tmpfs_vfsops.c revision 1.66 
1/*	$NetBSD: tmpfs_vfsops.c,v 1.66 2016/03/12 08:45:23 joerg Exp $	*/
2
3/*
4 * Copyright (c) 2005, 2006, 2007 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
44#include <sys/cdefs.h>
45__KERNEL_RCSID(0, "$NetBSD: tmpfs_vfsops.c,v 1.66 2016/03/12 08:45:23 joerg Exp $");
46
47#include <sys/param.h>
48#include <sys/atomic.h>
49#include <sys/types.h>
50#include <sys/kmem.h>
51#include <sys/mount.h>
52#include <sys/stat.h>
53#include <sys/systm.h>
54#include <sys/vnode.h>
55#include <sys/kauth.h>
56#include <sys/module.h>
57
58#include <miscfs/genfs/genfs.h>
59#include <fs/tmpfs/tmpfs.h>
60#include <fs/tmpfs/tmpfs_args.h>
61
62MODULE(MODULE_CLASS_VFS, tmpfs, NULL);
63
64struct pool	tmpfs_dirent_pool;
65struct pool	tmpfs_node_pool;
66
67void
68tmpfs_init(void)
69{
70
71	pool_init(&tmpfs_dirent_pool, sizeof(tmpfs_dirent_t), 0, 0, 0,
72	    "tmpfs_dirent", &pool_allocator_nointr, IPL_NONE);
73	pool_init(&tmpfs_node_pool, sizeof(tmpfs_node_t), 0, 0, 0,
74	    "tmpfs_node", &pool_allocator_nointr, IPL_NONE);
75}
76
77void
78tmpfs_done(void)
79{
80
81	pool_destroy(&tmpfs_dirent_pool);
82	pool_destroy(&tmpfs_node_pool);
83}
84
85int
86tmpfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
87{
88	struct tmpfs_args *args = data;
89	tmpfs_mount_t *tmp;
90	tmpfs_node_t *root;
91	struct vattr va;
92	struct vnode *vp;
93	uint64_t memlimit;
94	ino_t nodes;
95	int error;
96	bool set_memlimit;
97	bool set_nodes;
98
99	if (args == NULL)
100		return EINVAL;
101
102	/* Validate the version. */
103	if (*data_len < sizeof(*args) ||
104	    args->ta_version != TMPFS_ARGS_VERSION)
105		return EINVAL;
106
107	/* Handle retrieval of mount point arguments. */
108	if (mp->mnt_flag & MNT_GETARGS) {
109		if (mp->mnt_data == NULL)
110			return EIO;
111		tmp = VFS_TO_TMPFS(mp);
112
113		args->ta_version = TMPFS_ARGS_VERSION;
114		args->ta_nodes_max = tmp->tm_nodes_max;
115		args->ta_size_max = tmp->tm_mem_limit;
116
117		root = tmp->tm_root;
118		args->ta_root_uid = root->tn_uid;
119		args->ta_root_gid = root->tn_gid;
120		args->ta_root_mode = root->tn_mode;
121
122		*data_len = sizeof(*args);
123		return 0;
124	}
125
126
127	/* Prohibit mounts if there is not enough memory. */
128	if (tmpfs_mem_info(true) < uvmexp.freetarg)
129		return EINVAL;
130
131	/* Check for invalid uid and gid arguments */
132	if (args->ta_root_uid == VNOVAL || args->ta_root_gid == VNOVAL)
133		return EINVAL;
134
135	/* This can never happen? */
136	if ((args->ta_root_mode & ALLPERMS) == VNOVAL)
137		return EINVAL;
138
139	/* Get the memory usage limit for this file-system. */
140	if (args->ta_size_max < PAGE_SIZE) {
141		memlimit = UINT64_MAX;
142		set_memlimit = false;
143	} else {
144		memlimit = args->ta_size_max;
145		set_memlimit = true;
146	}
147	KASSERT(memlimit > 0);
148
149	if (args->ta_nodes_max <= 3) {
150		nodes = 3 + (memlimit / 1024);
151		set_nodes = false;
152	} else {
153		nodes = args->ta_nodes_max;
154		set_nodes = true;
155	}
156	nodes = MIN(nodes, INT_MAX);
157	KASSERT(nodes >= 3);
158
159	if (mp->mnt_flag & MNT_UPDATE) {
160		tmp = VFS_TO_TMPFS(mp);
161		if (set_nodes && nodes < tmp->tm_nodes_cnt)
162			return EBUSY;
163		if (set_memlimit) {
164			if ((error = tmpfs_mntmem_set(tmp, memlimit)) != 0)
165				return error;
166		}
167		if (set_nodes)
168			tmp->tm_nodes_max = nodes;
169		root = tmp->tm_root;
170		root->tn_uid = args->ta_root_uid;
171		root->tn_gid = args->ta_root_gid;
172		root->tn_mode = args->ta_root_mode;
173		return 0;
174	}
175
176	/* Allocate the tmpfs mount structure and fill it. */
177	tmp = kmem_zalloc(sizeof(tmpfs_mount_t), KM_SLEEP);
178	if (tmp == NULL)
179		return ENOMEM;
180
181	tmp->tm_nodes_max = nodes;
182	tmp->tm_nodes_cnt = 0;
183	LIST_INIT(&tmp->tm_nodes);
184
185	mutex_init(&tmp->tm_lock, MUTEX_DEFAULT, IPL_NONE);
186	tmpfs_mntmem_init(tmp, memlimit);
187	mp->mnt_data = tmp;
188
189	/* Allocate the root node. */
190	vattr_null(&va);
191	va.va_type = VDIR;
192	va.va_mode = args->ta_root_mode & ALLPERMS;
193	va.va_uid = args->ta_root_uid;
194	va.va_gid = args->ta_root_gid;
195	error = vcache_new(mp, NULL, &va, NOCRED, &vp);
196	root = VP_TO_TMPFS_NODE(vp);
197	KASSERT(error == 0 && root != NULL);
198
199	/*
200	 * Parent of the root inode is itself.  Also, root inode has no
201	 * directory entry (i.e. is never attached), thus hold an extra
202	 * reference (link) for it.
203	 */
204	root->tn_links++;
205	root->tn_spec.tn_dir.tn_parent = root;
206	tmp->tm_root = root;
207	vrele(vp);
208
209	mp->mnt_flag |= MNT_LOCAL;
210	mp->mnt_stat.f_namemax = TMPFS_MAXNAMLEN;
211	mp->mnt_fs_bshift = PAGE_SHIFT;
212	mp->mnt_dev_bshift = DEV_BSHIFT;
213	mp->mnt_iflag |= IMNT_MPSAFE;
214	vfs_getnewfsid(mp);
215
216	error = set_statvfs_info(path, UIO_USERSPACE, "tmpfs", UIO_SYSSPACE,
217	    mp->mnt_op->vfs_name, mp, curlwp);
218	if (error) {
219		(void)tmpfs_unmount(mp, MNT_FORCE);
220	}
221	return error;
222}
223
224int
225tmpfs_start(struct mount *mp, int flags)
226{
227
228	return 0;
229}
230
231int
232tmpfs_unmount(struct mount *mp, int mntflags)
233{
234	tmpfs_mount_t *tmp = VFS_TO_TMPFS(mp);
235	tmpfs_node_t *node, *cnode;
236	int error, flags = 0;
237
238	/* Handle forced unmounts. */
239	if (mntflags & MNT_FORCE)
240		flags |= FORCECLOSE;
241
242	/* Finalize all pending I/O. */
243	error = vflush(mp, NULL, flags);
244	if (error != 0)
245		return error;
246
247	/*
248	 * First round, detach and destroy all directory entries.
249	 * Also, clear the pointers to the vnodes - they are gone.
250	 */
251	LIST_FOREACH(node, &tmp->tm_nodes, tn_entries) {
252		tmpfs_dirent_t *de;
253
254		node->tn_vnode = NULL;
255		if (node->tn_type != VDIR) {
256			continue;
257		}
258		while ((de = TAILQ_FIRST(&node->tn_spec.tn_dir.tn_dir)) != NULL) {
259			cnode = de->td_node;
260			if (cnode && cnode != TMPFS_NODE_WHITEOUT) {
261				cnode->tn_vnode = NULL;
262			}
263			tmpfs_dir_detach(node, de);
264			tmpfs_free_dirent(tmp, de);
265		}
266		/* Extra virtual entry (itself for the root). */
267		node->tn_links--;
268	}
269
270	/* Release the reference on root (diagnostic). */
271	node = tmp->tm_root;
272	node->tn_links--;
273
274	/* Second round, destroy all inodes. */
275	while ((node = LIST_FIRST(&tmp->tm_nodes)) != NULL) {
276		tmpfs_free_node(tmp, node);
277	}
278
279	/* Throw away the tmpfs_mount structure. */
280	tmpfs_mntmem_destroy(tmp);
281	mutex_destroy(&tmp->tm_lock);
282	kmem_free(tmp, sizeof(*tmp));
283	mp->mnt_data = NULL;
284
285	return 0;
286}
287
288int
289tmpfs_root(struct mount *mp, vnode_t **vpp)
290{
291	tmpfs_node_t *node = VFS_TO_TMPFS(mp)->tm_root;
292	int error;
293
294	error = vcache_get(mp, &node, sizeof(node), vpp);
295	if (error)
296		return error;
297	error = vn_lock(*vpp, LK_EXCLUSIVE);
298	if (error) {
299		vrele(*vpp);
300		*vpp = NULL;
301		return error;
302	}
303
304	return 0;
305}
306
307int
308tmpfs_vget(struct mount *mp, ino_t ino, vnode_t **vpp)
309{
310
311	return EOPNOTSUPP;
312}
313
314int
315tmpfs_fhtovp(struct mount *mp, struct fid *fhp, vnode_t **vpp)
316{
317	tmpfs_mount_t *tmp = VFS_TO_TMPFS(mp);
318	tmpfs_node_t *node;
319	tmpfs_fid_t tfh;
320	int error;
321
322	if (fhp->fid_len != sizeof(tmpfs_fid_t)) {
323		return EINVAL;
324	}
325	memcpy(&tfh, fhp, sizeof(tmpfs_fid_t));
326
327	mutex_enter(&tmp->tm_lock);
328	LIST_FOREACH(node, &tmp->tm_nodes, tn_entries) {
329		if (node->tn_id == tfh.tf_id) {
330			/* Prevent this node from disappearing. */
331			atomic_inc_32(&node->tn_holdcount);
332			break;
333		}
334	}
335	mutex_exit(&tmp->tm_lock);
336	if (node == NULL)
337		return ESTALE;
338
339	error = vcache_get(mp, &node, sizeof(node), vpp);
340	/* If this node has been reclaimed free it now. */
341	if (atomic_dec_32_nv(&node->tn_holdcount) == TMPFS_NODE_RECLAIMED) {
342		KASSERT(error != 0);
343		tmpfs_free_node(tmp, node);
344	}
345	if (error)
346		return (error == ENOENT ? ESTALE : error);
347	error = vn_lock(*vpp, LK_EXCLUSIVE);
348	if (error) {
349		vrele(*vpp);
350		*vpp = NULL;
351		return error;
352	}
353	if (TMPFS_NODE_GEN(node) != tfh.tf_gen) {
354		vput(*vpp);
355		*vpp = NULL;
356		return ESTALE;
357	}
358
359	return 0;
360}
361
362int
363tmpfs_vptofh(vnode_t *vp, struct fid *fhp, size_t *fh_size)
364{
365	tmpfs_fid_t tfh;
366	tmpfs_node_t *node;
367
368	if (*fh_size < sizeof(tmpfs_fid_t)) {
369		*fh_size = sizeof(tmpfs_fid_t);
370		return E2BIG;
371	}
372	*fh_size = sizeof(tmpfs_fid_t);
373	node = VP_TO_TMPFS_NODE(vp);
374
375	memset(&tfh, 0, sizeof(tfh));
376	tfh.tf_len = sizeof(tmpfs_fid_t);
377	tfh.tf_gen = TMPFS_NODE_GEN(node);
378	tfh.tf_id = node->tn_id;
379	memcpy(fhp, &tfh, sizeof(tfh));
380
381	return 0;
382}
383
384int
385tmpfs_statvfs(struct mount *mp, struct statvfs *sbp)
386{
387	tmpfs_mount_t *tmp;
388	fsfilcnt_t freenodes;
389	size_t avail;
390
391	tmp = VFS_TO_TMPFS(mp);
392
393	sbp->f_iosize = sbp->f_frsize = sbp->f_bsize = PAGE_SIZE;
394
395	mutex_enter(&tmp->tm_acc_lock);
396	avail =  tmpfs_pages_avail(tmp);
397	sbp->f_blocks = (tmpfs_bytes_max(tmp) >> PAGE_SHIFT);
398	sbp->f_bavail = sbp->f_bfree = avail;
399	sbp->f_bresvd = 0;
400
401	freenodes = MIN(tmp->tm_nodes_max - tmp->tm_nodes_cnt,
402	    avail * PAGE_SIZE / sizeof(tmpfs_node_t));
403
404	sbp->f_files = tmp->tm_nodes_cnt + freenodes;
405	sbp->f_favail = sbp->f_ffree = freenodes;
406	sbp->f_fresvd = 0;
407	mutex_exit(&tmp->tm_acc_lock);
408
409	copy_statvfs_info(sbp, mp);
410
411	return 0;
412}
413
414int
415tmpfs_sync(struct mount *mp, int waitfor, kauth_cred_t uc)
416{
417
418	return 0;
419}
420
421int
422tmpfs_snapshot(struct mount *mp, vnode_t *vp, struct timespec *ctime)
423{
424
425	return EOPNOTSUPP;
426}
427
428/*
429 * tmpfs vfs operations.
430 */
431
432extern const struct vnodeopv_desc tmpfs_fifoop_opv_desc;
433extern const struct vnodeopv_desc tmpfs_specop_opv_desc;
434extern const struct vnodeopv_desc tmpfs_vnodeop_opv_desc;
435
436const struct vnodeopv_desc * const tmpfs_vnodeopv_descs[] = {
437	&tmpfs_fifoop_opv_desc,
438	&tmpfs_specop_opv_desc,
439	&tmpfs_vnodeop_opv_desc,
440	NULL,
441};
442
443struct vfsops tmpfs_vfsops = {
444	.vfs_name = MOUNT_TMPFS,
445	.vfs_min_mount_data = sizeof (struct tmpfs_args),
446	.vfs_mount = tmpfs_mount,
447	.vfs_start = tmpfs_start,
448	.vfs_unmount = tmpfs_unmount,
449	.vfs_root = tmpfs_root,
450	.vfs_quotactl = (void *)eopnotsupp,
451	.vfs_statvfs = tmpfs_statvfs,
452	.vfs_sync = tmpfs_sync,
453	.vfs_vget = tmpfs_vget,
454	.vfs_loadvnode = tmpfs_loadvnode,
455	.vfs_newvnode = tmpfs_newvnode,
456	.vfs_fhtovp = tmpfs_fhtovp,
457	.vfs_vptofh = tmpfs_vptofh,
458	.vfs_init = tmpfs_init,
459	.vfs_done = tmpfs_done,
460	.vfs_snapshot = tmpfs_snapshot,
461	.vfs_extattrctl = vfs_stdextattrctl,
462	.vfs_suspendctl = (void *)eopnotsupp,
463	.vfs_renamelock_enter = genfs_renamelock_enter,
464	.vfs_renamelock_exit = genfs_renamelock_exit,
465	.vfs_fsync = (void *)eopnotsupp,
466	.vfs_opv_descs = tmpfs_vnodeopv_descs
467};
468
469static int
470tmpfs_modcmd(modcmd_t cmd, void *arg)
471{
472
473	switch (cmd) {
474	case MODULE_CMD_INIT:
475		return vfs_attach(&tmpfs_vfsops);
476	case MODULE_CMD_FINI:
477		return vfs_detach(&tmpfs_vfsops);
478	default:
479		return ENOTTY;
480	}
481}
482