Loading...
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 if (!proc_parse_options(options, ns)) {
125 deactivate_locked_super(sb);
126 return ERR_PTR(-EINVAL);
127 }
128
129 if (!sb->s_root) {
130 err = proc_fill_super(sb);
131 if (err) {
132 deactivate_locked_super(sb);
133 return ERR_PTR(err);
134 }
135
136 sb->s_flags |= MS_ACTIVE;
137 /* User space would break if executables appear on proc */
138 sb->s_iflags |= SB_I_NOEXEC;
139 }
140
141 return dget(sb->s_root);
142}
143
144static void proc_kill_sb(struct super_block *sb)
145{
146 struct pid_namespace *ns;
147
148 ns = (struct pid_namespace *)sb->s_fs_info;
149 if (ns->proc_self)
150 dput(ns->proc_self);
151 if (ns->proc_thread_self)
152 dput(ns->proc_thread_self);
153 kill_anon_super(sb);
154 put_pid_ns(ns);
155}
156
157static struct file_system_type proc_fs_type = {
158 .name = "proc",
159 .mount = proc_mount,
160 .kill_sb = proc_kill_sb,
161 .fs_flags = FS_USERNS_VISIBLE | FS_USERNS_MOUNT,
162};
163
164void __init proc_root_init(void)
165{
166 int err;
167
168 proc_init_inodecache();
169 err = register_filesystem(&proc_fs_type);
170 if (err)
171 return;
172
173 proc_self_init();
174 proc_thread_self_init();
175 proc_symlink("mounts", NULL, "self/mounts");
176
177 proc_net_init();
178
179#ifdef CONFIG_SYSVIPC
180 proc_mkdir("sysvipc", NULL);
181#endif
182 proc_mkdir("fs", NULL);
183 proc_mkdir("driver", NULL);
184 proc_create_mount_point("fs/nfsd"); /* somewhere for the nfsd filesystem to be mounted */
185#if defined(CONFIG_SUN_OPENPROMFS) || defined(CONFIG_SUN_OPENPROMFS_MODULE)
186 /* just give it a mountpoint */
187 proc_create_mount_point("openprom");
188#endif
189 proc_tty_init();
190 proc_mkdir("bus", NULL);
191 proc_sys_init();
192}
193
194static int proc_root_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat
195)
196{
197 generic_fillattr(d_inode(dentry), stat);
198 stat->nlink = proc_root.nlink + nr_processes();
199 return 0;
200}
201
202static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentry, unsigned int flags)
203{
204 if (!proc_pid_lookup(dir, dentry, flags))
205 return NULL;
206
207 return proc_lookup(dir, dentry, flags);
208}
209
210static int proc_root_readdir(struct file *file, struct dir_context *ctx)
211{
212 if (ctx->pos < FIRST_PROCESS_ENTRY) {
213 int error = proc_readdir(file, ctx);
214 if (unlikely(error <= 0))
215 return error;
216 ctx->pos = FIRST_PROCESS_ENTRY;
217 }
218
219 return proc_pid_readdir(file, ctx);
220}
221
222/*
223 * The root /proc directory is special, as it has the
224 * <pid> directories. Thus we don't use the generic
225 * directory handling functions for that..
226 */
227static const struct file_operations proc_root_operations = {
228 .read = generic_read_dir,
229 .iterate = proc_root_readdir,
230 .llseek = default_llseek,
231};
232
233/*
234 * proc root can do almost nothing..
235 */
236static const struct inode_operations proc_root_inode_operations = {
237 .lookup = proc_root_lookup,
238 .getattr = proc_root_getattr,
239};
240
241/*
242 * This is the root "inode" in the /proc tree..
243 */
244struct proc_dir_entry proc_root = {
245 .low_ino = PROC_ROOT_INO,
246 .namelen = 5,
247 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
248 .nlink = 2,
249 .count = ATOMIC_INIT(1),
250 .proc_iops = &proc_root_inode_operations,
251 .proc_fops = &proc_root_operations,
252 .parent = &proc_root,
253 .subdir = RB_ROOT,
254 .name = "/proc",
255};
256
257int pid_ns_prepare_proc(struct pid_namespace *ns)
258{
259 struct vfsmount *mnt;
260
261 mnt = kern_mount_data(&proc_fs_type, ns);
262 if (IS_ERR(mnt))
263 return PTR_ERR(mnt);
264
265 ns->proc_mnt = mnt;
266 return 0;
267}
268
269void pid_ns_release_proc(struct pid_namespace *ns)
270{
271 kern_unmount(ns->proc_mnt);
272}
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#include <linux/errno.h>
10#include <linux/time.h>
11#include <linux/proc_fs.h>
12#include <linux/stat.h>
13#include <linux/init.h>
14#include <linux/sched.h>
15#include <linux/sched/stat.h>
16#include <linux/module.h>
17#include <linux/bitops.h>
18#include <linux/user_namespace.h>
19#include <linux/fs_context.h>
20#include <linux/mount.h>
21#include <linux/pid_namespace.h>
22#include <linux/fs_parser.h>
23#include <linux/cred.h>
24#include <linux/magic.h>
25#include <linux/slab.h>
26
27#include "internal.h"
28
29struct proc_fs_context {
30 struct pid_namespace *pid_ns;
31 unsigned int mask;
32 enum proc_hidepid hidepid;
33 int gid;
34 enum proc_pidonly pidonly;
35};
36
37enum proc_param {
38 Opt_gid,
39 Opt_hidepid,
40 Opt_subset,
41};
42
43static const struct fs_parameter_spec proc_fs_parameters[] = {
44 fsparam_u32("gid", Opt_gid),
45 fsparam_string("hidepid", Opt_hidepid),
46 fsparam_string("subset", Opt_subset),
47 {}
48};
49
50static inline int valid_hidepid(unsigned int value)
51{
52 return (value == HIDEPID_OFF ||
53 value == HIDEPID_NO_ACCESS ||
54 value == HIDEPID_INVISIBLE ||
55 value == HIDEPID_NOT_PTRACEABLE);
56}
57
58static int proc_parse_hidepid_param(struct fs_context *fc, struct fs_parameter *param)
59{
60 struct proc_fs_context *ctx = fc->fs_private;
61 struct fs_parameter_spec hidepid_u32_spec = fsparam_u32("hidepid", Opt_hidepid);
62 struct fs_parse_result result;
63 int base = (unsigned long)hidepid_u32_spec.data;
64
65 if (param->type != fs_value_is_string)
66 return invalf(fc, "proc: unexpected type of hidepid value\n");
67
68 if (!kstrtouint(param->string, base, &result.uint_32)) {
69 if (!valid_hidepid(result.uint_32))
70 return invalf(fc, "proc: unknown value of hidepid - %s\n", param->string);
71 ctx->hidepid = result.uint_32;
72 return 0;
73 }
74
75 if (!strcmp(param->string, "off"))
76 ctx->hidepid = HIDEPID_OFF;
77 else if (!strcmp(param->string, "noaccess"))
78 ctx->hidepid = HIDEPID_NO_ACCESS;
79 else if (!strcmp(param->string, "invisible"))
80 ctx->hidepid = HIDEPID_INVISIBLE;
81 else if (!strcmp(param->string, "ptraceable"))
82 ctx->hidepid = HIDEPID_NOT_PTRACEABLE;
83 else
84 return invalf(fc, "proc: unknown value of hidepid - %s\n", param->string);
85
86 return 0;
87}
88
89static int proc_parse_subset_param(struct fs_context *fc, char *value)
90{
91 struct proc_fs_context *ctx = fc->fs_private;
92
93 while (value) {
94 char *ptr = strchr(value, ',');
95
96 if (ptr != NULL)
97 *ptr++ = '\0';
98
99 if (*value != '\0') {
100 if (!strcmp(value, "pid")) {
101 ctx->pidonly = PROC_PIDONLY_ON;
102 } else {
103 return invalf(fc, "proc: unsupported subset option - %s\n", value);
104 }
105 }
106 value = ptr;
107 }
108
109 return 0;
110}
111
112static int proc_parse_param(struct fs_context *fc, struct fs_parameter *param)
113{
114 struct proc_fs_context *ctx = fc->fs_private;
115 struct fs_parse_result result;
116 int opt;
117
118 opt = fs_parse(fc, proc_fs_parameters, param, &result);
119 if (opt < 0)
120 return opt;
121
122 switch (opt) {
123 case Opt_gid:
124 ctx->gid = result.uint_32;
125 break;
126
127 case Opt_hidepid:
128 if (proc_parse_hidepid_param(fc, param))
129 return -EINVAL;
130 break;
131
132 case Opt_subset:
133 if (proc_parse_subset_param(fc, param->string) < 0)
134 return -EINVAL;
135 break;
136
137 default:
138 return -EINVAL;
139 }
140
141 ctx->mask |= 1 << opt;
142 return 0;
143}
144
145static void proc_apply_options(struct proc_fs_info *fs_info,
146 struct fs_context *fc,
147 struct user_namespace *user_ns)
148{
149 struct proc_fs_context *ctx = fc->fs_private;
150
151 if (ctx->mask & (1 << Opt_gid))
152 fs_info->pid_gid = make_kgid(user_ns, ctx->gid);
153 if (ctx->mask & (1 << Opt_hidepid))
154 fs_info->hide_pid = ctx->hidepid;
155 if (ctx->mask & (1 << Opt_subset))
156 fs_info->pidonly = ctx->pidonly;
157}
158
159static int proc_fill_super(struct super_block *s, struct fs_context *fc)
160{
161 struct proc_fs_context *ctx = fc->fs_private;
162 struct inode *root_inode;
163 struct proc_fs_info *fs_info;
164 int ret;
165
166 fs_info = kzalloc(sizeof(*fs_info), GFP_KERNEL);
167 if (!fs_info)
168 return -ENOMEM;
169
170 fs_info->pid_ns = get_pid_ns(ctx->pid_ns);
171 proc_apply_options(fs_info, fc, current_user_ns());
172
173 /* User space would break if executables or devices appear on proc */
174 s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
175 s->s_flags |= SB_NODIRATIME | SB_NOSUID | SB_NOEXEC;
176 s->s_blocksize = 1024;
177 s->s_blocksize_bits = 10;
178 s->s_magic = PROC_SUPER_MAGIC;
179 s->s_op = &proc_sops;
180 s->s_time_gran = 1;
181 s->s_fs_info = fs_info;
182
183 /*
184 * procfs isn't actually a stacking filesystem; however, there is
185 * too much magic going on inside it to permit stacking things on
186 * top of it
187 */
188 s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
189
190 /* procfs dentries and inodes don't require IO to create */
191 s->s_shrink->seeks = 0;
192
193 pde_get(&proc_root);
194 root_inode = proc_get_inode(s, &proc_root);
195 if (!root_inode) {
196 pr_err("proc_fill_super: get root inode failed\n");
197 return -ENOMEM;
198 }
199
200 s->s_root = d_make_root(root_inode);
201 if (!s->s_root) {
202 pr_err("proc_fill_super: allocate dentry failed\n");
203 return -ENOMEM;
204 }
205
206 ret = proc_setup_self(s);
207 if (ret) {
208 return ret;
209 }
210 return proc_setup_thread_self(s);
211}
212
213static int proc_reconfigure(struct fs_context *fc)
214{
215 struct super_block *sb = fc->root->d_sb;
216 struct proc_fs_info *fs_info = proc_sb_info(sb);
217
218 sync_filesystem(sb);
219
220 proc_apply_options(fs_info, fc, current_user_ns());
221 return 0;
222}
223
224static int proc_get_tree(struct fs_context *fc)
225{
226 return get_tree_nodev(fc, proc_fill_super);
227}
228
229static void proc_fs_context_free(struct fs_context *fc)
230{
231 struct proc_fs_context *ctx = fc->fs_private;
232
233 put_pid_ns(ctx->pid_ns);
234 kfree(ctx);
235}
236
237static const struct fs_context_operations proc_fs_context_ops = {
238 .free = proc_fs_context_free,
239 .parse_param = proc_parse_param,
240 .get_tree = proc_get_tree,
241 .reconfigure = proc_reconfigure,
242};
243
244static int proc_init_fs_context(struct fs_context *fc)
245{
246 struct proc_fs_context *ctx;
247
248 ctx = kzalloc(sizeof(struct proc_fs_context), GFP_KERNEL);
249 if (!ctx)
250 return -ENOMEM;
251
252 ctx->pid_ns = get_pid_ns(task_active_pid_ns(current));
253 put_user_ns(fc->user_ns);
254 fc->user_ns = get_user_ns(ctx->pid_ns->user_ns);
255 fc->fs_private = ctx;
256 fc->ops = &proc_fs_context_ops;
257 return 0;
258}
259
260static void proc_kill_sb(struct super_block *sb)
261{
262 struct proc_fs_info *fs_info = proc_sb_info(sb);
263
264 if (!fs_info) {
265 kill_anon_super(sb);
266 return;
267 }
268
269 dput(fs_info->proc_self);
270 dput(fs_info->proc_thread_self);
271
272 kill_anon_super(sb);
273 put_pid_ns(fs_info->pid_ns);
274 kfree_rcu(fs_info, rcu);
275}
276
277static struct file_system_type proc_fs_type = {
278 .name = "proc",
279 .init_fs_context = proc_init_fs_context,
280 .parameters = proc_fs_parameters,
281 .kill_sb = proc_kill_sb,
282 .fs_flags = FS_USERNS_MOUNT | FS_DISALLOW_NOTIFY_PERM,
283};
284
285void __init proc_root_init(void)
286{
287 proc_init_kmemcache();
288 set_proc_pid_nlink();
289 proc_self_init();
290 proc_thread_self_init();
291 proc_symlink("mounts", NULL, "self/mounts");
292
293 proc_net_init();
294 proc_mkdir("fs", NULL);
295 proc_mkdir("driver", NULL);
296 proc_create_mount_point("fs/nfsd"); /* somewhere for the nfsd filesystem to be mounted */
297#if defined(CONFIG_SUN_OPENPROMFS) || defined(CONFIG_SUN_OPENPROMFS_MODULE)
298 /* just give it a mountpoint */
299 proc_create_mount_point("openprom");
300#endif
301 proc_tty_init();
302 proc_mkdir("bus", NULL);
303 proc_sys_init();
304
305 /*
306 * Last things last. It is not like userspace processes eager
307 * to open /proc files exist at this point but register last
308 * anyway.
309 */
310 register_filesystem(&proc_fs_type);
311}
312
313static int proc_root_getattr(struct mnt_idmap *idmap,
314 const struct path *path, struct kstat *stat,
315 u32 request_mask, unsigned int query_flags)
316{
317 generic_fillattr(&nop_mnt_idmap, request_mask, d_inode(path->dentry),
318 stat);
319 stat->nlink = proc_root.nlink + nr_processes();
320 return 0;
321}
322
323static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentry, unsigned int flags)
324{
325 if (!proc_pid_lookup(dentry, flags))
326 return NULL;
327
328 return proc_lookup(dir, dentry, flags);
329}
330
331static int proc_root_readdir(struct file *file, struct dir_context *ctx)
332{
333 if (ctx->pos < FIRST_PROCESS_ENTRY) {
334 int error = proc_readdir(file, ctx);
335 if (unlikely(error <= 0))
336 return error;
337 ctx->pos = FIRST_PROCESS_ENTRY;
338 }
339
340 return proc_pid_readdir(file, ctx);
341}
342
343/*
344 * The root /proc directory is special, as it has the
345 * <pid> directories. Thus we don't use the generic
346 * directory handling functions for that..
347 */
348static const struct file_operations proc_root_operations = {
349 .read = generic_read_dir,
350 .iterate_shared = proc_root_readdir,
351 .llseek = generic_file_llseek,
352};
353
354/*
355 * proc root can do almost nothing..
356 */
357static const struct inode_operations proc_root_inode_operations = {
358 .lookup = proc_root_lookup,
359 .getattr = proc_root_getattr,
360};
361
362/*
363 * This is the root "inode" in the /proc tree..
364 */
365struct proc_dir_entry proc_root = {
366 .low_ino = PROC_ROOT_INO,
367 .namelen = 5,
368 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
369 .nlink = 2,
370 .refcnt = REFCOUNT_INIT(1),
371 .proc_iops = &proc_root_inode_operations,
372 .proc_dir_ops = &proc_root_operations,
373 .parent = &proc_root,
374 .subdir = RB_ROOT,
375 .name = "/proc",
376};