tmpfs_vfsops.c revision 182739 
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: head/sys/fs/tmpfs/tmpfs_vfsops.c 182739 2008-09-03 18:53:48Z delphij $");
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
66MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
67MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
68
69/* --------------------------------------------------------------------- */
70
71static int	tmpfs_mount(struct mount *, struct thread *);
72static int	tmpfs_unmount(struct mount *, int, struct thread *);
73static int	tmpfs_root(struct mount *, int flags, struct vnode **,
74		    struct thread *);
75static int	tmpfs_fhtovp(struct mount *, struct fid *, struct vnode **);
76static int	tmpfs_statfs(struct mount *, struct statfs *, struct thread *);
77
78/* --------------------------------------------------------------------- */
79
80static const char *tmpfs_opts[] = {
81	"from", "size", "inodes", "uid", "gid", "mode", "export",
82	NULL
83};
84
85/* --------------------------------------------------------------------- */
86
87#define SWI_MAXMIB	3
88
89static u_int
90get_swpgtotal(void)
91{
92	struct xswdev xsd;
93	char *sname = "vm.swap_info";
94	int soid[SWI_MAXMIB], oid[2];
95	u_int unswdev, total, dmmax, nswapdev;
96	size_t mibi, len;
97
98	total = 0;
99
100	len = sizeof(dmmax);
101	if (kernel_sysctlbyname(curthread, "vm.dmmax", &dmmax, &len,
102				NULL, 0, NULL, 0) != 0)
103		return total;
104
105	len = sizeof(nswapdev);
106	if (kernel_sysctlbyname(curthread, "vm.nswapdev",
107				&nswapdev, &len,
108				NULL, 0, NULL, 0) != 0)
109		return total;
110
111	mibi = (SWI_MAXMIB - 1) * sizeof(int);
112	oid[0] = 0;
113	oid[1] = 3;
114
115	if (kernel_sysctl(curthread, oid, 2,
116			soid, &mibi, (void *)sname, strlen(sname),
117			NULL, 0) != 0)
118		return total;
119
120	mibi = (SWI_MAXMIB - 1);
121	for (unswdev = 0; unswdev < nswapdev; ++unswdev) {
122		soid[mibi] = unswdev;
123		len = sizeof(struct xswdev);
124		if (kernel_sysctl(curthread,
125				soid, mibi + 1, &xsd, &len, NULL, 0,
126				NULL, 0) != 0)
127			return total;
128		if (len == sizeof(struct xswdev))
129			total += (xsd.xsw_nblks - dmmax);
130	}
131
132	/* Not Reached */
133	return total;
134}
135
136/* --------------------------------------------------------------------- */
137static int
138tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
139{
140	struct tmpfs_node *node = (struct tmpfs_node *)mem;
141
142	node->tn_gen++;
143	node->tn_size = 0;
144	node->tn_status = 0;
145	node->tn_flags = 0;
146	node->tn_links = 0;
147	node->tn_vnode = NULL;
148	node->tn_vpstate = 0;
149
150	return (0);
151}
152
153static void
154tmpfs_node_dtor(void *mem, int size, void *arg)
155{
156	struct tmpfs_node *node = (struct tmpfs_node *)mem;
157	node->tn_type = VNON;
158}
159
160static int
161tmpfs_node_init(void *mem, int size, int flags)
162{
163	struct tmpfs_node *node = (struct tmpfs_node *)mem;
164	node->tn_id = 0;
165
166	mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
167	node->tn_gen = arc4random();
168
169	return (0);
170}
171
172static void
173tmpfs_node_fini(void *mem, int size)
174{
175	struct tmpfs_node *node = (struct tmpfs_node *)mem;
176
177	mtx_destroy(&node->tn_interlock);
178}
179
180static int
181tmpfs_mount(struct mount *mp, struct thread *td)
182{
183	struct tmpfs_mount *tmp;
184	struct tmpfs_node *root;
185	size_t pages, mem_size;
186	ino_t nodes;
187	int error;
188	/* Size counters. */
189	ino_t	nodes_max;
190	size_t	size_max;
191
192	/* Root node attributes. */
193	uid_t	root_uid;
194	gid_t	root_gid;
195	mode_t	root_mode;
196
197	struct vattr	va;
198
199	if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
200		return (EINVAL);
201
202	if (mp->mnt_flag & MNT_UPDATE) {
203		/* XXX: There is no support yet to update file system
204		 * settings.  Should be added. */
205
206		return EOPNOTSUPP;
207	}
208
209	printf("WARNING: TMPFS is considered to be a highly experimental "
210	    "feature in FreeBSD.\n");
211
212	vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
213	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
214	VOP_UNLOCK(mp->mnt_vnodecovered, 0);
215	if (error)
216		return (error);
217
218	if (mp->mnt_cred->cr_ruid != 0 ||
219	    vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
220		root_gid = va.va_gid;
221	if (mp->mnt_cred->cr_ruid != 0 ||
222	    vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
223		root_uid = va.va_uid;
224	if (mp->mnt_cred->cr_ruid != 0 ||
225	    vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
226		root_mode = va.va_mode;
227	if (vfs_scanopt(mp->mnt_optnew, "inodes", "%d", &nodes_max) != 1)
228		nodes_max = 0;
229	if (vfs_scanopt(mp->mnt_optnew, "size", "%qu", &size_max) != 1)
230		size_max = 0;
231
232	/* Do not allow mounts if we do not have enough memory to preserve
233	 * the minimum reserved pages. */
234	mem_size = cnt.v_free_count + cnt.v_inactive_count + get_swpgtotal();
235	mem_size -= mem_size > cnt.v_wire_count ? cnt.v_wire_count : mem_size;
236	if (mem_size < TMPFS_PAGES_RESERVED)
237		return ENOSPC;
238
239	/* Get the maximum number of memory pages this file system is
240	 * allowed to use, based on the maximum size the user passed in
241	 * the mount structure.  A value of zero is treated as if the
242	 * maximum available space was requested. */
243	if (size_max < PAGE_SIZE || size_max >= SIZE_MAX)
244		pages = SIZE_MAX;
245	else
246		pages = howmany(size_max, PAGE_SIZE);
247	MPASS(pages > 0);
248
249	if (nodes_max <= 3)
250		nodes = 3 + pages * PAGE_SIZE / 1024;
251	else
252		nodes = nodes_max;
253	MPASS(nodes >= 3);
254
255	/* Allocate the tmpfs mount structure and fill it. */
256	tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
257	    M_TMPFSMNT, M_WAITOK | M_ZERO);
258
259	mtx_init(&tmp->allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
260	tmp->tm_nodes_max = nodes;
261	tmp->tm_nodes_inuse = 0;
262	tmp->tm_maxfilesize = (u_int64_t)(cnt.v_page_count + get_swpgtotal()) * PAGE_SIZE;
263	LIST_INIT(&tmp->tm_nodes_used);
264
265	tmp->tm_pages_max = pages;
266	tmp->tm_pages_used = 0;
267	tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->allnode_lock);
268	tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
269	    sizeof(struct tmpfs_dirent),
270	    NULL, NULL, NULL, NULL,
271	    UMA_ALIGN_PTR, 0);
272	tmp->tm_node_pool = uma_zcreate("TMPFS node",
273	    sizeof(struct tmpfs_node),
274	    tmpfs_node_ctor, tmpfs_node_dtor,
275	    tmpfs_node_init, tmpfs_node_fini,
276	    UMA_ALIGN_PTR, 0);
277
278	/* Allocate the root node. */
279	error = tmpfs_alloc_node(tmp, VDIR, root_uid,
280	    root_gid, root_mode & ALLPERMS, NULL, NULL,
281	    VNOVAL, td, &root);
282
283	if (error != 0 || root == NULL) {
284	    uma_zdestroy(tmp->tm_node_pool);
285	    uma_zdestroy(tmp->tm_dirent_pool);
286	    delete_unrhdr(tmp->tm_ino_unr);
287	    free(tmp, M_TMPFSMNT);
288	    return error;
289	}
290	KASSERT(root->tn_id == 2, ("tmpfs root with invalid ino: %d", root->tn_id));
291	tmp->tm_root = root;
292
293	MNT_ILOCK(mp);
294	mp->mnt_flag |= MNT_LOCAL;
295	mp->mnt_kern_flag |= MNTK_MPSAFE;
296	MNT_IUNLOCK(mp);
297
298	mp->mnt_data = tmp;
299	mp->mnt_stat.f_namemax = MAXNAMLEN;
300	vfs_getnewfsid(mp);
301	vfs_mountedfrom(mp, "tmpfs");
302
303	return 0;
304}
305
306/* --------------------------------------------------------------------- */
307
308/* ARGSUSED2 */
309static int
310tmpfs_unmount(struct mount *mp, int mntflags, struct thread *l)
311{
312	int error;
313	int flags = 0;
314	struct tmpfs_mount *tmp;
315	struct tmpfs_node *node;
316
317	/* Handle forced unmounts. */
318	if (mntflags & MNT_FORCE)
319		flags |= FORCECLOSE;
320
321	/* Finalize all pending I/O. */
322	error = vflush(mp, 0, flags, l);
323	if (error != 0)
324		return error;
325
326	tmp = VFS_TO_TMPFS(mp);
327
328	/* Free all associated data.  The loop iterates over the linked list
329	 * we have containing all used nodes.  For each of them that is
330	 * a directory, we free all its directory entries.  Note that after
331	 * freeing a node, it will automatically go to the available list,
332	 * so we will later have to iterate over it to release its items. */
333	node = LIST_FIRST(&tmp->tm_nodes_used);
334	while (node != NULL) {
335		struct tmpfs_node *next;
336
337		if (node->tn_type == VDIR) {
338			struct tmpfs_dirent *de;
339
340			de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
341			while (de != NULL) {
342				struct tmpfs_dirent *nde;
343
344				nde = TAILQ_NEXT(de, td_entries);
345				tmpfs_free_dirent(tmp, de, FALSE);
346				de = nde;
347				node->tn_size -= sizeof(struct tmpfs_dirent);
348			}
349		}
350
351		next = LIST_NEXT(node, tn_entries);
352		tmpfs_free_node(tmp, node);
353		node = next;
354	}
355
356	uma_zdestroy(tmp->tm_dirent_pool);
357	uma_zdestroy(tmp->tm_node_pool);
358	delete_unrhdr(tmp->tm_ino_unr);
359
360	mtx_destroy(&tmp->allnode_lock);
361	MPASS(tmp->tm_pages_used == 0);
362	MPASS(tmp->tm_nodes_inuse == 0);
363
364	/* Throw away the tmpfs_mount structure. */
365	free(mp->mnt_data, M_TMPFSMNT);
366	mp->mnt_data = NULL;
367
368	MNT_ILOCK(mp);
369	mp->mnt_flag &= ~MNT_LOCAL;
370	MNT_IUNLOCK(mp);
371	return 0;
372}
373
374/* --------------------------------------------------------------------- */
375
376static int
377tmpfs_root(struct mount *mp, int flags, struct vnode **vpp, struct thread *td)
378{
379	int error;
380	error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp, td);
381
382	if (!error)
383		(*vpp)->v_vflag |= VV_ROOT;
384
385	return error;
386}
387
388/* --------------------------------------------------------------------- */
389
390static int
391tmpfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
392{
393	boolean_t found;
394	struct tmpfs_fid *tfhp;
395	struct tmpfs_mount *tmp;
396	struct tmpfs_node *node;
397
398	tmp = VFS_TO_TMPFS(mp);
399
400	tfhp = (struct tmpfs_fid *)fhp;
401	if (tfhp->tf_len != sizeof(struct tmpfs_fid))
402		return EINVAL;
403
404	if (tfhp->tf_id >= tmp->tm_nodes_max)
405		return EINVAL;
406
407	found = FALSE;
408
409	TMPFS_LOCK(tmp);
410	LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
411		if (node->tn_id == tfhp->tf_id &&
412		    node->tn_gen == tfhp->tf_gen) {
413			found = TRUE;
414			break;
415		}
416	}
417	TMPFS_UNLOCK(tmp);
418
419	if (found)
420		return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp, curthread));
421
422	return (EINVAL);
423}
424
425/* --------------------------------------------------------------------- */
426
427/* ARGSUSED2 */
428static int
429tmpfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *l)
430{
431	fsfilcnt_t freenodes;
432	struct tmpfs_mount *tmp;
433
434	tmp = VFS_TO_TMPFS(mp);
435
436	sbp->f_iosize = PAGE_SIZE;
437	sbp->f_bsize = PAGE_SIZE;
438
439	sbp->f_blocks = TMPFS_PAGES_MAX(tmp);
440	sbp->f_bavail = sbp->f_bfree = TMPFS_PAGES_AVAIL(tmp);
441
442	freenodes = MIN(tmp->tm_nodes_max - tmp->tm_nodes_inuse,
443	    TMPFS_PAGES_AVAIL(tmp) * PAGE_SIZE / sizeof(struct tmpfs_node));
444
445	sbp->f_files = freenodes + tmp->tm_nodes_inuse;
446	sbp->f_ffree = freenodes;
447	/* sbp->f_owner = tmp->tn_uid; */
448
449	return 0;
450}
451
452/* --------------------------------------------------------------------- */
453
454/*
455 * tmpfs vfs operations.
456 */
457
458struct vfsops tmpfs_vfsops = {
459	.vfs_mount =			tmpfs_mount,
460	.vfs_unmount =			tmpfs_unmount,
461	.vfs_root =			tmpfs_root,
462	.vfs_statfs =			tmpfs_statfs,
463	.vfs_fhtovp =			tmpfs_fhtovp,
464};
465VFS_SET(tmpfs_vfsops, tmpfs, 0);
466