xref: /freebsd/sys/fs/tmpfs/tmpfs_vfsops.c (revision 0677dfd1c4dadb62482e2c72fa4c6720902128a4)
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 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 static const char *tmpfs_opts[] = {
79 	"from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export",
80 	"union", NULL
81 };
82 
83 static const char *tmpfs_updateopts[] = {
84 	"from", "export", NULL
85 };
86 
87 static int
88 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
89 {
90 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
91 
92 	node->tn_gen++;
93 	node->tn_size = 0;
94 	node->tn_status = 0;
95 	node->tn_flags = 0;
96 	node->tn_links = 0;
97 	node->tn_vnode = NULL;
98 	node->tn_vpstate = 0;
99 
100 	return (0);
101 }
102 
103 static void
104 tmpfs_node_dtor(void *mem, int size, void *arg)
105 {
106 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
107 	node->tn_type = VNON;
108 }
109 
110 static int
111 tmpfs_node_init(void *mem, int size, int flags)
112 {
113 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
114 	node->tn_id = 0;
115 
116 	mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
117 	node->tn_gen = arc4random();
118 
119 	return (0);
120 }
121 
122 static void
123 tmpfs_node_fini(void *mem, int size)
124 {
125 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
126 
127 	mtx_destroy(&node->tn_interlock);
128 }
129 
130 static int
131 tmpfs_mount(struct mount *mp)
132 {
133 	const size_t nodes_per_page = howmany(PAGE_SIZE,
134 	    sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
135 	struct tmpfs_mount *tmp;
136 	struct tmpfs_node *root;
137 	struct thread *td = curthread;
138 	int error;
139 	/* Size counters. */
140 	u_quad_t pages;
141 	off_t nodes_max, size_max, maxfilesize;
142 
143 	/* Root node attributes. */
144 	uid_t root_uid;
145 	gid_t root_gid;
146 	mode_t root_mode;
147 
148 	struct vattr va;
149 
150 	if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_TMPFS))
151 		return (EPERM);
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 == 0 || size_max > OFF_MAX - PAGE_SIZE ||
198 	    (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
199 		pages = SIZE_MAX;
200 	else {
201 		size_max = roundup(size_max, PAGE_SIZE);
202 		pages = howmany(size_max, PAGE_SIZE);
203 	}
204 	MPASS(pages > 0);
205 
206 	if (nodes_max <= 3) {
207 		if (pages < INT_MAX / nodes_per_page)
208 			nodes_max = pages * nodes_per_page;
209 		else
210 			nodes_max = INT_MAX;
211 	}
212 	if (nodes_max > INT_MAX)
213 		nodes_max = INT_MAX;
214 	MPASS(nodes_max >= 3);
215 
216 	/* Allocate the tmpfs mount structure and fill it. */
217 	tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
218 	    M_TMPFSMNT, M_WAITOK | M_ZERO);
219 
220 	mtx_init(&tmp->allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
221 	tmp->tm_nodes_max = nodes_max;
222 	tmp->tm_nodes_inuse = 0;
223 	tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
224 	LIST_INIT(&tmp->tm_nodes_used);
225 
226 	tmp->tm_pages_max = pages;
227 	tmp->tm_pages_used = 0;
228 	tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->allnode_lock);
229 	tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
230 	    sizeof(struct tmpfs_dirent),
231 	    NULL, NULL, NULL, NULL,
232 	    UMA_ALIGN_PTR, 0);
233 	tmp->tm_node_pool = uma_zcreate("TMPFS node",
234 	    sizeof(struct tmpfs_node),
235 	    tmpfs_node_ctor, tmpfs_node_dtor,
236 	    tmpfs_node_init, tmpfs_node_fini,
237 	    UMA_ALIGN_PTR, 0);
238 	tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
239 
240 	/* Allocate the root node. */
241 	error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid,
242 	    root_gid, root_mode & ALLPERMS, NULL, NULL,
243 	    VNOVAL, &root);
244 
245 	if (error != 0 || root == NULL) {
246 	    uma_zdestroy(tmp->tm_node_pool);
247 	    uma_zdestroy(tmp->tm_dirent_pool);
248 	    delete_unrhdr(tmp->tm_ino_unr);
249 	    free(tmp, M_TMPFSMNT);
250 	    return error;
251 	}
252 	KASSERT(root->tn_id == 2,
253 	    ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
254 	tmp->tm_root = root;
255 
256 	MNT_ILOCK(mp);
257 	mp->mnt_flag |= MNT_LOCAL;
258 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_SUSPENDABLE;
259 	MNT_IUNLOCK(mp);
260 
261 	mp->mnt_data = tmp;
262 	mp->mnt_stat.f_namemax = MAXNAMLEN;
263 	vfs_getnewfsid(mp);
264 	vfs_mountedfrom(mp, "tmpfs");
265 
266 	return 0;
267 }
268 
269 /* ARGSUSED2 */
270 static int
271 tmpfs_unmount(struct mount *mp, int mntflags)
272 {
273 	struct tmpfs_mount *tmp;
274 	struct tmpfs_node *node;
275 	int error, flags;
276 
277 	flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0;
278 	tmp = VFS_TO_TMPFS(mp);
279 
280 	/* Stop writers */
281 	error = vfs_write_suspend_umnt(mp);
282 	if (error != 0)
283 		return (error);
284 	/*
285 	 * At this point, nodes cannot be destroyed by any other
286 	 * thread because write suspension is started.
287 	 */
288 
289 	for (;;) {
290 		error = vflush(mp, 0, flags, curthread);
291 		if (error != 0) {
292 			vfs_write_resume(mp, VR_START_WRITE);
293 			return (error);
294 		}
295 		MNT_ILOCK(mp);
296 		if (mp->mnt_nvnodelistsize == 0) {
297 			MNT_IUNLOCK(mp);
298 			break;
299 		}
300 		MNT_IUNLOCK(mp);
301 		if ((mntflags & MNT_FORCE) == 0) {
302 			vfs_write_resume(mp, VR_START_WRITE);
303 			return (EBUSY);
304 		}
305 	}
306 
307 	TMPFS_LOCK(tmp);
308 	while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
309 		TMPFS_UNLOCK(tmp);
310 		if (node->tn_type == VDIR)
311 			tmpfs_dir_destroy(tmp, node);
312 		tmpfs_free_node(tmp, node);
313 		TMPFS_LOCK(tmp);
314 	}
315 	TMPFS_UNLOCK(tmp);
316 
317 	uma_zdestroy(tmp->tm_dirent_pool);
318 	uma_zdestroy(tmp->tm_node_pool);
319 	delete_unrhdr(tmp->tm_ino_unr);
320 
321 	mtx_destroy(&tmp->allnode_lock);
322 	MPASS(tmp->tm_pages_used == 0);
323 	MPASS(tmp->tm_nodes_inuse == 0);
324 
325 	/* Throw away the tmpfs_mount structure. */
326 	free(mp->mnt_data, M_TMPFSMNT);
327 	mp->mnt_data = NULL;
328 	vfs_write_resume(mp, VR_START_WRITE);
329 
330 	MNT_ILOCK(mp);
331 	mp->mnt_flag &= ~MNT_LOCAL;
332 	MNT_IUNLOCK(mp);
333 
334 	return (0);
335 }
336 
337 static int
338 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
339 {
340 	int error;
341 	error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
342 
343 	if (!error)
344 		(*vpp)->v_vflag |= VV_ROOT;
345 
346 	return error;
347 }
348 
349 static int
350 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
351     struct vnode **vpp)
352 {
353 	boolean_t found;
354 	struct tmpfs_fid *tfhp;
355 	struct tmpfs_mount *tmp;
356 	struct tmpfs_node *node;
357 
358 	tmp = VFS_TO_TMPFS(mp);
359 
360 	tfhp = (struct tmpfs_fid *)fhp;
361 	if (tfhp->tf_len != sizeof(struct tmpfs_fid))
362 		return EINVAL;
363 
364 	if (tfhp->tf_id >= tmp->tm_nodes_max)
365 		return EINVAL;
366 
367 	found = FALSE;
368 
369 	TMPFS_LOCK(tmp);
370 	LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
371 		if (node->tn_id == tfhp->tf_id &&
372 		    node->tn_gen == tfhp->tf_gen) {
373 			found = TRUE;
374 			break;
375 		}
376 	}
377 	TMPFS_UNLOCK(tmp);
378 
379 	if (found)
380 		return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp));
381 
382 	return (EINVAL);
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 static int
419 tmpfs_sync(struct mount *mp, int waitfor)
420 {
421 
422 	if (waitfor == MNT_SUSPEND) {
423 		MNT_ILOCK(mp);
424 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
425 		MNT_IUNLOCK(mp);
426 	}
427 	return (0);
428 }
429 
430 /*
431  * tmpfs vfs operations.
432  */
433 
434 struct vfsops tmpfs_vfsops = {
435 	.vfs_mount =			tmpfs_mount,
436 	.vfs_unmount =			tmpfs_unmount,
437 	.vfs_root =			tmpfs_root,
438 	.vfs_statfs =			tmpfs_statfs,
439 	.vfs_fhtovp =			tmpfs_fhtovp,
440 	.vfs_sync =			tmpfs_sync,
441 };
442 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);
443