xref: /freebsd-14-stable/sys/fs/tmpfs/tmpfs_vfsops.c (revision 5e6079209e9b6db2bd2980e1b97e7cb4cf8b5364)
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/cdefs.h>
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/dirent.h>
53 #include <sys/file.h>
54 #include <sys/limits.h>
55 #include <sys/lock.h>
56 #include <sys/mount.h>
57 #include <sys/mutex.h>
58 #include <sys/proc.h>
59 #include <sys/jail.h>
60 #include <sys/kernel.h>
61 #include <sys/rwlock.h>
62 #include <sys/stat.h>
63 #include <sys/sx.h>
64 #include <sys/sysctl.h>
65 #include <sys/vnode.h>
66 
67 #include <vm/vm.h>
68 #include <vm/vm_param.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_param.h>
74 
75 #include <fs/tmpfs/tmpfs.h>
76 
77 /*
78  * Default permission for root node
79  */
80 #define TMPFS_DEFAULT_ROOT_MODE	(S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
81 
82 static MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
83 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
84 
85 static int	tmpfs_mount(struct mount *);
86 static int	tmpfs_unmount(struct mount *, int);
87 static int	tmpfs_root(struct mount *, int flags, struct vnode **);
88 static int	tmpfs_fhtovp(struct mount *, struct fid *, int,
89 		    struct vnode **);
90 static int	tmpfs_statfs(struct mount *, struct statfs *);
91 
92 static const char *tmpfs_opts[] = {
93 	"from", "easize", "size", "maxfilesize", "inodes", "uid", "gid", "mode",
94 	"export", "union", "nonc", "nomtime", "nosymfollow", "pgread", NULL
95 };
96 
97 static const char *tmpfs_updateopts[] = {
98 	"from", "easize", "export", "nomtime", "size", "nosymfollow", NULL
99 };
100 
101 static int
tmpfs_update_mtime_lazy_filter(struct vnode * vp,void * arg)102 tmpfs_update_mtime_lazy_filter(struct vnode *vp, void *arg)
103 {
104 	struct vm_object *obj;
105 
106 	if (vp->v_type != VREG)
107 		return (0);
108 
109 	obj = atomic_load_ptr(&vp->v_object);
110 	if (obj == NULL)
111 		return (0);
112 
113 	return (vm_object_mightbedirty_(obj));
114 }
115 
116 static void
tmpfs_update_mtime_lazy(struct mount * mp)117 tmpfs_update_mtime_lazy(struct mount *mp)
118 {
119 	struct vnode *vp, *mvp;
120 
121 	MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, tmpfs_update_mtime_lazy_filter, NULL) {
122 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK) != 0)
123 			continue;
124 		tmpfs_check_mtime(vp);
125 		vput(vp);
126 	}
127 }
128 
129 static void
tmpfs_update_mtime_all(struct mount * mp)130 tmpfs_update_mtime_all(struct mount *mp)
131 {
132 	struct vnode *vp, *mvp;
133 
134 	if (VFS_TO_TMPFS(mp)->tm_nomtime)
135 		return;
136 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
137 		if (vp->v_type != VREG) {
138 			VI_UNLOCK(vp);
139 			continue;
140 		}
141 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK) != 0)
142 			continue;
143 		tmpfs_check_mtime(vp);
144 		tmpfs_update(vp);
145 		vput(vp);
146 	}
147 }
148 
149 struct tmpfs_check_rw_maps_arg {
150 	bool found;
151 };
152 
153 static bool
tmpfs_check_rw_maps_cb(struct mount * mp __unused,vm_map_t map __unused,vm_map_entry_t entry __unused,void * arg)154 tmpfs_check_rw_maps_cb(struct mount *mp __unused, vm_map_t map __unused,
155     vm_map_entry_t entry __unused, void *arg)
156 {
157 	struct tmpfs_check_rw_maps_arg *a;
158 
159 	a = arg;
160 	a->found = true;
161 	return (true);
162 }
163 
164 /*
165  * Revoke write permissions from all mappings of regular files
166  * belonging to the specified tmpfs mount.
167  */
168 static bool
tmpfs_revoke_rw_maps_cb(struct mount * mp __unused,vm_map_t map,vm_map_entry_t entry,void * arg __unused)169 tmpfs_revoke_rw_maps_cb(struct mount *mp __unused, vm_map_t map,
170     vm_map_entry_t entry, void *arg __unused)
171 {
172 
173 	/*
174 	 * XXXKIB: might be invalidate the mapping
175 	 * instead ?  The process is not going to be
176 	 * happy in any case.
177 	 */
178 	entry->max_protection &= ~VM_PROT_WRITE;
179 	if ((entry->protection & VM_PROT_WRITE) != 0) {
180 		entry->protection &= ~VM_PROT_WRITE;
181 		pmap_protect(map->pmap, entry->start, entry->end,
182 		    entry->protection);
183 	}
184 	return (false);
185 }
186 
187 static void
tmpfs_all_rw_maps(struct mount * mp,bool (* cb)(struct mount * mp,vm_map_t,vm_map_entry_t,void *),void * cb_arg)188 tmpfs_all_rw_maps(struct mount *mp, bool (*cb)(struct mount *mp, vm_map_t,
189     vm_map_entry_t, void *), void *cb_arg)
190 {
191 	struct proc *p;
192 	struct vmspace *vm;
193 	vm_map_t map;
194 	vm_map_entry_t entry;
195 	vm_object_t object;
196 	struct vnode *vp;
197 	int gen;
198 	bool terminate;
199 
200 	terminate = false;
201 	sx_slock(&allproc_lock);
202 again:
203 	gen = allproc_gen;
204 	FOREACH_PROC_IN_SYSTEM(p) {
205 		PROC_LOCK(p);
206 		if (p->p_state != PRS_NORMAL || (p->p_flag & (P_INEXEC |
207 		    P_SYSTEM | P_WEXIT)) != 0) {
208 			PROC_UNLOCK(p);
209 			continue;
210 		}
211 		vm = vmspace_acquire_ref(p);
212 		_PHOLD_LITE(p);
213 		PROC_UNLOCK(p);
214 		if (vm == NULL) {
215 			PRELE(p);
216 			continue;
217 		}
218 		sx_sunlock(&allproc_lock);
219 		map = &vm->vm_map;
220 
221 		vm_map_lock(map);
222 		if (map->busy)
223 			vm_map_wait_busy(map);
224 		VM_MAP_ENTRY_FOREACH(entry, map) {
225 			if ((entry->eflags & (MAP_ENTRY_GUARD |
226 			    MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_COW)) != 0 ||
227 			    (entry->max_protection & VM_PROT_WRITE) == 0)
228 				continue;
229 			object = entry->object.vm_object;
230 			if (object == NULL || object->type != tmpfs_pager_type)
231 				continue;
232 			/*
233 			 * No need to dig into shadow chain, mapping
234 			 * of the object not at top is readonly.
235 			 */
236 
237 			VM_OBJECT_RLOCK(object);
238 			if (object->type == OBJT_DEAD) {
239 				VM_OBJECT_RUNLOCK(object);
240 				continue;
241 			}
242 			MPASS(object->ref_count > 1);
243 			if ((object->flags & OBJ_TMPFS) == 0) {
244 				VM_OBJECT_RUNLOCK(object);
245 				continue;
246 			}
247 			vp = VM_TO_TMPFS_VP(object);
248 			if (vp->v_mount != mp) {
249 				VM_OBJECT_RUNLOCK(object);
250 				continue;
251 			}
252 
253 			terminate = cb(mp, map, entry, cb_arg);
254 			VM_OBJECT_RUNLOCK(object);
255 			if (terminate)
256 				break;
257 		}
258 		vm_map_unlock(map);
259 
260 		vmspace_free(vm);
261 		sx_slock(&allproc_lock);
262 		PRELE(p);
263 		if (terminate)
264 			break;
265 	}
266 	if (!terminate && gen != allproc_gen)
267 		goto again;
268 	sx_sunlock(&allproc_lock);
269 }
270 
271 static bool
tmpfs_check_rw_maps(struct mount * mp)272 tmpfs_check_rw_maps(struct mount *mp)
273 {
274 	struct tmpfs_check_rw_maps_arg ca;
275 
276 	ca.found = false;
277 	tmpfs_all_rw_maps(mp, tmpfs_check_rw_maps_cb, &ca);
278 	return (ca.found);
279 }
280 
281 static int
tmpfs_rw_to_ro(struct mount * mp)282 tmpfs_rw_to_ro(struct mount *mp)
283 {
284 	int error, flags;
285 	bool forced;
286 
287 	forced = (mp->mnt_flag & MNT_FORCE) != 0;
288 	flags = WRITECLOSE | (forced ? FORCECLOSE : 0);
289 
290 	if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
291 		return (error);
292 	error = vfs_write_suspend_umnt(mp);
293 	if (error != 0)
294 		return (error);
295 	if (!forced && tmpfs_check_rw_maps(mp)) {
296 		error = EBUSY;
297 		goto out;
298 	}
299 	VFS_TO_TMPFS(mp)->tm_ronly = 1;
300 	MNT_ILOCK(mp);
301 	mp->mnt_flag |= MNT_RDONLY;
302 	MNT_IUNLOCK(mp);
303 	for (;;) {
304 		tmpfs_all_rw_maps(mp, tmpfs_revoke_rw_maps_cb, NULL);
305 		tmpfs_update_mtime_all(mp);
306 		error = vflush(mp, 0, flags, curthread);
307 		if (error != 0) {
308 			VFS_TO_TMPFS(mp)->tm_ronly = 0;
309 			MNT_ILOCK(mp);
310 			mp->mnt_flag &= ~MNT_RDONLY;
311 			MNT_IUNLOCK(mp);
312 			goto out;
313 		}
314 		if (!tmpfs_check_rw_maps(mp))
315 			break;
316 	}
317 out:
318 	vfs_write_resume(mp, 0);
319 	return (error);
320 }
321 
322 static int
tmpfs_mount(struct mount * mp)323 tmpfs_mount(struct mount *mp)
324 {
325 	const size_t nodes_per_page = howmany(PAGE_SIZE,
326 	    sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
327 	struct tmpfs_mount *tmp;
328 	struct tmpfs_node *root;
329 	int error;
330 	bool nomtime, nonc, pgread;
331 	/* Size counters. */
332 	u_quad_t pages;
333 	off_t nodes_max, size_max, maxfilesize, ea_max_size;
334 
335 	/* Root node attributes. */
336 	uid_t root_uid;
337 	gid_t root_gid;
338 	mode_t root_mode;
339 
340 	struct vattr va;
341 
342 	if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
343 		return (EINVAL);
344 
345 	if (mp->mnt_flag & MNT_UPDATE) {
346 		/* Only support update mounts for certain options. */
347 		if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
348 			return (EOPNOTSUPP);
349 		tmp = VFS_TO_TMPFS(mp);
350 		if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) == 0) {
351 			/*
352 			 * On-the-fly resizing is not supported (yet). We still
353 			 * need to have "size" listed as "supported", otherwise
354 			 * trying to update fs that is listed in fstab with size
355 			 * parameter, say trying to change rw to ro or vice
356 			 * versa, would cause vfs_filteropt() to bail.
357 			 */
358 			if (size_max != tmp->tm_size_max)
359 				return (EOPNOTSUPP);
360 		}
361 		if (vfs_getopt_size(mp->mnt_optnew, "easize", &ea_max_size) == 0) {
362 			tmp->tm_ea_memory_max = ea_max_size;
363 		}
364 		if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
365 		    !tmp->tm_ronly) {
366 			/* RW -> RO */
367 			return (tmpfs_rw_to_ro(mp));
368 		} else if (!vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
369 		    tmp->tm_ronly) {
370 			/* RO -> RW */
371 			tmp->tm_ronly = 0;
372 			MNT_ILOCK(mp);
373 			mp->mnt_flag &= ~MNT_RDONLY;
374 			MNT_IUNLOCK(mp);
375 		}
376 		tmp->tm_nomtime = vfs_getopt(mp->mnt_optnew, "nomtime", NULL,
377 		    0) == 0;
378 		MNT_ILOCK(mp);
379 		if ((mp->mnt_flag & MNT_UNION) == 0) {
380 			mp->mnt_kern_flag |= MNTK_FPLOOKUP;
381 		} else {
382 			mp->mnt_kern_flag &= ~MNTK_FPLOOKUP;
383 		}
384 		MNT_IUNLOCK(mp);
385 		return (0);
386 	}
387 
388 	vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
389 	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
390 	VOP_UNLOCK(mp->mnt_vnodecovered);
391 	if (error)
392 		return (error);
393 
394 	if (mp->mnt_cred->cr_ruid != 0 ||
395 	    vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
396 		root_gid = va.va_gid;
397 	if (mp->mnt_cred->cr_ruid != 0 ||
398 	    vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
399 		root_uid = va.va_uid;
400 	if (mp->mnt_cred->cr_ruid != 0 ||
401 	    vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
402 		root_mode = va.va_mode;
403 	if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
404 		nodes_max = 0;
405 	if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
406 		size_max = 0;
407 	if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
408 		maxfilesize = 0;
409 	if (vfs_getopt_size(mp->mnt_optnew, "easize", &ea_max_size) != 0)
410 		ea_max_size = 0;
411 	nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0;
412 	nomtime = vfs_getopt(mp->mnt_optnew, "nomtime", NULL, NULL) == 0;
413 	pgread = vfs_getopt(mp->mnt_optnew, "pgread", NULL, NULL) == 0;
414 
415 	/* Do not allow mounts if we do not have enough memory to preserve
416 	 * the minimum reserved pages. */
417 	if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
418 		return (ENOSPC);
419 
420 	/* Get the maximum number of memory pages this file system is
421 	 * allowed to use, based on the maximum size the user passed in
422 	 * the mount structure.  A value of zero is treated as if the
423 	 * maximum available space was requested. */
424 	if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
425 	    (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
426 		pages = SIZE_MAX;
427 	else {
428 		size_max = roundup(size_max, PAGE_SIZE);
429 		pages = howmany(size_max, PAGE_SIZE);
430 	}
431 	MPASS(pages > 0);
432 
433 	if (nodes_max <= 3) {
434 		if (pages < INT_MAX / nodes_per_page)
435 			nodes_max = pages * nodes_per_page;
436 		else
437 			nodes_max = INT_MAX;
438 	}
439 	if (nodes_max > INT_MAX)
440 		nodes_max = INT_MAX;
441 	MPASS(nodes_max >= 3);
442 
443 	/* Allocate the tmpfs mount structure and fill it. */
444 	tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
445 	    M_TMPFSMNT, M_WAITOK | M_ZERO);
446 
447 	mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
448 	tmp->tm_nodes_max = nodes_max;
449 	tmp->tm_nodes_inuse = 0;
450 	tmp->tm_ea_memory_inuse = 0;
451 	tmp->tm_refcount = 1;
452 	tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
453 	tmp->tm_ea_memory_max = ea_max_size > 0 ?
454 	    ea_max_size : TMPFS_EA_MEMORY_RESERVED;
455 	LIST_INIT(&tmp->tm_nodes_used);
456 
457 	tmp->tm_size_max = size_max;
458 	tmp->tm_pages_max = pages;
459 	tmp->tm_pages_used = 0;
460 	new_unrhdr64(&tmp->tm_ino_unr, 2);
461 	tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
462 	tmp->tm_nonc = nonc;
463 	tmp->tm_nomtime = nomtime;
464 	tmp->tm_pgread = pgread;
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
tmpfs_unmount(struct mount * mp,int mntflags)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
tmpfs_free_tmp(struct tmpfs_mount * tmp)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
tmpfs_root(struct mount * mp,int flags,struct vnode ** vpp)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
tmpfs_fhtovp(struct mount * mp,struct fid * fhp,int flags,struct vnode ** vpp)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 	tfd = (struct tmpfs_fid_data *)fhp;
598 
599 	tmp = VFS_TO_TMPFS(mp);
600 
601 	if (tfd->tfd_id >= tmp->tm_nodes_max)
602 		return (EINVAL);
603 
604 	TMPFS_LOCK(tmp);
605 	LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
606 		if (node->tn_id == tfd->tfd_id &&
607 		    node->tn_gen == tfd->tfd_gen) {
608 			tmpfs_ref_node(node);
609 			break;
610 		}
611 	}
612 	TMPFS_UNLOCK(tmp);
613 
614 	if (node != NULL) {
615 		error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
616 		tmpfs_free_node(tmp, node);
617 	} else
618 		error = EINVAL;
619 	return (error);
620 }
621 
622 /* ARGSUSED2 */
623 static int
tmpfs_statfs(struct mount * mp,struct statfs * sbp)624 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
625 {
626 	struct tmpfs_mount *tmp;
627 	size_t used;
628 
629 	tmp = VFS_TO_TMPFS(mp);
630 
631 	sbp->f_iosize = PAGE_SIZE;
632 	sbp->f_bsize = PAGE_SIZE;
633 
634 	used = tmpfs_pages_used(tmp);
635 	if (tmp->tm_pages_max != ULONG_MAX)
636 		 sbp->f_blocks = tmp->tm_pages_max;
637 	else
638 		 sbp->f_blocks = used + tmpfs_mem_avail();
639 	if (sbp->f_blocks <= used)
640 		sbp->f_bavail = 0;
641 	else
642 		sbp->f_bavail = sbp->f_blocks - used;
643 	sbp->f_bfree = sbp->f_bavail;
644 	used = tmp->tm_nodes_inuse;
645 	sbp->f_files = tmp->tm_nodes_max;
646 	if (sbp->f_files <= used)
647 		sbp->f_ffree = 0;
648 	else
649 		sbp->f_ffree = sbp->f_files - used;
650 	/* sbp->f_owner = tmp->tn_uid; */
651 
652 	return (0);
653 }
654 
655 static int
tmpfs_sync(struct mount * mp,int waitfor)656 tmpfs_sync(struct mount *mp, int waitfor)
657 {
658 
659 	if (waitfor == MNT_SUSPEND) {
660 		MNT_ILOCK(mp);
661 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
662 		MNT_IUNLOCK(mp);
663 	} else if (waitfor == MNT_LAZY) {
664 		tmpfs_update_mtime_lazy(mp);
665 	}
666 	return (0);
667 }
668 
669 static int
tmpfs_init(struct vfsconf * conf)670 tmpfs_init(struct vfsconf *conf)
671 {
672 	int res;
673 
674 	res = tmpfs_subr_init();
675 	if (res != 0)
676 		return (res);
677 	memcpy(&tmpfs_fnops, &vnops, sizeof(struct fileops));
678 	tmpfs_fnops.fo_close = tmpfs_fo_close;
679 	return (0);
680 }
681 
682 static int
tmpfs_uninit(struct vfsconf * conf)683 tmpfs_uninit(struct vfsconf *conf)
684 {
685 	tmpfs_subr_uninit();
686 	return (0);
687 }
688 
689 /*
690  * tmpfs vfs operations.
691  */
692 struct vfsops tmpfs_vfsops = {
693 	.vfs_mount =			tmpfs_mount,
694 	.vfs_unmount =			tmpfs_unmount,
695 	.vfs_root =			vfs_cache_root,
696 	.vfs_cachedroot =		tmpfs_root,
697 	.vfs_statfs =			tmpfs_statfs,
698 	.vfs_fhtovp =			tmpfs_fhtovp,
699 	.vfs_sync =			tmpfs_sync,
700 	.vfs_init =			tmpfs_init,
701 	.vfs_uninit =			tmpfs_uninit,
702 };
703 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);
704 
705 #ifdef DDB
706 #include <ddb/ddb.h>
707 
708 static void
db_print_tmpfs(struct mount * mp,struct tmpfs_mount * tmp)709 db_print_tmpfs(struct mount *mp, struct tmpfs_mount *tmp)
710 {
711 	db_printf("mp %p (%s) tmp %p\n", mp,
712 	    mp->mnt_stat.f_mntonname, tmp);
713 	db_printf(
714 	    "\tsize max %ju pages max %lu pages used %lu\n"
715 	    "\tinodes max %ju inodes inuse %ju ea inuse %ju refcount %ju\n"
716 	    "\tmaxfilesize %ju r%c %snamecache %smtime\n",
717 	    (uintmax_t)tmp->tm_size_max, tmp->tm_pages_max, tmp->tm_pages_used,
718 	    (uintmax_t)tmp->tm_nodes_max, (uintmax_t)tmp->tm_nodes_inuse,
719 	    (uintmax_t)tmp->tm_ea_memory_inuse, (uintmax_t)tmp->tm_refcount,
720 	    (uintmax_t)tmp->tm_maxfilesize,
721 	    tmp->tm_ronly ? 'o' : 'w', tmp->tm_nonc ? "no" : "",
722 	    tmp->tm_nomtime ? "no" : "");
723 }
724 
DB_SHOW_COMMAND(tmpfs,db_show_tmpfs)725 DB_SHOW_COMMAND(tmpfs, db_show_tmpfs)
726 {
727 	struct mount *mp;
728 	struct tmpfs_mount *tmp;
729 
730 	if (have_addr) {
731 		mp = (struct mount *)addr;
732 		tmp = VFS_TO_TMPFS(mp);
733 		db_print_tmpfs(mp, tmp);
734 		return;
735 	}
736 
737 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
738 		if (strcmp(mp->mnt_stat.f_fstypename, tmpfs_vfsconf.vfc_name) ==
739 		    0) {
740 			tmp = VFS_TO_TMPFS(mp);
741 			db_print_tmpfs(mp, tmp);
742 		}
743 	}
744 }
745 #endif	/* DDB */
746