cache_setvp(ap->a_nch, *vpp);
233 		tmpfs_knote(dvp, NOTE_WRITE);
234 	}
235 	return error;
236 }
237 
238 /* --------------------------------------------------------------------- */
239 
240 static int
241 tmpfs_open(struct vop_open_args *ap)
242 {
243 	struct vnode *vp = ap->a_vp;
244 	int mode = ap->a_mode;
245 	struct tmpfs_node *node;
246 	int error;
247 
248 	node = VP_TO_TMPFS_NODE(vp);
249 
250 #if 0
251 	/* The file is still active but all its names have been removed
252 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
253 	 * it is about to die. */
254 	if (node->tn_links < 1)
255 		return (ENOENT);
256 #endif
257 
258 	/* If the file is marked append-only, deny write requests. */
259 	if ((node->tn_flags & APPEND) &&
260 	    (mode & (FWRITE | O_APPEND)) == FWRITE) {
261 		error = EPERM;
262 	} else {
263 		if (node->tn_reg.tn_pages_in_aobj) {
264 			TMPFS_NODE_LOCK(node);
265 			if (node->tn_reg.tn_pages_in_aobj) {
266 				tmpfs_move_pages(node->tn_reg.tn_aobj,
267 						 vp->v_object,
268 						 TMPFS_MOVF_FROMBACKING);
269 				node->tn_reg.tn_pages_in_aobj = 0;
270 			}
271 			TMPFS_NODE_UNLOCK(node);
272 		}
273 		error = vop_stdopen(ap);
274 	}
275 
276 	return (error);
277 }
278 
279 /* --------------------------------------------------------------------- */
280 
281 static int
282 tmpfs_close(struct vop_close_args *ap)
283 {
284 	struct vnode *vp = ap->a_vp;
285 	struct tmpfs_node *node;
286 	int error;
287 
288 	node = VP_TO_TMPFS_NODE(vp);
289 
290 	if (node->tn_links > 0) {
291 		/*
292 		 * Update node times.  No need to do it if the node has
293 		 * been deleted, because it will vanish after we return.
294 		 */
295 		tmpfs_update(vp);
296 	}
297 
298 	error = vop_stdclose(ap);
299 
300 	return (error);
301 }
302 
303 /* --------------------------------------------------------------------- */
304 
305 int
306 tmpfs_access(struct vop_access_args *ap)
307 {
308 	struct vnode *vp = ap->a_vp;
309 	int error;
310 	struct tmpfs_node *node;
311 
312 	node = VP_TO_TMPFS_NODE(vp);
313 
314 	switch (vp->v_type) {
315 	case VDIR:
316 		/* FALLTHROUGH */
317 	case VLNK:
318 		/* FALLTHROUGH */
319 	case VREG:
320 		if ((ap->a_mode & VWRITE) &&
321 	            (vp->v_mount->mnt_flag & MNT_RDONLY)) {
322 			error = EROFS;
323 			goto out;
324 		}
325 		break;
326 
327 	case VBLK:
328 		/* FALLTHROUGH */
329 	case VCHR:
330 		/* FALLTHROUGH */
331 	case VSOCK:
332 		/* FALLTHROUGH */
333 	case VFIFO:
334 		break;
335 
336 	default:
337 		error = EINVAL;
338 		goto out;
339 	}
340 
341 	if ((ap->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
342 		error = EPERM;
343 		goto out;
344 	}
345 
346 	error = vop_helper_access(ap, node->tn_uid, node->tn_gid,
347 			          node->tn_mode, 0);
348 out:
349 	return error;
350 }
351 
352 /* --------------------------------------------------------------------- */
353 
354 int
355 tmpfs_getattr(struct vop_getattr_args *ap)
356 {
357 	struct vnode *vp = ap->a_vp;
358 	struct vattr *vap = ap->a_vap;
359 	struct tmpfs_node *node;
360 
361 	node = VP_TO_TMPFS_NODE(vp);
362 
363 	tmpfs_update(vp);
364 
365 	TMPFS_NODE_LOCK_SH(node);
366 	vap->va_type = vp->v_type;
367 	vap->va_mode = node->tn_mode;
368 	vap->va_nlink = node->tn_links;
369 	vap->va_uid = node->tn_uid;
370 	vap->va_gid = node->tn_gid;
371 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
372 	vap->va_fileid = node->tn_id;
373 	vap->va_size = node->tn_size;
374 	vap->va_blocksize = PAGE_SIZE;
375 	vap->va_atime.tv_sec = node->tn_atime;
376 	vap->va_atime.tv_nsec = node->tn_atimensec;
377 	vap->va_mtime.tv_sec = node->tn_mtime;
378 	vap->va_mtime.tv_nsec = node->tn_mtimensec;
379 	vap->va_ctime.tv_sec = node->tn_ctime;
380 	vap->va_ctime.tv_nsec = node->tn_ctimensec;
381 	vap->va_gen = node->tn_gen;
382 	vap->va_flags = node->tn_flags;
383 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
384 		vap->va_rmajor = umajor(node->tn_rdev);
385 		vap->va_rminor = uminor(node->tn_rdev);
386 	}
387 	vap->va_bytes = round_page(node->tn_size);
388 	vap->va_filerev = 0;
389 	TMPFS_NODE_UNLOCK(node);
390 
391 	return 0;
392 }
393 
394 /* --------------------------------------------------------------------- */
395 
396 int
397 tmpfs_setattr(struct vop_setattr_args *ap)
398 {
399 	struct vnode *vp = ap->a_vp;
400 	struct vattr *vap = ap->a_vap;
401 	struct ucred *cred = ap->a_cred;
402 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
403 	int error = 0;
404 	int kflags = 0;
405 
406 	TMPFS_NODE_LOCK(node);
407 	if (error == 0 && (vap->va_flags != VNOVAL)) {
408 		error = tmpfs_chflags(vp, vap->va_flags, cred);
409 		kflags |= NOTE_ATTRIB;
410 	}
411 
412 	if (error == 0 && (vap->va_size != VNOVAL)) {
413 		/* restore any saved pages before proceeding */
414 		if (node->tn_reg.tn_pages_in_aobj) {
415 			tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object,
416 					 TMPFS_MOVF_FROMBACKING |
417 					 TMPFS_MOVF_DEACTIVATE);
418 			node->tn_reg.tn_pages_in_aobj = 0;
419 		}
420 		if (vap->va_size > node->tn_size)
421 			kflags |= NOTE_WRITE | NOTE_EXTEND;
422 		else
423 			kflags |= NOTE_WRITE;
424 		error = tmpfs_chsize(vp, vap->va_size, cred);
425 	}
426 
427 	if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
428 			   vap->va_gid != (gid_t)VNOVAL)) {
429 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
430 		kflags |= NOTE_ATTRIB;
431 	}
432 
433 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
434 		error = tmpfs_chmod(vp, vap->va_mode, cred);
435 		kflags |= NOTE_ATTRIB;
436 	}
437 
438 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
439 	    vap->va_atime.tv_nsec != VNOVAL) ||
440 	    (vap->va_mtime.tv_sec != VNOVAL &&
441 	    vap->va_mtime.tv_nsec != VNOVAL) )) {
442 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
443 				      vap->va_vaflags, cred);
444 		kflags |= NOTE_ATTRIB;
445 	}
446 
447 	/*
448 	 * Update the node times.  We give preference to the error codes
449 	 * generated by this function rather than the ones that may arise
450 	 * from tmpfs_update.
451 	 */
452 	tmpfs_update(vp);
453 	TMPFS_NODE_UNLOCK(node);
454 	tmpfs_knote(vp, kflags);
455 
456 	return (error);
457 }
458 
459 /* --------------------------------------------------------------------- */
460 
461 /*
462  * fsync is usually a NOP, but we must take action when unmounting or
463  * when recycling.
464  */
465 static int
466 tmpfs_fsync(struct vop_fsync_args *ap)
467 {
468 	struct tmpfs_node *node;
469 	struct vnode *vp = ap->a_vp;
470 
471 	node = VP_TO_TMPFS_NODE(vp);
472 
473 	/*
474 	 * tmpfs vnodes typically remain dirty, avoid long syncer scans
475 	 * by forcing removal from the syncer list.
476 	 */
477 	vn_syncer_remove(vp, 1);
478 
479 	tmpfs_update(vp);
480 	if (vp->v_type == VREG) {
481 		if (vp->v_flag & VRECLAIMED) {
482 			if (node->tn_links == 0)
483 				tmpfs_truncate(vp, 0);
484 			else
485 				vfsync(ap->a_vp, ap->a_waitfor, 1, NULL, NULL);
486 		}
487 	}
488 
489 	return 0;
490 }
491 
492 /* --------------------------------------------------------------------- */
493 
494 static int
495 tmpfs_read(struct vop_read_args *ap)
496 {
497 	struct buf *bp;
498 	struct vnode *vp = ap->a_vp;
499 	struct uio *uio = ap->a_uio;
500 	struct tmpfs_node *node;
501 	off_t base_offset;
502 	size_t offset;
503 	size_t len;
504 	size_t resid;
505 	int error;
506 	int seqcount;
507 
508 	/*
509 	 * Check the basics
510 	 */
511 	if (uio->uio_offset < 0)
512 		return (EINVAL);
513 	if (vp->v_type != VREG)
514 		return (EINVAL);
515 
516 	/*
517 	 * Extract node, try to shortcut the operation through
518 	 * the VM page cache, allowing us to avoid buffer cache
519 	 * overheads.
520 	 */
521 	node = VP_TO_TMPFS_NODE(vp);
522         resid = uio->uio_resid;
523 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
524         error = vop_helper_read_shortcut(ap);
525         if (error)
526                 return error;
527         if (uio->uio_resid == 0) {
528 		if (resid)
529 			goto finished;
530 		return error;
531 	}
532 
533 	/*
534 	 * restore any saved pages before proceeding
535 	 */
536 	if (node->tn_reg.tn_pages_in_aobj) {
537 		TMPFS_NODE_LOCK(node);
538 		if (node->tn_reg.tn_pages_in_aobj) {
539 			tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object,
540 					 TMPFS_MOVF_FROMBACKING);
541 			node->tn_reg.tn_pages_in_aobj = 0;
542 		}
543 		TMPFS_NODE_UNLOCK(node);
544 	}
545 
546 	/*
547 	 * Fall-through to our normal read code.
548 	 */
549 	while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
550 		/*
551 		 * Use buffer cache I/O (via tmpfs_strategy)
552 		 */
553 		offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64;
554 		base_offset = (off_t)uio->uio_offset - offset;
555 		bp = getcacheblk(vp, base_offset,
556 				 node->tn_blksize, GETBLK_KVABIO);
557 		if (bp == NULL) {
558 			if (tmpfs_cluster_rd_enable) {
559 				error = cluster_readx(vp, node->tn_size,
560 						     base_offset,
561 						     node->tn_blksize,
562 						     B_NOTMETA | B_KVABIO,
563 						     uio->uio_resid,
564 						     seqcount * MAXBSIZE,
565 						     &bp);
566 			} else {
567 				error = bread_kvabio(vp, base_offset,
568 						     node->tn_blksize, &bp);
569 			}
570 			if (error) {
571 				brelse(bp);
572 				kprintf("tmpfs_read bread error %d\n", error);
573 				break;
574 			}
575 
576 			/*
577 			 * tmpfs pretty much fiddles directly with the VM
578 			 * system, don't let it exhaust it or we won't play
579 			 * nice with other processes.
580 			 *
581 			 * Only do this if the VOP is coming from a normal
582 			 * read/write.  The VM system handles the case for
583 			 * UIO_NOCOPY.
584 			 */
585 			if (uio->uio_segflg != UIO_NOCOPY)
586 				vm_wait_nominal();
587 		}
588 		bp->b_flags |= B_CLUSTEROK;
589 		bkvasync(bp);
590 
591 		/*
592 		 * Figure out how many bytes we can actually copy this loop.
593 		 */
594 		len = node->tn_blksize - offset;
595 		if (len > uio->uio_resid)
596 			len = uio->uio_resid;
597 		if (len > node->tn_size - uio->uio_offset)
598 			len = (size_t)(node->tn_size - uio->uio_offset);
599 
600 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
601 		bqrelse(bp);
602 		if (error) {
603 			kprintf("tmpfs_read uiomove error %d\n", error);
604 			break;
605 		}
606 	}
607 
608 finished:
609 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
610 		TMPFS_NODE_LOCK(node);
611 		node->tn_status |= TMPFS_NODE_ACCESSED;
612 		TMPFS_NODE_UNLOCK(node);
613 	}
614 	return (error);
615 }
616 
617 static int
618 tmpfs_write(struct vop_write_args *ap)
619 {
620 	struct buf *bp;
621 	struct vnode *vp = ap->a_vp;
622 	struct uio *uio = ap->a_uio;
623 	struct thread *td = uio->uio_td;
624 	struct tmpfs_node *node;
625 	boolean_t extended;
626 	off_t oldsize;
627 	int error;
628 	off_t base_offset;
629 	size_t offset;
630 	size_t len;
631 	struct rlimit limit;
632 	int trivial = 0;
633 	int kflags = 0;
634 	int seqcount;
635 
636 	error = 0;
637 	if (uio->uio_resid == 0) {
638 		return error;
639 	}
640 
641 	node = VP_TO_TMPFS_NODE(vp);
642 
643 	if (vp->v_type != VREG)
644 		return (EINVAL);
645 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
646 
647 	TMPFS_NODE_LOCK(node);
648 
649 	/*
650 	 * restore any saved pages before proceeding
651 	 */
652 	if (node->tn_reg.tn_pages_in_aobj) {
653 		tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object,
654 				 TMPFS_MOVF_FROMBACKING);
655 		node->tn_reg.tn_pages_in_aobj = 0;
656 	}
657 
658 	oldsize = node->tn_size;
659 	if (ap->a_ioflag & IO_APPEND)
660 		uio->uio_offset = node->tn_size;
661 
662 	/*
663 	 * Check for illegal write offsets.
664 	 */
665 	if (uio->uio_offset + uio->uio_resid >
666 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
667 		error = EFBIG;
668 		goto done;
669 	}
670 
671 	/*
672 	 * NOTE: Ignore if UIO does not come from a user thread (e.g. VN).
673 	 */
674 	if (vp->v_type == VREG && td != NULL && td->td_lwp != NULL) {
675 		error = kern_getrlimit(RLIMIT_FSIZE, &limit);
676 		if (error)
677 			goto done;
678 		if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
679 			ksignal(td->td_proc, SIGXFSZ);
680 			error = EFBIG;
681 			goto done;
682 		}
683 	}
684 
685 	/*
686 	 * Extend the file's size if necessary
687 	 */
688 	extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
689 
690 	while (uio->uio_resid > 0) {
691 		/*
692 		 * Don't completely blow out running buffer I/O
693 		 * when being hit from the pageout daemon.
694 		 */
695 		if (uio->uio_segflg == UIO_NOCOPY &&
696 		    (ap->a_ioflag & IO_RECURSE) == 0) {
697 			bwillwrite(node->tn_blksize);
698 		}
699 
700 		/*
701 		 * Use buffer cache I/O (via tmpfs_strategy)
702 		 *
703 		 * Calculate the maximum bytes we can write to the buffer at
704 		 * this offset (after resizing).
705 		 */
706 		offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64;
707 		base_offset = (off_t)uio->uio_offset - offset;
708 		len = uio->uio_resid;
709 		if (len > TMPFS_BLKSIZE - offset)
710 			len = TMPFS_BLKSIZE - offset;
711 
712 		if ((uio->uio_offset + len) > node->tn_size) {
713 			trivial = (uio->uio_offset <= node->tn_size);
714 			error = tmpfs_reg_resize(vp, uio->uio_offset + len,
715 						 trivial);
716 			if (error)
717 				break;
718 		}
719 
720 		/*
721 		 * Read to fill in any gaps.  Theoretically we could
722 		 * optimize this if the write covers the entire buffer
723 		 * and is not a UIO_NOCOPY write, however this can lead
724 		 * to a security violation exposing random kernel memory
725 		 * (whatever junk was in the backing VM pages before).
726 		 *
727 		 * So just use bread() to do the right thing.
728 		 */
729 		error = bread_kvabio(vp, base_offset, node->tn_blksize, &bp);
730 		bkvasync(bp);
731 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
732 		if (error) {
733 			kprintf("tmpfs_write uiomove error %d\n", error);
734 			brelse(bp);
735 			break;
736 		}
737 
738 		if (uio->uio_offset > node->tn_size) {
739 			node->tn_size = uio->uio_offset;
740 			kflags |= NOTE_EXTEND;
741 		}
742 		kflags |= NOTE_WRITE;
743 
744 		/*
745 		 * UIO_NOCOPY is a sensitive state due to potentially being
746 		 * issued from the pageout daemon while in a low-memory
747 		 * situation.  However, in order to cluster the I/O nicely
748 		 * (e.g. 64KB+ writes instead of 16KB writes), we still try
749 		 * to follow the same semantics that any other filesystem
750 		 * might use.
751 		 *
752 		 * For the normal case we buwrite(), dirtying the underlying
753 		 * VM pages instead of dirtying the buffer and releasing the
754 		 * buffer as a clean buffer.  This allows tmpfs to use
755 		 * essentially all available memory to cache file data.
756 		 * If we used bdwrite() the buffer cache would wind up
757 		 * flushing the data to swap too quickly.
758 		 *
759 		 * But because tmpfs can seriously load the VM system we
760 		 * fall-back to using bdwrite() when free memory starts
761 		 * to get low.  This shifts the load away from the VM system
762 		 * and makes tmpfs act more like a normal filesystem with
763 		 * regards to disk activity.
764 		 *
765 		 * tmpfs pretty much fiddles directly with the VM
766 		 * system, don't let it exhaust it or we won't play
767 		 * nice with other processes.  Only do this if the
768 		 * VOP is coming from a normal read/write.  The VM system
769 		 * handles the case for UIO_NOCOPY.
770 		 */
771 		bp->b_flags |= B_CLUSTEROK;
772 		if (uio->uio_segflg == UIO_NOCOPY) {
773 			/*
774 			 * Flush from the pageout daemon, deal with
775 			 * potentially very heavy tmpfs write activity
776 			 * causing long stalls in the pageout daemon
777 			 * before pages get to free/cache.
778 			 *
779 			 * (a) Under severe pressure setting B_DIRECT will
780 			 *     cause a buffer release to try to free the
781 			 *     underlying pages.
782 			 *
783 			 * (b) Under modest memory pressure the B_RELBUF
784 			 *     alone is sufficient to get the pages moved
785 			 *     to the cache.  We could also force this by
786 			 *     setting B_NOTMETA but that might have other
787 			 *     unintended side-effects (e.g. setting
788 			 *     PG_NOTMETA on the VM page).
789 			 *
790 			 * (c) For the pageout->putpages->generic_putpages->
791 			 *     UIO_NOCOPY-write (here), issuing an immediate
792 			 *     write prevents any real clustering from
793 			 *     happening because the buffers probably aren't
794 			 *     (yet) marked dirty, or lost due to prior use
795 			 *     of buwrite().  Try to use the normal
796 			 *     cluster_write() mechanism for performance.
797 			 *
798 			 * Hopefully this will unblock the VM system more
799 			 * quickly under extreme tmpfs write load.
800 			 */
801 			if (vm_page_count_min(vm_page_free_hysteresis))
802 				bp->b_flags |= B_DIRECT;
803 			bp->b_flags |= B_AGE | B_RELBUF | B_TTC;
804 			bp->b_act_count = 0;	/* buffer->deactivate pgs */
805 			if (tmpfs_cluster_wr_enable &&
806 			    (ap->a_ioflag & (IO_SYNC | IO_DIRECT)) == 0) {
807 				cluster_write(bp, node->tn_size,
808 					      node->tn_blksize, seqcount);
809 			} else {
810 				cluster_awrite(bp);
811 			}
812 		} else if (vm_pages_needed || vm_paging_needed(0) ||
813 			   tmpfs_bufcache_mode >= 2) {
814 			/*
815 			 * If the pageout daemon is running we cycle the
816 			 * write through the buffer cache normally to
817 			 * pipeline the flush, thus avoiding adding any
818 			 * more memory pressure to the pageout daemon.
819 			 */
820 			bp->b_act_count = 0;	/* buffer->deactivate pgs */
821 			if (tmpfs_cluster_wr_enable) {
822 				cluster_write(bp, node->tn_size,
823 					      node->tn_blksize, seqcount);
824 			} else {
825 				bdwrite(bp);
826 			}
827 		} else {
828 			/*
829 			 * Otherwise run the buffer directly through to the
830 			 * backing VM store, leaving the buffer clean so
831 			 * buffer limits do not force early flushes to swap.
832 			 */
833 			buwrite(bp);
834 			/*vm_wait_nominal();*/
835 		}
836 
837 		if (bp->b_error) {
838 			kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
839 			break;
840 		}
841 	}
842 
843 	if (error) {
844 		if (extended) {
845 			(void)tmpfs_reg_resize(vp, oldsize, trivial);
846 			kflags &= ~NOTE_EXTEND;
847 		}
848 		goto done;
849 	}
850 
851 	/*
852 	 * Currently we don't set the mtime on files modified via mmap()
853 	 * because we can't tell the difference between those modifications
854 	 * and an attempt by the pageout daemon to flush tmpfs pages to
855 	 * swap.
856 	 *
857 	 * This is because in order to defer flushes as long as possible
858 	 * buwrite() works by marking the underlying VM pages dirty in
859 	 * order to be able to dispose of the buffer cache buffer without
860 	 * flushing it.
861 	 */
862 	if (uio->uio_segflg == UIO_NOCOPY) {
863 		if (vp->v_flag & VLASTWRITETS) {
864 			node->tn_mtime = vp->v_lastwrite_ts.tv_sec;
865 			node->tn_mtimensec = vp->v_lastwrite_ts.tv_nsec;
866 		}
867 	} else {
868 		node->tn_status |= TMPFS_NODE_MODIFIED;
869 		vclrflags(vp, VLASTWRITETS);
870 	}
871 
872 	if (extended)
873 		node->tn_status |= TMPFS_NODE_CHANGED;
874 
875 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
876 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
877 			node->tn_mode &= ~(S_ISUID | S_ISGID);
878 	}
879 done:
880 	TMPFS_NODE_UNLOCK(node);
881 	if (kflags)
882 		tmpfs_knote(vp, kflags);
883 
884 	return(error);
885 }
886 
887 static int
888 tmpfs_advlock(struct vop_advlock_args *ap)
889 {
890 	struct tmpfs_node *node;
891 	struct vnode *vp = ap->a_vp;
892 	int error;
893 
894 	node = VP_TO_TMPFS_NODE(vp);
895 	error = (lf_advlock(ap, &node->tn_advlock, node->tn_size));
896 
897 	return (error);
898 }
899 
900 /*
901  * The strategy function is typically only called when memory pressure
902  * forces the system to attempt to pageout pages.  It can also be called
903  * by [n]vtruncbuf() when a truncation cuts a page in half.  Normal write
904  * operations
905  *
906  * We set VKVABIO for VREG files so bp->b_data may not be synchronized to
907  * our cpu.  swap_pager_strategy() is all we really use, and it directly
908  * supports this.
909  */
910 static int
911 tmpfs_strategy(struct vop_strategy_args *ap)
912 {
913 	struct bio *bio = ap->a_bio;
914 	struct bio *nbio;
915 	struct buf *bp = bio->bio_buf;
916 	struct vnode *vp = ap->a_vp;
917 	struct tmpfs_node *node;
918 	vm_object_t uobj;
919 	vm_page_t m;
920 	int i;
921 
922 	if (vp->v_type != VREG) {
923 		bp->b_resid = bp->b_bcount;
924 		bp->b_flags |= B_ERROR | B_INVAL;
925 		bp->b_error = EINVAL;
926 		biodone(bio);
927 		return(0);
928 	}
929 
930 	node = VP_TO_TMPFS_NODE(vp);
931 
932 	uobj = node->tn_reg.tn_aobj;
933 
934 	/*
935 	 * Don't bother flushing to swap if there is no swap, just
936 	 * ensure that the pages are marked as needing a commit (still).
937 	 */
938 	if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
939 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
940 			m = bp->b_xio.xio_pages[i];
941 			vm_page_need_commit(m);
942 		}
943 		bp->b_resid = 0;
944 		bp->b_error = 0;
945 		biodone(bio);
946 	} else {
947 		/*
948 		 * Tell the buffer cache to try to recycle the pages
949 		 * to PQ_CACHE on release.
950 		 */
951 		if (tmpfs_bufcache_mode >= 2 ||
952 		    (tmpfs_bufcache_mode == 1 && vm_paging_needed(0))) {
953 			bp->b_flags |= B_TTC;
954 		}
955 		nbio = push_bio(bio);
956 		nbio->bio_done = tmpfs_strategy_done;
957 		nbio->bio_offset = bio->bio_offset;
958 		swap_pager_strategy(uobj, nbio);
959 	}
960 	return 0;
961 }
962 
963 /*
964  * If we were unable to commit the pages to swap make sure they are marked
965  * as needing a commit (again).  If we were, clear the flag to allow the
966  * pages to be freed.
967  *
968  * Do not error-out the buffer.  In particular, vinvalbuf() needs to
969  * always work.
970  */
971 static void
972 tmpfs_strategy_done(struct bio *bio)
973 {
974 	struct buf *bp;
975 	vm_page_t m;
976 	int i;
977 
978 	bp = bio->bio_buf;
979 
980 	if (bp->b_flags & B_ERROR) {
981 		bp->b_flags &= ~B_ERROR;
982 		bp->b_error = 0;
983 		bp->b_resid = 0;
984 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
985 			m = bp->b_xio.xio_pages[i];
986 			vm_page_need_commit(m);
987 		}
988 	} else {
989 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
990 			m = bp->b_xio.xio_pages[i];
991 			vm_page_clear_commit(m);
992 		}
993 	}
994 	bio = pop_bio(bio);
995 	biodone(bio);
996 }
997 
998 /*
999  * To make write clustering work well make the backing store look
1000  * contiguous to the cluster_*() code.  The swap_strategy() function
1001  * will take it from there.
1002  *
1003  * Use MAXBSIZE-sized chunks as a micro-optimization to make random
1004  * flushes leave full-sized gaps.
1005  */
1006 static int
1007 tmpfs_bmap(struct vop_bmap_args *ap)
1008 {
1009 	if (ap->a_doffsetp != NULL)
1010 		*ap->a_doffsetp = ap->a_loffset;
1011 	if (ap->a_runp != NULL)
1012 		*ap->a_runp = MAXBSIZE - (ap->a_loffset & (MAXBSIZE - 1));
1013 	if (ap->a_runb != NULL)
1014 		*ap->a_runb = ap->a_loffset & (MAXBSIZE - 1);
1015 
1016 	return 0;
1017 }
1018 
1019 /* --------------------------------------------------------------------- */
1020 
1021 static int
1022 tmpfs_nremove(struct vop_nremove_args *ap)
1023 {
1024 	struct vnode *dvp = ap->a_dvp;
1025 	struct namecache *ncp = ap->a_nch->ncp;
1026 	struct vnode *vp;
1027 	int error;
1028 	struct tmpfs_dirent *de;
1029 	struct tmpfs_mount *tmp;
1030 	struct tmpfs_node *dnode;
1031 	struct tmpfs_node *node;
1032 
1033 	/*
1034 	 * We have to acquire the vp from ap->a_nch because we will likely
1035 	 * unresolve the namecache entry, and a vrele/vput is needed to
1036 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1037 	 *
1038 	 * We have to use vget to clear any inactive state on the vnode,
1039 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1040 	 * will not get called when we release it.
1041 	 */
1042 	error = cache_vget(ap->a_nch, ap->a_cred, LK_SHARED, &vp);
1043 	KKASSERT(vp->v_mount == dvp->v_mount);
1044 	KKASSERT(error == 0);
1045 	vn_unlock(vp);
1046 
1047 	if (vp->v_type == VDIR) {
1048 		error = EISDIR;
1049 		goto out2;
1050 	}
1051 
1052 	dnode = VP_TO_TMPFS_DIR(dvp);
1053 	node = VP_TO_TMPFS_NODE(vp);
1054 	tmp = VFS_TO_TMPFS(vp->v_mount);
1055 
1056 	TMPFS_NODE_LOCK(dnode);
1057 	de = tmpfs_dir_lookup(dnode, node, ncp);
1058 	if (de == NULL) {
1059 		error = ENOENT;
1060 		TMPFS_NODE_UNLOCK(dnode);
1061 		goto out;
1062 	}
1063 
1064 	/* Files marked as immutable or append-only cannot be deleted. */
1065 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
1066 	    (dnode->tn_flags & APPEND)) {
1067 		error = EPERM;
1068 		TMPFS_NODE_UNLOCK(dnode);
1069 		goto out;
1070 	}
1071 
1072 	/* Remove the entry from the directory; as it is a file, we do not
1073 	 * have to change the number of hard links of the directory. */
1074 	tmpfs_dir_detach(dnode, de);
1075 	TMPFS_NODE_UNLOCK(dnode);
1076 
1077 	/* Free the directory entry we just deleted.  Note that the node
1078 	 * referred by it will not be removed until the vnode is really
1079 	 * reclaimed. */
1080 	tmpfs_free_dirent(tmp, de);
1081 
1082 	if (node->tn_links > 0) {
1083 	        TMPFS_NODE_LOCK(node);
1084 		node->tn_status |= TMPFS_NODE_CHANGED;
1085 	        TMPFS_NODE_UNLOCK(node);
1086 	}
1087 
1088 	cache_unlink(ap->a_nch);
1089 	tmpfs_knote(vp, NOTE_DELETE);
1090 	error = 0;
1091 
1092 out:
1093 	if (error == 0)
1094 		tmpfs_knote(dvp, NOTE_WRITE);
1095 out2:
1096 	vrele(vp);
1097 
1098 	return error;
1099 }
1100 
1101 /* --------------------------------------------------------------------- */
1102 
1103 static int
1104 tmpfs_nlink(struct vop_nlink_args *ap)
1105 {
1106 	struct vnode *dvp = ap->a_dvp;
1107 	struct vnode *vp = ap->a_vp;
1108 	struct namecache *ncp = ap->a_nch->ncp;
1109 	struct tmpfs_dirent *de;
1110 	struct tmpfs_node *node;
1111 	struct tmpfs_node *dnode;
1112 	int error;
1113 
1114 	KKASSERT(dvp != vp); /* XXX When can this be false? */
1115 
1116 	node = VP_TO_TMPFS_NODE(vp);
1117 	dnode = VP_TO_TMPFS_NODE(dvp);
1118 	TMPFS_NODE_LOCK(dnode);
1119 
1120 	/* XXX: Why aren't the following two tests done by the caller? */
1121 
1122 	/* Hard links of directories are forbidden. */
1123 	if (vp->v_type == VDIR) {
1124 		error = EPERM;
1125 		goto out;
1126 	}
1127 
1128 	/* Cannot create cross-device links. */
1129 	if (dvp->v_mount != vp->v_mount) {
1130 		error = EXDEV;
1131 		goto out;
1132 	}
1133 
1134 	/* Ensure that we do not overflow the maximum number of links imposed
1135 	 * by the system. */
1136 	KKASSERT(node->tn_links <= LINK_MAX);
1137 	if (node->tn_links >= LINK_MAX) {
1138 		error = EMLINK;
1139 		goto out;
1140 	}
1141 
1142 	/* We cannot create links of files marked immutable or append-only. */
1143 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
1144 		error = EPERM;
1145 		goto out;
1146 	}
1147 
1148 	/* Allocate a new directory entry to represent the node. */
1149 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
1150 				   ncp->nc_name, ncp->nc_nlen, &de);
1151 	if (error != 0)
1152 		goto out;
1153 
1154 	/* Insert the new directory entry into the appropriate directory. */
1155 	tmpfs_dir_attach(dnode, de);
1156 
1157 	/* vp link count has changed, so update node times. */
1158 
1159 	TMPFS_NODE_LOCK(node);
1160 	node->tn_status |= TMPFS_NODE_CHANGED;
1161 	TMPFS_NODE_UNLOCK(node);
1162 	tmpfs_update(vp);
1163 
1164 	tmpfs_knote(vp, NOTE_LINK);
1165 	cache_setunresolved(ap->a_nch);
1166 	cache_setvp(ap->a_nch, vp);
1167 	error = 0;
1168 
1169 out:
1170 	TMPFS_NODE_UNLOCK(dnode);
1171 	if (error == 0)
1172 		tmpfs_knote(dvp, NOTE_WRITE);
1173 	return error;
1174 }
1175 
1176 /* --------------------------------------------------------------------- */
1177 
1178 static int
1179 tmpfs_nrename(struct vop_nrename_args *ap)
1180 {
1181 	struct vnode *fdvp = ap->a_fdvp;
1182 	struct namecache *fncp = ap->a_fnch->ncp;
1183 	struct vnode *fvp = fncp->nc_vp;
1184 	struct vnode *tdvp = ap->a_tdvp;
1185 	struct namecache *tncp = ap->a_tnch->ncp;
1186 	struct vnode *tvp;
1187 	struct tmpfs_dirent *de, *tde;
1188 	struct tmpfs_mount *tmp;
1189 	struct tmpfs_node *fdnode;
1190 	struct tmpfs_node *tdnode;
1191 	struct tmpfs_node *fnode;
1192 	struct tmpfs_node *tnode;
1193 	char *newname;
1194 	char *oldname;
1195 	int error;
1196 
1197 	KKASSERT(fdvp->v_mount == fvp->v_mount);
1198 
1199 	/*
1200 	 * Because tvp can get overwritten we have to vget it instead of
1201 	 * just vref or use it, otherwise it's VINACTIVE flag may not get
1202 	 * cleared and the node won't get destroyed.
1203 	 */
1204 	error = cache_vget(ap->a_tnch, ap->a_cred, LK_SHARED, &tvp);
1205 	if (error == 0) {
1206 		tnode = VP_TO_TMPFS_NODE(tvp);
1207 		vn_unlock(tvp);
1208 	} else {
1209 		tnode = NULL;
1210 	}
1211 
1212 	/* Disallow cross-device renames.
1213 	 * XXX Why isn't this done by the caller? */
1214 	if (fvp->v_mount != tdvp->v_mount ||
1215 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
1216 		error = EXDEV;
1217 		goto out;
1218 	}
1219 
1220 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
1221 	tdnode = VP_TO_TMPFS_DIR(tdvp);
1222 
1223 	/* If source and target are the same file, there is nothing to do. */
1224 	if (fvp == tvp) {
1225 		error = 0;
1226 		goto out;
1227 	}
1228 
1229 	fdnode = VP_TO_TMPFS_DIR(fdvp);
1230 	fnode = VP_TO_TMPFS_NODE(fvp);
1231 
1232 	tmpfs_lock4(fdnode, tdnode, fnode, tnode);
1233 
1234 	de = tmpfs_dir_lookup(fdnode, fnode, fncp);
1235 
1236 	/* Avoid manipulating '.' and '..' entries. */
1237 	if (de == NULL) {
1238 		error = ENOENT;
1239 		goto out_locked;
1240 	}
1241 	KKASSERT(de->td_node == fnode);
1242 
1243 	/*
1244 	 * If replacing an entry in the target directory and that entry
1245 	 * is a directory, it must be empty.
1246 	 *
1247 	 * Kern_rename gurantees the destination to be a directory
1248 	 * if the source is one (it does?).
1249 	 */
1250 	if (tvp != NULL) {
1251 		KKASSERT(tnode != NULL);
1252 
1253 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1254 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1255 			error = EPERM;
1256 			goto out_locked;
1257 		}
1258 
1259 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1260 			if (tnode->tn_size > 0) {
1261 				error = ENOTEMPTY;
1262 				goto out_locked;
1263 			}
1264 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1265 			error = ENOTDIR;
1266 			goto out_locked;
1267 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1268 			error = EISDIR;
1269 			goto out_locked;
1270 		} else {
1271 			KKASSERT(fnode->tn_type != VDIR &&
1272 				tnode->tn_type != VDIR);
1273 		}
1274 	}
1275 
1276 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1277 	    (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1278 		error = EPERM;
1279 		goto out_locked;
1280 	}
1281 
1282 	/*
1283 	 * Ensure that we have enough memory to hold the new name, if it
1284 	 * has to be changed.
1285 	 */
1286 	if (fncp->nc_nlen != tncp->nc_nlen ||
1287 	    bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1288 		newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1289 				  M_WAITOK | M_NULLOK);
1290 		if (newname == NULL) {
1291 			error = ENOSPC;
1292 			goto out_locked;
1293 		}
1294 		bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1295 		newname[tncp->nc_nlen] = '\0';
1296 	} else {
1297 		newname = NULL;
1298 	}
1299 
1300 	/*
1301 	 * Unlink entry from source directory.  Note that the kernel has
1302 	 * already checked for illegal recursion cases (renaming a directory
1303 	 * into a subdirectory of itself).
1304 	 */
1305 	if (fdnode != tdnode) {
1306 		tmpfs_dir_detach(fdnode, de);
1307 	} else {
1308 		/* XXX depend on namecache lock */
1309 		KKASSERT(de == tmpfs_dir_lookup(fdnode, fnode, fncp));
1310 		RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de);
1311 		RB_REMOVE(tmpfs_dirtree_cookie,
1312 			  &fdnode->tn_dir.tn_cookietree, de);
1313 	}
1314 
1315 	/*
1316 	 * Handle any name change.  Swap with newname, we will
1317 	 * deallocate it at the end.
1318 	 */
1319 	if (newname != NULL) {
1320 		oldname = de->td_name;
1321 		de->td_name = newname;
1322 		de->td_namelen = (uint16_t)tncp->nc_nlen;
1323 		newname = oldname;
1324 	}
1325 
1326 	/*
1327 	 * If we are overwriting an entry, we have to remove the old one
1328 	 * from the target directory.
1329 	 */
1330 	if (tvp != NULL) {
1331 		/* Remove the old entry from the target directory. */
1332 		tde = tmpfs_dir_lookup(tdnode, tnode, tncp);
1333 		tmpfs_dir_detach(tdnode, tde);
1334 		tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1335 
1336 		/*
1337 		 * Free the directory entry we just deleted.  Note that the
1338 		 * node referred by it will not be removed until the vnode is
1339 		 * really reclaimed.
1340 		 */
1341 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1342 		/*cache_inval_vp(tvp, CINV_DESTROY);*/
1343 	}
1344 
1345 	/*
1346 	 * Link entry to target directory.  If the entry
1347 	 * represents a directory move the parent linkage
1348 	 * as well.
1349 	 */
1350 	if (fdnode != tdnode) {
1351 		if (de->td_node->tn_type == VDIR) {
1352 			TMPFS_VALIDATE_DIR(fnode);
1353 		}
1354 		tmpfs_dir_attach(tdnode, de);
1355 	} else {
1356 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1357 		RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de);
1358 		RB_INSERT(tmpfs_dirtree_cookie,
1359 			  &tdnode->tn_dir.tn_cookietree, de);
1360 	}
1361 	tmpfs_unlock4(fdnode, tdnode, fnode, tnode);
1362 
1363 	/*
1364 	 * Finish up
1365 	 */
1366 	if (newname) {
1367 		kfree(newname, tmp->tm_name_zone);
1368 		newname = NULL;
1369 	}
1370 	cache_rename(ap->a_fnch, ap->a_tnch);
1371 	tmpfs_knote(ap->a_fdvp, NOTE_WRITE);
1372 	tmpfs_knote(ap->a_tdvp, NOTE_WRITE);
1373 	if (fnode->tn_vnode)
1374 		tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1375 	if (tvp)
1376 		vrele(tvp);
1377 	return 0;
1378 
1379 out_locked:
1380 	tmpfs_unlock4(fdnode, tdnode, fnode, tnode);
1381 out:
1382 	if (tvp)
1383 		vrele(tvp);
1384 	return error;
1385 }
1386 
1387 /* --------------------------------------------------------------------- */
1388 
1389 static int
1390 tmpfs_nmkdir(struct vop_nmkdir_args *ap)
1391 {
1392 	struct vnode *dvp = ap->a_dvp;
1393 	struct vnode **vpp = ap->a_vpp;
1394 	struct namecache *ncp = ap->a_nch->ncp;
1395 	struct vattr *vap = ap->a_vap;
1396 	struct ucred *cred = ap->a_cred;
1397 	int error;
1398 
1399 	KKASSERT(vap->va_type == VDIR);
1400 
1401 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1402 	if (error == 0) {
1403 		cache_setunresolved(ap->a_nch);
1404 		cache_setvp(ap->a_nch, *vpp);
1405 		tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1406 	}
1407 	return error;
1408 }
1409 
1410 /* --------------------------------------------------------------------- */
1411 
1412 static int
1413 tmpfs_nrmdir(struct vop_nrmdir_args *ap)
1414 {
1415 	struct vnode *dvp = ap->a_dvp;
1416 	struct namecache *ncp = ap->a_nch->ncp;
1417 	struct vnode *vp;
1418 	struct tmpfs_dirent *de;
1419 	struct tmpfs_mount *tmp;
1420 	struct tmpfs_node *dnode;
1421 	struct tmpfs_node *node;
1422 	int error;
1423 
1424 	/*
1425 	 * We have to acquire the vp from ap->a_nch because we will likely
1426 	 * unresolve the namecache entry, and a vrele/vput is needed to
1427 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1428 	 *
1429 	 * We have to use vget to clear any inactive state on the vnode,
1430 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1431 	 * will not get called when we release it.
1432 	 */
1433 	error = cache_vget(ap->a_nch, ap->a_cred, LK_SHARED, &vp);
1434 	KKASSERT(error == 0);
1435 	vn_unlock(vp);
1436 
1437 	/*
1438 	 * Prevalidate so we don't hit an assertion later
1439 	 */
1440 	if (vp->v_type != VDIR) {
1441 		error = ENOTDIR;
1442 		goto out;
1443 	}
1444 
1445 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1446 	dnode = VP_TO_TMPFS_DIR(dvp);
1447 	node = VP_TO_TMPFS_DIR(vp);
1448 
1449 	/*
1450 	 *
1451 	 */
1452 	TMPFS_NODE_LOCK(dnode);
1453 	TMPFS_NODE_LOCK(node);
1454 
1455 	/*
1456 	 * Only empty directories can be removed.
1457 	 */
1458 	if (node->tn_size > 0) {
1459 		error = ENOTEMPTY;
1460 		goto out_locked;
1461 	}
1462 
1463 	if ((dnode->tn_flags & APPEND)
1464 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1465 		error = EPERM;
1466 		goto out_locked;
1467 	}
1468 
1469 	/*
1470 	 * This invariant holds only if we are not trying to
1471 	 * remove "..".  We checked for that above so this is safe now.
1472 	 */
1473 	KKASSERT(node->tn_dir.tn_parent == dnode);
1474 
1475 	/*
1476 	 * Get the directory entry associated with node (vp)
1477 	 */
1478 	de = tmpfs_dir_lookup(dnode, node, ncp);
1479 	KKASSERT(TMPFS_DIRENT_MATCHES(de, ncp->nc_name, ncp->nc_nlen));
1480 
1481 	/* Check flags to see if we are allowed to remove the directory. */
1482 	if ((dnode->tn_flags & APPEND) ||
1483 	    node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1484 		error = EPERM;
1485 		goto out_locked;
1486 	}
1487 
1488 	/* Detach the directory entry from the directory (dnode). */
1489 	tmpfs_dir_detach(dnode, de);
1490 
1491 	/*
1492 	 * Must set parent linkage to NULL (tested by ncreate to disallow
1493 	 * the creation of new files/dirs in a deleted directory)
1494 	 */
1495 	node->tn_status |= TMPFS_NODE_CHANGED;
1496 
1497 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1498 			    TMPFS_NODE_MODIFIED;
1499 
1500 	/* Free the directory entry we just deleted.  Note that the node
1501 	 * referred by it will not be removed until the vnode is really
1502 	 * reclaimed. */
1503 	tmpfs_free_dirent(tmp, de);
1504 
1505 	/* Release the deleted vnode (will destroy the node, notify
1506 	 * interested parties and clean it from the cache). */
1507 
1508 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1509 
1510 	TMPFS_NODE_UNLOCK(node);
1511 	TMPFS_NODE_UNLOCK(dnode);
1512 
1513 	tmpfs_update(dvp);
1514 	cache_unlink(ap->a_nch);
1515 	tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1516 	vrele(vp);
1517 	return 0;
1518 
1519 out_locked:
1520 	TMPFS_NODE_UNLOCK(node);
1521 	TMPFS_NODE_UNLOCK(dnode);
1522 
1523 out:
1524 	vrele(vp);
1525 
1526 	return error;
1527 }
1528 
1529 /* --------------------------------------------------------------------- */
1530 
1531 static int
1532 tmpfs_nsymlink(struct vop_nsymlink_args *ap)
1533 {
1534 	struct vnode *dvp = ap->a_dvp;
1535 	struct vnode **vpp = ap->a_vpp;
1536 	struct namecache *ncp = ap->a_nch->ncp;
1537 	struct vattr *vap = ap->a_vap;
1538 	struct ucred *cred = ap->a_cred;
1539 	char *target = ap->a_target;
1540 	int error;
1541 
1542 	vap->va_type = VLNK;
1543 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1544 	if (error == 0) {
1545 		tmpfs_knote(*vpp, NOTE_WRITE);
1546 		cache_setunresolved(ap->a_nch);
1547 		cache_setvp(ap->a_nch, *vpp);
1548 	}
1549 	return error;
1550 }
1551 
1552 /* --------------------------------------------------------------------- */
1553 
1554 static int
1555 tmpfs_readdir(struct vop_readdir_args *ap)
1556 {
1557 	struct vnode *vp = ap->a_vp;
1558 	struct uio *uio = ap->a_uio;
1559 	int *eofflag = ap->a_eofflag;
1560 	off_t **cookies = ap->a_cookies;
1561 	int *ncookies = ap->a_ncookies;
1562 	struct tmpfs_mount *tmp;
1563 	int error;
1564 	off_t startoff;
1565 	off_t cnt = 0;
1566 	struct tmpfs_node *node;
1567 
1568 	/* This operation only makes sense on directory nodes. */
1569 	if (vp->v_type != VDIR) {
1570 		return ENOTDIR;
1571 	}
1572 
1573 	tmp = VFS_TO_TMPFS(vp->v_mount);
1574 	node = VP_TO_TMPFS_DIR(vp);
1575 	startoff = uio->uio_offset;
1576 
1577 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1578 		error = tmpfs_dir_getdotdent(node, uio);
1579 		if (error != 0) {
1580 			TMPFS_NODE_LOCK_SH(node);
1581 			goto outok;
1582 		}
1583 		cnt++;
1584 	}
1585 
1586 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1587 		/* may lock parent, cannot hold node lock */
1588 		error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1589 		if (error != 0) {
1590 			TMPFS_NODE_LOCK_SH(node);
1591 			goto outok;
1592 		}
1593 		cnt++;
1594 	}
1595 
1596 	TMPFS_NODE_LOCK_SH(node);
1597 	error = tmpfs_dir_getdents(node, uio, &cnt);
1598 
1599 outok:
1600 	KKASSERT(error >= -1);
1601 
1602 	if (error == -1)
1603 		error = 0;
1604 
1605 	if (eofflag != NULL)
1606 		*eofflag =
1607 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1608 
1609 	/* Update NFS-related variables. */
1610 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1611 		off_t i;
1612 		off_t off = startoff;
1613 		struct tmpfs_dirent *de = NULL;
1614 
1615 		*ncookies = cnt;
1616 		*cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1617 
1618 		for (i = 0; i < cnt; i++) {
1619 			KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1620 			if (off == TMPFS_DIRCOOKIE_DOT) {
1621 				off = TMPFS_DIRCOOKIE_DOTDOT;
1622 			} else {
1623 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1624 					de = RB_MIN(tmpfs_dirtree_cookie,
1625 						&node->tn_dir.tn_cookietree);
1626 				} else if (de != NULL) {
1627 					de = RB_NEXT(tmpfs_dirtree_cookie,
1628 					       &node->tn_dir.tn_cookietree, de);
1629 				} else {
1630 					de = tmpfs_dir_lookupbycookie(node,
1631 								      off);
1632 					KKASSERT(de != NULL);
1633 					de = RB_NEXT(tmpfs_dirtree_cookie,
1634 					       &node->tn_dir.tn_cookietree, de);
1635 				}
1636 				if (de == NULL)
1637 					off = TMPFS_DIRCOOKIE_EOF;
1638 				else
1639 					off = tmpfs_dircookie(de);
1640 			}
1641 			(*cookies)[i] = off;
1642 		}
1643 		KKASSERT(uio->uio_offset == off);
1644 	}
1645 	TMPFS_NODE_UNLOCK(node);
1646 
1647 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
1648 		TMPFS_NODE_LOCK(node);
1649 		node->tn_status |= TMPFS_NODE_ACCESSED;
1650 		TMPFS_NODE_UNLOCK(node);
1651 	}
1652 	return error;
1653 }
1654 
1655 /* --------------------------------------------------------------------- */
1656 
1657 static int
1658 tmpfs_readlink(struct vop_readlink_args *ap)
1659 {
1660 	struct vnode *vp = ap->a_vp;
1661 	struct uio *uio = ap->a_uio;
1662 	int error;
1663 	struct tmpfs_node *node;
1664 
1665 	KKASSERT(uio->uio_offset == 0);
1666 	KKASSERT(vp->v_type == VLNK);
1667 
1668 	node = VP_TO_TMPFS_NODE(vp);
1669 	TMPFS_NODE_LOCK_SH(node);
1670 	error = uiomove(node->tn_link,
1671 			MIN(node->tn_size, uio->uio_resid), uio);
1672 	TMPFS_NODE_UNLOCK(node);
1673 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
1674 		TMPFS_NODE_LOCK(node);
1675 		node->tn_status |= TMPFS_NODE_ACCESSED;
1676 		TMPFS_NODE_UNLOCK(node);
1677 	}
1678 	return error;
1679 }
1680 
1681 /* --------------------------------------------------------------------- */
1682 
1683 static int
1684 tmpfs_inactive(struct vop_inactive_args *ap)
1685 {
1686 	struct vnode *vp = ap->a_vp;
1687 	struct tmpfs_node *node;
1688 	struct mount *mp;
1689 
1690 	mp = vp->v_mount;
1691 	lwkt_gettoken(&mp->mnt_token);
1692 	node = VP_TO_TMPFS_NODE(vp);
1693 
1694 	/*
1695 	 * Degenerate case
1696 	 */
1697 	if (node == NULL) {
1698 		vrecycle(vp);
1699 		lwkt_reltoken(&mp->mnt_token);
1700 		return(0);
1701 	}
1702 
1703 	/*
1704 	 * Get rid of unreferenced deleted vnodes sooner rather than
1705 	 * later so the data memory can be recovered immediately.
1706 	 *
1707 	 * We must truncate the vnode to prevent the normal reclamation
1708 	 * path from flushing the data for the removed file to disk.
1709 	 */
1710 	TMPFS_NODE_LOCK(node);
1711 	if (node->tn_links == 0) {
1712 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1713 		TMPFS_NODE_UNLOCK(node);
1714 		if (node->tn_type == VREG)
1715 			tmpfs_truncate(vp, 0);
1716 		vrecycle(vp);
1717 	} else {
1718 		/*
1719 		 * We must retain any VM pages belonging to the vnode's
1720 		 * object as the vnode will destroy the object during a
1721 		 * later reclaim.  We call vinvalbuf(V_SAVE) to clean
1722 		 * out the buffer cache.
1723 		 *
1724 		 * On DragonFlyBSD, vnodes are not immediately deactivated
1725 		 * on the 1->0 refs, so this is a relatively optimal
1726 		 * operation.  We have to do this in tmpfs_inactive()
1727 		 * because the pages will have already been thrown away
1728 		 * at the time tmpfs_reclaim() is called.
1729 		 */
1730 		if (node->tn_type == VREG &&
1731 		    node->tn_reg.tn_pages_in_aobj == 0) {
1732 			vinvalbuf(vp, V_SAVE, 0, 0);
1733 			KKASSERT(RB_EMPTY(&vp->v_rbdirty_tree));
1734 			KKASSERT(RB_EMPTY(&vp->v_rbclean_tree));
1735 			tmpfs_move_pages(vp->v_object, node->tn_reg.tn_aobj,
1736 					 TMPFS_MOVF_DEACTIVATE);
1737 			node->tn_reg.tn_pages_in_aobj = 1;
1738 		}
1739 
1740 		TMPFS_NODE_UNLOCK(node);
1741 	}
1742 	lwkt_reltoken(&mp->mnt_token);
1743 
1744 	return 0;
1745 }
1746 
1747 /* --------------------------------------------------------------------- */
1748 
1749 int
1750 tmpfs_reclaim(struct vop_reclaim_args *ap)
1751 {
1752 	struct vnode *vp = ap->a_vp;
1753 	struct tmpfs_mount *tmp;
1754 	struct tmpfs_node *node;
1755 	struct mount *mp;
1756 
1757 	mp = vp->v_mount;
1758 	lwkt_gettoken(&mp->mnt_token);
1759 
1760 	node = VP_TO_TMPFS_NODE(vp);
1761 	tmp = VFS_TO_TMPFS(vp->v_mount);
1762 	KKASSERT(mp == tmp->tm_mount);
1763 
1764         TMPFS_NODE_LOCK(node);
1765 	KKASSERT(node->tn_vnode == vp);
1766         node->tn_vnode = NULL;
1767         vp->v_data = NULL;
1768 
1769 	/*
1770 	 * If the node referenced by this vnode was deleted by the
1771 	 * user, we must free its associated data structures now that
1772 	 * the vnode is being reclaimed.
1773 	 *
1774 	 * Directories have an extra link ref.
1775 	 */
1776 	if (node->tn_links == 0) {
1777 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1778 		tmpfs_free_node(tmp, node);
1779 		/* eats the lock */
1780 	} else {
1781 		TMPFS_NODE_UNLOCK(node);
1782 	}
1783 	lwkt_reltoken(&mp->mnt_token);
1784 
1785 	KKASSERT(vp->v_data == NULL);
1786 	return 0;
1787 }
1788 
1789 /* --------------------------------------------------------------------- */
1790 
1791 static int
1792 tmpfs_mountctl(struct vop_mountctl_args *ap)
1793 {
1794 	struct tmpfs_mount *tmp;
1795 	struct mount *mp;
1796 	int rc;
1797 
1798 	mp = ap->a_head.a_ops->head.vv_mount;
1799 	lwkt_gettoken(&mp->mnt_token);
1800 
1801 	switch (ap->a_op) {
1802 	case (MOUNTCTL_SET_EXPORT):
1803 		tmp = (struct tmpfs_mount *) mp->mnt_data;
1804 
1805 		if (ap->a_ctllen != sizeof(struct export_args))
1806 			rc = (EINVAL);
1807 		else
1808 			rc = vfs_export(mp, &tmp->tm_export,
1809 					(const struct export_args *) ap->a_ctl);
1810 		break;
1811 	default:
1812 		rc = vop_stdmountctl(ap);
1813 		break;
1814 	}
1815 
1816 	lwkt_reltoken(&mp->mnt_token);
1817 	return (rc);
1818 }
1819 
1820 /* --------------------------------------------------------------------- */
1821 
1822 static int
1823 tmpfs_print(struct vop_print_args *ap)
1824 {
1825 	struct vnode *vp = ap->a_vp;
1826 
1827 	struct tmpfs_node *node;
1828 
1829 	node = VP_TO_TMPFS_NODE(vp);
1830 
1831 	kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1832 	    node, node->tn_flags, node->tn_links);
1833 	kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1834 	    node->tn_mode, node->tn_uid, node->tn_gid,
1835 	    (uintmax_t)node->tn_size, node->tn_status);
1836 
1837 	if (vp->v_type == VFIFO)
1838 		fifo_printinfo(vp);
1839 
1840 	kprintf("\n");
1841 
1842 	return 0;
1843 }
1844 
1845 /* --------------------------------------------------------------------- */
1846 
1847 static int
1848 tmpfs_pathconf(struct vop_pathconf_args *ap)
1849 {
1850 	struct vnode *vp = ap->a_vp;
1851 	int name = ap->a_name;
1852 	register_t *retval = ap->a_retval;
1853 	struct tmpfs_mount *tmp;
1854 	int error;
1855 
1856 	error = 0;
1857 
1858 	switch (name) {
1859 	case _PC_CHOWN_RESTRICTED:
1860 		*retval = 1;
1861 		break;
1862 
1863 	case _PC_FILESIZEBITS:
1864 		tmp = VFS_TO_TMPFS(vp->v_mount);
1865 		*retval = max(32, flsll(tmp->tm_pages_max * PAGE_SIZE) + 1);
1866 		break;
1867 
1868 	case _PC_LINK_MAX:
1869 		*retval = LINK_MAX;
1870 		break;
1871 
1872 	case _PC_NAME_MAX:
1873 		*retval = NAME_MAX;
1874 		break;
1875 
1876 	case _PC_NO_TRUNC:
1877 		*retval = 1;
1878 		break;
1879 
1880 	case _PC_PATH_MAX:
1881 		*retval = PATH_MAX;
1882 		break;
1883 
1884 	case _PC_PIPE_BUF:
1885 		*retval = PIPE_BUF;
1886 		break;
1887 
1888 	case _PC_SYNC_IO:
1889 		*retval = 1;
1890 		break;
1891 
1892 	case _PC_2_SYMLINKS:
1893 		*retval = 1;
1894 		break;
1895 
1896 	default:
1897 		error = EINVAL;
1898 	}
1899 
1900 	return error;
1901 }
1902 
1903 /************************************************************************
1904  *                          KQFILTER OPS                                *
1905  ************************************************************************/
1906 
1907 static void filt_tmpfsdetach(struct knote *kn);
1908 static int filt_tmpfsread(struct knote *kn, long hint);
1909 static int filt_tmpfswrite(struct knote *kn, long hint);
1910 static int filt_tmpfsvnode(struct knote *kn, long hint);
1911 
1912 static struct filterops tmpfsread_filtops =
1913 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1914 	  NULL, filt_tmpfsdetach, filt_tmpfsread };
1915 static struct filterops tmpfswrite_filtops =
1916 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1917 	  NULL, filt_tmpfsdetach, filt_tmpfswrite };
1918 static struct filterops tmpfsvnode_filtops =
1919 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1920 	  NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1921 
1922 static int
1923 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1924 {
1925 	struct vnode *vp = ap->a_vp;
1926 	struct knote *kn = ap->a_kn;
1927 
1928 	switch (kn->kn_filter) {
1929 	case EVFILT_READ:
1930 		kn->kn_fop = &tmpfsread_filtops;
1931 		break;
1932 	case EVFILT_WRITE:
1933 		kn->kn_fop = &tmpfswrite_filtops;
1934 		break;
1935 	case EVFILT_VNODE:
1936 		kn->kn_fop = &tmpfsvnode_filtops;
1937 		break;
1938 	default:
1939 		return (EOPNOTSUPP);
1940 	}
1941 
1942 	kn->kn_hook = (caddr_t)vp;
1943 
1944 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1945 
1946 	return(0);
1947 }
1948 
1949 static void
1950 filt_tmpfsdetach(struct knote *kn)
1951 {
1952 	struct vnode *vp = (void *)kn->kn_hook;
1953 
1954 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1955 }
1956 
1957 static int
1958 filt_tmpfsread(struct knote *kn, long hint)
1959 {
1960 	struct vnode *vp = (void *)kn->kn_hook;
1961 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1962 	off_t off;
1963 
1964 	if (hint == NOTE_REVOKE) {
1965 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1966 		return(1);
1967 	}
1968 
1969 	/*
1970 	 * Interlock against MP races when performing this function.
1971 	 */
1972 	TMPFS_NODE_LOCK_SH(node);
1973 	off = node->tn_size - kn->kn_fp->f_offset;
1974 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1975 	if (kn->kn_sfflags & NOTE_OLDAPI) {
1976 		TMPFS_NODE_UNLOCK(node);
1977 		return(1);
1978 	}
1979 	if (kn->kn_data == 0) {
1980 		kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1981 	}
1982 	TMPFS_NODE_UNLOCK(node);
1983 	return (kn->kn_data != 0);
1984 }
1985 
1986 static int
1987 filt_tmpfswrite(struct knote *kn, long hint)
1988 {
1989 	if (hint == NOTE_REVOKE)
1990 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1991 	kn->kn_data = 0;
1992 	return (1);
1993 }
1994 
1995 static int
1996 filt_tmpfsvnode(struct knote *kn, long hint)
1997 {
1998 	if (kn->kn_sfflags & hint)
1999 		kn->kn_fflags |= hint;
2000 	if (hint == NOTE_REVOKE) {
2001 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2002 		return (1);
2003 	}
2004 	return (kn->kn_fflags != 0);
2005 }
2006 
2007 /*
2008  * Helper to move VM pages between objects
2009  *
2010  * NOTE: The vm_page_rename() dirties the page, so we can clear the
2011  *	 PG_NEED_COMMIT flag.  If the pages are being moved into tn_aobj,
2012  *	 the pageout daemon will be able to page them out.
2013  */
2014 static int
2015 tmpfs_move_pages_callback(vm_page_t p, void *data)
2016 {
2017 	struct rb_vm_page_scan_info *info = data;
2018 	vm_pindex_t pindex;
2019 
2020 	pindex = p->pindex;
2021 	if (vm_page_busy_try(p, TRUE)) {
2022 		vm_page_sleep_busy(p, TRUE, "tpgmov");
2023 		info->error = -1;
2024 		return -1;
2025 	}
2026 	if (p->object != info->object || p->pindex != pindex) {
2027 		vm_page_wakeup(p);
2028 		info->error = -1;
2029 		return -1;
2030 	}
2031 
2032 	if ((info->pagerflags & TMPFS_MOVF_FROMBACKING) &&
2033 	    (p->flags & PG_SWAPPED) &&
2034 	    (p->flags & PG_NEED_COMMIT) == 0 &&
2035 	    p->dirty == 0) {
2036 		/*
2037 		 * If the page in the backing aobj was paged out to swap
2038 		 * it will be clean and it is better to free it rather
2039 		 * than re-dirty it.  We will assume that the page was
2040 		 * paged out to swap for a reason!
2041 		 *
2042 		 * This helps avoid unnecessary swap thrashing on the page.
2043 		 */
2044 		vm_page_free(p);
2045 	} else if ((info->pagerflags & TMPFS_MOVF_FROMBACKING) == 0 &&
2046 		   (p->flags & PG_NEED_COMMIT) == 0 &&
2047 		   p->dirty == 0) {
2048 		/*
2049 		 * If the page associated with the vnode was cleaned via
2050 		 * a tmpfs_strategy() call, it exists as a swap block in
2051 		 * aobj and it is again better to free it rather than
2052 		 * re-dirty it.  We will assume that the page was
2053 		 * paged out to swap for a reason!
2054 		 *
2055 		 * This helps avoid unnecessary swap thrashing on the page.
2056 		 */
2057 		vm_page_free(p);
2058 	} else {
2059 		/*
2060 		 * Rename the page, which will also ensure that it is flagged
2061 		 * as dirty and check whether a swap block association exists
2062 		 * in the target object or not, setting appropriate flags if
2063 		 * it does.
2064 		 */
2065 		vm_page_rename(p, info->dest_object, pindex);
2066 		vm_page_clear_commit(p);
2067 		if (info->pagerflags & TMPFS_MOVF_DEACTIVATE)
2068 			vm_page_deactivate(p);
2069 		vm_page_wakeup(p);
2070 		/* page automaticaly made dirty */
2071 	}
2072 
2073 	return 0;
2074 }
2075 
2076 static
2077 void
2078 tmpfs_move_pages(vm_object_t src, vm_object_t dst, int movflags)
2079 {
2080 	struct rb_vm_page_scan_info info;
2081 
2082 	vm_object_hold(src);
2083 	vm_object_hold(dst);
2084 	info.object = src;
2085 	info.dest_object = dst;
2086 	info.pagerflags = movflags;
2087 	do {
2088 		if (src->paging_in_progress)
2089 			vm_object_pip_wait(src, "objtfs");
2090 		info.error = 1;
2091 		vm_page_rb_tree_RB_SCAN(&src->rb_memq, NULL,
2092 					tmpfs_move_pages_callback, &info);
2093 	} while (info.error < 0 || !RB_EMPTY(&src->rb_memq) ||
2094 		 src->paging_in_progress);
2095 	vm_object_drop(dst);
2096 	vm_object_drop(src);
2097 }
2098 
2099 /* --------------------------------------------------------------------- */
2100 
2101 /*
2102  * vnode operations vector used for files stored in a tmpfs file system.
2103  */
2104 struct vop_ops tmpfs_vnode_vops = {
2105 	.vop_default =			vop_defaultop,
2106 	.vop_getpages = 		vop_stdgetpages,
2107 	.vop_putpages = 		vop_stdputpages,
2108 	.vop_ncreate =			tmpfs_ncreate,
2109 	.vop_nresolve =			tmpfs_nresolve,
2110 	.vop_nlookupdotdot =		tmpfs_nlookupdotdot,
2111 	.vop_nmknod =			tmpfs_nmknod,
2112 	.vop_open =			tmpfs_open,
2113 	.vop_close =			tmpfs_close,
2114 	.vop_access =			tmpfs_access,
2115 	.vop_getattr =			tmpfs_getattr,
2116 	.vop_setattr =			tmpfs_setattr,
2117 	.vop_read =			tmpfs_read,
2118 	.vop_write =			tmpfs_write,
2119 	.vop_fsync =			tmpfs_fsync,
2120 	.vop_mountctl =			tmpfs_mountctl,
2121 	.vop_nremove =			tmpfs_nremove,
2122 	.vop_nlink =			tmpfs_nlink,
2123 	.vop_nrename =			tmpfs_nrename,
2124 	.vop_nmkdir =			tmpfs_nmkdir,
2125 	.vop_nrmdir =			tmpfs_nrmdir,
2126 	.vop_nsymlink =			tmpfs_nsymlink,
2127 	.vop_readdir =			tmpfs_readdir,
2128 	.vop_readlink =			tmpfs_readlink,
2129 	.vop_inactive =			tmpfs_inactive,
2130 	.vop_reclaim =			tmpfs_reclaim,
2131 	.vop_print =			tmpfs_print,
2132 	.vop_pathconf =			tmpfs_pathconf,
2133 	.vop_bmap =			tmpfs_bmap,
2134 	.vop_strategy =			tmpfs_strategy,
2135 	.vop_advlock =			tmpfs_advlock,
2136 	.vop_kqfilter =			tmpfs_kqfilter
2137 };
2138