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_args.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;
92 node->tn_vpstate = TMPFS_VNODE_WANT;
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 = objcache_malloc_alloc(args, flags);
114 lockinit(&node->tn_interlock, "tmpfs node interlock", 0, LK_CANRECURSE);
115 node->tn_gen = karc4random();
121 tmpfs_node_fini(void *obj, void *args)
123 struct tmpfs_node *node = (struct tmpfs_node *)obj;
124 lockuninit(&node->tn_interlock);
125 objcache_malloc_free(obj, args);
129 tmpfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
131 struct tmpfs_mount *tmp;
132 struct tmpfs_node *root;
133 struct tmpfs_args args;
135 vm_pindex_t pages_limit;
145 /* Root node attributes. */
146 uid_t root_uid = cred->cr_uid;
147 gid_t root_gid = cred->cr_gid;
148 mode_t root_mode = (VREAD | VWRITE);
150 if (mp->mnt_flag & MNT_UPDATE) {
151 /* XXX: There is no support yet to update file system
152 * settings. Should be added. */
160 bzero(&args, sizeof(args));
167 error = copyin(data, &args, sizeof(args));
171 size_max = args.ta_size_max;
172 nodes_max = args.ta_nodes_max;
173 maxfsize_max = args.ta_maxfsize_max;
174 root_uid = args.ta_root_uid;
175 root_gid = args.ta_root_gid;
176 root_mode = args.ta_root_mode;
180 * If mount by non-root, then verify that user has necessary
181 * permissions on the device.
183 if (cred->cr_uid != 0) {
185 if ((mp->mnt_flag & MNT_RDONLY) == 0)
189 pages_limit = vm_swap_max + vmstats.v_page_count / 2;
192 pages = pages_limit / 2;
193 } else if (size_max < PAGE_SIZE) {
195 } else if (OFF_TO_IDX(size_max) > pages_limit) {
197 * do not force pages = pages_limit for this case, otherwise
198 * we might not honor tmpfs size requests from /etc/fstab
199 * during boot because they are mounted prior to swap being
202 pages = OFF_TO_IDX(size_max);
204 pages = OFF_TO_IDX(size_max);
208 nodes = 3 + pages * PAGE_SIZE / 1024;
209 else if (nodes_max < 3)
211 else if (nodes_max > pages)
216 maxfsize = IDX_TO_OFF(pages_limit);
217 if (maxfsize_max != 0 && maxfsize > maxfsize_max)
218 maxfsize = maxfsize_max;
220 /* Allocate the tmpfs mount structure and fill it. */
221 tmp = kmalloc(sizeof(*tmp), M_TMPFSMNT, M_WAITOK | M_ZERO);
224 tmp->tm_nodes_max = nodes;
225 tmp->tm_nodes_inuse = 0;
226 tmp->tm_maxfilesize = maxfsize;
227 LIST_INIT(&tmp->tm_nodes_used);
229 tmp->tm_pages_max = pages;
230 tmp->tm_pages_used = 0;
232 kmalloc_create(&tmp->tm_node_zone, "tmpfs node");
233 kmalloc_create(&tmp->tm_dirent_zone, "tmpfs dirent");
234 kmalloc_create(&tmp->tm_name_zone, "tmpfs name zone");
236 kmalloc_raise_limit(tmp->tm_node_zone, sizeof(struct tmpfs_node) *
239 tmp->tm_node_zone_malloc_args.objsize = sizeof(struct tmpfs_node);
240 tmp->tm_node_zone_malloc_args.mtype = tmp->tm_node_zone;
242 tmp->tm_dirent_zone_malloc_args.objsize = sizeof(struct tmpfs_dirent);
243 tmp->tm_dirent_zone_malloc_args.mtype = tmp->tm_dirent_zone;
245 tmp->tm_dirent_pool = objcache_create( "tmpfs dirent cache",
248 objcache_malloc_alloc, objcache_malloc_free,
249 &tmp->tm_dirent_zone_malloc_args);
250 tmp->tm_node_pool = objcache_create( "tmpfs node cache",
252 tmpfs_node_ctor, tmpfs_node_dtor, NULL,
253 tmpfs_node_init, tmpfs_node_fini,
254 &tmp->tm_node_zone_malloc_args);
256 tmp->tm_ino = TMPFS_ROOTINO;
258 /* Allocate the root node. */
259 error = tmpfs_alloc_node(tmp, VDIR, root_uid, root_gid,
260 root_mode & ALLPERMS, NULL,
261 VNOVAL, VNOVAL, &root);
264 * We are backed by swap, set snocache chflags flag so we
265 * don't trip over swapcache.
267 root->tn_flags = SF_NOCACHE;
269 if (error != 0 || root == NULL) {
270 objcache_destroy(tmp->tm_node_pool);
271 objcache_destroy(tmp->tm_dirent_pool);
272 kfree(tmp, M_TMPFSMNT);
275 KASSERT(root->tn_id == TMPFS_ROOTINO,
276 ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
278 ++root->tn_links; /* prevent destruction */
281 mp->mnt_flag |= MNT_LOCAL;
282 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
283 mp->mnt_kern_flag |= MNTK_NOMSYNC;
284 mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */
285 mp->mnt_data = (qaddr_t)tmp;
288 vfs_add_vnodeops(mp, &tmpfs_vnode_vops, &mp->mnt_vn_norm_ops);
289 vfs_add_vnodeops(mp, &tmpfs_fifo_vops, &mp->mnt_vn_fifo_ops);
291 copystr("tmpfs", mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
292 bzero(mp->mnt_stat.f_mntfromname +size, MNAMELEN - size);
293 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
294 copyinstr(path, mp->mnt_stat.f_mntonname,
295 sizeof(mp->mnt_stat.f_mntonname) -1,
298 tmpfs_statfs(mp, &mp->mnt_stat, cred);
303 /* --------------------------------------------------------------------- */
307 tmpfs_unmount(struct mount *mp, int mntflags)
311 struct tmpfs_mount *tmp;
312 struct tmpfs_node *node;
316 tmp = VFS_TO_TMPFS(mp);
319 /* Handle forced unmounts. */
320 if (mntflags & MNT_FORCE)
324 * Finalize all pending I/O. In the case of tmpfs we want
325 * to throw all the data away so clean out the buffer cache
326 * and vm objects before calling vflush().
328 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
330 * tn_links is mnt_token protected
332 TMPFS_NODE_LOCK(node);
334 TMPFS_NODE_UNLOCK(node);
336 while (node->tn_type == VREG && node->tn_vnode) {
342 * vx_get/vx_put and tmpfs_truncate may block,
343 * releasing the tmpfs mountpoint token.
345 * Make sure the lock order is correct.
347 vx_get(vp); /* held vnode */
348 TMPFS_NODE_LOCK(node);
349 if (node->tn_vnode == vp) {
350 tmpfs_truncate(vp, 0);
355 TMPFS_NODE_UNLOCK(node);
363 TMPFS_NODE_LOCK(node);
365 TMPFS_NODE_UNLOCK(node);
369 * Flush all vnodes on the unmount.
371 * If we fail to flush, we cannot unmount, but all the nodes have
372 * already been truncated. Erroring out is the best we can do.
374 error = vflush(mp, 0, flags);
381 * First pass get rid of all the directory entries and
382 * vnode associations. This will also destroy the
383 * directory topology and should drop all link counts
384 * to 0 except for the root.
386 * No vnodes should remain after the vflush above.
388 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
391 TMPFS_NODE_LOCK(node);
393 if (node->tn_type == VDIR) {
394 struct tmpfs_dirent *de;
396 while ((de = RB_ROOT(&node->tn_dir.tn_dirtree)) != NULL) {
397 tmpfs_dir_detach(node, de);
398 tmpfs_free_dirent(tmp, de);
401 KKASSERT(node->tn_vnode == NULL);
404 TMPFS_NODE_UNLOCK(node);
408 * Allow the root node to be destroyed by dropping the link count
409 * we bumped in the mount code.
411 KKASSERT(tmp->tm_root);
412 TMPFS_NODE_LOCK(tmp->tm_root);
413 --tmp->tm_root->tn_links;
414 TMPFS_NODE_UNLOCK(tmp->tm_root);
417 * At this point all nodes, including the root node, should have a
418 * link count of 0. The root is not necessarily going to be last.
420 while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
422 panic("tmpfs: Dangling nodes during umount (%p)!\n",
425 TMPFS_NODE_LOCK(node);
426 tmpfs_free_node(tmp, node);
430 KKASSERT(tmp->tm_root == NULL);
432 objcache_destroy(tmp->tm_dirent_pool);
433 objcache_destroy(tmp->tm_node_pool);
435 kmalloc_destroy(&tmp->tm_name_zone);
436 kmalloc_destroy(&tmp->tm_dirent_zone);
437 kmalloc_destroy(&tmp->tm_node_zone);
439 tmp->tm_node_zone = tmp->tm_dirent_zone = NULL;
441 KKASSERT(tmp->tm_pages_used == 0);
442 KKASSERT(tmp->tm_nodes_inuse == 0);
446 /* Throw away the tmpfs_mount structure. */
447 kfree(tmp, M_TMPFSMNT);
450 mp->mnt_flag &= ~MNT_LOCAL;
454 /* --------------------------------------------------------------------- */
457 tmpfs_root(struct mount *mp, struct vnode **vpp)
459 struct tmpfs_mount *tmp;
462 tmp = VFS_TO_TMPFS(mp);
463 if (tmp->tm_root == NULL) {
464 kprintf("tmpfs_root: called without root node %p\n", mp);
469 error = tmpfs_alloc_vp(mp, tmp->tm_root, LK_EXCLUSIVE, vpp);
470 (*vpp)->v_flag |= VROOT;
471 (*vpp)->v_type = VDIR;
476 /* --------------------------------------------------------------------- */
479 tmpfs_fhtovp(struct mount *mp, struct vnode *rootvp, struct fid *fhp,
483 struct tmpfs_fid *tfhp;
484 struct tmpfs_mount *tmp;
485 struct tmpfs_node *node;
488 tmp = VFS_TO_TMPFS(mp);
490 tfhp = (struct tmpfs_fid *) fhp;
491 if (tfhp->tf_len != sizeof(struct tmpfs_fid))
494 if (tfhp->tf_id >= tmp->tm_nodes_max)
501 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
502 if (node->tn_id == tfhp->tf_id &&
503 node->tn_gen == tfhp->tf_gen) {
510 rc = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
517 /* --------------------------------------------------------------------- */
521 tmpfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
523 fsfilcnt_t freenodes;
524 struct tmpfs_mount *tmp;
526 tmp = VFS_TO_TMPFS(mp);
529 sbp->f_iosize = PAGE_SIZE;
530 sbp->f_bsize = PAGE_SIZE;
532 sbp->f_blocks = tmp->tm_pages_max;
533 sbp->f_bavail = tmp->tm_pages_max - tmp->tm_pages_used;
534 sbp->f_bfree = sbp->f_bavail;
536 freenodes = tmp->tm_nodes_max - tmp->tm_nodes_inuse;
538 sbp->f_files = freenodes + tmp->tm_nodes_inuse;
539 sbp->f_ffree = freenodes;
540 sbp->f_owner = tmp->tm_root->tn_uid;
547 /* --------------------------------------------------------------------- */
550 tmpfs_vptofh(struct vnode *vp, struct fid *fhp)
552 struct tmpfs_node *node;
553 struct tmpfs_fid tfh;
554 node = VP_TO_TMPFS_NODE(vp);
555 memset(&tfh, 0, sizeof(tfh));
556 tfh.tf_len = sizeof(struct tmpfs_fid);
557 tfh.tf_gen = node->tn_gen;
558 tfh.tf_id = node->tn_id;
559 memcpy(fhp, &tfh, sizeof(tfh));
563 /* --------------------------------------------------------------------- */
566 tmpfs_checkexp(struct mount *mp, struct sockaddr *nam, int *exflagsp,
567 struct ucred **credanonp)
569 struct tmpfs_mount *tmp;
572 tmp = (struct tmpfs_mount *) mp->mnt_data;
573 nc = vfs_export_lookup(mp, &tmp->tm_export, nam);
577 *exflagsp = nc->netc_exflags;
578 *credanonp = &nc->netc_anon;
583 /* --------------------------------------------------------------------- */
586 * tmpfs vfs operations.
589 static struct vfsops tmpfs_vfsops = {
590 .vfs_mount = tmpfs_mount,
591 .vfs_unmount = tmpfs_unmount,
592 .vfs_root = tmpfs_root,
593 .vfs_statfs = tmpfs_statfs,
594 .vfs_fhtovp = tmpfs_fhtovp,
595 .vfs_vptofh = tmpfs_vptofh,
596 .vfs_checkexp = tmpfs_checkexp,
599 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_MPSAFE);
600 MODULE_VERSION(tmpfs, 1);