xref: /dflybsd-src/sys/vfs/tmpfs/tmpfs_vnops.c (revision cfd59c5a2b3edde0431e7578d5c63f4600f8a262)
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 static 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, TMPFS_BLKSIZE, GETBLK_KVABIO);
556 		if (bp == NULL) {
557 			if (tmpfs_cluster_rd_enable) {
558 				error = cluster_readx(vp, node->tn_size,
559 						     base_offset,
560 						     TMPFS_BLKSIZE,
561 						     B_NOTMETA | B_KVABIO,
562 						     uio->uio_resid,
563 						     seqcount * MAXBSIZE,
564 						     &bp);
565 			} else {
566 				error = bread_kvabio(vp, base_offset,
567 						     TMPFS_BLKSIZE, &bp);
568 			}
569 			if (error) {
570 				brelse(bp);
571 				kprintf("tmpfs_read bread error %d\n", error);
572 				break;
573 			}
574 
575 			/*
576 			 * tmpfs pretty much fiddles directly with the VM
577 			 * system, don't let it exhaust it or we won't play
578 			 * nice with other processes.
579 			 *
580 			 * Only do this if the VOP is coming from a normal
581 			 * read/write.  The VM system handles the case for
582 			 * UIO_NOCOPY.
583 			 */
584 			if (uio->uio_segflg != UIO_NOCOPY)
585 				vm_wait_nominal();
586 		}
587 		bp->b_flags |= B_CLUSTEROK;
588 		bkvasync(bp);
589 
590 		/*
591 		 * Figure out how many bytes we can actually copy this loop.
592 		 */
593 		len = TMPFS_BLKSIZE - offset;
594 		if (len > uio->uio_resid)
595 			len = uio->uio_resid;
596 		if (len > node->tn_size - uio->uio_offset)
597 			len = (size_t)(node->tn_size - uio->uio_offset);
598 
599 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
600 		bqrelse(bp);
601 		if (error) {
602 			kprintf("tmpfs_read uiomove error %d\n", error);
603 			break;
604 		}
605 	}
606 
607 finished:
608 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
609 		TMPFS_NODE_LOCK(node);
610 		node->tn_status |= TMPFS_NODE_ACCESSED;
611 		TMPFS_NODE_UNLOCK(node);
612 	}
613 	return (error);
614 }
615 
616 static int
617 tmpfs_write(struct vop_write_args *ap)
618 {
619 	struct buf *bp;
620 	struct vnode *vp = ap->a_vp;
621 	struct uio *uio = ap->a_uio;
622 	struct thread *td = uio->uio_td;
623 	struct tmpfs_node *node;
624 	boolean_t extended;
625 	off_t oldsize;
626 	int error;
627 	off_t base_offset;
628 	size_t offset;
629 	size_t len;
630 	struct rlimit limit;
631 	int trivial = 0;
632 	int kflags = 0;
633 	int seqcount;
634 
635 	error = 0;
636 	if (uio->uio_resid == 0) {
637 		return error;
638 	}
639 
640 	node = VP_TO_TMPFS_NODE(vp);
641 
642 	if (vp->v_type != VREG)
643 		return (EINVAL);
644 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
645 
646 	TMPFS_NODE_LOCK(node);
647 
648 	/*
649 	 * restore any saved pages before proceeding
650 	 */
651 	if (node->tn_reg.tn_pages_in_aobj) {
652 		tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object,
653 				 TMPFS_MOVF_FROMBACKING);
654 		node->tn_reg.tn_pages_in_aobj = 0;
655 	}
656 
657 	oldsize = node->tn_size;
658 	if (ap->a_ioflag & IO_APPEND)
659 		uio->uio_offset = node->tn_size;
660 
661 	/*
662 	 * Check for illegal write offsets.
663 	 */
664 	if (uio->uio_offset + uio->uio_resid >
665 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
666 		error = EFBIG;
667 		goto done;
668 	}
669 
670 	/*
671 	 * NOTE: Ignore if UIO does not come from a user thread (e.g. VN).
672 	 */
673 	if (vp->v_type == VREG && td != NULL && td->td_lwp != NULL) {
674 		error = kern_getrlimit(RLIMIT_FSIZE, &limit);
675 		if (error)
676 			goto done;
677 		if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
678 			ksignal(td->td_proc, SIGXFSZ);
679 			error = EFBIG;
680 			goto done;
681 		}
682 	}
683 
684 	/*
685 	 * Extend the file's size if necessary
686 	 */
687 	extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
688 
689 	while (uio->uio_resid > 0) {
690 		/*
691 		 * Don't completely blow out running buffer I/O
692 		 * when being hit from the pageout daemon.
693 		 */
694 		if (uio->uio_segflg == UIO_NOCOPY &&
695 		    (ap->a_ioflag & IO_RECURSE) == 0) {
696 			bwillwrite(TMPFS_BLKSIZE);
697 		}
698 
699 		/*
700 		 * Use buffer cache I/O (via tmpfs_strategy)
701 		 */
702 		offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64;
703 		base_offset = (off_t)uio->uio_offset - offset;
704 		len = TMPFS_BLKSIZE - offset;
705 		if (len > uio->uio_resid)
706 			len = uio->uio_resid;
707 
708 		if ((uio->uio_offset + len) > node->tn_size) {
709 			trivial = (uio->uio_offset <= node->tn_size);
710 			error = tmpfs_reg_resize(vp, uio->uio_offset + len,
711 						 trivial);
712 			if (error)
713 				break;
714 		}
715 
716 		/*
717 		 * Read to fill in any gaps.  Theoretically we could
718 		 * optimize this if the write covers the entire buffer
719 		 * and is not a UIO_NOCOPY write, however this can lead
720 		 * to a security violation exposing random kernel memory
721 		 * (whatever junk was in the backing VM pages before).
722 		 *
723 		 * So just use bread() to do the right thing.
724 		 */
725 		error = bread_kvabio(vp, base_offset, TMPFS_BLKSIZE, &bp);
726 		bkvasync(bp);
727 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
728 		if (error) {
729 			kprintf("tmpfs_write uiomove error %d\n", error);
730 			brelse(bp);
731 			break;
732 		}
733 
734 		if (uio->uio_offset > node->tn_size) {
735 			node->tn_size = uio->uio_offset;
736 			kflags |= NOTE_EXTEND;
737 		}
738 		kflags |= NOTE_WRITE;
739 
740 		/*
741 		 * Always try to flush the page in the UIO_NOCOPY case.  This
742 		 * can come from the pageout daemon or during vnode eviction.
743 		 * It is not necessarily going to be marked IO_ASYNC/IO_SYNC.
744 		 *
745 		 * For the normal case we buwrite(), dirtying the underlying
746 		 * VM pages instead of dirtying the buffer and releasing the
747 		 * buffer as a clean buffer.  This allows tmpfs to use
748 		 * essentially all available memory to cache file data.
749 		 * If we used bdwrite() the buffer cache would wind up
750 		 * flushing the data to swap too quickly.
751 		 *
752 		 * But because tmpfs can seriously load the VM system we
753 		 * fall-back to using bdwrite() when free memory starts
754 		 * to get low.  This shifts the load away from the VM system
755 		 * and makes tmpfs act more like a normal filesystem with
756 		 * regards to disk activity.
757 		 *
758 		 * tmpfs pretty much fiddles directly with the VM
759 		 * system, don't let it exhaust it or we won't play
760 		 * nice with other processes.  Only do this if the
761 		 * VOP is coming from a normal read/write.  The VM system
762 		 * handles the case for UIO_NOCOPY.
763 		 */
764 		bp->b_flags |= B_CLUSTEROK;
765 		if (uio->uio_segflg == UIO_NOCOPY) {
766 			/*
767 			 * Flush from the pageout daemon, deal with
768 			 * potentially very heavy tmpfs write activity
769 			 * causing long stalls in the pageout daemon
770 			 * before pages get to free/cache.
771 			 *
772 			 * (a) Under severe pressure setting B_DIRECT will
773 			 *     cause a buffer release to try to free the
774 			 *     underlying pages.
775 			 *
776 			 * (b) Under modest memory pressure the B_RELBUF
777 			 *     alone is sufficient to get the pages moved
778 			 *     to the cache.  We could also force this by
779 			 *     setting B_NOTMETA but that might have other
780 			 *     unintended side-effects (e.g. setting
781 			 *     PG_NOTMETA on the VM page).
782 			 *
783 			 * Hopefully this will unblock the VM system more
784 			 * quickly under extreme tmpfs write load.
785 			 */
786 			if (vm_page_count_min(vm_page_free_hysteresis))
787 				bp->b_flags |= B_DIRECT;
788 			bp->b_flags |= B_AGE | B_RELBUF;
789 			bp->b_act_count = 0;	/* buffer->deactivate pgs */
790 			cluster_awrite(bp);
791 		} else if (vm_pages_needed || vm_paging_needed(0) ||
792 			   tmpfs_bufcache_mode >= 2) {
793 			/*
794 			 * If the pageout daemon is running we cycle the
795 			 * write through the buffer cache normally to
796 			 * pipeline the flush, thus avoiding adding any
797 			 * more memory pressure to the pageout daemon.
798 			 */
799 			bp->b_act_count = 0;	/* buffer->deactivate pgs */
800 			if (tmpfs_cluster_wr_enable) {
801 				cluster_write(bp, node->tn_size,
802 					      TMPFS_BLKSIZE, seqcount);
803 			} else {
804 				bdwrite(bp);
805 			}
806 		} else {
807 			/*
808 			 * Otherwise run the buffer directly through to the
809 			 * backing VM store, leaving the buffer clean so
810 			 * buffer limits do not force early flushes to swap.
811 			 */
812 			buwrite(bp);
813 			/*vm_wait_nominal();*/
814 		}
815 
816 		if (bp->b_error) {
817 			kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
818 			break;
819 		}
820 	}
821 
822 	if (error) {
823 		if (extended) {
824 			(void)tmpfs_reg_resize(vp, oldsize, trivial);
825 			kflags &= ~NOTE_EXTEND;
826 		}
827 		goto done;
828 	}
829 
830 	/*
831 	 * Currently we don't set the mtime on files modified via mmap()
832 	 * because we can't tell the difference between those modifications
833 	 * and an attempt by the pageout daemon to flush tmpfs pages to
834 	 * swap.
835 	 *
836 	 * This is because in order to defer flushes as long as possible
837 	 * buwrite() works by marking the underlying VM pages dirty in
838 	 * order to be able to dispose of the buffer cache buffer without
839 	 * flushing it.
840 	 */
841 	if (uio->uio_segflg == UIO_NOCOPY) {
842 		if (vp->v_flag & VLASTWRITETS) {
843 			node->tn_mtime = vp->v_lastwrite_ts.tv_sec;
844 			node->tn_mtimensec = vp->v_lastwrite_ts.tv_nsec;
845 		}
846 	} else {
847 		node->tn_status |= TMPFS_NODE_MODIFIED;
848 		vclrflags(vp, VLASTWRITETS);
849 	}
850 
851 	if (extended)
852 		node->tn_status |= TMPFS_NODE_CHANGED;
853 
854 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
855 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
856 			node->tn_mode &= ~(S_ISUID | S_ISGID);
857 	}
858 done:
859 	TMPFS_NODE_UNLOCK(node);
860 	if (kflags)
861 		tmpfs_knote(vp, kflags);
862 
863 	return(error);
864 }
865 
866 static int
867 tmpfs_advlock(struct vop_advlock_args *ap)
868 {
869 	struct tmpfs_node *node;
870 	struct vnode *vp = ap->a_vp;
871 	int error;
872 
873 	node = VP_TO_TMPFS_NODE(vp);
874 	error = (lf_advlock(ap, &node->tn_advlock, node->tn_size));
875 
876 	return (error);
877 }
878 
879 /*
880  * The strategy function is typically only called when memory pressure
881  * forces the system to attempt to pageout pages.  It can also be called
882  * by [n]vtruncbuf() when a truncation cuts a page in half.  Normal write
883  * operations
884  *
885  * We set VKVABIO for VREG files so bp->b_data may not be synchronized to
886  * our cpu.  swap_pager_strategy() is all we really use, and it directly
887  * supports this.
888  */
889 static int
890 tmpfs_strategy(struct vop_strategy_args *ap)
891 {
892 	struct bio *bio = ap->a_bio;
893 	struct bio *nbio;
894 	struct buf *bp = bio->bio_buf;
895 	struct vnode *vp = ap->a_vp;
896 	struct tmpfs_node *node;
897 	vm_object_t uobj;
898 	vm_page_t m;
899 	int i;
900 
901 	if (vp->v_type != VREG) {
902 		bp->b_resid = bp->b_bcount;
903 		bp->b_flags |= B_ERROR | B_INVAL;
904 		bp->b_error = EINVAL;
905 		biodone(bio);
906 		return(0);
907 	}
908 
909 	node = VP_TO_TMPFS_NODE(vp);
910 
911 	uobj = node->tn_reg.tn_aobj;
912 
913 	/*
914 	 * Don't bother flushing to swap if there is no swap, just
915 	 * ensure that the pages are marked as needing a commit (still).
916 	 */
917 	if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
918 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
919 			m = bp->b_xio.xio_pages[i];
920 			vm_page_need_commit(m);
921 		}
922 		bp->b_resid = 0;
923 		bp->b_error = 0;
924 		biodone(bio);
925 	} else {
926 		/*
927 		 * Tell the buffer cache to try to recycle the pages
928 		 * to PQ_CACHE on release.
929 		 */
930 		if (tmpfs_bufcache_mode >= 2 ||
931 		    (tmpfs_bufcache_mode == 1 && vm_paging_needed(0))) {
932 			bp->b_flags |= B_TTC;
933 		}
934 		nbio = push_bio(bio);
935 		nbio->bio_done = tmpfs_strategy_done;
936 		nbio->bio_offset = bio->bio_offset;
937 		swap_pager_strategy(uobj, nbio);
938 	}
939 	return 0;
940 }
941 
942 /*
943  * If we were unable to commit the pages to swap make sure they are marked
944  * as needing a commit (again).  If we were, clear the flag to allow the
945  * pages to be freed.
946  *
947  * Do not error-out the buffer.  In particular, vinvalbuf() needs to
948  * always work.
949  */
950 static void
951 tmpfs_strategy_done(struct bio *bio)
952 {
953 	struct buf *bp;
954 	vm_page_t m;
955 	int i;
956 
957 	bp = bio->bio_buf;
958 
959 	if (bp->b_flags & B_ERROR) {
960 		bp->b_flags &= ~B_ERROR;
961 		bp->b_error = 0;
962 		bp->b_resid = 0;
963 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
964 			m = bp->b_xio.xio_pages[i];
965 			vm_page_need_commit(m);
966 		}
967 	} else {
968 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
969 			m = bp->b_xio.xio_pages[i];
970 			vm_page_clear_commit(m);
971 		}
972 	}
973 	bio = pop_bio(bio);
974 	biodone(bio);
975 }
976 
977 static int
978 tmpfs_bmap(struct vop_bmap_args *ap)
979 {
980 	if (ap->a_doffsetp != NULL)
981 		*ap->a_doffsetp = ap->a_loffset;
982 	if (ap->a_runp != NULL)
983 		*ap->a_runp = 0;
984 	if (ap->a_runb != NULL)
985 		*ap->a_runb = 0;
986 
987 	return 0;
988 }
989 
990 /* --------------------------------------------------------------------- */
991 
992 static int
993 tmpfs_nremove(struct vop_nremove_args *ap)
994 {
995 	struct vnode *dvp = ap->a_dvp;
996 	struct namecache *ncp = ap->a_nch->ncp;
997 	struct vnode *vp;
998 	int error;
999 	struct tmpfs_dirent *de;
1000 	struct tmpfs_mount *tmp;
1001 	struct tmpfs_node *dnode;
1002 	struct tmpfs_node *node;
1003 
1004 	/*
1005 	 * We have to acquire the vp from ap->a_nch because we will likely
1006 	 * unresolve the namecache entry, and a vrele/vput is needed to
1007 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1008 	 *
1009 	 * We have to use vget to clear any inactive state on the vnode,
1010 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1011 	 * will not get called when we release it.
1012 	 */
1013 	error = cache_vget(ap->a_nch, ap->a_cred, LK_SHARED, &vp);
1014 	KKASSERT(vp->v_mount == dvp->v_mount);
1015 	KKASSERT(error == 0);
1016 	vn_unlock(vp);
1017 
1018 	if (vp->v_type == VDIR) {
1019 		error = EISDIR;
1020 		goto out2;
1021 	}
1022 
1023 	dnode = VP_TO_TMPFS_DIR(dvp);
1024 	node = VP_TO_TMPFS_NODE(vp);
1025 	tmp = VFS_TO_TMPFS(vp->v_mount);
1026 
1027 	TMPFS_NODE_LOCK(dnode);
1028 	de = tmpfs_dir_lookup(dnode, node, ncp);
1029 	if (de == NULL) {
1030 		error = ENOENT;
1031 		TMPFS_NODE_UNLOCK(dnode);
1032 		goto out;
1033 	}
1034 
1035 	/* Files marked as immutable or append-only cannot be deleted. */
1036 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
1037 	    (dnode->tn_flags & APPEND)) {
1038 		error = EPERM;
1039 		TMPFS_NODE_UNLOCK(dnode);
1040 		goto out;
1041 	}
1042 
1043 	/* Remove the entry from the directory; as it is a file, we do not
1044 	 * have to change the number of hard links of the directory. */
1045 	tmpfs_dir_detach(dnode, de);
1046 	TMPFS_NODE_UNLOCK(dnode);
1047 
1048 	/* Free the directory entry we just deleted.  Note that the node
1049 	 * referred by it will not be removed until the vnode is really
1050 	 * reclaimed. */
1051 	tmpfs_free_dirent(tmp, de);
1052 
1053 	if (node->tn_links > 0) {
1054 	        TMPFS_NODE_LOCK(node);
1055 		node->tn_status |= TMPFS_NODE_CHANGED;
1056 	        TMPFS_NODE_UNLOCK(node);
1057 	}
1058 
1059 	cache_unlink(ap->a_nch);
1060 	tmpfs_knote(vp, NOTE_DELETE);
1061 	error = 0;
1062 
1063 out:
1064 	if (error == 0)
1065 		tmpfs_knote(dvp, NOTE_WRITE);
1066 out2:
1067 	vrele(vp);
1068 
1069 	return error;
1070 }
1071 
1072 /* --------------------------------------------------------------------- */
1073 
1074 static int
1075 tmpfs_nlink(struct vop_nlink_args *ap)
1076 {
1077 	struct vnode *dvp = ap->a_dvp;
1078 	struct vnode *vp = ap->a_vp;
1079 	struct namecache *ncp = ap->a_nch->ncp;
1080 	struct tmpfs_dirent *de;
1081 	struct tmpfs_node *node;
1082 	struct tmpfs_node *dnode;
1083 	int error;
1084 
1085 	KKASSERT(dvp != vp); /* XXX When can this be false? */
1086 
1087 	node = VP_TO_TMPFS_NODE(vp);
1088 	dnode = VP_TO_TMPFS_NODE(dvp);
1089 	TMPFS_NODE_LOCK(dnode);
1090 
1091 	/* XXX: Why aren't the following two tests done by the caller? */
1092 
1093 	/* Hard links of directories are forbidden. */
1094 	if (vp->v_type == VDIR) {
1095 		error = EPERM;
1096 		goto out;
1097 	}
1098 
1099 	/* Cannot create cross-device links. */
1100 	if (dvp->v_mount != vp->v_mount) {
1101 		error = EXDEV;
1102 		goto out;
1103 	}
1104 
1105 	/* Ensure that we do not overflow the maximum number of links imposed
1106 	 * by the system. */
1107 	KKASSERT(node->tn_links <= LINK_MAX);
1108 	if (node->tn_links >= LINK_MAX) {
1109 		error = EMLINK;
1110 		goto out;
1111 	}
1112 
1113 	/* We cannot create links of files marked immutable or append-only. */
1114 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
1115 		error = EPERM;
1116 		goto out;
1117 	}
1118 
1119 	/* Allocate a new directory entry to represent the node. */
1120 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
1121 				   ncp->nc_name, ncp->nc_nlen, &de);
1122 	if (error != 0)
1123 		goto out;
1124 
1125 	/* Insert the new directory entry into the appropriate directory. */
1126 	tmpfs_dir_attach(dnode, de);
1127 
1128 	/* vp link count has changed, so update node times. */
1129 
1130 	TMPFS_NODE_LOCK(node);
1131 	node->tn_status |= TMPFS_NODE_CHANGED;
1132 	TMPFS_NODE_UNLOCK(node);
1133 	tmpfs_update(vp);
1134 
1135 	tmpfs_knote(vp, NOTE_LINK);
1136 	cache_setunresolved(ap->a_nch);
1137 	cache_setvp(ap->a_nch, vp);
1138 	error = 0;
1139 
1140 out:
1141 	TMPFS_NODE_UNLOCK(dnode);
1142 	if (error == 0)
1143 		tmpfs_knote(dvp, NOTE_WRITE);
1144 	return error;
1145 }
1146 
1147 /* --------------------------------------------------------------------- */
1148 
1149 static int
1150 tmpfs_nrename(struct vop_nrename_args *ap)
1151 {
1152 	struct vnode *fdvp = ap->a_fdvp;
1153 	struct namecache *fncp = ap->a_fnch->ncp;
1154 	struct vnode *fvp = fncp->nc_vp;
1155 	struct vnode *tdvp = ap->a_tdvp;
1156 	struct namecache *tncp = ap->a_tnch->ncp;
1157 	struct vnode *tvp;
1158 	struct tmpfs_dirent *de, *tde;
1159 	struct tmpfs_mount *tmp;
1160 	struct tmpfs_node *fdnode;
1161 	struct tmpfs_node *fnode;
1162 	struct tmpfs_node *tnode;
1163 	struct tmpfs_node *tdnode;
1164 	char *newname;
1165 	char *oldname;
1166 	int error;
1167 
1168 	KKASSERT(fdvp->v_mount == fvp->v_mount);
1169 
1170 	/*
1171 	 * Because tvp can get overwritten we have to vget it instead of
1172 	 * just vref or use it, otherwise it's VINACTIVE flag may not get
1173 	 * cleared and the node won't get destroyed.
1174 	 */
1175 	error = cache_vget(ap->a_tnch, ap->a_cred, LK_SHARED, &tvp);
1176 	if (error == 0) {
1177 		tnode = VP_TO_TMPFS_NODE(tvp);
1178 		vn_unlock(tvp);
1179 	} else {
1180 		tnode = NULL;
1181 	}
1182 
1183 	/* Disallow cross-device renames.
1184 	 * XXX Why isn't this done by the caller? */
1185 	if (fvp->v_mount != tdvp->v_mount ||
1186 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
1187 		error = EXDEV;
1188 		goto out;
1189 	}
1190 
1191 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
1192 	tdnode = VP_TO_TMPFS_DIR(tdvp);
1193 
1194 	/* If source and target are the same file, there is nothing to do. */
1195 	if (fvp == tvp) {
1196 		error = 0;
1197 		goto out;
1198 	}
1199 
1200 	fdnode = VP_TO_TMPFS_DIR(fdvp);
1201 	fnode = VP_TO_TMPFS_NODE(fvp);
1202 	TMPFS_NODE_LOCK(fdnode);
1203 	de = tmpfs_dir_lookup(fdnode, fnode, fncp);
1204 	TMPFS_NODE_UNLOCK(fdnode);	/* XXX depend on namecache lock */
1205 
1206 	/* Avoid manipulating '.' and '..' entries. */
1207 	if (de == NULL) {
1208 		error = ENOENT;
1209 		goto out_locked;
1210 	}
1211 	KKASSERT(de->td_node == fnode);
1212 
1213 	/*
1214 	 * If replacing an entry in the target directory and that entry
1215 	 * is a directory, it must be empty.
1216 	 *
1217 	 * Kern_rename gurantees the destination to be a directory
1218 	 * if the source is one (it does?).
1219 	 */
1220 	if (tvp != NULL) {
1221 		KKASSERT(tnode != NULL);
1222 
1223 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1224 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1225 			error = EPERM;
1226 			goto out_locked;
1227 		}
1228 
1229 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1230 			if (tnode->tn_size > 0) {
1231 				error = ENOTEMPTY;
1232 				goto out_locked;
1233 			}
1234 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1235 			error = ENOTDIR;
1236 			goto out_locked;
1237 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1238 			error = EISDIR;
1239 			goto out_locked;
1240 		} else {
1241 			KKASSERT(fnode->tn_type != VDIR &&
1242 				tnode->tn_type != VDIR);
1243 		}
1244 	}
1245 
1246 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1247 	    (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1248 		error = EPERM;
1249 		goto out_locked;
1250 	}
1251 
1252 	/*
1253 	 * Ensure that we have enough memory to hold the new name, if it
1254 	 * has to be changed.
1255 	 */
1256 	if (fncp->nc_nlen != tncp->nc_nlen ||
1257 	    bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1258 		newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1259 				  M_WAITOK | M_NULLOK);
1260 		if (newname == NULL) {
1261 			error = ENOSPC;
1262 			goto out_locked;
1263 		}
1264 		bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1265 		newname[tncp->nc_nlen] = '\0';
1266 	} else {
1267 		newname = NULL;
1268 	}
1269 
1270 	/*
1271 	 * Unlink entry from source directory.  Note that the kernel has
1272 	 * already checked for illegal recursion cases (renaming a directory
1273 	 * into a subdirectory of itself).
1274 	 */
1275 	if (fdnode != tdnode) {
1276 		tmpfs_dir_detach(fdnode, de);
1277 	} else {
1278 		/* XXX depend on namecache lock */
1279 		TMPFS_NODE_LOCK(fdnode);
1280 		KKASSERT(de == tmpfs_dir_lookup(fdnode, fnode, fncp));
1281 		RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de);
1282 		RB_REMOVE(tmpfs_dirtree_cookie,
1283 			  &fdnode->tn_dir.tn_cookietree, de);
1284 		TMPFS_NODE_UNLOCK(fdnode);
1285 	}
1286 
1287 	/*
1288 	 * Handle any name change.  Swap with newname, we will
1289 	 * deallocate it at the end.
1290 	 */
1291 	if (newname != NULL) {
1292 #if 0
1293 		TMPFS_NODE_LOCK(fnode);
1294 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1295 		TMPFS_NODE_UNLOCK(fnode);
1296 #endif
1297 		oldname = de->td_name;
1298 		de->td_name = newname;
1299 		de->td_namelen = (uint16_t)tncp->nc_nlen;
1300 		newname = oldname;
1301 	}
1302 
1303 	/*
1304 	 * If we are overwriting an entry, we have to remove the old one
1305 	 * from the target directory.
1306 	 */
1307 	if (tvp != NULL) {
1308 		/* Remove the old entry from the target directory. */
1309 		TMPFS_NODE_LOCK(tdnode);
1310 		tde = tmpfs_dir_lookup(tdnode, tnode, tncp);
1311 		tmpfs_dir_detach(tdnode, tde);
1312 		TMPFS_NODE_UNLOCK(tdnode);
1313 		tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1314 
1315 		/*
1316 		 * Free the directory entry we just deleted.  Note that the
1317 		 * node referred by it will not be removed until the vnode is
1318 		 * really reclaimed.
1319 		 */
1320 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1321 		/*cache_inval_vp(tvp, CINV_DESTROY);*/
1322 	}
1323 
1324 	/*
1325 	 * Link entry to target directory.  If the entry
1326 	 * represents a directory move the parent linkage
1327 	 * as well.
1328 	 */
1329 	if (fdnode != tdnode) {
1330 		if (de->td_node->tn_type == VDIR) {
1331 			TMPFS_VALIDATE_DIR(fnode);
1332 		}
1333 		tmpfs_dir_attach(tdnode, de);
1334 	} else {
1335 		TMPFS_NODE_LOCK(tdnode);
1336 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1337 		RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de);
1338 		RB_INSERT(tmpfs_dirtree_cookie,
1339 			  &tdnode->tn_dir.tn_cookietree, de);
1340 		TMPFS_NODE_UNLOCK(tdnode);
1341 	}
1342 
1343 	/*
1344 	 * Finish up
1345 	 */
1346 	if (newname) {
1347 		kfree(newname, tmp->tm_name_zone);
1348 		newname = NULL;
1349 	}
1350 	cache_rename(ap->a_fnch, ap->a_tnch);
1351 	tmpfs_knote(ap->a_fdvp, NOTE_WRITE);
1352 	tmpfs_knote(ap->a_tdvp, NOTE_WRITE);
1353 	if (fnode->tn_vnode)
1354 		tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1355 	error = 0;
1356 
1357 out_locked:
1358 	;
1359 out:
1360 	if (tvp)
1361 		vrele(tvp);
1362 	return error;
1363 }
1364 
1365 /* --------------------------------------------------------------------- */
1366 
1367 static int
1368 tmpfs_nmkdir(struct vop_nmkdir_args *ap)
1369 {
1370 	struct vnode *dvp = ap->a_dvp;
1371 	struct vnode **vpp = ap->a_vpp;
1372 	struct namecache *ncp = ap->a_nch->ncp;
1373 	struct vattr *vap = ap->a_vap;
1374 	struct ucred *cred = ap->a_cred;
1375 	int error;
1376 
1377 	KKASSERT(vap->va_type == VDIR);
1378 
1379 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1380 	if (error == 0) {
1381 		cache_setunresolved(ap->a_nch);
1382 		cache_setvp(ap->a_nch, *vpp);
1383 		tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1384 	}
1385 	return error;
1386 }
1387 
1388 /* --------------------------------------------------------------------- */
1389 
1390 static int
1391 tmpfs_nrmdir(struct vop_nrmdir_args *ap)
1392 {
1393 	struct vnode *dvp = ap->a_dvp;
1394 	struct namecache *ncp = ap->a_nch->ncp;
1395 	struct vnode *vp;
1396 	struct tmpfs_dirent *de;
1397 	struct tmpfs_mount *tmp;
1398 	struct tmpfs_node *dnode;
1399 	struct tmpfs_node *node;
1400 	int error;
1401 
1402 	/*
1403 	 * We have to acquire the vp from ap->a_nch because we will likely
1404 	 * unresolve the namecache entry, and a vrele/vput is needed to
1405 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1406 	 *
1407 	 * We have to use vget to clear any inactive state on the vnode,
1408 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1409 	 * will not get called when we release it.
1410 	 */
1411 	error = cache_vget(ap->a_nch, ap->a_cred, LK_SHARED, &vp);
1412 	KKASSERT(error == 0);
1413 	vn_unlock(vp);
1414 
1415 	/*
1416 	 * Prevalidate so we don't hit an assertion later
1417 	 */
1418 	if (vp->v_type != VDIR) {
1419 		error = ENOTDIR;
1420 		goto out;
1421 	}
1422 
1423 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1424 	dnode = VP_TO_TMPFS_DIR(dvp);
1425 	node = VP_TO_TMPFS_DIR(vp);
1426 
1427 	/*
1428 	 * Directories with more than two entries ('.' and '..') cannot
1429 	 * be removed.
1430 	 */
1431 	if (node->tn_size > 0) {
1432 		error = ENOTEMPTY;
1433 		goto out;
1434 	}
1435 
1436 	if ((dnode->tn_flags & APPEND)
1437 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1438 		error = EPERM;
1439 		goto out;
1440 	}
1441 
1442 	/*
1443 	 * This invariant holds only if we are not trying to
1444 	 * remove "..".  We checked for that above so this is safe now.
1445 	 */
1446 	KKASSERT(node->tn_dir.tn_parent == dnode);
1447 
1448 	/*
1449 	 * Get the directory entry associated with node (vp).  This
1450 	 * was filled by tmpfs_lookup while looking up the entry.
1451 	 */
1452 	TMPFS_NODE_LOCK(dnode);
1453 	de = tmpfs_dir_lookup(dnode, node, ncp);
1454 	KKASSERT(TMPFS_DIRENT_MATCHES(de, ncp->nc_name, ncp->nc_nlen));
1455 
1456 	/* Check flags to see if we are allowed to remove the directory. */
1457 	if ((dnode->tn_flags & APPEND) ||
1458 	    node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1459 		error = EPERM;
1460 		TMPFS_NODE_UNLOCK(dnode);
1461 		goto out;
1462 	}
1463 
1464 	/* Detach the directory entry from the directory (dnode). */
1465 	tmpfs_dir_detach(dnode, de);
1466 	TMPFS_NODE_UNLOCK(dnode);
1467 
1468 	/* No vnode should be allocated for this entry from this point */
1469 	TMPFS_NODE_LOCK(dnode);
1470 	TMPFS_ASSERT_ELOCKED(dnode);
1471 	TMPFS_NODE_LOCK(node);
1472 	TMPFS_ASSERT_ELOCKED(node);
1473 
1474 	/*
1475 	 * Must set parent linkage to NULL (tested by ncreate to disallow
1476 	 * the creation of new files/dirs in a deleted directory)
1477 	 */
1478 	node->tn_status |= TMPFS_NODE_CHANGED;
1479 
1480 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1481 			    TMPFS_NODE_MODIFIED;
1482 
1483 	TMPFS_NODE_UNLOCK(node);
1484 	TMPFS_NODE_UNLOCK(dnode);
1485 
1486 	/* Free the directory entry we just deleted.  Note that the node
1487 	 * referred by it will not be removed until the vnode is really
1488 	 * reclaimed. */
1489 	tmpfs_free_dirent(tmp, de);
1490 
1491 	/* Release the deleted vnode (will destroy the node, notify
1492 	 * interested parties and clean it from the cache). */
1493 
1494 	TMPFS_NODE_LOCK(dnode);
1495 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1496 	TMPFS_NODE_UNLOCK(dnode);
1497 	tmpfs_update(dvp);
1498 
1499 	cache_unlink(ap->a_nch);
1500 	tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1501 	error = 0;
1502 
1503 out:
1504 	vrele(vp);
1505 
1506 	return error;
1507 }
1508 
1509 /* --------------------------------------------------------------------- */
1510 
1511 static int
1512 tmpfs_nsymlink(struct vop_nsymlink_args *ap)
1513 {
1514 	struct vnode *dvp = ap->a_dvp;
1515 	struct vnode **vpp = ap->a_vpp;
1516 	struct namecache *ncp = ap->a_nch->ncp;
1517 	struct vattr *vap = ap->a_vap;
1518 	struct ucred *cred = ap->a_cred;
1519 	char *target = ap->a_target;
1520 	int error;
1521 
1522 	vap->va_type = VLNK;
1523 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1524 	if (error == 0) {
1525 		tmpfs_knote(*vpp, NOTE_WRITE);
1526 		cache_setunresolved(ap->a_nch);
1527 		cache_setvp(ap->a_nch, *vpp);
1528 	}
1529 	return error;
1530 }
1531 
1532 /* --------------------------------------------------------------------- */
1533 
1534 static int
1535 tmpfs_readdir(struct vop_readdir_args *ap)
1536 {
1537 	struct vnode *vp = ap->a_vp;
1538 	struct uio *uio = ap->a_uio;
1539 	int *eofflag = ap->a_eofflag;
1540 	off_t **cookies = ap->a_cookies;
1541 	int *ncookies = ap->a_ncookies;
1542 	struct tmpfs_mount *tmp;
1543 	int error;
1544 	off_t startoff;
1545 	off_t cnt = 0;
1546 	struct tmpfs_node *node;
1547 
1548 	/* This operation only makes sense on directory nodes. */
1549 	if (vp->v_type != VDIR) {
1550 		return ENOTDIR;
1551 	}
1552 
1553 	tmp = VFS_TO_TMPFS(vp->v_mount);
1554 	node = VP_TO_TMPFS_DIR(vp);
1555 	startoff = uio->uio_offset;
1556 
1557 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1558 		error = tmpfs_dir_getdotdent(node, uio);
1559 		if (error != 0) {
1560 			TMPFS_NODE_LOCK_SH(node);
1561 			goto outok;
1562 		}
1563 		cnt++;
1564 	}
1565 
1566 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1567 		/* may lock parent, cannot hold node lock */
1568 		error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1569 		if (error != 0) {
1570 			TMPFS_NODE_LOCK_SH(node);
1571 			goto outok;
1572 		}
1573 		cnt++;
1574 	}
1575 
1576 	TMPFS_NODE_LOCK_SH(node);
1577 	error = tmpfs_dir_getdents(node, uio, &cnt);
1578 
1579 outok:
1580 	KKASSERT(error >= -1);
1581 
1582 	if (error == -1)
1583 		error = 0;
1584 
1585 	if (eofflag != NULL)
1586 		*eofflag =
1587 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1588 
1589 	/* Update NFS-related variables. */
1590 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1591 		off_t i;
1592 		off_t off = startoff;
1593 		struct tmpfs_dirent *de = NULL;
1594 
1595 		*ncookies = cnt;
1596 		*cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1597 
1598 		for (i = 0; i < cnt; i++) {
1599 			KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1600 			if (off == TMPFS_DIRCOOKIE_DOT) {
1601 				off = TMPFS_DIRCOOKIE_DOTDOT;
1602 			} else {
1603 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1604 					de = RB_MIN(tmpfs_dirtree_cookie,
1605 						&node->tn_dir.tn_cookietree);
1606 				} else if (de != NULL) {
1607 					de = RB_NEXT(tmpfs_dirtree_cookie,
1608 					       &node->tn_dir.tn_cookietree, de);
1609 				} else {
1610 					de = tmpfs_dir_lookupbycookie(node,
1611 								      off);
1612 					KKASSERT(de != NULL);
1613 					de = RB_NEXT(tmpfs_dirtree_cookie,
1614 					       &node->tn_dir.tn_cookietree, de);
1615 				}
1616 				if (de == NULL)
1617 					off = TMPFS_DIRCOOKIE_EOF;
1618 				else
1619 					off = tmpfs_dircookie(de);
1620 			}
1621 			(*cookies)[i] = off;
1622 		}
1623 		KKASSERT(uio->uio_offset == off);
1624 	}
1625 	TMPFS_NODE_UNLOCK(node);
1626 
1627 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
1628 		TMPFS_NODE_LOCK(node);
1629 		node->tn_status |= TMPFS_NODE_ACCESSED;
1630 		TMPFS_NODE_UNLOCK(node);
1631 	}
1632 	return error;
1633 }
1634 
1635 /* --------------------------------------------------------------------- */
1636 
1637 static int
1638 tmpfs_readlink(struct vop_readlink_args *ap)
1639 {
1640 	struct vnode *vp = ap->a_vp;
1641 	struct uio *uio = ap->a_uio;
1642 	int error;
1643 	struct tmpfs_node *node;
1644 
1645 	KKASSERT(uio->uio_offset == 0);
1646 	KKASSERT(vp->v_type == VLNK);
1647 
1648 	node = VP_TO_TMPFS_NODE(vp);
1649 	TMPFS_NODE_LOCK_SH(node);
1650 	error = uiomove(node->tn_link,
1651 			MIN(node->tn_size, uio->uio_resid), uio);
1652 	TMPFS_NODE_UNLOCK(node);
1653 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
1654 		TMPFS_NODE_LOCK(node);
1655 		node->tn_status |= TMPFS_NODE_ACCESSED;
1656 		TMPFS_NODE_UNLOCK(node);
1657 	}
1658 	return error;
1659 }
1660 
1661 /* --------------------------------------------------------------------- */
1662 
1663 static int
1664 tmpfs_inactive(struct vop_inactive_args *ap)
1665 {
1666 	struct vnode *vp = ap->a_vp;
1667 	struct tmpfs_node *node;
1668 	struct mount *mp;
1669 
1670 	mp = vp->v_mount;
1671 	lwkt_gettoken(&mp->mnt_token);
1672 	node = VP_TO_TMPFS_NODE(vp);
1673 
1674 	/*
1675 	 * Degenerate case
1676 	 */
1677 	if (node == NULL) {
1678 		vrecycle(vp);
1679 		lwkt_reltoken(&mp->mnt_token);
1680 		return(0);
1681 	}
1682 
1683 	/*
1684 	 * Get rid of unreferenced deleted vnodes sooner rather than
1685 	 * later so the data memory can be recovered immediately.
1686 	 *
1687 	 * We must truncate the vnode to prevent the normal reclamation
1688 	 * path from flushing the data for the removed file to disk.
1689 	 */
1690 	TMPFS_NODE_LOCK(node);
1691 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1692 	    node->tn_links == 0)
1693 	{
1694 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1695 		TMPFS_NODE_UNLOCK(node);
1696 		if (node->tn_type == VREG)
1697 			tmpfs_truncate(vp, 0);
1698 		vrecycle(vp);
1699 	} else {
1700 		/*
1701 		 * We must retain any VM pages belonging to the vnode's
1702 		 * object as the vnode will destroy the object during a
1703 		 * later reclaim.  We call vinvalbuf(V_SAVE) to clean
1704 		 * out the buffer cache.
1705 		 *
1706 		 * On DragonFlyBSD, vnodes are not immediately deactivated
1707 		 * on the 1->0 refs, so this is a relatively optimal
1708 		 * operation.  We have to do this in tmpfs_inactive()
1709 		 * because the pages will have already been thrown away
1710 		 * at the time tmpfs_reclaim() is called.
1711 		 */
1712 		if (node->tn_type == VREG &&
1713 		    node->tn_reg.tn_pages_in_aobj == 0) {
1714 			vinvalbuf(vp, V_SAVE, 0, 0);
1715 			KKASSERT(RB_EMPTY(&vp->v_rbdirty_tree));
1716 			KKASSERT(RB_EMPTY(&vp->v_rbclean_tree));
1717 			tmpfs_move_pages(vp->v_object, node->tn_reg.tn_aobj,
1718 					 TMPFS_MOVF_DEACTIVATE);
1719 			node->tn_reg.tn_pages_in_aobj = 1;
1720 		}
1721 
1722 		TMPFS_NODE_UNLOCK(node);
1723 	}
1724 	lwkt_reltoken(&mp->mnt_token);
1725 
1726 	return 0;
1727 }
1728 
1729 /* --------------------------------------------------------------------- */
1730 
1731 int
1732 tmpfs_reclaim(struct vop_reclaim_args *ap)
1733 {
1734 	struct vnode *vp = ap->a_vp;
1735 	struct tmpfs_mount *tmp;
1736 	struct tmpfs_node *node;
1737 	struct mount *mp;
1738 
1739 	mp = vp->v_mount;
1740 	lwkt_gettoken(&mp->mnt_token);
1741 
1742 	node = VP_TO_TMPFS_NODE(vp);
1743 	tmp = VFS_TO_TMPFS(vp->v_mount);
1744 	KKASSERT(mp == tmp->tm_mount);
1745 
1746         TMPFS_NODE_LOCK(node);
1747 	KKASSERT(node->tn_vnode == vp);
1748         node->tn_vnode = NULL;
1749         vp->v_data = NULL;
1750 
1751 	/*
1752 	 * If the node referenced by this vnode was deleted by the
1753 	 * user, we must free its associated data structures now that
1754 	 * the vnode is being reclaimed.
1755 	 *
1756 	 * Directories have an extra link ref.
1757 	 */
1758 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1759 	    node->tn_links == 0) {
1760 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1761 		tmpfs_free_node(tmp, node);
1762 		/* eats the lock */
1763 	} else {
1764 		TMPFS_NODE_UNLOCK(node);
1765 	}
1766 	lwkt_reltoken(&mp->mnt_token);
1767 
1768 	KKASSERT(vp->v_data == NULL);
1769 	return 0;
1770 }
1771 
1772 /* --------------------------------------------------------------------- */
1773 
1774 static int
1775 tmpfs_mountctl(struct vop_mountctl_args *ap)
1776 {
1777 	struct tmpfs_mount *tmp;
1778 	struct mount *mp;
1779 	int rc;
1780 
1781 	mp = ap->a_head.a_ops->head.vv_mount;
1782 	lwkt_gettoken(&mp->mnt_token);
1783 
1784 	switch (ap->a_op) {
1785 	case (MOUNTCTL_SET_EXPORT):
1786 		tmp = (struct tmpfs_mount *) mp->mnt_data;
1787 
1788 		if (ap->a_ctllen != sizeof(struct export_args))
1789 			rc = (EINVAL);
1790 		else
1791 			rc = vfs_export(mp, &tmp->tm_export,
1792 					(const struct export_args *) ap->a_ctl);
1793 		break;
1794 	default:
1795 		rc = vop_stdmountctl(ap);
1796 		break;
1797 	}
1798 
1799 	lwkt_reltoken(&mp->mnt_token);
1800 	return (rc);
1801 }
1802 
1803 /* --------------------------------------------------------------------- */
1804 
1805 static int
1806 tmpfs_print(struct vop_print_args *ap)
1807 {
1808 	struct vnode *vp = ap->a_vp;
1809 
1810 	struct tmpfs_node *node;
1811 
1812 	node = VP_TO_TMPFS_NODE(vp);
1813 
1814 	kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1815 	    node, node->tn_flags, node->tn_links);
1816 	kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1817 	    node->tn_mode, node->tn_uid, node->tn_gid,
1818 	    (uintmax_t)node->tn_size, node->tn_status);
1819 
1820 	if (vp->v_type == VFIFO)
1821 		fifo_printinfo(vp);
1822 
1823 	kprintf("\n");
1824 
1825 	return 0;
1826 }
1827 
1828 /* --------------------------------------------------------------------- */
1829 
1830 static int
1831 tmpfs_pathconf(struct vop_pathconf_args *ap)
1832 {
1833 	struct vnode *vp = ap->a_vp;
1834 	int name = ap->a_name;
1835 	register_t *retval = ap->a_retval;
1836 	struct tmpfs_mount *tmp;
1837 	int error;
1838 
1839 	error = 0;
1840 
1841 	switch (name) {
1842 	case _PC_CHOWN_RESTRICTED:
1843 		*retval = 1;
1844 		break;
1845 
1846 	case _PC_FILESIZEBITS:
1847 		tmp = VFS_TO_TMPFS(vp->v_mount);
1848 		*retval = max(32, flsll(tmp->tm_pages_max * PAGE_SIZE) + 1);
1849 		break;
1850 
1851 	case _PC_LINK_MAX:
1852 		*retval = LINK_MAX;
1853 		break;
1854 
1855 	case _PC_NAME_MAX:
1856 		*retval = NAME_MAX;
1857 		break;
1858 
1859 	case _PC_NO_TRUNC:
1860 		*retval = 1;
1861 		break;
1862 
1863 	case _PC_PATH_MAX:
1864 		*retval = PATH_MAX;
1865 		break;
1866 
1867 	case _PC_PIPE_BUF:
1868 		*retval = PIPE_BUF;
1869 		break;
1870 
1871 	case _PC_SYNC_IO:
1872 		*retval = 1;
1873 		break;
1874 
1875 	case _PC_2_SYMLINKS:
1876 		*retval = 1;
1877 		break;
1878 
1879 	default:
1880 		error = EINVAL;
1881 	}
1882 
1883 	return error;
1884 }
1885 
1886 /************************************************************************
1887  *                          KQFILTER OPS                                *
1888  ************************************************************************/
1889 
1890 static void filt_tmpfsdetach(struct knote *kn);
1891 static int filt_tmpfsread(struct knote *kn, long hint);
1892 static int filt_tmpfswrite(struct knote *kn, long hint);
1893 static int filt_tmpfsvnode(struct knote *kn, long hint);
1894 
1895 static struct filterops tmpfsread_filtops =
1896 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1897 	  NULL, filt_tmpfsdetach, filt_tmpfsread };
1898 static struct filterops tmpfswrite_filtops =
1899 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1900 	  NULL, filt_tmpfsdetach, filt_tmpfswrite };
1901 static struct filterops tmpfsvnode_filtops =
1902 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1903 	  NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1904 
1905 static int
1906 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1907 {
1908 	struct vnode *vp = ap->a_vp;
1909 	struct knote *kn = ap->a_kn;
1910 
1911 	switch (kn->kn_filter) {
1912 	case EVFILT_READ:
1913 		kn->kn_fop = &tmpfsread_filtops;
1914 		break;
1915 	case EVFILT_WRITE:
1916 		kn->kn_fop = &tmpfswrite_filtops;
1917 		break;
1918 	case EVFILT_VNODE:
1919 		kn->kn_fop = &tmpfsvnode_filtops;
1920 		break;
1921 	default:
1922 		return (EOPNOTSUPP);
1923 	}
1924 
1925 	kn->kn_hook = (caddr_t)vp;
1926 
1927 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1928 
1929 	return(0);
1930 }
1931 
1932 static void
1933 filt_tmpfsdetach(struct knote *kn)
1934 {
1935 	struct vnode *vp = (void *)kn->kn_hook;
1936 
1937 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1938 }
1939 
1940 static int
1941 filt_tmpfsread(struct knote *kn, long hint)
1942 {
1943 	struct vnode *vp = (void *)kn->kn_hook;
1944 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1945 	off_t off;
1946 
1947 	if (hint == NOTE_REVOKE) {
1948 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1949 		return(1);
1950 	}
1951 
1952 	/*
1953 	 * Interlock against MP races when performing this function.
1954 	 */
1955 	TMPFS_NODE_LOCK_SH(node);
1956 	off = node->tn_size - kn->kn_fp->f_offset;
1957 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1958 	if (kn->kn_sfflags & NOTE_OLDAPI) {
1959 		TMPFS_NODE_UNLOCK(node);
1960 		return(1);
1961 	}
1962 	if (kn->kn_data == 0) {
1963 		kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1964 	}
1965 	TMPFS_NODE_UNLOCK(node);
1966 	return (kn->kn_data != 0);
1967 }
1968 
1969 static int
1970 filt_tmpfswrite(struct knote *kn, long hint)
1971 {
1972 	if (hint == NOTE_REVOKE)
1973 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1974 	kn->kn_data = 0;
1975 	return (1);
1976 }
1977 
1978 static int
1979 filt_tmpfsvnode(struct knote *kn, long hint)
1980 {
1981 	if (kn->kn_sfflags & hint)
1982 		kn->kn_fflags |= hint;
1983 	if (hint == NOTE_REVOKE) {
1984 		kn->kn_flags |= (EV_EOF | EV_NODATA);
1985 		return (1);
1986 	}
1987 	return (kn->kn_fflags != 0);
1988 }
1989 
1990 /*
1991  * Helper to move VM pages between objects
1992  *
1993  * NOTE: The vm_page_rename() dirties the page, so we can clear the
1994  *	 PG_NEED_COMMIT flag.  If the pages are being moved into tn_aobj,
1995  *	 the pageout daemon will be able to page them out.
1996  */
1997 static int
1998 tmpfs_move_pages_callback(vm_page_t p, void *data)
1999 {
2000 	struct rb_vm_page_scan_info *info = data;
2001 	vm_pindex_t pindex;
2002 
2003 	pindex = p->pindex;
2004 	if (vm_page_busy_try(p, TRUE)) {
2005 		vm_page_sleep_busy(p, TRUE, "tpgmov");
2006 		info->error = -1;
2007 		return -1;
2008 	}
2009 	if (p->object != info->object || p->pindex != pindex) {
2010 		vm_page_wakeup(p);
2011 		info->error = -1;
2012 		return -1;
2013 	}
2014 
2015 	if ((info->pagerflags & TMPFS_MOVF_FROMBACKING) &&
2016 	    (p->flags & PG_SWAPPED) &&
2017 	    (p->flags & PG_NEED_COMMIT) == 0 &&
2018 	    p->dirty == 0) {
2019 		/*
2020 		 * If the page in the backing aobj was paged out to swap
2021 		 * it will be clean and it is better to free it rather
2022 		 * than re-dirty it.  We will assume that the page was
2023 		 * paged out to swap for a reason!
2024 		 *
2025 		 * This helps avoid unnecessary swap thrashing on the page.
2026 		 */
2027 		vm_page_free(p);
2028 	} else if ((info->pagerflags & TMPFS_MOVF_FROMBACKING) == 0 &&
2029 		   (p->flags & PG_NEED_COMMIT) == 0 &&
2030 		   p->dirty == 0) {
2031 		/*
2032 		 * If the page associated with the vnode was cleaned via
2033 		 * a tmpfs_strategy() call, it exists as a swap block in
2034 		 * aobj and it is again better to free it rather than
2035 		 * re-dirty it.  We will assume that the page was
2036 		 * paged out to swap for a reason!
2037 		 *
2038 		 * This helps avoid unnecessary swap thrashing on the page.
2039 		 */
2040 		vm_page_free(p);
2041 	} else {
2042 		/*
2043 		 * Rename the page, which will also ensure that it is flagged
2044 		 * as dirty and check whether a swap block association exists
2045 		 * in the target object or not, setting appropriate flags if
2046 		 * it does.
2047 		 */
2048 		vm_page_rename(p, info->dest_object, pindex);
2049 		vm_page_clear_commit(p);
2050 		if (info->pagerflags & TMPFS_MOVF_DEACTIVATE)
2051 			vm_page_deactivate(p);
2052 		vm_page_wakeup(p);
2053 		/* page automaticaly made dirty */
2054 	}
2055 
2056 	return 0;
2057 }
2058 
2059 static
2060 void
2061 tmpfs_move_pages(vm_object_t src, vm_object_t dst, int movflags)
2062 {
2063 	struct rb_vm_page_scan_info info;
2064 
2065 	vm_object_hold(src);
2066 	vm_object_hold(dst);
2067 	info.object = src;
2068 	info.dest_object = dst;
2069 	info.pagerflags = movflags;
2070 	do {
2071 		if (src->paging_in_progress)
2072 			vm_object_pip_wait(src, "objtfs");
2073 		info.error = 1;
2074 		vm_page_rb_tree_RB_SCAN(&src->rb_memq, NULL,
2075 					tmpfs_move_pages_callback, &info);
2076 	} while (info.error < 0 || !RB_EMPTY(&src->rb_memq) ||
2077 		 src->paging_in_progress);
2078 	vm_object_drop(dst);
2079 	vm_object_drop(src);
2080 }
2081 
2082 /* --------------------------------------------------------------------- */
2083 
2084 /*
2085  * vnode operations vector used for files stored in a tmpfs file system.
2086  */
2087 struct vop_ops tmpfs_vnode_vops = {
2088 	.vop_default =			vop_defaultop,
2089 	.vop_getpages = 		vop_stdgetpages,
2090 	.vop_putpages = 		vop_stdputpages,
2091 	.vop_ncreate =			tmpfs_ncreate,
2092 	.vop_nresolve =			tmpfs_nresolve,
2093 	.vop_nlookupdotdot =		tmpfs_nlookupdotdot,
2094 	.vop_nmknod =			tmpfs_nmknod,
2095 	.vop_open =			tmpfs_open,
2096 	.vop_close =			tmpfs_close,
2097 	.vop_access =			tmpfs_access,
2098 	.vop_getattr =			tmpfs_getattr,
2099 	.vop_setattr =			tmpfs_setattr,
2100 	.vop_read =			tmpfs_read,
2101 	.vop_write =			tmpfs_write,
2102 	.vop_fsync =			tmpfs_fsync,
2103 	.vop_mountctl =			tmpfs_mountctl,
2104 	.vop_nremove =			tmpfs_nremove,
2105 	.vop_nlink =			tmpfs_nlink,
2106 	.vop_nrename =			tmpfs_nrename,
2107 	.vop_nmkdir =			tmpfs_nmkdir,
2108 	.vop_nrmdir =			tmpfs_nrmdir,
2109 	.vop_nsymlink =			tmpfs_nsymlink,
2110 	.vop_readdir =			tmpfs_readdir,
2111 	.vop_readlink =			tmpfs_readlink,
2112 	.vop_inactive =			tmpfs_inactive,
2113 	.vop_reclaim =			tmpfs_reclaim,
2114 	.vop_print =			tmpfs_print,
2115 	.vop_pathconf =			tmpfs_pathconf,
2116 	.vop_bmap =			tmpfs_bmap,
2117 	.vop_strategy =			tmpfs_strategy,
2118 	.vop_advlock =			tmpfs_advlock,
2119 	.vop_kqfilter =			tmpfs_kqfilter
2120 };
2121