xref: /openbsd/sys/tmpfs/tmpfs_vfsops.c (revision 73471bf0)
1 /*	$OpenBSD: tmpfs_vfsops.c,v 1.19 2021/10/24 15:41:47 patrick Exp $	*/
2 /*	$NetBSD: tmpfs_vfsops.c,v 1.52 2011/09/27 01:10:43 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 2005, 2006, 2007 The NetBSD Foundation, Inc.
6  * All rights reserved.
7  *
8  * This code is derived from software contributed to The NetBSD Foundation
9  * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
10  * 2005 program.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
23  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
24  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
25  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 /*
35  * Efficient memory file system.
36  *
37  * tmpfs is a file system that uses NetBSD's virtual memory sub-system
38  * (the well-known UVM) to store file data and metadata in an efficient
39  * way.  This means that it does not follow the structure of an on-disk
40  * file system because it simply does not need to.  Instead, it uses
41  * memory-specific data structures and algorithms to automatically
42  * allocate and release resources.
43  */
44 
45 #include <sys/param.h>
46 #include <sys/mount.h>
47 #include <sys/stat.h>
48 #include <sys/systm.h>
49 #include <sys/vnode.h>
50 #include <sys/malloc.h>
51 
52 #include <tmpfs/tmpfs.h>
53 
54 /* MODULE(MODULE_CLASS_VFS, tmpfs, NULL); */
55 
56 struct pool	tmpfs_dirent_pool;
57 struct pool	tmpfs_node_pool;
58 
59 int	tmpfs_mount(struct mount *, const char *, void *, struct nameidata *,
60 	    struct proc *);
61 int	tmpfs_start(struct mount *, int, struct proc *);
62 int	tmpfs_unmount(struct mount *, int, struct proc *);
63 int	tmpfs_root(struct mount *, struct vnode **);
64 int	tmpfs_vget(struct mount *, ino_t, struct vnode **);
65 int	tmpfs_fhtovp(struct mount *, struct fid *, struct vnode **);
66 int	tmpfs_vptofh(struct vnode *, struct fid *);
67 int	tmpfs_statfs(struct mount *, struct statfs *, struct proc *);
68 int	tmpfs_sync(struct mount *, int, int, struct ucred *, struct proc *);
69 int	tmpfs_init(struct vfsconf *);
70 int	tmpfs_mount_update(struct mount *);
71 
72 int
73 tmpfs_init(struct vfsconf *vfsp)
74 {
75 
76 	pool_init(&tmpfs_dirent_pool, sizeof(tmpfs_dirent_t), 0, IPL_NONE,
77 	    PR_WAITOK, "tmpfs_dirent", NULL);
78 	pool_init(&tmpfs_node_pool, sizeof(tmpfs_node_t), 0, IPL_NONE,
79 	    PR_WAITOK, "tmpfs_node", NULL);
80 
81 	return 0;
82 }
83 
84 int
85 tmpfs_mount_update(struct mount *mp)
86 {
87 	tmpfs_mount_t *tmp;
88 	struct vnode *rootvp;
89 	int error;
90 
91 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
92 		return EOPNOTSUPP;
93 
94 	/* ro->rw transition: nothing to do? */
95 	if (mp->mnt_flag & MNT_WANTRDWR)
96 		return 0;
97 
98 	tmp = mp->mnt_data;
99 	rootvp = tmp->tm_root->tn_vnode;
100 
101 	/* Lock root to prevent lookups. */
102 	error = vn_lock(rootvp, LK_EXCLUSIVE | LK_RETRY);
103 	if (error)
104 		return error;
105 
106 	/* Lock mount point to prevent nodes from being added/removed. */
107 	rw_enter_write(&tmp->tm_lock);
108 
109 	/* Flush files opened for writing; skip rootvp. */
110 	error = vflush(mp, rootvp, WRITECLOSE);
111 
112 	rw_exit_write(&tmp->tm_lock);
113 	VOP_UNLOCK(rootvp);
114 
115 	return error;
116 }
117 
118 int
119 tmpfs_mount(struct mount *mp, const char *path, void *data,
120     struct nameidata *ndp, struct proc *p)
121 {
122 	struct tmpfs_args *args = data;
123 	tmpfs_mount_t *tmp;
124 	tmpfs_node_t *root;
125 	uint64_t memlimit;
126 	uint64_t nodes;
127 	int error;
128 
129 	if (mp->mnt_flag & MNT_UPDATE)
130 		return (tmpfs_mount_update(mp));
131 
132 	/* Prohibit mounts if there is not enough memory. */
133 	if (tmpfs_mem_info(1) < TMPFS_PAGES_RESERVED)
134 		return EINVAL;
135 
136 	if (args->ta_root_uid == VNOVAL || args->ta_root_gid == VNOVAL ||
137 	    args->ta_root_mode == VNOVAL)
138 		return EINVAL;
139 
140 	/* Get the memory usage limit for this file-system. */
141 	if (args->ta_size_max < PAGE_SIZE) {
142 		memlimit = UINT64_MAX;
143 	} else {
144 		memlimit = args->ta_size_max;
145 	}
146 	KASSERT(memlimit > 0);
147 
148 	if (args->ta_nodes_max <= 3) {
149 		nodes = 3 + (memlimit / 1024);
150 	} else {
151 		nodes = args->ta_nodes_max;
152 	}
153 	nodes = MIN(nodes, INT_MAX);
154 	KASSERT(nodes >= 3);
155 
156 	/* Allocate the tmpfs mount structure and fill it. */
157 	tmp = malloc(sizeof(tmpfs_mount_t), M_MISCFSMNT, M_WAITOK);
158 
159 	tmp->tm_nodes_max = (ino_t)nodes;
160 	tmp->tm_nodes_cnt = 0;
161 	tmp->tm_highest_inode = 1;
162 	LIST_INIT(&tmp->tm_nodes);
163 
164 	rw_init(&tmp->tm_lock, "tmplk");
165 	tmpfs_mntmem_init(tmp, memlimit);
166 
167 	/* Allocate the root node. */
168 	error = tmpfs_alloc_node(tmp, VDIR, args->ta_root_uid,
169 	    args->ta_root_gid, args->ta_root_mode & ALLPERMS, NULL,
170 	    VNOVAL, &root);
171 	KASSERT(error == 0 && root != NULL);
172 
173 	/*
174 	 * Parent of the root inode is itself.  Also, root inode has no
175 	 * directory entry (i.e. is never attached), thus hold an extra
176 	 * reference (link) for it.
177 	 */
178 	root->tn_links++;
179 	root->tn_spec.tn_dir.tn_parent = root;
180 	tmp->tm_root = root;
181 
182 	mp->mnt_data = tmp;
183 	mp->mnt_flag |= MNT_LOCAL;
184 	mp->mnt_stat.f_namemax = TMPFS_MAXNAMLEN;
185 	vfs_getnewfsid(mp);
186 
187 	mp->mnt_stat.mount_info.tmpfs_args = *args;
188 
189 	bzero(&mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
190 	bzero(&mp->mnt_stat.f_mntfromname, sizeof(mp->mnt_stat.f_mntfromname));
191 	bzero(&mp->mnt_stat.f_mntfromspec, sizeof(mp->mnt_stat.f_mntfromspec));
192 
193 	strlcpy(mp->mnt_stat.f_mntonname, path,
194 	    sizeof(mp->mnt_stat.f_mntonname) - 1);
195 	strlcpy(mp->mnt_stat.f_mntfromname, "tmpfs",
196 	    sizeof(mp->mnt_stat.f_mntfromname) - 1);
197 	strlcpy(mp->mnt_stat.f_mntfromspec, "tmpfs",
198 	    sizeof(mp->mnt_stat.f_mntfromspec) - 1);
199 
200 	return error;
201 }
202 
203 int
204 tmpfs_start(struct mount *mp, int flags, struct proc *p)
205 {
206 	return 0;
207 }
208 
209 int
210 tmpfs_unmount(struct mount *mp, int mntflags, struct proc *p)
211 {
212 	tmpfs_mount_t *tmp = VFS_TO_TMPFS(mp);
213 	tmpfs_node_t *node, *cnode;
214 	int error, flags = 0;
215 
216 	/* Handle forced unmounts. */
217 	if (mntflags & MNT_FORCE)
218 		flags |= FORCECLOSE;
219 
220 	/* Finalize all pending I/O. */
221 	error = vflush(mp, NULL, flags);
222 	if (error != 0)
223 		return error;
224 
225 	/*
226 	 * First round, detach and destroy all directory entries.
227 	 * Also, clear the pointers to the vnodes - they are gone.
228 	 */
229 	LIST_FOREACH(node, &tmp->tm_nodes, tn_entries) {
230 		tmpfs_dirent_t *de;
231 
232 		node->tn_vnode = NULL;
233 		if (node->tn_type != VDIR) {
234 			continue;
235 		}
236 		while ((de = TAILQ_FIRST(&node->tn_spec.tn_dir.tn_dir)) != NULL) {
237 			cnode = de->td_node;
238 			if (cnode)
239 				cnode->tn_vnode = NULL;
240 			tmpfs_dir_detach(node, de);
241 			tmpfs_free_dirent(tmp, de);
242 		}
243 	}
244 
245 	/* Second round, destroy all inodes. */
246 	while ((node = LIST_FIRST(&tmp->tm_nodes)) != NULL) {
247 		tmpfs_free_node(tmp, node);
248 	}
249 
250 	/* Throw away the tmpfs_mount structure. */
251 	tmpfs_mntmem_destroy(tmp);
252 	/* mutex_destroy(&tmp->tm_lock); */
253 	/* kmem_free(tmp, sizeof(*tmp)); */
254 	free(tmp, M_MISCFSMNT, sizeof(tmpfs_mount_t));
255 	mp->mnt_data = NULL;
256 
257 	return 0;
258 }
259 
260 int
261 tmpfs_root(struct mount *mp, struct vnode **vpp)
262 {
263 	tmpfs_node_t *node = VFS_TO_TMPFS(mp)->tm_root;
264 
265 	rw_enter_write(&node->tn_nlock);
266 	return tmpfs_vnode_get(mp, node, vpp);
267 }
268 
269 int
270 tmpfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
271 {
272 
273 	printf("tmpfs_vget called; need for it unknown yet\n");
274 	return EOPNOTSUPP;
275 }
276 
277 int
278 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
279 {
280 	tmpfs_mount_t *tmp = VFS_TO_TMPFS(mp);
281 	tmpfs_node_t *node;
282 	tmpfs_fid_t tfh;
283 
284 	if (fhp->fid_len != sizeof(tmpfs_fid_t)) {
285 		return EINVAL;
286 	}
287 	memcpy(&tfh, fhp, sizeof(tmpfs_fid_t));
288 
289 	rw_enter_write(&tmp->tm_lock);
290 	LIST_FOREACH(node, &tmp->tm_nodes, tn_entries) {
291 		if (node->tn_id != tfh.tf_id) {
292 			continue;
293 		}
294 		if (TMPFS_NODE_GEN(node) != tfh.tf_gen) {
295 			continue;
296 		}
297 		rw_enter_write(&node->tn_nlock);
298 		break;
299 	}
300 	rw_exit_write(&tmp->tm_lock);
301 
302 	/* Will release the tn_nlock. */
303 	return node ? tmpfs_vnode_get(mp, node, vpp) : ESTALE;
304 }
305 
306 int
307 tmpfs_vptofh(struct vnode *vp, struct fid *fhp)
308 {
309 	tmpfs_fid_t tfh;
310 	tmpfs_node_t *node;
311 
312 	node = VP_TO_TMPFS_NODE(vp);
313 
314 	memset(&tfh, 0, sizeof(tfh));
315 	tfh.tf_len = sizeof(tmpfs_fid_t);
316 	tfh.tf_gen = TMPFS_NODE_GEN(node);
317 	tfh.tf_id = node->tn_id;
318 	memcpy(fhp, &tfh, sizeof(tfh));
319 
320 	return 0;
321 }
322 
323 int
324 tmpfs_statfs(struct mount *mp, struct statfs *sbp, struct proc *p)
325 {
326 	tmpfs_mount_t *tmp;
327 	fsfilcnt_t freenodes;
328 	uint64_t avail;
329 
330 	tmp = VFS_TO_TMPFS(mp);
331 
332 	sbp->f_iosize = sbp->f_bsize = PAGE_SIZE;
333 
334 	rw_enter_write(&tmp->tm_acc_lock);
335 	avail =  tmpfs_pages_avail(tmp);
336 	sbp->f_blocks = (tmpfs_bytes_max(tmp) >> PAGE_SHIFT);
337 	sbp->f_bfree = avail;
338 	sbp->f_bavail = avail & INT64_MAX; /* f_bavail is int64_t */
339 
340 	freenodes = MIN(tmp->tm_nodes_max - tmp->tm_nodes_cnt,
341 	    avail * PAGE_SIZE / sizeof(tmpfs_node_t));
342 
343 	sbp->f_files = tmp->tm_nodes_cnt + freenodes;
344 	sbp->f_ffree = freenodes;
345 	sbp->f_favail = freenodes & INT64_MAX; /* f_favail is int64_t */
346 	rw_exit_write(&tmp->tm_acc_lock);
347 
348 	copy_statfs_info(sbp, mp);
349 
350 	return 0;
351 }
352 
353 int
354 tmpfs_sync(struct mount *mp, int waitfor, int stall, struct ucred *cred,
355     struct proc *p)
356 {
357 
358 	return 0;
359 }
360 
361 /*
362  * tmpfs vfs operations.
363  */
364 
365 const struct vfsops tmpfs_vfsops = {
366 	.vfs_mount	= tmpfs_mount,
367 	.vfs_start	= tmpfs_start,
368 	.vfs_unmount	= tmpfs_unmount,
369 	.vfs_root	= tmpfs_root,
370 	.vfs_quotactl	= (void *)eopnotsupp,
371 	.vfs_statfs	= tmpfs_statfs,
372 	.vfs_sync	= tmpfs_sync,
373 	.vfs_vget	= tmpfs_vget,
374 	.vfs_fhtovp	= tmpfs_fhtovp,
375 	.vfs_vptofh	= tmpfs_vptofh,
376 	.vfs_init	= tmpfs_init,
377 	.vfs_sysctl	= (void *)eopnotsupp,
378 	.vfs_checkexp	= (void *)eopnotsupp,
379 };
380