1 /* $NetBSD: tmpfs_vfsops.c,v 1.10 2005/12/11 12:24:29 christos Exp $ */
4 * SPDX-License-Identifier: BSD-2-Clause-NetBSD
6 * Copyright (c) 2005 The NetBSD Foundation, Inc.
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
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
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.
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.
36 * Efficient memory file system.
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.
46 #include "opt_tmpfs.h"
48 #include <sys/cdefs.h>
49 __FBSDID("$FreeBSD$");
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/dirent.h>
54 #include <sys/limits.h>
56 #include <sys/mount.h>
57 #include <sys/mutex.h>
60 #include <sys/kernel.h>
61 #include <sys/rwlock.h>
63 #include <sys/sysctl.h>
64 #include <sys/vnode.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_param.h>
70 #include <fs/tmpfs/tmpfs.h>
73 * Default permission for root node
75 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
77 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
78 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
80 static int tmpfs_mount(struct mount *);
81 static int tmpfs_unmount(struct mount *, int);
82 static int tmpfs_root(struct mount *, int flags, struct vnode **);
83 static int tmpfs_fhtovp(struct mount *, struct fid *, int,
85 static int tmpfs_statfs(struct mount *, struct statfs *);
86 static void tmpfs_susp_clean(struct mount *);
88 static const char *tmpfs_opts[] = {
89 "from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export",
93 static const char *tmpfs_updateopts[] = {
94 "from", "export", "size", NULL
98 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
100 struct tmpfs_node *node = (struct tmpfs_node *)mem;
107 node->tn_vnode = NULL;
108 node->tn_vpstate = 0;
114 tmpfs_node_dtor(void *mem, int size, void *arg)
116 struct tmpfs_node *node = (struct tmpfs_node *)mem;
117 node->tn_type = VNON;
121 tmpfs_node_init(void *mem, int size, int flags)
123 struct tmpfs_node *node = (struct tmpfs_node *)mem;
126 mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
127 node->tn_gen = arc4random();
133 tmpfs_node_fini(void *mem, int size)
135 struct tmpfs_node *node = (struct tmpfs_node *)mem;
137 mtx_destroy(&node->tn_interlock);
140 #define TMPFS_SC(mp) ((struct tmpfs_mount *)(mp)->mnt_data)
143 tmpfs_mount(struct mount *mp)
145 const size_t nodes_per_page = howmany(PAGE_SIZE,
146 sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
147 struct tmpfs_mount *tmp;
148 struct tmpfs_node *root;
153 off_t nodes_max, size_max, maxfilesize;
155 /* Root node attributes. */
162 if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
165 if (mp->mnt_flag & MNT_UPDATE) {
166 /* Only support update mounts for certain options. */
167 if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
169 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) == 0) {
171 * On-the-fly resizing is not supported (yet). We still
172 * need to have "size" listed as "supported", otherwise
173 * trying to update fs that is listed in fstab with size
174 * parameter, say trying to change rw to ro or vice
175 * versa, would cause vfs_filteropt() to bail.
177 if (size_max != TMPFS_SC(mp)->tm_size_max)
180 if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
181 !(TMPFS_SC(mp)->tm_ronly)) {
183 error = VFS_SYNC(mp, MNT_WAIT);
187 if (mp->mnt_flag & MNT_FORCE)
189 error = vflush(mp, 0, flags, curthread);
192 TMPFS_SC(mp)->tm_ronly = 1;
194 mp->mnt_flag |= MNT_RDONLY;
196 } else if (!vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
197 TMPFS_SC(mp)->tm_ronly) {
199 TMPFS_SC(mp)->tm_ronly = 0;
201 mp->mnt_flag &= ~MNT_RDONLY;
207 vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
208 error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
209 VOP_UNLOCK(mp->mnt_vnodecovered, 0);
213 if (mp->mnt_cred->cr_ruid != 0 ||
214 vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
215 root_gid = va.va_gid;
216 if (mp->mnt_cred->cr_ruid != 0 ||
217 vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
218 root_uid = va.va_uid;
219 if (mp->mnt_cred->cr_ruid != 0 ||
220 vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
221 root_mode = va.va_mode;
222 if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
224 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
226 if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
228 nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0;
230 /* Do not allow mounts if we do not have enough memory to preserve
231 * the minimum reserved pages. */
232 if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
235 /* Get the maximum number of memory pages this file system is
236 * allowed to use, based on the maximum size the user passed in
237 * the mount structure. A value of zero is treated as if the
238 * maximum available space was requested. */
239 if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
240 (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
243 size_max = roundup(size_max, PAGE_SIZE);
244 pages = howmany(size_max, PAGE_SIZE);
248 if (nodes_max <= 3) {
249 if (pages < INT_MAX / nodes_per_page)
250 nodes_max = pages * nodes_per_page;
254 if (nodes_max > INT_MAX)
256 MPASS(nodes_max >= 3);
258 /* Allocate the tmpfs mount structure and fill it. */
259 tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
260 M_TMPFSMNT, M_WAITOK | M_ZERO);
262 mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
263 tmp->tm_nodes_max = nodes_max;
264 tmp->tm_nodes_inuse = 0;
265 tmp->tm_refcount = 1;
266 tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
267 LIST_INIT(&tmp->tm_nodes_used);
269 tmp->tm_size_max = size_max;
270 tmp->tm_pages_max = pages;
271 tmp->tm_pages_used = 0;
272 new_unrhdr64(&tmp->tm_ino_unr, 2);
273 tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
274 sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL,
276 tmp->tm_node_pool = uma_zcreate("TMPFS node",
277 sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
278 tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
279 tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
282 /* Allocate the root node. */
283 error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid,
284 root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root);
286 if (error != 0 || root == NULL) {
287 uma_zdestroy(tmp->tm_node_pool);
288 uma_zdestroy(tmp->tm_dirent_pool);
289 free(tmp, M_TMPFSMNT);
292 KASSERT(root->tn_id == 2,
293 ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
297 mp->mnt_flag |= MNT_LOCAL;
298 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED;
302 mp->mnt_stat.f_namemax = MAXNAMLEN;
304 vfs_mountedfrom(mp, "tmpfs");
311 tmpfs_unmount(struct mount *mp, int mntflags)
313 struct tmpfs_mount *tmp;
314 struct tmpfs_node *node;
317 flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0;
318 tmp = VFS_TO_TMPFS(mp);
321 error = vfs_write_suspend_umnt(mp);
325 * At this point, nodes cannot be destroyed by any other
326 * thread because write suspension is started.
330 error = vflush(mp, 0, flags, curthread);
332 vfs_write_resume(mp, VR_START_WRITE);
336 if (mp->mnt_nvnodelistsize == 0) {
341 if ((mntflags & MNT_FORCE) == 0) {
342 vfs_write_resume(mp, VR_START_WRITE);
348 while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
349 TMPFS_NODE_LOCK(node);
350 if (node->tn_type == VDIR)
351 tmpfs_dir_destroy(tmp, node);
352 if (tmpfs_free_node_locked(tmp, node, true))
355 TMPFS_NODE_UNLOCK(node);
360 vfs_write_resume(mp, VR_START_WRITE);
363 mp->mnt_flag &= ~MNT_LOCAL;
370 tmpfs_free_tmp(struct tmpfs_mount *tmp)
373 MPASS(tmp->tm_refcount > 0);
375 if (tmp->tm_refcount > 0) {
381 uma_zdestroy(tmp->tm_dirent_pool);
382 uma_zdestroy(tmp->tm_node_pool);
384 mtx_destroy(&tmp->tm_allnode_lock);
385 MPASS(tmp->tm_pages_used == 0);
386 MPASS(tmp->tm_nodes_inuse == 0);
388 free(tmp, M_TMPFSMNT);
392 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
396 error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
398 (*vpp)->v_vflag |= VV_ROOT;
403 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
406 struct tmpfs_fid *tfhp;
407 struct tmpfs_mount *tmp;
408 struct tmpfs_node *node;
411 tmp = VFS_TO_TMPFS(mp);
413 tfhp = (struct tmpfs_fid *)fhp;
414 if (tfhp->tf_len != sizeof(struct tmpfs_fid))
417 if (tfhp->tf_id >= tmp->tm_nodes_max)
421 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
422 if (node->tn_id == tfhp->tf_id &&
423 node->tn_gen == tfhp->tf_gen) {
424 tmpfs_ref_node(node);
431 error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
432 tmpfs_free_node(tmp, node);
440 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
442 struct tmpfs_mount *tmp;
445 tmp = VFS_TO_TMPFS(mp);
447 sbp->f_iosize = PAGE_SIZE;
448 sbp->f_bsize = PAGE_SIZE;
450 used = tmpfs_pages_used(tmp);
451 if (tmp->tm_pages_max != ULONG_MAX)
452 sbp->f_blocks = tmp->tm_pages_max;
454 sbp->f_blocks = used + tmpfs_mem_avail();
455 if (sbp->f_blocks <= used)
458 sbp->f_bavail = sbp->f_blocks - used;
459 sbp->f_bfree = sbp->f_bavail;
460 used = tmp->tm_nodes_inuse;
461 sbp->f_files = tmp->tm_nodes_max;
462 if (sbp->f_files <= used)
465 sbp->f_ffree = sbp->f_files - used;
466 /* sbp->f_owner = tmp->tn_uid; */
472 tmpfs_sync(struct mount *mp, int waitfor)
474 struct vnode *vp, *mvp;
475 struct vm_object *obj;
477 if (waitfor == MNT_SUSPEND) {
479 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
481 } else if (waitfor == MNT_LAZY) {
483 * Handle lazy updates of mtime from writes to mmaped
484 * regions. Use MNT_VNODE_FOREACH_ALL instead of
485 * MNT_VNODE_FOREACH_ACTIVE, since unmap of the
486 * tmpfs-backed vnode does not call vinactive(), due
487 * to vm object type is OBJT_SWAP.
489 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
490 if (vp->v_type != VREG) {
495 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
496 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
499 * Unlocked read, avoid taking vnode lock if
500 * not needed. Lost update will be handled on
503 if ((obj->flags & OBJ_TMPFS_DIRTY) == 0) {
507 if (vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
510 tmpfs_check_mtime(vp);
518 * The presence of a susp_clean method tells the VFS to track writes.
521 tmpfs_susp_clean(struct mount *mp __unused)
526 * tmpfs vfs operations.
529 struct vfsops tmpfs_vfsops = {
530 .vfs_mount = tmpfs_mount,
531 .vfs_unmount = tmpfs_unmount,
532 .vfs_root = tmpfs_root,
533 .vfs_statfs = tmpfs_statfs,
534 .vfs_fhtovp = tmpfs_fhtovp,
535 .vfs_sync = tmpfs_sync,
536 .vfs_susp_clean = tmpfs_susp_clean,
538 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);