1 /* $NetBSD: tmpfs_vfsops.c,v 1.10 2005/12/11 12:24:29 christos Exp $ */
4 * Copyright (c) 2005 The NetBSD Foundation, Inc.
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
34 * Efficient memory file system.
36 * tmpfs is a file system that uses NetBSD's virtual memory sub-system
37 * (the well-known UVM) to store file data and metadata in an efficient
38 * way. This means that it does not follow the structure of an on-disk
39 * file system because it simply does not need to. Instead, it uses
40 * memory-specific data structures and algorithms to automatically
41 * allocate and release resources.
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
46 #include <sys/param.h>
47 #include <sys/limits.h>
49 #include <sys/mutex.h>
50 #include <sys/kernel.h>
52 #include <sys/systm.h>
53 #include <sys/sysctl.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_param.h>
59 #include <fs/tmpfs/tmpfs.h>
62 * Default permission for root node
64 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
66 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
67 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
69 /* --------------------------------------------------------------------- */
71 static int tmpfs_mount(struct mount *, struct thread *);
72 static int tmpfs_unmount(struct mount *, int, struct thread *);
73 static int tmpfs_root(struct mount *, int flags, struct vnode **,
75 static int tmpfs_fhtovp(struct mount *, struct fid *, struct vnode **);
76 static int tmpfs_statfs(struct mount *, struct statfs *, struct thread *);
78 /* --------------------------------------------------------------------- */
80 static const char *tmpfs_opts[] = {
81 "from", "size", "inodes", "uid", "gid", "mode", "export",
85 /* --------------------------------------------------------------------- */
93 char *sname = "vm.swap_info";
94 int soid[SWI_MAXMIB], oid[2];
95 u_int unswdev, total, dmmax, nswapdev;
101 if (kernel_sysctlbyname(curthread, "vm.dmmax", &dmmax, &len,
102 NULL, 0, NULL, 0) != 0)
105 len = sizeof(nswapdev);
106 if (kernel_sysctlbyname(curthread, "vm.nswapdev",
108 NULL, 0, NULL, 0) != 0)
111 mibi = (SWI_MAXMIB - 1) * sizeof(int);
115 if (kernel_sysctl(curthread, oid, 2,
116 soid, &mibi, (void *)sname, strlen(sname),
120 mibi = (SWI_MAXMIB - 1);
121 for (unswdev = 0; unswdev < nswapdev; ++unswdev) {
122 soid[mibi] = unswdev;
123 len = sizeof(struct xswdev);
124 if (kernel_sysctl(curthread,
125 soid, mibi + 1, &xsd, &len, NULL, 0,
128 if (len == sizeof(struct xswdev))
129 total += (xsd.xsw_nblks - dmmax);
136 /* --------------------------------------------------------------------- */
138 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
140 struct tmpfs_node *node = (struct tmpfs_node *)mem;
147 node->tn_vnode = NULL;
148 node->tn_vpstate = 0;
154 tmpfs_node_dtor(void *mem, int size, void *arg)
156 struct tmpfs_node *node = (struct tmpfs_node *)mem;
157 node->tn_type = VNON;
161 tmpfs_node_init(void *mem, int size, int flags)
163 struct tmpfs_node *node = (struct tmpfs_node *)mem;
166 mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
167 node->tn_gen = arc4random();
173 tmpfs_node_fini(void *mem, int size)
175 struct tmpfs_node *node = (struct tmpfs_node *)mem;
177 mtx_destroy(&node->tn_interlock);
181 tmpfs_mount(struct mount *mp, struct thread *td)
183 struct tmpfs_mount *tmp;
184 struct tmpfs_node *root;
185 size_t pages, mem_size;
192 /* Root node attributes. */
199 if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
202 if (mp->mnt_flag & MNT_UPDATE) {
203 /* XXX: There is no support yet to update file system
204 * settings. Should be added. */
209 printf("WARNING: TMPFS is considered to be a highly experimental "
210 "feature in FreeBSD.\n");
212 vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
213 error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
214 VOP_UNLOCK(mp->mnt_vnodecovered, 0);
218 if (mp->mnt_cred->cr_ruid != 0 ||
219 vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
220 root_gid = va.va_gid;
221 if (mp->mnt_cred->cr_ruid != 0 ||
222 vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
223 root_uid = va.va_uid;
224 if (mp->mnt_cred->cr_ruid != 0 ||
225 vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
226 root_mode = va.va_mode;
227 if (vfs_scanopt(mp->mnt_optnew, "inodes", "%d", &nodes_max) != 1)
229 if (vfs_scanopt(mp->mnt_optnew, "size", "%qu", &size_max) != 1)
232 /* Do not allow mounts if we do not have enough memory to preserve
233 * the minimum reserved pages. */
234 mem_size = cnt.v_free_count + cnt.v_inactive_count + get_swpgtotal();
235 mem_size -= mem_size > cnt.v_wire_count ? cnt.v_wire_count : mem_size;
236 if (mem_size < TMPFS_PAGES_RESERVED)
239 /* Get the maximum number of memory pages this file system is
240 * allowed to use, based on the maximum size the user passed in
241 * the mount structure. A value of zero is treated as if the
242 * maximum available space was requested. */
243 if (size_max < PAGE_SIZE || size_max >= SIZE_MAX)
246 pages = howmany(size_max, PAGE_SIZE);
250 nodes = 3 + pages * PAGE_SIZE / 1024;
255 /* Allocate the tmpfs mount structure and fill it. */
256 tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
257 M_TMPFSMNT, M_WAITOK | M_ZERO);
259 mtx_init(&tmp->allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
260 tmp->tm_nodes_max = nodes;
261 tmp->tm_nodes_inuse = 0;
262 tmp->tm_maxfilesize = (u_int64_t)(cnt.v_page_count + get_swpgtotal()) * PAGE_SIZE;
263 LIST_INIT(&tmp->tm_nodes_used);
265 tmp->tm_pages_max = pages;
266 tmp->tm_pages_used = 0;
267 tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->allnode_lock);
268 tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
269 sizeof(struct tmpfs_dirent),
270 NULL, NULL, NULL, NULL,
272 tmp->tm_node_pool = uma_zcreate("TMPFS node",
273 sizeof(struct tmpfs_node),
274 tmpfs_node_ctor, tmpfs_node_dtor,
275 tmpfs_node_init, tmpfs_node_fini,
278 /* Allocate the root node. */
279 error = tmpfs_alloc_node(tmp, VDIR, root_uid,
280 root_gid, root_mode & ALLPERMS, NULL, NULL,
283 if (error != 0 || root == NULL) {
284 uma_zdestroy(tmp->tm_node_pool);
285 uma_zdestroy(tmp->tm_dirent_pool);
286 delete_unrhdr(tmp->tm_ino_unr);
287 free(tmp, M_TMPFSMNT);
290 KASSERT(root->tn_id == 2, ("tmpfs root with invalid ino: %d", root->tn_id));
294 mp->mnt_flag |= MNT_LOCAL;
295 mp->mnt_kern_flag |= MNTK_MPSAFE;
299 mp->mnt_stat.f_namemax = MAXNAMLEN;
301 vfs_mountedfrom(mp, "tmpfs");
306 /* --------------------------------------------------------------------- */
310 tmpfs_unmount(struct mount *mp, int mntflags, struct thread *l)
314 struct tmpfs_mount *tmp;
315 struct tmpfs_node *node;
317 /* Handle forced unmounts. */
318 if (mntflags & MNT_FORCE)
321 /* Finalize all pending I/O. */
322 error = vflush(mp, 0, flags, l);
326 tmp = VFS_TO_TMPFS(mp);
328 /* Free all associated data. The loop iterates over the linked list
329 * we have containing all used nodes. For each of them that is
330 * a directory, we free all its directory entries. Note that after
331 * freeing a node, it will automatically go to the available list,
332 * so we will later have to iterate over it to release its items. */
333 node = LIST_FIRST(&tmp->tm_nodes_used);
334 while (node != NULL) {
335 struct tmpfs_node *next;
337 if (node->tn_type == VDIR) {
338 struct tmpfs_dirent *de;
340 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
342 struct tmpfs_dirent *nde;
344 nde = TAILQ_NEXT(de, td_entries);
345 tmpfs_free_dirent(tmp, de, FALSE);
347 node->tn_size -= sizeof(struct tmpfs_dirent);
351 next = LIST_NEXT(node, tn_entries);
352 tmpfs_free_node(tmp, node);
356 uma_zdestroy(tmp->tm_dirent_pool);
357 uma_zdestroy(tmp->tm_node_pool);
358 delete_unrhdr(tmp->tm_ino_unr);
360 mtx_destroy(&tmp->allnode_lock);
361 MPASS(tmp->tm_pages_used == 0);
362 MPASS(tmp->tm_nodes_inuse == 0);
364 /* Throw away the tmpfs_mount structure. */
365 free(mp->mnt_data, M_TMPFSMNT);
369 mp->mnt_flag &= ~MNT_LOCAL;
374 /* --------------------------------------------------------------------- */
377 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp, struct thread *td)
380 error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp, td);
383 (*vpp)->v_vflag |= VV_ROOT;
388 /* --------------------------------------------------------------------- */
391 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
394 struct tmpfs_fid *tfhp;
395 struct tmpfs_mount *tmp;
396 struct tmpfs_node *node;
398 tmp = VFS_TO_TMPFS(mp);
400 tfhp = (struct tmpfs_fid *)fhp;
401 if (tfhp->tf_len != sizeof(struct tmpfs_fid))
404 if (tfhp->tf_id >= tmp->tm_nodes_max)
410 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
411 if (node->tn_id == tfhp->tf_id &&
412 node->tn_gen == tfhp->tf_gen) {
420 return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp, curthread));
425 /* --------------------------------------------------------------------- */
429 tmpfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *l)
431 fsfilcnt_t freenodes;
432 struct tmpfs_mount *tmp;
434 tmp = VFS_TO_TMPFS(mp);
436 sbp->f_iosize = PAGE_SIZE;
437 sbp->f_bsize = PAGE_SIZE;
439 sbp->f_blocks = TMPFS_PAGES_MAX(tmp);
440 sbp->f_bavail = sbp->f_bfree = TMPFS_PAGES_AVAIL(tmp);
442 freenodes = MIN(tmp->tm_nodes_max - tmp->tm_nodes_inuse,
443 TMPFS_PAGES_AVAIL(tmp) * PAGE_SIZE / sizeof(struct tmpfs_node));
445 sbp->f_files = freenodes + tmp->tm_nodes_inuse;
446 sbp->f_ffree = freenodes;
447 /* sbp->f_owner = tmp->tn_uid; */
452 /* --------------------------------------------------------------------- */
455 * tmpfs vfs operations.
458 struct vfsops tmpfs_vfsops = {
459 .vfs_mount = tmpfs_mount,
460 .vfs_unmount = tmpfs_unmount,
461 .vfs_root = tmpfs_root,
462 .vfs_statfs = tmpfs_statfs,
463 .vfs_fhtovp = tmpfs_fhtovp,
465 VFS_SET(tmpfs_vfsops, tmpfs, 0);