xref: /freebsd/sys/fs/tmpfs/tmpfs_vfsops.c (revision 9768746ba83efa02837c5b9c66348db6e900208f)
1 /*	$NetBSD: tmpfs_vfsops.c,v 1.10 2005/12/11 12:24:29 christos Exp $	*/
2 
3 /*-
4  * SPDX-License-Identifier: BSD-2-Clause-NetBSD
5  *
6  * Copyright (c) 2005 The NetBSD Foundation, Inc.
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to The NetBSD Foundation
10  * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11  * 2005 program.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
24  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
26  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 /*
36  * Efficient memory file system.
37  *
38  * tmpfs is a file system that uses FreeBSD's virtual memory
39  * sub-system to store file data and metadata in an efficient way.
40  * This means that it does not follow the structure of an on-disk file
41  * system because it simply does not need to.  Instead, it uses
42  * memory-specific data structures and algorithms to automatically
43  * allocate and release resources.
44  */
45 
46 #include "opt_ddb.h"
47 #include "opt_tmpfs.h"
48 
49 #include <sys/cdefs.h>
50 __FBSDID("$FreeBSD$");
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/dirent.h>
55 #include <sys/file.h>
56 #include <sys/limits.h>
57 #include <sys/lock.h>
58 #include <sys/mount.h>
59 #include <sys/mutex.h>
60 #include <sys/proc.h>
61 #include <sys/jail.h>
62 #include <sys/kernel.h>
63 #include <sys/rwlock.h>
64 #include <sys/stat.h>
65 #include <sys/sx.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_param.h>
71 #include <vm/pmap.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_map.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_param.h>
76 
77 #include <fs/tmpfs/tmpfs.h>
78 
79 /*
80  * Default permission for root node
81  */
82 #define TMPFS_DEFAULT_ROOT_MODE	(S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
83 
84 static MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
85 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
86 
87 static int	tmpfs_mount(struct mount *);
88 static int	tmpfs_unmount(struct mount *, int);
89 static int	tmpfs_root(struct mount *, int flags, struct vnode **);
90 static int	tmpfs_fhtovp(struct mount *, struct fid *, int,
91 		    struct vnode **);
92 static int	tmpfs_statfs(struct mount *, struct statfs *);
93 
94 static const char *tmpfs_opts[] = {
95 	"from", "easize", "size", "maxfilesize", "inodes", "uid", "gid", "mode",
96 	"export", "union", "nonc", "nomtime", NULL
97 };
98 
99 static const char *tmpfs_updateopts[] = {
100 	"from", "easize", "export", "nomtime", "size", NULL
101 };
102 
103 static int
104 tmpfs_update_mtime_lazy_filter(struct vnode *vp, void *arg)
105 {
106 	struct vm_object *obj;
107 
108 	if (vp->v_type != VREG)
109 		return (0);
110 
111 	obj = atomic_load_ptr(&vp->v_object);
112 	if (obj == NULL)
113 		return (0);
114 
115 	return (vm_object_mightbedirty_(obj));
116 }
117 
118 static void
119 tmpfs_update_mtime_lazy(struct mount *mp)
120 {
121 	struct vnode *vp, *mvp;
122 
123 	MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, tmpfs_update_mtime_lazy_filter, NULL) {
124 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK) != 0)
125 			continue;
126 		tmpfs_check_mtime(vp);
127 		vput(vp);
128 	}
129 }
130 
131 static void
132 tmpfs_update_mtime_all(struct mount *mp)
133 {
134 	struct vnode *vp, *mvp;
135 
136 	if (VFS_TO_TMPFS(mp)->tm_nomtime)
137 		return;
138 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
139 		if (vp->v_type != VREG) {
140 			VI_UNLOCK(vp);
141 			continue;
142 		}
143 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK) != 0)
144 			continue;
145 		tmpfs_check_mtime(vp);
146 		tmpfs_update(vp);
147 		vput(vp);
148 	}
149 }
150 
151 struct tmpfs_check_rw_maps_arg {
152 	bool found;
153 };
154 
155 static bool
156 tmpfs_check_rw_maps_cb(struct mount *mp __unused, vm_map_t map __unused,
157     vm_map_entry_t entry __unused, void *arg)
158 {
159 	struct tmpfs_check_rw_maps_arg *a;
160 
161 	a = arg;
162 	a->found = true;
163 	return (true);
164 }
165 
166 /*
167  * Revoke write permissions from all mappings of regular files
168  * belonging to the specified tmpfs mount.
169  */
170 static bool
171 tmpfs_revoke_rw_maps_cb(struct mount *mp __unused, vm_map_t map,
172     vm_map_entry_t entry, void *arg __unused)
173 {
174 
175 	/*
176 	 * XXXKIB: might be invalidate the mapping
177 	 * instead ?  The process is not going to be
178 	 * happy in any case.
179 	 */
180 	entry->max_protection &= ~VM_PROT_WRITE;
181 	if ((entry->protection & VM_PROT_WRITE) != 0) {
182 		entry->protection &= ~VM_PROT_WRITE;
183 		pmap_protect(map->pmap, entry->start, entry->end,
184 		    entry->protection);
185 	}
186 	return (false);
187 }
188 
189 static void
190 tmpfs_all_rw_maps(struct mount *mp, bool (*cb)(struct mount *mp, vm_map_t,
191     vm_map_entry_t, void *), void *cb_arg)
192 {
193 	struct proc *p;
194 	struct vmspace *vm;
195 	vm_map_t map;
196 	vm_map_entry_t entry;
197 	vm_object_t object;
198 	struct vnode *vp;
199 	int gen;
200 	bool terminate;
201 
202 	terminate = false;
203 	sx_slock(&allproc_lock);
204 again:
205 	gen = allproc_gen;
206 	FOREACH_PROC_IN_SYSTEM(p) {
207 		PROC_LOCK(p);
208 		if (p->p_state != PRS_NORMAL || (p->p_flag & (P_INEXEC |
209 		    P_SYSTEM | P_WEXIT)) != 0) {
210 			PROC_UNLOCK(p);
211 			continue;
212 		}
213 		vm = vmspace_acquire_ref(p);
214 		_PHOLD_LITE(p);
215 		PROC_UNLOCK(p);
216 		if (vm == NULL) {
217 			PRELE(p);
218 			continue;
219 		}
220 		sx_sunlock(&allproc_lock);
221 		map = &vm->vm_map;
222 
223 		vm_map_lock(map);
224 		if (map->busy)
225 			vm_map_wait_busy(map);
226 		VM_MAP_ENTRY_FOREACH(entry, map) {
227 			if ((entry->eflags & (MAP_ENTRY_GUARD |
228 			    MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_COW)) != 0 ||
229 			    (entry->max_protection & VM_PROT_WRITE) == 0)
230 				continue;
231 			object = entry->object.vm_object;
232 			if (object == NULL || object->type != tmpfs_pager_type)
233 				continue;
234 			/*
235 			 * No need to dig into shadow chain, mapping
236 			 * of the object not at top is readonly.
237 			 */
238 
239 			VM_OBJECT_RLOCK(object);
240 			if (object->type == OBJT_DEAD) {
241 				VM_OBJECT_RUNLOCK(object);
242 				continue;
243 			}
244 			MPASS(object->ref_count > 1);
245 			if ((object->flags & OBJ_TMPFS) == 0) {
246 				VM_OBJECT_RUNLOCK(object);
247 				continue;
248 			}
249 			vp = VM_TO_TMPFS_VP(object);
250 			if (vp->v_mount != mp) {
251 				VM_OBJECT_RUNLOCK(object);
252 				continue;
253 			}
254 
255 			terminate = cb(mp, map, entry, cb_arg);
256 			VM_OBJECT_RUNLOCK(object);
257 			if (terminate)
258 				break;
259 		}
260 		vm_map_unlock(map);
261 
262 		vmspace_free(vm);
263 		sx_slock(&allproc_lock);
264 		PRELE(p);
265 		if (terminate)
266 			break;
267 	}
268 	if (!terminate && gen != allproc_gen)
269 		goto again;
270 	sx_sunlock(&allproc_lock);
271 }
272 
273 static bool
274 tmpfs_check_rw_maps(struct mount *mp)
275 {
276 	struct tmpfs_check_rw_maps_arg ca;
277 
278 	ca.found = false;
279 	tmpfs_all_rw_maps(mp, tmpfs_check_rw_maps_cb, &ca);
280 	return (ca.found);
281 }
282 
283 static int
284 tmpfs_rw_to_ro(struct mount *mp)
285 {
286 	int error, flags;
287 	bool forced;
288 
289 	forced = (mp->mnt_flag & MNT_FORCE) != 0;
290 	flags = WRITECLOSE | (forced ? FORCECLOSE : 0);
291 
292 	if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
293 		return (error);
294 	error = vfs_write_suspend_umnt(mp);
295 	if (error != 0)
296 		return (error);
297 	if (!forced && tmpfs_check_rw_maps(mp)) {
298 		error = EBUSY;
299 		goto out;
300 	}
301 	VFS_TO_TMPFS(mp)->tm_ronly = 1;
302 	MNT_ILOCK(mp);
303 	mp->mnt_flag |= MNT_RDONLY;
304 	MNT_IUNLOCK(mp);
305 	for (;;) {
306 		tmpfs_all_rw_maps(mp, tmpfs_revoke_rw_maps_cb, NULL);
307 		tmpfs_update_mtime_all(mp);
308 		error = vflush(mp, 0, flags, curthread);
309 		if (error != 0) {
310 			VFS_TO_TMPFS(mp)->tm_ronly = 0;
311 			MNT_ILOCK(mp);
312 			mp->mnt_flag &= ~MNT_RDONLY;
313 			MNT_IUNLOCK(mp);
314 			goto out;
315 		}
316 		if (!tmpfs_check_rw_maps(mp))
317 			break;
318 	}
319 out:
320 	vfs_write_resume(mp, 0);
321 	return (error);
322 }
323 
324 static int
325 tmpfs_mount(struct mount *mp)
326 {
327 	const size_t nodes_per_page = howmany(PAGE_SIZE,
328 	    sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
329 	struct tmpfs_mount *tmp;
330 	struct tmpfs_node *root;
331 	int error;
332 	bool nomtime, nonc;
333 	/* Size counters. */
334 	u_quad_t pages;
335 	off_t nodes_max, size_max, maxfilesize, ea_max_size;
336 
337 	/* Root node attributes. */
338 	uid_t root_uid;
339 	gid_t root_gid;
340 	mode_t root_mode;
341 
342 	struct vattr va;
343 
344 	if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
345 		return (EINVAL);
346 
347 	if (mp->mnt_flag & MNT_UPDATE) {
348 		/* Only support update mounts for certain options. */
349 		if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
350 			return (EOPNOTSUPP);
351 		tmp = VFS_TO_TMPFS(mp);
352 		if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) == 0) {
353 			/*
354 			 * On-the-fly resizing is not supported (yet). We still
355 			 * need to have "size" listed as "supported", otherwise
356 			 * trying to update fs that is listed in fstab with size
357 			 * parameter, say trying to change rw to ro or vice
358 			 * versa, would cause vfs_filteropt() to bail.
359 			 */
360 			if (size_max != tmp->tm_size_max)
361 				return (EOPNOTSUPP);
362 		}
363 		if (vfs_getopt_size(mp->mnt_optnew, "easize", &ea_max_size) == 0) {
364 			tmp->tm_ea_memory_max = ea_max_size;
365 		}
366 		if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
367 		    !tmp->tm_ronly) {
368 			/* RW -> RO */
369 			return (tmpfs_rw_to_ro(mp));
370 		} else if (!vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
371 		    tmp->tm_ronly) {
372 			/* RO -> RW */
373 			tmp->tm_ronly = 0;
374 			MNT_ILOCK(mp);
375 			mp->mnt_flag &= ~MNT_RDONLY;
376 			MNT_IUNLOCK(mp);
377 		}
378 		tmp->tm_nomtime = vfs_getopt(mp->mnt_optnew, "nomtime", NULL,
379 		    0) == 0;
380 		MNT_ILOCK(mp);
381 		if ((mp->mnt_flag & MNT_UNION) == 0) {
382 			mp->mnt_kern_flag |= MNTK_FPLOOKUP;
383 		} else {
384 			mp->mnt_kern_flag &= ~MNTK_FPLOOKUP;
385 		}
386 		MNT_IUNLOCK(mp);
387 		return (0);
388 	}
389 
390 	vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
391 	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
392 	VOP_UNLOCK(mp->mnt_vnodecovered);
393 	if (error)
394 		return (error);
395 
396 	if (mp->mnt_cred->cr_ruid != 0 ||
397 	    vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
398 		root_gid = va.va_gid;
399 	if (mp->mnt_cred->cr_ruid != 0 ||
400 	    vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
401 		root_uid = va.va_uid;
402 	if (mp->mnt_cred->cr_ruid != 0 ||
403 	    vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
404 		root_mode = va.va_mode;
405 	if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
406 		nodes_max = 0;
407 	if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
408 		size_max = 0;
409 	if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
410 		maxfilesize = 0;
411 	if (vfs_getopt_size(mp->mnt_optnew, "easize", &ea_max_size) != 0)
412 		ea_max_size = 0;
413 	nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0;
414 	nomtime = vfs_getopt(mp->mnt_optnew, "nomtime", NULL, NULL) == 0;
415 
416 	/* Do not allow mounts if we do not have enough memory to preserve
417 	 * the minimum reserved pages. */
418 	if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
419 		return (ENOSPC);
420 
421 	/* Get the maximum number of memory pages this file system is
422 	 * allowed to use, based on the maximum size the user passed in
423 	 * the mount structure.  A value of zero is treated as if the
424 	 * maximum available space was requested. */
425 	if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
426 	    (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
427 		pages = SIZE_MAX;
428 	else {
429 		size_max = roundup(size_max, PAGE_SIZE);
430 		pages = howmany(size_max, PAGE_SIZE);
431 	}
432 	MPASS(pages > 0);
433 
434 	if (nodes_max <= 3) {
435 		if (pages < INT_MAX / nodes_per_page)
436 			nodes_max = pages * nodes_per_page;
437 		else
438 			nodes_max = INT_MAX;
439 	}
440 	if (nodes_max > INT_MAX)
441 		nodes_max = INT_MAX;
442 	MPASS(nodes_max >= 3);
443 
444 	/* Allocate the tmpfs mount structure and fill it. */
445 	tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
446 	    M_TMPFSMNT, M_WAITOK | M_ZERO);
447 
448 	mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
449 	tmp->tm_nodes_max = nodes_max;
450 	tmp->tm_nodes_inuse = 0;
451 	tmp->tm_ea_memory_inuse = 0;
452 	tmp->tm_refcount = 1;
453 	tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
454 	tmp->tm_ea_memory_max = ea_max_size > 0 ?
455 	    ea_max_size : TMPFS_EA_MEMORY_RESERVED;
456 	LIST_INIT(&tmp->tm_nodes_used);
457 
458 	tmp->tm_size_max = size_max;
459 	tmp->tm_pages_max = pages;
460 	tmp->tm_pages_used = 0;
461 	new_unrhdr64(&tmp->tm_ino_unr, 2);
462 	tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
463 	tmp->tm_nonc = nonc;
464 	tmp->tm_nomtime = nomtime;
465 
466 	/* Allocate the root node. */
467 	error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid,
468 	    root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root);
469 
470 	if (error != 0 || root == NULL) {
471 		free(tmp, M_TMPFSMNT);
472 		return (error);
473 	}
474 	KASSERT(root->tn_id == 2,
475 	    ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
476 	tmp->tm_root = root;
477 
478 	MNT_ILOCK(mp);
479 	mp->mnt_flag |= MNT_LOCAL;
480 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
481 	    MNTK_NOMSYNC;
482 	if (!nonc && (mp->mnt_flag & MNT_UNION) == 0)
483 		mp->mnt_kern_flag |= MNTK_FPLOOKUP;
484 	MNT_IUNLOCK(mp);
485 
486 	mp->mnt_data = tmp;
487 	mp->mnt_stat.f_namemax = MAXNAMLEN;
488 	vfs_getnewfsid(mp);
489 	vfs_mountedfrom(mp, "tmpfs");
490 
491 	return (0);
492 }
493 
494 /* ARGSUSED2 */
495 static int
496 tmpfs_unmount(struct mount *mp, int mntflags)
497 {
498 	struct tmpfs_mount *tmp;
499 	struct tmpfs_node *node;
500 	int error, flags;
501 
502 	flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0;
503 	tmp = VFS_TO_TMPFS(mp);
504 
505 	/* Stop writers */
506 	error = vfs_write_suspend_umnt(mp);
507 	if (error != 0)
508 		return (error);
509 	/*
510 	 * At this point, nodes cannot be destroyed by any other
511 	 * thread because write suspension is started.
512 	 */
513 
514 	for (;;) {
515 		error = vflush(mp, 0, flags, curthread);
516 		if (error != 0) {
517 			vfs_write_resume(mp, VR_START_WRITE);
518 			return (error);
519 		}
520 		MNT_ILOCK(mp);
521 		if (mp->mnt_nvnodelistsize == 0) {
522 			MNT_IUNLOCK(mp);
523 			break;
524 		}
525 		MNT_IUNLOCK(mp);
526 		if ((mntflags & MNT_FORCE) == 0) {
527 			vfs_write_resume(mp, VR_START_WRITE);
528 			return (EBUSY);
529 		}
530 	}
531 
532 	TMPFS_LOCK(tmp);
533 	while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
534 		TMPFS_NODE_LOCK(node);
535 		if (node->tn_type == VDIR)
536 			tmpfs_dir_destroy(tmp, node);
537 		if (tmpfs_free_node_locked(tmp, node, true))
538 			TMPFS_LOCK(tmp);
539 		else
540 			TMPFS_NODE_UNLOCK(node);
541 	}
542 
543 	mp->mnt_data = NULL;
544 	tmpfs_free_tmp(tmp);
545 	vfs_write_resume(mp, VR_START_WRITE);
546 
547 	return (0);
548 }
549 
550 void
551 tmpfs_free_tmp(struct tmpfs_mount *tmp)
552 {
553 	TMPFS_MP_ASSERT_LOCKED(tmp);
554 	MPASS(tmp->tm_refcount > 0);
555 
556 	tmp->tm_refcount--;
557 	if (tmp->tm_refcount > 0) {
558 		TMPFS_UNLOCK(tmp);
559 		return;
560 	}
561 	TMPFS_UNLOCK(tmp);
562 
563 	mtx_destroy(&tmp->tm_allnode_lock);
564 	/*
565 	 * We cannot assert that tmp->tm_pages_used == 0 there,
566 	 * because tmpfs vm_objects might be still mapped by some
567 	 * process and outlive the mount due to reference counting.
568 	 */
569 	MPASS(tmp->tm_nodes_inuse == 0);
570 
571 	free(tmp, M_TMPFSMNT);
572 }
573 
574 static int
575 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
576 {
577 	int error;
578 
579 	error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
580 	if (error == 0)
581 		(*vpp)->v_vflag |= VV_ROOT;
582 	return (error);
583 }
584 
585 static int
586 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
587     struct vnode **vpp)
588 {
589 	struct tmpfs_fid_data tfd;
590 	struct tmpfs_mount *tmp;
591 	struct tmpfs_node *node;
592 	int error;
593 
594 	if (fhp->fid_len != sizeof(tfd))
595 		return (EINVAL);
596 
597 	/*
598 	 * Copy from fid_data onto the stack to avoid unaligned pointer use.
599 	 * See the comment in sys/mount.h on struct fid for details.
600 	 */
601 	memcpy(&tfd, fhp->fid_data, fhp->fid_len);
602 
603 	tmp = VFS_TO_TMPFS(mp);
604 
605 	if (tfd.tfd_id >= tmp->tm_nodes_max)
606 		return (EINVAL);
607 
608 	TMPFS_LOCK(tmp);
609 	LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
610 		if (node->tn_id == tfd.tfd_id &&
611 		    node->tn_gen == tfd.tfd_gen) {
612 			tmpfs_ref_node(node);
613 			break;
614 		}
615 	}
616 	TMPFS_UNLOCK(tmp);
617 
618 	if (node != NULL) {
619 		error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
620 		tmpfs_free_node(tmp, node);
621 	} else
622 		error = EINVAL;
623 	return (error);
624 }
625 
626 /* ARGSUSED2 */
627 static int
628 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
629 {
630 	struct tmpfs_mount *tmp;
631 	size_t used;
632 
633 	tmp = VFS_TO_TMPFS(mp);
634 
635 	sbp->f_iosize = PAGE_SIZE;
636 	sbp->f_bsize = PAGE_SIZE;
637 
638 	used = tmpfs_pages_used(tmp);
639 	if (tmp->tm_pages_max != ULONG_MAX)
640 		 sbp->f_blocks = tmp->tm_pages_max;
641 	else
642 		 sbp->f_blocks = used + tmpfs_mem_avail();
643 	if (sbp->f_blocks <= used)
644 		sbp->f_bavail = 0;
645 	else
646 		sbp->f_bavail = sbp->f_blocks - used;
647 	sbp->f_bfree = sbp->f_bavail;
648 	used = tmp->tm_nodes_inuse;
649 	sbp->f_files = tmp->tm_nodes_max;
650 	if (sbp->f_files <= used)
651 		sbp->f_ffree = 0;
652 	else
653 		sbp->f_ffree = sbp->f_files - used;
654 	/* sbp->f_owner = tmp->tn_uid; */
655 
656 	return (0);
657 }
658 
659 static int
660 tmpfs_sync(struct mount *mp, int waitfor)
661 {
662 
663 	if (waitfor == MNT_SUSPEND) {
664 		MNT_ILOCK(mp);
665 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
666 		MNT_IUNLOCK(mp);
667 	} else if (waitfor == MNT_LAZY) {
668 		tmpfs_update_mtime_lazy(mp);
669 	}
670 	return (0);
671 }
672 
673 static int
674 tmpfs_init(struct vfsconf *conf)
675 {
676 	int res;
677 
678 	res = tmpfs_subr_init();
679 	if (res != 0)
680 		return (res);
681 	memcpy(&tmpfs_fnops, &vnops, sizeof(struct fileops));
682 	tmpfs_fnops.fo_close = tmpfs_fo_close;
683 	return (0);
684 }
685 
686 static int
687 tmpfs_uninit(struct vfsconf *conf)
688 {
689 	tmpfs_subr_uninit();
690 	return (0);
691 }
692 
693 /*
694  * tmpfs vfs operations.
695  */
696 struct vfsops tmpfs_vfsops = {
697 	.vfs_mount =			tmpfs_mount,
698 	.vfs_unmount =			tmpfs_unmount,
699 	.vfs_root =			vfs_cache_root,
700 	.vfs_cachedroot =		tmpfs_root,
701 	.vfs_statfs =			tmpfs_statfs,
702 	.vfs_fhtovp =			tmpfs_fhtovp,
703 	.vfs_sync =			tmpfs_sync,
704 	.vfs_init =			tmpfs_init,
705 	.vfs_uninit =			tmpfs_uninit,
706 };
707 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);
708 
709 #ifdef DDB
710 #include <ddb/ddb.h>
711 
712 static void
713 db_print_tmpfs(struct mount *mp, struct tmpfs_mount *tmp)
714 {
715 	db_printf("mp %p (%s) tmp %p\n", mp,
716 	    mp->mnt_stat.f_mntonname, tmp);
717 	db_printf(
718 	    "\tsize max %ju pages max %lu pages used %lu\n"
719 	    "\tinodes max %ju inodes inuse %ju ea inuse %ju refcount %ju\n"
720 	    "\tmaxfilesize %ju r%c %snamecache %smtime\n",
721 	    (uintmax_t)tmp->tm_size_max, tmp->tm_pages_max, tmp->tm_pages_used,
722 	    (uintmax_t)tmp->tm_nodes_max, (uintmax_t)tmp->tm_nodes_inuse,
723 	    (uintmax_t)tmp->tm_ea_memory_inuse, (uintmax_t)tmp->tm_refcount,
724 	    (uintmax_t)tmp->tm_maxfilesize,
725 	    tmp->tm_ronly ? 'o' : 'w', tmp->tm_nonc ? "no" : "",
726 	    tmp->tm_nomtime ? "no" : "");
727 }
728 
729 DB_SHOW_COMMAND(tmpfs, db_show_tmpfs)
730 {
731 	struct mount *mp;
732 	struct tmpfs_mount *tmp;
733 
734 	if (have_addr) {
735 		mp = (struct mount *)addr;
736 		tmp = VFS_TO_TMPFS(mp);
737 		db_print_tmpfs(mp, tmp);
738 		return;
739 	}
740 
741 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
742 		if (strcmp(mp->mnt_stat.f_fstypename, tmpfs_vfsconf.vfc_name) ==
743 		    0) {
744 			tmp = VFS_TO_TMPFS(mp);
745 			db_print_tmpfs(mp, tmp);
746 		}
747 	}
748 }
749 #endif	/* DDB */
750