xref: /dflybsd-src/sys/vfs/tmpfs/tmpfs_subr.c (revision 71c97a3cee892aedd0f1dc81caba4c8fbdf7fc00)
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 
37 #include <sys/kernel.h>
38 #include <sys/param.h>
39 #include <sys/namei.h>
40 #include <sys/priv.h>
41 #include <sys/proc.h>
42 #include <sys/stat.h>
43 #include <sys/systm.h>
44 #include <sys/vnode.h>
45 #include <sys/vmmeter.h>
46 
47 #include <vm/vm.h>
48 #include <vm/vm_object.h>
49 #include <vm/vm_page.h>
50 #include <vm/vm_pager.h>
51 #include <vm/vm_extern.h>
52 #include <vm/vm_pageout.h>
53 #include <vm/vm_page2.h>
54 
55 #include <vfs/tmpfs/tmpfs.h>
56 #include <vfs/tmpfs/tmpfs_vnops.h>
57 
58 static ino_t tmpfs_fetch_ino(struct tmpfs_mount *);
59 
60 static int tmpfs_dirtree_compare(struct tmpfs_dirent *a,
61 	struct tmpfs_dirent *b);
62 RB_GENERATE(tmpfs_dirtree, tmpfs_dirent, rb_node, tmpfs_dirtree_compare);
63 
64 static int tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a,
65 	struct tmpfs_dirent *b);
66 RB_GENERATE(tmpfs_dirtree_cookie, tmpfs_dirent,
67 	rb_cookienode, tmpfs_dirtree_compare_cookie);
68 
69 
70 /* --------------------------------------------------------------------- */
71 
72 /*
73  * Allocates a new node of type 'type' inside the 'tmp' mount point, with
74  * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
75  * using the credentials of the process 'p'.
76  *
77  * If the node type is set to 'VDIR', then the parent parameter must point
78  * to the parent directory of the node being created.  It may only be NULL
79  * while allocating the root node.
80  *
81  * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
82  * specifies the device the node represents.
83  *
84  * If the node type is set to 'VLNK', then the parameter target specifies
85  * the file name of the target file for the symbolic link that is being
86  * created.
87  *
88  * Note that new nodes are retrieved from the available list if it has
89  * items or, if it is empty, from the node pool as long as there is enough
90  * space to create them.
91  *
92  * Returns zero on success or an appropriate error code on failure.
93  */
94 int
95 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
96 		 uid_t uid, gid_t gid, mode_t mode,
97 		 char *target, int rmajor, int rminor,
98 		 struct tmpfs_node **node)
99 {
100 	struct tmpfs_node *nnode;
101 	struct timespec ts;
102 	dev_t rdev;
103 
104 	KKASSERT(IFF(type == VLNK, target != NULL));
105 	KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL));
106 
107 	if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
108 		return (ENOSPC);
109 
110 	nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK);
111 	if (nnode == NULL)
112 		return (ENOSPC);
113 
114 	/* Generic initialization. */
115 	nnode->tn_type = type;
116 	vfs_timestamp(&ts);
117 	nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime
118 		= ts.tv_sec;
119 	nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec
120 		= ts.tv_nsec;
121 	nnode->tn_uid = uid;
122 	nnode->tn_gid = gid;
123 	nnode->tn_mode = mode;
124 	nnode->tn_id = tmpfs_fetch_ino(tmp);
125 	nnode->tn_advlock.init_done = 0;
126 	KKASSERT(nnode->tn_links == 0);
127 
128 	/* Type-specific initialization. */
129 	switch (nnode->tn_type) {
130 	case VBLK:
131 	case VCHR:
132 		rdev = makeudev(rmajor, rminor);
133 		if (rdev == NOUDEV) {
134 			objcache_put(tmp->tm_node_pool, nnode);
135 			return(EINVAL);
136 		}
137 		nnode->tn_rdev = rdev;
138 		break;
139 
140 	case VDIR:
141 		RB_INIT(&nnode->tn_dir.tn_dirtree);
142 		RB_INIT(&nnode->tn_dir.tn_cookietree);
143 		nnode->tn_dir.tn_parent = NULL;
144 		nnode->tn_size = 0;
145 		break;
146 
147 	case VFIFO:
148 		/* FALLTHROUGH */
149 	case VSOCK:
150 		break;
151 
152 	case VLNK:
153 		nnode->tn_size = strlen(target);
154 		nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone,
155 					 M_WAITOK | M_NULLOK);
156 		if (nnode->tn_link == NULL) {
157 			objcache_put(tmp->tm_node_pool, nnode);
158 			return (ENOSPC);
159 		}
160 		bcopy(target, nnode->tn_link, nnode->tn_size);
161 		nnode->tn_link[nnode->tn_size] = '\0';
162 		break;
163 
164 	case VREG:
165 		nnode->tn_reg.tn_aobj = swap_pager_alloc(NULL, 0,
166 							 VM_PROT_DEFAULT, 0);
167 		nnode->tn_reg.tn_aobj_pages = 0;
168 		nnode->tn_size = 0;
169 		vm_object_set_flag(nnode->tn_reg.tn_aobj, OBJ_NOPAGEIN);
170 		break;
171 
172 	default:
173 		panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
174 	}
175 
176 	TMPFS_NODE_LOCK(nnode);
177 	TMPFS_LOCK(tmp);
178 	LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
179 	tmp->tm_nodes_inuse++;
180 	TMPFS_UNLOCK(tmp);
181 	TMPFS_NODE_UNLOCK(nnode);
182 
183 	*node = nnode;
184 	return 0;
185 }
186 
187 /* --------------------------------------------------------------------- */
188 
189 /*
190  * Destroys the node pointed to by node from the file system 'tmp'.
191  * If the node does not belong to the given mount point, the results are
192  * unpredicted.
193  *
194  * If the node references a directory; no entries are allowed because
195  * their removal could need a recursive algorithm, something forbidden in
196  * kernel space.  Furthermore, there is not need to provide such
197  * functionality (recursive removal) because the only primitives offered
198  * to the user are the removal of empty directories and the deletion of
199  * individual files.
200  *
201  * Note that nodes are not really deleted; in fact, when a node has been
202  * allocated, it cannot be deleted during the whole life of the file
203  * system.  Instead, they are moved to the available list and remain there
204  * until reused.
205  *
206  * A caller must have TMPFS_NODE_LOCK(node) and this function unlocks it.
207  */
208 void
209 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
210 {
211 	vm_pindex_t pages = 0;
212 
213 #ifdef INVARIANTS
214 	TMPFS_ASSERT_ELOCKED(node);
215 	KKASSERT(node->tn_vnode == NULL);
216 #endif
217 
218 	TMPFS_LOCK(tmp);
219 	LIST_REMOVE(node, tn_entries);
220 	tmp->tm_nodes_inuse--;
221 	TMPFS_UNLOCK(tmp);
222 	TMPFS_NODE_UNLOCK(node);  /* Caller has this lock */
223 
224 	switch (node->tn_type) {
225 	case VNON:
226 		/* Do not do anything.  VNON is provided to let the
227 		 * allocation routine clean itself easily by avoiding
228 		 * duplicating code in it. */
229 		/* FALLTHROUGH */
230 	case VBLK:
231 		/* FALLTHROUGH */
232 	case VCHR:
233 		/* FALLTHROUGH */
234 		break;
235 	case VDIR:
236 		/*
237 		 * The parent link can be NULL if this is the root
238 		 * node or if it is a directory node that was rmdir'd.
239 		 *
240 		 * XXX what if node is a directory which still contains
241 		 * directory entries (e.g. due to a forced umount) ?
242 		 */
243 		node->tn_size = 0;
244 		KKASSERT(node->tn_dir.tn_parent == NULL);
245 
246 		/*
247 		 * If the root node is being destroyed don't leave a
248 		 * dangling pointer in tmpfs_mount.
249 		 */
250 		if (node == tmp->tm_root)
251 			tmp->tm_root = NULL;
252 		break;
253 	case VFIFO:
254 		/* FALLTHROUGH */
255 	case VSOCK:
256 		break;
257 
258 	case VLNK:
259 		kfree(node->tn_link, tmp->tm_name_zone);
260 		node->tn_link = NULL;
261 		node->tn_size = 0;
262 		break;
263 
264 	case VREG:
265 		if (node->tn_reg.tn_aobj != NULL)
266 			vm_object_deallocate(node->tn_reg.tn_aobj);
267 		node->tn_reg.tn_aobj = NULL;
268 		pages = node->tn_reg.tn_aobj_pages;
269 		break;
270 
271 	default:
272 		panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
273 	}
274 
275 	/*
276 	 * Clean up fields for the next allocation.  The objcache only ctors
277 	 * new allocations.
278 	 */
279 	tmpfs_node_ctor(node, NULL, 0);
280 	objcache_put(tmp->tm_node_pool, node);
281 	/* node is now invalid */
282 
283 	if (pages)
284 		atomic_add_long(&tmp->tm_pages_used, -(long)pages);
285 }
286 
287 /* --------------------------------------------------------------------- */
288 
289 /*
290  * Allocates a new directory entry for the node node with a name of name.
291  * The new directory entry is returned in *de.
292  *
293  * The link count of node is increased by one to reflect the new object
294  * referencing it.
295  *
296  * Returns zero on success or an appropriate error code on failure.
297  */
298 int
299 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
300 		   const char *name, uint16_t len, struct tmpfs_dirent **de)
301 {
302 	struct tmpfs_dirent *nde;
303 
304 	nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK);
305 	nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK);
306 	if (nde->td_name == NULL) {
307 		objcache_put(tmp->tm_dirent_pool, nde);
308 		*de = NULL;
309 		return (ENOSPC);
310 	}
311 	nde->td_namelen = len;
312 	bcopy(name, nde->td_name, len);
313 	nde->td_name[len] = '\0';
314 
315 	nde->td_node = node;
316 
317 	atomic_add_int(&node->tn_links, 1);
318 
319 	*de = nde;
320 
321 	return 0;
322 }
323 
324 /* --------------------------------------------------------------------- */
325 
326 /*
327  * Frees a directory entry.  It is the caller's responsibility to destroy
328  * the node referenced by it if needed.
329  *
330  * The link count of node is decreased by one to reflect the removal of an
331  * object that referenced it.  This only happens if 'node_exists' is true;
332  * otherwise the function will not access the node referred to by the
333  * directory entry, as it may already have been released from the outside.
334  */
335 void
336 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
337 {
338 	struct tmpfs_node *node;
339 
340 	node = de->td_node;
341 
342 	KKASSERT(node->tn_links > 0);
343 	atomic_add_int(&node->tn_links, -1);
344 
345 	kfree(de->td_name, tmp->tm_name_zone);
346 	de->td_namelen = 0;
347 	de->td_name = NULL;
348 	de->td_node = NULL;
349 	objcache_put(tmp->tm_dirent_pool, de);
350 }
351 
352 /* --------------------------------------------------------------------- */
353 
354 /*
355  * Allocates a new vnode for the node node or returns a new reference to
356  * an existing one if the node had already a vnode referencing it.  The
357  * resulting locked vnode is returned in *vpp.
358  *
359  * Returns zero on success or an appropriate error code on failure.
360  *
361  * The caller must ensure that node cannot go away (usually by holding
362  * the related directory entry).
363  *
364  * If dnode is non-NULL this routine avoids deadlocking against it but
365  * can return EAGAIN.  Caller must try again.  The dnode lock will cycle
366  * in this case, it remains locked on return in all cases.  dnode must
367  * be shared-locked.
368  */
369 int
370 tmpfs_alloc_vp(struct mount *mp,
371 	       struct tmpfs_node *dnode, struct tmpfs_node *node, int lkflag,
372 	       struct vnode **vpp)
373 {
374 	int error = 0;
375 	struct vnode *vp;
376 
377 loop:
378 	vp = NULL;
379 	if (node->tn_vnode == NULL) {
380 		error = getnewvnode(VT_TMPFS, mp, &vp,
381 				    VLKTIMEOUT, LK_CANRECURSE);
382 		if (error)
383 			goto out;
384 	}
385 
386 	/*
387 	 * Interlocked extraction from node.  This can race many things.
388 	 * We have to get a soft reference on the vnode while we hold
389 	 * the node locked, then acquire it properly and check for races.
390 	 */
391 	TMPFS_NODE_LOCK(node);
392 	if (node->tn_vnode) {
393 		if (vp) {
394 			vp->v_type = VBAD;
395 			vx_put(vp);
396 		}
397 		vp = node->tn_vnode;
398 
399 		KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
400 		vhold(vp);
401 		TMPFS_NODE_UNLOCK(node);
402 
403 		if (dnode) {
404 			/*
405 			 * Special-case handling to avoid deadlocking against
406 			 * dnode.  This case has been validated and occurs
407 			 * every so often during synth builds.
408 			 */
409 			if (vget(vp, (lkflag & ~LK_RETRY) |
410 				     LK_NOWAIT |
411 				     LK_EXCLUSIVE) != 0) {
412 				TMPFS_NODE_UNLOCK(dnode);
413 				if (vget(vp, (lkflag & ~LK_RETRY) |
414 					     LK_SLEEPFAIL |
415 					     LK_EXCLUSIVE) == 0) {
416 					vn_unlock(vp);
417 				}
418 				vdrop(vp);
419 				TMPFS_NODE_LOCK_SH(dnode);
420 
421 				return EAGAIN;
422 			}
423 		} else {
424 			/*
425 			 * Normal path
426 			 */
427 			if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
428 				vdrop(vp);
429 				goto loop;
430 			}
431 		}
432 		if (node->tn_vnode != vp) {
433 			vput(vp);
434 			vdrop(vp);
435 			goto loop;
436 		}
437 		vdrop(vp);
438 		goto out;
439 	}
440 
441 	/*
442 	 * We need to assign node->tn_vnode.  If vp is NULL, loop up to
443 	 * allocate the vp.  This can happen due to SMP races.
444 	 */
445 	if (vp == NULL) {
446 		TMPFS_NODE_UNLOCK(node);
447 		goto loop;
448 	}
449 
450 	/*
451 	 * This should never happen.
452 	 */
453 	if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
454 		TMPFS_NODE_UNLOCK(node);
455 		vp->v_type = VBAD;
456 		vx_put(vp);
457 		error = ENOENT;
458 		goto out;
459 	}
460 
461 	KKASSERT(node->tn_vnode == NULL);
462 	KKASSERT(vp != NULL);
463 	vp->v_data = node;
464 	vp->v_type = node->tn_type;
465 
466 	/* Type-specific initialization. */
467 	switch (node->tn_type) {
468 	case VBLK:
469 		/* FALLTHROUGH */
470 	case VCHR:
471 		/* FALLTHROUGH */
472 	case VSOCK:
473 		break;
474 	case VREG:
475 		/*
476 		 * VMIO is mandatory.  Tmpfs also supports KVABIO
477 		 * for its tmpfs_strategy().
478 		 */
479 		vsetflags(vp, VKVABIO);
480 		vinitvmio(vp, node->tn_size, node->tn_blksize, -1);
481 		break;
482 	case VLNK:
483 		break;
484 	case VFIFO:
485 		vp->v_ops = &mp->mnt_vn_fifo_ops;
486 		break;
487 	case VDIR:
488 		break;
489 
490 	default:
491 		panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
492 	}
493 
494 	node->tn_vnode = vp;
495 	TMPFS_NODE_UNLOCK(node);
496 
497 	vx_downgrade(vp);
498 out:
499 	*vpp = vp;
500 	KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
501 
502 	return error;
503 }
504 
505 /* --------------------------------------------------------------------- */
506 
507 /*
508  * Allocates a new file of type 'type' and adds it to the parent directory
509  * 'dvp'; this addition is done using the component name given in 'cnp'.
510  * The ownership of the new file is automatically assigned based on the
511  * credentials of the caller (through 'cnp'), the group is set based on
512  * the parent directory and the mode is determined from the 'vap' argument.
513  * If successful, *vpp holds a vnode to the newly created file and zero
514  * is returned.  Otherwise *vpp is NULL and the function returns an
515  * appropriate error code.
516  */
517 int
518 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
519 		 struct namecache *ncp, struct ucred *cred, char *target)
520 {
521 	int error;
522 	struct tmpfs_dirent *de;
523 	struct tmpfs_mount *tmp;
524 	struct tmpfs_node *dnode;
525 	struct tmpfs_node *node;
526 
527 	tmp = VFS_TO_TMPFS(dvp->v_mount);
528 	dnode = VP_TO_TMPFS_DIR(dvp);
529 	*vpp = NULL;
530 
531 	/*
532 	 * If the directory was removed but a process was CD'd into it,
533 	 * we do not allow any more file/dir creation within it.  Otherwise
534 	 * we will lose track of it.
535 	 */
536 	KKASSERT(dnode->tn_type == VDIR);
537 	if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL)
538 		return ENOENT;
539 
540 	/*
541 	 * Make sure the link count does not overflow.
542 	 */
543 	if (vap->va_type == VDIR && dnode->tn_links >= LINK_MAX)
544 		return EMLINK;
545 
546 	/* Allocate a node that represents the new file. */
547 	error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
548 				 dnode->tn_gid, vap->va_mode, target,
549 				 vap->va_rmajor, vap->va_rminor, &node);
550 	if (error != 0)
551 		return error;
552 	TMPFS_NODE_LOCK(node);
553 
554 	/* Allocate a directory entry that points to the new file. */
555 	error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
556 	if (error != 0) {
557 		tmpfs_free_node(tmp, node);
558 		/* eats node lock */
559 		return error;
560 	}
561 
562 	/* Allocate a vnode for the new file. */
563 	error = tmpfs_alloc_vp(dvp->v_mount, NULL, node, LK_EXCLUSIVE, vpp);
564 	if (error != 0) {
565 		tmpfs_free_dirent(tmp, de);
566 		tmpfs_free_node(tmp, node);
567 		/* eats node lock */
568 		return error;
569 	}
570 
571 	/*
572 	 * Now that all required items are allocated, we can proceed to
573 	 * insert the new node into the directory, an operation that
574 	 * cannot fail.
575 	 */
576 	tmpfs_dir_attach(dnode, de);
577 	TMPFS_NODE_UNLOCK(node);
578 
579 	return error;
580 }
581 
582 /* --------------------------------------------------------------------- */
583 
584 /*
585  * Attaches the directory entry de to the directory represented by dnode.
586  * Note that this does not change the link count of the node pointed by
587  * the directory entry, as this is done by tmpfs_alloc_dirent.
588  */
589 void
590 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
591 {
592 	struct tmpfs_node *node = de->td_node;
593 
594 	TMPFS_NODE_LOCK(dnode);
595 	if (node && node->tn_type == VDIR) {
596 		TMPFS_NODE_LOCK(node);
597 		atomic_add_int(&node->tn_links, 1);
598 		node->tn_status |= TMPFS_NODE_CHANGED;
599 		node->tn_dir.tn_parent = dnode;
600 		atomic_add_int(&dnode->tn_links, 1);
601 		TMPFS_NODE_UNLOCK(node);
602 	}
603 	RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
604 	RB_INSERT(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de);
605 	dnode->tn_size += sizeof(struct tmpfs_dirent);
606 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
607 			    TMPFS_NODE_MODIFIED;
608 	TMPFS_NODE_UNLOCK(dnode);
609 }
610 
611 /* --------------------------------------------------------------------- */
612 
613 /*
614  * Detaches the directory entry de from the directory represented by dnode.
615  * Note that this does not change the link count of the node pointed by
616  * the directory entry, as this is done by tmpfs_free_dirent.
617  */
618 void
619 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
620 {
621 	struct tmpfs_node *node = de->td_node;
622 
623 	TMPFS_NODE_LOCK(dnode);
624 	RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
625 	RB_REMOVE(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de);
626 	dnode->tn_size -= sizeof(struct tmpfs_dirent);
627 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
628 			    TMPFS_NODE_MODIFIED;
629 	TMPFS_NODE_UNLOCK(dnode);
630 
631 	/*
632 	 * Clean out the tn_parent pointer immediately when removing a
633 	 * directory.
634 	 *
635 	 * Removal of the parent linkage also cleans out the extra tn_links
636 	 * count we had on both node and dnode.
637 	 *
638 	 * node can be NULL (typ during a forced umount), in which case
639 	 * the mount code is dealing with the linkages from a linked list
640 	 * scan.
641 	 */
642 	if (node && node->tn_type == VDIR && node->tn_dir.tn_parent) {
643 		TMPFS_NODE_LOCK(dnode);
644 		TMPFS_NODE_LOCK(node);
645 		KKASSERT(node->tn_dir.tn_parent == dnode);
646 		atomic_add_int(&dnode->tn_links, -1);
647 		atomic_add_int(&node->tn_links, -1);
648 		node->tn_dir.tn_parent = NULL;
649 		TMPFS_NODE_UNLOCK(node);
650 		TMPFS_NODE_UNLOCK(dnode);
651 	}
652 }
653 
654 /* --------------------------------------------------------------------- */
655 
656 /*
657  * Looks for a directory entry in the directory represented by node.
658  * 'ncp' describes the name of the entry to look for.  Note that the .
659  * and .. components are not allowed as they do not physically exist
660  * within directories.
661  *
662  * Returns a pointer to the entry when found, otherwise NULL.
663  *
664  * Caller must hold the node locked (shared ok)
665  */
666 struct tmpfs_dirent *
667 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
668 		 struct namecache *ncp)
669 {
670 	struct tmpfs_dirent *de;
671 	int len = ncp->nc_nlen;
672 	struct tmpfs_dirent wanted;
673 
674 	wanted.td_namelen = len;
675 	wanted.td_name = ncp->nc_name;
676 
677 	TMPFS_VALIDATE_DIR(node);
678 
679 	de = RB_FIND(tmpfs_dirtree, &node->tn_dir.tn_dirtree, &wanted);
680 
681 	KKASSERT(f == NULL || de == NULL || f == de->td_node);
682 
683 	return de;
684 }
685 
686 /* --------------------------------------------------------------------- */
687 
688 /*
689  * Helper function for tmpfs_readdir.  Creates a '.' entry for the given
690  * directory and returns it in the uio space.  The function returns 0
691  * on success, -1 if there was not enough space in the uio structure to
692  * hold the directory entry or an appropriate error code if another
693  * error happens.
694  */
695 int
696 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
697 {
698 	int error;
699 
700 	TMPFS_VALIDATE_DIR(node);
701 	KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
702 
703 	if (vop_write_dirent(&error, uio, node->tn_id, DT_DIR, 1, "."))
704 		return -1;
705 	if (error == 0)
706 		uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
707 	return error;
708 }
709 
710 /* --------------------------------------------------------------------- */
711 
712 /*
713  * Helper function for tmpfs_readdir.  Creates a '..' entry for the given
714  * directory and returns it in the uio space.  The function returns 0
715  * on success, -1 if there was not enough space in the uio structure to
716  * hold the directory entry or an appropriate error code if another
717  * error happens.
718  */
719 int
720 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
721 			struct uio *uio)
722 {
723 	int error;
724 	ino_t d_ino;
725 
726 	TMPFS_VALIDATE_DIR(node);
727 	KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
728 
729 	if (node->tn_dir.tn_parent) {
730 		TMPFS_NODE_LOCK(node);
731 		if (node->tn_dir.tn_parent)
732 			d_ino = node->tn_dir.tn_parent->tn_id;
733 		else
734 			d_ino = tmp->tm_root->tn_id;
735 		TMPFS_NODE_UNLOCK(node);
736 	} else {
737 		d_ino = tmp->tm_root->tn_id;
738 	}
739 
740 	if (vop_write_dirent(&error, uio, d_ino, DT_DIR, 2, ".."))
741 		return -1;
742 	if (error == 0) {
743 		struct tmpfs_dirent *de;
744 		de = RB_MIN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree);
745 		if (de == NULL)
746 			uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
747 		else
748 			uio->uio_offset = tmpfs_dircookie(de);
749 	}
750 	return error;
751 }
752 
753 /* --------------------------------------------------------------------- */
754 
755 /*
756  * Lookup a directory entry by its associated cookie.
757  *
758  * Must be called with the directory node locked (shared ok)
759  */
760 struct lubycookie_info {
761 	off_t	cookie;
762 	struct tmpfs_dirent *de;
763 };
764 
765 static int
766 lubycookie_cmp(struct tmpfs_dirent *de, void *arg)
767 {
768 	struct lubycookie_info *info = arg;
769 	off_t cookie = tmpfs_dircookie(de);
770 
771 	if (cookie < info->cookie)
772 		return(-1);
773 	if (cookie > info->cookie)
774 		return(1);
775 	return(0);
776 }
777 
778 static int
779 lubycookie_callback(struct tmpfs_dirent *de, void *arg)
780 {
781 	struct lubycookie_info *info = arg;
782 
783 	if (tmpfs_dircookie(de) == info->cookie) {
784 		info->de = de;
785 		return(-1);
786 	}
787 	return(0);
788 }
789 
790 struct tmpfs_dirent *
791 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
792 {
793 	struct lubycookie_info info;
794 
795 	info.cookie = cookie;
796 	info.de = NULL;
797 	RB_SCAN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree,
798 		lubycookie_cmp, lubycookie_callback, &info);
799 	return (info.de);
800 }
801 
802 /* --------------------------------------------------------------------- */
803 
804 /*
805  * Helper function for tmpfs_readdir.  Returns as much directory entries
806  * as can fit in the uio space.  The read starts at uio->uio_offset.
807  * The function returns 0 on success, -1 if there was not enough space
808  * in the uio structure to hold the directory entry or an appropriate
809  * error code if another error happens.
810  *
811  * Caller must hold the node locked (shared ok)
812  */
813 int
814 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
815 {
816 	int error;
817 	off_t startcookie;
818 	struct tmpfs_dirent *de;
819 
820 	TMPFS_VALIDATE_DIR(node);
821 
822 	/*
823 	 * Locate the first directory entry we have to return.  We have cached
824 	 * the last readdir in the node, so use those values if appropriate.
825 	 * Otherwise do a linear scan to find the requested entry.
826 	 */
827 	startcookie = uio->uio_offset;
828 	KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
829 	KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
830 
831 	if (startcookie == TMPFS_DIRCOOKIE_EOF)
832 		return 0;
833 
834 	de = tmpfs_dir_lookupbycookie(node, startcookie);
835 	if (de == NULL)
836 		return EINVAL;
837 
838 	/*
839 	 * Read as much entries as possible; i.e., until we reach the end of
840 	 * the directory or we exhaust uio space.
841 	 */
842 	do {
843 		ino_t d_ino;
844 		uint8_t d_type;
845 
846 		/* Create a dirent structure representing the current
847 		 * tmpfs_node and fill it. */
848 		d_ino = de->td_node->tn_id;
849 		switch (de->td_node->tn_type) {
850 		case VBLK:
851 			d_type = DT_BLK;
852 			break;
853 
854 		case VCHR:
855 			d_type = DT_CHR;
856 			break;
857 
858 		case VDIR:
859 			d_type = DT_DIR;
860 			break;
861 
862 		case VFIFO:
863 			d_type = DT_FIFO;
864 			break;
865 
866 		case VLNK:
867 			d_type = DT_LNK;
868 			break;
869 
870 		case VREG:
871 			d_type = DT_REG;
872 			break;
873 
874 		case VSOCK:
875 			d_type = DT_SOCK;
876 			break;
877 
878 		default:
879 			panic("tmpfs_dir_getdents: type %p %d",
880 			    de->td_node, (int)de->td_node->tn_type);
881 		}
882 		KKASSERT(de->td_namelen < 256); /* 255 + 1 */
883 
884 		if (vop_write_dirent(&error, uio, d_ino, d_type,
885 		    de->td_namelen, de->td_name)) {
886 			error = -1;
887 			break;
888 		}
889 
890 		(*cntp)++;
891 		de = RB_NEXT(tmpfs_dirtree_cookie,
892 			     node->tn_dir.tn_cookietree, de);
893 	} while (error == 0 && uio->uio_resid > 0 && de != NULL);
894 
895 	/* Update the offset and cache. */
896 	if (de == NULL) {
897 		uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
898 	} else {
899 		uio->uio_offset = tmpfs_dircookie(de);
900 	}
901 
902 	return error;
903 }
904 
905 /* --------------------------------------------------------------------- */
906 
907 /*
908  * Resizes the aobj associated to the regular file pointed to by vp to
909  * the size newsize.  'vp' must point to a vnode that represents a regular
910  * file.  'newsize' must be positive.
911  *
912  * pass NVEXTF_TRIVIAL when buf content will be overwritten, otherwise set 0
913  * to be zero filled.
914  *
915  * Returns zero on success or an appropriate error code on failure.
916  *
917  * Caller must hold the node exclusively locked.
918  */
919 int
920 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
921 {
922 	int error;
923 	vm_pindex_t newpages, oldpages;
924 	struct tmpfs_mount *tmp;
925 	struct tmpfs_node *node;
926 	off_t oldsize;
927 	int nvextflags;
928 
929 #ifdef INVARIANTS
930 	KKASSERT(vp->v_type == VREG);
931 	KKASSERT(newsize >= 0);
932 #endif
933 
934 	node = VP_TO_TMPFS_NODE(vp);
935 	tmp = VFS_TO_TMPFS(vp->v_mount);
936 
937 	/*
938 	 * Convert the old and new sizes to the number of pages needed to
939 	 * store them.  It may happen that we do not need to do anything
940 	 * because the last allocated page can accommodate the change on
941 	 * its own.
942 	 */
943 	oldsize = node->tn_size;
944 	oldpages = round_page64(oldsize) / PAGE_SIZE;
945 	KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
946 	newpages = round_page64(newsize) / PAGE_SIZE;
947 
948 	if (newpages > oldpages &&
949 	   tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
950 		error = ENOSPC;
951 		goto out;
952 	}
953 	node->tn_reg.tn_aobj_pages = newpages;
954 	node->tn_size = newsize;
955 
956 	if (newpages != oldpages)
957 		atomic_add_long(&tmp->tm_pages_used, (newpages - oldpages));
958 
959 	/*
960 	 * nvextflags to pass along for bdwrite() vs buwrite()
961 	 */
962 	if (vm_pages_needed || vm_paging_needed(0) ||
963 	    tmpfs_bufcache_mode >= 2) {
964 		nvextflags = 0;
965 	} else {
966 		nvextflags = NVEXTF_BUWRITE;
967 	}
968 
969 
970 	/*
971 	 * When adjusting the vnode filesize and its VM object we must
972 	 * also adjust our backing VM object (aobj).  The blocksize
973 	 * used must match the block sized we use for the buffer cache.
974 	 *
975 	 * The backing VM object may contain VM pages as well as swap
976 	 * assignments if we previously renamed main object pages into
977 	 * it during deactivation.
978 	 *
979 	 * To make things easier tmpfs uses a blksize in multiples of
980 	 * PAGE_SIZE, and will only increase the blksize as a small file
981 	 * increases in size.  Once a file has exceeded TMPFS_BLKSIZE (16KB),
982 	 * the blksize is maxed out.  Truncating the file does not reduce
983 	 * the blksize.
984 	 */
985 	if (newsize < oldsize) {
986 		vm_pindex_t osize;
987 		vm_pindex_t nsize;
988 		vm_object_t aobj;
989 
990 		error = nvtruncbuf(vp, newsize, node->tn_blksize,
991 				   -1, nvextflags);
992 		aobj = node->tn_reg.tn_aobj;
993 		if (aobj) {
994 			osize = aobj->size;
995 			nsize = vp->v_object->size;
996 			if (nsize < osize) {
997 				aobj->size = osize;
998 				swap_pager_freespace(aobj, nsize,
999 						     osize - nsize);
1000 				vm_object_page_remove(aobj, nsize, osize,
1001 						      FALSE);
1002 			}
1003 		}
1004 	} else {
1005 		vm_object_t aobj;
1006 		int nblksize;
1007 
1008 		/*
1009 		 * The first (and only the first) buffer in the file is resized
1010 		 * in multiples of PAGE_SIZE, up to TMPFS_BLKSIZE.
1011 		 */
1012 		nblksize = node->tn_blksize;
1013 		while (nblksize < TMPFS_BLKSIZE &&
1014 		       nblksize < newsize) {
1015 			nblksize += PAGE_SIZE;
1016 		}
1017 
1018 		if (trivial)
1019 			nvextflags |= NVEXTF_TRIVIAL;
1020 
1021 		error = nvextendbuf(vp, oldsize, newsize,
1022 				    node->tn_blksize, nblksize,
1023 				    -1, -1, nvextflags);
1024 		node->tn_blksize = nblksize;
1025 		aobj = node->tn_reg.tn_aobj;
1026 		if (aobj)
1027 			aobj->size = vp->v_object->size;
1028 	}
1029 
1030 out:
1031 	return error;
1032 }
1033 
1034 /* --------------------------------------------------------------------- */
1035 
1036 /*
1037  * Change flags of the given vnode.
1038  * Caller should execute tmpfs_update on vp after a successful execution.
1039  * The vnode must be locked on entry and remain locked on exit.
1040  */
1041 int
1042 tmpfs_chflags(struct vnode *vp, u_long vaflags, struct ucred *cred)
1043 {
1044 	int error;
1045 	struct tmpfs_node *node;
1046 	int flags;
1047 
1048 	KKASSERT(vn_islocked(vp));
1049 
1050 	node = VP_TO_TMPFS_NODE(vp);
1051 	flags = node->tn_flags;
1052 
1053 	/* Disallow this operation if the file system is mounted read-only. */
1054 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1055 		return EROFS;
1056 	error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred);
1057 
1058 	/* Actually change the flags on the node itself */
1059 	if (error == 0) {
1060 		TMPFS_NODE_LOCK(node);
1061 		node->tn_flags = flags;
1062 		node->tn_status |= TMPFS_NODE_CHANGED;
1063 		TMPFS_NODE_UNLOCK(node);
1064 	}
1065 
1066 	KKASSERT(vn_islocked(vp));
1067 
1068 	return error;
1069 }
1070 
1071 /* --------------------------------------------------------------------- */
1072 
1073 /*
1074  * Change access mode on the given vnode.
1075  * Caller should execute tmpfs_update on vp after a successful execution.
1076  * The vnode must be locked on entry and remain locked on exit.
1077  */
1078 int
1079 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred)
1080 {
1081 	struct tmpfs_node *node;
1082 	mode_t cur_mode;
1083 	int error;
1084 
1085 	KKASSERT(vn_islocked(vp));
1086 
1087 	node = VP_TO_TMPFS_NODE(vp);
1088 
1089 	/* Disallow this operation if the file system is mounted read-only. */
1090 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1091 		return EROFS;
1092 
1093 	/* Immutable or append-only files cannot be modified, either. */
1094 	if (node->tn_flags & (IMMUTABLE | APPEND))
1095 		return EPERM;
1096 
1097 	cur_mode = node->tn_mode;
1098 	error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid,
1099 				 &cur_mode);
1100 
1101 	if (error == 0 &&
1102 	    (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) {
1103 		TMPFS_NODE_LOCK(node);
1104 		node->tn_mode &= ~ALLPERMS;
1105 		node->tn_mode |= cur_mode & ALLPERMS;
1106 
1107 		node->tn_status |= TMPFS_NODE_CHANGED;
1108 		TMPFS_NODE_UNLOCK(node);
1109 	}
1110 
1111 	KKASSERT(vn_islocked(vp));
1112 
1113 	return 0;
1114 }
1115 
1116 /* --------------------------------------------------------------------- */
1117 
1118 /*
1119  * Change ownership of the given vnode.  At least one of uid or gid must
1120  * be different than VNOVAL.  If one is set to that value, the attribute
1121  * is unchanged.
1122  * Caller should execute tmpfs_update on vp after a successful execution.
1123  * The vnode must be locked on entry and remain locked on exit.
1124  */
1125 int
1126 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1127 {
1128 	mode_t cur_mode;
1129 	uid_t cur_uid;
1130 	gid_t cur_gid;
1131 	struct tmpfs_node *node;
1132 	int error;
1133 
1134 	KKASSERT(vn_islocked(vp));
1135 	node = VP_TO_TMPFS_NODE(vp);
1136 
1137 	/* Disallow this operation if the file system is mounted read-only. */
1138 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1139 		return EROFS;
1140 
1141 	/* Immutable or append-only files cannot be modified, either. */
1142 	if (node->tn_flags & (IMMUTABLE | APPEND))
1143 		return EPERM;
1144 
1145 	cur_uid = node->tn_uid;
1146 	cur_gid = node->tn_gid;
1147 	cur_mode = node->tn_mode;
1148 	error = vop_helper_chown(vp, uid, gid, cred,
1149 				 &cur_uid, &cur_gid, &cur_mode);
1150 
1151 	if (error == 0) {
1152 		TMPFS_NODE_LOCK(node);
1153 		if (cur_uid != node->tn_uid ||
1154 		    cur_gid != node->tn_gid ||
1155 		    cur_mode != node->tn_mode) {
1156 			node->tn_uid = cur_uid;
1157 			node->tn_gid = cur_gid;
1158 			node->tn_mode = cur_mode;
1159 			node->tn_status |= TMPFS_NODE_CHANGED;
1160 		}
1161 		TMPFS_NODE_UNLOCK(node);
1162 	}
1163 
1164 	return error;
1165 }
1166 
1167 /* --------------------------------------------------------------------- */
1168 
1169 /*
1170  * Change size of the given vnode.
1171  * Caller should execute tmpfs_update on vp after a successful execution.
1172  * The vnode must be locked on entry and remain locked on exit.
1173  */
1174 int
1175 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1176 {
1177 	int error;
1178 	struct tmpfs_node *node;
1179 
1180 	KKASSERT(vn_islocked(vp));
1181 
1182 	node = VP_TO_TMPFS_NODE(vp);
1183 
1184 	/* Decide whether this is a valid operation based on the file type. */
1185 	error = 0;
1186 	switch (vp->v_type) {
1187 	case VDIR:
1188 		return EISDIR;
1189 
1190 	case VREG:
1191 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
1192 			return EROFS;
1193 		break;
1194 
1195 	case VBLK:
1196 		/* FALLTHROUGH */
1197 	case VCHR:
1198 		/* FALLTHROUGH */
1199 	case VFIFO:
1200 		/* Allow modifications of special files even if in the file
1201 		 * system is mounted read-only (we are not modifying the
1202 		 * files themselves, but the objects they represent). */
1203 		return 0;
1204 
1205 	default:
1206 		/* Anything else is unsupported. */
1207 		return EOPNOTSUPP;
1208 	}
1209 
1210 	/* Immutable or append-only files cannot be modified, either. */
1211 	if (node->tn_flags & (IMMUTABLE | APPEND))
1212 		return EPERM;
1213 
1214 	error = tmpfs_truncate(vp, size);
1215 	/* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1216 	 * for us, as will update tn_status; no need to do that here. */
1217 
1218 	KKASSERT(vn_islocked(vp));
1219 
1220 	return error;
1221 }
1222 
1223 /* --------------------------------------------------------------------- */
1224 
1225 /*
1226  * Change access and modification times of the given vnode.
1227  * Caller should execute tmpfs_update on vp after a successful execution.
1228  * The vnode must be locked on entry and remain locked on exit.
1229  */
1230 int
1231 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1232 	      int vaflags, struct ucred *cred)
1233 {
1234 	struct tmpfs_node *node;
1235 
1236 	KKASSERT(vn_islocked(vp));
1237 
1238 	node = VP_TO_TMPFS_NODE(vp);
1239 
1240 	/* Disallow this operation if the file system is mounted read-only. */
1241 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1242 		return EROFS;
1243 
1244 	/* Immutable or append-only files cannot be modified, either. */
1245 	if (node->tn_flags & (IMMUTABLE | APPEND))
1246 		return EPERM;
1247 
1248 	TMPFS_NODE_LOCK(node);
1249 	if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1250 		node->tn_status |= TMPFS_NODE_ACCESSED;
1251 
1252 	if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) {
1253 		node->tn_status |= TMPFS_NODE_MODIFIED;
1254 		vclrflags(vp, VLASTWRITETS);
1255 	}
1256 
1257 	TMPFS_NODE_UNLOCK(node);
1258 
1259 	tmpfs_itimes(vp, atime, mtime);
1260 
1261 	KKASSERT(vn_islocked(vp));
1262 
1263 	return 0;
1264 }
1265 
1266 /* --------------------------------------------------------------------- */
1267 /* Sync timestamps */
1268 void
1269 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1270 	     const struct timespec *mod)
1271 {
1272 	struct tmpfs_node *node;
1273 	struct timespec now;
1274 
1275 	node = VP_TO_TMPFS_NODE(vp);
1276 
1277 	if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1278 	    TMPFS_NODE_CHANGED)) == 0) {
1279 		return;
1280 	}
1281 
1282 	vfs_timestamp(&now);
1283 
1284 	TMPFS_NODE_LOCK(node);
1285 	if (node->tn_status & TMPFS_NODE_ACCESSED) {
1286 		if (acc == NULL)
1287 			 acc = &now;
1288 		node->tn_atime = acc->tv_sec;
1289 		node->tn_atimensec = acc->tv_nsec;
1290 	}
1291 	if (node->tn_status & TMPFS_NODE_MODIFIED) {
1292 		if (mod == NULL)
1293 			mod = &now;
1294 		node->tn_mtime = mod->tv_sec;
1295 		node->tn_mtimensec = mod->tv_nsec;
1296 	}
1297 	if (node->tn_status & TMPFS_NODE_CHANGED) {
1298 		node->tn_ctime = now.tv_sec;
1299 		node->tn_ctimensec = now.tv_nsec;
1300 	}
1301 
1302 	node->tn_status &= ~(TMPFS_NODE_ACCESSED |
1303 			     TMPFS_NODE_MODIFIED |
1304 			     TMPFS_NODE_CHANGED);
1305 	TMPFS_NODE_UNLOCK(node);
1306 }
1307 
1308 /* --------------------------------------------------------------------- */
1309 
1310 void
1311 tmpfs_update(struct vnode *vp)
1312 {
1313 	tmpfs_itimes(vp, NULL, NULL);
1314 }
1315 
1316 /* --------------------------------------------------------------------- */
1317 
1318 /*
1319  * Caller must hold an exclusive node lock.
1320  */
1321 int
1322 tmpfs_truncate(struct vnode *vp, off_t length)
1323 {
1324 	int error;
1325 	struct tmpfs_node *node;
1326 
1327 	node = VP_TO_TMPFS_NODE(vp);
1328 
1329 	if (length < 0) {
1330 		error = EINVAL;
1331 		goto out;
1332 	}
1333 
1334 	if (node->tn_size == length) {
1335 		error = 0;
1336 		goto out;
1337 	}
1338 
1339 	if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1340 		return (EFBIG);
1341 
1342 
1343 	error = tmpfs_reg_resize(vp, length, 1);
1344 
1345 	if (error == 0)
1346 		node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1347 
1348 out:
1349 	tmpfs_update(vp);
1350 
1351 	return error;
1352 }
1353 
1354 /* --------------------------------------------------------------------- */
1355 
1356 static ino_t
1357 tmpfs_fetch_ino(struct tmpfs_mount *tmp)
1358 {
1359 	ino_t ret;
1360 
1361 	ret = atomic_fetchadd_64(&tmp->tm_ino, 1);
1362 
1363 	return (ret);
1364 }
1365 
1366 static int
1367 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1368 {
1369 	if (a->td_namelen > b->td_namelen)
1370 		return 1;
1371 	else if (a->td_namelen < b->td_namelen)
1372 		return -1;
1373 	else
1374 		return strncmp(a->td_name, b->td_name, a->td_namelen);
1375 }
1376 
1377 static int
1378 tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1379 {
1380 	if (a < b)
1381 		return(-1);
1382 	if (a > b)
1383 		return(1);
1384 	return 0;
1385 }
1386 
1387 /*
1388  * Lock for rename.  The namecache entries are already locked so
1389  * theoretically we should be able to lock the directories in any
1390  * order.  Underlying files must be locked after the related directory.
1391  */
1392 void
1393 tmpfs_lock4(struct tmpfs_node *node1, struct tmpfs_node *node2,
1394 	    struct tmpfs_node *node3, struct tmpfs_node *node4)
1395 {
1396 	TMPFS_NODE_LOCK(node1);		/* fdir */
1397 	TMPFS_NODE_LOCK(node3);		/* ffile */
1398 	TMPFS_NODE_LOCK(node2);		/* tdir */
1399 	if (node4)
1400 		TMPFS_NODE_LOCK(node4);	/* tfile */
1401 }
1402 
1403 void
1404 tmpfs_unlock4(struct tmpfs_node *node1, struct tmpfs_node *node2,
1405 	      struct tmpfs_node *node3, struct tmpfs_node *node4)
1406 {
1407 	if (node4)
1408 		TMPFS_NODE_UNLOCK(node4);
1409 	TMPFS_NODE_UNLOCK(node2);
1410 	TMPFS_NODE_UNLOCK(node3);
1411 	TMPFS_NODE_UNLOCK(node1);
1412 }
1413