xref: /freebsd/sys/fs/tmpfs/tmpfs_vfsops.c (revision 1cfc208a07b64a64377a4f8fe51c28f588a929cf)
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
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/param.h>
50 #include <sys/systm.h>
51 #include <sys/dirent.h>
52 #include <sys/file.h>
53 #include <sys/limits.h>
54 #include <sys/lock.h>
55 #include <sys/mount.h>
56 #include <sys/mutex.h>
57 #include <sys/proc.h>
58 #include <sys/jail.h>
59 #include <sys/kernel.h>
60 #include <sys/rwlock.h>
61 #include <sys/stat.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/vnode.h>
65 
66 #include <vm/vm.h>
67 #include <vm/vm_param.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_param.h>
73 
74 #include <fs/tmpfs/tmpfs.h>
75 
76 /*
77  * Default permission for root node
78  */
79 #define TMPFS_DEFAULT_ROOT_MODE	(S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
80 
81 static MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
82 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
83 
84 static int	tmpfs_mount(struct mount *);
85 static int	tmpfs_unmount(struct mount *, int);
86 static int	tmpfs_root(struct mount *, int flags, struct vnode **);
87 static int	tmpfs_fhtovp(struct mount *, struct fid *, int,
88 		    struct vnode **);
89 static int	tmpfs_statfs(struct mount *, struct statfs *);
90 
91 static const char *tmpfs_opts[] = {
92 	"from", "easize", "size", "maxfilesize", "inodes", "uid", "gid", "mode",
93 	"export", "union", "nonc", "nomtime", "nosymfollow", "pgread", NULL
94 };
95 
96 static int
97 tmpfs_update_mtime_lazy_filter(struct vnode *vp, void *arg)
98 {
99 	struct vm_object *obj;
100 
101 	if (vp->v_type != VREG)
102 		return (0);
103 
104 	obj = atomic_load_ptr(&vp->v_object);
105 	if (obj == NULL)
106 		return (0);
107 
108 	return (vm_object_mightbedirty_(obj));
109 }
110 
111 static void
112 tmpfs_update_mtime_lazy(struct mount *mp)
113 {
114 	struct vnode *vp, *mvp;
115 
116 	MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, tmpfs_update_mtime_lazy_filter, NULL) {
117 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK) != 0)
118 			continue;
119 		tmpfs_check_mtime(vp);
120 		vput(vp);
121 	}
122 }
123 
124 static void
125 tmpfs_update_mtime_all(struct mount *mp)
126 {
127 	struct vnode *vp, *mvp;
128 
129 	if (VFS_TO_TMPFS(mp)->tm_nomtime)
130 		return;
131 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
132 		if (vp->v_type != VREG) {
133 			VI_UNLOCK(vp);
134 			continue;
135 		}
136 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK) != 0)
137 			continue;
138 		tmpfs_check_mtime(vp);
139 		tmpfs_update(vp);
140 		vput(vp);
141 	}
142 }
143 
144 struct tmpfs_check_rw_maps_arg {
145 	bool found;
146 };
147 
148 static bool
149 tmpfs_check_rw_maps_cb(struct mount *mp __unused, vm_map_t map __unused,
150     vm_map_entry_t entry __unused, void *arg)
151 {
152 	struct tmpfs_check_rw_maps_arg *a;
153 
154 	a = arg;
155 	a->found = true;
156 	return (true);
157 }
158 
159 /*
160  * Revoke write permissions from all mappings of regular files
161  * belonging to the specified tmpfs mount.
162  */
163 static bool
164 tmpfs_revoke_rw_maps_cb(struct mount *mp __unused, vm_map_t map,
165     vm_map_entry_t entry, void *arg __unused)
166 {
167 
168 	/*
169 	 * XXXKIB: might be invalidate the mapping
170 	 * instead ?  The process is not going to be
171 	 * happy in any case.
172 	 */
173 	entry->max_protection &= ~VM_PROT_WRITE;
174 	if ((entry->protection & VM_PROT_WRITE) != 0) {
175 		entry->protection &= ~VM_PROT_WRITE;
176 		pmap_protect(map->pmap, entry->start, entry->end,
177 		    entry->protection);
178 	}
179 	return (false);
180 }
181 
182 static void
183 tmpfs_all_rw_maps(struct mount *mp, bool (*cb)(struct mount *mp, vm_map_t,
184     vm_map_entry_t, void *), void *cb_arg)
185 {
186 	struct proc *p;
187 	struct vmspace *vm;
188 	vm_map_t map;
189 	vm_map_entry_t entry;
190 	vm_object_t object;
191 	struct vnode *vp;
192 	int gen;
193 	bool terminate;
194 
195 	terminate = false;
196 	sx_slock(&allproc_lock);
197 again:
198 	gen = allproc_gen;
199 	FOREACH_PROC_IN_SYSTEM(p) {
200 		PROC_LOCK(p);
201 		if (p->p_state != PRS_NORMAL || (p->p_flag & (P_INEXEC |
202 		    P_SYSTEM | P_WEXIT)) != 0) {
203 			PROC_UNLOCK(p);
204 			continue;
205 		}
206 		vm = vmspace_acquire_ref(p);
207 		_PHOLD(p);
208 		PROC_UNLOCK(p);
209 		if (vm == NULL) {
210 			PRELE(p);
211 			continue;
212 		}
213 		sx_sunlock(&allproc_lock);
214 		map = &vm->vm_map;
215 
216 		vm_map_lock(map);
217 		if (map->busy)
218 			vm_map_wait_busy(map);
219 		VM_MAP_ENTRY_FOREACH(entry, map) {
220 			if ((entry->eflags & (MAP_ENTRY_GUARD |
221 			    MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_COW)) != 0 ||
222 			    (entry->max_protection & VM_PROT_WRITE) == 0)
223 				continue;
224 			object = entry->object.vm_object;
225 			if (object == NULL || object->type != tmpfs_pager_type)
226 				continue;
227 			/*
228 			 * No need to dig into shadow chain, mapping
229 			 * of the object not at top is readonly.
230 			 */
231 
232 			VM_OBJECT_RLOCK(object);
233 			if (object->type == OBJT_DEAD) {
234 				VM_OBJECT_RUNLOCK(object);
235 				continue;
236 			}
237 			MPASS(object->ref_count > 1);
238 			if ((object->flags & OBJ_TMPFS) == 0) {
239 				VM_OBJECT_RUNLOCK(object);
240 				continue;
241 			}
242 			vp = VM_TO_TMPFS_VP(object);
243 			if (vp->v_mount != mp) {
244 				VM_OBJECT_RUNLOCK(object);
245 				continue;
246 			}
247 
248 			terminate = cb(mp, map, entry, cb_arg);
249 			VM_OBJECT_RUNLOCK(object);
250 			if (terminate)
251 				break;
252 		}
253 		vm_map_unlock(map);
254 
255 		vmspace_free(vm);
256 		sx_slock(&allproc_lock);
257 		PRELE(p);
258 		if (terminate)
259 			break;
260 	}
261 	if (!terminate && gen != allproc_gen)
262 		goto again;
263 	sx_sunlock(&allproc_lock);
264 }
265 
266 static bool
267 tmpfs_check_rw_maps(struct mount *mp)
268 {
269 	struct tmpfs_check_rw_maps_arg ca;
270 
271 	ca.found = false;
272 	tmpfs_all_rw_maps(mp, tmpfs_check_rw_maps_cb, &ca);
273 	return (ca.found);
274 }
275 
276 static int
277 tmpfs_rw_to_ro(struct mount *mp)
278 {
279 	int error, flags;
280 	bool forced;
281 
282 	forced = (mp->mnt_flag & MNT_FORCE) != 0;
283 	flags = WRITECLOSE | (forced ? FORCECLOSE : 0);
284 
285 	if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
286 		return (error);
287 	error = vfs_write_suspend_umnt(mp);
288 	if (error != 0)
289 		return (error);
290 	if (!forced && tmpfs_check_rw_maps(mp)) {
291 		error = EBUSY;
292 		goto out;
293 	}
294 	VFS_TO_TMPFS(mp)->tm_ronly = 1;
295 	MNT_ILOCK(mp);
296 	mp->mnt_flag |= MNT_RDONLY;
297 	MNT_IUNLOCK(mp);
298 	for (;;) {
299 		tmpfs_all_rw_maps(mp, tmpfs_revoke_rw_maps_cb, NULL);
300 		tmpfs_update_mtime_all(mp);
301 		error = vflush(mp, 0, flags, curthread);
302 		if (error != 0) {
303 			VFS_TO_TMPFS(mp)->tm_ronly = 0;
304 			MNT_ILOCK(mp);
305 			mp->mnt_flag &= ~MNT_RDONLY;
306 			MNT_IUNLOCK(mp);
307 			goto out;
308 		}
309 		if (!tmpfs_check_rw_maps(mp))
310 			break;
311 	}
312 out:
313 	vfs_write_resume(mp, 0);
314 	return (error);
315 }
316 
317 static int
318 tmpfs_mount(struct mount *mp)
319 {
320 	const size_t nodes_per_page = howmany(PAGE_SIZE,
321 	    sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
322 	struct tmpfs_mount *tmp;
323 	struct tmpfs_node *root;
324 	struct vfsoptlist *opts;
325 	int error;
326 	bool nomtime, nonc, pgread;
327 	/* Size counters. */
328 	u_quad_t pages;
329 	off_t nodes_max, size_max, maxfilesize, ea_max_size;
330 
331 	/* Root node attributes. */
332 	uid_t root_uid;
333 	gid_t root_gid;
334 	mode_t root_mode;
335 
336 	struct vattr va;
337 
338 	opts = mp->mnt_optnew;
339 	if (vfs_filteropt(opts, tmpfs_opts))
340 		return (EINVAL);
341 
342 	if (mp->mnt_flag & MNT_UPDATE) {
343 		tmp = VFS_TO_TMPFS(mp);
344 
345 		/*
346 		 * These options cannot (yet) be modified on the fly, but
347 		 * mount(8) will still pass them when remounting, so we
348 		 * will silently ignore them as long as the value is
349 		 * unchanged.
350 		 */
351 		if (vfs_scanopt(opts, "gid", "%d", &root_gid) == 1 &&
352 		    root_gid != tmp->tm_root->tn_gid)
353 			return (EOPNOTSUPP);
354 		if (vfs_scanopt(opts, "uid", "%d", &root_uid) == 1 &&
355 		    root_uid != tmp->tm_root->tn_uid)
356 			return (EOPNOTSUPP);
357 		if (vfs_scanopt(opts, "mode", "%ho", &root_mode) == 1 &&
358 		    (root_mode & S_IFMT) != (tmp->tm_root->tn_mode & S_IFMT))
359 			return (EOPNOTSUPP);
360 		if (vfs_getopt_size(opts, "inodes", &nodes_max) == 0 &&
361 		    nodes_max != 0 && nodes_max != tmp->tm_nodes_max)
362 			return (EOPNOTSUPP);
363 		if (vfs_getopt_size(opts, "size", &size_max) == 0 &&
364 		    size_max != 0 && size_max != tmp->tm_size_max)
365 			return (EOPNOTSUPP);
366 		if (vfs_getopt_size(opts, "maxfilesize", &maxfilesize) == 0 &&
367 		    maxfilesize != 0 && maxfilesize != tmp->tm_maxfilesize)
368 			return (EOPNOTSUPP);
369 		if (tmp->tm_nonc !=
370 		    (vfs_getopt(opts, "nonc", NULL, NULL) == 0))
371 			return (EOPNOTSUPP);
372 		if (tmp->tm_pgread !=
373 		    (vfs_getopt(opts, "pgread", NULL, NULL) == 0))
374 			return (EOPNOTSUPP);
375 
376 		/*
377 		 * These options can be modified.
378 		 */
379 		if (vfs_getopt_size(opts, "easize", &ea_max_size) != 0)
380 			tmp->tm_ea_memory_max = ea_max_size;
381 		tmp->tm_nomtime = (vfs_getopt(opts, "nomtime", NULL, 0) == 0);
382 
383 		/*
384 		 * Handle read-write to read-only or vice versa.
385 		 */
386 		if (vfs_flagopt(opts, "ro", NULL, 0) && !tmp->tm_ronly) {
387 			/* RW -> RO */
388 			return (tmpfs_rw_to_ro(mp));
389 		}
390 		if (!vfs_flagopt(opts, "ro", NULL, 0) && tmp->tm_ronly) {
391 			/* RO -> RW */
392 			tmp->tm_ronly = 0;
393 			MNT_ILOCK(mp);
394 			mp->mnt_flag &= ~MNT_RDONLY;
395 			MNT_IUNLOCK(mp);
396 		}
397 
398 		/*
399 		 * Check if fast path lookup is still supported.
400 		 */
401 		MNT_ILOCK(mp);
402 		if (!tmp->tm_nonc && (mp->mnt_flag & MNT_UNION) == 0) {
403 			mp->mnt_kern_flag |= MNTK_FPLOOKUP;
404 		} else {
405 			mp->mnt_kern_flag &= ~MNTK_FPLOOKUP;
406 		}
407 		MNT_IUNLOCK(mp);
408 		return (0);
409 	}
410 
411 	vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
412 	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
413 	VOP_UNLOCK(mp->mnt_vnodecovered);
414 	if (error)
415 		return (error);
416 
417 	if (mp->mnt_cred->cr_ruid != 0 ||
418 	    vfs_scanopt(opts, "gid", "%d", &root_gid) != 1)
419 		root_gid = va.va_gid;
420 	if (mp->mnt_cred->cr_ruid != 0 ||
421 	    vfs_scanopt(opts, "uid", "%d", &root_uid) != 1)
422 		root_uid = va.va_uid;
423 	if (mp->mnt_cred->cr_ruid != 0 ||
424 	    vfs_scanopt(opts, "mode", "%ho", &root_mode) != 1)
425 		root_mode = va.va_mode;
426 	if (vfs_getopt_size(opts, "inodes", &nodes_max) != 0)
427 		nodes_max = 0;
428 	if (vfs_getopt_size(opts, "size", &size_max) != 0)
429 		size_max = 0;
430 	if (vfs_getopt_size(opts, "maxfilesize", &maxfilesize) != 0)
431 		maxfilesize = 0;
432 	if (vfs_getopt_size(opts, "easize", &ea_max_size) != 0)
433 		ea_max_size = 0;
434 	nonc = vfs_getopt(opts, "nonc", NULL, NULL) == 0;
435 	nomtime = vfs_getopt(opts, "nomtime", NULL, NULL) == 0;
436 	pgread = vfs_getopt(opts, "pgread", NULL, NULL) == 0;
437 
438 	/* Do not allow mounts if we do not have enough memory to preserve
439 	 * the minimum reserved pages. */
440 	if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
441 		return (ENOSPC);
442 
443 	/* Get the maximum number of memory pages this file system is
444 	 * allowed to use, based on the maximum size the user passed in
445 	 * the mount structure.  A value of zero is treated as if the
446 	 * maximum available space was requested. */
447 	if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
448 	    (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
449 		pages = SIZE_MAX;
450 	else {
451 		size_max = roundup(size_max, PAGE_SIZE);
452 		pages = howmany(size_max, PAGE_SIZE);
453 	}
454 	MPASS(pages > 0);
455 
456 	if (nodes_max <= 3) {
457 		if (pages < INT_MAX / nodes_per_page)
458 			nodes_max = pages * nodes_per_page;
459 		else
460 			nodes_max = INT_MAX;
461 	}
462 	if (nodes_max > INT_MAX)
463 		nodes_max = INT_MAX;
464 	MPASS(nodes_max >= 3);
465 
466 	/* Allocate the tmpfs mount structure and fill it. */
467 	tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
468 	    M_TMPFSMNT, M_WAITOK | M_ZERO);
469 
470 	mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
471 	tmp->tm_nodes_max = nodes_max;
472 	tmp->tm_nodes_inuse = 0;
473 	tmp->tm_ea_memory_inuse = 0;
474 	tmp->tm_refcount = 1;
475 	tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
476 	tmp->tm_ea_memory_max = ea_max_size > 0 ?
477 	    ea_max_size : TMPFS_EA_MEMORY_RESERVED;
478 	LIST_INIT(&tmp->tm_nodes_used);
479 
480 	tmp->tm_size_max = size_max;
481 	tmp->tm_pages_max = pages;
482 	tmp->tm_pages_used = 0;
483 	new_unrhdr64(&tmp->tm_ino_unr, 2);
484 	tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
485 	tmp->tm_nonc = nonc;
486 	tmp->tm_nomtime = nomtime;
487 	tmp->tm_pgread = pgread;
488 
489 	/* Allocate the root node. */
490 	error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid,
491 	    root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root);
492 
493 	if (error != 0 || root == NULL) {
494 		free(tmp, M_TMPFSMNT);
495 		return (error);
496 	}
497 	KASSERT(root->tn_id == 2,
498 	    ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
499 	tmp->tm_root = root;
500 
501 	MNT_ILOCK(mp);
502 	mp->mnt_flag |= MNT_LOCAL;
503 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
504 	    MNTK_NOMSYNC;
505 	if (!nonc && (mp->mnt_flag & MNT_UNION) == 0)
506 		mp->mnt_kern_flag |= MNTK_FPLOOKUP;
507 	MNT_IUNLOCK(mp);
508 
509 	mp->mnt_data = tmp;
510 	mp->mnt_stat.f_namemax = MAXNAMLEN;
511 	vfs_getnewfsid(mp);
512 	vfs_mountedfrom(mp, "tmpfs");
513 
514 	return (0);
515 }
516 
517 /* ARGSUSED2 */
518 static int
519 tmpfs_unmount(struct mount *mp, int mntflags)
520 {
521 	struct tmpfs_mount *tmp;
522 	struct tmpfs_node *node;
523 	int error, flags;
524 
525 	flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0;
526 	tmp = VFS_TO_TMPFS(mp);
527 
528 	/* Stop writers */
529 	error = vfs_write_suspend_umnt(mp);
530 	if (error != 0)
531 		return (error);
532 	/*
533 	 * At this point, nodes cannot be destroyed by any other
534 	 * thread because write suspension is started.
535 	 */
536 
537 	for (;;) {
538 		error = vflush(mp, 0, flags, curthread);
539 		if (error != 0) {
540 			vfs_write_resume(mp, VR_START_WRITE);
541 			return (error);
542 		}
543 		MNT_ILOCK(mp);
544 		if (mp->mnt_nvnodelistsize == 0) {
545 			MNT_IUNLOCK(mp);
546 			break;
547 		}
548 		MNT_IUNLOCK(mp);
549 		if ((mntflags & MNT_FORCE) == 0) {
550 			vfs_write_resume(mp, VR_START_WRITE);
551 			return (EBUSY);
552 		}
553 	}
554 
555 	TMPFS_LOCK(tmp);
556 	while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
557 		TMPFS_NODE_LOCK(node);
558 		if (node->tn_type == VDIR)
559 			tmpfs_dir_destroy(tmp, node);
560 		if (tmpfs_free_node_locked(tmp, node, true))
561 			TMPFS_LOCK(tmp);
562 		else
563 			TMPFS_NODE_UNLOCK(node);
564 	}
565 
566 	mp->mnt_data = NULL;
567 	tmpfs_free_tmp(tmp);
568 	vfs_write_resume(mp, VR_START_WRITE);
569 
570 	return (0);
571 }
572 
573 void
574 tmpfs_free_tmp(struct tmpfs_mount *tmp)
575 {
576 	TMPFS_MP_ASSERT_LOCKED(tmp);
577 	MPASS(tmp->tm_refcount > 0);
578 
579 	tmp->tm_refcount--;
580 	if (tmp->tm_refcount > 0) {
581 		TMPFS_UNLOCK(tmp);
582 		return;
583 	}
584 	TMPFS_UNLOCK(tmp);
585 
586 	mtx_destroy(&tmp->tm_allnode_lock);
587 	/*
588 	 * We cannot assert that tmp->tm_pages_used == 0 there,
589 	 * because tmpfs vm_objects might be still mapped by some
590 	 * process and outlive the mount due to reference counting.
591 	 */
592 	MPASS(tmp->tm_nodes_inuse == 0);
593 
594 	free(tmp, M_TMPFSMNT);
595 }
596 
597 static int
598 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
599 {
600 	int error;
601 
602 	error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
603 	if (error == 0)
604 		(*vpp)->v_vflag |= VV_ROOT;
605 	return (error);
606 }
607 
608 static int
609 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
610     struct vnode **vpp)
611 {
612 	struct tmpfs_fid_data *tfd;
613 	struct tmpfs_mount *tmp;
614 	struct tmpfs_node *node;
615 	int error;
616 
617 	if (fhp->fid_len != sizeof(*tfd))
618 		return (EINVAL);
619 
620 	tfd = (struct tmpfs_fid_data *)fhp;
621 
622 	tmp = VFS_TO_TMPFS(mp);
623 
624 	if (tfd->tfd_id >= tmp->tm_nodes_max)
625 		return (EINVAL);
626 
627 	TMPFS_LOCK(tmp);
628 	LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
629 		if (node->tn_id == tfd->tfd_id &&
630 		    node->tn_gen == tfd->tfd_gen) {
631 			tmpfs_ref_node(node);
632 			break;
633 		}
634 	}
635 	TMPFS_UNLOCK(tmp);
636 
637 	if (node != NULL) {
638 		error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
639 		tmpfs_free_node(tmp, node);
640 	} else
641 		error = EINVAL;
642 	return (error);
643 }
644 
645 /* ARGSUSED2 */
646 static int
647 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
648 {
649 	struct tmpfs_mount *tmp;
650 	size_t used;
651 
652 	tmp = VFS_TO_TMPFS(mp);
653 
654 	sbp->f_iosize = PAGE_SIZE;
655 	sbp->f_bsize = PAGE_SIZE;
656 
657 	used = tmpfs_pages_used(tmp);
658 	if (tmp->tm_pages_max != ULONG_MAX)
659 		 sbp->f_blocks = tmp->tm_pages_max;
660 	else
661 		 sbp->f_blocks = used + tmpfs_mem_avail();
662 	if (sbp->f_blocks <= used)
663 		sbp->f_bavail = 0;
664 	else
665 		sbp->f_bavail = sbp->f_blocks - used;
666 	sbp->f_bfree = sbp->f_bavail;
667 	used = tmp->tm_nodes_inuse;
668 	sbp->f_files = tmp->tm_nodes_max;
669 	if (sbp->f_files <= used)
670 		sbp->f_ffree = 0;
671 	else
672 		sbp->f_ffree = sbp->f_files - used;
673 	/* sbp->f_owner = tmp->tn_uid; */
674 
675 	return (0);
676 }
677 
678 static int
679 tmpfs_sync(struct mount *mp, int waitfor)
680 {
681 
682 	if (waitfor == MNT_SUSPEND) {
683 		MNT_ILOCK(mp);
684 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
685 		MNT_IUNLOCK(mp);
686 	} else if (waitfor == MNT_LAZY) {
687 		tmpfs_update_mtime_lazy(mp);
688 	}
689 	return (0);
690 }
691 
692 static int
693 tmpfs_init(struct vfsconf *conf)
694 {
695 	int res;
696 
697 	res = tmpfs_subr_init();
698 	if (res != 0)
699 		return (res);
700 	memcpy(&tmpfs_fnops, &vnops, sizeof(struct fileops));
701 	tmpfs_fnops.fo_close = tmpfs_fo_close;
702 	return (0);
703 }
704 
705 static int
706 tmpfs_uninit(struct vfsconf *conf)
707 {
708 	tmpfs_subr_uninit();
709 	return (0);
710 }
711 
712 /*
713  * tmpfs vfs operations.
714  */
715 struct vfsops tmpfs_vfsops = {
716 	.vfs_mount =			tmpfs_mount,
717 	.vfs_unmount =			tmpfs_unmount,
718 	.vfs_root =			vfs_cache_root,
719 	.vfs_cachedroot =		tmpfs_root,
720 	.vfs_statfs =			tmpfs_statfs,
721 	.vfs_fhtovp =			tmpfs_fhtovp,
722 	.vfs_sync =			tmpfs_sync,
723 	.vfs_init =			tmpfs_init,
724 	.vfs_uninit =			tmpfs_uninit,
725 };
726 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);
727 
728 #ifdef DDB
729 #include <ddb/ddb.h>
730 
731 static void
732 db_print_tmpfs(struct mount *mp, struct tmpfs_mount *tmp)
733 {
734 	db_printf("mp %p (%s) tmp %p\n", mp,
735 	    mp->mnt_stat.f_mntonname, tmp);
736 	db_printf(
737 	    "\tsize max %ju pages max %lu pages used %lu\n"
738 	    "\tinodes max %ju inodes inuse %ju ea inuse %ju refcount %ju\n"
739 	    "\tmaxfilesize %ju r%c %snamecache %smtime\n",
740 	    (uintmax_t)tmp->tm_size_max, tmp->tm_pages_max, tmp->tm_pages_used,
741 	    (uintmax_t)tmp->tm_nodes_max, (uintmax_t)tmp->tm_nodes_inuse,
742 	    (uintmax_t)tmp->tm_ea_memory_inuse, (uintmax_t)tmp->tm_refcount,
743 	    (uintmax_t)tmp->tm_maxfilesize,
744 	    tmp->tm_ronly ? 'o' : 'w', tmp->tm_nonc ? "no" : "",
745 	    tmp->tm_nomtime ? "no" : "");
746 }
747 
748 DB_SHOW_COMMAND(tmpfs, db_show_tmpfs)
749 {
750 	struct mount *mp;
751 	struct tmpfs_mount *tmp;
752 
753 	if (have_addr) {
754 		mp = (struct mount *)addr;
755 		tmp = VFS_TO_TMPFS(mp);
756 		db_print_tmpfs(mp, tmp);
757 		return;
758 	}
759 
760 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
761 		if (strcmp(mp->mnt_stat.f_fstypename, tmpfs_vfsconf.vfc_name) ==
762 		    0) {
763 			tmp = VFS_TO_TMPFS(mp);
764 			db_print_tmpfs(mp, tmp);
765 		}
766 	}
767 }
768 #endif	/* DDB */
769