xref: /netbsd-src/sys/fs/tmpfs/tmpfs_vnops.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: tmpfs_vnops.c,v 1.73 2010/07/14 16:03:49 pooka 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 1.73 2010/07/14 16:03:49 pooka 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/proc.h>
47 #include <sys/stat.h>
48 #include <sys/uio.h>
49 #include <sys/unistd.h>
50 #include <sys/vnode.h>
51 #include <sys/lockf.h>
52 #include <sys/kauth.h>
53 
54 #include <uvm/uvm.h>
55 
56 #include <miscfs/fifofs/fifo.h>
57 #include <miscfs/genfs/genfs.h>
58 #include <fs/tmpfs/tmpfs_vnops.h>
59 #include <fs/tmpfs/tmpfs.h>
60 
61 /* --------------------------------------------------------------------- */
62 
63 /*
64  * vnode operations vector used for files stored in a tmpfs file system.
65  */
66 int (**tmpfs_vnodeop_p)(void *);
67 const struct vnodeopv_entry_desc tmpfs_vnodeop_entries[] = {
68 	{ &vop_default_desc,		vn_default_error },
69 	{ &vop_lookup_desc,		tmpfs_lookup },
70 	{ &vop_create_desc,		tmpfs_create },
71 	{ &vop_mknod_desc,		tmpfs_mknod },
72 	{ &vop_open_desc,		tmpfs_open },
73 	{ &vop_close_desc,		tmpfs_close },
74 	{ &vop_access_desc,		tmpfs_access },
75 	{ &vop_getattr_desc,		tmpfs_getattr },
76 	{ &vop_setattr_desc,		tmpfs_setattr },
77 	{ &vop_read_desc,		tmpfs_read },
78 	{ &vop_write_desc,		tmpfs_write },
79 	{ &vop_ioctl_desc,		tmpfs_ioctl },
80 	{ &vop_fcntl_desc,		tmpfs_fcntl },
81 	{ &vop_poll_desc,		tmpfs_poll },
82 	{ &vop_kqfilter_desc,		tmpfs_kqfilter },
83 	{ &vop_revoke_desc,		tmpfs_revoke },
84 	{ &vop_mmap_desc,		tmpfs_mmap },
85 	{ &vop_fsync_desc,		tmpfs_fsync },
86 	{ &vop_seek_desc,		tmpfs_seek },
87 	{ &vop_remove_desc,		tmpfs_remove },
88 	{ &vop_link_desc,		tmpfs_link },
89 	{ &vop_rename_desc,		tmpfs_rename },
90 	{ &vop_mkdir_desc,		tmpfs_mkdir },
91 	{ &vop_rmdir_desc,		tmpfs_rmdir },
92 	{ &vop_symlink_desc,		tmpfs_symlink },
93 	{ &vop_readdir_desc,		tmpfs_readdir },
94 	{ &vop_readlink_desc,		tmpfs_readlink },
95 	{ &vop_abortop_desc,		tmpfs_abortop },
96 	{ &vop_inactive_desc,		tmpfs_inactive },
97 	{ &vop_reclaim_desc,		tmpfs_reclaim },
98 	{ &vop_lock_desc,		tmpfs_lock },
99 	{ &vop_unlock_desc,		tmpfs_unlock },
100 	{ &vop_bmap_desc,		tmpfs_bmap },
101 	{ &vop_strategy_desc,		tmpfs_strategy },
102 	{ &vop_print_desc,		tmpfs_print },
103 	{ &vop_pathconf_desc,		tmpfs_pathconf },
104 	{ &vop_islocked_desc,		tmpfs_islocked },
105 	{ &vop_advlock_desc,		tmpfs_advlock },
106 	{ &vop_bwrite_desc,		tmpfs_bwrite },
107 	{ &vop_getpages_desc,		tmpfs_getpages },
108 	{ &vop_putpages_desc,		tmpfs_putpages },
109 	{ NULL, NULL }
110 };
111 const struct vnodeopv_desc tmpfs_vnodeop_opv_desc =
112 	{ &tmpfs_vnodeop_p, tmpfs_vnodeop_entries };
113 
114 /*
115  * tmpfs_lookup: lookup routine.
116  *
117  * Arguments: dvp (directory being searched), vpp (result),
118  * cnp (component name - path).
119  *
120  * => Caller holds a reference and lock on dvp.
121  * => We return looked-up vnode (vpp) locked, with a reference held.
122  */
123 int
124 tmpfs_lookup(void *v)
125 {
126 	struct vop_lookup_args /* {
127 		struct vnode *a_dvp;
128 		struct vnode **a_vpp;
129 		struct componentname *a_cnp;
130 	} */ *ap = v;
131 	struct vnode *dvp = ap->a_dvp, **vpp = ap->a_vpp;
132 	struct componentname *cnp = ap->a_cnp;
133 	struct tmpfs_dirent *de;
134 	struct tmpfs_node *dnode;
135 	int error;
136 
137 	KASSERT(VOP_ISLOCKED(dvp));
138 
139 	dnode = VP_TO_TMPFS_DIR(dvp);
140 	*vpp = NULL;
141 
142 	/* Check accessibility of requested node as a first step. */
143 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
144 	if (error != 0)
145 		goto out;
146 
147 	/*
148 	 * If requesting the last path component on a read-only file system
149 	 * with a write operation, deny it.
150 	 */
151 	if ((cnp->cn_flags & ISLASTCN) &&
152 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
153 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
154 		error = EROFS;
155 		goto out;
156 	}
157 
158 	/*
159 	 * Avoid doing a linear scan of the directory if the requested
160 	 * directory/name couple is already in the cache.
161 	 */
162 	error = cache_lookup(dvp, vpp, cnp);
163 	if (error >= 0)
164 		goto out;
165 
166 	/* We cannot be requesting the parent directory of the root node. */
167 	KASSERT(IMPLIES(dnode->tn_type == VDIR &&
168 	    dnode->tn_spec.tn_dir.tn_parent == dnode,
169 	    !(cnp->cn_flags & ISDOTDOT)));
170 
171 	if (cnp->cn_flags & ISDOTDOT) {
172 		VOP_UNLOCK(dvp);
173 
174 		/* Allocate a new vnode on the matching entry. */
175 		error = tmpfs_alloc_vp(dvp->v_mount,
176 		    dnode->tn_spec.tn_dir.tn_parent, vpp);
177 
178 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
179 		goto done;
180 
181 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
182 		if ((cnp->cn_flags & ISLASTCN) &&
183 		    (cnp->cn_nameiop == RENAME)) {
184 			error = EISDIR;
185 			goto out;
186 		}
187 		vref(dvp);
188 		*vpp = dvp;
189 		error = 0;
190 		goto done;
191 	}
192 
193 	de = tmpfs_dir_lookup(dnode, cnp);
194 	if (de == NULL) {
195 		/*
196 		 * The entry was not found in the directory.  This is valid
197 		 * if we are creating or renaming an entry and are working
198 		 * on the last component of the path name.
199 		 */
200 		if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == CREATE ||
201 		    cnp->cn_nameiop == RENAME)) {
202 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
203 			if (error) {
204 				goto out;
205 			}
206 			/* Keep the component name for future uses. */
207 			cnp->cn_flags |= SAVENAME;
208 			error = EJUSTRETURN;
209 		} else {
210 			error = ENOENT;
211 		}
212 	} else {
213 		struct tmpfs_node *tnode = de->td_node;
214 
215 		/*
216 		 * If we are not at the last path component and found a
217 		 * non-directory or non-link entry (which may itself be
218 		 * pointing to a directory), raise an error.
219 		 */
220 		if ((tnode->tn_type != VDIR && tnode->tn_type != VLNK) &&
221 		    (cnp->cn_flags & ISLASTCN) == 0) {
222 			error = ENOTDIR;
223 			goto out;
224 		}
225 
226 		/* Check permissions. */
227 		if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == DELETE ||
228 		    cnp->cn_nameiop == RENAME)) {
229 			kauth_action_t action = 0;
230 
231 			/* This is the file-system's decision. */
232 			if ((dnode->tn_mode & S_ISTXT) != 0 &&
233 			    kauth_cred_geteuid(cnp->cn_cred) != dnode->tn_uid &&
234 			    kauth_cred_geteuid(cnp->cn_cred) != tnode->tn_uid)
235 				error = EPERM;
236 			else
237 				error = 0;
238 
239 			/* Only bother if we are not already failing it. */
240 			if (!error) {
241 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
242 			}
243 
244 			if (cnp->cn_nameiop == DELETE) {
245 				action |= KAUTH_VNODE_DELETE;
246 			} else {
247 				KASSERT(cnp->cn_nameiop == RENAME);
248 				action |= KAUTH_VNODE_RENAME;
249 			}
250 			error = kauth_authorize_vnode(cnp->cn_cred,
251 			    action, *vpp, dvp, error);
252 			if (error) {
253 				goto out;
254 			}
255 			cnp->cn_flags |= SAVENAME;
256 		}
257 		/* Allocate a new vnode on the matching entry. */
258 		error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp);
259 	}
260 done:
261 	/*
262 	 * Store the result of this lookup in the cache.  Avoid this if the
263 	 * request was for creation, as it does not improve timings on
264 	 * emprical tests.
265 	 */
266 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
267 	    (cnp->cn_flags & ISDOTDOT) == 0)
268 		cache_enter(dvp, *vpp, cnp);
269 
270 out:
271 	KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
272 	KASSERT(VOP_ISLOCKED(dvp));
273 	return error;
274 }
275 
276 int
277 tmpfs_create(void *v)
278 {
279 	struct vnode *dvp = ((struct vop_create_args *)v)->a_dvp;
280 	struct vnode **vpp = ((struct vop_create_args *)v)->a_vpp;
281 	struct componentname *cnp = ((struct vop_create_args *)v)->a_cnp;
282 	struct vattr *vap = ((struct vop_create_args *)v)->a_vap;
283 
284 	KASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
285 
286 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
287 }
288 /* --------------------------------------------------------------------- */
289 
290 int
291 tmpfs_mknod(void *v)
292 {
293 	struct vnode *dvp = ((struct vop_mknod_args *)v)->a_dvp;
294 	struct vnode **vpp = ((struct vop_mknod_args *)v)->a_vpp;
295 	struct componentname *cnp = ((struct vop_mknod_args *)v)->a_cnp;
296 	struct vattr *vap = ((struct vop_mknod_args *)v)->a_vap;
297 
298 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
299 	    vap->va_type != VFIFO) {
300 		vput(dvp);
301 		return EINVAL;
302 	}
303 
304 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
305 }
306 
307 /* --------------------------------------------------------------------- */
308 
309 int
310 tmpfs_open(void *v)
311 {
312 	struct vnode *vp = ((struct vop_open_args *)v)->a_vp;
313 	int mode = ((struct vop_open_args *)v)->a_mode;
314 
315 	int error;
316 	struct tmpfs_node *node;
317 
318 	KASSERT(VOP_ISLOCKED(vp));
319 
320 	node = VP_TO_TMPFS_NODE(vp);
321 
322 	/* The file is still active but all its names have been removed
323 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
324 	 * it is about to die. */
325 	if (node->tn_links < 1) {
326 		error = ENOENT;
327 		goto out;
328 	}
329 
330 	/* If the file is marked append-only, deny write requests. */
331 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
332 		error = EPERM;
333 	else
334 		error = 0;
335 
336 out:
337 	KASSERT(VOP_ISLOCKED(vp));
338 
339 	return error;
340 }
341 
342 /* --------------------------------------------------------------------- */
343 
344 int
345 tmpfs_close(void *v)
346 {
347 	struct vnode *vp = ((struct vop_close_args *)v)->a_vp;
348 
349 	struct tmpfs_node *node;
350 
351 	KASSERT(VOP_ISLOCKED(vp));
352 
353 	node = VP_TO_TMPFS_NODE(vp);
354 
355 	if (node->tn_links > 0) {
356 		/* Update node times.  No need to do it if the node has
357 		 * been deleted, because it will vanish after we return. */
358 		tmpfs_update(vp, NULL, NULL, NULL, UPDATE_CLOSE);
359 	}
360 
361 	return 0;
362 }
363 
364 /* --------------------------------------------------------------------- */
365 
366 static int
367 tmpfs_check_possible(struct vnode *vp, struct tmpfs_node *node, mode_t mode)
368 {
369 	int error = 0;
370 
371 	switch (vp->v_type) {
372 	case VDIR:
373 		/* FALLTHROUGH */
374 	case VLNK:
375 		/* FALLTHROUGH */
376 	case VREG:
377 		if (mode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
378 			error = EROFS;
379 			goto out;
380 		}
381 		break;
382 
383 	case VBLK:
384 		/* FALLTHROUGH */
385 	case VCHR:
386 		/* FALLTHROUGH */
387 	case VSOCK:
388 		/* FALLTHROUGH */
389 	case VFIFO:
390 		break;
391 
392 	default:
393 		error = EINVAL;
394 		goto out;
395 	}
396 
397 	if (mode & VWRITE && node->tn_flags & IMMUTABLE) {
398 		error = EPERM;
399 		goto out;
400 	}
401 
402  out:
403 	return error;
404 }
405 
406 static int
407 tmpfs_check_permitted(struct vnode *vp, struct tmpfs_node *node, mode_t mode,
408     kauth_cred_t cred)
409 {
410 
411 	return genfs_can_access(vp->v_type, node->tn_mode, node->tn_uid,
412 	    node->tn_gid, mode, cred);
413 }
414 
415 int
416 tmpfs_access(void *v)
417 {
418 	struct vnode *vp = ((struct vop_access_args *)v)->a_vp;
419 	int mode = ((struct vop_access_args *)v)->a_mode;
420 	kauth_cred_t cred = ((struct vop_access_args *)v)->a_cred;
421 
422 	int error;
423 	struct tmpfs_node *node;
424 
425 	KASSERT(VOP_ISLOCKED(vp));
426 
427 	node = VP_TO_TMPFS_NODE(vp);
428 
429 	error = tmpfs_check_possible(vp, node, mode);
430 	if (error)
431 		goto out;
432 
433 	error = tmpfs_check_permitted(vp, node, mode, cred);
434 
435 	error = kauth_authorize_vnode(cred, kauth_mode_to_action(mode), vp,
436 	    NULL, error);
437 
438 out:
439 	KASSERT(VOP_ISLOCKED(vp));
440 
441 	return error;
442 }
443 
444 /* --------------------------------------------------------------------- */
445 
446 int
447 tmpfs_getattr(void *v)
448 {
449 	struct vnode *vp = ((struct vop_getattr_args *)v)->a_vp;
450 	struct vattr *vap = ((struct vop_getattr_args *)v)->a_vap;
451 
452 	struct tmpfs_node *node;
453 
454 	node = VP_TO_TMPFS_NODE(vp);
455 
456 	vattr_null(vap);
457 
458 	tmpfs_itimes(vp, NULL, NULL, NULL);
459 
460 	vap->va_type = vp->v_type;
461 	vap->va_mode = node->tn_mode;
462 	vap->va_nlink = node->tn_links;
463 	vap->va_uid = node->tn_uid;
464 	vap->va_gid = node->tn_gid;
465 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
466 	vap->va_fileid = node->tn_id;
467 	vap->va_size = node->tn_size;
468 	vap->va_blocksize = PAGE_SIZE;
469 	vap->va_atime = node->tn_atime;
470 	vap->va_mtime = node->tn_mtime;
471 	vap->va_ctime = node->tn_ctime;
472 	vap->va_birthtime = node->tn_birthtime;
473 	vap->va_gen = node->tn_gen;
474 	vap->va_flags = node->tn_flags;
475 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
476 		node->tn_spec.tn_dev.tn_rdev : VNOVAL;
477 	vap->va_bytes = round_page(node->tn_size);
478 	vap->va_filerev = VNOVAL;
479 	vap->va_vaflags = 0;
480 	vap->va_spare = VNOVAL; /* XXX */
481 
482 	return 0;
483 }
484 
485 /* --------------------------------------------------------------------- */
486 
487 #define GOODTIME(tv)	((tv)->tv_sec != VNOVAL || (tv)->tv_nsec != VNOVAL)
488 /* XXX Should this operation be atomic?  I think it should, but code in
489  * XXX other places (e.g., ufs) doesn't seem to be... */
490 int
491 tmpfs_setattr(void *v)
492 {
493 	struct vnode *vp = ((struct vop_setattr_args *)v)->a_vp;
494 	struct vattr *vap = ((struct vop_setattr_args *)v)->a_vap;
495 	kauth_cred_t cred = ((struct vop_setattr_args *)v)->a_cred;
496 	struct lwp *l = curlwp;
497 
498 	int error;
499 
500 	KASSERT(VOP_ISLOCKED(vp));
501 
502 	error = 0;
503 
504 	/* Abort if any unsettable attribute is given. */
505 	if (vap->va_type != VNON ||
506 	    vap->va_nlink != VNOVAL ||
507 	    vap->va_fsid != VNOVAL ||
508 	    vap->va_fileid != VNOVAL ||
509 	    vap->va_blocksize != VNOVAL ||
510 	    GOODTIME(&vap->va_ctime) ||
511 	    vap->va_gen != VNOVAL ||
512 	    vap->va_rdev != VNOVAL ||
513 	    vap->va_bytes != VNOVAL)
514 		error = EINVAL;
515 
516 	if (error == 0 && (vap->va_flags != VNOVAL))
517 		error = tmpfs_chflags(vp, vap->va_flags, cred, l);
518 
519 	if (error == 0 && (vap->va_size != VNOVAL))
520 		error = tmpfs_chsize(vp, vap->va_size, cred, l);
521 
522 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
523 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
524 
525 	if (error == 0 && (vap->va_mode != VNOVAL))
526 		error = tmpfs_chmod(vp, vap->va_mode, cred, l);
527 
528 	if (error == 0 && (GOODTIME(&vap->va_atime) || GOODTIME(&vap->va_mtime)
529 	    || GOODTIME(&vap->va_birthtime)))
530 		if ((error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
531 		    &vap->va_birthtime, vap->va_vaflags, cred, l)) == 0)
532 			return 0;
533 
534 	/* Update the node times.  We give preference to the error codes
535 	 * generated by this function rather than the ones that may arise
536 	 * from tmpfs_update. */
537 	tmpfs_update(vp, NULL, NULL, NULL, 0);
538 
539 	KASSERT(VOP_ISLOCKED(vp));
540 
541 	return error;
542 }
543 
544 /* --------------------------------------------------------------------- */
545 
546 int
547 tmpfs_read(void *v)
548 {
549 	struct vnode *vp = ((struct vop_read_args *)v)->a_vp;
550 	struct uio *uio = ((struct vop_read_args *)v)->a_uio;
551 	int ioflag = ((struct vop_read_args *)v)->a_ioflag;
552 
553 	int error;
554 	struct tmpfs_node *node;
555 	struct uvm_object *uobj;
556 
557 	KASSERT(VOP_ISLOCKED(vp));
558 
559 	node = VP_TO_TMPFS_NODE(vp);
560 
561 	if (vp->v_type != VREG) {
562 		error = EISDIR;
563 		goto out;
564 	}
565 
566 	if (uio->uio_offset < 0) {
567 		error = EINVAL;
568 		goto out;
569 	}
570 
571 	node->tn_status |= TMPFS_NODE_ACCESSED;
572 
573 	uobj = node->tn_spec.tn_reg.tn_aobj;
574 	error = 0;
575 	while (error == 0 && uio->uio_resid > 0) {
576 		vsize_t len;
577 
578 		if (node->tn_size <= uio->uio_offset)
579 			break;
580 
581 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
582 		if (len == 0)
583 			break;
584 
585 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
586 		    UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
587 	}
588 
589 out:
590 	KASSERT(VOP_ISLOCKED(vp));
591 
592 	return error;
593 }
594 
595 /* --------------------------------------------------------------------- */
596 
597 int
598 tmpfs_write(void *v)
599 {
600 	struct vnode *vp = ((struct vop_write_args *)v)->a_vp;
601 	struct uio *uio = ((struct vop_write_args *)v)->a_uio;
602 	int ioflag = ((struct vop_write_args *)v)->a_ioflag;
603 
604 	bool extended;
605 	int error;
606 	off_t oldsize;
607 	struct tmpfs_node *node;
608 	struct uvm_object *uobj;
609 
610 	KASSERT(VOP_ISLOCKED(vp));
611 
612 	node = VP_TO_TMPFS_NODE(vp);
613 	oldsize = node->tn_size;
614 
615 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
616 		error = EINVAL;
617 		goto out;
618 	}
619 
620 	if (uio->uio_resid == 0) {
621 		error = 0;
622 		goto out;
623 	}
624 
625 	if (ioflag & IO_APPEND)
626 		uio->uio_offset = node->tn_size;
627 
628 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
629 	if (extended) {
630 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
631 		if (error != 0)
632 			goto out;
633 	}
634 
635 	uobj = node->tn_spec.tn_reg.tn_aobj;
636 	error = 0;
637 	while (error == 0 && uio->uio_resid > 0) {
638 		vsize_t len;
639 
640 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
641 		if (len == 0)
642 			break;
643 
644 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
645 		    UBC_WRITE | UBC_UNMAP_FLAG(vp));
646 	}
647 
648 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
649 	    (extended ? TMPFS_NODE_CHANGED : 0);
650 
651 	if (error != 0)
652 		(void)tmpfs_reg_resize(vp, oldsize);
653 
654 	VN_KNOTE(vp, NOTE_WRITE);
655 
656 out:
657 	KASSERT(VOP_ISLOCKED(vp));
658 	KASSERT(IMPLIES(error == 0, uio->uio_resid == 0));
659 	KASSERT(IMPLIES(error != 0, oldsize == node->tn_size));
660 
661 	return error;
662 }
663 
664 /* --------------------------------------------------------------------- */
665 
666 int
667 tmpfs_fsync(void *v)
668 {
669 	struct vnode *vp = ((struct vop_fsync_args *)v)->a_vp;
670 
671 	KASSERT(VOP_ISLOCKED(vp));
672 
673 	tmpfs_update(vp, NULL, NULL, NULL, 0);
674 
675 	return 0;
676 }
677 
678 /* --------------------------------------------------------------------- */
679 
680 int
681 tmpfs_remove(void *v)
682 {
683 	struct vnode *dvp = ((struct vop_remove_args *)v)->a_dvp;
684 	struct vnode *vp = ((struct vop_remove_args *)v)->a_vp;
685 	struct componentname *cnp = (((struct vop_remove_args *)v)->a_cnp);
686 
687 	int error;
688 	struct tmpfs_dirent *de;
689 	struct tmpfs_mount *tmp;
690 	struct tmpfs_node *dnode;
691 	struct tmpfs_node *node;
692 
693 	KASSERT(VOP_ISLOCKED(dvp));
694 	KASSERT(VOP_ISLOCKED(vp));
695 
696 	if (vp->v_type == VDIR) {
697 		error = EPERM;
698 		goto out;
699 	}
700 
701 	dnode = VP_TO_TMPFS_DIR(dvp);
702 	node = VP_TO_TMPFS_NODE(vp);
703 	tmp = VFS_TO_TMPFS(vp->v_mount);
704 	de = tmpfs_dir_lookup(dnode, cnp);
705 	KASSERT(de);
706 	KASSERT(de->td_node == node);
707 
708 	/* Files marked as immutable or append-only cannot be deleted. */
709 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
710 		error = EPERM;
711 		goto out;
712 	}
713 
714 	/* Remove the entry from the directory; as it is a file, we do not
715 	 * have to change the number of hard links of the directory. */
716 	tmpfs_dir_detach(dvp, de);
717 
718 	/* Free the directory entry we just deleted.  Note that the node
719 	 * referred by it will not be removed until the vnode is really
720 	 * reclaimed. */
721 	tmpfs_free_dirent(tmp, de, true);
722 
723 	error = 0;
724 
725 out:
726 	vput(vp);
727 	if (dvp == vp)
728 		vrele(dvp);
729 	else
730 		vput(dvp);
731 	if (cnp->cn_flags & HASBUF) {
732 		PNBUF_PUT(cnp->cn_pnbuf);
733 		cnp->cn_flags &= ~HASBUF;
734 	}
735 
736 	return error;
737 }
738 
739 /* --------------------------------------------------------------------- */
740 
741 int
742 tmpfs_link(void *v)
743 {
744 	struct vnode *dvp = ((struct vop_link_args *)v)->a_dvp;
745 	struct vnode *vp = ((struct vop_link_args *)v)->a_vp;
746 	struct componentname *cnp = ((struct vop_link_args *)v)->a_cnp;
747 
748 	int error;
749 	struct tmpfs_dirent *de;
750 	struct tmpfs_node *dnode;
751 	struct tmpfs_node *node;
752 
753 	KASSERT(VOP_ISLOCKED(dvp));
754 	KASSERT(cnp->cn_flags & HASBUF);
755 	KASSERT(dvp != vp); /* XXX When can this be false? */
756 
757 	dnode = VP_TO_TMPFS_DIR(dvp);
758 	node = VP_TO_TMPFS_NODE(vp);
759 
760 	/* Lock vp because we will need to run tmpfs_update over it, which
761 	 * needs the vnode to be locked. */
762 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
763 
764 	/* XXX: Why aren't the following two tests done by the caller? */
765 
766 	/* Hard links of directories are forbidden. */
767 	if (vp->v_type == VDIR) {
768 		error = EPERM;
769 		goto out;
770 	}
771 
772 	/* Cannot create cross-device links. */
773 	if (dvp->v_mount != vp->v_mount) {
774 		error = EXDEV;
775 		goto out;
776 	}
777 
778 	/* Ensure that we do not overflow the maximum number of links imposed
779 	 * by the system. */
780 	KASSERT(node->tn_links <= LINK_MAX);
781 	if (node->tn_links == LINK_MAX) {
782 		error = EMLINK;
783 		goto out;
784 	}
785 
786 	/* We cannot create links of files marked immutable or append-only. */
787 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
788 		error = EPERM;
789 		goto out;
790 	}
791 
792 	/* Allocate a new directory entry to represent the node. */
793 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
794 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
795 	if (error != 0)
796 		goto out;
797 
798 	/* Insert the new directory entry into the appropriate directory. */
799 	tmpfs_dir_attach(dvp, de);
800 
801 	/* vp link count has changed, so update node times. */
802 	node->tn_status |= TMPFS_NODE_CHANGED;
803 	tmpfs_update(vp, NULL, NULL, NULL, 0);
804 
805 	error = 0;
806 
807 out:
808 	VOP_UNLOCK(vp);
809 	PNBUF_PUT(cnp->cn_pnbuf);
810 	vput(dvp);
811 
812 	return error;
813 }
814 
815 /*
816  * tmpfs_rename: rename routine.
817  *
818  * Arguments: fdvp (from-parent vnode), fvp (from-leaf), tdvp (to-parent)
819  * and tvp (to-leaf), if exists (NULL if not).
820  *
821  * => Caller holds a reference on fdvp and fvp, they are unlocked.
822  *    Note: fdvp and fvp can refer to the same object (i.e. when it is root).
823  *
824  * => Both tdvp and tvp are referenced and locked.  It is our responsibility
825  *    to release the references and unlock them (or destroy).
826  */
827 int
828 tmpfs_rename(void *v)
829 {
830 	struct vnode *fdvp = ((struct vop_rename_args *)v)->a_fdvp;
831 	struct vnode *fvp = ((struct vop_rename_args *)v)->a_fvp;
832 	struct componentname *fcnp = ((struct vop_rename_args *)v)->a_fcnp;
833 	struct vnode *tdvp = ((struct vop_rename_args *)v)->a_tdvp;
834 	struct vnode *tvp = ((struct vop_rename_args *)v)->a_tvp;
835 	struct componentname *tcnp = ((struct vop_rename_args *)v)->a_tcnp;
836 
837 	char *newname;
838 	int error;
839 	struct tmpfs_dirent *de, *de2;
840 	struct tmpfs_mount *tmp;
841 	struct tmpfs_node *fdnode;
842 	struct tmpfs_node *fnode;
843 	struct tmpfs_node *tnode;
844 	struct tmpfs_node *tdnode;
845 	size_t namelen;
846 
847 	KASSERT(VOP_ISLOCKED(tdvp));
848 	KASSERT(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
849 	KASSERT(fcnp->cn_flags & HASBUF);
850 	KASSERT(tcnp->cn_flags & HASBUF);
851 
852 	newname = NULL;
853 	namelen = 0;
854 	tmp = NULL;
855 
856 	/* Disallow cross-device renames. */
857 	if (fvp->v_mount != tdvp->v_mount ||
858 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
859 		error = EXDEV;
860 		goto out_unlocked;
861 	}
862 
863 	fnode = VP_TO_TMPFS_NODE(fvp);
864 	fdnode = VP_TO_TMPFS_DIR(fdvp);
865 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
866 	tdnode = VP_TO_TMPFS_DIR(tdvp);
867 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
868 
869 	if (fdvp == tvp) {
870 		error = 0;
871 		goto out_unlocked;
872 	}
873 
874 	/* Allocate memory, if necessary, for a new name. */
875 	namelen = tcnp->cn_namelen;
876 	if (tmpfs_strname_neqlen(fcnp, tcnp)) {
877 		newname = tmpfs_strname_alloc(tmp, namelen);
878 		if (newname == NULL) {
879 			error = ENOSPC;
880 			goto out_unlocked;
881 		}
882 	}
883 
884 	/* If we need to move the directory between entries, lock the
885 	 * source so that we can safely operate on it. */
886 
887 	/* XXX: this is a potential locking order violation! */
888 	if (fdnode != tdnode) {
889 		vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
890 	}
891 
892 	/*
893 	 * If the node we were renaming has scarpered, just give up.
894 	 */
895 	de = tmpfs_dir_lookup(fdnode, fcnp);
896 	if (de == NULL || de->td_node != fnode) {
897 		error = ENOENT;
898 		goto out;
899 	}
900 
901 	/* If source and target is the same vnode, remove the source link. */
902 	if (fvp == tvp) {
903 		/*
904 		 * Detach and free the directory entry.  Drops the link
905 		 * count on the node.
906 		 */
907 		tmpfs_dir_detach(fdvp, de);
908 		tmpfs_free_dirent(VFS_TO_TMPFS(fvp->v_mount), de, true);
909 		VN_KNOTE(fdvp, NOTE_WRITE);
910 		goto out_ok;
911 	}
912 
913 	/* If replacing an existing entry, ensure we can do the operation. */
914 	if (tvp != NULL) {
915 		KASSERT(tnode != NULL);
916 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
917 			if (tnode->tn_size > 0) {
918 				error = ENOTEMPTY;
919 				goto out;
920 			}
921 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
922 			error = ENOTDIR;
923 			goto out;
924 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
925 			error = EISDIR;
926 			goto out;
927 		} else {
928 			KASSERT(fnode->tn_type != VDIR &&
929 			        tnode->tn_type != VDIR);
930 		}
931 	}
932 
933 	/* If the node is being moved to another directory, we have to do
934 	 * the move. */
935 	if (fdnode != tdnode) {
936 		/* In case we are moving a directory, we have to adjust its
937 		 * parent to point to the new parent. */
938 		if (de->td_node->tn_type == VDIR) {
939 			struct tmpfs_node *n;
940 
941 			/* Ensure the target directory is not a child of the
942 			 * directory being moved.  Otherwise, we'd end up
943 			 * with stale nodes. */
944 			n = tdnode;
945 			while (n != n->tn_spec.tn_dir.tn_parent) {
946 				if (n == fnode) {
947 					error = EINVAL;
948 					goto out;
949 				}
950 				n = n->tn_spec.tn_dir.tn_parent;
951 			}
952 
953 			/* Adjust the parent pointer. */
954 			TMPFS_VALIDATE_DIR(fnode);
955 			de->td_node->tn_spec.tn_dir.tn_parent = tdnode;
956 
957 			/* As a result of changing the target of the '..'
958 			 * entry, the link count of the source and target
959 			 * directories has to be adjusted. */
960 			fdnode->tn_links--;
961 			tdnode->tn_links++;
962 		}
963 
964 		/* Do the move: just remove the entry from the source directory
965 		 * and insert it into the target one. */
966 		tmpfs_dir_detach(fdvp, de);
967 		tmpfs_dir_attach(tdvp, de);
968 
969 		/* Notify listeners of fdvp about the change in the directory.
970 		 * We can do it at this point because we aren't touching fdvp
971 		 * any more below. */
972 		VN_KNOTE(fdvp, NOTE_WRITE);
973 	}
974 
975 	/* If we are overwriting an entry, we have to remove the old one
976 	 * from the target directory. */
977 	if (tvp != NULL) {
978 		KASSERT(tnode != NULL);
979 
980 		/* Remove the old entry from the target directory.
981 		 * Note! This relies on tmpfs_dir_attach() putting the new
982 		 * node on the end of the target's node list. */
983 		de2 = tmpfs_dir_lookup(tdnode, tcnp);
984 		KASSERT(de2 != NULL);
985 		KASSERT(de2->td_node == tnode);
986 		tmpfs_dir_detach(tdvp, de2);
987 
988 		/* Free the directory entry we just deleted.  Note that the
989 		 * node referred by it will not be removed until the vnode is
990 		 * really reclaimed. */
991 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de2, true);
992 	}
993 
994 	/* If the name has changed, we need to make it effective by changing
995 	 * it in the directory entry. */
996 	if (newname != NULL) {
997 		KASSERT(tcnp->cn_namelen < MAXNAMLEN);
998 		KASSERT(tcnp->cn_namelen < 0xffff);
999 
1000 		tmpfs_strname_free(tmp, de->td_name, de->td_namelen);
1001 		de->td_namelen = (uint16_t)namelen;
1002 		memcpy(newname, tcnp->cn_nameptr, namelen);
1003 		de->td_name = newname;
1004 		newname = NULL;
1005 
1006 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1007 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1008 	}
1009  out_ok:
1010 	/* Notify listeners of tdvp about the change in the directory (either
1011 	 * because a new entry was added or because one was removed) and
1012 	 * listeners of fvp about the rename. */
1013 	VN_KNOTE(tdvp, NOTE_WRITE);
1014 	VN_KNOTE(fvp, NOTE_RENAME);
1015 
1016 	error = 0;
1017 
1018  out:
1019 	if (fdnode != tdnode)
1020 		VOP_UNLOCK(fdvp);
1021 
1022  out_unlocked:
1023 	/* Release target nodes. */
1024 	if (tdvp == tvp)
1025 		vrele(tdvp);
1026 	else
1027 		vput(tdvp);
1028 	if (tvp != NULL)
1029 		vput(tvp);
1030 
1031 	/* Release source nodes. */
1032 	vrele(fdvp);
1033 	vrele(fvp);
1034 
1035 	if (newname != NULL) {
1036 		tmpfs_strname_free(tmp, newname, namelen);
1037 	}
1038 	return error;
1039 }
1040 
1041 /* --------------------------------------------------------------------- */
1042 
1043 int
1044 tmpfs_mkdir(void *v)
1045 {
1046 	struct vnode *dvp = ((struct vop_mkdir_args *)v)->a_dvp;
1047 	struct vnode **vpp = ((struct vop_mkdir_args *)v)->a_vpp;
1048 	struct componentname *cnp = ((struct vop_mkdir_args *)v)->a_cnp;
1049 	struct vattr *vap = ((struct vop_mkdir_args *)v)->a_vap;
1050 
1051 	KASSERT(vap->va_type == VDIR);
1052 
1053 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1054 }
1055 
1056 /* --------------------------------------------------------------------- */
1057 
1058 int
1059 tmpfs_rmdir(void *v)
1060 {
1061 	struct vnode *dvp = ((struct vop_rmdir_args *)v)->a_dvp;
1062 	struct vnode *vp = ((struct vop_rmdir_args *)v)->a_vp;
1063 	struct componentname *cnp = ((struct vop_rmdir_args *)v)->a_cnp;
1064 
1065 	int error;
1066 	struct tmpfs_dirent *de;
1067 	struct tmpfs_mount *tmp;
1068 	struct tmpfs_node *dnode;
1069 	struct tmpfs_node *node;
1070 
1071 	KASSERT(VOP_ISLOCKED(dvp));
1072 	KASSERT(VOP_ISLOCKED(vp));
1073 
1074 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1075 	dnode = VP_TO_TMPFS_DIR(dvp);
1076 	node = VP_TO_TMPFS_DIR(vp);
1077 	error = 0;
1078 
1079 	/* Directories with more than two entries ('.' and '..') cannot be
1080 	 * removed. */
1081 	if (node->tn_size > 0) {
1082 		error = ENOTEMPTY;
1083 		goto out;
1084 	}
1085 
1086 	/* This invariant holds only if we are not trying to remove "..".
1087 	 * We checked for that above so this is safe now. */
1088 	KASSERT(node->tn_spec.tn_dir.tn_parent == dnode);
1089 
1090 	/* Get the directory entry associated with node (vp). */
1091 	de = tmpfs_dir_lookup(dnode, cnp);
1092 	KASSERT(de);
1093 	KASSERT(de->td_node == node);
1094 
1095 	/* Check flags to see if we are allowed to remove the directory. */
1096 	if (dnode->tn_flags & APPEND || node->tn_flags & (IMMUTABLE | APPEND)) {
1097 		error = EPERM;
1098 		goto out;
1099 	}
1100 
1101 	/* Detach the directory entry from the directory (dnode). */
1102 	tmpfs_dir_detach(dvp, de);
1103 
1104 	node->tn_links--;
1105 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1106 	    TMPFS_NODE_MODIFIED;
1107 	node->tn_spec.tn_dir.tn_parent->tn_links--;
1108 	node->tn_spec.tn_dir.tn_parent->tn_status |= TMPFS_NODE_ACCESSED | \
1109 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1110 
1111 	/* Release the parent. */
1112 	cache_purge(dvp); /* XXX Is this needed? */
1113 
1114 	/* Free the directory entry we just deleted.  Note that the node
1115 	 * referred by it will not be removed until the vnode is really
1116 	 * reclaimed. */
1117 	tmpfs_free_dirent(tmp, de, true);
1118 
1119 	KASSERT(node->tn_links == 0);
1120  out:
1121 	/* Release the nodes. */
1122 	vput(dvp);
1123 	vput(vp);
1124 	PNBUF_PUT(cnp->cn_pnbuf);
1125 
1126 	return error;
1127 }
1128 
1129 /* --------------------------------------------------------------------- */
1130 
1131 int
1132 tmpfs_symlink(void *v)
1133 {
1134 	struct vnode *dvp = ((struct vop_symlink_args *)v)->a_dvp;
1135 	struct vnode **vpp = ((struct vop_symlink_args *)v)->a_vpp;
1136 	struct componentname *cnp = ((struct vop_symlink_args *)v)->a_cnp;
1137 	struct vattr *vap = ((struct vop_symlink_args *)v)->a_vap;
1138 	char *target = ((struct vop_symlink_args *)v)->a_target;
1139 
1140 	KASSERT(vap->va_type == VLNK);
1141 
1142 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1143 }
1144 
1145 /* --------------------------------------------------------------------- */
1146 
1147 int
1148 tmpfs_readdir(void *v)
1149 {
1150 	struct vnode *vp = ((struct vop_readdir_args *)v)->a_vp;
1151 	struct uio *uio = ((struct vop_readdir_args *)v)->a_uio;
1152 	int *eofflag = ((struct vop_readdir_args *)v)->a_eofflag;
1153 	off_t **cookies = ((struct vop_readdir_args *)v)->a_cookies;
1154 	int *ncookies = ((struct vop_readdir_args *)v)->a_ncookies;
1155 
1156 	int error;
1157 	off_t startoff;
1158 	off_t cnt;
1159 	struct tmpfs_node *node;
1160 
1161 	KASSERT(VOP_ISLOCKED(vp));
1162 
1163 	/* This operation only makes sense on directory nodes. */
1164 	if (vp->v_type != VDIR) {
1165 		error = ENOTDIR;
1166 		goto out;
1167 	}
1168 
1169 	node = VP_TO_TMPFS_DIR(vp);
1170 
1171 	startoff = uio->uio_offset;
1172 
1173 	cnt = 0;
1174 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1175 		error = tmpfs_dir_getdotdent(node, uio);
1176 		if (error == -1) {
1177 			error = 0;
1178 			goto outok;
1179 		} else if (error != 0)
1180 			goto outok;
1181 		cnt++;
1182 	}
1183 
1184 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1185 		error = tmpfs_dir_getdotdotdent(node, uio);
1186 		if (error == -1) {
1187 			error = 0;
1188 			goto outok;
1189 		} else if (error != 0)
1190 			goto outok;
1191 		cnt++;
1192 	}
1193 
1194 	error = tmpfs_dir_getdents(node, uio, &cnt);
1195 	if (error == -1)
1196 		error = 0;
1197 	KASSERT(error >= 0);
1198 
1199 outok:
1200 	/* This label assumes that startoff has been
1201 	 * initialized.  If the compiler didn't spit out warnings, we'd
1202 	 * simply make this one be 'out' and drop 'outok'. */
1203 
1204 	if (eofflag != NULL)
1205 		*eofflag =
1206 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1207 
1208 	/* Update NFS-related variables. */
1209 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1210 		off_t i;
1211 		off_t off = startoff;
1212 		struct tmpfs_dirent *de = NULL;
1213 
1214 		*ncookies = cnt;
1215 		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1216 
1217 		for (i = 0; i < cnt; i++) {
1218 			KASSERT(off != TMPFS_DIRCOOKIE_EOF);
1219 			if (off == TMPFS_DIRCOOKIE_DOT) {
1220 				off = TMPFS_DIRCOOKIE_DOTDOT;
1221 			} else {
1222 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1223 					de = TAILQ_FIRST(&node->tn_spec.
1224 					    tn_dir.tn_dir);
1225 				} else if (de != NULL) {
1226 					de = TAILQ_NEXT(de, td_entries);
1227 				} else {
1228 					de = tmpfs_dir_lookupbycookie(node,
1229 					    off);
1230 					KASSERT(de != NULL);
1231 					de = TAILQ_NEXT(de, td_entries);
1232 				}
1233 				if (de == NULL) {
1234 					off = TMPFS_DIRCOOKIE_EOF;
1235 				} else {
1236 					off = tmpfs_dircookie(de);
1237 				}
1238 			}
1239 
1240 			(*cookies)[i] = off;
1241 		}
1242 		KASSERT(uio->uio_offset == off);
1243 	}
1244 
1245 out:
1246 	KASSERT(VOP_ISLOCKED(vp));
1247 
1248 	return error;
1249 }
1250 
1251 /* --------------------------------------------------------------------- */
1252 
1253 int
1254 tmpfs_readlink(void *v)
1255 {
1256 	struct vnode *vp = ((struct vop_readlink_args *)v)->a_vp;
1257 	struct uio *uio = ((struct vop_readlink_args *)v)->a_uio;
1258 
1259 	int error;
1260 	struct tmpfs_node *node;
1261 
1262 	KASSERT(VOP_ISLOCKED(vp));
1263 	KASSERT(uio->uio_offset == 0);
1264 	KASSERT(vp->v_type == VLNK);
1265 
1266 	node = VP_TO_TMPFS_NODE(vp);
1267 
1268 	error = uiomove(node->tn_spec.tn_lnk.tn_link,
1269 	    MIN(node->tn_size, uio->uio_resid), uio);
1270 	node->tn_status |= TMPFS_NODE_ACCESSED;
1271 
1272 	KASSERT(VOP_ISLOCKED(vp));
1273 
1274 	return error;
1275 }
1276 
1277 /* --------------------------------------------------------------------- */
1278 
1279 int
1280 tmpfs_inactive(void *v)
1281 {
1282 	struct vnode *vp = ((struct vop_inactive_args *)v)->a_vp;
1283 
1284 	struct tmpfs_node *node;
1285 
1286 	KASSERT(VOP_ISLOCKED(vp));
1287 
1288 	node = VP_TO_TMPFS_NODE(vp);
1289 	*((struct vop_inactive_args *)v)->a_recycle = (node->tn_links == 0);
1290 	VOP_UNLOCK(vp);
1291 
1292 	return 0;
1293 }
1294 
1295 /* --------------------------------------------------------------------- */
1296 
1297 int
1298 tmpfs_reclaim(void *v)
1299 {
1300 	struct vnode *vp = ((struct vop_reclaim_args *)v)->a_vp;
1301 
1302 	struct tmpfs_mount *tmp;
1303 	struct tmpfs_node *node;
1304 
1305 	node = VP_TO_TMPFS_NODE(vp);
1306 	tmp = VFS_TO_TMPFS(vp->v_mount);
1307 
1308 	cache_purge(vp);
1309 	tmpfs_free_vp(vp);
1310 
1311 	/* If the node referenced by this vnode was deleted by the user,
1312 	 * we must free its associated data structures (now that the vnode
1313 	 * is being reclaimed). */
1314 	if (node->tn_links == 0)
1315 		tmpfs_free_node(tmp, node);
1316 
1317 	KASSERT(vp->v_data == NULL);
1318 
1319 	return 0;
1320 }
1321 
1322 /* --------------------------------------------------------------------- */
1323 
1324 int
1325 tmpfs_print(void *v)
1326 {
1327 	struct vnode *vp = ((struct vop_print_args *)v)->a_vp;
1328 
1329 	struct tmpfs_node *node;
1330 
1331 	node = VP_TO_TMPFS_NODE(vp);
1332 
1333 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1334 	    node, node->tn_flags, node->tn_links);
1335 	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1336 	    ", status 0x%x",
1337 	    node->tn_mode, node->tn_uid, node->tn_gid,
1338 	    (uintmax_t)node->tn_size, node->tn_status);
1339 	if (vp->v_type == VFIFO)
1340 		VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
1341 	printf("\n");
1342 
1343 	return 0;
1344 }
1345 
1346 /* --------------------------------------------------------------------- */
1347 
1348 int
1349 tmpfs_pathconf(void *v)
1350 {
1351 	int name = ((struct vop_pathconf_args *)v)->a_name;
1352 	register_t *retval = ((struct vop_pathconf_args *)v)->a_retval;
1353 
1354 	int error;
1355 
1356 	error = 0;
1357 
1358 	switch (name) {
1359 	case _PC_LINK_MAX:
1360 		*retval = LINK_MAX;
1361 		break;
1362 
1363 	case _PC_NAME_MAX:
1364 		*retval = NAME_MAX;
1365 		break;
1366 
1367 	case _PC_PATH_MAX:
1368 		*retval = PATH_MAX;
1369 		break;
1370 
1371 	case _PC_PIPE_BUF:
1372 		*retval = PIPE_BUF;
1373 		break;
1374 
1375 	case _PC_CHOWN_RESTRICTED:
1376 		*retval = 1;
1377 		break;
1378 
1379 	case _PC_NO_TRUNC:
1380 		*retval = 1;
1381 		break;
1382 
1383 	case _PC_SYNC_IO:
1384 		*retval = 1;
1385 		break;
1386 
1387 	case _PC_FILESIZEBITS:
1388 		*retval = 0; /* XXX Don't know which value should I return. */
1389 		break;
1390 
1391 	default:
1392 		error = EINVAL;
1393 	}
1394 
1395 	return error;
1396 }
1397 
1398 /* --------------------------------------------------------------------- */
1399 
1400 int
1401 tmpfs_advlock(void *v)
1402 {
1403 	struct vnode *vp = ((struct vop_advlock_args *)v)->a_vp;
1404 
1405 	struct tmpfs_node *node;
1406 
1407 	node = VP_TO_TMPFS_NODE(vp);
1408 
1409 	return lf_advlock(v, &node->tn_lockf, node->tn_size);
1410 }
1411 
1412 /* --------------------------------------------------------------------- */
1413 
1414 int
1415 tmpfs_getpages(void *v)
1416 {
1417 	struct vnode *vp = ((struct vop_getpages_args *)v)->a_vp;
1418 	voff_t offset = ((struct vop_getpages_args *)v)->a_offset;
1419 	struct vm_page **m = ((struct vop_getpages_args *)v)->a_m;
1420 	int *count = ((struct vop_getpages_args *)v)->a_count;
1421 	int centeridx = ((struct vop_getpages_args *)v)->a_centeridx;
1422 	vm_prot_t access_type = ((struct vop_getpages_args *)v)->a_access_type;
1423 	int advice = ((struct vop_getpages_args *)v)->a_advice;
1424 	int flags = ((struct vop_getpages_args *)v)->a_flags;
1425 
1426 	int error;
1427 	int i;
1428 	struct tmpfs_node *node;
1429 	struct uvm_object *uobj;
1430 	int npages = *count;
1431 
1432 	KASSERT(vp->v_type == VREG);
1433 	KASSERT(mutex_owned(&vp->v_interlock));
1434 
1435 	node = VP_TO_TMPFS_NODE(vp);
1436 	uobj = node->tn_spec.tn_reg.tn_aobj;
1437 
1438 	/* We currently don't rely on PGO_PASTEOF. */
1439 
1440 	if (vp->v_size <= offset + (centeridx << PAGE_SHIFT)) {
1441 		if ((flags & PGO_LOCKED) == 0)
1442 			mutex_exit(&vp->v_interlock);
1443 		return EINVAL;
1444 	}
1445 
1446 	if (vp->v_size < offset + (npages << PAGE_SHIFT)) {
1447 		npages = (round_page(vp->v_size) - offset) >> PAGE_SHIFT;
1448 	}
1449 
1450 	if ((flags & PGO_LOCKED) != 0)
1451 		return EBUSY;
1452 
1453 	if ((flags & PGO_NOTIMESTAMP) == 0) {
1454 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1455 			node->tn_status |= TMPFS_NODE_ACCESSED;
1456 
1457 		if ((access_type & VM_PROT_WRITE) != 0)
1458 			node->tn_status |= TMPFS_NODE_MODIFIED;
1459 	}
1460 
1461 	mutex_exit(&vp->v_interlock);
1462 
1463 	/*
1464 	 * Make sure that the array on which we will store the
1465 	 * gotten pages is clean.  Otherwise uao_get (pointed to by
1466 	 * the pgo_get below) gets confused and does not return the
1467 	 * appropriate pages.
1468 	 *
1469 	 * XXX This shall be revisited when kern/32166 is addressed
1470 	 * because the loop to clean m[i] will most likely be redundant
1471 	 * as well as the PGO_ALLPAGES flag.
1472 	 */
1473 	if (m != NULL)
1474 		for (i = 0; i < npages; i++)
1475 			m[i] = NULL;
1476 	mutex_enter(&uobj->vmobjlock);
1477 	error = (*uobj->pgops->pgo_get)(uobj, offset, m, &npages, centeridx,
1478 	    access_type, advice, flags | PGO_ALLPAGES);
1479 #if defined(DEBUG)
1480 	{
1481 		/* Make sure that all the pages we return are valid. */
1482 		int dbgi;
1483 		if (error == 0 && m != NULL)
1484 			for (dbgi = 0; dbgi < npages; dbgi++)
1485 				KASSERT(m[dbgi] != NULL);
1486 	}
1487 #endif
1488 
1489 	return error;
1490 }
1491 
1492 /* --------------------------------------------------------------------- */
1493 
1494 int
1495 tmpfs_putpages(void *v)
1496 {
1497 	struct vnode *vp = ((struct vop_putpages_args *)v)->a_vp;
1498 	voff_t offlo = ((struct vop_putpages_args *)v)->a_offlo;
1499 	voff_t offhi = ((struct vop_putpages_args *)v)->a_offhi;
1500 	int flags = ((struct vop_putpages_args *)v)->a_flags;
1501 
1502 	int error;
1503 	struct tmpfs_node *node;
1504 	struct uvm_object *uobj;
1505 
1506 	KASSERT(mutex_owned(&vp->v_interlock));
1507 
1508 	node = VP_TO_TMPFS_NODE(vp);
1509 
1510 	if (vp->v_type != VREG) {
1511 		mutex_exit(&vp->v_interlock);
1512 		return 0;
1513 	}
1514 
1515 	uobj = node->tn_spec.tn_reg.tn_aobj;
1516 	mutex_exit(&vp->v_interlock);
1517 
1518 	mutex_enter(&uobj->vmobjlock);
1519 	error = (*uobj->pgops->pgo_put)(uobj, offlo, offhi, flags);
1520 
1521 	/* XXX mtime */
1522 
1523 	return error;
1524 }
1525