xref: /freebsd/sys/fs/tmpfs/tmpfs_vfsops.c (revision 7750ad47a9a7dbc83f87158464170c8640723293)
1 /*	$NetBSD: tmpfs_vfsops.c,v 1.10 2005/12/11 12:24:29 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2005 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
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
9  * 2005 program.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
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.
19  *
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.
31  */
32 
33 /*
34  * Efficient memory file system.
35  *
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.
42  */
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include <sys/param.h>
47 #include <sys/limits.h>
48 #include <sys/lock.h>
49 #include <sys/mutex.h>
50 #include <sys/kernel.h>
51 #include <sys/stat.h>
52 #include <sys/systm.h>
53 #include <sys/sysctl.h>
54 
55 #include <vm/vm.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_param.h>
58 
59 #include <fs/tmpfs/tmpfs.h>
60 
61 /*
62  * Default permission for root node
63  */
64 #define TMPFS_DEFAULT_ROOT_MODE	(S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
65 
66 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
67 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
68 
69 /* --------------------------------------------------------------------- */
70 
71 static int	tmpfs_mount(struct mount *);
72 static int	tmpfs_unmount(struct mount *, int);
73 static int	tmpfs_root(struct mount *, int flags, struct vnode **);
74 static int	tmpfs_fhtovp(struct mount *, struct fid *, int,
75 		    struct vnode **);
76 static int	tmpfs_statfs(struct mount *, struct statfs *);
77 
78 /* --------------------------------------------------------------------- */
79 
80 static const char *tmpfs_opts[] = {
81 	"from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export",
82 	NULL
83 };
84 
85 static const char *tmpfs_updateopts[] = {
86 	"from", "export", NULL
87 };
88 
89 /* --------------------------------------------------------------------- */
90 
91 static int
92 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
93 {
94 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
95 
96 	node->tn_gen++;
97 	node->tn_size = 0;
98 	node->tn_status = 0;
99 	node->tn_flags = 0;
100 	node->tn_links = 0;
101 	node->tn_vnode = NULL;
102 	node->tn_vpstate = 0;
103 
104 	return (0);
105 }
106 
107 static void
108 tmpfs_node_dtor(void *mem, int size, void *arg)
109 {
110 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
111 	node->tn_type = VNON;
112 }
113 
114 static int
115 tmpfs_node_init(void *mem, int size, int flags)
116 {
117 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
118 	node->tn_id = 0;
119 
120 	mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
121 	node->tn_gen = arc4random();
122 
123 	return (0);
124 }
125 
126 static void
127 tmpfs_node_fini(void *mem, int size)
128 {
129 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
130 
131 	mtx_destroy(&node->tn_interlock);
132 }
133 
134 static int
135 tmpfs_mount(struct mount *mp)
136 {
137 	const size_t nodes_per_page = howmany(PAGE_SIZE,
138 	    sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
139 	struct tmpfs_mount *tmp;
140 	struct tmpfs_node *root;
141 	int error;
142 	/* Size counters. */
143 	u_quad_t pages;
144 	off_t nodes_max, size_max, maxfilesize;
145 
146 	/* Root node attributes. */
147 	uid_t root_uid;
148 	gid_t root_gid;
149 	mode_t root_mode;
150 
151 	struct vattr va;
152 
153 	if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
154 		return (EINVAL);
155 
156 	if (mp->mnt_flag & MNT_UPDATE) {
157 		/* Only support update mounts for certain options. */
158 		if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
159 			return (EOPNOTSUPP);
160 		if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) !=
161 		    ((struct tmpfs_mount *)mp->mnt_data)->tm_ronly)
162 			return (EOPNOTSUPP);
163 		return (0);
164 	}
165 
166 	vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
167 	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
168 	VOP_UNLOCK(mp->mnt_vnodecovered, 0);
169 	if (error)
170 		return (error);
171 
172 	if (mp->mnt_cred->cr_ruid != 0 ||
173 	    vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
174 		root_gid = va.va_gid;
175 	if (mp->mnt_cred->cr_ruid != 0 ||
176 	    vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
177 		root_uid = va.va_uid;
178 	if (mp->mnt_cred->cr_ruid != 0 ||
179 	    vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
180 		root_mode = va.va_mode;
181 	if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
182 		nodes_max = 0;
183 	if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
184 		size_max = 0;
185 	if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
186 		maxfilesize = 0;
187 
188 	/* Do not allow mounts if we do not have enough memory to preserve
189 	 * the minimum reserved pages. */
190 	if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
191 		return ENOSPC;
192 
193 	/* Get the maximum number of memory pages this file system is
194 	 * allowed to use, based on the maximum size the user passed in
195 	 * the mount structure.  A value of zero is treated as if the
196 	 * maximum available space was requested. */
197 	if (size_max < PAGE_SIZE || size_max > OFF_MAX - PAGE_SIZE ||
198 	    (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
199 		pages = SIZE_MAX;
200 	else
201 		pages = howmany(size_max, PAGE_SIZE);
202 	MPASS(pages > 0);
203 
204 	if (nodes_max <= 3) {
205 		if (pages < INT_MAX / nodes_per_page)
206 			nodes_max = pages * nodes_per_page;
207 		else
208 			nodes_max = INT_MAX;
209 	}
210 	if (nodes_max > INT_MAX)
211 		nodes_max = INT_MAX;
212 	MPASS(nodes_max >= 3);
213 
214 	/* Allocate the tmpfs mount structure and fill it. */
215 	tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
216 	    M_TMPFSMNT, M_WAITOK | M_ZERO);
217 
218 	mtx_init(&tmp->allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
219 	tmp->tm_nodes_max = nodes_max;
220 	tmp->tm_nodes_inuse = 0;
221 	tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
222 	LIST_INIT(&tmp->tm_nodes_used);
223 
224 	tmp->tm_pages_max = pages;
225 	tmp->tm_pages_used = 0;
226 	tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->allnode_lock);
227 	tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
228 	    sizeof(struct tmpfs_dirent),
229 	    NULL, NULL, NULL, NULL,
230 	    UMA_ALIGN_PTR, 0);
231 	tmp->tm_node_pool = uma_zcreate("TMPFS node",
232 	    sizeof(struct tmpfs_node),
233 	    tmpfs_node_ctor, tmpfs_node_dtor,
234 	    tmpfs_node_init, tmpfs_node_fini,
235 	    UMA_ALIGN_PTR, 0);
236 	tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
237 
238 	/* Allocate the root node. */
239 	error = tmpfs_alloc_node(tmp, VDIR, root_uid,
240 	    root_gid, root_mode & ALLPERMS, NULL, NULL,
241 	    VNOVAL, &root);
242 
243 	if (error != 0 || root == NULL) {
244 	    uma_zdestroy(tmp->tm_node_pool);
245 	    uma_zdestroy(tmp->tm_dirent_pool);
246 	    delete_unrhdr(tmp->tm_ino_unr);
247 	    free(tmp, M_TMPFSMNT);
248 	    return error;
249 	}
250 	KASSERT(root->tn_id == 2, ("tmpfs root with invalid ino: %d", root->tn_id));
251 	tmp->tm_root = root;
252 
253 	MNT_ILOCK(mp);
254 	mp->mnt_flag |= MNT_LOCAL;
255 	mp->mnt_kern_flag |= MNTK_MPSAFE;
256 	MNT_IUNLOCK(mp);
257 
258 	mp->mnt_data = tmp;
259 	mp->mnt_stat.f_namemax = MAXNAMLEN;
260 	vfs_getnewfsid(mp);
261 	vfs_mountedfrom(mp, "tmpfs");
262 
263 	return 0;
264 }
265 
266 /* --------------------------------------------------------------------- */
267 
268 /* ARGSUSED2 */
269 static int
270 tmpfs_unmount(struct mount *mp, int mntflags)
271 {
272 	int error;
273 	int flags = 0;
274 	struct tmpfs_mount *tmp;
275 	struct tmpfs_node *node;
276 
277 	/* Handle forced unmounts. */
278 	if (mntflags & MNT_FORCE)
279 		flags |= FORCECLOSE;
280 
281 	/* Finalize all pending I/O. */
282 	error = vflush(mp, 0, flags, curthread);
283 	if (error != 0)
284 		return error;
285 
286 	tmp = VFS_TO_TMPFS(mp);
287 
288 	/* Free all associated data.  The loop iterates over the linked list
289 	 * we have containing all used nodes.  For each of them that is
290 	 * a directory, we free all its directory entries.  Note that after
291 	 * freeing a node, it will automatically go to the available list,
292 	 * so we will later have to iterate over it to release its items. */
293 	node = LIST_FIRST(&tmp->tm_nodes_used);
294 	while (node != NULL) {
295 		struct tmpfs_node *next;
296 
297 		if (node->tn_type == VDIR) {
298 			struct tmpfs_dirent *de;
299 
300 			de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
301 			while (de != NULL) {
302 				struct tmpfs_dirent *nde;
303 
304 				nde = TAILQ_NEXT(de, td_entries);
305 				tmpfs_free_dirent(tmp, de, FALSE);
306 				de = nde;
307 				node->tn_size -= sizeof(struct tmpfs_dirent);
308 			}
309 		}
310 
311 		next = LIST_NEXT(node, tn_entries);
312 		tmpfs_free_node(tmp, node);
313 		node = next;
314 	}
315 
316 	uma_zdestroy(tmp->tm_dirent_pool);
317 	uma_zdestroy(tmp->tm_node_pool);
318 	delete_unrhdr(tmp->tm_ino_unr);
319 
320 	mtx_destroy(&tmp->allnode_lock);
321 	MPASS(tmp->tm_pages_used == 0);
322 	MPASS(tmp->tm_nodes_inuse == 0);
323 
324 	/* Throw away the tmpfs_mount structure. */
325 	free(mp->mnt_data, M_TMPFSMNT);
326 	mp->mnt_data = NULL;
327 
328 	MNT_ILOCK(mp);
329 	mp->mnt_flag &= ~MNT_LOCAL;
330 	MNT_IUNLOCK(mp);
331 	return 0;
332 }
333 
334 /* --------------------------------------------------------------------- */
335 
336 static int
337 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
338 {
339 	int error;
340 	error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
341 
342 	if (!error)
343 		(*vpp)->v_vflag |= VV_ROOT;
344 
345 	return error;
346 }
347 
348 /* --------------------------------------------------------------------- */
349 
350 static int
351 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
352     struct vnode **vpp)
353 {
354 	boolean_t found;
355 	struct tmpfs_fid *tfhp;
356 	struct tmpfs_mount *tmp;
357 	struct tmpfs_node *node;
358 
359 	tmp = VFS_TO_TMPFS(mp);
360 
361 	tfhp = (struct tmpfs_fid *)fhp;
362 	if (tfhp->tf_len != sizeof(struct tmpfs_fid))
363 		return EINVAL;
364 
365 	if (tfhp->tf_id >= tmp->tm_nodes_max)
366 		return EINVAL;
367 
368 	found = FALSE;
369 
370 	TMPFS_LOCK(tmp);
371 	LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
372 		if (node->tn_id == tfhp->tf_id &&
373 		    node->tn_gen == tfhp->tf_gen) {
374 			found = TRUE;
375 			break;
376 		}
377 	}
378 	TMPFS_UNLOCK(tmp);
379 
380 	if (found)
381 		return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp));
382 
383 	return (EINVAL);
384 }
385 
386 /* --------------------------------------------------------------------- */
387 
388 /* ARGSUSED2 */
389 static int
390 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
391 {
392 	struct tmpfs_mount *tmp;
393 	size_t used;
394 
395 	tmp = VFS_TO_TMPFS(mp);
396 
397 	sbp->f_iosize = PAGE_SIZE;
398 	sbp->f_bsize = PAGE_SIZE;
399 
400 	used = tmpfs_pages_used(tmp);
401 	if (tmp->tm_pages_max != SIZE_MAX)
402 		 sbp->f_blocks = tmp->tm_pages_max;
403 	else
404 		 sbp->f_blocks = used + tmpfs_mem_avail();
405 	if (sbp->f_blocks <= used)
406 		sbp->f_bavail = 0;
407 	else
408 		sbp->f_bavail = sbp->f_blocks - used;
409 	sbp->f_bfree = sbp->f_bavail;
410 	used = tmp->tm_nodes_inuse;
411 	sbp->f_files = tmp->tm_nodes_max;
412 	if (sbp->f_files <= used)
413 		sbp->f_ffree = 0;
414 	else
415 		sbp->f_ffree = sbp->f_files - used;
416 	/* sbp->f_owner = tmp->tn_uid; */
417 
418 	return 0;
419 }
420 
421 /* --------------------------------------------------------------------- */
422 
423 /*
424  * tmpfs vfs operations.
425  */
426 
427 struct vfsops tmpfs_vfsops = {
428 	.vfs_mount =			tmpfs_mount,
429 	.vfs_unmount =			tmpfs_unmount,
430 	.vfs_root =			tmpfs_root,
431 	.vfs_statfs =			tmpfs_statfs,
432 	.vfs_fhtovp =			tmpfs_fhtovp,
433 };
434 VFS_SET(tmpfs_vfsops, tmpfs, 0);
435