xref: /freebsd/sys/fs/tmpfs/tmpfs_vfsops.c (revision c6063d0da8e1f8d8452b0c2ee1c7ad718c1a7052)
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/proc.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/stat.h>
54 #include <sys/systm.h>
55 #include <sys/sysctl.h>
56 
57 #include <vm/vm.h>
58 #include <vm/vm_object.h>
59 #include <vm/vm_param.h>
60 
61 #include <fs/tmpfs/tmpfs.h>
62 
63 /*
64  * Default permission for root node
65  */
66 #define TMPFS_DEFAULT_ROOT_MODE	(S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
67 
68 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
69 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
70 
71 /* --------------------------------------------------------------------- */
72 
73 static int	tmpfs_mount(struct mount *);
74 static int	tmpfs_unmount(struct mount *, int);
75 static int	tmpfs_root(struct mount *, int flags, struct vnode **);
76 static int	tmpfs_fhtovp(struct mount *, struct fid *, int,
77 		    struct vnode **);
78 static int	tmpfs_statfs(struct mount *, struct statfs *);
79 
80 /* --------------------------------------------------------------------- */
81 
82 static const char *tmpfs_opts[] = {
83 	"from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export",
84 	"union", NULL
85 };
86 
87 static const char *tmpfs_updateopts[] = {
88 	"from", "export", NULL
89 };
90 
91 /* --------------------------------------------------------------------- */
92 
93 static int
94 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
95 {
96 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
97 
98 	node->tn_gen++;
99 	node->tn_size = 0;
100 	node->tn_status = 0;
101 	node->tn_flags = 0;
102 	node->tn_links = 0;
103 	node->tn_vnode = NULL;
104 	node->tn_vpstate = 0;
105 
106 	return (0);
107 }
108 
109 static void
110 tmpfs_node_dtor(void *mem, int size, void *arg)
111 {
112 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
113 	node->tn_type = VNON;
114 }
115 
116 static int
117 tmpfs_node_init(void *mem, int size, int flags)
118 {
119 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
120 	node->tn_id = 0;
121 
122 	mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
123 	node->tn_gen = arc4random();
124 
125 	return (0);
126 }
127 
128 static void
129 tmpfs_node_fini(void *mem, int size)
130 {
131 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
132 
133 	mtx_destroy(&node->tn_interlock);
134 }
135 
136 static int
137 tmpfs_mount(struct mount *mp)
138 {
139 	const size_t nodes_per_page = howmany(PAGE_SIZE,
140 	    sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
141 	struct tmpfs_mount *tmp;
142 	struct tmpfs_node *root;
143 	struct thread *td = curthread;
144 	int error;
145 	/* Size counters. */
146 	u_quad_t pages;
147 	off_t nodes_max, size_max, maxfilesize;
148 
149 	/* Root node attributes. */
150 	uid_t root_uid;
151 	gid_t root_gid;
152 	mode_t root_mode;
153 
154 	struct vattr va;
155 
156 	if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_TMPFS))
157 		return (EPERM);
158 
159 	if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
160 		return (EINVAL);
161 
162 	if (mp->mnt_flag & MNT_UPDATE) {
163 		/* Only support update mounts for certain options. */
164 		if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
165 			return (EOPNOTSUPP);
166 		if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) !=
167 		    ((struct tmpfs_mount *)mp->mnt_data)->tm_ronly)
168 			return (EOPNOTSUPP);
169 		return (0);
170 	}
171 
172 	vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
173 	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
174 	VOP_UNLOCK(mp->mnt_vnodecovered, 0);
175 	if (error)
176 		return (error);
177 
178 	if (mp->mnt_cred->cr_ruid != 0 ||
179 	    vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
180 		root_gid = va.va_gid;
181 	if (mp->mnt_cred->cr_ruid != 0 ||
182 	    vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
183 		root_uid = va.va_uid;
184 	if (mp->mnt_cred->cr_ruid != 0 ||
185 	    vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
186 		root_mode = va.va_mode;
187 	if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
188 		nodes_max = 0;
189 	if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
190 		size_max = 0;
191 	if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
192 		maxfilesize = 0;
193 
194 	/* Do not allow mounts if we do not have enough memory to preserve
195 	 * the minimum reserved pages. */
196 	if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
197 		return ENOSPC;
198 
199 	/* Get the maximum number of memory pages this file system is
200 	 * allowed to use, based on the maximum size the user passed in
201 	 * the mount structure.  A value of zero is treated as if the
202 	 * maximum available space was requested. */
203 	if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
204 	    (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
205 		pages = SIZE_MAX;
206 	else {
207 		size_max = roundup(size_max, PAGE_SIZE);
208 		pages = howmany(size_max, PAGE_SIZE);
209 	}
210 	MPASS(pages > 0);
211 
212 	if (nodes_max <= 3) {
213 		if (pages < INT_MAX / nodes_per_page)
214 			nodes_max = pages * nodes_per_page;
215 		else
216 			nodes_max = INT_MAX;
217 	}
218 	if (nodes_max > INT_MAX)
219 		nodes_max = INT_MAX;
220 	MPASS(nodes_max >= 3);
221 
222 	/* Allocate the tmpfs mount structure and fill it. */
223 	tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
224 	    M_TMPFSMNT, M_WAITOK | M_ZERO);
225 
226 	mtx_init(&tmp->allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
227 	tmp->tm_nodes_max = nodes_max;
228 	tmp->tm_nodes_inuse = 0;
229 	tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
230 	LIST_INIT(&tmp->tm_nodes_used);
231 
232 	tmp->tm_pages_max = pages;
233 	tmp->tm_pages_used = 0;
234 	tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->allnode_lock);
235 	tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
236 	    sizeof(struct tmpfs_dirent),
237 	    NULL, NULL, NULL, NULL,
238 	    UMA_ALIGN_PTR, 0);
239 	tmp->tm_node_pool = uma_zcreate("TMPFS node",
240 	    sizeof(struct tmpfs_node),
241 	    tmpfs_node_ctor, tmpfs_node_dtor,
242 	    tmpfs_node_init, tmpfs_node_fini,
243 	    UMA_ALIGN_PTR, 0);
244 	tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
245 
246 	/* Allocate the root node. */
247 	error = tmpfs_alloc_node(tmp, VDIR, root_uid,
248 	    root_gid, root_mode & ALLPERMS, NULL, NULL,
249 	    VNOVAL, &root);
250 
251 	if (error != 0 || root == NULL) {
252 	    uma_zdestroy(tmp->tm_node_pool);
253 	    uma_zdestroy(tmp->tm_dirent_pool);
254 	    delete_unrhdr(tmp->tm_ino_unr);
255 	    free(tmp, M_TMPFSMNT);
256 	    return error;
257 	}
258 	KASSERT(root->tn_id == 2,
259 	    ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
260 	tmp->tm_root = root;
261 
262 	MNT_ILOCK(mp);
263 	mp->mnt_flag |= MNT_LOCAL;
264 	MNT_IUNLOCK(mp);
265 
266 	mp->mnt_data = tmp;
267 	mp->mnt_stat.f_namemax = MAXNAMLEN;
268 	vfs_getnewfsid(mp);
269 	vfs_mountedfrom(mp, "tmpfs");
270 
271 	return 0;
272 }
273 
274 /* --------------------------------------------------------------------- */
275 
276 /* ARGSUSED2 */
277 static int
278 tmpfs_unmount(struct mount *mp, int mntflags)
279 {
280 	int error;
281 	int flags = 0;
282 	struct tmpfs_mount *tmp;
283 	struct tmpfs_node *node;
284 
285 	/* Handle forced unmounts. */
286 	if (mntflags & MNT_FORCE)
287 		flags |= FORCECLOSE;
288 
289 	/* Finalize all pending I/O. */
290 	error = vflush(mp, 0, flags, curthread);
291 	if (error != 0)
292 		return error;
293 
294 	tmp = VFS_TO_TMPFS(mp);
295 
296 	/* Free all associated data.  The loop iterates over the linked list
297 	 * we have containing all used nodes.  For each of them that is
298 	 * a directory, we free all its directory entries.  Note that after
299 	 * freeing a node, it will automatically go to the available list,
300 	 * so we will later have to iterate over it to release its items. */
301 	node = LIST_FIRST(&tmp->tm_nodes_used);
302 	while (node != NULL) {
303 		struct tmpfs_node *next;
304 
305 		if (node->tn_type == VDIR)
306 			tmpfs_dir_destroy(tmp, node);
307 
308 		next = LIST_NEXT(node, tn_entries);
309 		tmpfs_free_node(tmp, node);
310 		node = next;
311 	}
312 
313 	uma_zdestroy(tmp->tm_dirent_pool);
314 	uma_zdestroy(tmp->tm_node_pool);
315 	delete_unrhdr(tmp->tm_ino_unr);
316 
317 	mtx_destroy(&tmp->allnode_lock);
318 	MPASS(tmp->tm_pages_used == 0);
319 	MPASS(tmp->tm_nodes_inuse == 0);
320 
321 	/* Throw away the tmpfs_mount structure. */
322 	free(mp->mnt_data, M_TMPFSMNT);
323 	mp->mnt_data = NULL;
324 
325 	MNT_ILOCK(mp);
326 	mp->mnt_flag &= ~MNT_LOCAL;
327 	MNT_IUNLOCK(mp);
328 	return 0;
329 }
330 
331 /* --------------------------------------------------------------------- */
332 
333 static int
334 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
335 {
336 	int error;
337 	error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
338 
339 	if (!error)
340 		(*vpp)->v_vflag |= VV_ROOT;
341 
342 	return error;
343 }
344 
345 /* --------------------------------------------------------------------- */
346 
347 static int
348 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
349     struct vnode **vpp)
350 {
351 	boolean_t found;
352 	struct tmpfs_fid *tfhp;
353 	struct tmpfs_mount *tmp;
354 	struct tmpfs_node *node;
355 
356 	tmp = VFS_TO_TMPFS(mp);
357 
358 	tfhp = (struct tmpfs_fid *)fhp;
359 	if (tfhp->tf_len != sizeof(struct tmpfs_fid))
360 		return EINVAL;
361 
362 	if (tfhp->tf_id >= tmp->tm_nodes_max)
363 		return EINVAL;
364 
365 	found = FALSE;
366 
367 	TMPFS_LOCK(tmp);
368 	LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
369 		if (node->tn_id == tfhp->tf_id &&
370 		    node->tn_gen == tfhp->tf_gen) {
371 			found = TRUE;
372 			break;
373 		}
374 	}
375 	TMPFS_UNLOCK(tmp);
376 
377 	if (found)
378 		return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp));
379 
380 	return (EINVAL);
381 }
382 
383 /* --------------------------------------------------------------------- */
384 
385 /* ARGSUSED2 */
386 static int
387 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
388 {
389 	struct tmpfs_mount *tmp;
390 	size_t used;
391 
392 	tmp = VFS_TO_TMPFS(mp);
393 
394 	sbp->f_iosize = PAGE_SIZE;
395 	sbp->f_bsize = PAGE_SIZE;
396 
397 	used = tmpfs_pages_used(tmp);
398 	if (tmp->tm_pages_max != SIZE_MAX)
399 		 sbp->f_blocks = tmp->tm_pages_max;
400 	else
401 		 sbp->f_blocks = used + tmpfs_mem_avail();
402 	if (sbp->f_blocks <= used)
403 		sbp->f_bavail = 0;
404 	else
405 		sbp->f_bavail = sbp->f_blocks - used;
406 	sbp->f_bfree = sbp->f_bavail;
407 	used = tmp->tm_nodes_inuse;
408 	sbp->f_files = tmp->tm_nodes_max;
409 	if (sbp->f_files <= used)
410 		sbp->f_ffree = 0;
411 	else
412 		sbp->f_ffree = sbp->f_files - used;
413 	/* sbp->f_owner = tmp->tn_uid; */
414 
415 	return 0;
416 }
417 
418 /* --------------------------------------------------------------------- */
419 
420 /*
421  * tmpfs vfs operations.
422  */
423 
424 struct vfsops tmpfs_vfsops = {
425 	.vfs_mount =			tmpfs_mount,
426 	.vfs_unmount =			tmpfs_unmount,
427 	.vfs_root =			tmpfs_root,
428 	.vfs_statfs =			tmpfs_statfs,
429 	.vfs_fhtovp =			tmpfs_fhtovp,
430 };
431 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);
432