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>
45 #include <sys/param.h>
46 #include <sys/limits.h>
48 #include <sys/mutex.h>
49 #include <sys/kernel.h>
51 #include <sys/systm.h>
52 #include <sys/sysctl.h>
53 #include <sys/objcache.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_param.h>
59 #include <vfs/tmpfs/tmpfs.h>
60 #include <vfs/tmpfs/tmpfs_vnops.h>
61 #include <vfs/tmpfs/tmpfs_args.h>
64 * Default permission for root node
66 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
68 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
69 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
70 MALLOC_DEFINE(M_TMPFS_DIRENT, "tmpfs dirent", "tmpfs dirent structures");
71 MALLOC_DEFINE(M_TMPFS_NODE, "tmpfs node", "tmpfs node structures");
73 /* --------------------------------------------------------------------- */
75 static int tmpfs_mount(struct mount *, char *, caddr_t, struct ucred *);
76 static int tmpfs_unmount(struct mount *, int);
77 static int tmpfs_root(struct mount *, struct vnode **);
78 static int tmpfs_fhtovp(struct mount *, struct vnode *, struct fid *, struct vnode **);
79 static int tmpfs_statfs(struct mount *, struct statfs *, struct ucred *cred);
81 /* --------------------------------------------------------------------- */
83 tmpfs_node_ctor(void *obj, void *privdata, int flags)
85 struct tmpfs_node *node = (struct tmpfs_node *)obj;
92 node->tn_vnode = NULL;
93 node->tn_vpstate = TMPFS_VNODE_WANT;
94 bzero(&node->tn_spec, sizeof(node->tn_spec));
100 tmpfs_node_dtor(void *obj, void *privdata)
102 struct tmpfs_node *node = (struct tmpfs_node *)obj;
103 node->tn_type = VNON;
104 node->tn_vpstate = TMPFS_VNODE_DOOMED;
108 tmpfs_node_init(void *args, int flags)
110 struct tmpfs_node *node = (struct tmpfs_node *)objcache_malloc_alloc(args, flags);
113 lockinit(&node->tn_interlock, "tmpfs node interlock", 0, LK_CANRECURSE);
114 node->tn_gen = karc4random();
120 tmpfs_node_fini(void *obj, void *args)
122 struct tmpfs_node *node = (struct tmpfs_node *)obj;
123 lockuninit(&node->tn_interlock);
124 objcache_malloc_free(obj, args);
127 struct objcache_malloc_args tmpfs_dirent_pool_malloc_args =
128 { sizeof(struct tmpfs_dirent), M_TMPFS_DIRENT };
129 struct objcache_malloc_args tmpfs_node_pool_malloc_args =
130 { sizeof(struct tmpfs_node), M_TMPFS_NODE };
133 tmpfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
135 struct tmpfs_mount *tmp;
136 struct tmpfs_node *root;
137 struct tmpfs_args args;
139 vm_pindex_t pages_limit;
149 /* Root node attributes. */
150 uid_t root_uid = cred->cr_uid;
151 gid_t root_gid = cred->cr_gid;
152 mode_t root_mode = (VREAD | VWRITE);
154 if (mp->mnt_flag & MNT_UPDATE) {
155 /* XXX: There is no support yet to update file system
156 * settings. Should be added. */
164 bzero(&args, sizeof(args));
171 error = copyin(data, &args, sizeof(args));
175 size_max = args.ta_size_max;
176 nodes_max = args.ta_nodes_max;
177 maxfsize_max = args.ta_maxfsize_max;
178 root_uid = args.ta_root_uid;
179 root_gid = args.ta_root_gid;
180 root_mode = args.ta_root_mode;
184 * If mount by non-root, then verify that user has necessary
185 * permissions on the device.
187 if (cred->cr_uid != 0) {
189 if ((mp->mnt_flag & MNT_RDONLY) == 0)
193 pages_limit = vm_swap_max + vmstats.v_page_count / 2;
196 pages = pages_limit / 2;
197 else if (size_max < PAGE_SIZE)
199 else if (OFF_TO_IDX(size_max) > pages_limit)
202 pages = OFF_TO_IDX(size_max);
205 nodes = 3 + pages * PAGE_SIZE / 1024;
206 else if (nodes_max < 3)
208 else if (nodes_max > pages)
213 maxfsize = IDX_TO_OFF(pages_limit);
214 if (maxfsize_max != 0 && maxfsize > maxfsize_max)
215 maxfsize = maxfsize_max;
217 /* Allocate the tmpfs mount structure and fill it. */
218 tmp = kmalloc(sizeof(*tmp), M_TMPFSMNT, M_WAITOK | M_ZERO);
220 lockinit(&(tmp->allnode_lock), "tmpfs allnode lock", 0, LK_CANRECURSE);
221 tmp->tm_nodes_max = nodes;
222 tmp->tm_nodes_inuse = 0;
223 tmp->tm_maxfilesize = maxfsize;
224 LIST_INIT(&tmp->tm_nodes_used);
226 tmp->tm_pages_max = pages;
227 tmp->tm_pages_used = 0;
229 tmp->tm_dirent_pool = objcache_create( "tmpfs dirent cache",
232 objcache_malloc_alloc, objcache_malloc_free,
233 &tmpfs_dirent_pool_malloc_args);
234 tmp->tm_node_pool = objcache_create( "tmpfs node cache",
236 tmpfs_node_ctor, tmpfs_node_dtor, NULL,
237 tmpfs_node_init, tmpfs_node_fini,
238 &tmpfs_node_pool_malloc_args);
240 /* Allocate the root node. */
241 error = tmpfs_alloc_node(tmp, VDIR, root_uid, root_gid,
242 root_mode & ALLPERMS, NULL, NULL,
243 VNOVAL, VNOVAL, &root);
246 * We are backed by swap, set snocache chflags flag so we
247 * don't trip over swapcache.
249 root->tn_flags = SF_NOCACHE;
251 if (error != 0 || root == NULL) {
252 objcache_destroy(tmp->tm_node_pool);
253 objcache_destroy(tmp->tm_dirent_pool);
254 kfree(tmp, M_TMPFSMNT);
257 KASSERT(root->tn_id >= 0, ("tmpfs root with invalid ino: %d", (int)root->tn_id));
260 mp->mnt_flag |= MNT_LOCAL;
262 mp->mnt_kern_flag |= MNTK_RD_MPSAFE | MNTK_WR_MPSAFE | MNTK_GA_MPSAFE |
263 MNTK_IN_MPSAFE | MNTK_SG_MPSAFE;
265 mp->mnt_kern_flag |= MNTK_NOMSYNC;
266 mp->mnt_data = (qaddr_t)tmp;
270 vfs_add_vnodeops(mp, &tmpfs_vnode_vops, &mp->mnt_vn_norm_ops);
271 vfs_add_vnodeops(mp, &tmpfs_fifo_vops, &mp->mnt_vn_fifo_ops);
273 copystr("tmpfs", mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
274 bzero(mp->mnt_stat.f_mntfromname +size, MNAMELEN - size);
275 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
276 copyinstr(path, mp->mnt_stat.f_mntonname,
277 sizeof(mp->mnt_stat.f_mntonname) -1,
280 tmpfs_statfs(mp, &mp->mnt_stat, cred);
285 /* --------------------------------------------------------------------- */
289 tmpfs_unmount(struct mount *mp, int mntflags)
294 struct tmpfs_mount *tmp;
295 struct tmpfs_node *node;
297 /* Handle forced unmounts. */
298 if (mntflags & MNT_FORCE)
301 tmp = VFS_TO_TMPFS(mp);
304 * Finalize all pending I/O. In the case of tmpfs we want
305 * to throw all the data away so clean out the buffer cache
306 * and vm objects before calling vflush().
308 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
309 if (node->tn_type == VREG && node->tn_vnode) {
311 TMPFS_NODE_LOCK(node);
312 vx_get(node->tn_vnode);
313 tmpfs_truncate(node->tn_vnode, 0);
314 vx_put(node->tn_vnode);
315 TMPFS_NODE_UNLOCK(node);
319 error = vflush(mp, 0, flags);
324 * First pass get rid of all the directory entries and
325 * vnode associations. The directory structure will
326 * remain via the extra link count representing tn_dir.tn_parent.
328 * No vnodes should remain after the vflush above.
330 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
332 TMPFS_NODE_LOCK(node);
333 if (node->tn_type == VDIR) {
334 struct tmpfs_dirent *de;
336 while (!TAILQ_EMPTY(&node->tn_dir.tn_dirhead)) {
337 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
338 tmpfs_dir_detach(node, de);
339 tmpfs_free_dirent(tmp, de);
340 node->tn_size -= sizeof(struct tmpfs_dirent);
343 KKASSERT(node->tn_vnode == NULL);
349 node->tn_vnode = NULL;
352 TMPFS_NODE_UNLOCK(node);
357 * Now get rid of all nodes. We can remove any node with a
358 * link count of 0 or any directory node with a link count of
359 * 1. The parents will not be destroyed until all their children
360 * have been destroyed.
362 * Recursion in tmpfs_free_node() can further modify the list so
363 * we cannot use a next pointer here.
365 * The root node will be destroyed by this loop (it will be last).
367 while (!LIST_EMPTY(&tmp->tm_nodes_used)) {
369 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
370 if (node->tn_links == 0 ||
371 (node->tn_links == 1 && node->tn_type == VDIR)) {
372 TMPFS_NODE_LOCK(node);
373 tmpfs_free_node(tmp, node);
380 kprintf("tmpfs: Cannot free entire node tree!");
385 KKASSERT(tmp->tm_root == NULL);
387 objcache_destroy(tmp->tm_dirent_pool);
388 objcache_destroy(tmp->tm_node_pool);
390 lockuninit(&tmp->allnode_lock);
391 KKASSERT(tmp->tm_pages_used == 0);
392 KKASSERT(tmp->tm_nodes_inuse == 0);
394 /* Throw away the tmpfs_mount structure. */
395 kfree(tmp, M_TMPFSMNT);
398 mp->mnt_flag &= ~MNT_LOCAL;
402 /* --------------------------------------------------------------------- */
405 tmpfs_root(struct mount *mp, struct vnode **vpp)
407 struct tmpfs_mount *tmp;
410 tmp = VFS_TO_TMPFS(mp);
411 if (tmp->tm_root == NULL) {
412 kprintf("tmpfs_root: called without root node %p\n", mp);
417 error = tmpfs_alloc_vp(mp, tmp->tm_root, LK_EXCLUSIVE, vpp);
418 (*vpp)->v_flag |= VROOT;
419 (*vpp)->v_type = VDIR;
424 /* --------------------------------------------------------------------- */
427 tmpfs_fhtovp(struct mount *mp, struct vnode *rootvp, struct fid *fhp, struct vnode **vpp)
430 struct tmpfs_fid *tfhp;
431 struct tmpfs_mount *tmp;
432 struct tmpfs_node *node;
434 tmp = VFS_TO_TMPFS(mp);
436 tfhp = (struct tmpfs_fid *)fhp;
437 if (tfhp->tf_len != sizeof(struct tmpfs_fid))
440 if (tfhp->tf_id >= tmp->tm_nodes_max)
446 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
447 if (node->tn_id == tfhp->tf_id &&
448 node->tn_gen == tfhp->tf_gen) {
456 return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp));
461 /* --------------------------------------------------------------------- */
465 tmpfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
467 fsfilcnt_t freenodes;
468 struct tmpfs_mount *tmp;
470 tmp = VFS_TO_TMPFS(mp);
472 sbp->f_iosize = PAGE_SIZE;
473 sbp->f_bsize = PAGE_SIZE;
475 sbp->f_blocks = tmp->tm_pages_max;
476 sbp->f_bavail = tmp->tm_pages_max - tmp->tm_pages_used;
477 sbp->f_bfree = sbp->f_bavail;
479 freenodes = tmp->tm_nodes_max - tmp->tm_nodes_inuse;
481 sbp->f_files = freenodes + tmp->tm_nodes_inuse;
482 sbp->f_ffree = freenodes;
483 sbp->f_owner = tmp->tm_root->tn_uid;
488 /* --------------------------------------------------------------------- */
491 * tmpfs vfs operations.
494 static struct vfsops tmpfs_vfsops = {
495 .vfs_mount = tmpfs_mount,
496 .vfs_unmount = tmpfs_unmount,
497 .vfs_root = tmpfs_root,
498 .vfs_statfs = tmpfs_statfs,
499 .vfs_fhtovp = tmpfs_fhtovp,
500 .vfs_sync = vfs_stdsync
503 VFS_SET(tmpfs_vfsops, tmpfs, 0);