xref: /freebsd/sys/fs/tmpfs/tmpfs_subr.c (revision 6b22f423cfa76dc95ed00e5a8f43e563cc34b9a6)
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/fnv_hash.h>
41 #include <sys/lock.h>
42 #include <sys/namei.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/rwlock.h>
46 #include <sys/stat.h>
47 #include <sys/systm.h>
48 #include <sys/sysctl.h>
49 #include <sys/vnode.h>
50 #include <sys/vmmeter.h>
51 
52 #include <vm/vm.h>
53 #include <vm/vm_param.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pageout.h>
57 #include <vm/vm_pager.h>
58 #include <vm/vm_extern.h>
59 
60 #include <fs/tmpfs/tmpfs.h>
61 #include <fs/tmpfs/tmpfs_fifoops.h>
62 #include <fs/tmpfs/tmpfs_vnops.h>
63 
64 struct tmpfs_dir_cursor {
65 	struct tmpfs_dirent	*tdc_current;
66 	struct tmpfs_dirent	*tdc_tree;
67 };
68 
69 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system");
70 
71 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
72 
73 static int
74 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
75 {
76 	int error;
77 	long pages, bytes;
78 
79 	pages = *(long *)arg1;
80 	bytes = pages * PAGE_SIZE;
81 
82 	error = sysctl_handle_long(oidp, &bytes, 0, req);
83 	if (error || !req->newptr)
84 		return (error);
85 
86 	pages = bytes / PAGE_SIZE;
87 	if (pages < TMPFS_PAGES_MINRESERVED)
88 		return (EINVAL);
89 
90 	*(long *)arg1 = pages;
91 	return (0);
92 }
93 
94 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW,
95     &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L",
96     "Amount of available memory and swap below which tmpfs growth stops");
97 
98 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
99     struct tmpfs_dirent *b);
100 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
101 
102 size_t
103 tmpfs_mem_avail(void)
104 {
105 	vm_ooffset_t avail;
106 
107 	avail = swap_pager_avail + vm_cnt.v_free_count + vm_cnt.v_cache_count -
108 	    tmpfs_pages_reserved;
109 	if (__predict_false(avail < 0))
110 		avail = 0;
111 	return (avail);
112 }
113 
114 size_t
115 tmpfs_pages_used(struct tmpfs_mount *tmp)
116 {
117 	const size_t node_size = sizeof(struct tmpfs_node) +
118 	    sizeof(struct tmpfs_dirent);
119 	size_t meta_pages;
120 
121 	meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
122 	    PAGE_SIZE);
123 	return (meta_pages + tmp->tm_pages_used);
124 }
125 
126 static size_t
127 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
128 {
129 	if (tmpfs_mem_avail() < req_pages)
130 		return (0);
131 
132 	if (tmp->tm_pages_max != SIZE_MAX &&
133 	    tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
134 			return (0);
135 
136 	return (1);
137 }
138 
139 /* --------------------------------------------------------------------- */
140 
141 /*
142  * Allocates a new node of type 'type' inside the 'tmp' mount point, with
143  * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
144  * using the credentials of the process 'p'.
145  *
146  * If the node type is set to 'VDIR', then the parent parameter must point
147  * to the parent directory of the node being created.  It may only be NULL
148  * while allocating the root node.
149  *
150  * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
151  * specifies the device the node represents.
152  *
153  * If the node type is set to 'VLNK', then the parameter target specifies
154  * the file name of the target file for the symbolic link that is being
155  * created.
156  *
157  * Note that new nodes are retrieved from the available list if it has
158  * items or, if it is empty, from the node pool as long as there is enough
159  * space to create them.
160  *
161  * Returns zero on success or an appropriate error code on failure.
162  */
163 int
164 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
165     uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
166     char *target, dev_t rdev, struct tmpfs_node **node)
167 {
168 	struct tmpfs_node *nnode;
169 	vm_object_t obj;
170 
171 	/* If the root directory of the 'tmp' file system is not yet
172 	 * allocated, this must be the request to do it. */
173 	MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
174 
175 	MPASS(IFF(type == VLNK, target != NULL));
176 	MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
177 
178 	if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
179 		return (ENOSPC);
180 	if (tmpfs_pages_check_avail(tmp, 1) == 0)
181 		return (ENOSPC);
182 
183 	nnode = (struct tmpfs_node *)uma_zalloc_arg(
184 				tmp->tm_node_pool, tmp, M_WAITOK);
185 
186 	/* Generic initialization. */
187 	nnode->tn_type = type;
188 	vfs_timestamp(&nnode->tn_atime);
189 	nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
190 	    nnode->tn_atime;
191 	nnode->tn_uid = uid;
192 	nnode->tn_gid = gid;
193 	nnode->tn_mode = mode;
194 	nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
195 
196 	/* Type-specific initialization. */
197 	switch (nnode->tn_type) {
198 	case VBLK:
199 	case VCHR:
200 		nnode->tn_rdev = rdev;
201 		break;
202 
203 	case VDIR:
204 		RB_INIT(&nnode->tn_dir.tn_dirhead);
205 		LIST_INIT(&nnode->tn_dir.tn_dupindex);
206 		MPASS(parent != nnode);
207 		MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
208 		nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
209 		nnode->tn_dir.tn_readdir_lastn = 0;
210 		nnode->tn_dir.tn_readdir_lastp = NULL;
211 		nnode->tn_links++;
212 		TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
213 		nnode->tn_dir.tn_parent->tn_links++;
214 		TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
215 		break;
216 
217 	case VFIFO:
218 		/* FALLTHROUGH */
219 	case VSOCK:
220 		break;
221 
222 	case VLNK:
223 		MPASS(strlen(target) < MAXPATHLEN);
224 		nnode->tn_size = strlen(target);
225 		nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
226 		    M_WAITOK);
227 		memcpy(nnode->tn_link, target, nnode->tn_size);
228 		break;
229 
230 	case VREG:
231 		obj = nnode->tn_reg.tn_aobj =
232 		    vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
233 			NULL /* XXXKIB - tmpfs needs swap reservation */);
234 		VM_OBJECT_WLOCK(obj);
235 		/* OBJ_TMPFS is set together with the setting of vp->v_object */
236 		vm_object_set_flag(obj, OBJ_NOSPLIT);
237 		vm_object_clear_flag(obj, OBJ_ONEMAPPING);
238 		VM_OBJECT_WUNLOCK(obj);
239 		break;
240 
241 	default:
242 		panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
243 	}
244 
245 	TMPFS_LOCK(tmp);
246 	LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
247 	tmp->tm_nodes_inuse++;
248 	TMPFS_UNLOCK(tmp);
249 
250 	*node = nnode;
251 	return 0;
252 }
253 
254 /* --------------------------------------------------------------------- */
255 
256 /*
257  * Destroys the node pointed to by node from the file system 'tmp'.
258  * If the node does not belong to the given mount point, the results are
259  * unpredicted.
260  *
261  * If the node references a directory; no entries are allowed because
262  * their removal could need a recursive algorithm, something forbidden in
263  * kernel space.  Furthermore, there is not need to provide such
264  * functionality (recursive removal) because the only primitives offered
265  * to the user are the removal of empty directories and the deletion of
266  * individual files.
267  *
268  * Note that nodes are not really deleted; in fact, when a node has been
269  * allocated, it cannot be deleted during the whole life of the file
270  * system.  Instead, they are moved to the available list and remain there
271  * until reused.
272  */
273 void
274 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
275 {
276 	vm_object_t uobj;
277 
278 #ifdef INVARIANTS
279 	TMPFS_NODE_LOCK(node);
280 	MPASS(node->tn_vnode == NULL);
281 	MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
282 	TMPFS_NODE_UNLOCK(node);
283 #endif
284 
285 	TMPFS_LOCK(tmp);
286 	LIST_REMOVE(node, tn_entries);
287 	tmp->tm_nodes_inuse--;
288 	TMPFS_UNLOCK(tmp);
289 
290 	switch (node->tn_type) {
291 	case VNON:
292 		/* Do not do anything.  VNON is provided to let the
293 		 * allocation routine clean itself easily by avoiding
294 		 * duplicating code in it. */
295 		/* FALLTHROUGH */
296 	case VBLK:
297 		/* FALLTHROUGH */
298 	case VCHR:
299 		/* FALLTHROUGH */
300 	case VDIR:
301 		/* FALLTHROUGH */
302 	case VFIFO:
303 		/* FALLTHROUGH */
304 	case VSOCK:
305 		break;
306 
307 	case VLNK:
308 		free(node->tn_link, M_TMPFSNAME);
309 		break;
310 
311 	case VREG:
312 		uobj = node->tn_reg.tn_aobj;
313 		if (uobj != NULL) {
314 			TMPFS_LOCK(tmp);
315 			tmp->tm_pages_used -= uobj->size;
316 			TMPFS_UNLOCK(tmp);
317 			KASSERT((uobj->flags & OBJ_TMPFS) == 0,
318 			    ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
319 			vm_object_deallocate(uobj);
320 		}
321 		break;
322 
323 	default:
324 		panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
325 	}
326 
327 	free_unr(tmp->tm_ino_unr, node->tn_id);
328 	uma_zfree(tmp->tm_node_pool, node);
329 }
330 
331 /* --------------------------------------------------------------------- */
332 
333 static __inline uint32_t
334 tmpfs_dirent_hash(const char *name, u_int len)
335 {
336 	uint32_t hash;
337 
338 	hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
339 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
340 	hash &= 0xf;
341 #endif
342 	if (hash < TMPFS_DIRCOOKIE_MIN)
343 		hash += TMPFS_DIRCOOKIE_MIN;
344 
345 	return (hash);
346 }
347 
348 static __inline off_t
349 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
350 {
351 	if (de == NULL)
352 		return (TMPFS_DIRCOOKIE_EOF);
353 
354 	MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
355 
356 	return (de->td_cookie);
357 }
358 
359 static __inline boolean_t
360 tmpfs_dirent_dup(struct tmpfs_dirent *de)
361 {
362 	return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
363 }
364 
365 static __inline boolean_t
366 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
367 {
368 	return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
369 }
370 
371 void
372 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
373 {
374 	de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
375 	memcpy(de->ud.td_name, name, namelen);
376 	de->td_namelen = namelen;
377 }
378 
379 /*
380  * Allocates a new directory entry for the node node with a name of name.
381  * The new directory entry is returned in *de.
382  *
383  * The link count of node is increased by one to reflect the new object
384  * referencing it.
385  *
386  * Returns zero on success or an appropriate error code on failure.
387  */
388 int
389 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
390     const char *name, u_int len, struct tmpfs_dirent **de)
391 {
392 	struct tmpfs_dirent *nde;
393 
394 	nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK);
395 	nde->td_node = node;
396 	if (name != NULL) {
397 		nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
398 		tmpfs_dirent_init(nde, name, len);
399 	} else
400 		nde->td_namelen = 0;
401 	if (node != NULL)
402 		node->tn_links++;
403 
404 	*de = nde;
405 
406 	return 0;
407 }
408 
409 /* --------------------------------------------------------------------- */
410 
411 /*
412  * Frees a directory entry.  It is the caller's responsibility to destroy
413  * the node referenced by it if needed.
414  *
415  * The link count of node is decreased by one to reflect the removal of an
416  * object that referenced it.  This only happens if 'node_exists' is true;
417  * otherwise the function will not access the node referred to by the
418  * directory entry, as it may already have been released from the outside.
419  */
420 void
421 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
422 {
423 	struct tmpfs_node *node;
424 
425 	node = de->td_node;
426 	if (node != NULL) {
427 		MPASS(node->tn_links > 0);
428 		node->tn_links--;
429 	}
430 	if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
431 		free(de->ud.td_name, M_TMPFSNAME);
432 	uma_zfree(tmp->tm_dirent_pool, de);
433 }
434 
435 /* --------------------------------------------------------------------- */
436 
437 void
438 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
439 {
440 
441 	if (vp->v_type != VREG || obj == NULL)
442 		return;
443 
444 	VM_OBJECT_WLOCK(obj);
445 	VI_LOCK(vp);
446 	vm_object_clear_flag(obj, OBJ_TMPFS);
447 	obj->un_pager.swp.swp_tmpfs = NULL;
448 	VI_UNLOCK(vp);
449 	VM_OBJECT_WUNLOCK(obj);
450 }
451 
452 /*
453  * Need to clear v_object for insmntque failure.
454  */
455 static void
456 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
457 {
458 
459 	tmpfs_destroy_vobject(vp, vp->v_object);
460 	vp->v_object = NULL;
461 	vp->v_data = NULL;
462 	vp->v_op = &dead_vnodeops;
463 	vgone(vp);
464 	vput(vp);
465 }
466 
467 /*
468  * Allocates a new vnode for the node node or returns a new reference to
469  * an existing one if the node had already a vnode referencing it.  The
470  * resulting locked vnode is returned in *vpp.
471  *
472  * Returns zero on success or an appropriate error code on failure.
473  */
474 int
475 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
476     struct vnode **vpp)
477 {
478 	struct vnode *vp;
479 	vm_object_t object;
480 	int error;
481 
482 	error = 0;
483 loop:
484 	TMPFS_NODE_LOCK(node);
485 loop1:
486 	if ((vp = node->tn_vnode) != NULL) {
487 		MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
488 		VI_LOCK(vp);
489 		if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
490 		    ((vp->v_iflag & VI_DOOMED) != 0 &&
491 		    (lkflag & LK_NOWAIT) != 0)) {
492 			VI_UNLOCK(vp);
493 			TMPFS_NODE_UNLOCK(node);
494 			error = ENOENT;
495 			vp = NULL;
496 			goto out;
497 		}
498 		if ((vp->v_iflag & VI_DOOMED) != 0) {
499 			VI_UNLOCK(vp);
500 			node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
501 			while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
502 				msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
503 				    0, "tmpfsE", 0);
504 			}
505 			goto loop1;
506 		}
507 		TMPFS_NODE_UNLOCK(node);
508 		error = vget(vp, lkflag | LK_INTERLOCK, curthread);
509 		if (error == ENOENT)
510 			goto loop;
511 		if (error != 0) {
512 			vp = NULL;
513 			goto out;
514 		}
515 
516 		/*
517 		 * Make sure the vnode is still there after
518 		 * getting the interlock to avoid racing a free.
519 		 */
520 		if (node->tn_vnode == NULL || node->tn_vnode != vp) {
521 			vput(vp);
522 			goto loop;
523 		}
524 
525 		goto out;
526 	}
527 
528 	if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
529 	    (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
530 		TMPFS_NODE_UNLOCK(node);
531 		error = ENOENT;
532 		vp = NULL;
533 		goto out;
534 	}
535 
536 	/*
537 	 * otherwise lock the vp list while we call getnewvnode
538 	 * since that can block.
539 	 */
540 	if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
541 		node->tn_vpstate |= TMPFS_VNODE_WANT;
542 		error = msleep((caddr_t) &node->tn_vpstate,
543 		    TMPFS_NODE_MTX(node), PDROP | PCATCH,
544 		    "tmpfs_alloc_vp", 0);
545 		if (error)
546 			return error;
547 
548 		goto loop;
549 	} else
550 		node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
551 
552 	TMPFS_NODE_UNLOCK(node);
553 
554 	/* Get a new vnode and associate it with our node. */
555 	error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
556 	if (error != 0)
557 		goto unlock;
558 	MPASS(vp != NULL);
559 
560 	(void) vn_lock(vp, lkflag | LK_RETRY);
561 
562 	vp->v_data = node;
563 	vp->v_type = node->tn_type;
564 
565 	/* Type-specific initialization. */
566 	switch (node->tn_type) {
567 	case VBLK:
568 		/* FALLTHROUGH */
569 	case VCHR:
570 		/* FALLTHROUGH */
571 	case VLNK:
572 		/* FALLTHROUGH */
573 	case VSOCK:
574 		break;
575 	case VFIFO:
576 		vp->v_op = &tmpfs_fifoop_entries;
577 		break;
578 	case VREG:
579 		object = node->tn_reg.tn_aobj;
580 		VM_OBJECT_WLOCK(object);
581 		VI_LOCK(vp);
582 		KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
583 		vp->v_object = object;
584 		object->un_pager.swp.swp_tmpfs = vp;
585 		vm_object_set_flag(object, OBJ_TMPFS);
586 		VI_UNLOCK(vp);
587 		VM_OBJECT_WUNLOCK(object);
588 		break;
589 	case VDIR:
590 		MPASS(node->tn_dir.tn_parent != NULL);
591 		if (node->tn_dir.tn_parent == node)
592 			vp->v_vflag |= VV_ROOT;
593 		break;
594 
595 	default:
596 		panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
597 	}
598 	if (vp->v_type != VFIFO)
599 		VN_LOCK_ASHARE(vp);
600 
601 	error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
602 	if (error)
603 		vp = NULL;
604 
605 unlock:
606 	TMPFS_NODE_LOCK(node);
607 
608 	MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
609 	node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
610 	node->tn_vnode = vp;
611 
612 	if (node->tn_vpstate & TMPFS_VNODE_WANT) {
613 		node->tn_vpstate &= ~TMPFS_VNODE_WANT;
614 		TMPFS_NODE_UNLOCK(node);
615 		wakeup((caddr_t) &node->tn_vpstate);
616 	} else
617 		TMPFS_NODE_UNLOCK(node);
618 
619 out:
620 	*vpp = vp;
621 
622 #ifdef INVARIANTS
623 	if (error == 0) {
624 		MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
625 		TMPFS_NODE_LOCK(node);
626 		MPASS(*vpp == node->tn_vnode);
627 		TMPFS_NODE_UNLOCK(node);
628 	}
629 #endif
630 
631 	return error;
632 }
633 
634 /* --------------------------------------------------------------------- */
635 
636 /*
637  * Destroys the association between the vnode vp and the node it
638  * references.
639  */
640 void
641 tmpfs_free_vp(struct vnode *vp)
642 {
643 	struct tmpfs_node *node;
644 
645 	node = VP_TO_TMPFS_NODE(vp);
646 
647 	mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED);
648 	node->tn_vnode = NULL;
649 	if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
650 		wakeup(&node->tn_vnode);
651 	node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
652 	vp->v_data = NULL;
653 }
654 
655 /* --------------------------------------------------------------------- */
656 
657 /*
658  * Allocates a new file of type 'type' and adds it to the parent directory
659  * 'dvp'; this addition is done using the component name given in 'cnp'.
660  * The ownership of the new file is automatically assigned based on the
661  * credentials of the caller (through 'cnp'), the group is set based on
662  * the parent directory and the mode is determined from the 'vap' argument.
663  * If successful, *vpp holds a vnode to the newly created file and zero
664  * is returned.  Otherwise *vpp is NULL and the function returns an
665  * appropriate error code.
666  */
667 int
668 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
669     struct componentname *cnp, char *target)
670 {
671 	int error;
672 	struct tmpfs_dirent *de;
673 	struct tmpfs_mount *tmp;
674 	struct tmpfs_node *dnode;
675 	struct tmpfs_node *node;
676 	struct tmpfs_node *parent;
677 
678 	MPASS(VOP_ISLOCKED(dvp));
679 	MPASS(cnp->cn_flags & HASBUF);
680 
681 	tmp = VFS_TO_TMPFS(dvp->v_mount);
682 	dnode = VP_TO_TMPFS_DIR(dvp);
683 	*vpp = NULL;
684 
685 	/* If the entry we are creating is a directory, we cannot overflow
686 	 * the number of links of its parent, because it will get a new
687 	 * link. */
688 	if (vap->va_type == VDIR) {
689 		/* Ensure that we do not overflow the maximum number of links
690 		 * imposed by the system. */
691 		MPASS(dnode->tn_links <= LINK_MAX);
692 		if (dnode->tn_links == LINK_MAX) {
693 			error = EMLINK;
694 			goto out;
695 		}
696 
697 		parent = dnode;
698 		MPASS(parent != NULL);
699 	} else
700 		parent = NULL;
701 
702 	/* Allocate a node that represents the new file. */
703 	error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid,
704 	    dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node);
705 	if (error != 0)
706 		goto out;
707 
708 	/* Allocate a directory entry that points to the new file. */
709 	error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
710 	    &de);
711 	if (error != 0) {
712 		tmpfs_free_node(tmp, node);
713 		goto out;
714 	}
715 
716 	/* Allocate a vnode for the new file. */
717 	error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
718 	if (error != 0) {
719 		tmpfs_free_dirent(tmp, de);
720 		tmpfs_free_node(tmp, node);
721 		goto out;
722 	}
723 
724 	/* Now that all required items are allocated, we can proceed to
725 	 * insert the new node into the directory, an operation that
726 	 * cannot fail. */
727 	if (cnp->cn_flags & ISWHITEOUT)
728 		tmpfs_dir_whiteout_remove(dvp, cnp);
729 	tmpfs_dir_attach(dvp, de);
730 
731 out:
732 
733 	return error;
734 }
735 
736 /* --------------------------------------------------------------------- */
737 
738 static struct tmpfs_dirent *
739 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
740 {
741 	struct tmpfs_dirent *de;
742 
743 	de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
744 	dc->tdc_tree = de;
745 	if (de != NULL && tmpfs_dirent_duphead(de))
746 		de = LIST_FIRST(&de->ud.td_duphead);
747 	dc->tdc_current = de;
748 
749 	return (dc->tdc_current);
750 }
751 
752 static struct tmpfs_dirent *
753 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
754 {
755 	struct tmpfs_dirent *de;
756 
757 	MPASS(dc->tdc_tree != NULL);
758 	if (tmpfs_dirent_dup(dc->tdc_current)) {
759 		dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
760 		if (dc->tdc_current != NULL)
761 			return (dc->tdc_current);
762 	}
763 	dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
764 	    &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
765 	if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
766 		dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
767 		MPASS(dc->tdc_current != NULL);
768 	}
769 
770 	return (dc->tdc_current);
771 }
772 
773 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
774 static struct tmpfs_dirent *
775 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
776 {
777 	struct tmpfs_dirent *de, dekey;
778 
779 	dekey.td_hash = hash;
780 	de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
781 	return (de);
782 }
783 
784 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
785 static struct tmpfs_dirent *
786 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
787     struct tmpfs_dir_cursor *dc)
788 {
789 	struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
790 	struct tmpfs_dirent *de, dekey;
791 
792 	MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
793 
794 	if (cookie == node->tn_dir.tn_readdir_lastn &&
795 	    (de = node->tn_dir.tn_readdir_lastp) != NULL) {
796 		/* Protect against possible race, tn_readdir_last[pn]
797 		 * may be updated with only shared vnode lock held. */
798 		if (cookie == tmpfs_dirent_cookie(de))
799 			goto out;
800 	}
801 
802 	if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
803 		LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
804 		    uh.td_dup.index_entries) {
805 			MPASS(tmpfs_dirent_dup(de));
806 			if (de->td_cookie == cookie)
807 				goto out;
808 			/* dupindex list is sorted. */
809 			if (de->td_cookie < cookie) {
810 				de = NULL;
811 				goto out;
812 			}
813 		}
814 		MPASS(de == NULL);
815 		goto out;
816 	}
817 
818 	MPASS((cookie & TMPFS_DIRCOOKIE_MASK) == cookie);
819 	dekey.td_hash = cookie;
820 	/* Recover if direntry for cookie was removed */
821 	de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
822 	dc->tdc_tree = de;
823 	dc->tdc_current = de;
824 	if (de != NULL && tmpfs_dirent_duphead(de)) {
825 		dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
826 		MPASS(dc->tdc_current != NULL);
827 	}
828 	return (dc->tdc_current);
829 
830 out:
831 	dc->tdc_tree = de;
832 	dc->tdc_current = de;
833 	if (de != NULL && tmpfs_dirent_dup(de))
834 		dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
835 		    de->td_hash);
836 	return (dc->tdc_current);
837 }
838 
839 /*
840  * Looks for a directory entry in the directory represented by node.
841  * 'cnp' describes the name of the entry to look for.  Note that the .
842  * and .. components are not allowed as they do not physically exist
843  * within directories.
844  *
845  * Returns a pointer to the entry when found, otherwise NULL.
846  */
847 struct tmpfs_dirent *
848 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
849     struct componentname *cnp)
850 {
851 	struct tmpfs_dir_duphead *duphead;
852 	struct tmpfs_dirent *de;
853 	uint32_t hash;
854 
855 	MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
856 	MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
857 	    cnp->cn_nameptr[1] == '.')));
858 	TMPFS_VALIDATE_DIR(node);
859 
860 	hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
861 	de = tmpfs_dir_xlookup_hash(node, hash);
862 	if (de != NULL && tmpfs_dirent_duphead(de)) {
863 		duphead = &de->ud.td_duphead;
864 		LIST_FOREACH(de, duphead, uh.td_dup.entries) {
865 			if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
866 			    cnp->cn_namelen))
867 				break;
868 		}
869 	} else if (de != NULL) {
870 		if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
871 		    cnp->cn_namelen))
872 			de = NULL;
873 	}
874 	if (de != NULL && f != NULL && de->td_node != f)
875 		de = NULL;
876 
877 	return (de);
878 }
879 
880 /*
881  * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
882  * list, allocate new cookie value.
883  */
884 static void
885 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
886     struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
887 {
888 	struct tmpfs_dir_duphead *dupindex;
889 	struct tmpfs_dirent *de, *pde;
890 
891 	dupindex = &dnode->tn_dir.tn_dupindex;
892 	de = LIST_FIRST(dupindex);
893 	if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
894 		if (de == NULL)
895 			nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
896 		else
897 			nde->td_cookie = de->td_cookie + 1;
898 		MPASS(tmpfs_dirent_dup(nde));
899 		LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
900 		LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
901 		return;
902 	}
903 
904 	/*
905 	 * Cookie numbers are near exhaustion. Scan dupindex list for unused
906 	 * numbers. dupindex list is sorted in descending order. Keep it so
907 	 * after inserting nde.
908 	 */
909 	while (1) {
910 		pde = de;
911 		de = LIST_NEXT(de, uh.td_dup.index_entries);
912 		if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
913 			/*
914 			 * Last element of the index doesn't have minimal cookie
915 			 * value, use it.
916 			 */
917 			nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
918 			LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
919 			LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
920 			return;
921 		} else if (de == NULL) {
922 			/*
923 			 * We are so lucky have 2^30 hash duplicates in single
924 			 * directory :) Return largest possible cookie value.
925 			 * It should be fine except possible issues with
926 			 * VOP_READDIR restart.
927 			 */
928 			nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
929 			LIST_INSERT_HEAD(dupindex, nde,
930 			    uh.td_dup.index_entries);
931 			LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
932 			return;
933 		}
934 		if (de->td_cookie + 1 == pde->td_cookie ||
935 		    de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
936 			continue;	/* No hole or invalid cookie. */
937 		nde->td_cookie = de->td_cookie + 1;
938 		MPASS(tmpfs_dirent_dup(nde));
939 		MPASS(pde->td_cookie > nde->td_cookie);
940 		MPASS(nde->td_cookie > de->td_cookie);
941 		LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
942 		LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
943 		return;
944 	};
945 }
946 
947 /*
948  * Attaches the directory entry de to the directory represented by vp.
949  * Note that this does not change the link count of the node pointed by
950  * the directory entry, as this is done by tmpfs_alloc_dirent.
951  */
952 void
953 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
954 {
955 	struct tmpfs_node *dnode;
956 	struct tmpfs_dirent *xde, *nde;
957 
958 	ASSERT_VOP_ELOCKED(vp, __func__);
959 	MPASS(de->td_namelen > 0);
960 	MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
961 	MPASS(de->td_cookie == de->td_hash);
962 
963 	dnode = VP_TO_TMPFS_DIR(vp);
964 	dnode->tn_dir.tn_readdir_lastn = 0;
965 	dnode->tn_dir.tn_readdir_lastp = NULL;
966 
967 	MPASS(!tmpfs_dirent_dup(de));
968 	xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
969 	if (xde != NULL && tmpfs_dirent_duphead(xde))
970 		tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
971 	else if (xde != NULL) {
972 		/*
973 		 * Allocate new duphead. Swap xde with duphead to avoid
974 		 * adding/removing elements with the same hash.
975 		 */
976 		MPASS(!tmpfs_dirent_dup(xde));
977 		tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
978 		    &nde);
979 		/* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
980 		memcpy(nde, xde, sizeof(*xde));
981 		xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
982 		LIST_INIT(&xde->ud.td_duphead);
983 		xde->td_namelen = 0;
984 		xde->td_node = NULL;
985 		tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
986 		tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
987 	}
988 	dnode->tn_size += sizeof(struct tmpfs_dirent);
989 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
990 	    TMPFS_NODE_MODIFIED;
991 }
992 
993 /* --------------------------------------------------------------------- */
994 
995 /*
996  * Detaches the directory entry de from the directory represented by vp.
997  * Note that this does not change the link count of the node pointed by
998  * the directory entry, as this is done by tmpfs_free_dirent.
999  */
1000 void
1001 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1002 {
1003 	struct tmpfs_mount *tmp;
1004 	struct tmpfs_dir *head;
1005 	struct tmpfs_node *dnode;
1006 	struct tmpfs_dirent *xde;
1007 
1008 	ASSERT_VOP_ELOCKED(vp, __func__);
1009 
1010 	dnode = VP_TO_TMPFS_DIR(vp);
1011 	head = &dnode->tn_dir.tn_dirhead;
1012 	dnode->tn_dir.tn_readdir_lastn = 0;
1013 	dnode->tn_dir.tn_readdir_lastp = NULL;
1014 
1015 	if (tmpfs_dirent_dup(de)) {
1016 		/* Remove duphead if de was last entry. */
1017 		if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1018 			xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1019 			MPASS(tmpfs_dirent_duphead(xde));
1020 		} else
1021 			xde = NULL;
1022 		LIST_REMOVE(de, uh.td_dup.entries);
1023 		LIST_REMOVE(de, uh.td_dup.index_entries);
1024 		if (xde != NULL) {
1025 			if (LIST_EMPTY(&xde->ud.td_duphead)) {
1026 				RB_REMOVE(tmpfs_dir, head, xde);
1027 				tmp = VFS_TO_TMPFS(vp->v_mount);
1028 				MPASS(xde->td_node == NULL);
1029 				tmpfs_free_dirent(tmp, xde);
1030 			}
1031 		}
1032 	} else
1033 		RB_REMOVE(tmpfs_dir, head, de);
1034 
1035 	dnode->tn_size -= sizeof(struct tmpfs_dirent);
1036 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1037 	    TMPFS_NODE_MODIFIED;
1038 }
1039 
1040 void
1041 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1042 {
1043 	struct tmpfs_dirent *de, *dde, *nde;
1044 
1045 	RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1046 		RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1047 		/* Node may already be destroyed. */
1048 		de->td_node = NULL;
1049 		if (tmpfs_dirent_duphead(de)) {
1050 			while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1051 				LIST_REMOVE(dde, uh.td_dup.entries);
1052 				dde->td_node = NULL;
1053 				tmpfs_free_dirent(tmp, dde);
1054 			}
1055 		}
1056 		tmpfs_free_dirent(tmp, de);
1057 	}
1058 }
1059 
1060 /* --------------------------------------------------------------------- */
1061 
1062 /*
1063  * Helper function for tmpfs_readdir.  Creates a '.' entry for the given
1064  * directory and returns it in the uio space.  The function returns 0
1065  * on success, -1 if there was not enough space in the uio structure to
1066  * hold the directory entry or an appropriate error code if another
1067  * error happens.
1068  */
1069 static int
1070 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
1071 {
1072 	int error;
1073 	struct dirent dent;
1074 
1075 	TMPFS_VALIDATE_DIR(node);
1076 	MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1077 
1078 	dent.d_fileno = node->tn_id;
1079 	dent.d_type = DT_DIR;
1080 	dent.d_namlen = 1;
1081 	dent.d_name[0] = '.';
1082 	dent.d_name[1] = '\0';
1083 	dent.d_reclen = GENERIC_DIRSIZ(&dent);
1084 
1085 	if (dent.d_reclen > uio->uio_resid)
1086 		error = EJUSTRETURN;
1087 	else
1088 		error = uiomove(&dent, dent.d_reclen, uio);
1089 
1090 	node->tn_status |= TMPFS_NODE_ACCESSED;
1091 
1092 	return error;
1093 }
1094 
1095 /* --------------------------------------------------------------------- */
1096 
1097 /*
1098  * Helper function for tmpfs_readdir.  Creates a '..' entry for the given
1099  * directory and returns it in the uio space.  The function returns 0
1100  * on success, -1 if there was not enough space in the uio structure to
1101  * hold the directory entry or an appropriate error code if another
1102  * error happens.
1103  */
1104 static int
1105 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
1106 {
1107 	int error;
1108 	struct dirent dent;
1109 
1110 	TMPFS_VALIDATE_DIR(node);
1111 	MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1112 
1113 	/*
1114 	 * Return ENOENT if the current node is already removed.
1115 	 */
1116 	TMPFS_ASSERT_LOCKED(node);
1117 	if (node->tn_dir.tn_parent == NULL) {
1118 		return (ENOENT);
1119 	}
1120 
1121 	TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
1122 	dent.d_fileno = node->tn_dir.tn_parent->tn_id;
1123 	TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
1124 
1125 	dent.d_type = DT_DIR;
1126 	dent.d_namlen = 2;
1127 	dent.d_name[0] = '.';
1128 	dent.d_name[1] = '.';
1129 	dent.d_name[2] = '\0';
1130 	dent.d_reclen = GENERIC_DIRSIZ(&dent);
1131 
1132 	if (dent.d_reclen > uio->uio_resid)
1133 		error = EJUSTRETURN;
1134 	else
1135 		error = uiomove(&dent, dent.d_reclen, uio);
1136 
1137 	node->tn_status |= TMPFS_NODE_ACCESSED;
1138 
1139 	return error;
1140 }
1141 
1142 /* --------------------------------------------------------------------- */
1143 
1144 /*
1145  * Helper function for tmpfs_readdir.  Returns as much directory entries
1146  * as can fit in the uio space.  The read starts at uio->uio_offset.
1147  * The function returns 0 on success, -1 if there was not enough space
1148  * in the uio structure to hold the directory entry or an appropriate
1149  * error code if another error happens.
1150  */
1151 int
1152 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies,
1153     u_long *cookies, int *ncookies)
1154 {
1155 	struct tmpfs_dir_cursor dc;
1156 	struct tmpfs_dirent *de;
1157 	off_t off;
1158 	int error;
1159 
1160 	TMPFS_VALIDATE_DIR(node);
1161 
1162 	off = 0;
1163 
1164 	/*
1165 	 * Lookup the node from the current offset.  The starting offset of
1166 	 * 0 will lookup both '.' and '..', and then the first real entry,
1167 	 * or EOF if there are none.  Then find all entries for the dir that
1168 	 * fit into the buffer.  Once no more entries are found (de == NULL),
1169 	 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1170 	 * call to return 0.
1171 	 */
1172 	switch (uio->uio_offset) {
1173 	case TMPFS_DIRCOOKIE_DOT:
1174 		error = tmpfs_dir_getdotdent(node, uio);
1175 		if (error != 0)
1176 			return (error);
1177 		uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1178 		if (cookies != NULL)
1179 			cookies[(*ncookies)++] = off = uio->uio_offset;
1180 		/* FALLTHROUGH */
1181 	case TMPFS_DIRCOOKIE_DOTDOT:
1182 		error = tmpfs_dir_getdotdotdent(node, uio);
1183 		if (error != 0)
1184 			return (error);
1185 		de = tmpfs_dir_first(node, &dc);
1186 		uio->uio_offset = tmpfs_dirent_cookie(de);
1187 		if (cookies != NULL)
1188 			cookies[(*ncookies)++] = off = uio->uio_offset;
1189 		/* EOF. */
1190 		if (de == NULL)
1191 			return (0);
1192 		break;
1193 	case TMPFS_DIRCOOKIE_EOF:
1194 		return (0);
1195 	default:
1196 		de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1197 		if (de == NULL)
1198 			return (EINVAL);
1199 		if (cookies != NULL)
1200 			off = tmpfs_dirent_cookie(de);
1201 	}
1202 
1203 	/* Read as much entries as possible; i.e., until we reach the end of
1204 	 * the directory or we exhaust uio space. */
1205 	do {
1206 		struct dirent d;
1207 
1208 		/* Create a dirent structure representing the current
1209 		 * tmpfs_node and fill it. */
1210 		if (de->td_node == NULL) {
1211 			d.d_fileno = 1;
1212 			d.d_type = DT_WHT;
1213 		} else {
1214 			d.d_fileno = de->td_node->tn_id;
1215 			switch (de->td_node->tn_type) {
1216 			case VBLK:
1217 				d.d_type = DT_BLK;
1218 				break;
1219 
1220 			case VCHR:
1221 				d.d_type = DT_CHR;
1222 				break;
1223 
1224 			case VDIR:
1225 				d.d_type = DT_DIR;
1226 				break;
1227 
1228 			case VFIFO:
1229 				d.d_type = DT_FIFO;
1230 				break;
1231 
1232 			case VLNK:
1233 				d.d_type = DT_LNK;
1234 				break;
1235 
1236 			case VREG:
1237 				d.d_type = DT_REG;
1238 				break;
1239 
1240 			case VSOCK:
1241 				d.d_type = DT_SOCK;
1242 				break;
1243 
1244 			default:
1245 				panic("tmpfs_dir_getdents: type %p %d",
1246 				    de->td_node, (int)de->td_node->tn_type);
1247 			}
1248 		}
1249 		d.d_namlen = de->td_namelen;
1250 		MPASS(de->td_namelen < sizeof(d.d_name));
1251 		(void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1252 		d.d_name[de->td_namelen] = '\0';
1253 		d.d_reclen = GENERIC_DIRSIZ(&d);
1254 
1255 		/* Stop reading if the directory entry we are treating is
1256 		 * bigger than the amount of data that can be returned. */
1257 		if (d.d_reclen > uio->uio_resid) {
1258 			error = EJUSTRETURN;
1259 			break;
1260 		}
1261 
1262 		/* Copy the new dirent structure into the output buffer and
1263 		 * advance pointers. */
1264 		error = uiomove(&d, d.d_reclen, uio);
1265 		if (error == 0) {
1266 			de = tmpfs_dir_next(node, &dc);
1267 			if (cookies != NULL) {
1268 				off = tmpfs_dirent_cookie(de);
1269 				MPASS(*ncookies < maxcookies);
1270 				cookies[(*ncookies)++] = off;
1271 			}
1272 		}
1273 	} while (error == 0 && uio->uio_resid > 0 && de != NULL);
1274 
1275 	/* Skip setting off when using cookies as it is already done above. */
1276 	if (cookies == NULL)
1277 		off = tmpfs_dirent_cookie(de);
1278 
1279 	/* Update the offset and cache. */
1280 	uio->uio_offset = off;
1281 	node->tn_dir.tn_readdir_lastn = off;
1282 	node->tn_dir.tn_readdir_lastp = de;
1283 
1284 	node->tn_status |= TMPFS_NODE_ACCESSED;
1285 	return error;
1286 }
1287 
1288 int
1289 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1290 {
1291 	struct tmpfs_dirent *de;
1292 	int error;
1293 
1294 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1295 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
1296 	if (error != 0)
1297 		return (error);
1298 	tmpfs_dir_attach(dvp, de);
1299 	return (0);
1300 }
1301 
1302 void
1303 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1304 {
1305 	struct tmpfs_dirent *de;
1306 
1307 	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1308 	MPASS(de != NULL && de->td_node == NULL);
1309 	tmpfs_dir_detach(dvp, de);
1310 	tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1311 }
1312 
1313 /* --------------------------------------------------------------------- */
1314 
1315 /*
1316  * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1317  * size 'newsize'.  'vp' must point to a vnode that represents a regular file.
1318  * 'newsize' must be positive.
1319  *
1320  * Returns zero on success or an appropriate error code on failure.
1321  */
1322 int
1323 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1324 {
1325 	struct tmpfs_mount *tmp;
1326 	struct tmpfs_node *node;
1327 	vm_object_t uobj;
1328 	vm_page_t m, ma[1];
1329 	vm_pindex_t idx, newpages, oldpages;
1330 	off_t oldsize;
1331 	int base, rv;
1332 
1333 	MPASS(vp->v_type == VREG);
1334 	MPASS(newsize >= 0);
1335 
1336 	node = VP_TO_TMPFS_NODE(vp);
1337 	uobj = node->tn_reg.tn_aobj;
1338 	tmp = VFS_TO_TMPFS(vp->v_mount);
1339 
1340 	/*
1341 	 * Convert the old and new sizes to the number of pages needed to
1342 	 * store them.  It may happen that we do not need to do anything
1343 	 * because the last allocated page can accommodate the change on
1344 	 * its own.
1345 	 */
1346 	oldsize = node->tn_size;
1347 	oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1348 	MPASS(oldpages == uobj->size);
1349 	newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1350 	if (newpages > oldpages &&
1351 	    tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1352 		return (ENOSPC);
1353 
1354 	VM_OBJECT_WLOCK(uobj);
1355 	if (newsize < oldsize) {
1356 		/*
1357 		 * Zero the truncated part of the last page.
1358 		 */
1359 		base = newsize & PAGE_MASK;
1360 		if (base != 0) {
1361 			idx = OFF_TO_IDX(newsize);
1362 retry:
1363 			m = vm_page_lookup(uobj, idx);
1364 			if (m != NULL) {
1365 				if (vm_page_sleep_if_busy(m, "tmfssz"))
1366 					goto retry;
1367 				MPASS(m->valid == VM_PAGE_BITS_ALL);
1368 			} else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1369 				m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL);
1370 				if (m == NULL) {
1371 					VM_OBJECT_WUNLOCK(uobj);
1372 					VM_WAIT;
1373 					VM_OBJECT_WLOCK(uobj);
1374 					goto retry;
1375 				} else if (m->valid != VM_PAGE_BITS_ALL) {
1376 					ma[0] = m;
1377 					rv = vm_pager_get_pages(uobj, ma, 1, 0);
1378 					m = vm_page_lookup(uobj, idx);
1379 				} else
1380 					/* A cached page was reactivated. */
1381 					rv = VM_PAGER_OK;
1382 				vm_page_lock(m);
1383 				if (rv == VM_PAGER_OK) {
1384 					vm_page_deactivate(m);
1385 					vm_page_unlock(m);
1386 					vm_page_xunbusy(m);
1387 				} else {
1388 					vm_page_free(m);
1389 					vm_page_unlock(m);
1390 					if (ignerr)
1391 						m = NULL;
1392 					else {
1393 						VM_OBJECT_WUNLOCK(uobj);
1394 						return (EIO);
1395 					}
1396 				}
1397 			}
1398 			if (m != NULL) {
1399 				pmap_zero_page_area(m, base, PAGE_SIZE - base);
1400 				vm_page_dirty(m);
1401 				vm_pager_page_unswapped(m);
1402 			}
1403 		}
1404 
1405 		/*
1406 		 * Release any swap space and free any whole pages.
1407 		 */
1408 		if (newpages < oldpages) {
1409 			swap_pager_freespace(uobj, newpages, oldpages -
1410 			    newpages);
1411 			vm_object_page_remove(uobj, newpages, 0, 0);
1412 		}
1413 	}
1414 	uobj->size = newpages;
1415 	VM_OBJECT_WUNLOCK(uobj);
1416 
1417 	TMPFS_LOCK(tmp);
1418 	tmp->tm_pages_used += (newpages - oldpages);
1419 	TMPFS_UNLOCK(tmp);
1420 
1421 	node->tn_size = newsize;
1422 	return (0);
1423 }
1424 
1425 /* --------------------------------------------------------------------- */
1426 
1427 /*
1428  * Change flags of the given vnode.
1429  * Caller should execute tmpfs_update on vp after a successful execution.
1430  * The vnode must be locked on entry and remain locked on exit.
1431  */
1432 int
1433 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1434     struct thread *p)
1435 {
1436 	int error;
1437 	struct tmpfs_node *node;
1438 
1439 	MPASS(VOP_ISLOCKED(vp));
1440 
1441 	node = VP_TO_TMPFS_NODE(vp);
1442 
1443 	if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1444 	    UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1445 	    UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1446 	    UF_SPARSE | UF_SYSTEM)) != 0)
1447 		return (EOPNOTSUPP);
1448 
1449 	/* Disallow this operation if the file system is mounted read-only. */
1450 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1451 		return EROFS;
1452 
1453 	/*
1454 	 * Callers may only modify the file flags on objects they
1455 	 * have VADMIN rights for.
1456 	 */
1457 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1458 		return (error);
1459 	/*
1460 	 * Unprivileged processes are not permitted to unset system
1461 	 * flags, or modify flags if any system flags are set.
1462 	 */
1463 	if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1464 		if (node->tn_flags &
1465 		    (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1466 			error = securelevel_gt(cred, 0);
1467 			if (error)
1468 				return (error);
1469 		}
1470 	} else {
1471 		if (node->tn_flags &
1472 		    (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1473 		    ((flags ^ node->tn_flags) & SF_SETTABLE))
1474 			return (EPERM);
1475 	}
1476 	node->tn_flags = flags;
1477 	node->tn_status |= TMPFS_NODE_CHANGED;
1478 
1479 	MPASS(VOP_ISLOCKED(vp));
1480 
1481 	return 0;
1482 }
1483 
1484 /* --------------------------------------------------------------------- */
1485 
1486 /*
1487  * Change access mode on the given vnode.
1488  * Caller should execute tmpfs_update on vp after a successful execution.
1489  * The vnode must be locked on entry and remain locked on exit.
1490  */
1491 int
1492 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1493 {
1494 	int error;
1495 	struct tmpfs_node *node;
1496 
1497 	MPASS(VOP_ISLOCKED(vp));
1498 
1499 	node = VP_TO_TMPFS_NODE(vp);
1500 
1501 	/* Disallow this operation if the file system is mounted read-only. */
1502 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1503 		return EROFS;
1504 
1505 	/* Immutable or append-only files cannot be modified, either. */
1506 	if (node->tn_flags & (IMMUTABLE | APPEND))
1507 		return EPERM;
1508 
1509 	/*
1510 	 * To modify the permissions on a file, must possess VADMIN
1511 	 * for that file.
1512 	 */
1513 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1514 		return (error);
1515 
1516 	/*
1517 	 * Privileged processes may set the sticky bit on non-directories,
1518 	 * as well as set the setgid bit on a file with a group that the
1519 	 * process is not a member of.
1520 	 */
1521 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1522 		if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1523 			return (EFTYPE);
1524 	}
1525 	if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1526 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1527 		if (error)
1528 			return (error);
1529 	}
1530 
1531 
1532 	node->tn_mode &= ~ALLPERMS;
1533 	node->tn_mode |= mode & ALLPERMS;
1534 
1535 	node->tn_status |= TMPFS_NODE_CHANGED;
1536 
1537 	MPASS(VOP_ISLOCKED(vp));
1538 
1539 	return 0;
1540 }
1541 
1542 /* --------------------------------------------------------------------- */
1543 
1544 /*
1545  * Change ownership of the given vnode.  At least one of uid or gid must
1546  * be different than VNOVAL.  If one is set to that value, the attribute
1547  * is unchanged.
1548  * Caller should execute tmpfs_update on vp after a successful execution.
1549  * The vnode must be locked on entry and remain locked on exit.
1550  */
1551 int
1552 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1553     struct thread *p)
1554 {
1555 	int error;
1556 	struct tmpfs_node *node;
1557 	uid_t ouid;
1558 	gid_t ogid;
1559 
1560 	MPASS(VOP_ISLOCKED(vp));
1561 
1562 	node = VP_TO_TMPFS_NODE(vp);
1563 
1564 	/* Assign default values if they are unknown. */
1565 	MPASS(uid != VNOVAL || gid != VNOVAL);
1566 	if (uid == VNOVAL)
1567 		uid = node->tn_uid;
1568 	if (gid == VNOVAL)
1569 		gid = node->tn_gid;
1570 	MPASS(uid != VNOVAL && gid != VNOVAL);
1571 
1572 	/* Disallow this operation if the file system is mounted read-only. */
1573 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1574 		return EROFS;
1575 
1576 	/* Immutable or append-only files cannot be modified, either. */
1577 	if (node->tn_flags & (IMMUTABLE | APPEND))
1578 		return EPERM;
1579 
1580 	/*
1581 	 * To modify the ownership of a file, must possess VADMIN for that
1582 	 * file.
1583 	 */
1584 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1585 		return (error);
1586 
1587 	/*
1588 	 * To change the owner of a file, or change the group of a file to a
1589 	 * group of which we are not a member, the caller must have
1590 	 * privilege.
1591 	 */
1592 	if ((uid != node->tn_uid ||
1593 	    (gid != node->tn_gid && !groupmember(gid, cred))) &&
1594 	    (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1595 		return (error);
1596 
1597 	ogid = node->tn_gid;
1598 	ouid = node->tn_uid;
1599 
1600 	node->tn_uid = uid;
1601 	node->tn_gid = gid;
1602 
1603 	node->tn_status |= TMPFS_NODE_CHANGED;
1604 
1605 	if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1606 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1607 			node->tn_mode &= ~(S_ISUID | S_ISGID);
1608 	}
1609 
1610 	MPASS(VOP_ISLOCKED(vp));
1611 
1612 	return 0;
1613 }
1614 
1615 /* --------------------------------------------------------------------- */
1616 
1617 /*
1618  * Change size of the given vnode.
1619  * Caller should execute tmpfs_update on vp after a successful execution.
1620  * The vnode must be locked on entry and remain locked on exit.
1621  */
1622 int
1623 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1624     struct thread *p)
1625 {
1626 	int error;
1627 	struct tmpfs_node *node;
1628 
1629 	MPASS(VOP_ISLOCKED(vp));
1630 
1631 	node = VP_TO_TMPFS_NODE(vp);
1632 
1633 	/* Decide whether this is a valid operation based on the file type. */
1634 	error = 0;
1635 	switch (vp->v_type) {
1636 	case VDIR:
1637 		return EISDIR;
1638 
1639 	case VREG:
1640 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
1641 			return EROFS;
1642 		break;
1643 
1644 	case VBLK:
1645 		/* FALLTHROUGH */
1646 	case VCHR:
1647 		/* FALLTHROUGH */
1648 	case VFIFO:
1649 		/* Allow modifications of special files even if in the file
1650 		 * system is mounted read-only (we are not modifying the
1651 		 * files themselves, but the objects they represent). */
1652 		return 0;
1653 
1654 	default:
1655 		/* Anything else is unsupported. */
1656 		return EOPNOTSUPP;
1657 	}
1658 
1659 	/* Immutable or append-only files cannot be modified, either. */
1660 	if (node->tn_flags & (IMMUTABLE | APPEND))
1661 		return EPERM;
1662 
1663 	error = tmpfs_truncate(vp, size);
1664 	/* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1665 	 * for us, as will update tn_status; no need to do that here. */
1666 
1667 	MPASS(VOP_ISLOCKED(vp));
1668 
1669 	return error;
1670 }
1671 
1672 /* --------------------------------------------------------------------- */
1673 
1674 /*
1675  * Change access and modification times of the given vnode.
1676  * Caller should execute tmpfs_update on vp after a successful execution.
1677  * The vnode must be locked on entry and remain locked on exit.
1678  */
1679 int
1680 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
1681     struct ucred *cred, struct thread *l)
1682 {
1683 	int error;
1684 	struct tmpfs_node *node;
1685 
1686 	MPASS(VOP_ISLOCKED(vp));
1687 
1688 	node = VP_TO_TMPFS_NODE(vp);
1689 
1690 	/* Disallow this operation if the file system is mounted read-only. */
1691 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1692 		return EROFS;
1693 
1694 	/* Immutable or append-only files cannot be modified, either. */
1695 	if (node->tn_flags & (IMMUTABLE | APPEND))
1696 		return EPERM;
1697 
1698 	error = vn_utimes_perm(vp, vap, cred, l);
1699 	if (error != 0)
1700 		return (error);
1701 
1702 	if (vap->va_atime.tv_sec != VNOVAL && vap->va_atime.tv_nsec != VNOVAL)
1703 		node->tn_status |= TMPFS_NODE_ACCESSED;
1704 
1705 	if (vap->va_mtime.tv_sec != VNOVAL && vap->va_mtime.tv_nsec != VNOVAL)
1706 		node->tn_status |= TMPFS_NODE_MODIFIED;
1707 
1708 	if (vap->va_birthtime.tv_nsec != VNOVAL &&
1709 	    vap->va_birthtime.tv_nsec != VNOVAL)
1710 		node->tn_status |= TMPFS_NODE_MODIFIED;
1711 
1712 	tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1713 
1714 	if (vap->va_birthtime.tv_nsec != VNOVAL &&
1715 	    vap->va_birthtime.tv_nsec != VNOVAL)
1716 		node->tn_birthtime = vap->va_birthtime;
1717 	MPASS(VOP_ISLOCKED(vp));
1718 
1719 	return 0;
1720 }
1721 
1722 /* --------------------------------------------------------------------- */
1723 /* Sync timestamps */
1724 void
1725 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1726     const struct timespec *mod)
1727 {
1728 	struct tmpfs_node *node;
1729 	struct timespec now;
1730 
1731 	node = VP_TO_TMPFS_NODE(vp);
1732 
1733 	if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1734 	    TMPFS_NODE_CHANGED)) == 0)
1735 		return;
1736 
1737 	vfs_timestamp(&now);
1738 	if (node->tn_status & TMPFS_NODE_ACCESSED) {
1739 		if (acc == NULL)
1740 			 acc = &now;
1741 		node->tn_atime = *acc;
1742 	}
1743 	if (node->tn_status & TMPFS_NODE_MODIFIED) {
1744 		if (mod == NULL)
1745 			mod = &now;
1746 		node->tn_mtime = *mod;
1747 	}
1748 	if (node->tn_status & TMPFS_NODE_CHANGED) {
1749 		node->tn_ctime = now;
1750 	}
1751 	node->tn_status &=
1752 	    ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1753 }
1754 
1755 /* --------------------------------------------------------------------- */
1756 
1757 void
1758 tmpfs_update(struct vnode *vp)
1759 {
1760 
1761 	tmpfs_itimes(vp, NULL, NULL);
1762 }
1763 
1764 /* --------------------------------------------------------------------- */
1765 
1766 int
1767 tmpfs_truncate(struct vnode *vp, off_t length)
1768 {
1769 	int error;
1770 	struct tmpfs_node *node;
1771 
1772 	node = VP_TO_TMPFS_NODE(vp);
1773 
1774 	if (length < 0) {
1775 		error = EINVAL;
1776 		goto out;
1777 	}
1778 
1779 	if (node->tn_size == length) {
1780 		error = 0;
1781 		goto out;
1782 	}
1783 
1784 	if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1785 		return (EFBIG);
1786 
1787 	error = tmpfs_reg_resize(vp, length, FALSE);
1788 	if (error == 0) {
1789 		node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1790 	}
1791 
1792 out:
1793 	tmpfs_update(vp);
1794 
1795 	return error;
1796 }
1797 
1798 static __inline int
1799 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1800 {
1801 	if (a->td_hash > b->td_hash)
1802 		return (1);
1803 	else if (a->td_hash < b->td_hash)
1804 		return (-1);
1805 	return (0);
1806 }
1807 
1808 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
1809