xref: /dragonfly/sys/vfs/tmpfs/tmpfs_vnops.c (revision 0de61e28)
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.  Run through the buffer
76  *		cache if free memory is under the minimum.
77  *
78  *	1	Try to keep in memory, but run through the buffer cache if
79  *		the system is under memory pressure (though this might just
80  *		require inactive cleaning).
81  *
82  *	2	Be a bit more aggressive when running writes through the
83  *		buffer cache when the system is under memory pressure.
84  *
85  *	3	Always run tmpfs writes through the buffer cache, thus forcing
86  *		them out to swap.
87  */
88 __read_mostly static int tmpfs_cluster_rd_enable = 1;
89 __read_mostly static int tmpfs_cluster_wr_enable = 1;
90 __read_mostly int tmpfs_bufcache_mode = 0;
91 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "TMPFS filesystem");
92 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, cluster_rd_enable, CTLFLAG_RW,
93 		&tmpfs_cluster_rd_enable, 0, "");
94 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, cluster_wr_enable, CTLFLAG_RW,
95 		&tmpfs_cluster_wr_enable, 0, "");
96 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, bufcache_mode, CTLFLAG_RW,
97 		&tmpfs_bufcache_mode, 0, "");
98 
99 #define TMPFS_MOVF_FROMBACKING	0x0001
100 #define TMPFS_MOVF_DEACTIVATE	0x0002
101 
102 
103 static __inline
104 void
105 tmpfs_knote(struct vnode *vp, int flags)
106 {
107 	if (flags)
108 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
109 }
110 
111 
112 /* --------------------------------------------------------------------- */
113 
114 static int
115 tmpfs_nresolve(struct vop_nresolve_args *ap)
116 {
117 	struct vnode *dvp = ap->a_dvp;
118 	struct vnode *vp = NULL;
119 	struct namecache *ncp = ap->a_nch->ncp;
120 	struct tmpfs_node *tnode;
121 	struct tmpfs_dirent *de;
122 	struct tmpfs_node *dnode;
123 	int error;
124 
125 	dnode = VP_TO_TMPFS_DIR(dvp);
126 
127 	TMPFS_NODE_LOCK_SH(dnode);
128 loop:
129 	de = tmpfs_dir_lookup(dnode, NULL, ncp);
130 	if (de == NULL) {
131 		error = ENOENT;
132 	} else {
133 		/*
134 		 * Allocate a vnode for the node we found.  Use
135 		 * tmpfs_alloc_vp()'s deadlock handling mode.
136 		 */
137 		tnode = de->td_node;
138 		error = tmpfs_alloc_vp(dvp->v_mount, dnode, tnode,
139 				       LK_EXCLUSIVE | LK_RETRY, &vp);
140 		if (error == EAGAIN)
141 			goto loop;
142 		if (error)
143 			goto out;
144 		KKASSERT(vp);
145 	}
146 
147 out:
148 	TMPFS_NODE_UNLOCK(dnode);
149 
150 	if ((dnode->tn_status & TMPFS_NODE_ACCESSED) == 0) {
151 		TMPFS_NODE_LOCK(dnode);
152 		dnode->tn_status |= TMPFS_NODE_ACCESSED;
153 		TMPFS_NODE_UNLOCK(dnode);
154 	}
155 
156 	/*
157 	 * Store the result of this lookup in the cache.  Avoid this if the
158 	 * request was for creation, as it does not improve timings on
159 	 * emprical tests.
160 	 */
161 	if (vp) {
162 		vn_unlock(vp);
163 		cache_setvp(ap->a_nch, vp);
164 		vrele(vp);
165 	} else if (error == ENOENT) {
166 		cache_setvp(ap->a_nch, NULL);
167 	}
168 	return (error);
169 }
170 
171 static int
172 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
173 {
174 	struct vnode *dvp = ap->a_dvp;
175 	struct vnode **vpp = ap->a_vpp;
176 	struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
177 	struct ucred *cred = ap->a_cred;
178 	int error;
179 
180 	*vpp = NULL;
181 
182 	/* Check accessibility of requested node as a first step. */
183 	error = VOP_ACCESS(dvp, VEXEC, cred);
184 	if (error != 0)
185 		return error;
186 
187 	if (dnode->tn_dir.tn_parent != NULL) {
188 		/* Allocate a new vnode on the matching entry. */
189 		error = tmpfs_alloc_vp(dvp->v_mount,
190 				       NULL, dnode->tn_dir.tn_parent,
191 				       LK_EXCLUSIVE | LK_RETRY, vpp);
192 
193 		if (*vpp)
194 			vn_unlock(*vpp);
195 	}
196 	return (*vpp == NULL) ? ENOENT : 0;
197 }
198 
199 /* --------------------------------------------------------------------- */
200 
201 static int
202 tmpfs_ncreate(struct vop_ncreate_args *ap)
203 {
204 	struct vnode *dvp = ap->a_dvp;
205 	struct vnode **vpp = ap->a_vpp;
206 	struct namecache *ncp = ap->a_nch->ncp;
207 	struct vattr *vap = ap->a_vap;
208 	struct ucred *cred = ap->a_cred;
209 	int error;
210 
211 	KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
212 
213 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
214 	if (error == 0) {
215 		cache_setunresolved(ap->a_nch);
216 		cache_setvp(ap->a_nch, *vpp);
217 		tmpfs_knote(dvp, NOTE_WRITE);
218 	}
219 	return (error);
220 }
221 /* --------------------------------------------------------------------- */
222 
223 static int
224 tmpfs_nmknod(struct vop_nmknod_args *ap)
225 {
226 	struct vnode *dvp = ap->a_dvp;
227 	struct vnode **vpp = ap->a_vpp;
228 	struct namecache *ncp = ap->a_nch->ncp;
229 	struct vattr *vap = ap->a_vap;
230 	struct ucred *cred = ap->a_cred;
231 	int error;
232 
233 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
234 	    vap->va_type != VFIFO) {
235 		return (EINVAL);
236 	}
237 
238 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
239 	if (error == 0) {
240 		cache_setunresolved(ap->a_nch);
241 		cache_setvp(ap->a_nch, *vpp);
242 		tmpfs_knote(dvp, NOTE_WRITE);
243 	}
244 	return error;
245 }
246 
247 /* --------------------------------------------------------------------- */
248 
249 static int
250 tmpfs_open(struct vop_open_args *ap)
251 {
252 	struct vnode *vp = ap->a_vp;
253 	int mode = ap->a_mode;
254 	struct tmpfs_node *node;
255 	int error;
256 
257 	node = VP_TO_TMPFS_NODE(vp);
258 
259 #if 0
260 	/* The file is still active but all its names have been removed
261 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
262 	 * it is about to die. */
263 	if (node->tn_links < 1)
264 		return (ENOENT);
265 #endif
266 
267 	/* If the file is marked append-only, deny write requests. */
268 	if ((node->tn_flags & APPEND) &&
269 	    (mode & (FWRITE | O_APPEND)) == FWRITE) {
270 		error = EPERM;
271 	} else {
272 		if (node->tn_reg.tn_pages_in_aobj) {
273 			TMPFS_NODE_LOCK(node);
274 			if (node->tn_reg.tn_pages_in_aobj) {
275 				tmpfs_move_pages(node->tn_reg.tn_aobj,
276 						 vp->v_object,
277 						 TMPFS_MOVF_FROMBACKING);
278 				node->tn_reg.tn_pages_in_aobj = 0;
279 			}
280 			TMPFS_NODE_UNLOCK(node);
281 		}
282 		error = vop_stdopen(ap);
283 	}
284 
285 	return (error);
286 }
287 
288 /* --------------------------------------------------------------------- */
289 
290 static int
291 tmpfs_close(struct vop_close_args *ap)
292 {
293 	struct vnode *vp = ap->a_vp;
294 	struct tmpfs_node *node;
295 	int error;
296 
297 	node = VP_TO_TMPFS_NODE(vp);
298 
299 	if (node->tn_links > 0) {
300 		/*
301 		 * Update node times.  No need to do it if the node has
302 		 * been deleted, because it will vanish after we return.
303 		 */
304 		tmpfs_update(vp);
305 	}
306 
307 	error = vop_stdclose(ap);
308 
309 	return (error);
310 }
311 
312 /* --------------------------------------------------------------------- */
313 
314 int
315 tmpfs_access(struct vop_access_args *ap)
316 {
317 	struct vnode *vp = ap->a_vp;
318 	int error;
319 	struct tmpfs_node *node;
320 
321 	node = VP_TO_TMPFS_NODE(vp);
322 
323 	switch (vp->v_type) {
324 	case VDIR:
325 		/* FALLTHROUGH */
326 	case VLNK:
327 		/* FALLTHROUGH */
328 	case VREG:
329 		if ((ap->a_mode & VWRITE) &&
330 	            (vp->v_mount->mnt_flag & MNT_RDONLY)) {
331 			error = EROFS;
332 			goto out;
333 		}
334 		break;
335 
336 	case VBLK:
337 		/* FALLTHROUGH */
338 	case VCHR:
339 		/* FALLTHROUGH */
340 	case VSOCK:
341 		/* FALLTHROUGH */
342 	case VFIFO:
343 		break;
344 
345 	default:
346 		error = EINVAL;
347 		goto out;
348 	}
349 
350 	if ((ap->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
351 		error = EPERM;
352 		goto out;
353 	}
354 
355 	error = vop_helper_access(ap, node->tn_uid, node->tn_gid,
356 			          node->tn_mode, 0);
357 out:
358 	return error;
359 }
360 
361 /* --------------------------------------------------------------------- */
362 
363 int
364 tmpfs_getattr(struct vop_getattr_args *ap)
365 {
366 	struct vnode *vp = ap->a_vp;
367 	struct vattr *vap = ap->a_vap;
368 	struct tmpfs_node *node;
369 
370 	node = VP_TO_TMPFS_NODE(vp);
371 
372 	tmpfs_update(vp);
373 
374 	vap->va_type = vp->v_type;
375 	vap->va_mode = node->tn_mode;
376 	vap->va_nlink = node->tn_links;
377 	vap->va_uid = node->tn_uid;
378 	vap->va_gid = node->tn_gid;
379 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
380 	vap->va_fileid = node->tn_id;
381 	vap->va_size = node->tn_size;
382 	vap->va_blocksize = PAGE_SIZE;
383 	vap->va_atime.tv_sec = node->tn_atime;
384 	vap->va_atime.tv_nsec = node->tn_atimensec;
385 	vap->va_mtime.tv_sec = node->tn_mtime;
386 	vap->va_mtime.tv_nsec = node->tn_mtimensec;
387 	vap->va_ctime.tv_sec = node->tn_ctime;
388 	vap->va_ctime.tv_nsec = node->tn_ctimensec;
389 	vap->va_gen = node->tn_gen;
390 	vap->va_flags = node->tn_flags;
391 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
392 		vap->va_rmajor = umajor(node->tn_rdev);
393 		vap->va_rminor = uminor(node->tn_rdev);
394 	}
395 	vap->va_bytes = round_page(node->tn_size);
396 	vap->va_filerev = 0;
397 
398 	return 0;
399 }
400 
401 /* --------------------------------------------------------------------- */
402 
403 int
404 tmpfs_getattr_quick(struct vop_getattr_args *ap)
405 {
406 	struct vnode *vp = ap->a_vp;
407 	struct vattr *vap = ap->a_vap;
408 	struct tmpfs_node *node;
409 
410 	node = VP_TO_TMPFS_NODE(vp);
411 
412 	tmpfs_update(vp);
413 
414 	vap->va_type = vp->v_type;
415 	vap->va_mode = node->tn_mode;
416 	vap->va_nlink = node->tn_links;
417 	vap->va_uid = node->tn_uid;
418 	vap->va_gid = node->tn_gid;
419 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
420 	vap->va_fileid = node->tn_id;
421 	vap->va_size = node->tn_size;
422 	vap->va_blocksize = PAGE_SIZE;
423 	vap->va_gen = node->tn_gen;
424 	vap->va_flags = node->tn_flags;
425 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
426 		vap->va_rmajor = umajor(node->tn_rdev);
427 		vap->va_rminor = uminor(node->tn_rdev);
428 	}
429 	vap->va_bytes = -1;
430 	vap->va_filerev = 0;
431 
432 	return 0;
433 }
434 
435 
436 /* --------------------------------------------------------------------- */
437 
438 int
439 tmpfs_setattr(struct vop_setattr_args *ap)
440 {
441 	struct vnode *vp = ap->a_vp;
442 	struct vattr *vap = ap->a_vap;
443 	struct ucred *cred = ap->a_cred;
444 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
445 	int error = 0;
446 	int kflags = 0;
447 
448 	TMPFS_NODE_LOCK(node);
449 	if (error == 0 && (vap->va_flags != VNOVAL)) {
450 		error = tmpfs_chflags(vp, vap->va_flags, cred);
451 		kflags |= NOTE_ATTRIB;
452 	}
453 
454 	if (error == 0 && (vap->va_size != VNOVAL)) {
455 		/* restore any saved pages before proceeding */
456 		if (node->tn_reg.tn_pages_in_aobj) {
457 			tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object,
458 					 TMPFS_MOVF_FROMBACKING |
459 					 TMPFS_MOVF_DEACTIVATE);
460 			node->tn_reg.tn_pages_in_aobj = 0;
461 		}
462 		if (vap->va_size > node->tn_size)
463 			kflags |= NOTE_WRITE | NOTE_EXTEND;
464 		else
465 			kflags |= NOTE_WRITE;
466 		error = tmpfs_chsize(vp, vap->va_size, cred);
467 	}
468 
469 	if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
470 			   vap->va_gid != (gid_t)VNOVAL)) {
471 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
472 		kflags |= NOTE_ATTRIB;
473 	}
474 
475 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
476 		error = tmpfs_chmod(vp, vap->va_mode, cred);
477 		kflags |= NOTE_ATTRIB;
478 	}
479 
480 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
481 	    vap->va_atime.tv_nsec != VNOVAL) ||
482 	    (vap->va_mtime.tv_sec != VNOVAL &&
483 	    vap->va_mtime.tv_nsec != VNOVAL) )) {
484 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
485 				      vap->va_vaflags, cred);
486 		kflags |= NOTE_ATTRIB;
487 	}
488 
489 	/*
490 	 * Update the node times.  We give preference to the error codes
491 	 * generated by this function rather than the ones that may arise
492 	 * from tmpfs_update.
493 	 */
494 	tmpfs_update(vp);
495 	TMPFS_NODE_UNLOCK(node);
496 	tmpfs_knote(vp, kflags);
497 
498 	return (error);
499 }
500 
501 /* --------------------------------------------------------------------- */
502 
503 /*
504  * fsync is usually a NOP, but we must take action when unmounting or
505  * when recycling.
506  */
507 static int
508 tmpfs_fsync(struct vop_fsync_args *ap)
509 {
510 	struct tmpfs_node *node;
511 	struct vnode *vp = ap->a_vp;
512 
513 	node = VP_TO_TMPFS_NODE(vp);
514 
515 	/*
516 	 * tmpfs vnodes typically remain dirty, avoid long syncer scans
517 	 * by forcing removal from the syncer list.
518 	 */
519 	vn_syncer_remove(vp, 1);
520 
521 	tmpfs_update(vp);
522 	if (vp->v_type == VREG) {
523 		if (vp->v_flag & VRECLAIMED) {
524 			if (node->tn_links == 0)
525 				tmpfs_truncate(vp, 0);
526 			else
527 				vfsync(ap->a_vp, ap->a_waitfor, 1, NULL, NULL);
528 		}
529 	}
530 
531 	return 0;
532 }
533 
534 /* --------------------------------------------------------------------- */
535 
536 static int
537 tmpfs_read(struct vop_read_args *ap)
538 {
539 	struct buf *bp;
540 	struct vnode *vp = ap->a_vp;
541 	struct uio *uio = ap->a_uio;
542 	struct tmpfs_node *node;
543 	off_t base_offset;
544 	size_t offset;
545 	size_t len;
546 	size_t resid;
547 	int error;
548 	int seqcount;
549 
550 	/*
551 	 * Check the basics
552 	 */
553 	if (uio->uio_offset < 0)
554 		return (EINVAL);
555 	if (vp->v_type != VREG)
556 		return (EINVAL);
557 
558 	/*
559 	 * Extract node, try to shortcut the operation through
560 	 * the VM page cache, allowing us to avoid buffer cache
561 	 * overheads.
562 	 */
563 	node = VP_TO_TMPFS_NODE(vp);
564         resid = uio->uio_resid;
565 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
566         error = vop_helper_read_shortcut(ap);
567         if (error)
568                 return error;
569         if (uio->uio_resid == 0) {
570 		if (resid)
571 			goto finished;
572 		return error;
573 	}
574 
575 	/*
576 	 * restore any saved pages before proceeding
577 	 */
578 	if (node->tn_reg.tn_pages_in_aobj) {
579 		TMPFS_NODE_LOCK(node);
580 		if (node->tn_reg.tn_pages_in_aobj) {
581 			tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object,
582 					 TMPFS_MOVF_FROMBACKING);
583 			node->tn_reg.tn_pages_in_aobj = 0;
584 		}
585 		TMPFS_NODE_UNLOCK(node);
586 	}
587 
588 	/*
589 	 * Fall-through to our normal read code.
590 	 */
591 	while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
592 		/*
593 		 * Use buffer cache I/O (via tmpfs_strategy)
594 		 */
595 		offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64;
596 		base_offset = (off_t)uio->uio_offset - offset;
597 		bp = getcacheblk(vp, base_offset,
598 				 node->tn_blksize, GETBLK_KVABIO);
599 		if (bp == NULL) {
600 			if (tmpfs_cluster_rd_enable) {
601 				error = cluster_readx(vp, node->tn_size,
602 						     base_offset,
603 						     node->tn_blksize,
604 						     B_NOTMETA | B_KVABIO,
605 						     uio->uio_resid,
606 						     seqcount * MAXBSIZE,
607 						     &bp);
608 			} else {
609 				error = bread_kvabio(vp, base_offset,
610 						     node->tn_blksize, &bp);
611 			}
612 			if (error) {
613 				brelse(bp);
614 				kprintf("tmpfs_read bread error %d\n", error);
615 				break;
616 			}
617 
618 			/*
619 			 * tmpfs pretty much fiddles directly with the VM
620 			 * system, don't let it exhaust it or we won't play
621 			 * nice with other processes.
622 			 *
623 			 * Only do this if the VOP is coming from a normal
624 			 * read/write.  The VM system handles the case for
625 			 * UIO_NOCOPY.
626 			 */
627 			if (uio->uio_segflg != UIO_NOCOPY)
628 				vm_wait_nominal();
629 		}
630 		bp->b_flags |= B_CLUSTEROK;
631 		bkvasync(bp);
632 
633 		/*
634 		 * Figure out how many bytes we can actually copy this loop.
635 		 */
636 		len = node->tn_blksize - offset;
637 		if (len > uio->uio_resid)
638 			len = uio->uio_resid;
639 		if (len > node->tn_size - uio->uio_offset)
640 			len = (size_t)(node->tn_size - uio->uio_offset);
641 
642 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
643 		bqrelse(bp);
644 		if (error) {
645 			kprintf("tmpfs_read uiomove error %d\n", error);
646 			break;
647 		}
648 	}
649 
650 finished:
651 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
652 		TMPFS_NODE_LOCK(node);
653 		node->tn_status |= TMPFS_NODE_ACCESSED;
654 		TMPFS_NODE_UNLOCK(node);
655 	}
656 	return (error);
657 }
658 
659 static int
660 tmpfs_write(struct vop_write_args *ap)
661 {
662 	struct buf *bp;
663 	struct vnode *vp = ap->a_vp;
664 	struct uio *uio = ap->a_uio;
665 	struct thread *td = uio->uio_td;
666 	struct tmpfs_node *node;
667 	boolean_t extended;
668 	off_t oldsize;
669 	int error;
670 	off_t base_offset;
671 	size_t offset;
672 	size_t len;
673 	struct rlimit limit;
674 	int trivial = 0;
675 	int kflags = 0;
676 	int seqcount;
677 
678 	error = 0;
679 	if (uio->uio_resid == 0) {
680 		return error;
681 	}
682 
683 	node = VP_TO_TMPFS_NODE(vp);
684 
685 	if (vp->v_type != VREG)
686 		return (EINVAL);
687 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
688 
689 	TMPFS_NODE_LOCK(node);
690 
691 	/*
692 	 * restore any saved pages before proceeding
693 	 */
694 	if (node->tn_reg.tn_pages_in_aobj) {
695 		tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object,
696 				 TMPFS_MOVF_FROMBACKING);
697 		node->tn_reg.tn_pages_in_aobj = 0;
698 	}
699 
700 	oldsize = node->tn_size;
701 	if (ap->a_ioflag & IO_APPEND)
702 		uio->uio_offset = node->tn_size;
703 
704 	/*
705 	 * Check for illegal write offsets.
706 	 */
707 	if (uio->uio_offset + uio->uio_resid >
708 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
709 		error = EFBIG;
710 		goto done;
711 	}
712 
713 	/*
714 	 * NOTE: Ignore if UIO does not come from a user thread (e.g. VN).
715 	 */
716 	if (vp->v_type == VREG && td != NULL && td->td_lwp != NULL) {
717 		error = kern_getrlimit(RLIMIT_FSIZE, &limit);
718 		if (error)
719 			goto done;
720 		if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
721 			ksignal(td->td_proc, SIGXFSZ);
722 			error = EFBIG;
723 			goto done;
724 		}
725 	}
726 
727 	/*
728 	 * Extend the file's size if necessary
729 	 */
730 	extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
731 
732 	while (uio->uio_resid > 0) {
733 		/*
734 		 * Don't completely blow out running buffer I/O
735 		 * when being hit from the pageout daemon.
736 		 */
737 		if (uio->uio_segflg == UIO_NOCOPY &&
738 		    (ap->a_ioflag & IO_RECURSE) == 0) {
739 			bwillwrite(node->tn_blksize);
740 		}
741 
742 		/*
743 		 * Use buffer cache I/O (via tmpfs_strategy)
744 		 *
745 		 * Calculate the maximum bytes we can write to the buffer at
746 		 * this offset (after resizing).
747 		 */
748 		offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64;
749 		base_offset = (off_t)uio->uio_offset - offset;
750 		len = uio->uio_resid;
751 		if (len > TMPFS_BLKSIZE - offset)
752 			len = TMPFS_BLKSIZE - offset;
753 
754 		if ((uio->uio_offset + len) > node->tn_size) {
755 			trivial = (uio->uio_offset <= node->tn_size);
756 			error = tmpfs_reg_resize(vp, uio->uio_offset + len,
757 						 trivial);
758 			if (error)
759 				break;
760 		}
761 
762 		/*
763 		 * Read to fill in any gaps.  Theoretically we could
764 		 * optimize this if the write covers the entire buffer
765 		 * and is not a UIO_NOCOPY write, however this can lead
766 		 * to a security violation exposing random kernel memory
767 		 * (whatever junk was in the backing VM pages before).
768 		 *
769 		 * So just use bread() to do the right thing.
770 		 */
771 		error = bread_kvabio(vp, base_offset, node->tn_blksize, &bp);
772 		bkvasync(bp);
773 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
774 		if (error) {
775 			kprintf("tmpfs_write uiomove error %d\n", error);
776 			brelse(bp);
777 			break;
778 		}
779 
780 		if (uio->uio_offset > node->tn_size) {
781 			node->tn_size = uio->uio_offset;
782 			kflags |= NOTE_EXTEND;
783 		}
784 		kflags |= NOTE_WRITE;
785 
786 		/*
787 		 * UIO_NOCOPY is a sensitive state due to potentially being
788 		 * issued from the pageout daemon while in a low-memory
789 		 * situation.  However, in order to cluster the I/O nicely
790 		 * (e.g. 64KB+ writes instead of 16KB writes), we still try
791 		 * to follow the same semantics that any other filesystem
792 		 * might use.
793 		 *
794 		 * For the normal case we buwrite(), dirtying the underlying
795 		 * VM pages instead of dirtying the buffer and releasing the
796 		 * buffer as a clean buffer.  This allows tmpfs to use
797 		 * essentially all available memory to cache file data.
798 		 * If we used bdwrite() the buffer cache would wind up
799 		 * flushing the data to swap too quickly.
800 		 *
801 		 * But because tmpfs can seriously load the VM system we
802 		 * fall-back to using bdwrite() when free memory starts
803 		 * to get low.  This shifts the load away from the VM system
804 		 * and makes tmpfs act more like a normal filesystem with
805 		 * regards to disk activity.
806 		 *
807 		 * tmpfs pretty much fiddles directly with the VM
808 		 * system, don't let it exhaust it or we won't play
809 		 * nice with other processes.  Only do this if the
810 		 * VOP is coming from a normal read/write.  The VM system
811 		 * handles the case for UIO_NOCOPY.
812 		 */
813 		bp->b_flags |= B_CLUSTEROK;
814 		if (uio->uio_segflg == UIO_NOCOPY) {
815 			/*
816 			 * Flush from the pageout daemon, deal with potentially
817 			 * very heavy tmpfs write activity causing long stalls
818 			 * in the pageout daemon before pages get to free/cache.
819 			 *
820 			 * We have to be careful not to bypass the page queues
821 			 * entirely or we can cause write-read thrashing and
822 			 * delay the paging of data that is more pageable then
823 			 * our current data.
824 			 *
825 			 * (a) Under severe pressure setting B_DIRECT will
826 			 *     cause a buffer release to try to free the
827 			 *     underlying pages.
828 			 *
829 			 * (b) Under modest memory pressure the B_AGE flag
830 			 *     we retire the buffer and its underlying pages
831 			 *     more quickly than normal.
832 			 *
833 			 *     We could also force this by setting B_NOTMETA
834 			 *     but that might have other unintended side-
835 			 *     effects (e.g. setting PG_NOTMETA on the VM page).
836 			 *
837 			 * (c) For the pageout->putpages->generic_putpages->
838 			 *     UIO_NOCOPY-write (here), issuing an immediate
839 			 *     write prevents any real clustering from
840 			 *     happening because the buffers probably aren't
841 			 *     (yet) marked dirty, or lost due to prior use
842 			 *     of buwrite().  Try to use the normal
843 			 *     cluster_write() mechanism for performance.
844 			 *
845 			 * Hopefully this will unblock the VM system more
846 			 * quickly under extreme tmpfs write load.
847 			 */
848 			if (tmpfs_bufcache_mode >= 2) {
849 				if (vm_page_count_min(vm_page_free_hysteresis))
850 					bp->b_flags |= B_DIRECT | B_TTC;
851 				if (vm_pages_needed || vm_paging_needed(0))
852 					bp->b_flags |= B_AGE;
853 			}
854 			bp->b_flags |= B_RELBUF;
855 			bp->b_act_count = 0;	/* buffer->deactivate pgs */
856 			if (tmpfs_cluster_wr_enable &&
857 			    (ap->a_ioflag & (IO_SYNC | IO_DIRECT)) == 0) {
858 				cluster_write(bp, node->tn_size,
859 					      node->tn_blksize, seqcount);
860 			} else {
861 				cluster_awrite(bp);
862 			}
863 		} else if (vm_page_count_min(0) ||
864 			   ((vm_pages_needed || vm_paging_needed(0)) &&
865 			    tmpfs_bufcache_mode >= 1)) {
866 			/*
867 			 * If the pageout daemon is running we cycle the
868 			 * write through the buffer cache normally to
869 			 * pipeline the flush, thus avoiding adding any
870 			 * more memory pressure to the pageout daemon.
871 			 */
872 			bp->b_act_count = 0;	/* buffer->deactivate pgs */
873 			if (tmpfs_cluster_wr_enable) {
874 				cluster_write(bp, node->tn_size,
875 					      node->tn_blksize, seqcount);
876 			} else {
877 				bdwrite(bp);
878 			}
879 		} else {
880 			/*
881 			 * Otherwise run the buffer directly through to the
882 			 * backing VM store, leaving the buffer clean so
883 			 * buffer limits do not force early flushes to swap.
884 			 */
885 			buwrite(bp);
886 			/*vm_wait_nominal();*/
887 		}
888 
889 		if (bp->b_error) {
890 			kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
891 			break;
892 		}
893 	}
894 
895 	if (error) {
896 		if (extended) {
897 			(void)tmpfs_reg_resize(vp, oldsize, trivial);
898 			kflags &= ~NOTE_EXTEND;
899 		}
900 		goto done;
901 	}
902 
903 	/*
904 	 * Currently we don't set the mtime on files modified via mmap()
905 	 * because we can't tell the difference between those modifications
906 	 * and an attempt by the pageout daemon to flush tmpfs pages to
907 	 * swap.
908 	 *
909 	 * This is because in order to defer flushes as long as possible
910 	 * buwrite() works by marking the underlying VM pages dirty in
911 	 * order to be able to dispose of the buffer cache buffer without
912 	 * flushing it.
913 	 */
914 	if (uio->uio_segflg == UIO_NOCOPY) {
915 		if (vp->v_flag & VLASTWRITETS) {
916 			node->tn_mtime = vp->v_lastwrite_ts.tv_sec;
917 			node->tn_mtimensec = vp->v_lastwrite_ts.tv_nsec;
918 		}
919 	} else {
920 		node->tn_status |= TMPFS_NODE_MODIFIED;
921 		vclrflags(vp, VLASTWRITETS);
922 	}
923 
924 	if (extended)
925 		node->tn_status |= TMPFS_NODE_CHANGED;
926 
927 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
928 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
929 			node->tn_mode &= ~(S_ISUID | S_ISGID);
930 	}
931 done:
932 	TMPFS_NODE_UNLOCK(node);
933 	if (kflags)
934 		tmpfs_knote(vp, kflags);
935 
936 	return(error);
937 }
938 
939 static int
940 tmpfs_advlock(struct vop_advlock_args *ap)
941 {
942 	struct tmpfs_node *node;
943 	struct vnode *vp = ap->a_vp;
944 	int error;
945 
946 	node = VP_TO_TMPFS_NODE(vp);
947 	error = (lf_advlock(ap, &node->tn_advlock, node->tn_size));
948 
949 	return (error);
950 }
951 
952 /*
953  * The strategy function is typically only called when memory pressure
954  * forces the system to attempt to pageout pages.  It can also be called
955  * by [n]vtruncbuf() when a truncation cuts a page in half.  Normal write
956  * operations
957  *
958  * We set VKVABIO for VREG files so bp->b_data may not be synchronized to
959  * our cpu.  swap_pager_strategy() is all we really use, and it directly
960  * supports this.
961  */
962 static int
963 tmpfs_strategy(struct vop_strategy_args *ap)
964 {
965 	struct bio *bio = ap->a_bio;
966 	struct bio *nbio;
967 	struct buf *bp = bio->bio_buf;
968 	struct vnode *vp = ap->a_vp;
969 	struct tmpfs_node *node;
970 	vm_object_t uobj;
971 	vm_page_t m;
972 	int i;
973 
974 	if (vp->v_type != VREG) {
975 		bp->b_resid = bp->b_bcount;
976 		bp->b_flags |= B_ERROR | B_INVAL;
977 		bp->b_error = EINVAL;
978 		biodone(bio);
979 		return(0);
980 	}
981 
982 	node = VP_TO_TMPFS_NODE(vp);
983 
984 	uobj = node->tn_reg.tn_aobj;
985 
986 	/*
987 	 * Don't bother flushing to swap if there is no swap, just
988 	 * ensure that the pages are marked as needing a commit (still).
989 	 */
990 	if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
991 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
992 			m = bp->b_xio.xio_pages[i];
993 			vm_page_need_commit(m);
994 		}
995 		bp->b_resid = 0;
996 		bp->b_error = 0;
997 		biodone(bio);
998 	} else {
999 #if 0
1000 		/*
1001 		 * XXX removed, this does not work well because under heavy
1002 		 * filesystem loads it often
1003 		 * forces the data to be read right back in again after
1004 		 * being written due to bypassing normal LRU operation.
1005 		 *
1006 		 * Tell the buffer cache to try to recycle the pages
1007 		 * to PQ_CACHE on release.
1008 		 */
1009 		if (tmpfs_bufcache_mode >= 2 ||
1010 		    (tmpfs_bufcache_mode == 1 && vm_paging_needed(0))) {
1011 			bp->b_flags |= B_TTC;
1012 		}
1013 #endif
1014 		nbio = push_bio(bio);
1015 		nbio->bio_done = tmpfs_strategy_done;
1016 		nbio->bio_offset = bio->bio_offset;
1017 		swap_pager_strategy(uobj, nbio);
1018 	}
1019 	return 0;
1020 }
1021 
1022 /*
1023  * If we were unable to commit the pages to swap make sure they are marked
1024  * as needing a commit (again).  If we were, clear the flag to allow the
1025  * pages to be freed.
1026  *
1027  * Do not error-out the buffer.  In particular, vinvalbuf() needs to
1028  * always work.
1029  */
1030 static void
1031 tmpfs_strategy_done(struct bio *bio)
1032 {
1033 	struct buf *bp;
1034 	vm_page_t m;
1035 	int i;
1036 
1037 	bp = bio->bio_buf;
1038 
1039 	if (bp->b_flags & B_ERROR) {
1040 		bp->b_flags &= ~B_ERROR;
1041 		bp->b_error = 0;
1042 		bp->b_resid = 0;
1043 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
1044 			m = bp->b_xio.xio_pages[i];
1045 			vm_page_need_commit(m);
1046 		}
1047 	} else {
1048 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
1049 			m = bp->b_xio.xio_pages[i];
1050 			vm_page_clear_commit(m);
1051 		}
1052 	}
1053 	bio = pop_bio(bio);
1054 	biodone(bio);
1055 }
1056 
1057 /*
1058  * To make write clustering work well make the backing store look
1059  * contiguous to the cluster_*() code.  The swap_strategy() function
1060  * will take it from there.
1061  *
1062  * Use MAXBSIZE-sized chunks as a micro-optimization to make random
1063  * flushes leave full-sized gaps.
1064  */
1065 static int
1066 tmpfs_bmap(struct vop_bmap_args *ap)
1067 {
1068 	if (ap->a_doffsetp != NULL)
1069 		*ap->a_doffsetp = ap->a_loffset;
1070 	if (ap->a_runp != NULL)
1071 		*ap->a_runp = MAXBSIZE - (ap->a_loffset & (MAXBSIZE - 1));
1072 	if (ap->a_runb != NULL)
1073 		*ap->a_runb = ap->a_loffset & (MAXBSIZE - 1);
1074 
1075 	return 0;
1076 }
1077 
1078 /* --------------------------------------------------------------------- */
1079 
1080 static int
1081 tmpfs_nremove(struct vop_nremove_args *ap)
1082 {
1083 	struct vnode *dvp = ap->a_dvp;
1084 	struct namecache *ncp = ap->a_nch->ncp;
1085 	struct vnode *vp;
1086 	int error;
1087 	struct tmpfs_dirent *de;
1088 	struct tmpfs_mount *tmp;
1089 	struct tmpfs_node *dnode;
1090 	struct tmpfs_node *node;
1091 
1092 	/*
1093 	 * We have to acquire the vp from ap->a_nch because we will likely
1094 	 * unresolve the namecache entry, and a vrele/vput is needed to
1095 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1096 	 *
1097 	 * We have to use vget to clear any inactive state on the vnode,
1098 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1099 	 * will not get called when we release it.
1100 	 */
1101 	error = cache_vget(ap->a_nch, ap->a_cred, LK_SHARED, &vp);
1102 	KKASSERT(vp->v_mount == dvp->v_mount);
1103 	KKASSERT(error == 0);
1104 	vn_unlock(vp);
1105 
1106 	if (vp->v_type == VDIR) {
1107 		error = EISDIR;
1108 		goto out2;
1109 	}
1110 
1111 	dnode = VP_TO_TMPFS_DIR(dvp);
1112 	node = VP_TO_TMPFS_NODE(vp);
1113 	tmp = VFS_TO_TMPFS(vp->v_mount);
1114 
1115 	TMPFS_NODE_LOCK(dnode);
1116 	TMPFS_NODE_LOCK(node);
1117 	de = tmpfs_dir_lookup(dnode, node, ncp);
1118 	if (de == NULL) {
1119 		error = ENOENT;
1120 		TMPFS_NODE_UNLOCK(node);
1121 		TMPFS_NODE_UNLOCK(dnode);
1122 		goto out;
1123 	}
1124 
1125 	/* Files marked as immutable or append-only cannot be deleted. */
1126 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
1127 	    (dnode->tn_flags & APPEND)) {
1128 		error = EPERM;
1129 		TMPFS_NODE_UNLOCK(node);
1130 		TMPFS_NODE_UNLOCK(dnode);
1131 		goto out;
1132 	}
1133 
1134 	/* Remove the entry from the directory; as it is a file, we do not
1135 	 * have to change the number of hard links of the directory. */
1136 	tmpfs_dir_detach_locked(dnode, de);
1137 	TMPFS_NODE_UNLOCK(dnode);
1138 
1139 	/* Free the directory entry we just deleted.  Note that the node
1140 	 * referred by it will not be removed until the vnode is really
1141 	 * reclaimed. */
1142 	tmpfs_free_dirent(tmp, de);
1143 
1144 	if (node->tn_links > 0)
1145 		node->tn_status |= TMPFS_NODE_CHANGED;
1146 	TMPFS_NODE_UNLOCK(node);
1147 
1148 	cache_unlink(ap->a_nch);
1149 	tmpfs_knote(vp, NOTE_DELETE);
1150 	error = 0;
1151 
1152 out:
1153 	if (error == 0)
1154 		tmpfs_knote(dvp, NOTE_WRITE);
1155 out2:
1156 	vrele(vp);
1157 
1158 	return error;
1159 }
1160 
1161 /* --------------------------------------------------------------------- */
1162 
1163 static int
1164 tmpfs_nlink(struct vop_nlink_args *ap)
1165 {
1166 	struct vnode *dvp = ap->a_dvp;
1167 	struct vnode *vp = ap->a_vp;
1168 	struct tmpfs_mount *tmp = VFS_TO_TMPFS(vp->v_mount);
1169 	struct namecache *ncp = ap->a_nch->ncp;
1170 	struct tmpfs_dirent *de;
1171 	struct tmpfs_node *node;
1172 	struct tmpfs_node *dnode;
1173 	int error;
1174 
1175 	KKASSERT(dvp != vp); /* XXX When can this be false? */
1176 
1177 	node = VP_TO_TMPFS_NODE(vp);
1178 	dnode = VP_TO_TMPFS_NODE(dvp);
1179 	TMPFS_NODE_LOCK(dnode);
1180 
1181 	/* XXX: Why aren't the following two tests done by the caller? */
1182 
1183 	/* Hard links of directories are forbidden. */
1184 	if (vp->v_type == VDIR) {
1185 		error = EPERM;
1186 		goto out;
1187 	}
1188 
1189 	/* Cannot create cross-device links. */
1190 	if (dvp->v_mount != vp->v_mount) {
1191 		error = EXDEV;
1192 		goto out;
1193 	}
1194 
1195 	/* Cannot hard-link into a deleted directory */
1196 	if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL) {
1197 		error = ENOENT;
1198 		goto out;
1199 	}
1200 
1201 	/* Ensure that we do not overflow the maximum number of links imposed
1202 	 * by the system. */
1203 	KKASSERT(node->tn_links <= LINK_MAX);
1204 	if (node->tn_links >= LINK_MAX) {
1205 		error = EMLINK;
1206 		goto out;
1207 	}
1208 
1209 	/* We cannot create links of files marked immutable or append-only. */
1210 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
1211 		error = EPERM;
1212 		goto out;
1213 	}
1214 
1215 	/* Allocate a new directory entry to represent the node. */
1216 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
1217 				   ncp->nc_name, ncp->nc_nlen, &de);
1218 	if (error != 0)
1219 		goto out;
1220 
1221 	/* Insert the new directory entry into the appropriate directory. */
1222 	tmpfs_dir_attach_locked(dnode, de);
1223 
1224 	/* vp link count has changed, so update node times. */
1225 
1226 	TMPFS_NODE_LOCK(node);
1227 	node->tn_status |= TMPFS_NODE_CHANGED;
1228 	TMPFS_NODE_UNLOCK(node);
1229 	tmpfs_update(vp);
1230 
1231 	tmpfs_knote(vp, NOTE_LINK);
1232 	cache_setunresolved(ap->a_nch);
1233 	cache_setvp(ap->a_nch, vp);
1234 	error = 0;
1235 
1236 out:
1237 	TMPFS_NODE_UNLOCK(dnode);
1238 	if (error == 0)
1239 		tmpfs_knote(dvp, NOTE_WRITE);
1240 	return error;
1241 }
1242 
1243 /* --------------------------------------------------------------------- */
1244 
1245 static int
1246 tmpfs_nrename(struct vop_nrename_args *ap)
1247 {
1248 	struct vnode *fdvp = ap->a_fdvp;
1249 	struct namecache *fncp = ap->a_fnch->ncp;
1250 	struct vnode *fvp = fncp->nc_vp;
1251 	struct vnode *tdvp = ap->a_tdvp;
1252 	struct namecache *tncp = ap->a_tnch->ncp;
1253 	struct vnode *tvp;
1254 	struct tmpfs_dirent *de, *tde;
1255 	struct tmpfs_mount *tmp;
1256 	struct tmpfs_node *fdnode;
1257 	struct tmpfs_node *tdnode;
1258 	struct tmpfs_node *fnode;
1259 	struct tmpfs_node *tnode;
1260 	char *newname;
1261 	char *oldname;
1262 	int error;
1263 
1264 	KKASSERT(fdvp->v_mount == fvp->v_mount);
1265 
1266 	/*
1267 	 * Because tvp can get overwritten we have to vget it instead of
1268 	 * just vref or use it, otherwise it's VINACTIVE flag may not get
1269 	 * cleared and the node won't get destroyed.
1270 	 */
1271 	error = cache_vget(ap->a_tnch, ap->a_cred, LK_SHARED, &tvp);
1272 	if (error == 0) {
1273 		tnode = VP_TO_TMPFS_NODE(tvp);
1274 		vn_unlock(tvp);
1275 	} else {
1276 		tnode = NULL;
1277 	}
1278 
1279 	/* Disallow cross-device renames.
1280 	 * XXX Why isn't this done by the caller? */
1281 	if (fvp->v_mount != tdvp->v_mount ||
1282 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
1283 		error = EXDEV;
1284 		goto out;
1285 	}
1286 
1287 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
1288 	tdnode = VP_TO_TMPFS_DIR(tdvp);
1289 
1290 	/* If source and target are the same file, there is nothing to do. */
1291 	if (fvp == tvp) {
1292 		error = 0;
1293 		goto out;
1294 	}
1295 
1296 	fdnode = VP_TO_TMPFS_DIR(fdvp);
1297 	fnode = VP_TO_TMPFS_NODE(fvp);
1298 
1299 	tmpfs_lock4(fdnode, tdnode, fnode, tnode);
1300 
1301 	/*
1302 	 * Cannot rename into a deleted directory
1303 	 */
1304 	if (tdnode != tmp->tm_root && tdnode->tn_dir.tn_parent == NULL) {
1305 		error = ENOENT;
1306 		goto out_locked;
1307 	}
1308 
1309 	/* Avoid manipulating '.' and '..' entries. */
1310 	de = tmpfs_dir_lookup(fdnode, fnode, fncp);
1311 	if (de == NULL) {
1312 		error = ENOENT;
1313 		goto out_locked;
1314 	}
1315 	KKASSERT(de->td_node == fnode);
1316 
1317 	/*
1318 	 * If replacing an entry in the target directory and that entry
1319 	 * is a directory, it must be empty.
1320 	 *
1321 	 * Kern_rename gurantees the destination to be a directory
1322 	 * if the source is one (it does?).
1323 	 */
1324 	if (tvp != NULL) {
1325 		KKASSERT(tnode != NULL);
1326 
1327 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1328 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1329 			error = EPERM;
1330 			goto out_locked;
1331 		}
1332 
1333 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1334 			if (tnode->tn_size > 0) {
1335 				error = ENOTEMPTY;
1336 				goto out_locked;
1337 			}
1338 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1339 			error = ENOTDIR;
1340 			goto out_locked;
1341 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1342 			error = EISDIR;
1343 			goto out_locked;
1344 		} else {
1345 			KKASSERT(fnode->tn_type != VDIR &&
1346 				tnode->tn_type != VDIR);
1347 		}
1348 	}
1349 
1350 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1351 	    (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1352 		error = EPERM;
1353 		goto out_locked;
1354 	}
1355 
1356 	/*
1357 	 * Ensure that we have enough memory to hold the new name, if it
1358 	 * has to be changed.
1359 	 */
1360 	if (fncp->nc_nlen != tncp->nc_nlen ||
1361 	    bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1362 		newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1363 				  M_WAITOK | M_NULLOK);
1364 		if (newname == NULL) {
1365 			error = ENOSPC;
1366 			goto out_locked;
1367 		}
1368 		bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1369 		newname[tncp->nc_nlen] = '\0';
1370 	} else {
1371 		newname = NULL;
1372 	}
1373 
1374 	/*
1375 	 * Unlink entry from source directory.  Note that the kernel has
1376 	 * already checked for illegal recursion cases (renaming a directory
1377 	 * into a subdirectory of itself).
1378 	 */
1379 	if (fdnode != tdnode) {
1380 		tmpfs_dir_detach_locked(fdnode, de);
1381 	} else {
1382 		/* XXX depend on namecache lock */
1383 		KKASSERT(de == tmpfs_dir_lookup(fdnode, fnode, fncp));
1384 		RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de);
1385 		RB_REMOVE(tmpfs_dirtree_cookie,
1386 			  &fdnode->tn_dir.tn_cookietree, de);
1387 	}
1388 
1389 	/*
1390 	 * Handle any name change.  Swap with newname, we will
1391 	 * deallocate it at the end.
1392 	 */
1393 	if (newname != NULL) {
1394 		oldname = de->td_name;
1395 		de->td_name = newname;
1396 		de->td_namelen = (uint16_t)tncp->nc_nlen;
1397 		newname = oldname;
1398 	}
1399 
1400 	/*
1401 	 * If we are overwriting an entry, we have to remove the old one
1402 	 * from the target directory.
1403 	 */
1404 	if (tvp != NULL) {
1405 		/* Remove the old entry from the target directory. */
1406 		tde = tmpfs_dir_lookup(tdnode, tnode, tncp);
1407 		tmpfs_dir_detach_locked(tdnode, tde);
1408 		tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1409 
1410 		/*
1411 		 * Free the directory entry we just deleted.  Note that the
1412 		 * node referred by it will not be removed until the vnode is
1413 		 * really reclaimed.
1414 		 */
1415 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1416 		/*cache_inval_vp(tvp, CINV_DESTROY);*/
1417 	}
1418 
1419 	/*
1420 	 * Link entry to target directory.  If the entry
1421 	 * represents a directory move the parent linkage
1422 	 * as well.
1423 	 */
1424 	if (fdnode != tdnode) {
1425 		if (de->td_node->tn_type == VDIR) {
1426 			TMPFS_VALIDATE_DIR(fnode);
1427 		}
1428 		tmpfs_dir_attach_locked(tdnode, de);
1429 	} else {
1430 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1431 		RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de);
1432 		RB_INSERT(tmpfs_dirtree_cookie,
1433 			  &tdnode->tn_dir.tn_cookietree, de);
1434 	}
1435 	tmpfs_unlock4(fdnode, tdnode, fnode, tnode);
1436 
1437 	/*
1438 	 * Finish up
1439 	 */
1440 	if (newname) {
1441 		kfree(newname, tmp->tm_name_zone);
1442 		newname = NULL;
1443 	}
1444 	cache_rename(ap->a_fnch, ap->a_tnch);
1445 	tmpfs_knote(ap->a_fdvp, NOTE_WRITE);
1446 	tmpfs_knote(ap->a_tdvp, NOTE_WRITE);
1447 	if (fnode->tn_vnode)
1448 		tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1449 	if (tvp)
1450 		vrele(tvp);
1451 	return 0;
1452 
1453 out_locked:
1454 	tmpfs_unlock4(fdnode, tdnode, fnode, tnode);
1455 out:
1456 	if (tvp)
1457 		vrele(tvp);
1458 	return error;
1459 }
1460 
1461 /* --------------------------------------------------------------------- */
1462 
1463 static int
1464 tmpfs_nmkdir(struct vop_nmkdir_args *ap)
1465 {
1466 	struct vnode *dvp = ap->a_dvp;
1467 	struct vnode **vpp = ap->a_vpp;
1468 	struct namecache *ncp = ap->a_nch->ncp;
1469 	struct vattr *vap = ap->a_vap;
1470 	struct ucred *cred = ap->a_cred;
1471 	int error;
1472 
1473 	KKASSERT(vap->va_type == VDIR);
1474 
1475 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1476 	if (error == 0) {
1477 		cache_setunresolved(ap->a_nch);
1478 		cache_setvp(ap->a_nch, *vpp);
1479 		tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1480 	}
1481 	return error;
1482 }
1483 
1484 /* --------------------------------------------------------------------- */
1485 
1486 static int
1487 tmpfs_nrmdir(struct vop_nrmdir_args *ap)
1488 {
1489 	struct vnode *dvp = ap->a_dvp;
1490 	struct namecache *ncp = ap->a_nch->ncp;
1491 	struct vnode *vp;
1492 	struct tmpfs_dirent *de;
1493 	struct tmpfs_mount *tmp;
1494 	struct tmpfs_node *dnode;
1495 	struct tmpfs_node *node;
1496 	int error;
1497 
1498 	/*
1499 	 * We have to acquire the vp from ap->a_nch because we will likely
1500 	 * unresolve the namecache entry, and a vrele/vput is needed to
1501 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1502 	 *
1503 	 * We have to use vget to clear any inactive state on the vnode,
1504 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1505 	 * will not get called when we release it.
1506 	 */
1507 	error = cache_vget(ap->a_nch, ap->a_cred, LK_SHARED, &vp);
1508 	KKASSERT(error == 0);
1509 	vn_unlock(vp);
1510 
1511 	/*
1512 	 * Prevalidate so we don't hit an assertion later
1513 	 */
1514 	if (vp->v_type != VDIR) {
1515 		error = ENOTDIR;
1516 		goto out;
1517 	}
1518 
1519 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1520 	dnode = VP_TO_TMPFS_DIR(dvp);
1521 	node = VP_TO_TMPFS_DIR(vp);
1522 
1523 	/*
1524 	 *
1525 	 */
1526 	TMPFS_NODE_LOCK(dnode);
1527 	TMPFS_NODE_LOCK(node);
1528 
1529 	/*
1530 	 * Only empty directories can be removed.
1531 	 */
1532 	if (node->tn_size > 0) {
1533 		error = ENOTEMPTY;
1534 		goto out_locked;
1535 	}
1536 
1537 	if ((dnode->tn_flags & APPEND)
1538 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1539 		error = EPERM;
1540 		goto out_locked;
1541 	}
1542 
1543 	/*
1544 	 * This invariant holds only if we are not trying to
1545 	 * remove "..".  We checked for that above so this is safe now.
1546 	 */
1547 	KKASSERT(node->tn_dir.tn_parent == dnode);
1548 
1549 	/*
1550 	 * Get the directory entry associated with node (vp)
1551 	 */
1552 	de = tmpfs_dir_lookup(dnode, node, ncp);
1553 	KKASSERT(TMPFS_DIRENT_MATCHES(de, ncp->nc_name, ncp->nc_nlen));
1554 
1555 	/* Check flags to see if we are allowed to remove the directory. */
1556 	if ((dnode->tn_flags & APPEND) ||
1557 	    node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1558 		error = EPERM;
1559 		goto out_locked;
1560 	}
1561 
1562 	/* Detach the directory entry from the directory (dnode). */
1563 	tmpfs_dir_detach_locked(dnode, de);
1564 
1565 	/*
1566 	 * Must set parent linkage to NULL (tested by ncreate to disallow
1567 	 * the creation of new files/dirs in a deleted directory)
1568 	 */
1569 	node->tn_status |= TMPFS_NODE_CHANGED;
1570 
1571 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1572 			    TMPFS_NODE_MODIFIED;
1573 
1574 	/* Free the directory entry we just deleted.  Note that the node
1575 	 * referred by it will not be removed until the vnode is really
1576 	 * reclaimed. */
1577 	tmpfs_free_dirent(tmp, de);
1578 
1579 	/* Release the deleted vnode (will destroy the node, notify
1580 	 * interested parties and clean it from the cache). */
1581 
1582 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1583 
1584 	TMPFS_NODE_UNLOCK(node);
1585 	TMPFS_NODE_UNLOCK(dnode);
1586 
1587 	tmpfs_update(dvp);
1588 	cache_unlink(ap->a_nch);
1589 	tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1590 	vrele(vp);
1591 	return 0;
1592 
1593 out_locked:
1594 	TMPFS_NODE_UNLOCK(node);
1595 	TMPFS_NODE_UNLOCK(dnode);
1596 
1597 out:
1598 	vrele(vp);
1599 
1600 	return error;
1601 }
1602 
1603 /* --------------------------------------------------------------------- */
1604 
1605 static int
1606 tmpfs_nsymlink(struct vop_nsymlink_args *ap)
1607 {
1608 	struct vnode *dvp = ap->a_dvp;
1609 	struct vnode **vpp = ap->a_vpp;
1610 	struct namecache *ncp = ap->a_nch->ncp;
1611 	struct vattr *vap = ap->a_vap;
1612 	struct ucred *cred = ap->a_cred;
1613 	char *target = ap->a_target;
1614 	int error;
1615 
1616 	vap->va_type = VLNK;
1617 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1618 	if (error == 0) {
1619 		tmpfs_knote(*vpp, NOTE_WRITE);
1620 		cache_setunresolved(ap->a_nch);
1621 		cache_setvp(ap->a_nch, *vpp);
1622 	}
1623 	return error;
1624 }
1625 
1626 /* --------------------------------------------------------------------- */
1627 
1628 static int
1629 tmpfs_readdir(struct vop_readdir_args *ap)
1630 {
1631 	struct vnode *vp = ap->a_vp;
1632 	struct uio *uio = ap->a_uio;
1633 	int *eofflag = ap->a_eofflag;
1634 	off_t **cookies = ap->a_cookies;
1635 	int *ncookies = ap->a_ncookies;
1636 	struct tmpfs_mount *tmp;
1637 	int error;
1638 	off_t startoff;
1639 	off_t cnt = 0;
1640 	struct tmpfs_node *node;
1641 
1642 	/* This operation only makes sense on directory nodes. */
1643 	if (vp->v_type != VDIR) {
1644 		return ENOTDIR;
1645 	}
1646 
1647 	tmp = VFS_TO_TMPFS(vp->v_mount);
1648 	node = VP_TO_TMPFS_DIR(vp);
1649 	startoff = uio->uio_offset;
1650 
1651 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1652 		error = tmpfs_dir_getdotdent(node, uio);
1653 		if (error != 0) {
1654 			TMPFS_NODE_LOCK_SH(node);
1655 			goto outok;
1656 		}
1657 		cnt++;
1658 	}
1659 
1660 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1661 		/* may lock parent, cannot hold node lock */
1662 		error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1663 		if (error != 0) {
1664 			TMPFS_NODE_LOCK_SH(node);
1665 			goto outok;
1666 		}
1667 		cnt++;
1668 	}
1669 
1670 	TMPFS_NODE_LOCK_SH(node);
1671 	error = tmpfs_dir_getdents(node, uio, &cnt);
1672 
1673 outok:
1674 	KKASSERT(error >= -1);
1675 
1676 	if (error == -1)
1677 		error = 0;
1678 
1679 	if (eofflag != NULL)
1680 		*eofflag =
1681 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1682 
1683 	/* Update NFS-related variables. */
1684 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1685 		off_t i;
1686 		off_t off = startoff;
1687 		struct tmpfs_dirent *de = NULL;
1688 
1689 		*ncookies = cnt;
1690 		*cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1691 
1692 		for (i = 0; i < cnt; i++) {
1693 			KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1694 			if (off == TMPFS_DIRCOOKIE_DOT) {
1695 				off = TMPFS_DIRCOOKIE_DOTDOT;
1696 			} else {
1697 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1698 					de = RB_MIN(tmpfs_dirtree_cookie,
1699 						&node->tn_dir.tn_cookietree);
1700 				} else if (de != NULL) {
1701 					de = RB_NEXT(tmpfs_dirtree_cookie,
1702 					       &node->tn_dir.tn_cookietree, de);
1703 				} else {
1704 					de = tmpfs_dir_lookupbycookie(node,
1705 								      off);
1706 					KKASSERT(de != NULL);
1707 					de = RB_NEXT(tmpfs_dirtree_cookie,
1708 					       &node->tn_dir.tn_cookietree, de);
1709 				}
1710 				if (de == NULL)
1711 					off = TMPFS_DIRCOOKIE_EOF;
1712 				else
1713 					off = tmpfs_dircookie(de);
1714 			}
1715 			(*cookies)[i] = off;
1716 		}
1717 		KKASSERT(uio->uio_offset == off);
1718 	}
1719 	TMPFS_NODE_UNLOCK(node);
1720 
1721 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
1722 		TMPFS_NODE_LOCK(node);
1723 		node->tn_status |= TMPFS_NODE_ACCESSED;
1724 		TMPFS_NODE_UNLOCK(node);
1725 	}
1726 	return error;
1727 }
1728 
1729 /* --------------------------------------------------------------------- */
1730 
1731 static int
1732 tmpfs_readlink(struct vop_readlink_args *ap)
1733 {
1734 	struct vnode *vp = ap->a_vp;
1735 	struct uio *uio = ap->a_uio;
1736 	int error;
1737 	struct tmpfs_node *node;
1738 
1739 	KKASSERT(uio->uio_offset == 0);
1740 	KKASSERT(vp->v_type == VLNK);
1741 
1742 	node = VP_TO_TMPFS_NODE(vp);
1743 	TMPFS_NODE_LOCK_SH(node);
1744 	error = uiomove(node->tn_link,
1745 			MIN(node->tn_size, uio->uio_resid), uio);
1746 	TMPFS_NODE_UNLOCK(node);
1747 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
1748 		TMPFS_NODE_LOCK(node);
1749 		node->tn_status |= TMPFS_NODE_ACCESSED;
1750 		TMPFS_NODE_UNLOCK(node);
1751 	}
1752 	return error;
1753 }
1754 
1755 /* --------------------------------------------------------------------- */
1756 
1757 static int
1758 tmpfs_inactive(struct vop_inactive_args *ap)
1759 {
1760 	struct vnode *vp = ap->a_vp;
1761 	struct tmpfs_node *node;
1762 	struct mount *mp;
1763 
1764 	mp = vp->v_mount;
1765 	lwkt_gettoken(&mp->mnt_token);
1766 	node = VP_TO_TMPFS_NODE(vp);
1767 
1768 	/*
1769 	 * Degenerate case
1770 	 */
1771 	if (node == NULL) {
1772 		vrecycle(vp);
1773 		lwkt_reltoken(&mp->mnt_token);
1774 		return(0);
1775 	}
1776 
1777 	/*
1778 	 * Get rid of unreferenced deleted vnodes sooner rather than
1779 	 * later so the data memory can be recovered immediately.
1780 	 *
1781 	 * We must truncate the vnode to prevent the normal reclamation
1782 	 * path from flushing the data for the removed file to disk.
1783 	 */
1784 	TMPFS_NODE_LOCK(node);
1785 	if (node->tn_links == 0) {
1786 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1787 		TMPFS_NODE_UNLOCK(node);
1788 		if (node->tn_type == VREG)
1789 			tmpfs_truncate(vp, 0);
1790 		vrecycle(vp);
1791 	} else {
1792 		/*
1793 		 * We must retain any VM pages belonging to the vnode's
1794 		 * object as the vnode will destroy the object during a
1795 		 * later reclaim.  We call vinvalbuf(V_SAVE) to clean
1796 		 * out the buffer cache.
1797 		 *
1798 		 * On DragonFlyBSD, vnodes are not immediately deactivated
1799 		 * on the 1->0 refs, so this is a relatively optimal
1800 		 * operation.  We have to do this in tmpfs_inactive()
1801 		 * because the pages will have already been thrown away
1802 		 * at the time tmpfs_reclaim() is called.
1803 		 */
1804 		if (node->tn_type == VREG &&
1805 		    node->tn_reg.tn_pages_in_aobj == 0) {
1806 			vinvalbuf(vp, V_SAVE, 0, 0);
1807 			KKASSERT(RB_EMPTY(&vp->v_rbdirty_tree));
1808 			KKASSERT(RB_EMPTY(&vp->v_rbclean_tree));
1809 			tmpfs_move_pages(vp->v_object, node->tn_reg.tn_aobj,
1810 					 TMPFS_MOVF_DEACTIVATE);
1811 			node->tn_reg.tn_pages_in_aobj = 1;
1812 		}
1813 
1814 		TMPFS_NODE_UNLOCK(node);
1815 	}
1816 	lwkt_reltoken(&mp->mnt_token);
1817 
1818 	return 0;
1819 }
1820 
1821 /* --------------------------------------------------------------------- */
1822 
1823 int
1824 tmpfs_reclaim(struct vop_reclaim_args *ap)
1825 {
1826 	struct vnode *vp = ap->a_vp;
1827 	struct tmpfs_mount *tmp;
1828 	struct tmpfs_node *node;
1829 	struct mount *mp;
1830 
1831 	mp = vp->v_mount;
1832 	lwkt_gettoken(&mp->mnt_token);
1833 
1834 	node = VP_TO_TMPFS_NODE(vp);
1835 	tmp = VFS_TO_TMPFS(vp->v_mount);
1836 	KKASSERT(mp == tmp->tm_mount);
1837 
1838         TMPFS_NODE_LOCK(node);
1839 	KKASSERT(node->tn_vnode == vp);
1840         node->tn_vnode = NULL;
1841         vp->v_data = NULL;
1842 
1843 	/*
1844 	 * If the node referenced by this vnode was deleted by the
1845 	 * user, we must free its associated data structures now that
1846 	 * the vnode is being reclaimed.
1847 	 *
1848 	 * Directories have an extra link ref.
1849 	 */
1850 	if (node->tn_links == 0) {
1851 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1852 		tmpfs_free_node(tmp, node);
1853 		/* eats the lock */
1854 	} else {
1855 		TMPFS_NODE_UNLOCK(node);
1856 	}
1857 	lwkt_reltoken(&mp->mnt_token);
1858 
1859 	KKASSERT(vp->v_data == NULL);
1860 	return 0;
1861 }
1862 
1863 /* --------------------------------------------------------------------- */
1864 
1865 static int
1866 tmpfs_mountctl(struct vop_mountctl_args *ap)
1867 {
1868 	struct tmpfs_mount *tmp;
1869 	struct mount *mp;
1870 	int rc;
1871 
1872 	mp = ap->a_head.a_ops->head.vv_mount;
1873 	lwkt_gettoken(&mp->mnt_token);
1874 
1875 	switch (ap->a_op) {
1876 	case (MOUNTCTL_SET_EXPORT):
1877 		tmp = (struct tmpfs_mount *) mp->mnt_data;
1878 
1879 		if (ap->a_ctllen != sizeof(struct export_args))
1880 			rc = (EINVAL);
1881 		else
1882 			rc = vfs_export(mp, &tmp->tm_export,
1883 					(const struct export_args *) ap->a_ctl);
1884 		break;
1885 	default:
1886 		rc = vop_stdmountctl(ap);
1887 		break;
1888 	}
1889 
1890 	lwkt_reltoken(&mp->mnt_token);
1891 	return (rc);
1892 }
1893 
1894 /* --------------------------------------------------------------------- */
1895 
1896 static int
1897 tmpfs_print(struct vop_print_args *ap)
1898 {
1899 	struct vnode *vp = ap->a_vp;
1900 
1901 	struct tmpfs_node *node;
1902 
1903 	node = VP_TO_TMPFS_NODE(vp);
1904 
1905 	kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1906 	    node, node->tn_flags, node->tn_links);
1907 	kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1908 	    node->tn_mode, node->tn_uid, node->tn_gid,
1909 	    (uintmax_t)node->tn_size, node->tn_status);
1910 
1911 	if (vp->v_type == VFIFO)
1912 		fifo_printinfo(vp);
1913 
1914 	kprintf("\n");
1915 
1916 	return 0;
1917 }
1918 
1919 /* --------------------------------------------------------------------- */
1920 
1921 static int
1922 tmpfs_pathconf(struct vop_pathconf_args *ap)
1923 {
1924 	struct vnode *vp = ap->a_vp;
1925 	int name = ap->a_name;
1926 	register_t *retval = ap->a_retval;
1927 	struct tmpfs_mount *tmp;
1928 	int error;
1929 
1930 	error = 0;
1931 
1932 	switch (name) {
1933 	case _PC_CHOWN_RESTRICTED:
1934 		*retval = 1;
1935 		break;
1936 
1937 	case _PC_FILESIZEBITS:
1938 		tmp = VFS_TO_TMPFS(vp->v_mount);
1939 		*retval = max(32, flsll(tmp->tm_pages_max * PAGE_SIZE) + 1);
1940 		break;
1941 
1942 	case _PC_LINK_MAX:
1943 		*retval = LINK_MAX;
1944 		break;
1945 
1946 	case _PC_NAME_MAX:
1947 		*retval = NAME_MAX;
1948 		break;
1949 
1950 	case _PC_NO_TRUNC:
1951 		*retval = 1;
1952 		break;
1953 
1954 	case _PC_PATH_MAX:
1955 		*retval = PATH_MAX;
1956 		break;
1957 
1958 	case _PC_PIPE_BUF:
1959 		*retval = PIPE_BUF;
1960 		break;
1961 
1962 	case _PC_SYNC_IO:
1963 		*retval = 1;
1964 		break;
1965 
1966 	case _PC_2_SYMLINKS:
1967 		*retval = 1;
1968 		break;
1969 
1970 	default:
1971 		error = EINVAL;
1972 	}
1973 
1974 	return error;
1975 }
1976 
1977 /************************************************************************
1978  *                          KQFILTER OPS                                *
1979  ************************************************************************/
1980 
1981 static void filt_tmpfsdetach(struct knote *kn);
1982 static int filt_tmpfsread(struct knote *kn, long hint);
1983 static int filt_tmpfswrite(struct knote *kn, long hint);
1984 static int filt_tmpfsvnode(struct knote *kn, long hint);
1985 
1986 static struct filterops tmpfsread_filtops =
1987 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1988 	  NULL, filt_tmpfsdetach, filt_tmpfsread };
1989 static struct filterops tmpfswrite_filtops =
1990 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1991 	  NULL, filt_tmpfsdetach, filt_tmpfswrite };
1992 static struct filterops tmpfsvnode_filtops =
1993 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1994 	  NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1995 
1996 static int
1997 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1998 {
1999 	struct vnode *vp = ap->a_vp;
2000 	struct knote *kn = ap->a_kn;
2001 
2002 	switch (kn->kn_filter) {
2003 	case EVFILT_READ:
2004 		kn->kn_fop = &tmpfsread_filtops;
2005 		break;
2006 	case EVFILT_WRITE:
2007 		kn->kn_fop = &tmpfswrite_filtops;
2008 		break;
2009 	case EVFILT_VNODE:
2010 		kn->kn_fop = &tmpfsvnode_filtops;
2011 		break;
2012 	default:
2013 		return (EOPNOTSUPP);
2014 	}
2015 
2016 	kn->kn_hook = (caddr_t)vp;
2017 
2018 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2019 
2020 	return(0);
2021 }
2022 
2023 static void
2024 filt_tmpfsdetach(struct knote *kn)
2025 {
2026 	struct vnode *vp = (void *)kn->kn_hook;
2027 
2028 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2029 }
2030 
2031 static int
2032 filt_tmpfsread(struct knote *kn, long hint)
2033 {
2034 	struct vnode *vp = (void *)kn->kn_hook;
2035 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
2036 	off_t off;
2037 
2038 	if (hint == NOTE_REVOKE) {
2039 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2040 		return(1);
2041 	}
2042 
2043 	/*
2044 	 * Interlock against MP races when performing this function.
2045 	 */
2046 	TMPFS_NODE_LOCK_SH(node);
2047 	off = node->tn_size - kn->kn_fp->f_offset;
2048 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2049 	if (kn->kn_sfflags & NOTE_OLDAPI) {
2050 		TMPFS_NODE_UNLOCK(node);
2051 		return(1);
2052 	}
2053 	if (kn->kn_data == 0) {
2054 		kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2055 	}
2056 	TMPFS_NODE_UNLOCK(node);
2057 	return (kn->kn_data != 0);
2058 }
2059 
2060 static int
2061 filt_tmpfswrite(struct knote *kn, long hint)
2062 {
2063 	if (hint == NOTE_REVOKE)
2064 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2065 	kn->kn_data = 0;
2066 	return (1);
2067 }
2068 
2069 static int
2070 filt_tmpfsvnode(struct knote *kn, long hint)
2071 {
2072 	if (kn->kn_sfflags & hint)
2073 		kn->kn_fflags |= hint;
2074 	if (hint == NOTE_REVOKE) {
2075 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2076 		return (1);
2077 	}
2078 	return (kn->kn_fflags != 0);
2079 }
2080 
2081 /*
2082  * Helper to move VM pages between objects
2083  *
2084  * NOTE: The vm_page_rename() dirties the page, so we can clear the
2085  *	 PG_NEED_COMMIT flag.  If the pages are being moved into tn_aobj,
2086  *	 the pageout daemon will be able to page them out.
2087  */
2088 static int
2089 tmpfs_move_pages_callback(vm_page_t p, void *data)
2090 {
2091 	struct rb_vm_page_scan_info *info = data;
2092 	vm_pindex_t pindex;
2093 
2094 	/*
2095 	 * Take control of the page
2096 	 */
2097 	pindex = p->pindex;
2098 	if (vm_page_busy_try(p, TRUE)) {
2099 		vm_page_sleep_busy(p, TRUE, "tpgmov");
2100 		info->error = -1;
2101 		return -1;
2102 	}
2103 	if (p->object != info->object || p->pindex != pindex) {
2104 		vm_page_wakeup(p);
2105 		info->error = -1;
2106 		return -1;
2107 	}
2108 
2109 	/*
2110 	 * Make sure the page is not mapped.  These flags might also still be
2111 	 * set heuristically even if we know the page is not mapped and must
2112 	 * be properly cleaned up.
2113 	 */
2114 	if (__predict_false((p->flags & (PG_MAPPED|PG_WRITEABLE)) != 0))
2115 		vm_page_protect(p, VM_PROT_NONE);
2116 
2117 	/*
2118 	 * Free or rename the page as appropriate
2119 	 */
2120 	if ((info->pagerflags & TMPFS_MOVF_FROMBACKING) &&
2121 	    (p->flags & PG_SWAPPED) &&
2122 	    (p->flags & PG_NEED_COMMIT) == 0 &&
2123 	    p->dirty == 0) {
2124 		/*
2125 		 * If the page in the backing aobj was paged out to swap
2126 		 * it will be clean and it is better to free it rather
2127 		 * than re-dirty it.  We will assume that the page was
2128 		 * paged out to swap for a reason!
2129 		 *
2130 		 * This helps avoid unnecessary swap thrashing on the page.
2131 		 */
2132 		vm_page_free(p);
2133 	} else if ((info->pagerflags & TMPFS_MOVF_FROMBACKING) == 0 &&
2134 		   (p->flags & PG_NEED_COMMIT) == 0 &&
2135 		   p->dirty == 0) {
2136 		/*
2137 		 * If the page associated with the vnode was cleaned via
2138 		 * a tmpfs_strategy() call, it exists as a swap block in
2139 		 * aobj and it is again better to free it rather than
2140 		 * re-dirty it.  We will assume that the page was
2141 		 * paged out to swap for a reason!
2142 		 *
2143 		 * This helps avoid unnecessary swap thrashing on the page.
2144 		 */
2145 		vm_page_free(p);
2146 	} else {
2147 		/*
2148 		 * Rename the page, which will also ensure that it is flagged
2149 		 * as dirty and check whether a swap block association exists
2150 		 * in the target object or not, setting appropriate flags if
2151 		 * it does.
2152 		 */
2153 		vm_page_rename(p, info->dest_object, pindex);
2154 		vm_page_clear_commit(p);
2155 		if (info->pagerflags & TMPFS_MOVF_DEACTIVATE)
2156 			vm_page_deactivate(p);
2157 		vm_page_wakeup(p);
2158 		/* page automaticaly made dirty */
2159 	}
2160 
2161 	return 0;
2162 }
2163 
2164 static
2165 void
2166 tmpfs_move_pages(vm_object_t src, vm_object_t dst, int movflags)
2167 {
2168 	struct rb_vm_page_scan_info info;
2169 
2170 	vm_object_hold(src);
2171 	vm_object_hold(dst);
2172 	info.object = src;
2173 	info.dest_object = dst;
2174 	info.pagerflags = movflags;
2175 	do {
2176 		if (src->paging_in_progress)
2177 			vm_object_pip_wait(src, "objtfs");
2178 		info.error = 1;
2179 		vm_page_rb_tree_RB_SCAN(&src->rb_memq, NULL,
2180 					tmpfs_move_pages_callback, &info);
2181 	} while (info.error < 0 || !RB_EMPTY(&src->rb_memq) ||
2182 		 src->paging_in_progress);
2183 	vm_object_drop(dst);
2184 	vm_object_drop(src);
2185 }
2186 
2187 /* --------------------------------------------------------------------- */
2188 
2189 /*
2190  * vnode operations vector used for files stored in a tmpfs file system.
2191  */
2192 struct vop_ops tmpfs_vnode_vops = {
2193 	.vop_default =			vop_defaultop,
2194 	.vop_getpages = 		vop_stdgetpages,
2195 	.vop_putpages = 		vop_stdputpages,
2196 	.vop_ncreate =			tmpfs_ncreate,
2197 	.vop_nresolve =			tmpfs_nresolve,
2198 	.vop_nlookupdotdot =		tmpfs_nlookupdotdot,
2199 	.vop_nmknod =			tmpfs_nmknod,
2200 	.vop_open =			tmpfs_open,
2201 	.vop_close =			tmpfs_close,
2202 	.vop_access =			tmpfs_access,
2203 	.vop_getattr =			tmpfs_getattr,
2204 	.vop_getattr_quick =		tmpfs_getattr_quick,
2205 	.vop_setattr =			tmpfs_setattr,
2206 	.vop_read =			tmpfs_read,
2207 	.vop_write =			tmpfs_write,
2208 	.vop_fsync =			tmpfs_fsync,
2209 	.vop_mountctl =			tmpfs_mountctl,
2210 	.vop_nremove =			tmpfs_nremove,
2211 	.vop_nlink =			tmpfs_nlink,
2212 	.vop_nrename =			tmpfs_nrename,
2213 	.vop_nmkdir =			tmpfs_nmkdir,
2214 	.vop_nrmdir =			tmpfs_nrmdir,
2215 	.vop_nsymlink =			tmpfs_nsymlink,
2216 	.vop_readdir =			tmpfs_readdir,
2217 	.vop_readlink =			tmpfs_readlink,
2218 	.vop_inactive =			tmpfs_inactive,
2219 	.vop_reclaim =			tmpfs_reclaim,
2220 	.vop_print =			tmpfs_print,
2221 	.vop_pathconf =			tmpfs_pathconf,
2222 	.vop_bmap =			tmpfs_bmap,
2223 	.vop_strategy =			tmpfs_strategy,
2224 	.vop_advlock =			tmpfs_advlock,
2225 	.vop_kqfilter =			tmpfs_kqfilter
2226 };
2227