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
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 int hidepid;
36 int gid;
37};
38
39enum proc_param {
40 Opt_gid,
41 Opt_hidepid,
42};
43
44static const struct fs_parameter_spec proc_param_specs[] = {
45 fsparam_u32("gid", Opt_gid),
46 fsparam_u32("hidepid", Opt_hidepid),
47 {}
48};
49
50static const struct fs_parameter_description proc_fs_parameters = {
51 .name = "proc",
52 .specs = proc_param_specs,
53};
54
55static int proc_parse_param(struct fs_context *fc, struct fs_parameter *param)
56{
57 struct proc_fs_context *ctx = fc->fs_private;
58 struct fs_parse_result result;
59 int opt;
60
61 opt = fs_parse(fc, &proc_fs_parameters, param, &result);
62 if (opt < 0)
63 return opt;
64
65 switch (opt) {
66 case Opt_gid:
67 ctx->gid = result.uint_32;
68 break;
69
70 case Opt_hidepid:
71 ctx->hidepid = result.uint_32;
72 if (ctx->hidepid < HIDEPID_OFF ||
73 ctx->hidepid > HIDEPID_INVISIBLE)
74 return invalf(fc, "proc: hidepid value must be between 0 and 2.\n");
75 break;
76
77 default:
78 return -EINVAL;
79 }
80
81 ctx->mask |= 1 << opt;
82 return 0;
83}
84
85static void proc_apply_options(struct super_block *s,
86 struct fs_context *fc,
87 struct pid_namespace *pid_ns,
88 struct user_namespace *user_ns)
89{
90 struct proc_fs_context *ctx = fc->fs_private;
91
92 if (ctx->mask & (1 << Opt_gid))
93 pid_ns->pid_gid = make_kgid(user_ns, ctx->gid);
94 if (ctx->mask & (1 << Opt_hidepid))
95 pid_ns->hide_pid = ctx->hidepid;
96}
97
98static int proc_fill_super(struct super_block *s, struct fs_context *fc)
99{
100 struct pid_namespace *pid_ns = get_pid_ns(s->s_fs_info);
101 struct inode *root_inode;
102 int ret;
103
104 proc_apply_options(s, fc, pid_ns, current_user_ns());
105
106 /* User space would break if executables or devices appear on proc */
107 s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
108 s->s_flags |= SB_NODIRATIME | SB_NOSUID | SB_NOEXEC;
109 s->s_blocksize = 1024;
110 s->s_blocksize_bits = 10;
111 s->s_magic = PROC_SUPER_MAGIC;
112 s->s_op = &proc_sops;
113 s->s_time_gran = 1;
114
115 /*
116 * procfs isn't actually a stacking filesystem; however, there is
117 * too much magic going on inside it to permit stacking things on
118 * top of it
119 */
120 s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
121
122 /* procfs dentries and inodes don't require IO to create */
123 s->s_shrink.seeks = 0;
124
125 pde_get(&proc_root);
126 root_inode = proc_get_inode(s, &proc_root);
127 if (!root_inode) {
128 pr_err("proc_fill_super: get root inode failed\n");
129 return -ENOMEM;
130 }
131
132 s->s_root = d_make_root(root_inode);
133 if (!s->s_root) {
134 pr_err("proc_fill_super: allocate dentry failed\n");
135 return -ENOMEM;
136 }
137
138 ret = proc_setup_self(s);
139 if (ret) {
140 return ret;
141 }
142 return proc_setup_thread_self(s);
143}
144
145static int proc_reconfigure(struct fs_context *fc)
146{
147 struct super_block *sb = fc->root->d_sb;
148 struct pid_namespace *pid = sb->s_fs_info;
149
150 sync_filesystem(sb);
151
152 proc_apply_options(sb, fc, pid, current_user_ns());
153 return 0;
154}
155
156static int proc_get_tree(struct fs_context *fc)
157{
158 struct proc_fs_context *ctx = fc->fs_private;
159
160 return get_tree_keyed(fc, proc_fill_super, ctx->pid_ns);
161}
162
163static void proc_fs_context_free(struct fs_context *fc)
164{
165 struct proc_fs_context *ctx = fc->fs_private;
166
167 put_pid_ns(ctx->pid_ns);
168 kfree(ctx);
169}
170
171static const struct fs_context_operations proc_fs_context_ops = {
172 .free = proc_fs_context_free,
173 .parse_param = proc_parse_param,
174 .get_tree = proc_get_tree,
175 .reconfigure = proc_reconfigure,
176};
177
178static int proc_init_fs_context(struct fs_context *fc)
179{
180 struct proc_fs_context *ctx;
181
182 ctx = kzalloc(sizeof(struct proc_fs_context), GFP_KERNEL);
183 if (!ctx)
184 return -ENOMEM;
185
186 ctx->pid_ns = get_pid_ns(task_active_pid_ns(current));
187 put_user_ns(fc->user_ns);
188 fc->user_ns = get_user_ns(ctx->pid_ns->user_ns);
189 fc->fs_private = ctx;
190 fc->ops = &proc_fs_context_ops;
191 return 0;
192}
193
194static void proc_kill_sb(struct super_block *sb)
195{
196 struct pid_namespace *ns;
197
198 ns = (struct pid_namespace *)sb->s_fs_info;
199 if (ns->proc_self)
200 dput(ns->proc_self);
201 if (ns->proc_thread_self)
202 dput(ns->proc_thread_self);
203 kill_anon_super(sb);
204 put_pid_ns(ns);
205}
206
207static struct file_system_type proc_fs_type = {
208 .name = "proc",
209 .init_fs_context = proc_init_fs_context,
210 .parameters = &proc_fs_parameters,
211 .kill_sb = proc_kill_sb,
212 .fs_flags = FS_USERNS_MOUNT | FS_DISALLOW_NOTIFY_PERM,
213};
214
215void __init proc_root_init(void)
216{
217 proc_init_kmemcache();
218 set_proc_pid_nlink();
219 proc_self_init();
220 proc_thread_self_init();
221 proc_symlink("mounts", NULL, "self/mounts");
222
223 proc_net_init();
224 proc_mkdir("fs", NULL);
225 proc_mkdir("driver", NULL);
226 proc_create_mount_point("fs/nfsd"); /* somewhere for the nfsd filesystem to be mounted */
227#if defined(CONFIG_SUN_OPENPROMFS) || defined(CONFIG_SUN_OPENPROMFS_MODULE)
228 /* just give it a mountpoint */
229 proc_create_mount_point("openprom");
230#endif
231 proc_tty_init();
232 proc_mkdir("bus", NULL);
233 proc_sys_init();
234
235 register_filesystem(&proc_fs_type);
236}
237
238static int proc_root_getattr(const struct path *path, struct kstat *stat,
239 u32 request_mask, unsigned int query_flags)
240{
241 generic_fillattr(d_inode(path->dentry), stat);
242 stat->nlink = proc_root.nlink + nr_processes();
243 return 0;
244}
245
246static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentry, unsigned int flags)
247{
248 if (!proc_pid_lookup(dentry, flags))
249 return NULL;
250
251 return proc_lookup(dir, dentry, flags);
252}
253
254static int proc_root_readdir(struct file *file, struct dir_context *ctx)
255{
256 if (ctx->pos < FIRST_PROCESS_ENTRY) {
257 int error = proc_readdir(file, ctx);
258 if (unlikely(error <= 0))
259 return error;
260 ctx->pos = FIRST_PROCESS_ENTRY;
261 }
262
263 return proc_pid_readdir(file, ctx);
264}
265
266/*
267 * The root /proc directory is special, as it has the
268 * <pid> directories. Thus we don't use the generic
269 * directory handling functions for that..
270 */
271static const struct file_operations proc_root_operations = {
272 .read = generic_read_dir,
273 .iterate_shared = proc_root_readdir,
274 .llseek = generic_file_llseek,
275};
276
277/*
278 * proc root can do almost nothing..
279 */
280static const struct inode_operations proc_root_inode_operations = {
281 .lookup = proc_root_lookup,
282 .getattr = proc_root_getattr,
283};
284
285/*
286 * This is the root "inode" in the /proc tree..
287 */
288struct proc_dir_entry proc_root = {
289 .low_ino = PROC_ROOT_INO,
290 .namelen = 5,
291 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
292 .nlink = 2,
293 .refcnt = REFCOUNT_INIT(1),
294 .proc_iops = &proc_root_inode_operations,
295 .proc_fops = &proc_root_operations,
296 .parent = &proc_root,
297 .subdir = RB_ROOT,
298 .name = "/proc",
299};
300
301int pid_ns_prepare_proc(struct pid_namespace *ns)
302{
303 struct proc_fs_context *ctx;
304 struct fs_context *fc;
305 struct vfsmount *mnt;
306
307 fc = fs_context_for_mount(&proc_fs_type, SB_KERNMOUNT);
308 if (IS_ERR(fc))
309 return PTR_ERR(fc);
310
311 if (fc->user_ns != ns->user_ns) {
312 put_user_ns(fc->user_ns);
313 fc->user_ns = get_user_ns(ns->user_ns);
314 }
315
316 ctx = fc->fs_private;
317 if (ctx->pid_ns != ns) {
318 put_pid_ns(ctx->pid_ns);
319 get_pid_ns(ns);
320 ctx->pid_ns = ns;
321 }
322
323 mnt = fc_mount(fc);
324 put_fs_context(fc);
325 if (IS_ERR(mnt))
326 return PTR_ERR(mnt);
327
328 ns->proc_mnt = mnt;
329 return 0;
330}
331
332void pid_ns_release_proc(struct pid_namespace *ns)
333{
334 kern_unmount(ns->proc_mnt);
335}