xref: /kernel/linux/linux-5.10/fs/proc/root.c (revision 8c2ecf20) 
1// SPDX-License-Identifier: GPL-2.0
2/*
3 *  linux/fs/proc/root.c
4 *
5 *  Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 *  proc root directory handling functions
8 */
9
10#include <linux/uaccess.h>
11
12#include <linux/errno.h>
13#include <linux/time.h>
14#include <linux/proc_fs.h>
15#include <linux/stat.h>
16#include <linux/init.h>
17#include <linux/sched.h>
18#include <linux/sched/stat.h>
19#include <linux/module.h>
20#include <linux/bitops.h>
21#include <linux/user_namespace.h>
22#include <linux/fs_context.h>
23#include <linux/mount.h>
24#include <linux/pid_namespace.h>
25#include <linux/fs_parser.h>
26#include <linux/cred.h>
27#include <linux/magic.h>
28#include <linux/slab.h>
29
30#include "internal.h"
31
32struct proc_fs_context {
33	struct pid_namespace	*pid_ns;
34	unsigned int		mask;
35	enum proc_hidepid	hidepid;
36	int			gid;
37	enum proc_pidonly	pidonly;
38};
39
40enum proc_param {
41	Opt_gid,
42	Opt_hidepid,
43	Opt_subset,
44};
45
46static const struct fs_parameter_spec proc_fs_parameters[] = {
47	fsparam_u32("gid",	Opt_gid),
48	fsparam_string("hidepid",	Opt_hidepid),
49	fsparam_string("subset",	Opt_subset),
50	{}
51};
52
53static inline int valid_hidepid(unsigned int value)
54{
55	return (value == HIDEPID_OFF ||
56		value == HIDEPID_NO_ACCESS ||
57		value == HIDEPID_INVISIBLE ||
58		value == HIDEPID_NOT_PTRACEABLE);
59}
60
61static int proc_parse_hidepid_param(struct fs_context *fc, struct fs_parameter *param)
62{
63	struct proc_fs_context *ctx = fc->fs_private;
64	struct fs_parameter_spec hidepid_u32_spec = fsparam_u32("hidepid", Opt_hidepid);
65	struct fs_parse_result result;
66	int base = (unsigned long)hidepid_u32_spec.data;
67
68	if (param->type != fs_value_is_string)
69		return invalf(fc, "proc: unexpected type of hidepid value\n");
70
71	if (!kstrtouint(param->string, base, &result.uint_32)) {
72		if (!valid_hidepid(result.uint_32))
73			return invalf(fc, "proc: unknown value of hidepid - %s\n", param->string);
74		ctx->hidepid = result.uint_32;
75		return 0;
76	}
77
78	if (!strcmp(param->string, "off"))
79		ctx->hidepid = HIDEPID_OFF;
80	else if (!strcmp(param->string, "noaccess"))
81		ctx->hidepid = HIDEPID_NO_ACCESS;
82	else if (!strcmp(param->string, "invisible"))
83		ctx->hidepid = HIDEPID_INVISIBLE;
84	else if (!strcmp(param->string, "ptraceable"))
85		ctx->hidepid = HIDEPID_NOT_PTRACEABLE;
86	else
87		return invalf(fc, "proc: unknown value of hidepid - %s\n", param->string);
88
89	return 0;
90}
91
92static int proc_parse_subset_param(struct fs_context *fc, char *value)
93{
94	struct proc_fs_context *ctx = fc->fs_private;
95
96	while (value) {
97		char *ptr = strchr(value, ',');
98
99		if (ptr != NULL)
100			*ptr++ = '\0';
101
102		if (*value != '\0') {
103			if (!strcmp(value, "pid")) {
104				ctx->pidonly = PROC_PIDONLY_ON;
105			} else {
106				return invalf(fc, "proc: unsupported subset option - %s\n", value);
107			}
108		}
109		value = ptr;
110	}
111
112	return 0;
113}
114
115static int proc_parse_param(struct fs_context *fc, struct fs_parameter *param)
116{
117	struct proc_fs_context *ctx = fc->fs_private;
118	struct fs_parse_result result;
119	int opt;
120
121	opt = fs_parse(fc, proc_fs_parameters, param, &result);
122	if (opt < 0)
123		return opt;
124
125	switch (opt) {
126	case Opt_gid:
127		ctx->gid = result.uint_32;
128		break;
129
130	case Opt_hidepid:
131		if (proc_parse_hidepid_param(fc, param))
132			return -EINVAL;
133		break;
134
135	case Opt_subset:
136		if (proc_parse_subset_param(fc, param->string) < 0)
137			return -EINVAL;
138		break;
139
140	default:
141		return -EINVAL;
142	}
143
144	ctx->mask |= 1 << opt;
145	return 0;
146}
147
148static void proc_apply_options(struct proc_fs_info *fs_info,
149			       struct fs_context *fc,
150			       struct user_namespace *user_ns)
151{
152	struct proc_fs_context *ctx = fc->fs_private;
153
154	if (ctx->mask & (1 << Opt_gid))
155		fs_info->pid_gid = make_kgid(user_ns, ctx->gid);
156	if (ctx->mask & (1 << Opt_hidepid))
157		fs_info->hide_pid = ctx->hidepid;
158	if (ctx->mask & (1 << Opt_subset))
159		fs_info->pidonly = ctx->pidonly;
160}
161
162static int proc_fill_super(struct super_block *s, struct fs_context *fc)
163{
164	struct proc_fs_context *ctx = fc->fs_private;
165	struct inode *root_inode;
166	struct proc_fs_info *fs_info;
167	int ret;
168
169	fs_info = kzalloc(sizeof(*fs_info), GFP_KERNEL);
170	if (!fs_info)
171		return -ENOMEM;
172
173	fs_info->pid_ns = get_pid_ns(ctx->pid_ns);
174	proc_apply_options(fs_info, fc, current_user_ns());
175
176	/* User space would break if executables or devices appear on proc */
177	s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
178	s->s_flags |= SB_NODIRATIME | SB_NOSUID | SB_NOEXEC;
179	s->s_blocksize = 1024;
180	s->s_blocksize_bits = 10;
181	s->s_magic = PROC_SUPER_MAGIC;
182	s->s_op = &proc_sops;
183	s->s_time_gran = 1;
184	s->s_fs_info = fs_info;
185
186	/*
187	 * procfs isn't actually a stacking filesystem; however, there is
188	 * too much magic going on inside it to permit stacking things on
189	 * top of it
190	 */
191	s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
192
193	/* procfs dentries and inodes don't require IO to create */
194	s->s_shrink.seeks = 0;
195
196	pde_get(&proc_root);
197	root_inode = proc_get_inode(s, &proc_root);
198	if (!root_inode) {
199		pr_err("proc_fill_super: get root inode failed\n");
200		return -ENOMEM;
201	}
202
203	s->s_root = d_make_root(root_inode);
204	if (!s->s_root) {
205		pr_err("proc_fill_super: allocate dentry failed\n");
206		return -ENOMEM;
207	}
208
209	ret = proc_setup_self(s);
210	if (ret) {
211		return ret;
212	}
213	return proc_setup_thread_self(s);
214}
215
216static int proc_reconfigure(struct fs_context *fc)
217{
218	struct super_block *sb = fc->root->d_sb;
219	struct proc_fs_info *fs_info = proc_sb_info(sb);
220
221	sync_filesystem(sb);
222
223	proc_apply_options(fs_info, fc, current_user_ns());
224	return 0;
225}
226
227static int proc_get_tree(struct fs_context *fc)
228{
229	return get_tree_nodev(fc, proc_fill_super);
230}
231
232static void proc_fs_context_free(struct fs_context *fc)
233{
234	struct proc_fs_context *ctx = fc->fs_private;
235
236	put_pid_ns(ctx->pid_ns);
237	kfree(ctx);
238}
239
240static const struct fs_context_operations proc_fs_context_ops = {
241	.free		= proc_fs_context_free,
242	.parse_param	= proc_parse_param,
243	.get_tree	= proc_get_tree,
244	.reconfigure	= proc_reconfigure,
245};
246
247static int proc_init_fs_context(struct fs_context *fc)
248{
249	struct proc_fs_context *ctx;
250
251	ctx = kzalloc(sizeof(struct proc_fs_context), GFP_KERNEL);
252	if (!ctx)
253		return -ENOMEM;
254
255	ctx->pid_ns = get_pid_ns(task_active_pid_ns(current));
256	put_user_ns(fc->user_ns);
257	fc->user_ns = get_user_ns(ctx->pid_ns->user_ns);
258	fc->fs_private = ctx;
259	fc->ops = &proc_fs_context_ops;
260	return 0;
261}
262
263static void proc_kill_sb(struct super_block *sb)
264{
265	struct proc_fs_info *fs_info = proc_sb_info(sb);
266
267	if (!fs_info) {
268		kill_anon_super(sb);
269		return;
270	}
271
272	dput(fs_info->proc_self);
273	dput(fs_info->proc_thread_self);
274
275	kill_anon_super(sb);
276	put_pid_ns(fs_info->pid_ns);
277	kfree(fs_info);
278}
279
280static struct file_system_type proc_fs_type = {
281	.name			= "proc",
282	.init_fs_context	= proc_init_fs_context,
283	.parameters		= proc_fs_parameters,
284	.kill_sb		= proc_kill_sb,
285	.fs_flags		= FS_USERNS_MOUNT | FS_DISALLOW_NOTIFY_PERM,
286};
287
288void __init proc_root_init(void)
289{
290	proc_init_kmemcache();
291	set_proc_pid_nlink();
292	proc_self_init();
293	proc_thread_self_init();
294	proc_symlink("mounts", NULL, "self/mounts");
295
296	proc_net_init();
297	proc_mkdir("fs", NULL);
298	proc_mkdir("driver", NULL);
299	proc_create_mount_point("fs/nfsd"); /* somewhere for the nfsd filesystem to be mounted */
300#if defined(CONFIG_SUN_OPENPROMFS) || defined(CONFIG_SUN_OPENPROMFS_MODULE)
301	/* just give it a mountpoint */
302	proc_create_mount_point("openprom");
303#endif
304	proc_tty_init();
305	proc_mkdir("bus", NULL);
306	proc_sys_init();
307
308	register_filesystem(&proc_fs_type);
309}
310
311static int proc_root_getattr(const struct path *path, struct kstat *stat,
312			     u32 request_mask, unsigned int query_flags)
313{
314	generic_fillattr(d_inode(path->dentry), stat);
315	stat->nlink = proc_root.nlink + nr_processes();
316	return 0;
317}
318
319static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentry, unsigned int flags)
320{
321	if (!proc_pid_lookup(dentry, flags))
322		return NULL;
323
324	return proc_lookup(dir, dentry, flags);
325}
326
327static int proc_root_readdir(struct file *file, struct dir_context *ctx)
328{
329	if (ctx->pos < FIRST_PROCESS_ENTRY) {
330		int error = proc_readdir(file, ctx);
331		if (unlikely(error <= 0))
332			return error;
333		ctx->pos = FIRST_PROCESS_ENTRY;
334	}
335
336	return proc_pid_readdir(file, ctx);
337}
338
339/*
340 * The root /proc directory is special, as it has the
341 * <pid> directories. Thus we don't use the generic
342 * directory handling functions for that..
343 */
344static const struct file_operations proc_root_operations = {
345	.read		 = generic_read_dir,
346	.iterate_shared	 = proc_root_readdir,
347	.llseek		= generic_file_llseek,
348};
349
350/*
351 * proc root can do almost nothing..
352 */
353static const struct inode_operations proc_root_inode_operations = {
354	.lookup		= proc_root_lookup,
355	.getattr	= proc_root_getattr,
356};
357
358/*
359 * This is the root "inode" in the /proc tree..
360 */
361struct proc_dir_entry proc_root = {
362	.low_ino	= PROC_ROOT_INO,
363	.namelen	= 5,
364	.mode		= S_IFDIR | S_IRUGO | S_IXUGO,
365	.nlink		= 2,
366	.refcnt		= REFCOUNT_INIT(1),
367	.proc_iops	= &proc_root_inode_operations,
368	.proc_dir_ops	= &proc_root_operations,
369	.parent		= &proc_root,
370	.subdir		= RB_ROOT,
371	.name		= "/proc",
372};
373