xref: /freebsd-src/sys/fs/tmpfs/tmpfs_subr.c (revision cf771f223b1a69e11fd1b70b0a274a73fe844335)
1 /*	$NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 2005 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  * Efficient memory file system supporting functions.
35  */
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/param.h>
40 #include <sys/namei.h>
41 #include <sys/priv.h>
42 #include <sys/proc.h>
43 #include <sys/stat.h>
44 #include <sys/systm.h>
45 #include <sys/sysctl.h>
46 #include <sys/vnode.h>
47 #include <sys/vmmeter.h>
48 
49 #include <vm/vm.h>
50 #include <vm/vm_object.h>
51 #include <vm/vm_page.h>
52 #include <vm/vm_pageout.h>
53 #include <vm/vm_pager.h>
54 #include <vm/vm_extern.h>
55 
56 #include <fs/tmpfs/tmpfs.h>
57 #include <fs/tmpfs/tmpfs_fifoops.h>
58 #include <fs/tmpfs/tmpfs_vnops.h>
59 
60 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system");
61 
62 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
63 
64 static int
65 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
66 {
67 	int error;
68 	long pages, bytes;
69 
70 	pages = *(long *)arg1;
71 	bytes = pages * PAGE_SIZE;
72 
73 	error = sysctl_handle_long(oidp, &bytes, 0, req);
74 	if (error || !req->newptr)
75 		return (error);
76 
77 	pages = bytes / PAGE_SIZE;
78 	if (pages < TMPFS_PAGES_MINRESERVED)
79 		return (EINVAL);
80 
81 	*(long *)arg1 = pages;
82 	return (0);
83 }
84 
85 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW,
86     &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L", "reserved memory");
87 
88 size_t
89 tmpfs_mem_avail(void)
90 {
91 	vm_ooffset_t avail;
92 
93 	avail = swap_pager_avail + cnt.v_free_count + cnt.v_cache_count -
94 	    tmpfs_pages_reserved;
95 	if (__predict_false(avail < 0))
96 		avail = 0;
97 	return (avail);
98 }
99 
100 size_t
101 tmpfs_pages_used(struct tmpfs_mount *tmp)
102 {
103 	const size_t node_size = sizeof(struct tmpfs_node) +
104 	    sizeof(struct tmpfs_dirent);
105 	size_t meta_pages;
106 
107 	meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
108 	    PAGE_SIZE);
109 	return (meta_pages + tmp->tm_pages_used);
110 }
111 
112 static size_t
113 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
114 {
115 	if (tmpfs_mem_avail() < req_pages)
116 		return (0);
117 
118 	if (tmp->tm_pages_max != SIZE_MAX &&
119 	    tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
120 			return (0);
121 
122 	return (1);
123 }
124 
125 /* --------------------------------------------------------------------- */
126 
127 /*
128  * Allocates a new node of type 'type' inside the 'tmp' mount point, with
129  * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
130  * using the credentials of the process 'p'.
131  *
132  * If the node type is set to 'VDIR', then the parent parameter must point
133  * to the parent directory of the node being created.  It may only be NULL
134  * while allocating the root node.
135  *
136  * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
137  * specifies the device the node represents.
138  *
139  * If the node type is set to 'VLNK', then the parameter target specifies
140  * the file name of the target file for the symbolic link that is being
141  * created.
142  *
143  * Note that new nodes are retrieved from the available list if it has
144  * items or, if it is empty, from the node pool as long as there is enough
145  * space to create them.
146  *
147  * Returns zero on success or an appropriate error code on failure.
148  */
149 int
150 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
151     uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
152     char *target, dev_t rdev, struct tmpfs_node **node)
153 {
154 	struct tmpfs_node *nnode;
155 
156 	/* If the root directory of the 'tmp' file system is not yet
157 	 * allocated, this must be the request to do it. */
158 	MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
159 
160 	MPASS(IFF(type == VLNK, target != NULL));
161 	MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
162 
163 	if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
164 		return (ENOSPC);
165 	if (tmpfs_pages_check_avail(tmp, 1) == 0)
166 		return (ENOSPC);
167 
168 	nnode = (struct tmpfs_node *)uma_zalloc_arg(
169 				tmp->tm_node_pool, tmp, M_WAITOK);
170 
171 	/* Generic initialization. */
172 	nnode->tn_type = type;
173 	vfs_timestamp(&nnode->tn_atime);
174 	nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
175 	    nnode->tn_atime;
176 	nnode->tn_uid = uid;
177 	nnode->tn_gid = gid;
178 	nnode->tn_mode = mode;
179 	nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
180 
181 	/* Type-specific initialization. */
182 	switch (nnode->tn_type) {
183 	case VBLK:
184 	case VCHR:
185 		nnode->tn_rdev = rdev;
186 		break;
187 
188 	case VDIR:
189 		TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
190 		MPASS(parent != nnode);
191 		MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
192 		nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
193 		nnode->tn_dir.tn_readdir_lastn = 0;
194 		nnode->tn_dir.tn_readdir_lastp = NULL;
195 		nnode->tn_links++;
196 		TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
197 		nnode->tn_dir.tn_parent->tn_links++;
198 		TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
199 		break;
200 
201 	case VFIFO:
202 		/* FALLTHROUGH */
203 	case VSOCK:
204 		break;
205 
206 	case VLNK:
207 		MPASS(strlen(target) < MAXPATHLEN);
208 		nnode->tn_size = strlen(target);
209 		nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
210 		    M_WAITOK);
211 		memcpy(nnode->tn_link, target, nnode->tn_size);
212 		break;
213 
214 	case VREG:
215 		nnode->tn_reg.tn_aobj =
216 		    vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
217 			NULL /* XXXKIB - tmpfs needs swap reservation */);
218 		break;
219 
220 	default:
221 		panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
222 	}
223 
224 	TMPFS_LOCK(tmp);
225 	LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
226 	tmp->tm_nodes_inuse++;
227 	TMPFS_UNLOCK(tmp);
228 
229 	*node = nnode;
230 	return 0;
231 }
232 
233 /* --------------------------------------------------------------------- */
234 
235 /*
236  * Destroys the node pointed to by node from the file system 'tmp'.
237  * If the node does not belong to the given mount point, the results are
238  * unpredicted.
239  *
240  * If the node references a directory; no entries are allowed because
241  * their removal could need a recursive algorithm, something forbidden in
242  * kernel space.  Furthermore, there is not need to provide such
243  * functionality (recursive removal) because the only primitives offered
244  * to the user are the removal of empty directories and the deletion of
245  * individual files.
246  *
247  * Note that nodes are not really deleted; in fact, when a node has been
248  * allocated, it cannot be deleted during the whole life of the file
249  * system.  Instead, they are moved to the available list and remain there
250  * until reused.
251  */
252 void
253 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
254 {
255 	vm_object_t uobj;
256 
257 #ifdef INVARIANTS
258 	TMPFS_NODE_LOCK(node);
259 	MPASS(node->tn_vnode == NULL);
260 	MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
261 	TMPFS_NODE_UNLOCK(node);
262 #endif
263 
264 	TMPFS_LOCK(tmp);
265 	LIST_REMOVE(node, tn_entries);
266 	tmp->tm_nodes_inuse--;
267 	TMPFS_UNLOCK(tmp);
268 
269 	switch (node->tn_type) {
270 	case VNON:
271 		/* Do not do anything.  VNON is provided to let the
272 		 * allocation routine clean itself easily by avoiding
273 		 * duplicating code in it. */
274 		/* FALLTHROUGH */
275 	case VBLK:
276 		/* FALLTHROUGH */
277 	case VCHR:
278 		/* FALLTHROUGH */
279 	case VDIR:
280 		/* FALLTHROUGH */
281 	case VFIFO:
282 		/* FALLTHROUGH */
283 	case VSOCK:
284 		break;
285 
286 	case VLNK:
287 		free(node->tn_link, M_TMPFSNAME);
288 		break;
289 
290 	case VREG:
291 		uobj = node->tn_reg.tn_aobj;
292 		if (uobj != NULL) {
293 			TMPFS_LOCK(tmp);
294 			tmp->tm_pages_used -= uobj->size;
295 			TMPFS_UNLOCK(tmp);
296 			vm_object_deallocate(uobj);
297 		}
298 		break;
299 
300 	default:
301 		panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
302 	}
303 
304 	free_unr(tmp->tm_ino_unr, node->tn_id);
305 	uma_zfree(tmp->tm_node_pool, node);
306 }
307 
308 /* --------------------------------------------------------------------- */
309 
310 /*
311  * Allocates a new directory entry for the node node with a name of name.
312  * The new directory entry is returned in *de.
313  *
314  * The link count of node is increased by one to reflect the new object
315  * referencing it.
316  *
317  * Returns zero on success or an appropriate error code on failure.
318  */
319 int
320 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
321     const char *name, uint16_t len, struct tmpfs_dirent **de)
322 {
323 	struct tmpfs_dirent *nde;
324 
325 	nde = (struct tmpfs_dirent *)uma_zalloc(
326 					tmp->tm_dirent_pool, M_WAITOK);
327 	nde->td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
328 	nde->td_namelen = len;
329 	memcpy(nde->td_name, name, len);
330 
331 	nde->td_node = node;
332 	if (node != NULL)
333 		node->tn_links++;
334 
335 	*de = nde;
336 
337 	return 0;
338 }
339 
340 /* --------------------------------------------------------------------- */
341 
342 /*
343  * Frees a directory entry.  It is the caller's responsibility to destroy
344  * the node referenced by it if needed.
345  *
346  * The link count of node is decreased by one to reflect the removal of an
347  * object that referenced it.  This only happens if 'node_exists' is true;
348  * otherwise the function will not access the node referred to by the
349  * directory entry, as it may already have been released from the outside.
350  */
351 void
352 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de,
353     boolean_t node_exists)
354 {
355 	if (node_exists) {
356 		struct tmpfs_node *node;
357 
358 		node = de->td_node;
359 		if (node != NULL) {
360 			MPASS(node->tn_links > 0);
361 			node->tn_links--;
362 		}
363 	}
364 
365 	free(de->td_name, M_TMPFSNAME);
366 	uma_zfree(tmp->tm_dirent_pool, de);
367 }
368 
369 /* --------------------------------------------------------------------- */
370 
371 /*
372  * Allocates a new vnode for the node node or returns a new reference to
373  * an existing one if the node had already a vnode referencing it.  The
374  * resulting locked vnode is returned in *vpp.
375  *
376  * Returns zero on success or an appropriate error code on failure.
377  */
378 int
379 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
380     struct vnode **vpp)
381 {
382 	int error = 0;
383 	struct vnode *vp;
384 
385 loop:
386 	TMPFS_NODE_LOCK(node);
387 	if ((vp = node->tn_vnode) != NULL) {
388 		MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
389 		VI_LOCK(vp);
390 		TMPFS_NODE_UNLOCK(node);
391 		error = vget(vp, lkflag | LK_INTERLOCK, curthread);
392 		if (error != 0) {
393 			vp = NULL;
394 			goto out;
395 		}
396 
397 		/*
398 		 * Make sure the vnode is still there after
399 		 * getting the interlock to avoid racing a free.
400 		 */
401 		if (node->tn_vnode == NULL || node->tn_vnode != vp) {
402 			vput(vp);
403 			goto loop;
404 		}
405 
406 		goto out;
407 	}
408 
409 	if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
410 	    (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
411 		TMPFS_NODE_UNLOCK(node);
412 		error = ENOENT;
413 		vp = NULL;
414 		goto out;
415 	}
416 
417 	/*
418 	 * otherwise lock the vp list while we call getnewvnode
419 	 * since that can block.
420 	 */
421 	if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
422 		node->tn_vpstate |= TMPFS_VNODE_WANT;
423 		error = msleep((caddr_t) &node->tn_vpstate,
424 		    TMPFS_NODE_MTX(node), PDROP | PCATCH,
425 		    "tmpfs_alloc_vp", 0);
426 		if (error)
427 			return error;
428 
429 		goto loop;
430 	} else
431 		node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
432 
433 	TMPFS_NODE_UNLOCK(node);
434 
435 	/* Get a new vnode and associate it with our node. */
436 	error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
437 	if (error != 0)
438 		goto unlock;
439 	MPASS(vp != NULL);
440 
441 	(void) vn_lock(vp, lkflag | LK_RETRY);
442 
443 	vp->v_data = node;
444 	vp->v_type = node->tn_type;
445 
446 	/* Type-specific initialization. */
447 	switch (node->tn_type) {
448 	case VBLK:
449 		/* FALLTHROUGH */
450 	case VCHR:
451 		/* FALLTHROUGH */
452 	case VLNK:
453 		/* FALLTHROUGH */
454 	case VREG:
455 		/* FALLTHROUGH */
456 	case VSOCK:
457 		break;
458 	case VFIFO:
459 		vp->v_op = &tmpfs_fifoop_entries;
460 		break;
461 	case VDIR:
462 		MPASS(node->tn_dir.tn_parent != NULL);
463 		if (node->tn_dir.tn_parent == node)
464 			vp->v_vflag |= VV_ROOT;
465 		break;
466 
467 	default:
468 		panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
469 	}
470 
471 	vnode_pager_setsize(vp, node->tn_size);
472 	error = insmntque(vp, mp);
473 	if (error)
474 		vp = NULL;
475 
476 unlock:
477 	TMPFS_NODE_LOCK(node);
478 
479 	MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
480 	node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
481 	node->tn_vnode = vp;
482 
483 	if (node->tn_vpstate & TMPFS_VNODE_WANT) {
484 		node->tn_vpstate &= ~TMPFS_VNODE_WANT;
485 		TMPFS_NODE_UNLOCK(node);
486 		wakeup((caddr_t) &node->tn_vpstate);
487 	} else
488 		TMPFS_NODE_UNLOCK(node);
489 
490 out:
491 	*vpp = vp;
492 
493 #ifdef INVARIANTS
494 	if (error == 0) {
495 		MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
496 		TMPFS_NODE_LOCK(node);
497 		MPASS(*vpp == node->tn_vnode);
498 		TMPFS_NODE_UNLOCK(node);
499 	}
500 #endif
501 
502 	return error;
503 }
504 
505 /* --------------------------------------------------------------------- */
506 
507 /*
508  * Destroys the association between the vnode vp and the node it
509  * references.
510  */
511 void
512 tmpfs_free_vp(struct vnode *vp)
513 {
514 	struct tmpfs_node *node;
515 
516 	node = VP_TO_TMPFS_NODE(vp);
517 
518 	mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED);
519 	node->tn_vnode = NULL;
520 	vp->v_data = NULL;
521 }
522 
523 /* --------------------------------------------------------------------- */
524 
525 /*
526  * Allocates a new file of type 'type' and adds it to the parent directory
527  * 'dvp'; this addition is done using the component name given in 'cnp'.
528  * The ownership of the new file is automatically assigned based on the
529  * credentials of the caller (through 'cnp'), the group is set based on
530  * the parent directory and the mode is determined from the 'vap' argument.
531  * If successful, *vpp holds a vnode to the newly created file and zero
532  * is returned.  Otherwise *vpp is NULL and the function returns an
533  * appropriate error code.
534  */
535 int
536 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
537     struct componentname *cnp, char *target)
538 {
539 	int error;
540 	struct tmpfs_dirent *de;
541 	struct tmpfs_mount *tmp;
542 	struct tmpfs_node *dnode;
543 	struct tmpfs_node *node;
544 	struct tmpfs_node *parent;
545 
546 	MPASS(VOP_ISLOCKED(dvp));
547 	MPASS(cnp->cn_flags & HASBUF);
548 
549 	tmp = VFS_TO_TMPFS(dvp->v_mount);
550 	dnode = VP_TO_TMPFS_DIR(dvp);
551 	*vpp = NULL;
552 
553 	/* If the entry we are creating is a directory, we cannot overflow
554 	 * the number of links of its parent, because it will get a new
555 	 * link. */
556 	if (vap->va_type == VDIR) {
557 		/* Ensure that we do not overflow the maximum number of links
558 		 * imposed by the system. */
559 		MPASS(dnode->tn_links <= LINK_MAX);
560 		if (dnode->tn_links == LINK_MAX) {
561 			error = EMLINK;
562 			goto out;
563 		}
564 
565 		parent = dnode;
566 		MPASS(parent != NULL);
567 	} else
568 		parent = NULL;
569 
570 	/* Allocate a node that represents the new file. */
571 	error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid,
572 	    dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node);
573 	if (error != 0)
574 		goto out;
575 
576 	/* Allocate a directory entry that points to the new file. */
577 	error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
578 	    &de);
579 	if (error != 0) {
580 		tmpfs_free_node(tmp, node);
581 		goto out;
582 	}
583 
584 	/* Allocate a vnode for the new file. */
585 	error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
586 	if (error != 0) {
587 		tmpfs_free_dirent(tmp, de, TRUE);
588 		tmpfs_free_node(tmp, node);
589 		goto out;
590 	}
591 
592 	/* Now that all required items are allocated, we can proceed to
593 	 * insert the new node into the directory, an operation that
594 	 * cannot fail. */
595 	if (cnp->cn_flags & ISWHITEOUT)
596 		tmpfs_dir_whiteout_remove(dvp, cnp);
597 	tmpfs_dir_attach(dvp, de);
598 
599 out:
600 
601 	return error;
602 }
603 
604 /* --------------------------------------------------------------------- */
605 
606 /*
607  * Attaches the directory entry de to the directory represented by vp.
608  * Note that this does not change the link count of the node pointed by
609  * the directory entry, as this is done by tmpfs_alloc_dirent.
610  */
611 void
612 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
613 {
614 	struct tmpfs_node *dnode;
615 
616 	ASSERT_VOP_ELOCKED(vp, __func__);
617 	dnode = VP_TO_TMPFS_DIR(vp);
618 	TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
619 	dnode->tn_size += sizeof(struct tmpfs_dirent);
620 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
621 	    TMPFS_NODE_MODIFIED;
622 }
623 
624 /* --------------------------------------------------------------------- */
625 
626 /*
627  * Detaches the directory entry de from the directory represented by vp.
628  * Note that this does not change the link count of the node pointed by
629  * the directory entry, as this is done by tmpfs_free_dirent.
630  */
631 void
632 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
633 {
634 	struct tmpfs_node *dnode;
635 
636 	ASSERT_VOP_ELOCKED(vp, __func__);
637 	dnode = VP_TO_TMPFS_DIR(vp);
638 
639 	if (dnode->tn_dir.tn_readdir_lastp == de) {
640 		dnode->tn_dir.tn_readdir_lastn = 0;
641 		dnode->tn_dir.tn_readdir_lastp = NULL;
642 	}
643 
644 	TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
645 	dnode->tn_size -= sizeof(struct tmpfs_dirent);
646 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
647 	    TMPFS_NODE_MODIFIED;
648 }
649 
650 /* --------------------------------------------------------------------- */
651 
652 /*
653  * Looks for a directory entry in the directory represented by node.
654  * 'cnp' describes the name of the entry to look for.  Note that the .
655  * and .. components are not allowed as they do not physically exist
656  * within directories.
657  *
658  * Returns a pointer to the entry when found, otherwise NULL.
659  */
660 struct tmpfs_dirent *
661 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
662     struct componentname *cnp)
663 {
664 	boolean_t found;
665 	struct tmpfs_dirent *de;
666 
667 	MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
668 	MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
669 	    cnp->cn_nameptr[1] == '.')));
670 	TMPFS_VALIDATE_DIR(node);
671 
672 	found = 0;
673 	TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
674 		if (f != NULL && de->td_node != f)
675 		    continue;
676 		MPASS(cnp->cn_namelen < 0xffff);
677 		if (de->td_namelen == (uint16_t)cnp->cn_namelen &&
678 		    bcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) {
679 			found = 1;
680 			break;
681 		}
682 	}
683 	node->tn_status |= TMPFS_NODE_ACCESSED;
684 
685 	return found ? de : NULL;
686 }
687 
688 /* --------------------------------------------------------------------- */
689 
690 /*
691  * Helper function for tmpfs_readdir.  Creates a '.' entry for the given
692  * directory and returns it in the uio space.  The function returns 0
693  * on success, -1 if there was not enough space in the uio structure to
694  * hold the directory entry or an appropriate error code if another
695  * error happens.
696  */
697 int
698 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
699 {
700 	int error;
701 	struct dirent dent;
702 
703 	TMPFS_VALIDATE_DIR(node);
704 	MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
705 
706 	dent.d_fileno = node->tn_id;
707 	dent.d_type = DT_DIR;
708 	dent.d_namlen = 1;
709 	dent.d_name[0] = '.';
710 	dent.d_name[1] = '\0';
711 	dent.d_reclen = GENERIC_DIRSIZ(&dent);
712 
713 	if (dent.d_reclen > uio->uio_resid)
714 		error = -1;
715 	else {
716 		error = uiomove(&dent, dent.d_reclen, uio);
717 		if (error == 0)
718 			uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
719 	}
720 
721 	node->tn_status |= TMPFS_NODE_ACCESSED;
722 
723 	return error;
724 }
725 
726 /* --------------------------------------------------------------------- */
727 
728 /*
729  * Helper function for tmpfs_readdir.  Creates a '..' entry for the given
730  * directory and returns it in the uio space.  The function returns 0
731  * on success, -1 if there was not enough space in the uio structure to
732  * hold the directory entry or an appropriate error code if another
733  * error happens.
734  */
735 int
736 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
737 {
738 	int error;
739 	struct dirent dent;
740 
741 	TMPFS_VALIDATE_DIR(node);
742 	MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
743 
744 	/*
745 	 * Return ENOENT if the current node is already removed.
746 	 */
747 	TMPFS_ASSERT_LOCKED(node);
748 	if (node->tn_dir.tn_parent == NULL) {
749 		return (ENOENT);
750 	}
751 
752 	TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
753 	dent.d_fileno = node->tn_dir.tn_parent->tn_id;
754 	TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
755 
756 	dent.d_type = DT_DIR;
757 	dent.d_namlen = 2;
758 	dent.d_name[0] = '.';
759 	dent.d_name[1] = '.';
760 	dent.d_name[2] = '\0';
761 	dent.d_reclen = GENERIC_DIRSIZ(&dent);
762 
763 	if (dent.d_reclen > uio->uio_resid)
764 		error = -1;
765 	else {
766 		error = uiomove(&dent, dent.d_reclen, uio);
767 		if (error == 0) {
768 			struct tmpfs_dirent *de;
769 
770 			de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
771 			if (de == NULL)
772 				uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
773 			else
774 				uio->uio_offset = tmpfs_dircookie(de);
775 		}
776 	}
777 
778 	node->tn_status |= TMPFS_NODE_ACCESSED;
779 
780 	return error;
781 }
782 
783 /* --------------------------------------------------------------------- */
784 
785 /*
786  * Lookup a directory entry by its associated cookie.
787  */
788 struct tmpfs_dirent *
789 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
790 {
791 	struct tmpfs_dirent *de;
792 
793 	if (cookie == node->tn_dir.tn_readdir_lastn &&
794 	    node->tn_dir.tn_readdir_lastp != NULL) {
795 		return node->tn_dir.tn_readdir_lastp;
796 	}
797 
798 	TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
799 		if (tmpfs_dircookie(de) == cookie) {
800 			break;
801 		}
802 	}
803 
804 	return de;
805 }
806 
807 /* --------------------------------------------------------------------- */
808 
809 /*
810  * Helper function for tmpfs_readdir.  Returns as much directory entries
811  * as can fit in the uio space.  The read starts at uio->uio_offset.
812  * The function returns 0 on success, -1 if there was not enough space
813  * in the uio structure to hold the directory entry or an appropriate
814  * error code if another error happens.
815  */
816 int
817 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
818 {
819 	int error;
820 	off_t startcookie;
821 	struct tmpfs_dirent *de;
822 
823 	TMPFS_VALIDATE_DIR(node);
824 
825 	/* Locate the first directory entry we have to return.  We have cached
826 	 * the last readdir in the node, so use those values if appropriate.
827 	 * Otherwise do a linear scan to find the requested entry. */
828 	startcookie = uio->uio_offset;
829 	MPASS(startcookie != TMPFS_DIRCOOKIE_DOT);
830 	MPASS(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
831 	if (startcookie == TMPFS_DIRCOOKIE_EOF) {
832 		return 0;
833 	} else {
834 		de = tmpfs_dir_lookupbycookie(node, startcookie);
835 	}
836 	if (de == NULL) {
837 		return EINVAL;
838 	}
839 
840 	/* Read as much entries as possible; i.e., until we reach the end of
841 	 * the directory or we exhaust uio space. */
842 	do {
843 		struct dirent d;
844 
845 		/* Create a dirent structure representing the current
846 		 * tmpfs_node and fill it. */
847 		if (de->td_node == NULL) {
848 			d.d_fileno = 1;
849 			d.d_type = DT_WHT;
850 		} else {
851 			d.d_fileno = de->td_node->tn_id;
852 			switch (de->td_node->tn_type) {
853 			case VBLK:
854 				d.d_type = DT_BLK;
855 				break;
856 
857 			case VCHR:
858 				d.d_type = DT_CHR;
859 				break;
860 
861 			case VDIR:
862 				d.d_type = DT_DIR;
863 				break;
864 
865 			case VFIFO:
866 				d.d_type = DT_FIFO;
867 				break;
868 
869 			case VLNK:
870 				d.d_type = DT_LNK;
871 				break;
872 
873 			case VREG:
874 				d.d_type = DT_REG;
875 				break;
876 
877 			case VSOCK:
878 				d.d_type = DT_SOCK;
879 				break;
880 
881 			default:
882 				panic("tmpfs_dir_getdents: type %p %d",
883 				    de->td_node, (int)de->td_node->tn_type);
884 			}
885 		}
886 		d.d_namlen = de->td_namelen;
887 		MPASS(de->td_namelen < sizeof(d.d_name));
888 		(void)memcpy(d.d_name, de->td_name, de->td_namelen);
889 		d.d_name[de->td_namelen] = '\0';
890 		d.d_reclen = GENERIC_DIRSIZ(&d);
891 
892 		/* Stop reading if the directory entry we are treating is
893 		 * bigger than the amount of data that can be returned. */
894 		if (d.d_reclen > uio->uio_resid) {
895 			error = -1;
896 			break;
897 		}
898 
899 		/* Copy the new dirent structure into the output buffer and
900 		 * advance pointers. */
901 		error = uiomove(&d, d.d_reclen, uio);
902 		if (error == 0) {
903 			(*cntp)++;
904 			de = TAILQ_NEXT(de, td_entries);
905 		}
906 	} while (error == 0 && uio->uio_resid > 0 && de != NULL);
907 
908 	/* Update the offset and cache. */
909 	if (de == NULL) {
910 		uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
911 		node->tn_dir.tn_readdir_lastn = 0;
912 		node->tn_dir.tn_readdir_lastp = NULL;
913 	} else {
914 		node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
915 		node->tn_dir.tn_readdir_lastp = de;
916 	}
917 
918 	node->tn_status |= TMPFS_NODE_ACCESSED;
919 	return error;
920 }
921 
922 int
923 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
924 {
925 	struct tmpfs_dirent *de;
926 	int error;
927 
928 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
929 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
930 	if (error != 0)
931 		return (error);
932 	tmpfs_dir_attach(dvp, de);
933 	return (0);
934 }
935 
936 void
937 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
938 {
939 	struct tmpfs_dirent *de;
940 
941 	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
942 	MPASS(de != NULL && de->td_node == NULL);
943 	tmpfs_dir_detach(dvp, de);
944 	tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de, TRUE);
945 }
946 
947 /* --------------------------------------------------------------------- */
948 
949 /*
950  * Resizes the aobj associated with the regular file pointed to by 'vp' to the
951  * size 'newsize'.  'vp' must point to a vnode that represents a regular file.
952  * 'newsize' must be positive.
953  *
954  * Returns zero on success or an appropriate error code on failure.
955  */
956 int
957 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
958 {
959 	struct tmpfs_mount *tmp;
960 	struct tmpfs_node *node;
961 	vm_object_t uobj;
962 	vm_page_t m, ma[1];
963 	vm_pindex_t idx, newpages, oldpages;
964 	off_t oldsize;
965 	int base, rv;
966 
967 	MPASS(vp->v_type == VREG);
968 	MPASS(newsize >= 0);
969 
970 	node = VP_TO_TMPFS_NODE(vp);
971 	uobj = node->tn_reg.tn_aobj;
972 	tmp = VFS_TO_TMPFS(vp->v_mount);
973 
974 	/*
975 	 * Convert the old and new sizes to the number of pages needed to
976 	 * store them.  It may happen that we do not need to do anything
977 	 * because the last allocated page can accommodate the change on
978 	 * its own.
979 	 */
980 	oldsize = node->tn_size;
981 	oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
982 	MPASS(oldpages == uobj->size);
983 	newpages = OFF_TO_IDX(newsize + PAGE_MASK);
984 	if (newpages > oldpages &&
985 	    tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
986 		return (ENOSPC);
987 
988 	VM_OBJECT_LOCK(uobj);
989 	if (newsize < oldsize) {
990 		/*
991 		 * Zero the truncated part of the last page.
992 		 */
993 		base = newsize & PAGE_MASK;
994 		if (base != 0) {
995 			idx = OFF_TO_IDX(newsize);
996 retry:
997 			m = vm_page_lookup(uobj, idx);
998 			if (m != NULL) {
999 				if ((m->oflags & VPO_BUSY) != 0 ||
1000 				    m->busy != 0) {
1001 					vm_page_sleep(m, "tmfssz");
1002 					goto retry;
1003 				}
1004 				MPASS(m->valid == VM_PAGE_BITS_ALL);
1005 			} else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1006 				m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL);
1007 				if (m == NULL) {
1008 					VM_OBJECT_UNLOCK(uobj);
1009 					VM_WAIT;
1010 					VM_OBJECT_LOCK(uobj);
1011 					goto retry;
1012 				} else if (m->valid != VM_PAGE_BITS_ALL) {
1013 					ma[0] = m;
1014 					rv = vm_pager_get_pages(uobj, ma, 1, 0);
1015 					m = vm_page_lookup(uobj, idx);
1016 				} else
1017 					/* A cached page was reactivated. */
1018 					rv = VM_PAGER_OK;
1019 				vm_page_lock(m);
1020 				if (rv == VM_PAGER_OK) {
1021 					vm_page_deactivate(m);
1022 					vm_page_unlock(m);
1023 					vm_page_wakeup(m);
1024 				} else {
1025 					vm_page_free(m);
1026 					vm_page_unlock(m);
1027 					if (ignerr)
1028 						m = NULL;
1029 					else {
1030 						VM_OBJECT_UNLOCK(uobj);
1031 						return (EIO);
1032 					}
1033 				}
1034 			}
1035 			if (m != NULL) {
1036 				pmap_zero_page_area(m, base, PAGE_SIZE - base);
1037 				vm_page_dirty(m);
1038 				vm_pager_page_unswapped(m);
1039 			}
1040 		}
1041 
1042 		/*
1043 		 * Release any swap space and free any whole pages.
1044 		 */
1045 		if (newpages < oldpages) {
1046 			swap_pager_freespace(uobj, newpages, oldpages -
1047 			    newpages);
1048 			vm_object_page_remove(uobj, newpages, 0, 0);
1049 		}
1050 	}
1051 	uobj->size = newpages;
1052 	VM_OBJECT_UNLOCK(uobj);
1053 
1054 	TMPFS_LOCK(tmp);
1055 	tmp->tm_pages_used += (newpages - oldpages);
1056 	TMPFS_UNLOCK(tmp);
1057 
1058 	node->tn_size = newsize;
1059 	vnode_pager_setsize(vp, newsize);
1060 	return (0);
1061 }
1062 
1063 /* --------------------------------------------------------------------- */
1064 
1065 /*
1066  * Change flags of the given vnode.
1067  * Caller should execute tmpfs_update on vp after a successful execution.
1068  * The vnode must be locked on entry and remain locked on exit.
1069  */
1070 int
1071 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred, struct thread *p)
1072 {
1073 	int error;
1074 	struct tmpfs_node *node;
1075 
1076 	MPASS(VOP_ISLOCKED(vp));
1077 
1078 	node = VP_TO_TMPFS_NODE(vp);
1079 
1080 	/* Disallow this operation if the file system is mounted read-only. */
1081 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1082 		return EROFS;
1083 
1084 	/*
1085 	 * Callers may only modify the file flags on objects they
1086 	 * have VADMIN rights for.
1087 	 */
1088 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1089 		return (error);
1090 	/*
1091 	 * Unprivileged processes are not permitted to unset system
1092 	 * flags, or modify flags if any system flags are set.
1093 	 */
1094 	if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1095 		if (node->tn_flags
1096 		  & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1097 			error = securelevel_gt(cred, 0);
1098 			if (error)
1099 				return (error);
1100 		}
1101 		/* Snapshot flag cannot be set or cleared */
1102 		if (((flags & SF_SNAPSHOT) != 0 &&
1103 		  (node->tn_flags & SF_SNAPSHOT) == 0) ||
1104 		  ((flags & SF_SNAPSHOT) == 0 &&
1105 		  (node->tn_flags & SF_SNAPSHOT) != 0))
1106 			return (EPERM);
1107 		node->tn_flags = flags;
1108 	} else {
1109 		if (node->tn_flags
1110 		  & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1111 		  (flags & UF_SETTABLE) != flags)
1112 			return (EPERM);
1113 		node->tn_flags &= SF_SETTABLE;
1114 		node->tn_flags |= (flags & UF_SETTABLE);
1115 	}
1116 	node->tn_status |= TMPFS_NODE_CHANGED;
1117 
1118 	MPASS(VOP_ISLOCKED(vp));
1119 
1120 	return 0;
1121 }
1122 
1123 /* --------------------------------------------------------------------- */
1124 
1125 /*
1126  * Change access mode on the given vnode.
1127  * Caller should execute tmpfs_update on vp after a successful execution.
1128  * The vnode must be locked on entry and remain locked on exit.
1129  */
1130 int
1131 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1132 {
1133 	int error;
1134 	struct tmpfs_node *node;
1135 
1136 	MPASS(VOP_ISLOCKED(vp));
1137 
1138 	node = VP_TO_TMPFS_NODE(vp);
1139 
1140 	/* Disallow this operation if the file system is mounted read-only. */
1141 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1142 		return EROFS;
1143 
1144 	/* Immutable or append-only files cannot be modified, either. */
1145 	if (node->tn_flags & (IMMUTABLE | APPEND))
1146 		return EPERM;
1147 
1148 	/*
1149 	 * To modify the permissions on a file, must possess VADMIN
1150 	 * for that file.
1151 	 */
1152 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1153 		return (error);
1154 
1155 	/*
1156 	 * Privileged processes may set the sticky bit on non-directories,
1157 	 * as well as set the setgid bit on a file with a group that the
1158 	 * process is not a member of.
1159 	 */
1160 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1161 		if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1162 			return (EFTYPE);
1163 	}
1164 	if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1165 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1166 		if (error)
1167 			return (error);
1168 	}
1169 
1170 
1171 	node->tn_mode &= ~ALLPERMS;
1172 	node->tn_mode |= mode & ALLPERMS;
1173 
1174 	node->tn_status |= TMPFS_NODE_CHANGED;
1175 
1176 	MPASS(VOP_ISLOCKED(vp));
1177 
1178 	return 0;
1179 }
1180 
1181 /* --------------------------------------------------------------------- */
1182 
1183 /*
1184  * Change ownership of the given vnode.  At least one of uid or gid must
1185  * be different than VNOVAL.  If one is set to that value, the attribute
1186  * is unchanged.
1187  * Caller should execute tmpfs_update on vp after a successful execution.
1188  * The vnode must be locked on entry and remain locked on exit.
1189  */
1190 int
1191 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1192     struct thread *p)
1193 {
1194 	int error;
1195 	struct tmpfs_node *node;
1196 	uid_t ouid;
1197 	gid_t ogid;
1198 
1199 	MPASS(VOP_ISLOCKED(vp));
1200 
1201 	node = VP_TO_TMPFS_NODE(vp);
1202 
1203 	/* Assign default values if they are unknown. */
1204 	MPASS(uid != VNOVAL || gid != VNOVAL);
1205 	if (uid == VNOVAL)
1206 		uid = node->tn_uid;
1207 	if (gid == VNOVAL)
1208 		gid = node->tn_gid;
1209 	MPASS(uid != VNOVAL && gid != VNOVAL);
1210 
1211 	/* Disallow this operation if the file system is mounted read-only. */
1212 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1213 		return EROFS;
1214 
1215 	/* Immutable or append-only files cannot be modified, either. */
1216 	if (node->tn_flags & (IMMUTABLE | APPEND))
1217 		return EPERM;
1218 
1219 	/*
1220 	 * To modify the ownership of a file, must possess VADMIN for that
1221 	 * file.
1222 	 */
1223 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1224 		return (error);
1225 
1226 	/*
1227 	 * To change the owner of a file, or change the group of a file to a
1228 	 * group of which we are not a member, the caller must have
1229 	 * privilege.
1230 	 */
1231 	if ((uid != node->tn_uid ||
1232 	    (gid != node->tn_gid && !groupmember(gid, cred))) &&
1233 	    (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1234 		return (error);
1235 
1236 	ogid = node->tn_gid;
1237 	ouid = node->tn_uid;
1238 
1239 	node->tn_uid = uid;
1240 	node->tn_gid = gid;
1241 
1242 	node->tn_status |= TMPFS_NODE_CHANGED;
1243 
1244 	if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1245 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1246 			node->tn_mode &= ~(S_ISUID | S_ISGID);
1247 	}
1248 
1249 	MPASS(VOP_ISLOCKED(vp));
1250 
1251 	return 0;
1252 }
1253 
1254 /* --------------------------------------------------------------------- */
1255 
1256 /*
1257  * Change size of the given vnode.
1258  * Caller should execute tmpfs_update on vp after a successful execution.
1259  * The vnode must be locked on entry and remain locked on exit.
1260  */
1261 int
1262 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1263     struct thread *p)
1264 {
1265 	int error;
1266 	struct tmpfs_node *node;
1267 
1268 	MPASS(VOP_ISLOCKED(vp));
1269 
1270 	node = VP_TO_TMPFS_NODE(vp);
1271 
1272 	/* Decide whether this is a valid operation based on the file type. */
1273 	error = 0;
1274 	switch (vp->v_type) {
1275 	case VDIR:
1276 		return EISDIR;
1277 
1278 	case VREG:
1279 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
1280 			return EROFS;
1281 		break;
1282 
1283 	case VBLK:
1284 		/* FALLTHROUGH */
1285 	case VCHR:
1286 		/* FALLTHROUGH */
1287 	case VFIFO:
1288 		/* Allow modifications of special files even if in the file
1289 		 * system is mounted read-only (we are not modifying the
1290 		 * files themselves, but the objects they represent). */
1291 		return 0;
1292 
1293 	default:
1294 		/* Anything else is unsupported. */
1295 		return EOPNOTSUPP;
1296 	}
1297 
1298 	/* Immutable or append-only files cannot be modified, either. */
1299 	if (node->tn_flags & (IMMUTABLE | APPEND))
1300 		return EPERM;
1301 
1302 	error = tmpfs_truncate(vp, size);
1303 	/* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1304 	 * for us, as will update tn_status; no need to do that here. */
1305 
1306 	MPASS(VOP_ISLOCKED(vp));
1307 
1308 	return error;
1309 }
1310 
1311 /* --------------------------------------------------------------------- */
1312 
1313 /*
1314  * Change access and modification times of the given vnode.
1315  * Caller should execute tmpfs_update on vp after a successful execution.
1316  * The vnode must be locked on entry and remain locked on exit.
1317  */
1318 int
1319 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1320 	struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l)
1321 {
1322 	int error;
1323 	struct tmpfs_node *node;
1324 
1325 	MPASS(VOP_ISLOCKED(vp));
1326 
1327 	node = VP_TO_TMPFS_NODE(vp);
1328 
1329 	/* Disallow this operation if the file system is mounted read-only. */
1330 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1331 		return EROFS;
1332 
1333 	/* Immutable or append-only files cannot be modified, either. */
1334 	if (node->tn_flags & (IMMUTABLE | APPEND))
1335 		return EPERM;
1336 
1337 	/* Determine if the user have proper privilege to update time. */
1338 	if (vaflags & VA_UTIMES_NULL) {
1339 		error = VOP_ACCESS(vp, VADMIN, cred, l);
1340 		if (error)
1341 			error = VOP_ACCESS(vp, VWRITE, cred, l);
1342 	} else
1343 		error = VOP_ACCESS(vp, VADMIN, cred, l);
1344 	if (error)
1345 		return (error);
1346 
1347 	if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1348 		node->tn_status |= TMPFS_NODE_ACCESSED;
1349 
1350 	if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1351 		node->tn_status |= TMPFS_NODE_MODIFIED;
1352 
1353 	if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1354 		node->tn_status |= TMPFS_NODE_MODIFIED;
1355 
1356 	tmpfs_itimes(vp, atime, mtime);
1357 
1358 	if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1359 		node->tn_birthtime = *birthtime;
1360 	MPASS(VOP_ISLOCKED(vp));
1361 
1362 	return 0;
1363 }
1364 
1365 /* --------------------------------------------------------------------- */
1366 /* Sync timestamps */
1367 void
1368 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1369     const struct timespec *mod)
1370 {
1371 	struct tmpfs_node *node;
1372 	struct timespec now;
1373 
1374 	node = VP_TO_TMPFS_NODE(vp);
1375 
1376 	if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1377 	    TMPFS_NODE_CHANGED)) == 0)
1378 		return;
1379 
1380 	vfs_timestamp(&now);
1381 	if (node->tn_status & TMPFS_NODE_ACCESSED) {
1382 		if (acc == NULL)
1383 			 acc = &now;
1384 		node->tn_atime = *acc;
1385 	}
1386 	if (node->tn_status & TMPFS_NODE_MODIFIED) {
1387 		if (mod == NULL)
1388 			mod = &now;
1389 		node->tn_mtime = *mod;
1390 	}
1391 	if (node->tn_status & TMPFS_NODE_CHANGED) {
1392 		node->tn_ctime = now;
1393 	}
1394 	node->tn_status &=
1395 	    ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1396 }
1397 
1398 /* --------------------------------------------------------------------- */
1399 
1400 void
1401 tmpfs_update(struct vnode *vp)
1402 {
1403 
1404 	tmpfs_itimes(vp, NULL, NULL);
1405 }
1406 
1407 /* --------------------------------------------------------------------- */
1408 
1409 int
1410 tmpfs_truncate(struct vnode *vp, off_t length)
1411 {
1412 	int error;
1413 	struct tmpfs_node *node;
1414 
1415 	node = VP_TO_TMPFS_NODE(vp);
1416 
1417 	if (length < 0) {
1418 		error = EINVAL;
1419 		goto out;
1420 	}
1421 
1422 	if (node->tn_size == length) {
1423 		error = 0;
1424 		goto out;
1425 	}
1426 
1427 	if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1428 		return (EFBIG);
1429 
1430 	error = tmpfs_reg_resize(vp, length, FALSE);
1431 	if (error == 0) {
1432 		node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1433 	}
1434 
1435 out:
1436 	tmpfs_update(vp);
1437 
1438 	return error;
1439 }
1440