tmpfs_vfsops.c revision 182371 
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 * 3. All advertising materials mentioning features or use of this software
20 *    must display the following acknowledgement:
21 *        This product includes software developed by the NetBSD
22 *        Foundation, Inc. and its contributors.
23 * 4. Neither the name of The NetBSD Foundation nor the names of its
24 *    contributors may be used to endorse or promote products derived
25 *    from this software without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37 * POSSIBILITY OF SUCH DAMAGE.
38 */
39
40/*
41 * Efficient memory file system.
42 *
43 * tmpfs is a file system that uses NetBSD's virtual memory sub-system
44 * (the well-known UVM) to store file data and metadata in an efficient
45 * way.  This means that it does not follow the structure of an on-disk
46 * file system because it simply does not need to.  Instead, it uses
47 * memory-specific data structures and algorithms to automatically
48 * allocate and release resources.
49 */
50#include <sys/cdefs.h>
51__FBSDID("$FreeBSD: head/sys/fs/tmpfs/tmpfs_vfsops.c 182371 2008-08-28 15:23:18Z attilio $");
52
53#include <sys/param.h>
54#include <sys/limits.h>
55#include <sys/lock.h>
56#include <sys/mutex.h>
57#include <sys/kernel.h>
58#include <sys/stat.h>
59#include <sys/systm.h>
60#include <sys/sysctl.h>
61
62#include <vm/vm.h>
63#include <vm/vm_object.h>
64#include <vm/vm_param.h>
65
66#include <fs/tmpfs/tmpfs.h>
67
68/*
69 * Default permission for root node
70 */
71#define TMPFS_DEFAULT_ROOT_MODE	(S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
72
73MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
74MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
75
76/* --------------------------------------------------------------------- */
77
78static int	tmpfs_mount(struct mount *, struct thread *);
79static int	tmpfs_unmount(struct mount *, int, struct thread *);
80static int	tmpfs_root(struct mount *, int flags, struct vnode **,
81		    struct thread *);
82static int	tmpfs_fhtovp(struct mount *, struct fid *, struct vnode **);
83static int	tmpfs_statfs(struct mount *, struct statfs *, struct thread *);
84
85/* --------------------------------------------------------------------- */
86
87static const char *tmpfs_opts[] = {
88	"from", "size", "inodes", "uid", "gid", "mode", "export",
89	NULL
90};
91
92/* --------------------------------------------------------------------- */
93
94#define SWI_MAXMIB	3
95
96static u_int
97get_swpgtotal(void)
98{
99	struct xswdev xsd;
100	char *sname = "vm.swap_info";
101	int soid[SWI_MAXMIB], oid[2];
102	u_int unswdev, total, dmmax, nswapdev;
103	size_t mibi, len;
104
105	total = 0;
106
107	len = sizeof(dmmax);
108	if (kernel_sysctlbyname(curthread, "vm.dmmax", &dmmax, &len,
109				NULL, 0, NULL, 0) != 0)
110		return total;
111
112	len = sizeof(nswapdev);
113	if (kernel_sysctlbyname(curthread, "vm.nswapdev",
114				&nswapdev, &len,
115				NULL, 0, NULL, 0) != 0)
116		return total;
117
118	mibi = (SWI_MAXMIB - 1) * sizeof(int);
119	oid[0] = 0;
120	oid[1] = 3;
121
122	if (kernel_sysctl(curthread, oid, 2,
123			soid, &mibi, (void *)sname, strlen(sname),
124			NULL, 0) != 0)
125		return total;
126
127	mibi = (SWI_MAXMIB - 1);
128	for (unswdev = 0; unswdev < nswapdev; ++unswdev) {
129		soid[mibi] = unswdev;
130		len = sizeof(struct xswdev);
131		if (kernel_sysctl(curthread,
132				soid, mibi + 1, &xsd, &len, NULL, 0,
133				NULL, 0) != 0)
134			return total;
135		if (len == sizeof(struct xswdev))
136			total += (xsd.xsw_nblks - dmmax);
137	}
138
139	/* Not Reached */
140	return total;
141}
142
143/* --------------------------------------------------------------------- */
144static int
145tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
146{
147	struct tmpfs_node *node = (struct tmpfs_node *)mem;
148
149	node->tn_gen++;
150	node->tn_size = 0;
151	node->tn_status = 0;
152	node->tn_flags = 0;
153	node->tn_links = 0;
154	node->tn_vnode = NULL;
155	node->tn_vpstate = 0;
156
157	return (0);
158}
159
160static void
161tmpfs_node_dtor(void *mem, int size, void *arg)
162{
163	struct tmpfs_node *node = (struct tmpfs_node *)mem;
164	node->tn_type = VNON;
165}
166
167static int
168tmpfs_node_init(void *mem, int size, int flags)
169{
170	struct tmpfs_node *node = (struct tmpfs_node *)mem;
171	node->tn_id = 0;
172
173	mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
174	node->tn_gen = arc4random();
175
176	return (0);
177}
178
179static void
180tmpfs_node_fini(void *mem, int size)
181{
182	struct tmpfs_node *node = (struct tmpfs_node *)mem;
183
184	mtx_destroy(&node->tn_interlock);
185}
186
187static int
188tmpfs_mount(struct mount *mp, struct thread *td)
189{
190	struct tmpfs_mount *tmp;
191	struct tmpfs_node *root;
192	size_t pages, mem_size;
193	ino_t nodes;
194	int error;
195	/* Size counters. */
196	ino_t	nodes_max;
197	size_t	size_max;
198
199	/* Root node attributes. */
200	uid_t	root_uid;
201	gid_t	root_gid;
202	mode_t	root_mode;
203
204	struct vattr	va;
205
206	if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
207		return (EINVAL);
208
209	if (mp->mnt_flag & MNT_UPDATE) {
210		/* XXX: There is no support yet to update file system
211		 * settings.  Should be added. */
212
213		return EOPNOTSUPP;
214	}
215
216	printf("WARNING: TMPFS is considered to be a highly experimental "
217	    "feature in FreeBSD.\n");
218
219	vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
220	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
221	VOP_UNLOCK(mp->mnt_vnodecovered, 0);
222	if (error)
223		return (error);
224
225	if (mp->mnt_cred->cr_ruid != 0 ||
226	    vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
227		root_gid = va.va_gid;
228	if (mp->mnt_cred->cr_ruid != 0 ||
229	    vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
230		root_uid = va.va_uid;
231	if (mp->mnt_cred->cr_ruid != 0 ||
232	    vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
233		root_mode = va.va_mode;
234	if (vfs_scanopt(mp->mnt_optnew, "inodes", "%d", &nodes_max) != 1)
235		nodes_max = 0;
236	if (vfs_scanopt(mp->mnt_optnew, "size", "%qu", &size_max) != 1)
237		size_max = 0;
238
239	/* Do not allow mounts if we do not have enough memory to preserve
240	 * the minimum reserved pages. */
241	mem_size = cnt.v_free_count + cnt.v_inactive_count + get_swpgtotal();
242	mem_size -= mem_size > cnt.v_wire_count ? cnt.v_wire_count : mem_size;
243	if (mem_size < TMPFS_PAGES_RESERVED)
244		return ENOSPC;
245
246	/* Get the maximum number of memory pages this file system is
247	 * allowed to use, based on the maximum size the user passed in
248	 * the mount structure.  A value of zero is treated as if the
249	 * maximum available space was requested. */
250	if (size_max < PAGE_SIZE || size_max >= SIZE_MAX)
251		pages = SIZE_MAX;
252	else
253		pages = howmany(size_max, PAGE_SIZE);
254	MPASS(pages > 0);
255
256	if (nodes_max <= 3)
257		nodes = 3 + pages * PAGE_SIZE / 1024;
258	else
259		nodes = nodes_max;
260	MPASS(nodes >= 3);
261
262	/* Allocate the tmpfs mount structure and fill it. */
263	tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
264	    M_TMPFSMNT, M_WAITOK | M_ZERO);
265
266	mtx_init(&tmp->allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
267	tmp->tm_nodes_max = nodes;
268	tmp->tm_nodes_inuse = 0;
269	tmp->tm_maxfilesize = (u_int64_t)(cnt.v_page_count + get_swpgtotal()) * PAGE_SIZE;
270	LIST_INIT(&tmp->tm_nodes_used);
271
272	tmp->tm_pages_max = pages;
273	tmp->tm_pages_used = 0;
274	tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->allnode_lock);
275	tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
276	    sizeof(struct tmpfs_dirent),
277	    NULL, NULL, NULL, NULL,
278	    UMA_ALIGN_PTR, 0);
279	tmp->tm_node_pool = uma_zcreate("TMPFS node",
280	    sizeof(struct tmpfs_node),
281	    tmpfs_node_ctor, tmpfs_node_dtor,
282	    tmpfs_node_init, tmpfs_node_fini,
283	    UMA_ALIGN_PTR, 0);
284
285	/* Allocate the root node. */
286	error = tmpfs_alloc_node(tmp, VDIR, root_uid,
287	    root_gid, root_mode & ALLPERMS, NULL, NULL,
288	    VNOVAL, td, &root);
289
290	if (error != 0 || root == NULL) {
291	    uma_zdestroy(tmp->tm_node_pool);
292	    uma_zdestroy(tmp->tm_dirent_pool);
293	    delete_unrhdr(tmp->tm_ino_unr);
294	    free(tmp, M_TMPFSMNT);
295	    return error;
296	}
297	KASSERT(root->tn_id == 2, ("tmpfs root with invalid ino: %d", root->tn_id));
298	tmp->tm_root = root;
299
300	MNT_ILOCK(mp);
301	mp->mnt_flag |= MNT_LOCAL;
302	mp->mnt_kern_flag |= MNTK_MPSAFE;
303	MNT_IUNLOCK(mp);
304
305	mp->mnt_data = tmp;
306	mp->mnt_stat.f_namemax = MAXNAMLEN;
307	vfs_getnewfsid(mp);
308	vfs_mountedfrom(mp, "tmpfs");
309
310	return 0;
311}
312
313/* --------------------------------------------------------------------- */
314
315/* ARGSUSED2 */
316static int
317tmpfs_unmount(struct mount *mp, int mntflags, struct thread *l)
318{
319	int error;
320	int flags = 0;
321	struct tmpfs_mount *tmp;
322	struct tmpfs_node *node;
323
324	/* Handle forced unmounts. */
325	if (mntflags & MNT_FORCE)
326		flags |= FORCECLOSE;
327
328	/* Finalize all pending I/O. */
329	error = vflush(mp, 0, flags, l);
330	if (error != 0)
331		return error;
332
333	tmp = VFS_TO_TMPFS(mp);
334
335	/* Free all associated data.  The loop iterates over the linked list
336	 * we have containing all used nodes.  For each of them that is
337	 * a directory, we free all its directory entries.  Note that after
338	 * freeing a node, it will automatically go to the available list,
339	 * so we will later have to iterate over it to release its items. */
340	node = LIST_FIRST(&tmp->tm_nodes_used);
341	while (node != NULL) {
342		struct tmpfs_node *next;
343
344		if (node->tn_type == VDIR) {
345			struct tmpfs_dirent *de;
346
347			de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
348			while (de != NULL) {
349				struct tmpfs_dirent *nde;
350
351				nde = TAILQ_NEXT(de, td_entries);
352				tmpfs_free_dirent(tmp, de, FALSE);
353				de = nde;
354				node->tn_size -= sizeof(struct tmpfs_dirent);
355			}
356		}
357
358		next = LIST_NEXT(node, tn_entries);
359		tmpfs_free_node(tmp, node);
360		node = next;
361	}
362
363	uma_zdestroy(tmp->tm_dirent_pool);
364	uma_zdestroy(tmp->tm_node_pool);
365	delete_unrhdr(tmp->tm_ino_unr);
366
367	mtx_destroy(&tmp->allnode_lock);
368	MPASS(tmp->tm_pages_used == 0);
369	MPASS(tmp->tm_nodes_inuse == 0);
370
371	/* Throw away the tmpfs_mount structure. */
372	free(mp->mnt_data, M_TMPFSMNT);
373	mp->mnt_data = NULL;
374
375	MNT_ILOCK(mp);
376	mp->mnt_flag &= ~MNT_LOCAL;
377	MNT_IUNLOCK(mp);
378	return 0;
379}
380
381/* --------------------------------------------------------------------- */
382
383static int
384tmpfs_root(struct mount *mp, int flags, struct vnode **vpp, struct thread *td)
385{
386	int error;
387	error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp, td);
388
389	if (!error)
390		(*vpp)->v_vflag |= VV_ROOT;
391
392	return error;
393}
394
395/* --------------------------------------------------------------------- */
396
397static int
398tmpfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
399{
400	boolean_t found;
401	struct tmpfs_fid *tfhp;
402	struct tmpfs_mount *tmp;
403	struct tmpfs_node *node;
404
405	tmp = VFS_TO_TMPFS(mp);
406
407	tfhp = (struct tmpfs_fid *)fhp;
408	if (tfhp->tf_len != sizeof(struct tmpfs_fid))
409		return EINVAL;
410
411	if (tfhp->tf_id >= tmp->tm_nodes_max)
412		return EINVAL;
413
414	found = FALSE;
415
416	TMPFS_LOCK(tmp);
417	LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
418		if (node->tn_id == tfhp->tf_id &&
419		    node->tn_gen == tfhp->tf_gen) {
420			found = TRUE;
421			break;
422		}
423	}
424	TMPFS_UNLOCK(tmp);
425
426	if (found)
427		return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp, curthread));
428
429	return (EINVAL);
430}
431
432/* --------------------------------------------------------------------- */
433
434/* ARGSUSED2 */
435static int
436tmpfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *l)
437{
438	fsfilcnt_t freenodes;
439	struct tmpfs_mount *tmp;
440
441	tmp = VFS_TO_TMPFS(mp);
442
443	sbp->f_iosize = PAGE_SIZE;
444	sbp->f_bsize = PAGE_SIZE;
445
446	sbp->f_blocks = TMPFS_PAGES_MAX(tmp);
447	sbp->f_bavail = sbp->f_bfree = TMPFS_PAGES_AVAIL(tmp);
448
449	freenodes = MIN(tmp->tm_nodes_max - tmp->tm_nodes_inuse,
450	    TMPFS_PAGES_AVAIL(tmp) * PAGE_SIZE / sizeof(struct tmpfs_node));
451
452	sbp->f_files = freenodes + tmp->tm_nodes_inuse;
453	sbp->f_ffree = freenodes;
454	/* sbp->f_owner = tmp->tn_uid; */
455
456	return 0;
457}
458
459/* --------------------------------------------------------------------- */
460
461/*
462 * tmpfs vfs operations.
463 */
464
465struct vfsops tmpfs_vfsops = {
466	.vfs_mount =			tmpfs_mount,
467	.vfs_unmount =			tmpfs_unmount,
468	.vfs_root =			tmpfs_root,
469	.vfs_statfs =			tmpfs_statfs,
470	.vfs_fhtovp =			tmpfs_fhtovp,
471};
472VFS_SET(tmpfs_vfsops, tmpfs, 0);
473