1/*
2 *  linux/fs/proc/root.c
3 *
4 *  Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 *  proc root directory handling functions
7 */
8
9#include <asm/uaccess.h>
10
11#include <linux/errno.h>
12#include <linux/time.h>
13#include <linux/proc_fs.h>
14#include <linux/stat.h>
15#include <linux/init.h>
16#include <linux/sched.h>
17#include <linux/module.h>
18#include <linux/bitops.h>
19#include <linux/user_namespace.h>
20#include <linux/mount.h>
21#include <linux/pid_namespace.h>
22#include <linux/parser.h>
23
24#include "internal.h"
25
26static int proc_test_super(struct super_block *sb, void *data)
27{
28	return sb->s_fs_info == data;
29}
30
31static int proc_set_super(struct super_block *sb, void *data)
32{
33	int err = set_anon_super(sb, NULL);
34	if (!err) {
35		struct pid_namespace *ns = (struct pid_namespace *)data;
36		sb->s_fs_info = get_pid_ns(ns);
37	}
38	return err;
39}
40
41enum {
42	Opt_gid, Opt_hidepid, Opt_err,
43};
44
45static const match_table_t tokens = {
46	{Opt_hidepid, "hidepid=%u"},
47	{Opt_gid, "gid=%u"},
48	{Opt_err, NULL},
49};
50
51static int proc_parse_options(char *options, struct pid_namespace *pid)
52{
53	char *p;
54	substring_t args[MAX_OPT_ARGS];
55	int option;
56
57	if (!options)
58		return 1;
59
60	while ((p = strsep(&options, ",")) != NULL) {
61		int token;
62		if (!*p)
63			continue;
64
65		args[0].to = args[0].from = NULL;
66		token = match_token(p, tokens, args);
67		switch (token) {
68		case Opt_gid:
69			if (match_int(&args[0], &option))
70				return 0;
71			pid->pid_gid = make_kgid(current_user_ns(), option);
72			break;
73		case Opt_hidepid:
74			if (match_int(&args[0], &option))
75				return 0;
76			if (option < 0 || option > 2) {
77				pr_err("proc: hidepid value must be between 0 and 2.\n");
78				return 0;
79			}
80			pid->hide_pid = option;
81			break;
82		default:
83			pr_err("proc: unrecognized mount option \"%s\" "
84			       "or missing value\n", p);
85			return 0;
86		}
87	}
88
89	return 1;
90}
91
92int proc_remount(struct super_block *sb, int *flags, char *data)
93{
94	struct pid_namespace *pid = sb->s_fs_info;
95
96	sync_filesystem(sb);
97	return !proc_parse_options(data, pid);
98}
99
100static struct dentry *proc_mount(struct file_system_type *fs_type,
101	int flags, const char *dev_name, void *data)
102{
103	int err;
104	struct super_block *sb;
105	struct pid_namespace *ns;
106	char *options;
107
108	if (flags & MS_KERNMOUNT) {
109		ns = (struct pid_namespace *)data;
110		options = NULL;
111	} else {
112		ns = task_active_pid_ns(current);
113		options = data;
114
115		/* Does the mounter have privilege over the pid namespace? */
116		if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
117			return ERR_PTR(-EPERM);
118	}
119
120	sb = sget(fs_type, proc_test_super, proc_set_super, flags, ns);
121	if (IS_ERR(sb))
122		return ERR_CAST(sb);
123
124	/*
125	 * procfs isn't actually a stacking filesystem; however, there is
126	 * too much magic going on inside it to permit stacking things on
127	 * top of it
128	 */
129	sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
130
131	if (!proc_parse_options(options, ns)) {
132		deactivate_locked_super(sb);
133		return ERR_PTR(-EINVAL);
134	}
135
136	if (!sb->s_root) {
137		err = proc_fill_super(sb);
138		if (err) {
139			deactivate_locked_super(sb);
140			return ERR_PTR(err);
141		}
142
143		sb->s_flags |= MS_ACTIVE;
144		/* User space would break if executables appear on proc */
145		sb->s_iflags |= SB_I_NOEXEC;
146	}
147
148	return dget(sb->s_root);
149}
150
151static void proc_kill_sb(struct super_block *sb)
152{
153	struct pid_namespace *ns;
154
155	ns = (struct pid_namespace *)sb->s_fs_info;
156	if (ns->proc_self)
157		dput(ns->proc_self);
158	if (ns->proc_thread_self)
159		dput(ns->proc_thread_self);
160	kill_anon_super(sb);
161	put_pid_ns(ns);
162}
163
164static struct file_system_type proc_fs_type = {
165	.name		= "proc",
166	.mount		= proc_mount,
167	.kill_sb	= proc_kill_sb,
168	.fs_flags	= FS_USERNS_VISIBLE | FS_USERNS_MOUNT,
169};
170
171void __init proc_root_init(void)
172{
173	int err;
174
175	proc_init_inodecache();
176	err = register_filesystem(&proc_fs_type);
177	if (err)
178		return;
179
180	proc_self_init();
181	proc_thread_self_init();
182	proc_symlink("mounts", NULL, "self/mounts");
183
184	proc_net_init();
185
186#ifdef CONFIG_SYSVIPC
187	proc_mkdir("sysvipc", NULL);
188#endif
189	proc_mkdir("fs", NULL);
190	proc_mkdir("driver", NULL);
191	proc_create_mount_point("fs/nfsd"); /* somewhere for the nfsd filesystem to be mounted */
192#if defined(CONFIG_SUN_OPENPROMFS) || defined(CONFIG_SUN_OPENPROMFS_MODULE)
193	/* just give it a mountpoint */
194	proc_create_mount_point("openprom");
195#endif
196	proc_tty_init();
197	proc_mkdir("bus", NULL);
198	proc_sys_init();
199}
200
201static int proc_root_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat
202)
203{
204	generic_fillattr(d_inode(dentry), stat);
205	stat->nlink = proc_root.nlink + nr_processes();
206	return 0;
207}
208
209static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentry, unsigned int flags)
210{
211	if (!proc_pid_lookup(dir, dentry, flags))
212		return NULL;
213
214	return proc_lookup(dir, dentry, flags);
215}
216
217static int proc_root_readdir(struct file *file, struct dir_context *ctx)
218{
219	if (ctx->pos < FIRST_PROCESS_ENTRY) {
220		int error = proc_readdir(file, ctx);
221		if (unlikely(error <= 0))
222			return error;
223		ctx->pos = FIRST_PROCESS_ENTRY;
224	}
225
226	return proc_pid_readdir(file, ctx);
227}
228
229/*
230 * The root /proc directory is special, as it has the
231 * <pid> directories. Thus we don't use the generic
232 * directory handling functions for that..
233 */
234static const struct file_operations proc_root_operations = {
235	.read		 = generic_read_dir,
236	.iterate	 = proc_root_readdir,
237	.llseek		= default_llseek,
238};
239
240/*
241 * proc root can do almost nothing..
242 */
243static const struct inode_operations proc_root_inode_operations = {
244	.lookup		= proc_root_lookup,
245	.getattr	= proc_root_getattr,
246};
247
248/*
249 * This is the root "inode" in the /proc tree..
250 */
251struct proc_dir_entry proc_root = {
252	.low_ino	= PROC_ROOT_INO,
253	.namelen	= 5,
254	.mode		= S_IFDIR | S_IRUGO | S_IXUGO,
255	.nlink		= 2,
256	.count		= ATOMIC_INIT(1),
257	.proc_iops	= &proc_root_inode_operations,
258	.proc_fops	= &proc_root_operations,
259	.parent		= &proc_root,
260	.subdir		= RB_ROOT,
261	.name		= "/proc",
262};
263
264int pid_ns_prepare_proc(struct pid_namespace *ns)
265{
266	struct vfsmount *mnt;
267
268	mnt = kern_mount_data(&proc_fs_type, ns);
269	if (IS_ERR(mnt))
270		return PTR_ERR(mnt);
271
272	ns->proc_mnt = mnt;
273	return 0;
274}
275
276void pid_ns_release_proc(struct pid_namespace *ns)
277{
278	kern_unmount(ns->proc_mnt);
279}
280