xref: /netbsd-src/sys/fs/tmpfs/tmpfs_vnops.c (revision 5c46dd73a9bcb28b2994504ea090f64066b17a77)
1 /*	$NetBSD: tmpfs_vnops.c,v 1.69 2010/04/23 15:38:47 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 1.69 2010/04/23 15:38:47 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 
116 int
117 tmpfs_lookup(void *v)
118 {
119 	struct vnode *dvp = ((struct vop_lookup_args *)v)->a_dvp;
120 	struct vnode **vpp = ((struct vop_lookup_args *)v)->a_vpp;
121 	struct componentname *cnp = ((struct vop_lookup_args *)v)->a_cnp;
122 
123 	int error;
124 	struct tmpfs_dirent *de;
125 	struct tmpfs_node *dnode;
126 
127 	KASSERT(VOP_ISLOCKED(dvp));
128 
129 	dnode = VP_TO_TMPFS_DIR(dvp);
130 	*vpp = NULL;
131 
132 	/* Check accessibility of requested node as a first step. */
133 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
134 	if (error != 0)
135 		goto out;
136 
137 	/* If requesting the last path component on a read-only file system
138 	 * with a write operation, deny it. */
139 	if ((cnp->cn_flags & ISLASTCN) &&
140 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
141 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
142 		error = EROFS;
143 		goto out;
144 	}
145 
146 	/* Avoid doing a linear scan of the directory if the requested
147 	 * directory/name couple is already in the cache. */
148 	error = cache_lookup(dvp, vpp, cnp);
149 	if (error >= 0)
150 		goto out;
151 
152 	/* We cannot be requesting the parent directory of the root node. */
153 	KASSERT(IMPLIES(dnode->tn_type == VDIR &&
154 	    dnode->tn_spec.tn_dir.tn_parent == dnode,
155 	    !(cnp->cn_flags & ISDOTDOT)));
156 
157 	if (cnp->cn_flags & ISDOTDOT) {
158 		VOP_UNLOCK(dvp, 0);
159 
160 		/* Allocate a new vnode on the matching entry. */
161 		error = tmpfs_alloc_vp(dvp->v_mount,
162 		    dnode->tn_spec.tn_dir.tn_parent, vpp);
163 
164 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
165 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
166 		vref(dvp);
167 		*vpp = dvp;
168 		error = 0;
169 	} else {
170 		de = tmpfs_dir_lookup(dnode, cnp);
171 		if (de == NULL) {
172 			/* The entry was not found in the directory.
173 			 * This is OK iff we are creating or renaming an
174 			 * entry and are working on the last component of
175 			 * the path name. */
176 			if ((cnp->cn_flags & ISLASTCN) &&
177 			    (cnp->cn_nameiop == CREATE || \
178 			    cnp->cn_nameiop == RENAME)) {
179 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
180 				if (error != 0)
181 					goto out;
182 
183 				/* Keep the component name in the buffer for
184 				 * future uses. */
185 				cnp->cn_flags |= SAVENAME;
186 
187 				error = EJUSTRETURN;
188 			} else
189 				error = ENOENT;
190 		} else {
191 			struct tmpfs_node *tnode;
192 
193 			/* The entry was found, so get its associated
194 			 * tmpfs_node. */
195 			tnode = de->td_node;
196 
197 			/* If we are not at the last path component and
198 			 * found a non-directory or non-link entry (which
199 			 * may itself be pointing to a directory), raise
200 			 * an error. */
201 			if ((tnode->tn_type != VDIR &&
202 			    tnode->tn_type != VLNK) &&
203 			    !(cnp->cn_flags & ISLASTCN)) {
204 				error = ENOTDIR;
205 				goto out;
206 			}
207 
208 			/* Check permissions */
209 			if ((cnp->cn_flags & ISLASTCN) &&
210 			    (cnp->cn_nameiop == DELETE ||
211 			    cnp->cn_nameiop == RENAME)) {
212 				kauth_action_t action = 0;
213 
214 				/* This is the file-system's decision. */
215 				if ((dnode->tn_mode & S_ISTXT) != 0 &&
216 				    kauth_cred_geteuid(cnp->cn_cred) != dnode->tn_uid &&
217 				    kauth_cred_geteuid(cnp->cn_cred) != tnode->tn_uid)
218 					error = EPERM;
219 				else
220 					error = 0;
221 
222 				/* Only bother if we're not already failing it. */
223 				if (!error) {
224 					error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
225 				}
226 
227 				if (cnp->cn_nameiop == DELETE)
228 					action |= KAUTH_VNODE_DELETE;
229 				else /* if (cnp->cn_nameiop == RENAME) */
230 					action |= KAUTH_VNODE_RENAME;
231 
232 				error = kauth_authorize_vnode(cnp->cn_cred,
233 				    action, *vpp, dvp, error);
234 				if (error != 0)
235 					goto out;
236 
237 				cnp->cn_flags |= SAVENAME;
238 			} else
239 				de = NULL;
240 
241 			/* Allocate a new vnode on the matching entry. */
242 			error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp);
243 		}
244 	}
245 
246 	/* Store the result of this lookup in the cache.  Avoid this if the
247 	 * request was for creation, as it does not improve timings on
248 	 * emprical tests. */
249 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
250 	    (cnp->cn_flags & ISDOTDOT) == 0)
251 		cache_enter(dvp, *vpp, cnp);
252 
253 out:
254 	/* If there were no errors, *vpp cannot be null and it must be
255 	 * locked. */
256 	KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
257 
258 	/* dvp must always be locked. */
259 	KASSERT(VOP_ISLOCKED(dvp));
260 
261 	return error;
262 }
263 
264 /* --------------------------------------------------------------------- */
265 
266 int
267 tmpfs_create(void *v)
268 {
269 	struct vnode *dvp = ((struct vop_create_args *)v)->a_dvp;
270 	struct vnode **vpp = ((struct vop_create_args *)v)->a_vpp;
271 	struct componentname *cnp = ((struct vop_create_args *)v)->a_cnp;
272 	struct vattr *vap = ((struct vop_create_args *)v)->a_vap;
273 
274 	KASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
275 
276 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
277 }
278 /* --------------------------------------------------------------------- */
279 
280 int
281 tmpfs_mknod(void *v)
282 {
283 	struct vnode *dvp = ((struct vop_mknod_args *)v)->a_dvp;
284 	struct vnode **vpp = ((struct vop_mknod_args *)v)->a_vpp;
285 	struct componentname *cnp = ((struct vop_mknod_args *)v)->a_cnp;
286 	struct vattr *vap = ((struct vop_mknod_args *)v)->a_vap;
287 
288 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
289 	    vap->va_type != VFIFO) {
290 		vput(dvp);
291 		return EINVAL;
292 	}
293 
294 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
295 }
296 
297 /* --------------------------------------------------------------------- */
298 
299 int
300 tmpfs_open(void *v)
301 {
302 	struct vnode *vp = ((struct vop_open_args *)v)->a_vp;
303 	int mode = ((struct vop_open_args *)v)->a_mode;
304 
305 	int error;
306 	struct tmpfs_node *node;
307 
308 	KASSERT(VOP_ISLOCKED(vp));
309 
310 	node = VP_TO_TMPFS_NODE(vp);
311 
312 	/* The file is still active but all its names have been removed
313 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
314 	 * it is about to die. */
315 	if (node->tn_links < 1) {
316 		error = ENOENT;
317 		goto out;
318 	}
319 
320 	/* If the file is marked append-only, deny write requests. */
321 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
322 		error = EPERM;
323 	else
324 		error = 0;
325 
326 out:
327 	KASSERT(VOP_ISLOCKED(vp));
328 
329 	return error;
330 }
331 
332 /* --------------------------------------------------------------------- */
333 
334 int
335 tmpfs_close(void *v)
336 {
337 	struct vnode *vp = ((struct vop_close_args *)v)->a_vp;
338 
339 	struct tmpfs_node *node;
340 
341 	KASSERT(VOP_ISLOCKED(vp));
342 
343 	node = VP_TO_TMPFS_NODE(vp);
344 
345 	if (node->tn_links > 0) {
346 		/* Update node times.  No need to do it if the node has
347 		 * been deleted, because it will vanish after we return. */
348 		tmpfs_update(vp, NULL, NULL, NULL, UPDATE_CLOSE);
349 	}
350 
351 	return 0;
352 }
353 
354 /* --------------------------------------------------------------------- */
355 
356 static int
357 tmpfs_check_possible(struct vnode *vp, struct tmpfs_node *node, mode_t mode)
358 {
359 	int error = 0;
360 
361 	switch (vp->v_type) {
362 	case VDIR:
363 		/* FALLTHROUGH */
364 	case VLNK:
365 		/* FALLTHROUGH */
366 	case VREG:
367 		if (mode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
368 			error = EROFS;
369 			goto out;
370 		}
371 		break;
372 
373 	case VBLK:
374 		/* FALLTHROUGH */
375 	case VCHR:
376 		/* FALLTHROUGH */
377 	case VSOCK:
378 		/* FALLTHROUGH */
379 	case VFIFO:
380 		break;
381 
382 	default:
383 		error = EINVAL;
384 		goto out;
385 	}
386 
387 	if (mode & VWRITE && node->tn_flags & IMMUTABLE) {
388 		error = EPERM;
389 		goto out;
390 	}
391 
392  out:
393 	return error;
394 }
395 
396 static int
397 tmpfs_check_permitted(struct vnode *vp, struct tmpfs_node *node, mode_t mode,
398     kauth_cred_t cred)
399 {
400 
401 	return genfs_can_access(vp->v_type, node->tn_mode, node->tn_uid,
402 	    node->tn_gid, mode, cred);
403 }
404 
405 int
406 tmpfs_access(void *v)
407 {
408 	struct vnode *vp = ((struct vop_access_args *)v)->a_vp;
409 	int mode = ((struct vop_access_args *)v)->a_mode;
410 	kauth_cred_t cred = ((struct vop_access_args *)v)->a_cred;
411 
412 	int error;
413 	struct tmpfs_node *node;
414 
415 	KASSERT(VOP_ISLOCKED(vp));
416 
417 	node = VP_TO_TMPFS_NODE(vp);
418 
419 	error = tmpfs_check_possible(vp, node, mode);
420 	if (error)
421 		goto out;
422 
423 	error = tmpfs_check_permitted(vp, node, mode, cred);
424 
425 	error = kauth_authorize_vnode(cred, kauth_mode_to_action(mode), vp,
426 	    NULL, error);
427 
428 out:
429 	KASSERT(VOP_ISLOCKED(vp));
430 
431 	return error;
432 }
433 
434 /* --------------------------------------------------------------------- */
435 
436 int
437 tmpfs_getattr(void *v)
438 {
439 	struct vnode *vp = ((struct vop_getattr_args *)v)->a_vp;
440 	struct vattr *vap = ((struct vop_getattr_args *)v)->a_vap;
441 
442 	struct tmpfs_node *node;
443 
444 	node = VP_TO_TMPFS_NODE(vp);
445 
446 	vattr_null(vap);
447 
448 	tmpfs_itimes(vp, NULL, NULL, NULL);
449 
450 	vap->va_type = vp->v_type;
451 	vap->va_mode = node->tn_mode;
452 	vap->va_nlink = node->tn_links;
453 	vap->va_uid = node->tn_uid;
454 	vap->va_gid = node->tn_gid;
455 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
456 	vap->va_fileid = node->tn_id;
457 	vap->va_size = node->tn_size;
458 	vap->va_blocksize = PAGE_SIZE;
459 	vap->va_atime = node->tn_atime;
460 	vap->va_mtime = node->tn_mtime;
461 	vap->va_ctime = node->tn_ctime;
462 	vap->va_birthtime = node->tn_birthtime;
463 	vap->va_gen = node->tn_gen;
464 	vap->va_flags = node->tn_flags;
465 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
466 		node->tn_spec.tn_dev.tn_rdev : VNOVAL;
467 	vap->va_bytes = round_page(node->tn_size);
468 	vap->va_filerev = VNOVAL;
469 	vap->va_vaflags = 0;
470 	vap->va_spare = VNOVAL; /* XXX */
471 
472 	return 0;
473 }
474 
475 /* --------------------------------------------------------------------- */
476 
477 #define GOODTIME(tv)	((tv)->tv_sec != VNOVAL || (tv)->tv_nsec != VNOVAL)
478 /* XXX Should this operation be atomic?  I think it should, but code in
479  * XXX other places (e.g., ufs) doesn't seem to be... */
480 int
481 tmpfs_setattr(void *v)
482 {
483 	struct vnode *vp = ((struct vop_setattr_args *)v)->a_vp;
484 	struct vattr *vap = ((struct vop_setattr_args *)v)->a_vap;
485 	kauth_cred_t cred = ((struct vop_setattr_args *)v)->a_cred;
486 	struct lwp *l = curlwp;
487 
488 	int error;
489 
490 	KASSERT(VOP_ISLOCKED(vp));
491 
492 	error = 0;
493 
494 	/* Abort if any unsettable attribute is given. */
495 	if (vap->va_type != VNON ||
496 	    vap->va_nlink != VNOVAL ||
497 	    vap->va_fsid != VNOVAL ||
498 	    vap->va_fileid != VNOVAL ||
499 	    vap->va_blocksize != VNOVAL ||
500 	    GOODTIME(&vap->va_ctime) ||
501 	    vap->va_gen != VNOVAL ||
502 	    vap->va_rdev != VNOVAL ||
503 	    vap->va_bytes != VNOVAL)
504 		error = EINVAL;
505 
506 	if (error == 0 && (vap->va_flags != VNOVAL))
507 		error = tmpfs_chflags(vp, vap->va_flags, cred, l);
508 
509 	if (error == 0 && (vap->va_size != VNOVAL))
510 		error = tmpfs_chsize(vp, vap->va_size, cred, l);
511 
512 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
513 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
514 
515 	if (error == 0 && (vap->va_mode != VNOVAL))
516 		error = tmpfs_chmod(vp, vap->va_mode, cred, l);
517 
518 	if (error == 0 && (GOODTIME(&vap->va_atime) || GOODTIME(&vap->va_mtime)
519 	    || GOODTIME(&vap->va_birthtime)))
520 		if ((error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
521 		    &vap->va_birthtime, vap->va_vaflags, cred, l)) == 0)
522 			return 0;
523 
524 	/* Update the node times.  We give preference to the error codes
525 	 * generated by this function rather than the ones that may arise
526 	 * from tmpfs_update. */
527 	tmpfs_update(vp, NULL, NULL, NULL, 0);
528 
529 	KASSERT(VOP_ISLOCKED(vp));
530 
531 	return error;
532 }
533 
534 /* --------------------------------------------------------------------- */
535 
536 int
537 tmpfs_read(void *v)
538 {
539 	struct vnode *vp = ((struct vop_read_args *)v)->a_vp;
540 	struct uio *uio = ((struct vop_read_args *)v)->a_uio;
541 	int ioflag = ((struct vop_read_args *)v)->a_ioflag;
542 
543 	int error;
544 	struct tmpfs_node *node;
545 	struct uvm_object *uobj;
546 
547 	KASSERT(VOP_ISLOCKED(vp));
548 
549 	node = VP_TO_TMPFS_NODE(vp);
550 
551 	if (vp->v_type != VREG) {
552 		error = EISDIR;
553 		goto out;
554 	}
555 
556 	if (uio->uio_offset < 0) {
557 		error = EINVAL;
558 		goto out;
559 	}
560 
561 	node->tn_status |= TMPFS_NODE_ACCESSED;
562 
563 	uobj = node->tn_spec.tn_reg.tn_aobj;
564 	error = 0;
565 	while (error == 0 && uio->uio_resid > 0) {
566 		vsize_t len;
567 
568 		if (node->tn_size <= uio->uio_offset)
569 			break;
570 
571 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
572 		if (len == 0)
573 			break;
574 
575 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
576 		    UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
577 	}
578 
579 out:
580 	KASSERT(VOP_ISLOCKED(vp));
581 
582 	return error;
583 }
584 
585 /* --------------------------------------------------------------------- */
586 
587 int
588 tmpfs_write(void *v)
589 {
590 	struct vnode *vp = ((struct vop_write_args *)v)->a_vp;
591 	struct uio *uio = ((struct vop_write_args *)v)->a_uio;
592 	int ioflag = ((struct vop_write_args *)v)->a_ioflag;
593 
594 	bool extended;
595 	int error;
596 	off_t oldsize;
597 	struct tmpfs_node *node;
598 	struct uvm_object *uobj;
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 
610 	if (uio->uio_resid == 0) {
611 		error = 0;
612 		goto out;
613 	}
614 
615 	if (ioflag & IO_APPEND)
616 		uio->uio_offset = node->tn_size;
617 
618 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
619 	if (extended) {
620 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
621 		if (error != 0)
622 			goto out;
623 	}
624 
625 	uobj = node->tn_spec.tn_reg.tn_aobj;
626 	error = 0;
627 	while (error == 0 && uio->uio_resid > 0) {
628 		vsize_t len;
629 
630 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
631 		if (len == 0)
632 			break;
633 
634 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
635 		    UBC_WRITE | UBC_UNMAP_FLAG(vp));
636 	}
637 
638 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
639 	    (extended ? TMPFS_NODE_CHANGED : 0);
640 
641 	if (error != 0)
642 		(void)tmpfs_reg_resize(vp, oldsize);
643 
644 	VN_KNOTE(vp, NOTE_WRITE);
645 
646 out:
647 	KASSERT(VOP_ISLOCKED(vp));
648 	KASSERT(IMPLIES(error == 0, uio->uio_resid == 0));
649 	KASSERT(IMPLIES(error != 0, oldsize == node->tn_size));
650 
651 	return error;
652 }
653 
654 /* --------------------------------------------------------------------- */
655 
656 int
657 tmpfs_fsync(void *v)
658 {
659 	struct vnode *vp = ((struct vop_fsync_args *)v)->a_vp;
660 
661 	KASSERT(VOP_ISLOCKED(vp));
662 
663 	tmpfs_update(vp, NULL, NULL, NULL, 0);
664 
665 	return 0;
666 }
667 
668 /* --------------------------------------------------------------------- */
669 
670 int
671 tmpfs_remove(void *v)
672 {
673 	struct vnode *dvp = ((struct vop_remove_args *)v)->a_dvp;
674 	struct vnode *vp = ((struct vop_remove_args *)v)->a_vp;
675 	struct componentname *cnp = (((struct vop_remove_args *)v)->a_cnp);
676 
677 	int error;
678 	struct tmpfs_dirent *de;
679 	struct tmpfs_mount *tmp;
680 	struct tmpfs_node *dnode;
681 	struct tmpfs_node *node;
682 
683 	KASSERT(VOP_ISLOCKED(dvp));
684 	KASSERT(VOP_ISLOCKED(vp));
685 
686 	if (vp->v_type == VDIR) {
687 		error = EPERM;
688 		goto out;
689 	}
690 
691 	dnode = VP_TO_TMPFS_DIR(dvp);
692 	node = VP_TO_TMPFS_NODE(vp);
693 	tmp = VFS_TO_TMPFS(vp->v_mount);
694 	de = tmpfs_dir_lookup(dnode, cnp);
695 	KASSERT(de);
696 	KASSERT(de->td_node == node);
697 
698 	/* Files marked as immutable or append-only cannot be deleted. */
699 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
700 		error = EPERM;
701 		goto out;
702 	}
703 
704 	/* Remove the entry from the directory; as it is a file, we do not
705 	 * have to change the number of hard links of the directory. */
706 	tmpfs_dir_detach(dvp, de);
707 
708 	/* Free the directory entry we just deleted.  Note that the node
709 	 * referred by it will not be removed until the vnode is really
710 	 * reclaimed. */
711 	tmpfs_free_dirent(tmp, de, true);
712 
713 	error = 0;
714 
715 out:
716 	vput(vp);
717 	if (dvp == vp)
718 		vrele(dvp);
719 	else
720 		vput(dvp);
721 	if (cnp->cn_flags & HASBUF) {
722 		PNBUF_PUT(cnp->cn_pnbuf);
723 		cnp->cn_flags &= ~HASBUF;
724 	}
725 
726 	return error;
727 }
728 
729 /* --------------------------------------------------------------------- */
730 
731 int
732 tmpfs_link(void *v)
733 {
734 	struct vnode *dvp = ((struct vop_link_args *)v)->a_dvp;
735 	struct vnode *vp = ((struct vop_link_args *)v)->a_vp;
736 	struct componentname *cnp = ((struct vop_link_args *)v)->a_cnp;
737 
738 	int error;
739 	struct tmpfs_dirent *de;
740 	struct tmpfs_node *dnode;
741 	struct tmpfs_node *node;
742 
743 	KASSERT(VOP_ISLOCKED(dvp));
744 	KASSERT(cnp->cn_flags & HASBUF);
745 	KASSERT(dvp != vp); /* XXX When can this be false? */
746 
747 	dnode = VP_TO_TMPFS_DIR(dvp);
748 	node = VP_TO_TMPFS_NODE(vp);
749 
750 	/* Lock vp because we will need to run tmpfs_update over it, which
751 	 * needs the vnode to be locked. */
752 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
753 
754 	/* XXX: Why aren't the following two tests done by the caller? */
755 
756 	/* Hard links of directories are forbidden. */
757 	if (vp->v_type == VDIR) {
758 		error = EPERM;
759 		goto out;
760 	}
761 
762 	/* Cannot create cross-device links. */
763 	if (dvp->v_mount != vp->v_mount) {
764 		error = EXDEV;
765 		goto out;
766 	}
767 
768 	/* Ensure that we do not overflow the maximum number of links imposed
769 	 * by the system. */
770 	KASSERT(node->tn_links <= LINK_MAX);
771 	if (node->tn_links == LINK_MAX) {
772 		error = EMLINK;
773 		goto out;
774 	}
775 
776 	/* We cannot create links of files marked immutable or append-only. */
777 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
778 		error = EPERM;
779 		goto out;
780 	}
781 
782 	/* Allocate a new directory entry to represent the node. */
783 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
784 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
785 	if (error != 0)
786 		goto out;
787 
788 	/* Insert the new directory entry into the appropriate directory. */
789 	tmpfs_dir_attach(dvp, de);
790 
791 	/* vp link count has changed, so update node times. */
792 	node->tn_status |= TMPFS_NODE_CHANGED;
793 	tmpfs_update(vp, NULL, NULL, NULL, 0);
794 
795 	error = 0;
796 
797 out:
798 	VOP_UNLOCK(vp, 0);
799 	PNBUF_PUT(cnp->cn_pnbuf);
800 	vput(dvp);
801 
802 	return error;
803 }
804 
805 /*
806  * tmpfs_rename: rename routine.
807  *
808  * Arguments: fdvp (from-parent vnode), fvp (from-leaf), tdvp (to-parent)
809  * and tvp (to-leaf), if exists (NULL if not).
810  *
811  * => Caller holds a reference on fdvp and fvp, they are unlocked.
812  *    Note: fdvp and fvp can refer to the same object (i.e. when it is root).
813  *
814  * => Both tdvp and tvp are referenced and locked.  It is our responsibility
815  *    to release the references and unlock them (or destroy).
816  */
817 int
818 tmpfs_rename(void *v)
819 {
820 	struct vnode *fdvp = ((struct vop_rename_args *)v)->a_fdvp;
821 	struct vnode *fvp = ((struct vop_rename_args *)v)->a_fvp;
822 	struct componentname *fcnp = ((struct vop_rename_args *)v)->a_fcnp;
823 	struct vnode *tdvp = ((struct vop_rename_args *)v)->a_tdvp;
824 	struct vnode *tvp = ((struct vop_rename_args *)v)->a_tvp;
825 	struct componentname *tcnp = ((struct vop_rename_args *)v)->a_tcnp;
826 
827 	char *newname;
828 	int error;
829 	struct tmpfs_dirent *de, *de2;
830 	struct tmpfs_mount *tmp;
831 	struct tmpfs_node *fdnode;
832 	struct tmpfs_node *fnode;
833 	struct tmpfs_node *tnode;
834 	struct tmpfs_node *tdnode;
835 	size_t namelen;
836 
837 	KASSERT(VOP_ISLOCKED(tdvp));
838 	KASSERT(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
839 	KASSERT(fcnp->cn_flags & HASBUF);
840 	KASSERT(tcnp->cn_flags & HASBUF);
841 
842 	newname = NULL;
843 	namelen = 0;
844 	tmp = NULL;
845 
846 	/* Disallow cross-device renames. */
847 	if (fvp->v_mount != tdvp->v_mount ||
848 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
849 		error = EXDEV;
850 		goto out_unlocked;
851 	}
852 
853 	fnode = VP_TO_TMPFS_NODE(fvp);
854 	fdnode = VP_TO_TMPFS_DIR(fdvp);
855 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
856 	tdnode = VP_TO_TMPFS_DIR(tdvp);
857 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
858 
859 	if (fdvp == tvp) {
860 		error = 0;
861 		goto out_unlocked;
862 	}
863 
864 	/* If we need to move the directory between entries, lock the
865 	 * source so that we can safely operate on it. */
866 
867 	/* XXX: this is a potential locking order violation! */
868 	if (fdnode != tdnode) {
869 		vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
870 	}
871 
872 	/*
873 	 * If the node we were renaming has scarpered, just give up.
874 	 */
875 	de = tmpfs_dir_lookup(fdnode, fcnp);
876 	if (de == NULL || de->td_node != fnode) {
877 		error = ENOENT;
878 		goto out;
879 	}
880 
881 	/* If source and target is the same vnode, remove the source link. */
882 	if (fvp == tvp) {
883 		/*
884 		 * Detach and free the directory entry.  Drops the link
885 		 * count on the node.
886 		 */
887 		tmpfs_dir_detach(fdvp, de);
888 		tmpfs_free_dirent(VFS_TO_TMPFS(fvp->v_mount), de, true);
889 		VN_KNOTE(fdvp, NOTE_WRITE);
890 		goto out_ok;
891 	}
892 
893 	/* If replacing an existing entry, ensure we can do the operation. */
894 	if (tvp != NULL) {
895 		KASSERT(tnode != NULL);
896 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
897 			if (tnode->tn_size > 0) {
898 				error = ENOTEMPTY;
899 				goto out;
900 			}
901 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
902 			error = ENOTDIR;
903 			goto out;
904 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
905 			error = EISDIR;
906 			goto out;
907 		} else {
908 			KASSERT(fnode->tn_type != VDIR &&
909 			        tnode->tn_type != VDIR);
910 		}
911 	}
912 
913 	/* Ensure that we have enough memory to hold the new name, if it
914 	 * has to be changed. */
915 	namelen = tcnp->cn_namelen;
916 	if (fcnp->cn_namelen != tcnp->cn_namelen ||
917 	    memcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
918 		newname = tmpfs_str_pool_get(&tmp->tm_str_pool, namelen, 0);
919 		if (newname == NULL) {
920 			error = ENOSPC;
921 			goto out;
922 		}
923 	}
924 
925 	/* If the node is being moved to another directory, we have to do
926 	 * the move. */
927 	if (fdnode != tdnode) {
928 		/* In case we are moving a directory, we have to adjust its
929 		 * parent to point to the new parent. */
930 		if (de->td_node->tn_type == VDIR) {
931 			struct tmpfs_node *n;
932 
933 			/* Ensure the target directory is not a child of the
934 			 * directory being moved.  Otherwise, we'd end up
935 			 * with stale nodes. */
936 			n = tdnode;
937 			while (n != n->tn_spec.tn_dir.tn_parent) {
938 				if (n == fnode) {
939 					error = EINVAL;
940 					goto out;
941 				}
942 				n = n->tn_spec.tn_dir.tn_parent;
943 			}
944 
945 			/* Adjust the parent pointer. */
946 			TMPFS_VALIDATE_DIR(fnode);
947 			de->td_node->tn_spec.tn_dir.tn_parent = tdnode;
948 
949 			/* As a result of changing the target of the '..'
950 			 * entry, the link count of the source and target
951 			 * directories has to be adjusted. */
952 			fdnode->tn_links--;
953 			tdnode->tn_links++;
954 		}
955 
956 		/* Do the move: just remove the entry from the source directory
957 		 * and insert it into the target one. */
958 		tmpfs_dir_detach(fdvp, de);
959 		tmpfs_dir_attach(tdvp, de);
960 
961 		/* Notify listeners of fdvp about the change in the directory.
962 		 * We can do it at this point because we aren't touching fdvp
963 		 * any more below. */
964 		VN_KNOTE(fdvp, NOTE_WRITE);
965 	}
966 
967 	/* If we are overwriting an entry, we have to remove the old one
968 	 * from the target directory. */
969 	if (tvp != NULL) {
970 		KASSERT(tnode != NULL);
971 
972 		/* Remove the old entry from the target directory.
973 		 * Note! This relies on tmpfs_dir_attach() putting the new
974 		 * node on the end of the target's node list. */
975 		de2 = tmpfs_dir_lookup(tdnode, tcnp);
976 		KASSERT(de2 != NULL);
977 		KASSERT(de2->td_node == tnode);
978 		tmpfs_dir_detach(tdvp, de2);
979 
980 		/* Free the directory entry we just deleted.  Note that the
981 		 * node referred by it will not be removed until the vnode is
982 		 * really reclaimed. */
983 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de2, true);
984 	}
985 
986 	/* If the name has changed, we need to make it effective by changing
987 	 * it in the directory entry. */
988 	if (newname != NULL) {
989 		KASSERT(tcnp->cn_namelen < MAXNAMLEN);
990 		KASSERT(tcnp->cn_namelen < 0xffff);
991 
992 		tmpfs_str_pool_put(&tmp->tm_str_pool, de->td_name,
993 		    de->td_namelen);
994 		de->td_namelen = (uint16_t)namelen;
995 		memcpy(newname, tcnp->cn_nameptr, namelen);
996 		de->td_name = newname;
997 		newname = NULL;
998 
999 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1000 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1001 	}
1002  out_ok:
1003 	/* Notify listeners of tdvp about the change in the directory (either
1004 	 * because a new entry was added or because one was removed) and
1005 	 * listeners of fvp about the rename. */
1006 	VN_KNOTE(tdvp, NOTE_WRITE);
1007 	VN_KNOTE(fvp, NOTE_RENAME);
1008 
1009 	error = 0;
1010 
1011  out:
1012 	if (fdnode != tdnode)
1013 		VOP_UNLOCK(fdvp, 0);
1014 
1015  out_unlocked:
1016 	/* Release target nodes. */
1017 	if (tdvp == tvp)
1018 		vrele(tdvp);
1019 	else
1020 		vput(tdvp);
1021 	if (tvp != NULL)
1022 		vput(tvp);
1023 
1024 	/* Release source nodes. */
1025 	vrele(fdvp);
1026 	vrele(fvp);
1027 
1028 	if (newname != NULL)
1029 		tmpfs_str_pool_put(&tmp->tm_str_pool, newname, namelen);
1030 
1031 	return error;
1032 }
1033 
1034 /* --------------------------------------------------------------------- */
1035 
1036 int
1037 tmpfs_mkdir(void *v)
1038 {
1039 	struct vnode *dvp = ((struct vop_mkdir_args *)v)->a_dvp;
1040 	struct vnode **vpp = ((struct vop_mkdir_args *)v)->a_vpp;
1041 	struct componentname *cnp = ((struct vop_mkdir_args *)v)->a_cnp;
1042 	struct vattr *vap = ((struct vop_mkdir_args *)v)->a_vap;
1043 
1044 	KASSERT(vap->va_type == VDIR);
1045 
1046 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1047 }
1048 
1049 /* --------------------------------------------------------------------- */
1050 
1051 int
1052 tmpfs_rmdir(void *v)
1053 {
1054 	struct vnode *dvp = ((struct vop_rmdir_args *)v)->a_dvp;
1055 	struct vnode *vp = ((struct vop_rmdir_args *)v)->a_vp;
1056 	struct componentname *cnp = ((struct vop_rmdir_args *)v)->a_cnp;
1057 
1058 	int error;
1059 	struct tmpfs_dirent *de;
1060 	struct tmpfs_mount *tmp;
1061 	struct tmpfs_node *dnode;
1062 	struct tmpfs_node *node;
1063 
1064 	KASSERT(VOP_ISLOCKED(dvp));
1065 	KASSERT(VOP_ISLOCKED(vp));
1066 
1067 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1068 	dnode = VP_TO_TMPFS_DIR(dvp);
1069 	node = VP_TO_TMPFS_DIR(vp);
1070 	error = 0;
1071 
1072 	/* Directories with more than two entries ('.' and '..') cannot be
1073 	 * removed. */
1074 	if (node->tn_size > 0) {
1075 		error = ENOTEMPTY;
1076 		goto out;
1077 	}
1078 
1079 	/* This invariant holds only if we are not trying to remove "..".
1080 	 * We checked for that above so this is safe now. */
1081 	KASSERT(node->tn_spec.tn_dir.tn_parent == dnode);
1082 
1083 	/* Get the directory entry associated with node (vp). */
1084 	de = tmpfs_dir_lookup(dnode, cnp);
1085 	KASSERT(de);
1086 	KASSERT(de->td_node == node);
1087 
1088 	/* Check flags to see if we are allowed to remove the directory. */
1089 	if (dnode->tn_flags & APPEND || node->tn_flags & (IMMUTABLE | APPEND)) {
1090 		error = EPERM;
1091 		goto out;
1092 	}
1093 
1094 	/* Detach the directory entry from the directory (dnode). */
1095 	tmpfs_dir_detach(dvp, de);
1096 
1097 	node->tn_links--;
1098 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1099 	    TMPFS_NODE_MODIFIED;
1100 	node->tn_spec.tn_dir.tn_parent->tn_links--;
1101 	node->tn_spec.tn_dir.tn_parent->tn_status |= TMPFS_NODE_ACCESSED | \
1102 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1103 
1104 	/* Release the parent. */
1105 	cache_purge(dvp); /* XXX Is this needed? */
1106 
1107 	/* Free the directory entry we just deleted.  Note that the node
1108 	 * referred by it will not be removed until the vnode is really
1109 	 * reclaimed. */
1110 	tmpfs_free_dirent(tmp, de, true);
1111 
1112 	KASSERT(node->tn_links == 0);
1113  out:
1114 	/* Release the nodes. */
1115 	vput(dvp);
1116 	vput(vp);
1117 	PNBUF_PUT(cnp->cn_pnbuf);
1118 
1119 	return error;
1120 }
1121 
1122 /* --------------------------------------------------------------------- */
1123 
1124 int
1125 tmpfs_symlink(void *v)
1126 {
1127 	struct vnode *dvp = ((struct vop_symlink_args *)v)->a_dvp;
1128 	struct vnode **vpp = ((struct vop_symlink_args *)v)->a_vpp;
1129 	struct componentname *cnp = ((struct vop_symlink_args *)v)->a_cnp;
1130 	struct vattr *vap = ((struct vop_symlink_args *)v)->a_vap;
1131 	char *target = ((struct vop_symlink_args *)v)->a_target;
1132 
1133 	KASSERT(vap->va_type == VLNK);
1134 
1135 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1136 }
1137 
1138 /* --------------------------------------------------------------------- */
1139 
1140 int
1141 tmpfs_readdir(void *v)
1142 {
1143 	struct vnode *vp = ((struct vop_readdir_args *)v)->a_vp;
1144 	struct uio *uio = ((struct vop_readdir_args *)v)->a_uio;
1145 	int *eofflag = ((struct vop_readdir_args *)v)->a_eofflag;
1146 	off_t **cookies = ((struct vop_readdir_args *)v)->a_cookies;
1147 	int *ncookies = ((struct vop_readdir_args *)v)->a_ncookies;
1148 
1149 	int error;
1150 	off_t startoff;
1151 	off_t cnt;
1152 	struct tmpfs_node *node;
1153 
1154 	KASSERT(VOP_ISLOCKED(vp));
1155 
1156 	/* This operation only makes sense on directory nodes. */
1157 	if (vp->v_type != VDIR) {
1158 		error = ENOTDIR;
1159 		goto out;
1160 	}
1161 
1162 	node = VP_TO_TMPFS_DIR(vp);
1163 
1164 	startoff = uio->uio_offset;
1165 
1166 	cnt = 0;
1167 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1168 		error = tmpfs_dir_getdotdent(node, uio);
1169 		if (error == -1) {
1170 			error = 0;
1171 			goto outok;
1172 		} else if (error != 0)
1173 			goto outok;
1174 		cnt++;
1175 	}
1176 
1177 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1178 		error = tmpfs_dir_getdotdotdent(node, uio);
1179 		if (error == -1) {
1180 			error = 0;
1181 			goto outok;
1182 		} else if (error != 0)
1183 			goto outok;
1184 		cnt++;
1185 	}
1186 
1187 	error = tmpfs_dir_getdents(node, uio, &cnt);
1188 	if (error == -1)
1189 		error = 0;
1190 	KASSERT(error >= 0);
1191 
1192 outok:
1193 	/* This label assumes that startoff has been
1194 	 * initialized.  If the compiler didn't spit out warnings, we'd
1195 	 * simply make this one be 'out' and drop 'outok'. */
1196 
1197 	if (eofflag != NULL)
1198 		*eofflag =
1199 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1200 
1201 	/* Update NFS-related variables. */
1202 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1203 		off_t i;
1204 		off_t off = startoff;
1205 		struct tmpfs_dirent *de = NULL;
1206 
1207 		*ncookies = cnt;
1208 		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1209 
1210 		for (i = 0; i < cnt; i++) {
1211 			KASSERT(off != TMPFS_DIRCOOKIE_EOF);
1212 			if (off == TMPFS_DIRCOOKIE_DOT) {
1213 				off = TMPFS_DIRCOOKIE_DOTDOT;
1214 			} else {
1215 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1216 					de = TAILQ_FIRST(&node->tn_spec.
1217 					    tn_dir.tn_dir);
1218 				} else if (de != NULL) {
1219 					de = TAILQ_NEXT(de, td_entries);
1220 				} else {
1221 					de = tmpfs_dir_lookupbycookie(node,
1222 					    off);
1223 					KASSERT(de != NULL);
1224 					de = TAILQ_NEXT(de, td_entries);
1225 				}
1226 				if (de == NULL) {
1227 					off = TMPFS_DIRCOOKIE_EOF;
1228 				} else {
1229 					off = tmpfs_dircookie(de);
1230 				}
1231 			}
1232 
1233 			(*cookies)[i] = off;
1234 		}
1235 		KASSERT(uio->uio_offset == off);
1236 	}
1237 
1238 out:
1239 	KASSERT(VOP_ISLOCKED(vp));
1240 
1241 	return error;
1242 }
1243 
1244 /* --------------------------------------------------------------------- */
1245 
1246 int
1247 tmpfs_readlink(void *v)
1248 {
1249 	struct vnode *vp = ((struct vop_readlink_args *)v)->a_vp;
1250 	struct uio *uio = ((struct vop_readlink_args *)v)->a_uio;
1251 
1252 	int error;
1253 	struct tmpfs_node *node;
1254 
1255 	KASSERT(VOP_ISLOCKED(vp));
1256 	KASSERT(uio->uio_offset == 0);
1257 	KASSERT(vp->v_type == VLNK);
1258 
1259 	node = VP_TO_TMPFS_NODE(vp);
1260 
1261 	error = uiomove(node->tn_spec.tn_lnk.tn_link,
1262 	    MIN(node->tn_size, uio->uio_resid), uio);
1263 	node->tn_status |= TMPFS_NODE_ACCESSED;
1264 
1265 	KASSERT(VOP_ISLOCKED(vp));
1266 
1267 	return error;
1268 }
1269 
1270 /* --------------------------------------------------------------------- */
1271 
1272 int
1273 tmpfs_inactive(void *v)
1274 {
1275 	struct vnode *vp = ((struct vop_inactive_args *)v)->a_vp;
1276 
1277 	struct tmpfs_node *node;
1278 
1279 	KASSERT(VOP_ISLOCKED(vp));
1280 
1281 	node = VP_TO_TMPFS_NODE(vp);
1282 	*((struct vop_inactive_args *)v)->a_recycle = (node->tn_links == 0);
1283 	VOP_UNLOCK(vp, 0);
1284 
1285 	return 0;
1286 }
1287 
1288 /* --------------------------------------------------------------------- */
1289 
1290 int
1291 tmpfs_reclaim(void *v)
1292 {
1293 	struct vnode *vp = ((struct vop_reclaim_args *)v)->a_vp;
1294 
1295 	struct tmpfs_mount *tmp;
1296 	struct tmpfs_node *node;
1297 
1298 	node = VP_TO_TMPFS_NODE(vp);
1299 	tmp = VFS_TO_TMPFS(vp->v_mount);
1300 
1301 	cache_purge(vp);
1302 	tmpfs_free_vp(vp);
1303 
1304 	/* If the node referenced by this vnode was deleted by the user,
1305 	 * we must free its associated data structures (now that the vnode
1306 	 * is being reclaimed). */
1307 	if (node->tn_links == 0)
1308 		tmpfs_free_node(tmp, node);
1309 
1310 	KASSERT(vp->v_data == NULL);
1311 
1312 	return 0;
1313 }
1314 
1315 /* --------------------------------------------------------------------- */
1316 
1317 int
1318 tmpfs_print(void *v)
1319 {
1320 	struct vnode *vp = ((struct vop_print_args *)v)->a_vp;
1321 
1322 	struct tmpfs_node *node;
1323 
1324 	node = VP_TO_TMPFS_NODE(vp);
1325 
1326 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1327 	    node, node->tn_flags, node->tn_links);
1328 	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1329 	    ", status 0x%x",
1330 	    node->tn_mode, node->tn_uid, node->tn_gid,
1331 	    (uintmax_t)node->tn_size, node->tn_status);
1332 	if (vp->v_type == VFIFO)
1333 		VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
1334 	printf("\n");
1335 
1336 	return 0;
1337 }
1338 
1339 /* --------------------------------------------------------------------- */
1340 
1341 int
1342 tmpfs_pathconf(void *v)
1343 {
1344 	int name = ((struct vop_pathconf_args *)v)->a_name;
1345 	register_t *retval = ((struct vop_pathconf_args *)v)->a_retval;
1346 
1347 	int error;
1348 
1349 	error = 0;
1350 
1351 	switch (name) {
1352 	case _PC_LINK_MAX:
1353 		*retval = LINK_MAX;
1354 		break;
1355 
1356 	case _PC_NAME_MAX:
1357 		*retval = NAME_MAX;
1358 		break;
1359 
1360 	case _PC_PATH_MAX:
1361 		*retval = PATH_MAX;
1362 		break;
1363 
1364 	case _PC_PIPE_BUF:
1365 		*retval = PIPE_BUF;
1366 		break;
1367 
1368 	case _PC_CHOWN_RESTRICTED:
1369 		*retval = 1;
1370 		break;
1371 
1372 	case _PC_NO_TRUNC:
1373 		*retval = 1;
1374 		break;
1375 
1376 	case _PC_SYNC_IO:
1377 		*retval = 1;
1378 		break;
1379 
1380 	case _PC_FILESIZEBITS:
1381 		*retval = 0; /* XXX Don't know which value should I return. */
1382 		break;
1383 
1384 	default:
1385 		error = EINVAL;
1386 	}
1387 
1388 	return error;
1389 }
1390 
1391 /* --------------------------------------------------------------------- */
1392 
1393 int
1394 tmpfs_advlock(void *v)
1395 {
1396 	struct vnode *vp = ((struct vop_advlock_args *)v)->a_vp;
1397 
1398 	struct tmpfs_node *node;
1399 
1400 	node = VP_TO_TMPFS_NODE(vp);
1401 
1402 	return lf_advlock(v, &node->tn_lockf, node->tn_size);
1403 }
1404 
1405 /* --------------------------------------------------------------------- */
1406 
1407 int
1408 tmpfs_getpages(void *v)
1409 {
1410 	struct vnode *vp = ((struct vop_getpages_args *)v)->a_vp;
1411 	voff_t offset = ((struct vop_getpages_args *)v)->a_offset;
1412 	struct vm_page **m = ((struct vop_getpages_args *)v)->a_m;
1413 	int *count = ((struct vop_getpages_args *)v)->a_count;
1414 	int centeridx = ((struct vop_getpages_args *)v)->a_centeridx;
1415 	vm_prot_t access_type = ((struct vop_getpages_args *)v)->a_access_type;
1416 	int advice = ((struct vop_getpages_args *)v)->a_advice;
1417 	int flags = ((struct vop_getpages_args *)v)->a_flags;
1418 
1419 	int error;
1420 	int i;
1421 	struct tmpfs_node *node;
1422 	struct uvm_object *uobj;
1423 	int npages = *count;
1424 
1425 	KASSERT(vp->v_type == VREG);
1426 	KASSERT(mutex_owned(&vp->v_interlock));
1427 
1428 	node = VP_TO_TMPFS_NODE(vp);
1429 	uobj = node->tn_spec.tn_reg.tn_aobj;
1430 
1431 	/* We currently don't rely on PGO_PASTEOF. */
1432 
1433 	if (vp->v_size <= offset + (centeridx << PAGE_SHIFT)) {
1434 		if ((flags & PGO_LOCKED) == 0)
1435 			mutex_exit(&vp->v_interlock);
1436 		return EINVAL;
1437 	}
1438 
1439 	if (vp->v_size < offset + (npages << PAGE_SHIFT)) {
1440 		npages = (round_page(vp->v_size) - offset) >> PAGE_SHIFT;
1441 	}
1442 
1443 	if ((flags & PGO_LOCKED) != 0)
1444 		return EBUSY;
1445 
1446 	if ((flags & PGO_NOTIMESTAMP) == 0) {
1447 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1448 			node->tn_status |= TMPFS_NODE_ACCESSED;
1449 
1450 		if ((access_type & VM_PROT_WRITE) != 0)
1451 			node->tn_status |= TMPFS_NODE_MODIFIED;
1452 	}
1453 
1454 	mutex_exit(&vp->v_interlock);
1455 
1456 	/*
1457 	 * Make sure that the array on which we will store the
1458 	 * gotten pages is clean.  Otherwise uao_get (pointed to by
1459 	 * the pgo_get below) gets confused and does not return the
1460 	 * appropriate pages.
1461 	 *
1462 	 * XXX This shall be revisited when kern/32166 is addressed
1463 	 * because the loop to clean m[i] will most likely be redundant
1464 	 * as well as the PGO_ALLPAGES flag.
1465 	 */
1466 	if (m != NULL)
1467 		for (i = 0; i < npages; i++)
1468 			m[i] = NULL;
1469 	mutex_enter(&uobj->vmobjlock);
1470 	error = (*uobj->pgops->pgo_get)(uobj, offset, m, &npages, centeridx,
1471 	    access_type, advice, flags | PGO_ALLPAGES);
1472 #if defined(DEBUG)
1473 	{
1474 		/* Make sure that all the pages we return are valid. */
1475 		int dbgi;
1476 		if (error == 0 && m != NULL)
1477 			for (dbgi = 0; dbgi < npages; dbgi++)
1478 				KASSERT(m[dbgi] != NULL);
1479 	}
1480 #endif
1481 
1482 	return error;
1483 }
1484 
1485 /* --------------------------------------------------------------------- */
1486 
1487 int
1488 tmpfs_putpages(void *v)
1489 {
1490 	struct vnode *vp = ((struct vop_putpages_args *)v)->a_vp;
1491 	voff_t offlo = ((struct vop_putpages_args *)v)->a_offlo;
1492 	voff_t offhi = ((struct vop_putpages_args *)v)->a_offhi;
1493 	int flags = ((struct vop_putpages_args *)v)->a_flags;
1494 
1495 	int error;
1496 	struct tmpfs_node *node;
1497 	struct uvm_object *uobj;
1498 
1499 	KASSERT(mutex_owned(&vp->v_interlock));
1500 
1501 	node = VP_TO_TMPFS_NODE(vp);
1502 
1503 	if (vp->v_type != VREG) {
1504 		mutex_exit(&vp->v_interlock);
1505 		return 0;
1506 	}
1507 
1508 	uobj = node->tn_spec.tn_reg.tn_aobj;
1509 	mutex_exit(&vp->v_interlock);
1510 
1511 	mutex_enter(&uobj->vmobjlock);
1512 	error = (*uobj->pgops->pgo_put)(uobj, offlo, offhi, flags);
1513 
1514 	/* XXX mtime */
1515 
1516 	return error;
1517 }
1518