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 virtual memory to store file data and
37 * metadata efficiently. It does not follow the structure of an on-disk
38 * file system because it simply does not need to. Instead, it uses
39 * memory-specific data structures and algorithms to automatically
40 * allocate and release resources.
44 #include <sys/param.h>
45 #include <sys/limits.h>
47 #include <sys/mutex.h>
48 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/sysctl.h>
52 #include <sys/objcache.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_param.h>
59 #include <vfs/tmpfs/tmpfs.h>
62 #include <vfs/tmpfs/tmpfs_vnops.h>
63 #include <vfs/tmpfs/tmpfs_mount.h>
66 * Default permission for root node
68 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
70 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
72 /* --------------------------------------------------------------------- */
74 static int tmpfs_mount(struct mount *, char *, caddr_t, struct ucred *);
75 static int tmpfs_unmount(struct mount *, int);
76 static int tmpfs_root(struct mount *, struct vnode **);
77 static int tmpfs_fhtovp(struct mount *, struct vnode *, struct fid *, struct vnode **);
78 static int tmpfs_statfs(struct mount *, struct statfs *, struct ucred *cred);
80 /* --------------------------------------------------------------------- */
82 tmpfs_node_ctor(void *obj, void *privdata, int flags)
84 struct tmpfs_node *node = obj;
91 node->tn_vnode = NULL;
93 bzero(&node->tn_spec, sizeof(node->tn_spec));
99 tmpfs_node_dtor(void *obj, void *privdata)
101 struct tmpfs_node *node = (struct tmpfs_node *)obj;
102 node->tn_type = VNON;
103 node->tn_vpstate = TMPFS_VNODE_DOOMED;
107 tmpfs_node_init(void *args, int flags)
109 struct tmpfs_node *node;
111 node = objcache_malloc_alloc(args, flags);
115 node->tn_blksize = PAGE_SIZE; /* start small */
117 lockinit(&node->tn_interlock, "tmpfs node interlock", 0, LK_CANRECURSE);
118 node->tn_gen = karc4random();
124 tmpfs_node_fini(void *obj, void *args)
126 struct tmpfs_node *node = (struct tmpfs_node *)obj;
127 lockuninit(&node->tn_interlock);
128 objcache_malloc_free(obj, args);
132 tmpfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
134 struct tmpfs_mount *tmp;
135 struct tmpfs_node *root;
136 struct tmpfs_mount_info args;
138 vm_pindex_t pages_limit;
148 /* Root node attributes. */
149 uid_t root_uid = cred->cr_uid;
150 gid_t root_gid = cred->cr_gid;
151 mode_t root_mode = (VREAD | VWRITE);
153 if (mp->mnt_flag & MNT_UPDATE) {
154 /* XXX: There is no support yet to update file system
155 * settings. Should be added. */
163 bzero(&args, sizeof(args));
170 error = copyin(data, &args, sizeof(args));
174 size_max = args.ta_size_max;
175 nodes_max = args.ta_nodes_max;
176 maxfsize_max = args.ta_maxfsize_max;
177 root_uid = args.ta_root_uid;
178 root_gid = args.ta_root_gid;
179 root_mode = args.ta_root_mode;
183 * If mount by non-root, then verify that user has necessary
184 * permissions on the device.
186 if (cred->cr_uid != 0) {
188 if ((mp->mnt_flag & MNT_RDONLY) == 0)
192 pages_limit = vm_swap_max + vmstats.v_page_count / 2;
195 pages = pages_limit / 2;
196 } else if (size_max < PAGE_SIZE) {
198 } else if (OFF_TO_IDX(size_max) > pages_limit) {
200 * do not force pages = pages_limit for this case, otherwise
201 * we might not honor tmpfs size requests from /etc/fstab
202 * during boot because they are mounted prior to swap being
205 pages = OFF_TO_IDX(size_max);
207 pages = OFF_TO_IDX(size_max);
211 nodes = 3 + pages * PAGE_SIZE / 1024;
212 else if (nodes_max < 3)
214 else if (nodes_max > pages)
219 maxfsize = 0x7FFFFFFFFFFFFFFFLLU - TMPFS_BLKSIZE;
220 if (maxfsize_max != 0 && maxfsize > maxfsize_max)
221 maxfsize = maxfsize_max;
223 /* Allocate the tmpfs mount structure and fill it. */
224 tmp = kmalloc(sizeof(*tmp), M_TMPFSMNT, M_WAITOK | M_ZERO);
227 tmp->tm_nodes_max = nodes;
228 tmp->tm_nodes_inuse = 0;
229 tmp->tm_maxfilesize = maxfsize;
230 LIST_INIT(&tmp->tm_nodes_used);
232 tmp->tm_pages_max = pages;
233 tmp->tm_pages_used = 0;
235 kmalloc_create(&tmp->tm_node_zone, "tmpfs node");
236 kmalloc_create(&tmp->tm_dirent_zone, "tmpfs dirent");
237 kmalloc_create(&tmp->tm_name_zone, "tmpfs name zone");
239 kmalloc_raise_limit(tmp->tm_node_zone, sizeof(struct tmpfs_node) *
242 tmp->tm_node_zone_malloc_args.objsize = sizeof(struct tmpfs_node);
243 tmp->tm_node_zone_malloc_args.mtype = tmp->tm_node_zone;
245 tmp->tm_dirent_zone_malloc_args.objsize = sizeof(struct tmpfs_dirent);
246 tmp->tm_dirent_zone_malloc_args.mtype = tmp->tm_dirent_zone;
248 tmp->tm_dirent_pool = objcache_create( "tmpfs dirent cache",
251 objcache_malloc_alloc, objcache_malloc_free,
252 &tmp->tm_dirent_zone_malloc_args);
253 tmp->tm_node_pool = objcache_create( "tmpfs node cache",
255 tmpfs_node_ctor, tmpfs_node_dtor, NULL,
256 tmpfs_node_init, tmpfs_node_fini,
257 &tmp->tm_node_zone_malloc_args);
259 tmp->tm_ino = TMPFS_ROOTINO;
261 /* Allocate the root node. */
262 error = tmpfs_alloc_node(tmp, VDIR, root_uid, root_gid,
263 root_mode & ALLPERMS, NULL,
264 VNOVAL, VNOVAL, &root);
267 * We are backed by swap, set snocache chflags flag so we
268 * don't trip over swapcache.
270 root->tn_flags = SF_NOCACHE;
272 if (error != 0 || root == NULL) {
273 objcache_destroy(tmp->tm_node_pool);
274 objcache_destroy(tmp->tm_dirent_pool);
275 kfree(tmp, M_TMPFSMNT);
278 KASSERT(root->tn_id == TMPFS_ROOTINO,
279 ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
281 ++root->tn_links; /* prevent destruction */
284 mp->mnt_flag |= MNT_LOCAL;
285 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
286 mp->mnt_kern_flag |= MNTK_NOMSYNC;
287 mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */
288 mp->mnt_kern_flag |= MNTK_QUICKHALT; /* no teardown needed on halt */
289 mp->mnt_data = (qaddr_t)tmp;
290 mp->mnt_iosize_max = MAXBSIZE;
293 vfs_add_vnodeops(mp, &tmpfs_vnode_vops, &mp->mnt_vn_norm_ops);
294 vfs_add_vnodeops(mp, &tmpfs_fifo_vops, &mp->mnt_vn_fifo_ops);
296 copystr("tmpfs", mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
297 bzero(mp->mnt_stat.f_mntfromname +size, MNAMELEN - size);
298 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
299 copyinstr(path, mp->mnt_stat.f_mntonname,
300 sizeof(mp->mnt_stat.f_mntonname) -1,
303 tmpfs_statfs(mp, &mp->mnt_stat, cred);
308 /* --------------------------------------------------------------------- */
312 tmpfs_unmount(struct mount *mp, int mntflags)
316 struct tmpfs_mount *tmp;
317 struct tmpfs_node *node;
321 tmp = VFS_TO_TMPFS(mp);
324 /* Handle forced unmounts. */
325 if (mntflags & MNT_FORCE)
329 * Finalize all pending I/O. In the case of tmpfs we want
330 * to throw all the data away so clean out the buffer cache
331 * and vm objects before calling vflush().
333 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
335 * tn_links is mnt_token protected
337 TMPFS_NODE_LOCK(node);
340 while (node->tn_type == VREG && node->tn_vnode) {
343 TMPFS_NODE_UNLOCK(node);
347 * vx_get/vx_put and tmpfs_truncate may block,
348 * releasing the tmpfs mountpoint token.
350 * Make sure the lock order is correct.
352 vx_get(vp); /* held vnode */
353 TMPFS_NODE_LOCK(node);
354 if (node->tn_vnode == vp) {
355 tmpfs_truncate(vp, 0);
360 TMPFS_NODE_UNLOCK(node);
363 TMPFS_NODE_LOCK(node);
370 TMPFS_NODE_UNLOCK(node);
374 * Flush all vnodes on the unmount.
376 * If we fail to flush, we cannot unmount, but all the nodes have
377 * already been truncated. Erroring out is the best we can do.
379 error = vflush(mp, 0, flags);
386 * First pass get rid of all the directory entries and
387 * vnode associations. This will also destroy the
388 * directory topology and should drop all link counts
389 * to 0 except for the root.
391 * No vnodes should remain after the vflush above.
393 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
396 TMPFS_NODE_LOCK(node);
398 if (node->tn_type == VDIR) {
399 struct tmpfs_dirent *de;
401 while ((de = RB_ROOT(&node->tn_dir.tn_dirtree)) != NULL) {
402 tmpfs_dir_detach(node, de);
403 tmpfs_free_dirent(tmp, de);
406 KKASSERT(node->tn_vnode == NULL);
409 TMPFS_NODE_UNLOCK(node);
413 * Allow the root node to be destroyed by dropping the link count
414 * we bumped in the mount code.
416 KKASSERT(tmp->tm_root);
417 TMPFS_NODE_LOCK(tmp->tm_root);
418 --tmp->tm_root->tn_links;
419 TMPFS_NODE_UNLOCK(tmp->tm_root);
422 * At this point all nodes, including the root node, should have a
423 * link count of 0. The root is not necessarily going to be last.
425 while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
427 panic("tmpfs: Dangling nodes during umount (%p)!\n",
430 TMPFS_NODE_LOCK(node);
431 tmpfs_free_node(tmp, node);
435 KKASSERT(tmp->tm_root == NULL);
437 objcache_destroy(tmp->tm_dirent_pool);
438 objcache_destroy(tmp->tm_node_pool);
440 kmalloc_destroy(&tmp->tm_name_zone);
441 kmalloc_destroy(&tmp->tm_dirent_zone);
442 kmalloc_destroy(&tmp->tm_node_zone);
444 tmp->tm_node_zone = tmp->tm_dirent_zone = NULL;
446 KKASSERT(tmp->tm_pages_used == 0);
447 KKASSERT(tmp->tm_nodes_inuse == 0);
451 /* Throw away the tmpfs_mount structure. */
452 kfree(tmp, M_TMPFSMNT);
455 mp->mnt_flag &= ~MNT_LOCAL;
459 /* --------------------------------------------------------------------- */
462 tmpfs_root(struct mount *mp, struct vnode **vpp)
464 struct tmpfs_mount *tmp;
467 tmp = VFS_TO_TMPFS(mp);
468 if (tmp->tm_root == NULL) {
469 kprintf("tmpfs_root: called without root node %p\n", mp);
474 error = tmpfs_alloc_vp(mp, NULL, tmp->tm_root,
476 (*vpp)->v_flag |= VROOT;
477 (*vpp)->v_type = VDIR;
482 /* --------------------------------------------------------------------- */
485 tmpfs_fhtovp(struct mount *mp, struct vnode *rootvp, struct fid *fhp,
489 struct tmpfs_fid *tfhp;
490 struct tmpfs_mount *tmp;
491 struct tmpfs_node *node;
494 tmp = VFS_TO_TMPFS(mp);
496 tfhp = (struct tmpfs_fid *) fhp;
497 if (tfhp->tf_len != sizeof(struct tmpfs_fid))
504 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
505 if (node->tn_id == tfhp->tf_id &&
506 node->tn_gen == tfhp->tf_gen) {
513 rc = tmpfs_alloc_vp(mp, NULL, node, LK_EXCLUSIVE, vpp);
520 /* --------------------------------------------------------------------- */
524 tmpfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
526 fsfilcnt_t freenodes;
527 struct tmpfs_mount *tmp;
529 tmp = VFS_TO_TMPFS(mp);
532 sbp->f_iosize = PAGE_SIZE;
533 sbp->f_bsize = PAGE_SIZE;
535 sbp->f_blocks = tmp->tm_pages_max;
536 sbp->f_bavail = tmp->tm_pages_max - tmp->tm_pages_used;
537 sbp->f_bfree = sbp->f_bavail;
539 freenodes = tmp->tm_nodes_max - tmp->tm_nodes_inuse;
541 sbp->f_files = freenodes + tmp->tm_nodes_inuse;
542 sbp->f_ffree = freenodes;
543 sbp->f_owner = tmp->tm_root->tn_uid;
550 /* --------------------------------------------------------------------- */
553 tmpfs_vptofh(struct vnode *vp, struct fid *fhp)
555 struct tmpfs_node *node;
556 struct tmpfs_fid tfh;
557 node = VP_TO_TMPFS_NODE(vp);
558 memset(&tfh, 0, sizeof(tfh));
559 tfh.tf_len = sizeof(struct tmpfs_fid);
560 tfh.tf_gen = node->tn_gen;
561 tfh.tf_id = node->tn_id;
562 memcpy(fhp, &tfh, sizeof(tfh));
566 /* --------------------------------------------------------------------- */
569 tmpfs_checkexp(struct mount *mp, struct sockaddr *nam, int *exflagsp,
570 struct ucred **credanonp)
572 struct tmpfs_mount *tmp;
575 tmp = (struct tmpfs_mount *) mp->mnt_data;
576 nc = vfs_export_lookup(mp, &tmp->tm_export, nam);
580 *exflagsp = nc->netc_exflags;
581 *credanonp = &nc->netc_anon;
586 /* --------------------------------------------------------------------- */
589 * tmpfs vfs operations.
592 static struct vfsops tmpfs_vfsops = {
594 .vfs_mount = tmpfs_mount,
595 .vfs_unmount = tmpfs_unmount,
596 .vfs_root = tmpfs_root,
597 .vfs_statfs = tmpfs_statfs,
598 .vfs_fhtovp = tmpfs_fhtovp,
599 .vfs_vptofh = tmpfs_vptofh,
600 .vfs_checkexp = tmpfs_checkexp,
603 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_MPSAFE);
604 MODULE_VERSION(tmpfs, 1);