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 | |
29 | struct 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 | |
37 | enum proc_param { |
38 | Opt_gid, |
39 | Opt_hidepid, |
40 | Opt_subset, |
41 | }; |
42 | |
43 | static 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 | |
50 | static 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 | |
58 | static 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(s: param->string, base, res: &result.uint_32)) { |
69 | if (!valid_hidepid(value: 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 | |
89 | static 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 | |
112 | static 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, desc: proc_fs_parameters, param, result: &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, value: 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 | |
145 | static 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(from: user_ns, gid: 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 | |
159 | static 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(size: sizeof(*fs_info), GFP_KERNEL); |
167 | if (!fs_info) |
168 | return -ENOMEM; |
169 | |
170 | fs_info->pid_ns = get_pid_ns(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(pde: &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 | |
213 | static 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 | |
224 | static int proc_get_tree(struct fs_context *fc) |
225 | { |
226 | return get_tree_nodev(fc, fill_super: proc_fill_super); |
227 | } |
228 | |
229 | static void proc_fs_context_free(struct fs_context *fc) |
230 | { |
231 | struct proc_fs_context *ctx = fc->fs_private; |
232 | |
233 | put_pid_ns(ns: ctx->pid_ns); |
234 | kfree(objp: ctx); |
235 | } |
236 | |
237 | static 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 | |
244 | static int proc_init_fs_context(struct fs_context *fc) |
245 | { |
246 | struct proc_fs_context *ctx; |
247 | |
248 | ctx = kzalloc(size: sizeof(struct proc_fs_context), GFP_KERNEL); |
249 | if (!ctx) |
250 | return -ENOMEM; |
251 | |
252 | ctx->pid_ns = get_pid_ns(ns: task_active_pid_ns(current)); |
253 | put_user_ns(ns: fc->user_ns); |
254 | fc->user_ns = get_user_ns(ns: ctx->pid_ns->user_ns); |
255 | fc->fs_private = ctx; |
256 | fc->ops = &proc_fs_context_ops; |
257 | return 0; |
258 | } |
259 | |
260 | static 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(ns: fs_info->pid_ns); |
274 | kfree_rcu(fs_info, rcu); |
275 | } |
276 | |
277 | static 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 | |
285 | void __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(name: "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 | |
313 | static 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(dentry: path->dentry), |
318 | stat); |
319 | stat->nlink = proc_root.nlink + nr_processes(); |
320 | return 0; |
321 | } |
322 | |
323 | static 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 | |
331 | static 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 | */ |
348 | static 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 | */ |
357 | static 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 | */ |
365 | struct 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 | }; |
377 | |