xref: /netbsd-src/sys/fs/tmpfs/tmpfs_vnops.c (revision 88fcb00c0357f2d7c1774f86a352637bfda96184)
1 /*	$NetBSD: tmpfs_vnops.c,v 1.87 2011/06/12 03:35:54 rmind Exp $	*/
2 
3 /*
4  * Copyright (c) 2005, 2006, 2007 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9  * 2005 program.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * tmpfs vnode interface.
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: tmpfs_vnops.c,v 1.87 2011/06/12 03:35:54 rmind Exp $");
39 
40 #include <sys/param.h>
41 #include <sys/dirent.h>
42 #include <sys/fcntl.h>
43 #include <sys/event.h>
44 #include <sys/malloc.h>
45 #include <sys/namei.h>
46 #include <sys/stat.h>
47 #include <sys/uio.h>
48 #include <sys/unistd.h>
49 #include <sys/vnode.h>
50 #include <sys/lockf.h>
51 #include <sys/kauth.h>
52 
53 #include <uvm/uvm.h>
54 
55 #include <miscfs/fifofs/fifo.h>
56 #include <miscfs/genfs/genfs.h>
57 #include <fs/tmpfs/tmpfs_vnops.h>
58 #include <fs/tmpfs/tmpfs.h>
59 
60 /*
61  * vnode operations vector used for files stored in a tmpfs file system.
62  */
63 int (**tmpfs_vnodeop_p)(void *);
64 const struct vnodeopv_entry_desc tmpfs_vnodeop_entries[] = {
65 	{ &vop_default_desc,		vn_default_error },
66 	{ &vop_lookup_desc,		tmpfs_lookup },
67 	{ &vop_create_desc,		tmpfs_create },
68 	{ &vop_mknod_desc,		tmpfs_mknod },
69 	{ &vop_open_desc,		tmpfs_open },
70 	{ &vop_close_desc,		tmpfs_close },
71 	{ &vop_access_desc,		tmpfs_access },
72 	{ &vop_getattr_desc,		tmpfs_getattr },
73 	{ &vop_setattr_desc,		tmpfs_setattr },
74 	{ &vop_read_desc,		tmpfs_read },
75 	{ &vop_write_desc,		tmpfs_write },
76 	{ &vop_ioctl_desc,		tmpfs_ioctl },
77 	{ &vop_fcntl_desc,		tmpfs_fcntl },
78 	{ &vop_poll_desc,		tmpfs_poll },
79 	{ &vop_kqfilter_desc,		tmpfs_kqfilter },
80 	{ &vop_revoke_desc,		tmpfs_revoke },
81 	{ &vop_mmap_desc,		tmpfs_mmap },
82 	{ &vop_fsync_desc,		tmpfs_fsync },
83 	{ &vop_seek_desc,		tmpfs_seek },
84 	{ &vop_remove_desc,		tmpfs_remove },
85 	{ &vop_link_desc,		tmpfs_link },
86 	{ &vop_rename_desc,		tmpfs_rename },
87 	{ &vop_mkdir_desc,		tmpfs_mkdir },
88 	{ &vop_rmdir_desc,		tmpfs_rmdir },
89 	{ &vop_symlink_desc,		tmpfs_symlink },
90 	{ &vop_readdir_desc,		tmpfs_readdir },
91 	{ &vop_readlink_desc,		tmpfs_readlink },
92 	{ &vop_abortop_desc,		tmpfs_abortop },
93 	{ &vop_inactive_desc,		tmpfs_inactive },
94 	{ &vop_reclaim_desc,		tmpfs_reclaim },
95 	{ &vop_lock_desc,		tmpfs_lock },
96 	{ &vop_unlock_desc,		tmpfs_unlock },
97 	{ &vop_bmap_desc,		tmpfs_bmap },
98 	{ &vop_strategy_desc,		tmpfs_strategy },
99 	{ &vop_print_desc,		tmpfs_print },
100 	{ &vop_pathconf_desc,		tmpfs_pathconf },
101 	{ &vop_islocked_desc,		tmpfs_islocked },
102 	{ &vop_advlock_desc,		tmpfs_advlock },
103 	{ &vop_bwrite_desc,		tmpfs_bwrite },
104 	{ &vop_getpages_desc,		tmpfs_getpages },
105 	{ &vop_putpages_desc,		tmpfs_putpages },
106 #if TMPFS_WHITEOUT
107 	{ &vop_whiteout_desc,		tmpfs_whiteout },
108 #endif
109 	{ NULL, NULL }
110 };
111 
112 const struct vnodeopv_desc tmpfs_vnodeop_opv_desc = {
113 	&tmpfs_vnodeop_p, tmpfs_vnodeop_entries
114 };
115 
116 /*
117  * tmpfs_lookup: path name traversal routine.
118  *
119  * Arguments: dvp (directory being searched), vpp (result),
120  * cnp (component name - path).
121  *
122  * => Caller holds a reference and lock on dvp.
123  * => We return looked-up vnode (vpp) locked, with a reference held.
124  */
125 int
126 tmpfs_lookup(void *v)
127 {
128 	struct vop_lookup_args /* {
129 		struct vnode *a_dvp;
130 		struct vnode **a_vpp;
131 		struct componentname *a_cnp;
132 	} */ *ap = v;
133 	vnode_t *dvp = ap->a_dvp, **vpp = ap->a_vpp;
134 	struct componentname *cnp = ap->a_cnp;
135 	const bool lastcn = (cnp->cn_flags & ISLASTCN) != 0;
136 	tmpfs_node_t *dnode, *tnode;
137 	tmpfs_dirent_t *de;
138 	int error;
139 
140 	KASSERT(VOP_ISLOCKED(dvp));
141 
142 	dnode = VP_TO_TMPFS_DIR(dvp);
143 	*vpp = NULL;
144 
145 	/* Check accessibility of directory. */
146 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
147 	if (error) {
148 		goto out;
149 	}
150 
151 	/*
152 	 * If requesting the last path component on a read-only file system
153 	 * with a write operation, deny it.
154 	 */
155 	if (lastcn && (dvp->v_mount->mnt_flag & MNT_RDONLY) != 0 &&
156 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
157 		error = EROFS;
158 		goto out;
159 	}
160 
161 	/*
162 	 * Avoid doing a linear scan of the directory if the requested
163 	 * directory/name couple is already in the cache.
164 	 */
165 	error = cache_lookup(dvp, vpp, cnp);
166 	if (error >= 0) {
167 		/* Both cache-hit or an error case. */
168 		goto out;
169 	}
170 
171 	if (cnp->cn_flags & ISDOTDOT) {
172 		tmpfs_node_t *pnode;
173 
174 		/*
175 		 * Lookup of ".." case.
176 		 */
177 		if (lastcn && cnp->cn_nameiop == RENAME) {
178 			error = EINVAL;
179 			goto out;
180 		}
181 		KASSERT(dnode->tn_type == VDIR);
182 		pnode = dnode->tn_spec.tn_dir.tn_parent;
183 		if (pnode == NULL) {
184 			error = ENOENT;
185 			goto out;
186 		}
187 
188 		/*
189 		 * Lock the parent tn_vlock before releasing the vnode lock,
190 		 * and thus prevents parent from disappearing.
191 		 */
192 		mutex_enter(&pnode->tn_vlock);
193 		VOP_UNLOCK(dvp);
194 
195 		/*
196 		 * Get a vnode of the '..' entry and re-acquire the lock.
197 		 * Release the tn_vlock.
198 		 */
199 		error = tmpfs_vnode_get(dvp->v_mount, pnode, vpp);
200 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
201 		goto out;
202 
203 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
204 		/*
205 		 * Lookup of "." case.
206 		 */
207 		if (lastcn && cnp->cn_nameiop == RENAME) {
208 			error = EISDIR;
209 			goto out;
210 		}
211 		vref(dvp);
212 		*vpp = dvp;
213 		error = 0;
214 		goto done;
215 	}
216 
217 	/*
218 	 * Other lookup cases: perform directory scan.
219 	 */
220 	de = tmpfs_dir_lookup(dnode, cnp);
221 	if (de == NULL || de->td_node == TMPFS_NODE_WHITEOUT) {
222 		/*
223 		 * The entry was not found in the directory.  This is valid
224 		 * if we are creating or renaming an entry and are working
225 		 * on the last component of the path name.
226 		 */
227 		if (lastcn && (cnp->cn_nameiop == CREATE ||
228 		    cnp->cn_nameiop == RENAME)) {
229 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
230 			if (error) {
231 				goto out;
232 			}
233 			error = EJUSTRETURN;
234 		} else {
235 			error = ENOENT;
236 		}
237 		if (de) {
238 			KASSERT(de->td_node == TMPFS_NODE_WHITEOUT);
239 			cnp->cn_flags |= ISWHITEOUT;
240 		}
241 		goto done;
242 	}
243 
244 	tnode = de->td_node;
245 
246 	/*
247 	 * If it is not the last path component and found a non-directory
248 	 * or non-link entry (which may itself be pointing to a directory),
249 	 * raise an error.
250 	 */
251 	if (!lastcn && tnode->tn_type != VDIR && tnode->tn_type != VLNK) {
252 		error = ENOTDIR;
253 		goto out;
254 	}
255 
256 	/* Check the permissions. */
257 	if (lastcn && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
258 		kauth_action_t action = 0;
259 
260 		/* This is the file-system's decision. */
261 		if ((dnode->tn_mode & S_ISTXT) != 0 &&
262 		    kauth_cred_geteuid(cnp->cn_cred) != dnode->tn_uid &&
263 		    kauth_cred_geteuid(cnp->cn_cred) != tnode->tn_uid) {
264 			error = EPERM;
265 		} else {
266 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
267 		}
268 
269 		if (cnp->cn_nameiop == DELETE) {
270 			action |= KAUTH_VNODE_DELETE;
271 		} else {
272 			KASSERT(cnp->cn_nameiop == RENAME);
273 			action |= KAUTH_VNODE_RENAME;
274 		}
275 		error = kauth_authorize_vnode(cnp->cn_cred,
276 		    action, *vpp, dvp, error);
277 		if (error) {
278 			goto out;
279 		}
280 	}
281 
282 	/* Get a vnode for the matching entry. */
283 	mutex_enter(&tnode->tn_vlock);
284 	error = tmpfs_vnode_get(dvp->v_mount, tnode, vpp);
285 done:
286 	/*
287 	 * Cache the result, unless request was for creation (as it does
288 	 * not improve the performance).
289 	 */
290 	if ((cnp->cn_flags & MAKEENTRY) != 0 && cnp->cn_nameiop != CREATE) {
291 		cache_enter(dvp, *vpp, cnp);
292 	}
293 out:
294 	KASSERT((*vpp && VOP_ISLOCKED(*vpp)) || error);
295 	KASSERT(VOP_ISLOCKED(dvp));
296 
297 	return error;
298 }
299 
300 int
301 tmpfs_create(void *v)
302 {
303 	struct vop_create_args /* {
304 		struct vnode		*a_dvp;
305 		struct vnode		**a_vpp;
306 		struct componentname	*a_cnp;
307 		struct vattr		*a_vap;
308 	} */ *ap = v;
309 	vnode_t *dvp = ap->a_dvp, **vpp = ap->a_vpp;
310 	struct componentname *cnp = ap->a_cnp;
311 	struct vattr *vap = ap->a_vap;
312 
313 	KASSERT(VOP_ISLOCKED(dvp));
314 	KASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
315 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
316 }
317 
318 int
319 tmpfs_mknod(void *v)
320 {
321 	struct vop_mknod_args /* {
322 		struct vnode		*a_dvp;
323 		struct vnode		**a_vpp;
324 		struct componentname	*a_cnp;
325 		struct vattr		*a_vap;
326 	} */ *ap = v;
327 	vnode_t *dvp = ap->a_dvp, **vpp = ap->a_vpp;
328 	struct componentname *cnp = ap->a_cnp;
329 	struct vattr *vap = ap->a_vap;
330 	enum vtype vt = vap->va_type;
331 
332 	if (vt != VBLK && vt != VCHR && vt != VFIFO) {
333 		vput(dvp);
334 		return EINVAL;
335 	}
336 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
337 }
338 
339 int
340 tmpfs_open(void *v)
341 {
342 	struct vop_open_args /* {
343 		struct vnode	*a_vp;
344 		int		a_mode;
345 		kauth_cred_t	a_cred;
346 	} */ *ap = v;
347 	vnode_t *vp = ap->a_vp;
348 	mode_t mode = ap->a_mode;
349 	tmpfs_node_t *node;
350 
351 	KASSERT(VOP_ISLOCKED(vp));
352 
353 	node = VP_TO_TMPFS_NODE(vp);
354 	if (node->tn_links < 1) {
355 		/*
356 		 * The file is still active, but all its names have been
357 		 * removed (e.g. by a "rmdir $(pwd)").  It cannot be opened
358 		 * any more, as it is about to be destroyed.
359 		 */
360 		return ENOENT;
361 	}
362 
363 	/* If the file is marked append-only, deny write requests. */
364 	if ((node->tn_flags & APPEND) != 0 &&
365 	    (mode & (FWRITE | O_APPEND)) == FWRITE) {
366 		return EPERM;
367 	}
368 	return 0;
369 }
370 
371 int
372 tmpfs_close(void *v)
373 {
374 	struct vop_close_args /* {
375 		struct vnode	*a_vp;
376 		int		a_fflag;
377 		kauth_cred_t	a_cred;
378 	} */ *ap = v;
379 	vnode_t *vp = ap->a_vp;
380 
381 	KASSERT(VOP_ISLOCKED(vp));
382 
383 	tmpfs_update(vp, NULL, NULL, NULL, UPDATE_CLOSE);
384 	return 0;
385 }
386 
387 static int
388 tmpfs_check_possible(vnode_t *vp, tmpfs_node_t *node, mode_t mode)
389 {
390 	const bool writing = (mode & VWRITE) != 0;
391 
392 	switch (vp->v_type) {
393 	case VDIR:
394 	case VLNK:
395 	case VREG:
396 		if (writing && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
397 			return EROFS;
398 		}
399 		break;
400 	case VBLK:
401 	case VCHR:
402 	case VSOCK:
403 	case VFIFO:
404 		break;
405 	default:
406 		return EINVAL;
407 	}
408 	return (writing && (node->tn_flags & IMMUTABLE) != 0) ? EPERM : 0;
409 }
410 
411 static int
412 tmpfs_check_permitted(vnode_t *vp, tmpfs_node_t *node, mode_t mode,
413     kauth_cred_t cred)
414 {
415 
416 	return genfs_can_access(vp->v_type, node->tn_mode, node->tn_uid,
417 	    node->tn_gid, mode, cred);
418 }
419 
420 int
421 tmpfs_access(void *v)
422 {
423 	struct vop_access_args /* {
424 		struct vnode	*a_vp;
425 		int		a_mode;
426 		kauth_cred_t	a_cred;
427 	} */ *ap = v;
428 	vnode_t *vp = ap->a_vp;
429 	mode_t mode = ap->a_mode;
430 	kauth_cred_t cred = ap->a_cred;
431 	tmpfs_node_t *node;
432 	int error;
433 
434 	KASSERT(VOP_ISLOCKED(vp));
435 
436 	node = VP_TO_TMPFS_NODE(vp);
437 	error = tmpfs_check_possible(vp, node, mode);
438 	if (error) {
439 		return error;
440 	}
441 	return kauth_authorize_vnode(cred, kauth_mode_to_action(mode), vp,
442 	    NULL, tmpfs_check_permitted(vp, node, mode, cred));
443 }
444 
445 int
446 tmpfs_getattr(void *v)
447 {
448 	struct vop_getattr_args /* {
449 		struct vnode	*a_vp;
450 		struct vattr	*a_vap;
451 		kauth_cred_t	a_cred;
452 	} */ *ap = v;
453 	vnode_t *vp = ap->a_vp;
454 	struct vattr *vap = ap->a_vap;
455 	tmpfs_node_t *node = VP_TO_TMPFS_NODE(vp);
456 
457 	vattr_null(vap);
458 
459 	tmpfs_update(vp, NULL, NULL, NULL, 0);
460 
461 	vap->va_type = vp->v_type;
462 	vap->va_mode = node->tn_mode;
463 	vap->va_nlink = node->tn_links;
464 	vap->va_uid = node->tn_uid;
465 	vap->va_gid = node->tn_gid;
466 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
467 	vap->va_fileid = node->tn_id;
468 	vap->va_size = node->tn_size;
469 	vap->va_blocksize = PAGE_SIZE;
470 	vap->va_atime = node->tn_atime;
471 	vap->va_mtime = node->tn_mtime;
472 	vap->va_ctime = node->tn_ctime;
473 	vap->va_birthtime = node->tn_birthtime;
474 	vap->va_gen = TMPFS_NODE_GEN(node);
475 	vap->va_flags = node->tn_flags;
476 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
477 	    node->tn_spec.tn_dev.tn_rdev : VNOVAL;
478 	vap->va_bytes = round_page(node->tn_size);
479 	vap->va_filerev = VNOVAL;
480 	vap->va_vaflags = 0;
481 	vap->va_spare = VNOVAL; /* XXX */
482 
483 	return 0;
484 }
485 
486 #define GOODTIME(tv)	((tv)->tv_sec != VNOVAL || (tv)->tv_nsec != VNOVAL)
487 /* XXX Should this operation be atomic?  I think it should, but code in
488  * XXX other places (e.g., ufs) doesn't seem to be... */
489 int
490 tmpfs_setattr(void *v)
491 {
492 	struct vop_setattr_args /* {
493 		struct vnode	*a_vp;
494 		struct vattr	*a_vap;
495 		kauth_cred_t	a_cred;
496 	} */ *ap = v;
497 	vnode_t *vp = ap->a_vp;
498 	struct vattr *vap = ap->a_vap;
499 	kauth_cred_t cred = ap->a_cred;
500 	lwp_t *l = curlwp;
501 	int error = 0;
502 
503 	KASSERT(VOP_ISLOCKED(vp));
504 
505 	/* Abort if any unsettable attribute is given. */
506 	if (vap->va_type != VNON || vap->va_nlink != VNOVAL ||
507 	    vap->va_fsid != VNOVAL || vap->va_fileid != VNOVAL ||
508 	    vap->va_blocksize != VNOVAL || GOODTIME(&vap->va_ctime) ||
509 	    vap->va_gen != VNOVAL || vap->va_rdev != VNOVAL ||
510 	    vap->va_bytes != VNOVAL) {
511 		return EINVAL;
512 	}
513 	if (error == 0 && (vap->va_flags != VNOVAL))
514 		error = tmpfs_chflags(vp, vap->va_flags, cred, l);
515 
516 	if (error == 0 && (vap->va_size != VNOVAL))
517 		error = tmpfs_chsize(vp, vap->va_size, cred, l);
518 
519 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
520 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
521 
522 	if (error == 0 && (vap->va_mode != VNOVAL))
523 		error = tmpfs_chmod(vp, vap->va_mode, cred, l);
524 
525 	if (error == 0 && (GOODTIME(&vap->va_atime) || GOODTIME(&vap->va_mtime)
526 	    || GOODTIME(&vap->va_birthtime))) {
527 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
528 		    &vap->va_birthtime, vap->va_vaflags, cred, l);
529 		if (error == 0)
530 			return 0;
531 	}
532 	tmpfs_update(vp, NULL, NULL, NULL, 0);
533 	return error;
534 }
535 
536 int
537 tmpfs_read(void *v)
538 {
539 	struct vop_read_args /* {
540 		struct vnode *a_vp;
541 		struct uio *a_uio;
542 		int a_ioflag;
543 		kauth_cred_t a_cred;
544 	} */ *ap = v;
545 	vnode_t *vp = ap->a_vp;
546 	struct uio *uio = ap->a_uio;
547 	const int ioflag = ap->a_ioflag;
548 	tmpfs_node_t *node;
549 	struct uvm_object *uobj;
550 	int error;
551 
552 	KASSERT(VOP_ISLOCKED(vp));
553 
554 	if (vp->v_type != VREG) {
555 		return EISDIR;
556 	}
557 	if (uio->uio_offset < 0) {
558 		return EINVAL;
559 	}
560 
561 	node = VP_TO_TMPFS_NODE(vp);
562 	node->tn_status |= TMPFS_NODE_ACCESSED;
563 	uobj = node->tn_spec.tn_reg.tn_aobj;
564 	error = 0;
565 
566 	while (error == 0 && uio->uio_resid > 0) {
567 		vsize_t len;
568 
569 		if (node->tn_size <= uio->uio_offset) {
570 			break;
571 		}
572 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
573 		if (len == 0) {
574 			break;
575 		}
576 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
577 		    UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
578 	}
579 	return error;
580 }
581 
582 int
583 tmpfs_write(void *v)
584 {
585 	struct vop_write_args /* {
586 		struct vnode	*a_vp;
587 		struct uio	*a_uio;
588 		int		a_ioflag;
589 		kauth_cred_t	a_cred;
590 	} */ *ap = v;
591 	vnode_t *vp = ap->a_vp;
592 	struct uio *uio = ap->a_uio;
593 	const int ioflag = ap->a_ioflag;
594 	tmpfs_node_t *node;
595 	struct uvm_object *uobj;
596 	off_t oldsize;
597 	bool extended;
598 	int error;
599 
600 	KASSERT(VOP_ISLOCKED(vp));
601 
602 	node = VP_TO_TMPFS_NODE(vp);
603 	oldsize = node->tn_size;
604 
605 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
606 		error = EINVAL;
607 		goto out;
608 	}
609 	if (uio->uio_resid == 0) {
610 		error = 0;
611 		goto out;
612 	}
613 	if (ioflag & IO_APPEND) {
614 		uio->uio_offset = node->tn_size;
615 	}
616 
617 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
618 	if (extended) {
619 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
620 		if (error)
621 			goto out;
622 	}
623 
624 	uobj = node->tn_spec.tn_reg.tn_aobj;
625 	error = 0;
626 	while (error == 0 && uio->uio_resid > 0) {
627 		vsize_t len;
628 
629 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
630 		if (len == 0) {
631 			break;
632 		}
633 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
634 		    UBC_WRITE | UBC_UNMAP_FLAG(vp));
635 	}
636 	if (error) {
637 		(void)tmpfs_reg_resize(vp, oldsize);
638 	}
639 
640 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
641 	    (extended ? TMPFS_NODE_CHANGED : 0);
642 	VN_KNOTE(vp, NOTE_WRITE);
643 out:
644 	if (error) {
645 		KASSERT(oldsize == node->tn_size);
646 	} else {
647 		KASSERT(uio->uio_resid == 0);
648 	}
649 	return error;
650 }
651 
652 int
653 tmpfs_fsync(void *v)
654 {
655 	struct vop_fsync_args /* {
656 		struct vnode *a_vp;
657 		kauth_cred_t a_cred;
658 		int a_flags;
659 		off_t a_offlo;
660 		off_t a_offhi;
661 		struct lwp *a_l;
662 	} */ *ap = v;
663 	vnode_t *vp = ap->a_vp;
664 
665 	/* Nothing to do.  Just update. */
666 	KASSERT(VOP_ISLOCKED(vp));
667 	tmpfs_update(vp, NULL, NULL, NULL, 0);
668 	return 0;
669 }
670 
671 /*
672  * tmpfs_remove: unlink a file.
673  *
674  * => Both directory (dvp) and file (vp) are locked.
675  * => We unlock and drop the reference on both.
676  */
677 int
678 tmpfs_remove(void *v)
679 {
680 	struct vop_remove_args /* {
681 		struct vnode *a_dvp;
682 		struct vnode *a_vp;
683 		struct componentname *a_cnp;
684 	} */ *ap = v;
685 	vnode_t *dvp = ap->a_dvp, *vp = ap->a_vp;
686 	tmpfs_node_t *node;
687 	tmpfs_dirent_t *de;
688 	int error;
689 
690 	KASSERT(VOP_ISLOCKED(dvp));
691 	KASSERT(VOP_ISLOCKED(vp));
692 
693 	if (vp->v_type == VDIR) {
694 		error = EPERM;
695 		goto out;
696 	}
697 	node = VP_TO_TMPFS_NODE(vp);
698 
699 	/* Files marked as immutable or append-only cannot be deleted. */
700 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
701 		error = EPERM;
702 		goto out;
703 	}
704 
705 	/* Lookup the directory entry (check the cached hint first). */
706 	de = tmpfs_dir_cached(node);
707 	if (de == NULL) {
708 		tmpfs_node_t *dnode = VP_TO_TMPFS_DIR(dvp);
709 		struct componentname *cnp = ap->a_cnp;
710 		de = tmpfs_dir_lookup(dnode, cnp);
711 	}
712 	KASSERT(de && de->td_node == node);
713 
714 	/*
715 	 * Remove the entry from the directory (drops the link count) and
716 	 * destroy it.  Note: the inode referred by it will not be destroyed
717 	 * until the vnode is reclaimed/recycled.
718 	 */
719 	tmpfs_dir_detach(dvp, de);
720 	tmpfs_free_dirent(VFS_TO_TMPFS(vp->v_mount), de);
721 	error = 0;
722 out:
723 	/* Drop the references and unlock the vnodes. */
724 	vput(vp);
725 	if (dvp == vp) {
726 		vrele(dvp);
727 	} else {
728 		vput(dvp);
729 	}
730 	return error;
731 }
732 
733 /*
734  * tmpfs_link: create a hard link.
735  */
736 int
737 tmpfs_link(void *v)
738 {
739 	struct vop_link_args /* {
740 		struct vnode *a_dvp;
741 		struct vnode *a_vp;
742 		struct componentname *a_cnp;
743 	} */ *ap = v;
744 	vnode_t *dvp = ap->a_dvp;
745 	vnode_t *vp = ap->a_vp;
746 	struct componentname *cnp = ap->a_cnp;
747 	tmpfs_node_t *dnode, *node;
748 	tmpfs_dirent_t *de;
749 	int error;
750 
751 	KASSERT(dvp != vp);
752 	KASSERT(VOP_ISLOCKED(dvp));
753 	KASSERT(vp->v_type != VDIR);
754 	KASSERT(dvp->v_mount == vp->v_mount);
755 
756 	dnode = VP_TO_TMPFS_DIR(dvp);
757 	node = VP_TO_TMPFS_NODE(vp);
758 
759 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
760 
761 	/* Check for maximum number of links limit. */
762 	if (node->tn_links == LINK_MAX) {
763 		error = EMLINK;
764 		goto out;
765 	}
766 	KASSERT(node->tn_links < LINK_MAX);
767 
768 	/* We cannot create links of files marked immutable or append-only. */
769 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
770 		error = EPERM;
771 		goto out;
772 	}
773 
774 	/* Allocate a new directory entry to represent the inode. */
775 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount),
776 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
777 	if (error) {
778 		goto out;
779 	}
780 
781 	/*
782 	 * Insert the entry into the directory.
783 	 * It will increase the inode link count.
784 	 */
785 	tmpfs_dir_attach(dvp, de, node);
786 
787 	/* Update the timestamps and trigger the event. */
788 	if (node->tn_vnode) {
789 		VN_KNOTE(node->tn_vnode, NOTE_LINK);
790 	}
791 	node->tn_status |= TMPFS_NODE_CHANGED;
792 	tmpfs_update(vp, NULL, NULL, NULL, 0);
793 	error = 0;
794 out:
795 	VOP_UNLOCK(vp);
796 	vput(dvp);
797 	return error;
798 }
799 
800 /*
801  * tmpfs_parentcheck_p: check if 'lower' is a descendent of 'upper'.
802  *
803  * => Returns 'true' if parent, and 'false' otherwise.
804  */
805 static inline bool
806 tmpfs_parentcheck_p(tmpfs_node_t *lower, tmpfs_node_t *upper)
807 {
808 	tmpfs_node_t *un = lower;
809 
810 	while (un != un->tn_spec.tn_dir.tn_parent) {
811 		KASSERT(un->tn_type == VDIR);
812 		if (un == upper) {
813 			return true;
814 		}
815 		un = un->tn_spec.tn_dir.tn_parent;
816 	}
817 	return false;
818 }
819 
820 /*
821  * tmpfs_rename: rename routine.
822  *
823  * Arguments: fdvp (from-parent vnode), fvp (from-leaf), tdvp (to-parent)
824  * and tvp (to-leaf), if exists (NULL if not).
825  *
826  * => Caller holds a reference on fdvp and fvp, they are unlocked.
827  *    Note: fdvp and fvp can refer to the same object (i.e. when it is root).
828  *
829  * => Both tdvp and tvp are referenced and locked.  It is our responsibility
830  *    to release the references and unlock them (or destroy).
831  */
832 int
833 tmpfs_rename(void *v)
834 {
835 	struct vop_rename_args  /* {
836 		struct vnode		*a_fdvp;
837 		struct vnode		*a_fvp;
838 		struct componentname	*a_fcnp;
839 		struct vnode		*a_tdvp;
840 		struct vnode		*a_tvp;
841 		struct componentname	*a_tcnp;
842 	} */ *ap = v;
843 	vnode_t *fdvp = ap->a_fdvp;
844 	vnode_t *fvp = ap->a_fvp;
845 	struct componentname *fcnp = ap->a_fcnp;
846 	vnode_t *tdvp = ap->a_tdvp;
847 	vnode_t *tvp = ap->a_tvp;
848 	struct componentname *tcnp = ap->a_tcnp;
849 	tmpfs_node_t *fdnode, *fnode, *tnode, *tdnode;
850 	tmpfs_dirent_t *de;
851 	tmpfs_mount_t *tmp;
852 	size_t namelen;
853 	char *newname;
854 	int error;
855 
856 	KASSERT(VOP_ISLOCKED(tdvp));
857 	KASSERT(tvp == NULL || VOP_ISLOCKED(tvp) == LK_EXCLUSIVE);
858 	KASSERT((fcnp->cn_flags & ISDOTDOT) == 0);
859 	KASSERT((tcnp->cn_flags & ISDOTDOT) == 0);
860 
861 	newname = NULL;
862 	namelen = 0;
863 	tmp = NULL;
864 
865 	/* Disallow cross-device renames. */
866 	if (fvp->v_mount != tdvp->v_mount ||
867 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
868 		error = EXDEV;
869 		goto out_unlocked;
870 	}
871 
872 	fnode = VP_TO_TMPFS_NODE(fvp);
873 	fdnode = VP_TO_TMPFS_DIR(fdvp);
874 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
875 	tdnode = VP_TO_TMPFS_DIR(tdvp);
876 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
877 
878 	if (fdvp == tvp) {
879 		error = 0;
880 		goto out_unlocked;
881 	}
882 
883 	/* Allocate memory, if necessary, for a new name. */
884 	namelen = tcnp->cn_namelen;
885 	if (tmpfs_strname_neqlen(fcnp, tcnp)) {
886 		newname = tmpfs_strname_alloc(tmp, namelen);
887 		if (newname == NULL) {
888 			error = ENOSPC;
889 			goto out_unlocked;
890 		}
891 	}
892 
893 	/* XXX: Lock order violation! */
894 	if (fdvp != tdvp) {
895 		vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
896 	}
897 	if (fvp != tvp) {
898 		vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
899 	}
900 
901 	/* If the inode we were renaming has scarpered, just give up. */
902 	de = tmpfs_dir_lookup(fdnode, fcnp);
903 	if (de == NULL || de->td_node != fnode) {
904 		error = ENOENT;
905 		goto out;
906 	}
907 
908 	/*
909 	 * If source and target is the same vnode - it is either invalid
910 	 * rename of a directory, or a hard link.  Remove the source link,
911 	 * if the later.
912 	 */
913 	if (fvp == tvp) {
914 		if (fvp->v_type == VDIR) {
915 			error = EINVAL;
916 			goto out;
917 		}
918 		/*
919 		 * Detach and free the directory entry.  Drops the link
920 		 * count on the inode.
921 		 */
922 		KASSERT(fnode == tnode);
923 		tmpfs_dir_detach(fdvp, de);
924 		tmpfs_free_dirent(tmp, de);
925 		goto out_ok;
926 	}
927 
928 	/* If replacing an existing entry, ensure we can do the operation. */
929 	if (tvp != NULL) {
930 		KASSERT(tnode != NULL);
931 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
932 			if (tnode->tn_size > 0) {
933 				error = ENOTEMPTY;
934 				goto out;
935 			}
936 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
937 			error = ENOTDIR;
938 			goto out;
939 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
940 			error = EISDIR;
941 			goto out;
942 		} else {
943 			KASSERT(fnode->tn_type != VDIR);
944 			KASSERT(tnode->tn_type != VDIR);
945 		}
946 	}
947 
948 	/* Are we moving the inode to a different directory? */
949 	if (fdvp != tdvp) {
950 		/*
951 		 * If we are moving a directory - ensure that it is not
952 		 * parent of a target directory.  Otherwise, it would
953 		 * result in stale nodes.
954 		 */
955 		if (fnode->tn_type == VDIR &&
956 		    tmpfs_parentcheck_p(tdnode, fnode)) {
957 			error = EINVAL;
958 			goto out;
959 		}
960 
961 		/*
962 		 * Perform the move: detach from the source directory and
963 		 * attach into the target directory.
964 		 */
965 		tmpfs_dir_detach(fdvp, de);
966 		tmpfs_dir_attach(tdvp, de, fnode);
967 
968 	} else if (tvp == NULL) {
969 		/* Trigger the event, if not overwriting. */
970 		VN_KNOTE(tdvp, NOTE_WRITE);
971 	}
972 
973 	/* Are we overwriting the entry? */
974 	if (tvp != NULL) {
975 		tmpfs_dirent_t *tde;
976 
977 		tde = tmpfs_dir_cached(tnode);
978 		if (tde == NULL) {
979 			tde = tmpfs_dir_lookup(tdnode, tcnp);
980 		}
981 		KASSERT(tde && tde->td_node == tnode);
982 		KASSERT(tnode->tn_type == fnode->tn_type);
983 
984 		/*
985 		 * Remove and destroy the directory entry on the target
986 		 * directory, since we overwrite it.
987 		 */
988 		tmpfs_dir_detach(tdvp, tde);
989 		tmpfs_free_dirent(tmp, tde);
990 	}
991 
992 	/* If the name has changed, update directory entry. */
993 	if (newname != NULL) {
994 		KASSERT(tcnp->cn_namelen < MAXNAMLEN);
995 
996 		tmpfs_strname_free(tmp, de->td_name, de->td_namelen);
997 		de->td_namelen = (uint16_t)namelen;
998 		memcpy(newname, tcnp->cn_nameptr, namelen);
999 		de->td_name = newname;
1000 		newname = NULL;
1001 
1002 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1003 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1004 	}
1005 out_ok:
1006 	/* Trigger the rename event. */
1007 	VN_KNOTE(fvp, NOTE_RENAME);
1008 	error = 0;
1009 out:
1010 	if (fdvp != tdvp) {
1011 		VOP_UNLOCK(fdvp);
1012 	}
1013 	if (fvp != tvp) {
1014 		VOP_UNLOCK(fvp);
1015 	}
1016 out_unlocked:
1017 	/* Release target nodes. */
1018 	if (tdvp == tvp) {
1019 		vrele(tdvp);
1020 	} else {
1021 		vput(tdvp);
1022 	}
1023 	if (tvp) {
1024 		vput(tvp);
1025 	}
1026 
1027 	/* Release source nodes. */
1028 	vrele(fdvp);
1029 	vrele(fvp);
1030 
1031 	if (newname != NULL) {
1032 		tmpfs_strname_free(tmp, newname, namelen);
1033 	}
1034 	return error;
1035 }
1036 
1037 int
1038 tmpfs_mkdir(void *v)
1039 {
1040 	struct vop_mkdir_args /* {
1041 		struct vnode		*a_dvp;
1042 		struct vnode		**a_vpp;
1043 		struct componentname	*a_cnp;
1044 		struct vattr		*a_vap;
1045 	} */ *ap = v;
1046 	vnode_t *dvp = ap->a_dvp;
1047 	vnode_t **vpp = ap->a_vpp;
1048 	struct componentname *cnp = ap->a_cnp;
1049 	struct vattr *vap = ap->a_vap;
1050 
1051 	KASSERT(vap->va_type == VDIR);
1052 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1053 }
1054 
1055 int
1056 tmpfs_rmdir(void *v)
1057 {
1058 	struct vop_rmdir_args /* {
1059 		struct vnode		*a_dvp;
1060 		struct vnode		*a_vp;
1061 		struct componentname	*a_cnp;
1062 	} */ *ap = v;
1063 	vnode_t *dvp = ap->a_dvp;
1064 	vnode_t *vp = ap->a_vp;
1065 	tmpfs_mount_t *tmp = VFS_TO_TMPFS(dvp->v_mount);
1066 	tmpfs_node_t *dnode = VP_TO_TMPFS_DIR(dvp);
1067 	tmpfs_node_t *node = VP_TO_TMPFS_DIR(vp);
1068 	tmpfs_dirent_t *de;
1069 	int error = 0;
1070 
1071 	KASSERT(VOP_ISLOCKED(dvp));
1072 	KASSERT(VOP_ISLOCKED(vp));
1073 	KASSERT(node->tn_spec.tn_dir.tn_parent == dnode);
1074 
1075 	/*
1076 	 * Directories with more than two entries ('.' and '..') cannot
1077 	 * be removed.
1078 	 */
1079 	if (node->tn_size > 0) {
1080 		error = ENOTEMPTY;
1081 		goto out;
1082 	}
1083 
1084 	/* Lookup the directory entry (check the cached hint first). */
1085 	de = tmpfs_dir_cached(node);
1086 	if (de == NULL) {
1087 		struct componentname *cnp = ap->a_cnp;
1088 		de = tmpfs_dir_lookup(dnode, cnp);
1089 	}
1090 	KASSERT(de && de->td_node == node);
1091 
1092 	/* Check flags to see if we are allowed to remove the directory. */
1093 	if (dnode->tn_flags & APPEND || node->tn_flags & (IMMUTABLE | APPEND)) {
1094 		error = EPERM;
1095 		goto out;
1096 	}
1097 
1098 	/* Decrement the link count for the virtual '.' entry. */
1099 	node->tn_links--;
1100 	node->tn_status |= TMPFS_NODE_STATUSALL;
1101 
1102 	/* Detach the directory entry from the directory. */
1103 	tmpfs_dir_detach(dvp, de);
1104 
1105 	/* Purge the cache for parent. */
1106 	cache_purge(dvp);
1107 
1108 	/*
1109 	 * Destroy the directory entry.  Note: the inode referred by it
1110 	 * will not be destroyed until the vnode is reclaimed.
1111 	 */
1112 	tmpfs_free_dirent(tmp, de);
1113 	KASSERT(node->tn_links == 0);
1114 out:
1115 	/* Release the nodes. */
1116 	vput(dvp);
1117 	vput(vp);
1118 	return error;
1119 }
1120 
1121 int
1122 tmpfs_symlink(void *v)
1123 {
1124 	struct vop_symlink_args /* {
1125 		struct vnode		*a_dvp;
1126 		struct vnode		**a_vpp;
1127 		struct componentname	*a_cnp;
1128 		struct vattr		*a_vap;
1129 		char			*a_target;
1130 	} */ *ap = v;
1131 	vnode_t *dvp = ap->a_dvp;
1132 	vnode_t **vpp = ap->a_vpp;
1133 	struct componentname *cnp = ap->a_cnp;
1134 	struct vattr *vap = ap->a_vap;
1135 	char *target = ap->a_target;
1136 
1137 	KASSERT(vap->va_type == VLNK);
1138 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1139 }
1140 
1141 int
1142 tmpfs_readdir(void *v)
1143 {
1144 	struct vop_readdir_args /* {
1145 		struct vnode	*a_vp;
1146 		struct uio	*a_uio;
1147 		kauth_cred_t	a_cred;
1148 		int		*a_eofflag;
1149 		off_t		**a_cookies;
1150 		int		*ncookies;
1151 	} */ *ap = v;
1152 	vnode_t *vp = ap->a_vp;
1153 	struct uio *uio = ap->a_uio;
1154 	int *eofflag = ap->a_eofflag;
1155 	off_t **cookies = ap->a_cookies;
1156 	int *ncookies = ap->a_ncookies;
1157 	off_t startoff, cnt;
1158 	tmpfs_node_t *node;
1159 	int error;
1160 
1161 	KASSERT(VOP_ISLOCKED(vp));
1162 
1163 	/* This operation only makes sense on directory nodes. */
1164 	if (vp->v_type != VDIR) {
1165 		return ENOTDIR;
1166 	}
1167 	node = VP_TO_TMPFS_DIR(vp);
1168 	startoff = uio->uio_offset;
1169 	cnt = 0;
1170 
1171 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1172 		error = tmpfs_dir_getdotdent(node, uio);
1173 		if (error != 0) {
1174 			if (error == -1)
1175 				error = 0;
1176 			goto out;
1177 		}
1178 		cnt++;
1179 	}
1180 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1181 		error = tmpfs_dir_getdotdotdent(node, uio);
1182 		if (error != 0) {
1183 			if (error == -1)
1184 				error = 0;
1185 			goto out;
1186 		}
1187 		cnt++;
1188 	}
1189 	error = tmpfs_dir_getdents(node, uio, &cnt);
1190 	if (error == -1) {
1191 		error = 0;
1192 	}
1193 	KASSERT(error >= 0);
1194 out:
1195 	if (eofflag != NULL) {
1196 		*eofflag = (!error && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1197 	}
1198 	if (error || cookies == NULL || ncookies == NULL) {
1199 		return error;
1200 	}
1201 
1202 	/* Update NFS-related variables, if any. */
1203 	off_t i, off = startoff;
1204 	tmpfs_dirent_t *de = NULL;
1205 
1206 	*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1207 	*ncookies = cnt;
1208 
1209 	for (i = 0; i < cnt; i++) {
1210 		KASSERT(off != TMPFS_DIRCOOKIE_EOF);
1211 		if (off != TMPFS_DIRCOOKIE_DOT) {
1212 			if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1213 				de = TAILQ_FIRST(&node->tn_spec.tn_dir.tn_dir);
1214 			} else if (de != NULL) {
1215 				de = TAILQ_NEXT(de, td_entries);
1216 			} else {
1217 				de = tmpfs_dir_lookupbycookie(node, off);
1218 				KASSERT(de != NULL);
1219 				de = TAILQ_NEXT(de, td_entries);
1220 			}
1221 			if (de == NULL) {
1222 				off = TMPFS_DIRCOOKIE_EOF;
1223 			} else {
1224 				off = tmpfs_dircookie(de);
1225 			}
1226 		} else {
1227 			off = TMPFS_DIRCOOKIE_DOTDOT;
1228 		}
1229 		(*cookies)[i] = off;
1230 	}
1231 	KASSERT(uio->uio_offset == off);
1232 	return error;
1233 }
1234 
1235 int
1236 tmpfs_readlink(void *v)
1237 {
1238 	struct vop_readlink_args /* {
1239 		struct vnode	*a_vp;
1240 		struct uio	*a_uio;
1241 		kauth_cred_t	a_cred;
1242 	} */ *ap = v;
1243 	vnode_t *vp = ap->a_vp;
1244 	struct uio *uio = ap->a_uio;
1245 	tmpfs_node_t *node;
1246 	int error;
1247 
1248 	KASSERT(VOP_ISLOCKED(vp));
1249 	KASSERT(uio->uio_offset == 0);
1250 	KASSERT(vp->v_type == VLNK);
1251 
1252 	node = VP_TO_TMPFS_NODE(vp);
1253 	error = uiomove(node->tn_spec.tn_lnk.tn_link,
1254 	    MIN(node->tn_size, uio->uio_resid), uio);
1255 	node->tn_status |= TMPFS_NODE_ACCESSED;
1256 
1257 	return error;
1258 }
1259 
1260 int
1261 tmpfs_inactive(void *v)
1262 {
1263 	struct vop_inactive_args /* {
1264 		struct vnode *a_vp;
1265 		bool *a_recycle;
1266 	} */ *ap = v;
1267 	vnode_t *vp = ap->a_vp;
1268 	tmpfs_node_t *node;
1269 
1270 	KASSERT(VOP_ISLOCKED(vp));
1271 
1272 	node = VP_TO_TMPFS_NODE(vp);
1273 	*ap->a_recycle = (node->tn_links == 0);
1274 	VOP_UNLOCK(vp);
1275 
1276 	return 0;
1277 }
1278 
1279 int
1280 tmpfs_reclaim(void *v)
1281 {
1282 	struct vop_reclaim_args /* {
1283 		struct vnode *a_vp;
1284 	} */ *ap = v;
1285 	vnode_t *vp = ap->a_vp;
1286 	tmpfs_mount_t *tmp = VFS_TO_TMPFS(vp->v_mount);
1287 	tmpfs_node_t *node = VP_TO_TMPFS_NODE(vp);
1288 	bool racing;
1289 
1290 	/* Disassociate inode from vnode. */
1291 	mutex_enter(&node->tn_vlock);
1292 	node->tn_vnode = NULL;
1293 	vp->v_data = NULL;
1294 	/* Check if tmpfs_vnode_get() is racing with us. */
1295 	racing = TMPFS_NODE_RECLAIMING(node);
1296 	mutex_exit(&node->tn_vlock);
1297 
1298 	/*
1299 	 * If inode is not referenced, i.e. no links, then destroy it.
1300 	 * Note: if racing - inode is about to get a new vnode, leave it.
1301 	 */
1302 	if (node->tn_links == 0 && !racing) {
1303 		tmpfs_free_node(tmp, node);
1304 	}
1305 	return 0;
1306 }
1307 
1308 int
1309 tmpfs_pathconf(void *v)
1310 {
1311 	struct vop_pathconf_args /* {
1312 		struct vnode	*a_vp;
1313 		int		a_name;
1314 		register_t	*a_retval;
1315 	} */ *ap = v;
1316 	const int name = ap->a_name;
1317 	register_t *retval = ap->a_retval;
1318 	int error = 0;
1319 
1320 	switch (name) {
1321 	case _PC_LINK_MAX:
1322 		*retval = LINK_MAX;
1323 		break;
1324 	case _PC_NAME_MAX:
1325 		*retval = NAME_MAX;
1326 		break;
1327 	case _PC_PATH_MAX:
1328 		*retval = PATH_MAX;
1329 		break;
1330 	case _PC_PIPE_BUF:
1331 		*retval = PIPE_BUF;
1332 		break;
1333 	case _PC_CHOWN_RESTRICTED:
1334 		*retval = 1;
1335 		break;
1336 	case _PC_NO_TRUNC:
1337 		*retval = 1;
1338 		break;
1339 	case _PC_SYNC_IO:
1340 		*retval = 1;
1341 		break;
1342 	case _PC_FILESIZEBITS:
1343 		*retval = sizeof(off_t) * CHAR_BIT;
1344 		break;
1345 	default:
1346 		error = EINVAL;
1347 	}
1348 	return error;
1349 }
1350 
1351 int
1352 tmpfs_advlock(void *v)
1353 {
1354 	struct vop_advlock_args /* {
1355 		struct vnode	*a_vp;
1356 		void *		a_id;
1357 		int		a_op;
1358 		struct flock	*a_fl;
1359 		int		a_flags;
1360 	} */ *ap = v;
1361 	vnode_t *vp = ap->a_vp;
1362 	tmpfs_node_t *node = VP_TO_TMPFS_NODE(vp);
1363 
1364 	return lf_advlock(v, &node->tn_lockf, node->tn_size);
1365 }
1366 
1367 int
1368 tmpfs_getpages(void *v)
1369 {
1370 	struct vop_getpages_args /* {
1371 		struct vnode *a_vp;
1372 		voff_t a_offset;
1373 		struct vm_page **a_m;
1374 		int *a_count;
1375 		int a_centeridx;
1376 		vm_prot_t a_access_type;
1377 		int a_advice;
1378 		int a_flags;
1379 	} */ * const ap = v;
1380 	vnode_t *vp = ap->a_vp;
1381 	const voff_t offset = ap->a_offset;
1382 	struct vm_page **pgs = ap->a_m;
1383 	const int centeridx = ap->a_centeridx;
1384 	const vm_prot_t access_type = ap->a_access_type;
1385 	const int advice = ap->a_advice;
1386 	const int flags = ap->a_flags;
1387 	int error, npages = *ap->a_count;
1388 	tmpfs_node_t *node;
1389 	struct uvm_object *uobj;
1390 
1391 	KASSERT(vp->v_type == VREG);
1392 	KASSERT(mutex_owned(vp->v_interlock));
1393 
1394 	node = VP_TO_TMPFS_NODE(vp);
1395 	uobj = node->tn_spec.tn_reg.tn_aobj;
1396 
1397 	/*
1398 	 * Currently, PGO_PASTEOF is not supported.
1399 	 */
1400 	if (vp->v_size <= offset + (centeridx << PAGE_SHIFT)) {
1401 		if ((flags & PGO_LOCKED) == 0)
1402 			mutex_exit(vp->v_interlock);
1403 		return EINVAL;
1404 	}
1405 
1406 	if (vp->v_size < offset + (npages << PAGE_SHIFT)) {
1407 		npages = (round_page(vp->v_size) - offset) >> PAGE_SHIFT;
1408 	}
1409 
1410 	if ((flags & PGO_LOCKED) != 0)
1411 		return EBUSY;
1412 
1413 	if ((flags & PGO_NOTIMESTAMP) == 0) {
1414 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1415 			node->tn_status |= TMPFS_NODE_ACCESSED;
1416 
1417 		if ((access_type & VM_PROT_WRITE) != 0)
1418 			node->tn_status |= TMPFS_NODE_MODIFIED;
1419 	}
1420 
1421 	/*
1422 	 * Invoke the pager.
1423 	 *
1424 	 * Clean the array of pages before.  XXX: PR/32166
1425 	 * Note that vnode lock is shared with underlying UVM object.
1426 	 */
1427 	if (pgs) {
1428 		memset(pgs, 0, sizeof(struct vm_pages *) * npages);
1429 	}
1430 	KASSERT(vp->v_interlock == uobj->vmobjlock);
1431 
1432 	error = (*uobj->pgops->pgo_get)(uobj, offset, pgs, &npages, centeridx,
1433 	    access_type, advice, flags | PGO_ALLPAGES);
1434 
1435 #if defined(DEBUG)
1436 	if (!error && pgs) {
1437 		for (int i = 0; i < npages; i++) {
1438 			KASSERT(pgs[i] != NULL);
1439 		}
1440 	}
1441 #endif
1442 	return error;
1443 }
1444 
1445 int
1446 tmpfs_putpages(void *v)
1447 {
1448 	struct vop_putpages_args /* {
1449 		struct vnode *a_vp;
1450 		voff_t a_offlo;
1451 		voff_t a_offhi;
1452 		int a_flags;
1453 	} */ * const ap = v;
1454 	vnode_t *vp = ap->a_vp;
1455 	const voff_t offlo = ap->a_offlo;
1456 	const voff_t offhi = ap->a_offhi;
1457 	const int flags = ap->a_flags;
1458 	tmpfs_node_t *node;
1459 	struct uvm_object *uobj;
1460 	int error;
1461 
1462 	KASSERT(mutex_owned(vp->v_interlock));
1463 
1464 	if (vp->v_type != VREG) {
1465 		mutex_exit(vp->v_interlock);
1466 		return 0;
1467 	}
1468 
1469 	node = VP_TO_TMPFS_NODE(vp);
1470 	uobj = node->tn_spec.tn_reg.tn_aobj;
1471 
1472 	KASSERT(vp->v_interlock == uobj->vmobjlock);
1473 	error = (*uobj->pgops->pgo_put)(uobj, offlo, offhi, flags);
1474 
1475 	/* XXX mtime */
1476 
1477 	return error;
1478 }
1479 
1480 #ifdef TMPFS_WHITEOUT
1481 int
1482 tmpfs_whiteout(void *v)
1483 {
1484 	struct vop_whiteout_args /* {
1485 		struct vnode		*a_dvp;
1486 		struct componentname	*a_cnp;
1487 		int			a_flags;
1488 	} */ *ap = v;
1489 	vnode_t *dvp = ap->a_dvp;
1490 	struct componentname *cnp = ap->a_cnp;
1491 	const int flags = ap->a_flags;
1492 	tmpfs_mount_t *tmp = VFS_TO_TMPFS(dvp->v_mount);
1493 	tmpfs_dirent_t *de;
1494 	int error;
1495 
1496 	switch (flags) {
1497 	case LOOKUP:
1498 		break;
1499 	case CREATE:
1500 		error = tmpfs_alloc_dirent(tmp, cnp->cn_nameptr,
1501 		    cnp->cn_namelen, &de);
1502 		if (error)
1503 			return error;
1504 		tmpfs_dir_attach(dvp, de, TMPFS_NODE_WHITEOUT);
1505 		break;
1506 	case DELETE:
1507 		cnp->cn_flags &= ~DOWHITEOUT; /* when in doubt, cargo cult */
1508 		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), cnp);
1509 		if (de == NULL)
1510 			return ENOENT;
1511 		tmpfs_dir_detach(dvp, de);
1512 		tmpfs_free_dirent(tmp, de);
1513 		break;
1514 	}
1515 	return 0;
1516 }
1517 #endif
1518 
1519 int
1520 tmpfs_print(void *v)
1521 {
1522 	struct vop_print_args /* {
1523 		struct vnode	*a_vp;
1524 	} */ *ap = v;
1525 	vnode_t *vp = ap->a_vp;
1526 	tmpfs_node_t *node = VP_TO_TMPFS_NODE(vp);
1527 
1528 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n"
1529 	    "\tmode 0%o, owner %d, group %d, size %" PRIdMAX ", status 0x%x",
1530 	    node, node->tn_flags, node->tn_links, node->tn_mode, node->tn_uid,
1531 	    node->tn_gid, (uintmax_t)node->tn_size, node->tn_status);
1532 	if (vp->v_type == VFIFO) {
1533 		VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
1534 	}
1535 	printf("\n");
1536 	return 0;
1537 }
1538