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.
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
48 #include <sys/param.h>
49 #include <sys/limits.h>
51 #include <sys/mutex.h>
54 #include <sys/kernel.h>
55 #include <sys/rwlock.h>
57 #include <sys/systm.h>
58 #include <sys/sysctl.h>
61 #include <vm/vm_object.h>
62 #include <vm/vm_param.h>
64 #include <fs/tmpfs/tmpfs.h>
67 * Default permission for root node
69 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
71 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
72 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
74 static int tmpfs_mount(struct mount *);
75 static int tmpfs_unmount(struct mount *, int);
76 static int tmpfs_root(struct mount *, int flags, struct vnode **);
77 static int tmpfs_fhtovp(struct mount *, struct fid *, int,
79 static int tmpfs_statfs(struct mount *, struct statfs *);
80 static void tmpfs_susp_clean(struct mount *);
82 static const char *tmpfs_opts[] = {
83 "from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export",
87 static const char *tmpfs_updateopts[] = {
88 "from", "export", NULL
92 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
94 struct tmpfs_node *node = (struct tmpfs_node *)mem;
101 node->tn_vnode = NULL;
102 node->tn_vpstate = 0;
108 tmpfs_node_dtor(void *mem, int size, void *arg)
110 struct tmpfs_node *node = (struct tmpfs_node *)mem;
111 node->tn_type = VNON;
115 tmpfs_node_init(void *mem, int size, int flags)
117 struct tmpfs_node *node = (struct tmpfs_node *)mem;
120 mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
121 node->tn_gen = arc4random();
127 tmpfs_node_fini(void *mem, int size)
129 struct tmpfs_node *node = (struct tmpfs_node *)mem;
131 mtx_destroy(&node->tn_interlock);
135 tmpfs_mount(struct mount *mp)
137 const size_t nodes_per_page = howmany(PAGE_SIZE,
138 sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
139 struct tmpfs_mount *tmp;
140 struct tmpfs_node *root;
141 struct thread *td = curthread;
146 off_t nodes_max, size_max, maxfilesize;
148 /* Root node attributes. */
155 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_TMPFS))
158 if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
161 if (mp->mnt_flag & MNT_UPDATE) {
162 /* Only support update mounts for certain options. */
163 if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
165 if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) !=
166 ((struct tmpfs_mount *)mp->mnt_data)->tm_ronly)
171 vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
172 error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
173 VOP_UNLOCK(mp->mnt_vnodecovered, 0);
177 if (mp->mnt_cred->cr_ruid != 0 ||
178 vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
179 root_gid = va.va_gid;
180 if (mp->mnt_cred->cr_ruid != 0 ||
181 vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
182 root_uid = va.va_uid;
183 if (mp->mnt_cred->cr_ruid != 0 ||
184 vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
185 root_mode = va.va_mode;
186 if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
188 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
190 if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
192 nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0;
194 /* Do not allow mounts if we do not have enough memory to preserve
195 * the minimum reserved pages. */
196 if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
199 /* Get the maximum number of memory pages this file system is
200 * allowed to use, based on the maximum size the user passed in
201 * the mount structure. A value of zero is treated as if the
202 * maximum available space was requested. */
203 if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
204 (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
207 size_max = roundup(size_max, PAGE_SIZE);
208 pages = howmany(size_max, PAGE_SIZE);
212 if (nodes_max <= 3) {
213 if (pages < INT_MAX / nodes_per_page)
214 nodes_max = pages * nodes_per_page;
218 if (nodes_max > INT_MAX)
220 MPASS(nodes_max >= 3);
222 /* Allocate the tmpfs mount structure and fill it. */
223 tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
224 M_TMPFSMNT, M_WAITOK | M_ZERO);
226 mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
227 tmp->tm_nodes_max = nodes_max;
228 tmp->tm_nodes_inuse = 0;
229 tmp->tm_refcount = 1;
230 tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
231 LIST_INIT(&tmp->tm_nodes_used);
233 tmp->tm_pages_max = pages;
234 tmp->tm_pages_used = 0;
235 tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->tm_allnode_lock);
236 tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
237 sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL,
239 tmp->tm_node_pool = uma_zcreate("TMPFS node",
240 sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
241 tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
242 tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
245 /* Allocate the root node. */
246 error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid,
247 root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root);
249 if (error != 0 || root == NULL) {
250 uma_zdestroy(tmp->tm_node_pool);
251 uma_zdestroy(tmp->tm_dirent_pool);
252 delete_unrhdr(tmp->tm_ino_unr);
253 free(tmp, M_TMPFSMNT);
256 KASSERT(root->tn_id == 2,
257 ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
261 mp->mnt_flag |= MNT_LOCAL;
262 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED;
266 mp->mnt_stat.f_namemax = MAXNAMLEN;
268 vfs_mountedfrom(mp, "tmpfs");
275 tmpfs_unmount(struct mount *mp, int mntflags)
277 struct tmpfs_mount *tmp;
278 struct tmpfs_node *node;
281 flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0;
282 tmp = VFS_TO_TMPFS(mp);
285 error = vfs_write_suspend_umnt(mp);
289 * At this point, nodes cannot be destroyed by any other
290 * thread because write suspension is started.
294 error = vflush(mp, 0, flags, curthread);
296 vfs_write_resume(mp, VR_START_WRITE);
300 if (mp->mnt_nvnodelistsize == 0) {
305 if ((mntflags & MNT_FORCE) == 0) {
306 vfs_write_resume(mp, VR_START_WRITE);
312 while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
313 TMPFS_NODE_LOCK(node);
314 if (node->tn_type == VDIR)
315 tmpfs_dir_destroy(tmp, node);
316 if (tmpfs_free_node_locked(tmp, node, true))
319 TMPFS_NODE_UNLOCK(node);
324 vfs_write_resume(mp, VR_START_WRITE);
327 mp->mnt_flag &= ~MNT_LOCAL;
334 tmpfs_free_tmp(struct tmpfs_mount *tmp)
337 MPASS(tmp->tm_refcount > 0);
339 if (tmp->tm_refcount > 0) {
345 uma_zdestroy(tmp->tm_dirent_pool);
346 uma_zdestroy(tmp->tm_node_pool);
347 clear_unrhdr(tmp->tm_ino_unr);
348 delete_unrhdr(tmp->tm_ino_unr);
350 mtx_destroy(&tmp->tm_allnode_lock);
351 MPASS(tmp->tm_pages_used == 0);
352 MPASS(tmp->tm_nodes_inuse == 0);
354 free(tmp, M_TMPFSMNT);
358 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
362 error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
364 (*vpp)->v_vflag |= VV_ROOT;
369 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
372 struct tmpfs_fid *tfhp;
373 struct tmpfs_mount *tmp;
374 struct tmpfs_node *node;
377 tmp = VFS_TO_TMPFS(mp);
379 tfhp = (struct tmpfs_fid *)fhp;
380 if (tfhp->tf_len != sizeof(struct tmpfs_fid))
383 if (tfhp->tf_id >= tmp->tm_nodes_max)
387 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
388 if (node->tn_id == tfhp->tf_id &&
389 node->tn_gen == tfhp->tf_gen) {
390 tmpfs_ref_node(node);
397 error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
398 tmpfs_free_node(tmp, node);
406 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
408 struct tmpfs_mount *tmp;
411 tmp = VFS_TO_TMPFS(mp);
413 sbp->f_iosize = PAGE_SIZE;
414 sbp->f_bsize = PAGE_SIZE;
416 used = tmpfs_pages_used(tmp);
417 if (tmp->tm_pages_max != ULONG_MAX)
418 sbp->f_blocks = tmp->tm_pages_max;
420 sbp->f_blocks = used + tmpfs_mem_avail();
421 if (sbp->f_blocks <= used)
424 sbp->f_bavail = sbp->f_blocks - used;
425 sbp->f_bfree = sbp->f_bavail;
426 used = tmp->tm_nodes_inuse;
427 sbp->f_files = tmp->tm_nodes_max;
428 if (sbp->f_files <= used)
431 sbp->f_ffree = sbp->f_files - used;
432 /* sbp->f_owner = tmp->tn_uid; */
438 tmpfs_sync(struct mount *mp, int waitfor)
440 struct vnode *vp, *mvp;
441 struct vm_object *obj;
443 if (waitfor == MNT_SUSPEND) {
445 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
447 } else if (waitfor == MNT_LAZY) {
449 * Handle lazy updates of mtime from writes to mmaped
450 * regions. Use MNT_VNODE_FOREACH_ALL instead of
451 * MNT_VNODE_FOREACH_ACTIVE, since unmap of the
452 * tmpfs-backed vnode does not call vinactive(), due
453 * to vm object type is OBJT_SWAP.
455 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
456 if (vp->v_type != VREG) {
461 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
462 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
465 * Unlocked read, avoid taking vnode lock if
466 * not needed. Lost update will be handled on
469 if ((obj->flags & OBJ_TMPFS_DIRTY) == 0) {
473 if (vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
476 tmpfs_check_mtime(vp);
484 * The presence of a susp_clean method tells the VFS to track writes.
487 tmpfs_susp_clean(struct mount *mp __unused)
492 * tmpfs vfs operations.
495 struct vfsops tmpfs_vfsops = {
496 .vfs_mount = tmpfs_mount,
497 .vfs_unmount = tmpfs_unmount,
498 .vfs_root = tmpfs_root,
499 .vfs_statfs = tmpfs_statfs,
500 .vfs_fhtovp = tmpfs_fhtovp,
501 .vfs_sync = tmpfs_sync,
502 .vfs_susp_clean = tmpfs_susp_clean,
504 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);