Loading...
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(fs_info);
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 user_namespace *mnt_userns,
314 const struct path *path, struct kstat *stat,
315 u32 request_mask, unsigned int query_flags)
316{
317 generic_fillattr(&init_user_ns, d_inode(path->dentry), stat);
318 stat->nlink = proc_root.nlink + nr_processes();
319 return 0;
320}
321
322static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentry, unsigned int flags)
323{
324 if (!proc_pid_lookup(dentry, flags))
325 return NULL;
326
327 return proc_lookup(dir, dentry, flags);
328}
329
330static int proc_root_readdir(struct file *file, struct dir_context *ctx)
331{
332 if (ctx->pos < FIRST_PROCESS_ENTRY) {
333 int error = proc_readdir(file, ctx);
334 if (unlikely(error <= 0))
335 return error;
336 ctx->pos = FIRST_PROCESS_ENTRY;
337 }
338
339 return proc_pid_readdir(file, ctx);
340}
341
342/*
343 * The root /proc directory is special, as it has the
344 * <pid> directories. Thus we don't use the generic
345 * directory handling functions for that..
346 */
347static const struct file_operations proc_root_operations = {
348 .read = generic_read_dir,
349 .iterate_shared = proc_root_readdir,
350 .llseek = generic_file_llseek,
351};
352
353/*
354 * proc root can do almost nothing..
355 */
356static const struct inode_operations proc_root_inode_operations = {
357 .lookup = proc_root_lookup,
358 .getattr = proc_root_getattr,
359};
360
361/*
362 * This is the root "inode" in the /proc tree..
363 */
364struct proc_dir_entry proc_root = {
365 .low_ino = PROC_ROOT_INO,
366 .namelen = 5,
367 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
368 .nlink = 2,
369 .refcnt = REFCOUNT_INIT(1),
370 .proc_iops = &proc_root_inode_operations,
371 .proc_dir_ops = &proc_root_operations,
372 .parent = &proc_root,
373 .subdir = RB_ROOT,
374 .name = "/proc",
375};
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}