xref: /dflybsd-src/sys/vfs/tmpfs/tmpfs_vnops.c (revision ff837cd5d97df9cfc8d66fc9ebc73557657d023e)
1 /*-
2  * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
3  * All rights reserved.
4  *
5  * This code is derived from software contributed to The NetBSD Foundation
6  * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
7  * 2005 program.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
22  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28  * POSSIBILITY OF SUCH DAMAGE.
29  *
30  * $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $
31  */
32 
33 /*
34  * tmpfs vnode interface.
35  */
36 
37 #include <sys/kernel.h>
38 #include <sys/kern_syscall.h>
39 #include <sys/param.h>
40 #include <sys/fcntl.h>
41 #include <sys/lockf.h>
42 #include <sys/priv.h>
43 #include <sys/proc.h>
44 #include <sys/resourcevar.h>
45 #include <sys/sched.h>
46 #include <sys/stat.h>
47 #include <sys/systm.h>
48 #include <sys/unistd.h>
49 #include <sys/vfsops.h>
50 #include <sys/vnode.h>
51 #include <sys/mountctl.h>
52 
53 #include <vm/vm.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_page.h>
57 #include <vm/vm_pageout.h>
58 #include <vm/vm_pager.h>
59 #include <vm/swap_pager.h>
60 
61 #include <sys/buf2.h>
62 #include <vm/vm_page2.h>
63 
64 #include <vfs/fifofs/fifo.h>
65 #include <vfs/tmpfs/tmpfs_vnops.h>
66 #if 0
67 #include <vfs/tmpfs/tmpfs.h>
68 #endif
69 #include "tmpfs.h"
70 
71 static void tmpfs_strategy_done(struct bio *bio);
72 
73 static __inline
74 void
75 tmpfs_knote(struct vnode *vp, int flags)
76 {
77 	if (flags)
78 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
79 }
80 
81 
82 /* --------------------------------------------------------------------- */
83 
84 static int
85 tmpfs_nresolve(struct vop_nresolve_args *v)
86 {
87 	struct vnode *dvp = v->a_dvp;
88 	struct vnode *vp = NULL;
89 	struct namecache *ncp = v->a_nch->ncp;
90 	struct tmpfs_node *tnode;
91 	struct mount *mp;
92 	struct tmpfs_dirent *de;
93 	struct tmpfs_node *dnode;
94 	int error;
95 
96 	mp = dvp->v_mount;
97 
98 	dnode = VP_TO_TMPFS_DIR(dvp);
99 
100 	TMPFS_NODE_LOCK_SH(dnode);
101 	de = tmpfs_dir_lookup(dnode, NULL, ncp);
102 	if (de == NULL) {
103 		error = ENOENT;
104 	} else {
105 		/*
106 		 * Allocate a vnode for the node we found.
107 		 */
108 		tnode = de->td_node;
109 		error = tmpfs_alloc_vp(dvp->v_mount, tnode,
110 				       LK_EXCLUSIVE | LK_RETRY, &vp);
111 		if (error)
112 			goto out;
113 		KKASSERT(vp);
114 	}
115 
116 out:
117 	TMPFS_NODE_UNLOCK(dnode);
118 
119 	if ((dnode->tn_status & TMPFS_NODE_ACCESSED) == 0) {
120 		TMPFS_NODE_LOCK(dnode);
121 		dnode->tn_status |= TMPFS_NODE_ACCESSED;
122 		TMPFS_NODE_UNLOCK(dnode);
123 	}
124 
125 	/*
126 	 * Store the result of this lookup in the cache.  Avoid this if the
127 	 * request was for creation, as it does not improve timings on
128 	 * emprical tests.
129 	 */
130 	if (vp) {
131 		vn_unlock(vp);
132 		cache_setvp(v->a_nch, vp);
133 		vrele(vp);
134 	} else if (error == ENOENT) {
135 		cache_setvp(v->a_nch, NULL);
136 	}
137 	return (error);
138 }
139 
140 static int
141 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
142 {
143 	struct vnode *dvp = v->a_dvp;
144 	struct vnode **vpp = v->a_vpp;
145 	struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
146 	struct ucred *cred = v->a_cred;
147 	struct mount *mp;
148 	int error;
149 
150 	*vpp = NULL;
151 
152 	mp = dvp->v_mount;
153 
154 	/* Check accessibility of requested node as a first step. */
155 	error = VOP_ACCESS(dvp, VEXEC, cred);
156 	if (error != 0)
157 		return error;
158 
159 	if (dnode->tn_dir.tn_parent != NULL) {
160 		/* Allocate a new vnode on the matching entry. */
161 		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
162 				       LK_EXCLUSIVE | LK_RETRY, vpp);
163 
164 		if (*vpp)
165 			vn_unlock(*vpp);
166 	}
167 	return (*vpp == NULL) ? ENOENT : 0;
168 }
169 
170 /* --------------------------------------------------------------------- */
171 
172 static int
173 tmpfs_ncreate(struct vop_ncreate_args *v)
174 {
175 	struct vnode *dvp = v->a_dvp;
176 	struct vnode **vpp = v->a_vpp;
177 	struct namecache *ncp = v->a_nch->ncp;
178 	struct vattr *vap = v->a_vap;
179 	struct ucred *cred = v->a_cred;
180 	struct mount *mp;
181 	int error;
182 
183 	mp = dvp->v_mount;
184 
185 	KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
186 
187 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
188 	if (error == 0) {
189 		cache_setunresolved(v->a_nch);
190 		cache_setvp(v->a_nch, *vpp);
191 		tmpfs_knote(dvp, NOTE_WRITE);
192 	}
193 	return (error);
194 }
195 /* --------------------------------------------------------------------- */
196 
197 static int
198 tmpfs_nmknod(struct vop_nmknod_args *v)
199 {
200 	struct vnode *dvp = v->a_dvp;
201 	struct vnode **vpp = v->a_vpp;
202 	struct namecache *ncp = v->a_nch->ncp;
203 	struct vattr *vap = v->a_vap;
204 	struct ucred *cred = v->a_cred;
205 	int error;
206 
207 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
208 	    vap->va_type != VFIFO) {
209 		return (EINVAL);
210 	}
211 
212 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
213 	if (error == 0) {
214 		cache_setunresolved(v->a_nch);
215 		cache_setvp(v->a_nch, *vpp);
216 		tmpfs_knote(dvp, NOTE_WRITE);
217 	}
218 	return error;
219 }
220 
221 /* --------------------------------------------------------------------- */
222 
223 static int
224 tmpfs_open(struct vop_open_args *v)
225 {
226 	struct vnode *vp = v->a_vp;
227 	int mode = v->a_mode;
228 	struct tmpfs_node *node;
229 	int error;
230 
231 	node = VP_TO_TMPFS_NODE(vp);
232 
233 #if 0
234 	/* The file is still active but all its names have been removed
235 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
236 	 * it is about to die. */
237 	if (node->tn_links < 1)
238 		return (ENOENT);
239 #endif
240 
241 	/* If the file is marked append-only, deny write requests. */
242 	if ((node->tn_flags & APPEND) &&
243 	    (mode & (FWRITE | O_APPEND)) == FWRITE) {
244 		error = EPERM;
245 	} else {
246 		error = (vop_stdopen(v));
247 	}
248 
249 	return (error);
250 }
251 
252 /* --------------------------------------------------------------------- */
253 
254 static int
255 tmpfs_close(struct vop_close_args *v)
256 {
257 	struct vnode *vp = v->a_vp;
258 	struct tmpfs_node *node;
259 	int error;
260 
261 	node = VP_TO_TMPFS_NODE(vp);
262 
263 	if (node->tn_links > 0) {
264 		/*
265 		 * Update node times.  No need to do it if the node has
266 		 * been deleted, because it will vanish after we return.
267 		 */
268 		tmpfs_update(vp);
269 	}
270 
271 	error = vop_stdclose(v);
272 
273 	return (error);
274 }
275 
276 /* --------------------------------------------------------------------- */
277 
278 int
279 tmpfs_access(struct vop_access_args *v)
280 {
281 	struct vnode *vp = v->a_vp;
282 	int error;
283 	struct tmpfs_node *node;
284 
285 	node = VP_TO_TMPFS_NODE(vp);
286 
287 	switch (vp->v_type) {
288 	case VDIR:
289 		/* FALLTHROUGH */
290 	case VLNK:
291 		/* FALLTHROUGH */
292 	case VREG:
293 		if ((v->a_mode & VWRITE) &&
294 	            (vp->v_mount->mnt_flag & MNT_RDONLY)) {
295 			error = EROFS;
296 			goto out;
297 		}
298 		break;
299 
300 	case VBLK:
301 		/* FALLTHROUGH */
302 	case VCHR:
303 		/* FALLTHROUGH */
304 	case VSOCK:
305 		/* FALLTHROUGH */
306 	case VFIFO:
307 		break;
308 
309 	default:
310 		error = EINVAL;
311 		goto out;
312 	}
313 
314 	if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
315 		error = EPERM;
316 		goto out;
317 	}
318 
319 	error = vop_helper_access(v, node->tn_uid, node->tn_gid,
320 			          node->tn_mode, 0);
321 out:
322 	return error;
323 }
324 
325 /* --------------------------------------------------------------------- */
326 
327 int
328 tmpfs_getattr(struct vop_getattr_args *v)
329 {
330 	struct vnode *vp = v->a_vp;
331 	struct vattr *vap = v->a_vap;
332 	struct tmpfs_node *node;
333 
334 	node = VP_TO_TMPFS_NODE(vp);
335 
336 	tmpfs_update(vp);
337 
338 	TMPFS_NODE_LOCK_SH(node);
339 	vap->va_type = vp->v_type;
340 	vap->va_mode = node->tn_mode;
341 	vap->va_nlink = node->tn_links;
342 	vap->va_uid = node->tn_uid;
343 	vap->va_gid = node->tn_gid;
344 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
345 	vap->va_fileid = node->tn_id;
346 	vap->va_size = node->tn_size;
347 	vap->va_blocksize = PAGE_SIZE;
348 	vap->va_atime.tv_sec = node->tn_atime;
349 	vap->va_atime.tv_nsec = node->tn_atimensec;
350 	vap->va_mtime.tv_sec = node->tn_mtime;
351 	vap->va_mtime.tv_nsec = node->tn_mtimensec;
352 	vap->va_ctime.tv_sec = node->tn_ctime;
353 	vap->va_ctime.tv_nsec = node->tn_ctimensec;
354 	vap->va_gen = node->tn_gen;
355 	vap->va_flags = node->tn_flags;
356 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
357 		vap->va_rmajor = umajor(node->tn_rdev);
358 		vap->va_rminor = uminor(node->tn_rdev);
359 	}
360 	vap->va_bytes = round_page(node->tn_size);
361 	vap->va_filerev = 0;
362 	TMPFS_NODE_UNLOCK(node);
363 
364 	return 0;
365 }
366 
367 /* --------------------------------------------------------------------- */
368 
369 int
370 tmpfs_setattr(struct vop_setattr_args *v)
371 {
372 	struct vnode *vp = v->a_vp;
373 	struct vattr *vap = v->a_vap;
374 	struct ucred *cred = v->a_cred;
375 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
376 	int error = 0;
377 	int kflags = 0;
378 
379 	TMPFS_NODE_LOCK(node);
380 	if (error == 0 && (vap->va_flags != VNOVAL)) {
381 		error = tmpfs_chflags(vp, vap->va_flags, cred);
382 		kflags |= NOTE_ATTRIB;
383 	}
384 
385 	if (error == 0 && (vap->va_size != VNOVAL)) {
386 		if (vap->va_size > node->tn_size)
387 			kflags |= NOTE_WRITE | NOTE_EXTEND;
388 		else
389 			kflags |= NOTE_WRITE;
390 		error = tmpfs_chsize(vp, vap->va_size, cred);
391 	}
392 
393 	if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
394 			   vap->va_gid != (gid_t)VNOVAL)) {
395 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
396 		kflags |= NOTE_ATTRIB;
397 	}
398 
399 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
400 		error = tmpfs_chmod(vp, vap->va_mode, cred);
401 		kflags |= NOTE_ATTRIB;
402 	}
403 
404 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
405 	    vap->va_atime.tv_nsec != VNOVAL) ||
406 	    (vap->va_mtime.tv_sec != VNOVAL &&
407 	    vap->va_mtime.tv_nsec != VNOVAL) )) {
408 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
409 				      vap->va_vaflags, cred);
410 		kflags |= NOTE_ATTRIB;
411 	}
412 
413 	/*
414 	 * Update the node times.  We give preference to the error codes
415 	 * generated by this function rather than the ones that may arise
416 	 * from tmpfs_update.
417 	 */
418 	tmpfs_update(vp);
419 	TMPFS_NODE_UNLOCK(node);
420 	tmpfs_knote(vp, kflags);
421 
422 	return (error);
423 }
424 
425 /* --------------------------------------------------------------------- */
426 
427 /*
428  * fsync is usually a NOP, but we must take action when unmounting or
429  * when recycling.
430  */
431 static int
432 tmpfs_fsync(struct vop_fsync_args *v)
433 {
434 	struct tmpfs_node *node;
435 	struct vnode *vp = v->a_vp;
436 
437 	node = VP_TO_TMPFS_NODE(vp);
438 
439 	tmpfs_update(vp);
440 	if (vp->v_type == VREG) {
441 		if (vp->v_flag & VRECLAIMED) {
442 			if (node->tn_links == 0)
443 				tmpfs_truncate(vp, 0);
444 			else
445 				vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
446 		}
447 	}
448 	return 0;
449 }
450 
451 /* --------------------------------------------------------------------- */
452 
453 static int
454 tmpfs_read (struct vop_read_args *ap)
455 {
456 	struct buf *bp;
457 	struct vnode *vp = ap->a_vp;
458 	struct uio *uio = ap->a_uio;
459 	struct tmpfs_node *node;
460 	off_t base_offset;
461 	size_t offset;
462 	size_t len;
463 	size_t resid;
464 	int error;
465 
466 	/*
467 	 * Check the basics
468 	 */
469 	if (uio->uio_offset < 0)
470 		return (EINVAL);
471 	if (vp->v_type != VREG)
472 		return (EINVAL);
473 
474 	/*
475 	 * Extract node, try to shortcut the operation through
476 	 * the VM page cache, allowing us to avoid buffer cache
477 	 * overheads.
478 	 */
479 	node = VP_TO_TMPFS_NODE(vp);
480         resid = uio->uio_resid;
481         error = vop_helper_read_shortcut(ap);
482         if (error)
483                 return error;
484         if (uio->uio_resid == 0) {
485 		if (resid)
486 			goto finished;
487 		return error;
488 	}
489 
490 	/*
491 	 * Fall-through to our normal read code.
492 	 */
493 	while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
494 		/*
495 		 * Use buffer cache I/O (via tmpfs_strategy)
496 		 */
497 		offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64;
498 		base_offset = (off_t)uio->uio_offset - offset;
499 		bp = getcacheblk(vp, base_offset, TMPFS_BLKSIZE, 0);
500 		if (bp == NULL) {
501 			error = bread(vp, base_offset, TMPFS_BLKSIZE, &bp);
502 			if (error) {
503 				brelse(bp);
504 				kprintf("tmpfs_read bread error %d\n", error);
505 				break;
506 			}
507 
508 			/*
509 			 * tmpfs pretty much fiddles directly with the VM
510 			 * system, don't let it exhaust it or we won't play
511 			 * nice with other processes.
512 			 *
513 			 * Only do this if the VOP is coming from a normal
514 			 * read/write.  The VM system handles the case for
515 			 * UIO_NOCOPY.
516 			 */
517 			if (uio->uio_segflg != UIO_NOCOPY)
518 				vm_wait_nominal();
519 		}
520 		bp->b_flags |= B_CLUSTEROK;
521 
522 		/*
523 		 * Figure out how many bytes we can actually copy this loop.
524 		 */
525 		len = TMPFS_BLKSIZE - offset;
526 		if (len > uio->uio_resid)
527 			len = uio->uio_resid;
528 		if (len > node->tn_size - uio->uio_offset)
529 			len = (size_t)(node->tn_size - uio->uio_offset);
530 
531 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
532 		bqrelse(bp);
533 		if (error) {
534 			kprintf("tmpfs_read uiomove error %d\n", error);
535 			break;
536 		}
537 	}
538 
539 finished:
540 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
541 		TMPFS_NODE_LOCK(node);
542 		node->tn_status |= TMPFS_NODE_ACCESSED;
543 		TMPFS_NODE_UNLOCK(node);
544 	}
545 	return (error);
546 }
547 
548 static int
549 tmpfs_write (struct vop_write_args *ap)
550 {
551 	struct buf *bp;
552 	struct vnode *vp = ap->a_vp;
553 	struct uio *uio = ap->a_uio;
554 	struct thread *td = uio->uio_td;
555 	struct tmpfs_node *node;
556 	boolean_t extended;
557 	off_t oldsize;
558 	int error;
559 	off_t base_offset;
560 	size_t offset;
561 	size_t len;
562 	struct rlimit limit;
563 	int trivial = 0;
564 	int kflags = 0;
565 	int seqcount;
566 
567 	error = 0;
568 	if (uio->uio_resid == 0) {
569 		return error;
570 	}
571 
572 	node = VP_TO_TMPFS_NODE(vp);
573 
574 	if (vp->v_type != VREG)
575 		return (EINVAL);
576 	seqcount = ap->a_ioflag >> 16;
577 
578 	oldsize = node->tn_size;
579 	if (ap->a_ioflag & IO_APPEND)
580 		uio->uio_offset = node->tn_size;
581 
582 	/*
583 	 * Check for illegal write offsets.
584 	 */
585 	if (uio->uio_offset + uio->uio_resid >
586 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
587 		return (EFBIG);
588 	}
589 
590 	/*
591 	 * NOTE: Ignore if UIO does not come from a user thread (e.g. VN).
592 	 */
593 	if (vp->v_type == VREG && td != NULL && td->td_lwp != NULL) {
594 		error = kern_getrlimit(RLIMIT_FSIZE, &limit);
595 		if (error != 0) {
596 			return error;
597 		}
598 		if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
599 			ksignal(td->td_proc, SIGXFSZ);
600 			return (EFBIG);
601 		}
602 	}
603 
604 
605 	/*
606 	 * Extend the file's size if necessary
607 	 */
608 	extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
609 
610 	while (uio->uio_resid > 0) {
611 		/*
612 		 * Don't completely blow out running buffer I/O
613 		 * when being hit from the pageout daemon.
614 		 */
615 		if (uio->uio_segflg == UIO_NOCOPY &&
616 		    (ap->a_ioflag & IO_RECURSE) == 0) {
617 			bwillwrite(TMPFS_BLKSIZE);
618 		}
619 
620 		/*
621 		 * Use buffer cache I/O (via tmpfs_strategy)
622 		 */
623 		offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64;
624 		base_offset = (off_t)uio->uio_offset - offset;
625 		len = TMPFS_BLKSIZE - offset;
626 		if (len > uio->uio_resid)
627 			len = uio->uio_resid;
628 
629 		if ((uio->uio_offset + len) > node->tn_size) {
630 			trivial = (uio->uio_offset <= node->tn_size);
631 			error = tmpfs_reg_resize(vp, uio->uio_offset + len,  trivial);
632 			if (error)
633 				break;
634 		}
635 
636 		/*
637 		 * Read to fill in any gaps.  Theoretically we could
638 		 * optimize this if the write covers the entire buffer
639 		 * and is not a UIO_NOCOPY write, however this can lead
640 		 * to a security violation exposing random kernel memory
641 		 * (whatever junk was in the backing VM pages before).
642 		 *
643 		 * So just use bread() to do the right thing.
644 		 */
645 		error = bread(vp, base_offset, TMPFS_BLKSIZE, &bp);
646 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
647 		if (error) {
648 			kprintf("tmpfs_write uiomove error %d\n", error);
649 			brelse(bp);
650 			break;
651 		}
652 
653 		if (uio->uio_offset > node->tn_size) {
654 			node->tn_size = uio->uio_offset;
655 			kflags |= NOTE_EXTEND;
656 		}
657 		kflags |= NOTE_WRITE;
658 
659 		/*
660 		 * Always try to flush the page in the UIO_NOCOPY case.  This
661 		 * can come from the pageout daemon or during vnode eviction.
662 		 * It is not necessarily going to be marked IO_ASYNC/IO_SYNC.
663 		 *
664 		 * For the normal case we buwrite(), dirtying the underlying
665 		 * VM pages instead of dirtying the buffer and releasing the
666 		 * buffer as a clean buffer.  This allows tmpfs to use
667 		 * essentially all available memory to cache file data.
668 		 * If we used bdwrite() the buffer cache would wind up
669 		 * flushing the data to swap too quickly.
670 		 *
671 		 * But because tmpfs can seriously load the VM system we
672 		 * fall-back to using bdwrite() when free memory starts
673 		 * to get low.  This shifts the load away from the VM system
674 		 * and makes tmpfs act more like a normal filesystem with
675 		 * regards to disk activity.
676 		 *
677 		 * tmpfs pretty much fiddles directly with the VM
678 		 * system, don't let it exhaust it or we won't play
679 		 * nice with other processes.  Only do this if the
680 		 * VOP is coming from a normal read/write.  The VM system
681 		 * handles the case for UIO_NOCOPY.
682 		 */
683 		bp->b_flags |= B_CLUSTEROK;
684 		if (uio->uio_segflg == UIO_NOCOPY) {
685 			/*
686 			 * Flush from the pageout daemon, deal with
687 			 * potentially very heavy tmpfs write activity
688 			 * causing long stalls in the pageout daemon
689 			 * before pages get to free/cache.
690 			 *
691 			 * (a) Under severe pressure setting B_DIRECT will
692 			 *     cause a buffer release to try to free the
693 			 *     underlying pages.
694 			 *
695 			 * (b) Under modest memory pressure the B_RELBUF
696 			 *     alone is sufficient to get the pages moved
697 			 *     to the cache.  We could also force this by
698 			 *     setting B_NOTMETA but that might have other
699 			 *     unintended side-effects (e.g. setting
700 			 *     PG_NOTMETA on the VM page).
701 			 *
702 			 * Hopefully this will unblock the VM system more
703 			 * quickly under extreme tmpfs write load.
704 			 */
705 			if (vm_page_count_min(vm_page_free_hysteresis))
706 				bp->b_flags |= B_DIRECT;
707 			bp->b_flags |= B_AGE | B_RELBUF;
708 			bp->b_act_count = 0;	/* buffer->deactivate pgs */
709 			cluster_awrite(bp);
710 		} else if (vm_page_count_target()) {
711 			/*
712 			 * Normal (userland) write but we are low on memory,
713 			 * run the buffer the buffer cache.
714 			 */
715 			bp->b_act_count = 0;	/* buffer->deactivate pgs */
716 			bdwrite(bp);
717 		} else {
718 			/*
719 			 * Otherwise run the buffer directly through to the
720 			 * backing VM store.
721 			 */
722 			buwrite(bp);
723 			/*vm_wait_nominal();*/
724 		}
725 
726 		if (bp->b_error) {
727 			kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
728 			break;
729 		}
730 	}
731 
732 	if (error) {
733 		if (extended) {
734 			(void)tmpfs_reg_resize(vp, oldsize, trivial);
735 			kflags &= ~NOTE_EXTEND;
736 		}
737 		goto done;
738 	}
739 
740 	/*
741 	 * Currently we don't set the mtime on files modified via mmap()
742 	 * because we can't tell the difference between those modifications
743 	 * and an attempt by the pageout daemon to flush tmpfs pages to
744 	 * swap.
745 	 *
746 	 * This is because in order to defer flushes as long as possible
747 	 * buwrite() works by marking the underlying VM pages dirty in
748 	 * order to be able to dispose of the buffer cache buffer without
749 	 * flushing it.
750 	 */
751 	TMPFS_NODE_LOCK(node);
752 	if (uio->uio_segflg != UIO_NOCOPY)
753 		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED;
754 	if (extended)
755 		node->tn_status |= TMPFS_NODE_CHANGED;
756 
757 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
758 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
759 			node->tn_mode &= ~(S_ISUID | S_ISGID);
760 	}
761 	TMPFS_NODE_UNLOCK(node);
762 done:
763 	tmpfs_knote(vp, kflags);
764 
765 	return(error);
766 }
767 
768 static int
769 tmpfs_advlock (struct vop_advlock_args *ap)
770 {
771 	struct tmpfs_node *node;
772 	struct vnode *vp = ap->a_vp;
773 	int error;
774 
775 	node = VP_TO_TMPFS_NODE(vp);
776 	error = (lf_advlock(ap, &node->tn_advlock, node->tn_size));
777 
778 	return (error);
779 }
780 
781 /*
782  * The strategy function is typically only called when memory pressure
783  * forces the system to attempt to pageout pages.  It can also be called
784  * by [n]vtruncbuf() when a truncation cuts a page in half.  Normal write
785  * operations
786  */
787 static int
788 tmpfs_strategy(struct vop_strategy_args *ap)
789 {
790 	struct bio *bio = ap->a_bio;
791 	struct bio *nbio;
792 	struct buf *bp = bio->bio_buf;
793 	struct vnode *vp = ap->a_vp;
794 	struct tmpfs_node *node;
795 	vm_object_t uobj;
796 	vm_page_t m;
797 	int i;
798 
799 	if (vp->v_type != VREG) {
800 		bp->b_resid = bp->b_bcount;
801 		bp->b_flags |= B_ERROR | B_INVAL;
802 		bp->b_error = EINVAL;
803 		biodone(bio);
804 		return(0);
805 	}
806 
807 	node = VP_TO_TMPFS_NODE(vp);
808 
809 	uobj = node->tn_reg.tn_aobj;
810 
811 	/*
812 	 * Don't bother flushing to swap if there is no swap, just
813 	 * ensure that the pages are marked as needing a commit (still).
814 	 */
815 	if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
816 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
817 			m = bp->b_xio.xio_pages[i];
818 			vm_page_need_commit(m);
819 		}
820 		bp->b_resid = 0;
821 		bp->b_error = 0;
822 		biodone(bio);
823 	} else {
824 		nbio = push_bio(bio);
825 		nbio->bio_done = tmpfs_strategy_done;
826 		nbio->bio_offset = bio->bio_offset;
827 		swap_pager_strategy(uobj, nbio);
828 	}
829 	return 0;
830 }
831 
832 /*
833  * If we were unable to commit the pages to swap make sure they are marked
834  * as needing a commit (again).  If we were, clear the flag to allow the
835  * pages to be freed.
836  */
837 static void
838 tmpfs_strategy_done(struct bio *bio)
839 {
840 	struct buf *bp;
841 	vm_page_t m;
842 	int i;
843 
844 	bp = bio->bio_buf;
845 
846 	if (bp->b_flags & B_ERROR) {
847 		bp->b_flags &= ~B_ERROR;
848 		bp->b_error = 0;
849 		bp->b_resid = 0;
850 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
851 			m = bp->b_xio.xio_pages[i];
852 			vm_page_need_commit(m);
853 		}
854 	} else {
855 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
856 			m = bp->b_xio.xio_pages[i];
857 			vm_page_clear_commit(m);
858 		}
859 	}
860 	bio = pop_bio(bio);
861 	biodone(bio);
862 }
863 
864 static int
865 tmpfs_bmap(struct vop_bmap_args *ap)
866 {
867 	if (ap->a_doffsetp != NULL)
868 		*ap->a_doffsetp = ap->a_loffset;
869 	if (ap->a_runp != NULL)
870 		*ap->a_runp = 0;
871 	if (ap->a_runb != NULL)
872 		*ap->a_runb = 0;
873 
874 	return 0;
875 }
876 
877 /* --------------------------------------------------------------------- */
878 
879 static int
880 tmpfs_nremove(struct vop_nremove_args *v)
881 {
882 	struct vnode *dvp = v->a_dvp;
883 	struct namecache *ncp = v->a_nch->ncp;
884 	struct vnode *vp;
885 	int error;
886 	struct tmpfs_dirent *de;
887 	struct tmpfs_mount *tmp;
888 	struct tmpfs_node *dnode;
889 	struct tmpfs_node *node;
890 	struct mount *mp;
891 
892 	mp = dvp->v_mount;
893 
894 	/*
895 	 * We have to acquire the vp from v->a_nch because we will likely
896 	 * unresolve the namecache entry, and a vrele/vput is needed to
897 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
898 	 *
899 	 * We have to use vget to clear any inactive state on the vnode,
900 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
901 	 * will not get called when we release it.
902 	 */
903 	error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
904 	KKASSERT(vp->v_mount == dvp->v_mount);
905 	KKASSERT(error == 0);
906 	vn_unlock(vp);
907 
908 	if (vp->v_type == VDIR) {
909 		error = EISDIR;
910 		goto out2;
911 	}
912 
913 	dnode = VP_TO_TMPFS_DIR(dvp);
914 	node = VP_TO_TMPFS_NODE(vp);
915 	tmp = VFS_TO_TMPFS(vp->v_mount);
916 
917 	TMPFS_NODE_LOCK(dnode);
918 	de = tmpfs_dir_lookup(dnode, node, ncp);
919 	if (de == NULL) {
920 		error = ENOENT;
921 		goto out;
922 	}
923 
924 	/* Files marked as immutable or append-only cannot be deleted. */
925 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
926 	    (dnode->tn_flags & APPEND)) {
927 		error = EPERM;
928 		goto out;
929 	}
930 
931 	/* Remove the entry from the directory; as it is a file, we do not
932 	 * have to change the number of hard links of the directory. */
933 	tmpfs_dir_detach(dnode, de);
934 
935 	/* Free the directory entry we just deleted.  Note that the node
936 	 * referred by it will not be removed until the vnode is really
937 	 * reclaimed. */
938 	tmpfs_free_dirent(tmp, de);
939 
940 	if (node->tn_links > 0) {
941 	        TMPFS_NODE_LOCK(node);
942 		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
943 	                TMPFS_NODE_MODIFIED;
944 	        TMPFS_NODE_UNLOCK(node);
945 	}
946 
947 	cache_unlink(v->a_nch);
948 	tmpfs_knote(vp, NOTE_DELETE);
949 	error = 0;
950 
951 out:
952 	TMPFS_NODE_UNLOCK(dnode);
953 	if (error == 0)
954 		tmpfs_knote(dvp, NOTE_WRITE);
955 out2:
956 	vrele(vp);
957 
958 	return error;
959 }
960 
961 /* --------------------------------------------------------------------- */
962 
963 static int
964 tmpfs_nlink(struct vop_nlink_args *v)
965 {
966 	struct vnode *dvp = v->a_dvp;
967 	struct vnode *vp = v->a_vp;
968 	struct namecache *ncp = v->a_nch->ncp;
969 	struct tmpfs_dirent *de;
970 	struct tmpfs_node *node;
971 	struct tmpfs_node *dnode;
972 	struct mount *mp;
973 	int error;
974 
975 	if (dvp->v_mount != vp->v_mount)
976 		return(EXDEV);
977 	mp = dvp->v_mount;
978 
979 	KKASSERT(dvp != vp); /* XXX When can this be false? */
980 
981 	node = VP_TO_TMPFS_NODE(vp);
982 	dnode = VP_TO_TMPFS_NODE(dvp);
983 	TMPFS_NODE_LOCK(dnode);
984 
985 	/* XXX: Why aren't the following two tests done by the caller? */
986 
987 	/* Hard links of directories are forbidden. */
988 	if (vp->v_type == VDIR) {
989 		error = EPERM;
990 		goto out;
991 	}
992 
993 	/* Cannot create cross-device links. */
994 	if (dvp->v_mount != vp->v_mount) {
995 		error = EXDEV;
996 		goto out;
997 	}
998 
999 	/* Ensure that we do not overflow the maximum number of links imposed
1000 	 * by the system. */
1001 	KKASSERT(node->tn_links <= LINK_MAX);
1002 	if (node->tn_links >= LINK_MAX) {
1003 		error = EMLINK;
1004 		goto out;
1005 	}
1006 
1007 	/* We cannot create links of files marked immutable or append-only. */
1008 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
1009 		error = EPERM;
1010 		goto out;
1011 	}
1012 
1013 	/* Allocate a new directory entry to represent the node. */
1014 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
1015 				   ncp->nc_name, ncp->nc_nlen, &de);
1016 	if (error != 0)
1017 		goto out;
1018 
1019 	/* Insert the new directory entry into the appropriate directory. */
1020 	tmpfs_dir_attach(dnode, de);
1021 
1022 	/* vp link count has changed, so update node times. */
1023 
1024 	TMPFS_NODE_LOCK(node);
1025 	node->tn_status |= TMPFS_NODE_CHANGED;
1026 	TMPFS_NODE_UNLOCK(node);
1027 	tmpfs_update(vp);
1028 
1029 	tmpfs_knote(vp, NOTE_LINK);
1030 	cache_setunresolved(v->a_nch);
1031 	cache_setvp(v->a_nch, vp);
1032 	error = 0;
1033 
1034 out:
1035 	TMPFS_NODE_UNLOCK(dnode);
1036 	if (error == 0)
1037 		tmpfs_knote(dvp, NOTE_WRITE);
1038 	return error;
1039 }
1040 
1041 /* --------------------------------------------------------------------- */
1042 
1043 static int
1044 tmpfs_nrename(struct vop_nrename_args *v)
1045 {
1046 	struct vnode *fdvp = v->a_fdvp;
1047 	struct namecache *fncp = v->a_fnch->ncp;
1048 	struct vnode *fvp = fncp->nc_vp;
1049 	struct vnode *tdvp = v->a_tdvp;
1050 	struct namecache *tncp = v->a_tnch->ncp;
1051 	struct vnode *tvp;
1052 	struct tmpfs_dirent *de, *tde;
1053 	struct tmpfs_mount *tmp;
1054 	struct tmpfs_node *fdnode;
1055 	struct tmpfs_node *fnode;
1056 	struct tmpfs_node *tnode;
1057 	struct tmpfs_node *tdnode;
1058 	struct mount *mp;
1059 	char *newname;
1060 	char *oldname;
1061 	int error;
1062 
1063 	mp = fdvp->v_mount;
1064 	KKASSERT(fdvp->v_mount == fvp->v_mount);
1065 
1066 	/*
1067 	 * Because tvp can get overwritten we have to vget it instead of
1068 	 * just vref or use it, otherwise it's VINACTIVE flag may not get
1069 	 * cleared and the node won't get destroyed.
1070 	 */
1071 	error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp);
1072 	if (error == 0) {
1073 		tnode = VP_TO_TMPFS_NODE(tvp);
1074 		vn_unlock(tvp);
1075 	} else {
1076 		tnode = NULL;
1077 	}
1078 
1079 	/* Disallow cross-device renames.
1080 	 * XXX Why isn't this done by the caller? */
1081 	if (fvp->v_mount != tdvp->v_mount ||
1082 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
1083 		error = EXDEV;
1084 		goto out;
1085 	}
1086 
1087 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
1088 	tdnode = VP_TO_TMPFS_DIR(tdvp);
1089 
1090 	/* If source and target are the same file, there is nothing to do. */
1091 	if (fvp == tvp) {
1092 		error = 0;
1093 		goto out;
1094 	}
1095 
1096 	fdnode = VP_TO_TMPFS_DIR(fdvp);
1097 	fnode = VP_TO_TMPFS_NODE(fvp);
1098 	TMPFS_NODE_LOCK(fdnode);
1099 	de = tmpfs_dir_lookup(fdnode, fnode, fncp);
1100 	TMPFS_NODE_UNLOCK(fdnode);	/* XXX depend on namecache lock */
1101 
1102 	/* Avoid manipulating '.' and '..' entries. */
1103 	if (de == NULL) {
1104 		error = ENOENT;
1105 		goto out_locked;
1106 	}
1107 	KKASSERT(de->td_node == fnode);
1108 
1109 	/*
1110 	 * If replacing an entry in the target directory and that entry
1111 	 * is a directory, it must be empty.
1112 	 *
1113 	 * Kern_rename gurantees the destination to be a directory
1114 	 * if the source is one (it does?).
1115 	 */
1116 	if (tvp != NULL) {
1117 		KKASSERT(tnode != NULL);
1118 
1119 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1120 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1121 			error = EPERM;
1122 			goto out_locked;
1123 		}
1124 
1125 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1126 			if (tnode->tn_size > 0) {
1127 				error = ENOTEMPTY;
1128 				goto out_locked;
1129 			}
1130 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1131 			error = ENOTDIR;
1132 			goto out_locked;
1133 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1134 			error = EISDIR;
1135 			goto out_locked;
1136 		} else {
1137 			KKASSERT(fnode->tn_type != VDIR &&
1138 				tnode->tn_type != VDIR);
1139 		}
1140 	}
1141 
1142 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1143 	    (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1144 		error = EPERM;
1145 		goto out_locked;
1146 	}
1147 
1148 	/*
1149 	 * Ensure that we have enough memory to hold the new name, if it
1150 	 * has to be changed.
1151 	 */
1152 	if (fncp->nc_nlen != tncp->nc_nlen ||
1153 	    bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1154 		newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1155 				  M_WAITOK | M_NULLOK);
1156 		if (newname == NULL) {
1157 			error = ENOSPC;
1158 			goto out_locked;
1159 		}
1160 		bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1161 		newname[tncp->nc_nlen] = '\0';
1162 	} else {
1163 		newname = NULL;
1164 	}
1165 
1166 	/*
1167 	 * Unlink entry from source directory.  Note that the kernel has
1168 	 * already checked for illegal recursion cases (renaming a directory
1169 	 * into a subdirectory of itself).
1170 	 */
1171 	if (fdnode != tdnode) {
1172 		tmpfs_dir_detach(fdnode, de);
1173 	} else {
1174 		RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de);
1175 	}
1176 
1177 	/*
1178 	 * Handle any name change.  Swap with newname, we will
1179 	 * deallocate it at the end.
1180 	 */
1181 	if (newname != NULL) {
1182 #if 0
1183 		TMPFS_NODE_LOCK(fnode);
1184 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1185 		TMPFS_NODE_UNLOCK(fnode);
1186 #endif
1187 		oldname = de->td_name;
1188 		de->td_name = newname;
1189 		de->td_namelen = (uint16_t)tncp->nc_nlen;
1190 		newname = oldname;
1191 	}
1192 
1193 	/*
1194 	 * If we are overwriting an entry, we have to remove the old one
1195 	 * from the target directory.
1196 	 */
1197 	if (tvp != NULL) {
1198 		/* Remove the old entry from the target directory. */
1199 		TMPFS_NODE_LOCK(tdnode);
1200 		tde = tmpfs_dir_lookup(tdnode, tnode, tncp);
1201 		tmpfs_dir_detach(tdnode, tde);
1202 		TMPFS_NODE_UNLOCK(tdnode);
1203 		tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1204 
1205 		/*
1206 		 * Free the directory entry we just deleted.  Note that the
1207 		 * node referred by it will not be removed until the vnode is
1208 		 * really reclaimed.
1209 		 */
1210 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1211 		/*cache_inval_vp(tvp, CINV_DESTROY);*/
1212 	}
1213 
1214 	/*
1215 	 * Link entry to target directory.  If the entry
1216 	 * represents a directory move the parent linkage
1217 	 * as well.
1218 	 */
1219 	if (fdnode != tdnode) {
1220 		if (de->td_node->tn_type == VDIR) {
1221 			TMPFS_VALIDATE_DIR(fnode);
1222 		}
1223 		tmpfs_dir_attach(tdnode, de);
1224 	} else {
1225 		TMPFS_NODE_LOCK(tdnode);
1226 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1227 		RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de);
1228 		TMPFS_NODE_UNLOCK(tdnode);
1229 	}
1230 
1231 	/*
1232 	 * Finish up
1233 	 */
1234 	if (newname) {
1235 		kfree(newname, tmp->tm_name_zone);
1236 		newname = NULL;
1237 	}
1238 	cache_rename(v->a_fnch, v->a_tnch);
1239 	tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1240 	tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1241 	if (fnode->tn_vnode)
1242 		tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1243 	error = 0;
1244 
1245 out_locked:
1246 	;
1247 out:
1248 	if (tvp)
1249 		vrele(tvp);
1250 	return error;
1251 }
1252 
1253 /* --------------------------------------------------------------------- */
1254 
1255 static int
1256 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1257 {
1258 	struct vnode *dvp = v->a_dvp;
1259 	struct vnode **vpp = v->a_vpp;
1260 	struct namecache *ncp = v->a_nch->ncp;
1261 	struct vattr *vap = v->a_vap;
1262 	struct ucred *cred = v->a_cred;
1263 	struct mount *mp;
1264 	int error;
1265 
1266 	mp = dvp->v_mount;
1267 
1268 	KKASSERT(vap->va_type == VDIR);
1269 
1270 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1271 	if (error == 0) {
1272 		cache_setunresolved(v->a_nch);
1273 		cache_setvp(v->a_nch, *vpp);
1274 		tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1275 	}
1276 	return error;
1277 }
1278 
1279 /* --------------------------------------------------------------------- */
1280 
1281 static int
1282 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1283 {
1284 	struct vnode *dvp = v->a_dvp;
1285 	struct namecache *ncp = v->a_nch->ncp;
1286 	struct vnode *vp;
1287 	struct tmpfs_dirent *de;
1288 	struct tmpfs_mount *tmp;
1289 	struct tmpfs_node *dnode;
1290 	struct tmpfs_node *node;
1291 	struct mount *mp;
1292 	int error;
1293 
1294 	mp = dvp->v_mount;
1295 
1296 	/*
1297 	 * We have to acquire the vp from v->a_nch because we will likely
1298 	 * unresolve the namecache entry, and a vrele/vput is needed to
1299 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1300 	 *
1301 	 * We have to use vget to clear any inactive state on the vnode,
1302 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1303 	 * will not get called when we release it.
1304 	 */
1305 	error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
1306 	KKASSERT(error == 0);
1307 	vn_unlock(vp);
1308 
1309 	/*
1310 	 * Prevalidate so we don't hit an assertion later
1311 	 */
1312 	if (vp->v_type != VDIR) {
1313 		error = ENOTDIR;
1314 		goto out;
1315 	}
1316 
1317 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1318 	dnode = VP_TO_TMPFS_DIR(dvp);
1319 	node = VP_TO_TMPFS_DIR(vp);
1320 
1321 	/*
1322 	 * Directories with more than two entries ('.' and '..') cannot
1323 	 * be removed.
1324 	 */
1325 	if (node->tn_size > 0) {
1326 		error = ENOTEMPTY;
1327 		goto out;
1328 	}
1329 
1330 	if ((dnode->tn_flags & APPEND)
1331 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1332 		error = EPERM;
1333 		goto out;
1334 	}
1335 
1336 	/*
1337 	 * This invariant holds only if we are not trying to
1338 	 * remove "..".  We checked for that above so this is safe now.
1339 	 */
1340 	KKASSERT(node->tn_dir.tn_parent == dnode);
1341 
1342 	/*
1343 	 * Get the directory entry associated with node (vp).  This
1344 	 * was filled by tmpfs_lookup while looking up the entry.
1345 	 */
1346 	TMPFS_NODE_LOCK(dnode);
1347 	de = tmpfs_dir_lookup(dnode, node, ncp);
1348 	KKASSERT(TMPFS_DIRENT_MATCHES(de, ncp->nc_name, ncp->nc_nlen));
1349 
1350 	/* Check flags to see if we are allowed to remove the directory. */
1351 	if ((dnode->tn_flags & APPEND) ||
1352 	    node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1353 		error = EPERM;
1354 		TMPFS_NODE_UNLOCK(dnode);
1355 		goto out;
1356 	}
1357 
1358 	/* Detach the directory entry from the directory (dnode). */
1359 	tmpfs_dir_detach(dnode, de);
1360 	TMPFS_NODE_UNLOCK(dnode);
1361 
1362 	/* No vnode should be allocated for this entry from this point */
1363 	TMPFS_NODE_LOCK(node);
1364 	TMPFS_ASSERT_ELOCKED(node);
1365 	TMPFS_NODE_LOCK(dnode);
1366 	TMPFS_ASSERT_ELOCKED(dnode);
1367 
1368 	/*
1369 	 * Must set parent linkage to NULL (tested by ncreate to disallow
1370 	 * the creation of new files/dirs in a deleted directory)
1371 	 */
1372 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1373 			   TMPFS_NODE_MODIFIED;
1374 
1375 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1376 			    TMPFS_NODE_MODIFIED;
1377 
1378 	TMPFS_NODE_UNLOCK(dnode);
1379 	TMPFS_NODE_UNLOCK(node);
1380 
1381 	/* Free the directory entry we just deleted.  Note that the node
1382 	 * referred by it will not be removed until the vnode is really
1383 	 * reclaimed. */
1384 	tmpfs_free_dirent(tmp, de);
1385 
1386 	/* Release the deleted vnode (will destroy the node, notify
1387 	 * interested parties and clean it from the cache). */
1388 
1389 	TMPFS_NODE_LOCK(dnode);
1390 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1391 	TMPFS_NODE_UNLOCK(dnode);
1392 	tmpfs_update(dvp);
1393 
1394 	cache_unlink(v->a_nch);
1395 	tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1396 	error = 0;
1397 
1398 out:
1399 	vrele(vp);
1400 
1401 	return error;
1402 }
1403 
1404 /* --------------------------------------------------------------------- */
1405 
1406 static int
1407 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1408 {
1409 	struct vnode *dvp = v->a_dvp;
1410 	struct vnode **vpp = v->a_vpp;
1411 	struct namecache *ncp = v->a_nch->ncp;
1412 	struct vattr *vap = v->a_vap;
1413 	struct ucred *cred = v->a_cred;
1414 	char *target = v->a_target;
1415 	int error;
1416 
1417 	vap->va_type = VLNK;
1418 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1419 	if (error == 0) {
1420 		tmpfs_knote(*vpp, NOTE_WRITE);
1421 		cache_setunresolved(v->a_nch);
1422 		cache_setvp(v->a_nch, *vpp);
1423 	}
1424 	return error;
1425 }
1426 
1427 /* --------------------------------------------------------------------- */
1428 
1429 static int
1430 tmpfs_readdir(struct vop_readdir_args *v)
1431 {
1432 	struct vnode *vp = v->a_vp;
1433 	struct uio *uio = v->a_uio;
1434 	int *eofflag = v->a_eofflag;
1435 	off_t **cookies = v->a_cookies;
1436 	int *ncookies = v->a_ncookies;
1437 	struct tmpfs_mount *tmp;
1438 	int error;
1439 	off_t startoff;
1440 	off_t cnt = 0;
1441 	struct tmpfs_node *node;
1442 
1443 	/* This operation only makes sense on directory nodes. */
1444 	if (vp->v_type != VDIR) {
1445 		return ENOTDIR;
1446 	}
1447 
1448 	tmp = VFS_TO_TMPFS(vp->v_mount);
1449 	node = VP_TO_TMPFS_DIR(vp);
1450 	startoff = uio->uio_offset;
1451 
1452 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1453 		error = tmpfs_dir_getdotdent(node, uio);
1454 		if (error != 0) {
1455 			TMPFS_NODE_LOCK_SH(node);
1456 			goto outok;
1457 		}
1458 		cnt++;
1459 	}
1460 
1461 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1462 		/* may lock parent, cannot hold node lock */
1463 		error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1464 		if (error != 0) {
1465 			TMPFS_NODE_LOCK_SH(node);
1466 			goto outok;
1467 		}
1468 		cnt++;
1469 	}
1470 
1471 	TMPFS_NODE_LOCK_SH(node);
1472 	error = tmpfs_dir_getdents(node, uio, &cnt);
1473 
1474 outok:
1475 	KKASSERT(error >= -1);
1476 
1477 	if (error == -1)
1478 		error = 0;
1479 
1480 	if (eofflag != NULL)
1481 		*eofflag =
1482 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1483 
1484 	/* Update NFS-related variables. */
1485 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1486 		off_t i;
1487 		off_t off = startoff;
1488 		struct tmpfs_dirent *de = NULL;
1489 
1490 		*ncookies = cnt;
1491 		*cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1492 
1493 		for (i = 0; i < cnt; i++) {
1494 			KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1495 			if (off == TMPFS_DIRCOOKIE_DOT) {
1496 				off = TMPFS_DIRCOOKIE_DOTDOT;
1497 			} else {
1498 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1499 					de = RB_MIN(tmpfs_dirtree, &node->tn_dir.tn_dirtree);
1500 				} else if (de != NULL) {
1501 					de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
1502 				} else {
1503 					de = tmpfs_dir_lookupbycookie(node,
1504 					    off);
1505 					KKASSERT(de != NULL);
1506 					de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
1507 				}
1508 				if (de == NULL)
1509 					off = TMPFS_DIRCOOKIE_EOF;
1510 				else
1511 					off = tmpfs_dircookie(de);
1512 			}
1513 			(*cookies)[i] = off;
1514 		}
1515 		KKASSERT(uio->uio_offset == off);
1516 	}
1517 	TMPFS_NODE_UNLOCK(node);
1518 
1519 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
1520 		TMPFS_NODE_LOCK(node);
1521 		node->tn_status |= TMPFS_NODE_ACCESSED;
1522 		TMPFS_NODE_UNLOCK(node);
1523 	}
1524 	return error;
1525 }
1526 
1527 /* --------------------------------------------------------------------- */
1528 
1529 static int
1530 tmpfs_readlink(struct vop_readlink_args *v)
1531 {
1532 	struct vnode *vp = v->a_vp;
1533 	struct uio *uio = v->a_uio;
1534 	int error;
1535 	struct tmpfs_node *node;
1536 
1537 	KKASSERT(uio->uio_offset == 0);
1538 	KKASSERT(vp->v_type == VLNK);
1539 
1540 	node = VP_TO_TMPFS_NODE(vp);
1541 	TMPFS_NODE_LOCK_SH(node);
1542 	error = uiomove(node->tn_link,
1543 			MIN(node->tn_size, uio->uio_resid), uio);
1544 	TMPFS_NODE_UNLOCK(node);
1545 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
1546 		TMPFS_NODE_LOCK(node);
1547 		node->tn_status |= TMPFS_NODE_ACCESSED;
1548 		TMPFS_NODE_UNLOCK(node);
1549 	}
1550 	return error;
1551 }
1552 
1553 /* --------------------------------------------------------------------- */
1554 
1555 static int
1556 tmpfs_inactive(struct vop_inactive_args *v)
1557 {
1558 	struct vnode *vp = v->a_vp;
1559 	struct tmpfs_node *node;
1560 	struct mount *mp;
1561 
1562 	mp = vp->v_mount;
1563 	lwkt_gettoken(&mp->mnt_token);
1564 	node = VP_TO_TMPFS_NODE(vp);
1565 
1566 	/*
1567 	 * Degenerate case
1568 	 */
1569 	if (node == NULL) {
1570 		vrecycle(vp);
1571 		lwkt_reltoken(&mp->mnt_token);
1572 		return(0);
1573 	}
1574 
1575 	/*
1576 	 * Get rid of unreferenced deleted vnodes sooner rather than
1577 	 * later so the data memory can be recovered immediately.
1578 	 *
1579 	 * We must truncate the vnode to prevent the normal reclamation
1580 	 * path from flushing the data for the removed file to disk.
1581 	 */
1582 	TMPFS_NODE_LOCK(node);
1583 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1584 	    node->tn_links == 0)
1585 	{
1586 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1587 		TMPFS_NODE_UNLOCK(node);
1588 		if (node->tn_type == VREG)
1589 			tmpfs_truncate(vp, 0);
1590 		vrecycle(vp);
1591 	} else {
1592 		TMPFS_NODE_UNLOCK(node);
1593 	}
1594 	lwkt_reltoken(&mp->mnt_token);
1595 
1596 	return 0;
1597 }
1598 
1599 /* --------------------------------------------------------------------- */
1600 
1601 int
1602 tmpfs_reclaim(struct vop_reclaim_args *v)
1603 {
1604 	struct vnode *vp = v->a_vp;
1605 	struct tmpfs_mount *tmp;
1606 	struct tmpfs_node *node;
1607 	struct mount *mp;
1608 
1609 	mp = vp->v_mount;
1610 	lwkt_gettoken(&mp->mnt_token);
1611 
1612 	node = VP_TO_TMPFS_NODE(vp);
1613 	tmp = VFS_TO_TMPFS(vp->v_mount);
1614 	KKASSERT(mp == tmp->tm_mount);
1615 
1616 	tmpfs_free_vp(vp);
1617 
1618 	/*
1619 	 * If the node referenced by this vnode was deleted by the
1620 	 * user, we must free its associated data structures now that
1621 	 * the vnode is being reclaimed.
1622 	 *
1623 	 * Directories have an extra link ref.
1624 	 */
1625 	TMPFS_NODE_LOCK(node);
1626 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1627 	    node->tn_links == 0) {
1628 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1629 		tmpfs_free_node(tmp, node);
1630 		/* eats the lock */
1631 	} else {
1632 		TMPFS_NODE_UNLOCK(node);
1633 	}
1634 	lwkt_reltoken(&mp->mnt_token);
1635 
1636 	KKASSERT(vp->v_data == NULL);
1637 	return 0;
1638 }
1639 
1640 /* --------------------------------------------------------------------- */
1641 
1642 static int
1643 tmpfs_mountctl(struct vop_mountctl_args *ap)
1644 {
1645 	struct tmpfs_mount *tmp;
1646 	struct mount *mp;
1647 	int rc;
1648 
1649 	mp = ap->a_head.a_ops->head.vv_mount;
1650 	lwkt_gettoken(&mp->mnt_token);
1651 
1652 	switch (ap->a_op) {
1653 	case (MOUNTCTL_SET_EXPORT):
1654 		tmp = (struct tmpfs_mount *) mp->mnt_data;
1655 
1656 		if (ap->a_ctllen != sizeof(struct export_args))
1657 			rc = (EINVAL);
1658 		else
1659 			rc = vfs_export(mp, &tmp->tm_export,
1660 					(const struct export_args *) ap->a_ctl);
1661 		break;
1662 	default:
1663 		rc = vop_stdmountctl(ap);
1664 		break;
1665 	}
1666 
1667 	lwkt_reltoken(&mp->mnt_token);
1668 	return (rc);
1669 }
1670 
1671 /* --------------------------------------------------------------------- */
1672 
1673 static int
1674 tmpfs_print(struct vop_print_args *v)
1675 {
1676 	struct vnode *vp = v->a_vp;
1677 
1678 	struct tmpfs_node *node;
1679 
1680 	node = VP_TO_TMPFS_NODE(vp);
1681 
1682 	kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1683 	    node, node->tn_flags, node->tn_links);
1684 	kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1685 	    node->tn_mode, node->tn_uid, node->tn_gid,
1686 	    (uintmax_t)node->tn_size, node->tn_status);
1687 
1688 	if (vp->v_type == VFIFO)
1689 		fifo_printinfo(vp);
1690 
1691 	kprintf("\n");
1692 
1693 	return 0;
1694 }
1695 
1696 /* --------------------------------------------------------------------- */
1697 
1698 static int
1699 tmpfs_pathconf(struct vop_pathconf_args *v)
1700 {
1701 	int name = v->a_name;
1702 	register_t *retval = v->a_retval;
1703 
1704 	int error;
1705 
1706 	error = 0;
1707 
1708 	switch (name) {
1709 	case _PC_LINK_MAX:
1710 		*retval = LINK_MAX;
1711 		break;
1712 
1713 	case _PC_NAME_MAX:
1714 		*retval = NAME_MAX;
1715 		break;
1716 
1717 	case _PC_PATH_MAX:
1718 		*retval = PATH_MAX;
1719 		break;
1720 
1721 	case _PC_PIPE_BUF:
1722 		*retval = PIPE_BUF;
1723 		break;
1724 
1725 	case _PC_CHOWN_RESTRICTED:
1726 		*retval = 1;
1727 		break;
1728 
1729 	case _PC_NO_TRUNC:
1730 		*retval = 1;
1731 		break;
1732 
1733 	case _PC_SYNC_IO:
1734 		*retval = 1;
1735 		break;
1736 
1737 	case _PC_FILESIZEBITS:
1738 		*retval = 0; /* XXX Don't know which value should I return. */
1739 		break;
1740 
1741 	default:
1742 		error = EINVAL;
1743 	}
1744 
1745 	return error;
1746 }
1747 
1748 /************************************************************************
1749  *                          KQFILTER OPS                                *
1750  ************************************************************************/
1751 
1752 static void filt_tmpfsdetach(struct knote *kn);
1753 static int filt_tmpfsread(struct knote *kn, long hint);
1754 static int filt_tmpfswrite(struct knote *kn, long hint);
1755 static int filt_tmpfsvnode(struct knote *kn, long hint);
1756 
1757 static struct filterops tmpfsread_filtops =
1758 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1759 	  NULL, filt_tmpfsdetach, filt_tmpfsread };
1760 static struct filterops tmpfswrite_filtops =
1761 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1762 	  NULL, filt_tmpfsdetach, filt_tmpfswrite };
1763 static struct filterops tmpfsvnode_filtops =
1764 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1765 	  NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1766 
1767 static int
1768 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1769 {
1770 	struct vnode *vp = ap->a_vp;
1771 	struct knote *kn = ap->a_kn;
1772 
1773 	switch (kn->kn_filter) {
1774 	case EVFILT_READ:
1775 		kn->kn_fop = &tmpfsread_filtops;
1776 		break;
1777 	case EVFILT_WRITE:
1778 		kn->kn_fop = &tmpfswrite_filtops;
1779 		break;
1780 	case EVFILT_VNODE:
1781 		kn->kn_fop = &tmpfsvnode_filtops;
1782 		break;
1783 	default:
1784 		return (EOPNOTSUPP);
1785 	}
1786 
1787 	kn->kn_hook = (caddr_t)vp;
1788 
1789 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1790 
1791 	return(0);
1792 }
1793 
1794 static void
1795 filt_tmpfsdetach(struct knote *kn)
1796 {
1797 	struct vnode *vp = (void *)kn->kn_hook;
1798 
1799 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1800 }
1801 
1802 static int
1803 filt_tmpfsread(struct knote *kn, long hint)
1804 {
1805 	struct vnode *vp = (void *)kn->kn_hook;
1806 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1807 	off_t off;
1808 
1809 	if (hint == NOTE_REVOKE) {
1810 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1811 		return(1);
1812 	}
1813 
1814 	/*
1815 	 * Interlock against MP races when performing this function.
1816 	 */
1817 	TMPFS_NODE_LOCK_SH(node);
1818 	off = node->tn_size - kn->kn_fp->f_offset;
1819 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1820 	if (kn->kn_sfflags & NOTE_OLDAPI) {
1821 		TMPFS_NODE_UNLOCK(node);
1822 		return(1);
1823 	}
1824 	if (kn->kn_data == 0) {
1825 		kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1826 	}
1827 	TMPFS_NODE_UNLOCK(node);
1828 	return (kn->kn_data != 0);
1829 }
1830 
1831 static int
1832 filt_tmpfswrite(struct knote *kn, long hint)
1833 {
1834 	if (hint == NOTE_REVOKE)
1835 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1836 	kn->kn_data = 0;
1837 	return (1);
1838 }
1839 
1840 static int
1841 filt_tmpfsvnode(struct knote *kn, long hint)
1842 {
1843 	if (kn->kn_sfflags & hint)
1844 		kn->kn_fflags |= hint;
1845 	if (hint == NOTE_REVOKE) {
1846 		kn->kn_flags |= (EV_EOF | EV_NODATA);
1847 		return (1);
1848 	}
1849 	return (kn->kn_fflags != 0);
1850 }
1851 
1852 
1853 /* --------------------------------------------------------------------- */
1854 
1855 /*
1856  * vnode operations vector used for files stored in a tmpfs file system.
1857  */
1858 struct vop_ops tmpfs_vnode_vops = {
1859 	.vop_default =			vop_defaultop,
1860 	.vop_getpages = 		vop_stdgetpages,
1861 	.vop_putpages = 		vop_stdputpages,
1862 	.vop_ncreate =			tmpfs_ncreate,
1863 	.vop_nresolve =			tmpfs_nresolve,
1864 	.vop_nlookupdotdot =		tmpfs_nlookupdotdot,
1865 	.vop_nmknod =			tmpfs_nmknod,
1866 	.vop_open =			tmpfs_open,
1867 	.vop_close =			tmpfs_close,
1868 	.vop_access =			tmpfs_access,
1869 	.vop_getattr =			tmpfs_getattr,
1870 	.vop_setattr =			tmpfs_setattr,
1871 	.vop_read =			tmpfs_read,
1872 	.vop_write =			tmpfs_write,
1873 	.vop_fsync =			tmpfs_fsync,
1874 	.vop_mountctl =			tmpfs_mountctl,
1875 	.vop_nremove =			tmpfs_nremove,
1876 	.vop_nlink =			tmpfs_nlink,
1877 	.vop_nrename =			tmpfs_nrename,
1878 	.vop_nmkdir =			tmpfs_nmkdir,
1879 	.vop_nrmdir =			tmpfs_nrmdir,
1880 	.vop_nsymlink =			tmpfs_nsymlink,
1881 	.vop_readdir =			tmpfs_readdir,
1882 	.vop_readlink =			tmpfs_readlink,
1883 	.vop_inactive =			tmpfs_inactive,
1884 	.vop_reclaim =			tmpfs_reclaim,
1885 	.vop_print =			tmpfs_print,
1886 	.vop_pathconf =			tmpfs_pathconf,
1887 	.vop_bmap =			tmpfs_bmap,
1888 	.vop_strategy =			tmpfs_strategy,
1889 	.vop_advlock =			tmpfs_advlock,
1890 	.vop_kqfilter =			tmpfs_kqfilter
1891 };
1892