xref: /linux/fs/proc/root.c (revision 24bce201d79807b668bf9d9e0aca801c5c0d5f78)
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 
32 struct 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 
40 enum proc_param {
41 	Opt_gid,
42 	Opt_hidepid,
43 	Opt_subset,
44 };
45 
46 static 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 
53 static 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 
61 static 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 
92 static 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 
115 static 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 
148 static 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 
162 static 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 
216 static 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 
227 static int proc_get_tree(struct fs_context *fc)
228 {
229 	return get_tree_nodev(fc, proc_fill_super);
230 }
231 
232 static 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 
240 static 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 
247 static 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 
263 static 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 
280 static 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 
288 void __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 
311 static int proc_root_getattr(struct user_namespace *mnt_userns,
312 			     const struct path *path, struct kstat *stat,
313 			     u32 request_mask, unsigned int query_flags)
314 {
315 	generic_fillattr(&init_user_ns, d_inode(path->dentry), stat);
316 	stat->nlink = proc_root.nlink + nr_processes();
317 	return 0;
318 }
319 
320 static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentry, unsigned int flags)
321 {
322 	if (!proc_pid_lookup(dentry, flags))
323 		return NULL;
324 
325 	return proc_lookup(dir, dentry, flags);
326 }
327 
328 static int proc_root_readdir(struct file *file, struct dir_context *ctx)
329 {
330 	if (ctx->pos < FIRST_PROCESS_ENTRY) {
331 		int error = proc_readdir(file, ctx);
332 		if (unlikely(error <= 0))
333 			return error;
334 		ctx->pos = FIRST_PROCESS_ENTRY;
335 	}
336 
337 	return proc_pid_readdir(file, ctx);
338 }
339 
340 /*
341  * The root /proc directory is special, as it has the
342  * <pid> directories. Thus we don't use the generic
343  * directory handling functions for that..
344  */
345 static const struct file_operations proc_root_operations = {
346 	.read		 = generic_read_dir,
347 	.iterate_shared	 = proc_root_readdir,
348 	.llseek		= generic_file_llseek,
349 };
350 
351 /*
352  * proc root can do almost nothing..
353  */
354 static const struct inode_operations proc_root_inode_operations = {
355 	.lookup		= proc_root_lookup,
356 	.getattr	= proc_root_getattr,
357 };
358 
359 /*
360  * This is the root "inode" in the /proc tree..
361  */
362 struct proc_dir_entry proc_root = {
363 	.low_ino	= PROC_ROOT_INO,
364 	.namelen	= 5,
365 	.mode		= S_IFDIR | S_IRUGO | S_IXUGO,
366 	.nlink		= 2,
367 	.refcnt		= REFCOUNT_INIT(1),
368 	.proc_iops	= &proc_root_inode_operations,
369 	.proc_dir_ops	= &proc_root_operations,
370 	.parent		= &proc_root,
371 	.subdir		= RB_ROOT,
372 	.name		= "/proc",
373 };
374