1/*
2 *  fs/anon_inodes.c
3 *
4 *  Copyright (C) 2007  Davide Libenzi <davidel@xmailserver.org>
5 *
6 *  Thanks to Arnd Bergmann for code review and suggestions.
7 *  More changes for Thomas Gleixner suggestions.
8 *
9 */
10
11#include <linux/file.h>
12#include <linux/poll.h>
13#include <linux/slab.h>
14#include <linux/init.h>
15#include <linux/fs.h>
16#include <linux/mount.h>
17#include <linux/module.h>
18#include <linux/kernel.h>
19#include <linux/magic.h>
20#include <linux/anon_inodes.h>
21
22#include <asm/uaccess.h>
23
24static struct vfsmount *anon_inode_mnt __read_mostly;
25static struct inode *anon_inode_inode;
26static const struct file_operations anon_inode_fops;
27
28static int anon_inodefs_get_sb(struct file_system_type *fs_type, int flags,
29			       const char *dev_name, void *data,
30			       struct vfsmount *mnt)
31{
32	return get_sb_pseudo(fs_type, "anon_inode:", NULL, ANON_INODE_FS_MAGIC,
33			     mnt);
34}
35
36static int anon_inodefs_delete_dentry(struct dentry *dentry)
37{
38	/*
39	 * We faked vfs to believe the dentry was hashed when we created it.
40	 * Now we restore the flag so that dput() will work correctly.
41	 */
42	dentry->d_flags |= DCACHE_UNHASHED;
43	return 1;
44}
45
46static struct file_system_type anon_inode_fs_type = {
47	.name		= "anon_inodefs",
48	.get_sb		= anon_inodefs_get_sb,
49	.kill_sb	= kill_anon_super,
50};
51static struct dentry_operations anon_inodefs_dentry_operations = {
52	.d_delete	= anon_inodefs_delete_dentry,
53};
54
55/**
56 * anon_inode_getfd - creates a new file instance by hooking it up to and
57 *                    anonymous inode, and a dentry that describe the "class"
58 *                    of the file
59 *
60 * @pfd:     [out]   pointer to the file descriptor
61 * @dpinode: [out]   pointer to the inode
62 * @pfile:   [out]   pointer to the file struct
63 * @name:    [in]    name of the "class" of the new file
64 * @fops     [in]    file operations for the new file
65 * @priv     [in]    private data for the new file (will be file's private_data)
66 *
67 * Creates a new file by hooking it on a single inode. This is useful for files
68 * that do not need to have a full-fledged inode in order to operate correctly.
69 * All the files created with anon_inode_getfd() will share a single inode, by
70 * hence saving memory and avoiding code duplication for the file/inode/dentry
71 * setup.
72 */
73int anon_inode_getfd(int *pfd, struct inode **pinode, struct file **pfile,
74		     const char *name, const struct file_operations *fops,
75		     void *priv)
76{
77	struct qstr this;
78	struct dentry *dentry;
79	struct inode *inode;
80	struct file *file;
81	int error, fd;
82
83	if (IS_ERR(anon_inode_inode))
84		return -ENODEV;
85	file = get_empty_filp();
86	if (!file)
87		return -ENFILE;
88
89	inode = igrab(anon_inode_inode);
90	if (IS_ERR(inode)) {
91		error = PTR_ERR(inode);
92		goto err_put_filp;
93	}
94
95	error = get_unused_fd();
96	if (error < 0)
97		goto err_iput;
98	fd = error;
99
100	/*
101	 * Link the inode to a directory entry by creating a unique name
102	 * using the inode sequence number.
103	 */
104	error = -ENOMEM;
105	this.name = name;
106	this.len = strlen(name);
107	this.hash = 0;
108	dentry = d_alloc(anon_inode_mnt->mnt_sb->s_root, &this);
109	if (!dentry)
110		goto err_put_unused_fd;
111	dentry->d_op = &anon_inodefs_dentry_operations;
112	/* Do not publish this dentry inside the global dentry hash table */
113	dentry->d_flags &= ~DCACHE_UNHASHED;
114	d_instantiate(dentry, inode);
115
116	file->f_path.mnt = mntget(anon_inode_mnt);
117	file->f_path.dentry = dentry;
118	file->f_mapping = inode->i_mapping;
119
120	file->f_pos = 0;
121	file->f_flags = O_RDWR;
122	file->f_op = fops;
123	file->f_mode = FMODE_READ | FMODE_WRITE;
124	file->f_version = 0;
125	file->private_data = priv;
126
127	fd_install(fd, file);
128
129	*pfd = fd;
130	*pinode = inode;
131	*pfile = file;
132	return 0;
133
134err_put_unused_fd:
135	put_unused_fd(fd);
136err_iput:
137	iput(inode);
138err_put_filp:
139	put_filp(file);
140	return error;
141}
142
143/*
144 * A single inode exist for all anon_inode files. Contrary to pipes,
145 * anon_inode inodes has no per-instance data associated, so we can avoid
146 * the allocation of multiple of them.
147 */
148static struct inode *anon_inode_mkinode(void)
149{
150	struct inode *inode = new_inode(anon_inode_mnt->mnt_sb);
151
152	if (!inode)
153		return ERR_PTR(-ENOMEM);
154
155	inode->i_fop = &anon_inode_fops;
156
157	/*
158	 * Mark the inode dirty from the very beginning,
159	 * that way it will never be moved to the dirty
160	 * list because mark_inode_dirty() will think
161	 * that it already _is_ on the dirty list.
162	 */
163	inode->i_state = I_DIRTY;
164	inode->i_mode = S_IRUSR | S_IWUSR;
165	inode->i_uid = current->fsuid;
166	inode->i_gid = current->fsgid;
167	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
168	return inode;
169}
170
171static int __init anon_inode_init(void)
172{
173	int error;
174
175	error = register_filesystem(&anon_inode_fs_type);
176	if (error)
177		goto err_exit;
178	anon_inode_mnt = kern_mount(&anon_inode_fs_type);
179	if (IS_ERR(anon_inode_mnt)) {
180		error = PTR_ERR(anon_inode_mnt);
181		goto err_unregister_filesystem;
182	}
183	anon_inode_inode = anon_inode_mkinode();
184	if (IS_ERR(anon_inode_inode)) {
185		error = PTR_ERR(anon_inode_inode);
186		goto err_mntput;
187	}
188
189	return 0;
190
191err_mntput:
192	mntput(anon_inode_mnt);
193err_unregister_filesystem:
194	unregister_filesystem(&anon_inode_fs_type);
195err_exit:
196	panic(KERN_ERR "anon_inode_init() failed (%d)\n", error);
197}
198
199fs_initcall(anon_inode_init);
200