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 int hidepid;
36 int gid;
37 };
38
39 enum proc_param {
40 Opt_gid,
41 Opt_hidepid,
42 };
43
44 static const struct fs_parameter_spec proc_param_specs[] = {
45 fsparam_u32("gid", Opt_gid),
46 fsparam_u32("hidepid", Opt_hidepid),
47 {}
48 };
49
50 static const struct fs_parameter_description proc_fs_parameters = {
51 .name = "proc",
52 .specs = proc_param_specs,
53 };
54
proc_parse_param(struct fs_context * fc,struct fs_parameter * param)55 static 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
proc_apply_options(struct fs_context * fc,struct pid_namespace * pid_ns,struct user_namespace * user_ns)85 static void proc_apply_options(struct fs_context *fc,
86 struct pid_namespace *pid_ns,
87 struct user_namespace *user_ns)
88 {
89 struct proc_fs_context *ctx = fc->fs_private;
90
91 if (ctx->mask & (1 << Opt_gid))
92 pid_ns->pid_gid = make_kgid(user_ns, ctx->gid);
93 if (ctx->mask & (1 << Opt_hidepid))
94 pid_ns->hide_pid = ctx->hidepid;
95 }
96
proc_fill_super(struct super_block * s,struct fs_context * fc)97 static int proc_fill_super(struct super_block *s, struct fs_context *fc)
98 {
99 struct pid_namespace *pid_ns = get_pid_ns(s->s_fs_info);
100 struct inode *root_inode;
101 int ret;
102
103 proc_apply_options(fc, pid_ns, current_user_ns());
104
105 /* User space would break if executables or devices appear on proc */
106 s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
107 s->s_flags |= SB_NODIRATIME | SB_NOSUID | SB_NOEXEC;
108 s->s_blocksize = 1024;
109 s->s_blocksize_bits = 10;
110 s->s_magic = PROC_SUPER_MAGIC;
111 s->s_op = &proc_sops;
112 s->s_time_gran = 1;
113
114 /*
115 * procfs isn't actually a stacking filesystem; however, there is
116 * too much magic going on inside it to permit stacking things on
117 * top of it
118 */
119 s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
120
121 /* procfs dentries and inodes don't require IO to create */
122 s->s_shrink.seeks = 0;
123
124 pde_get(&proc_root);
125 root_inode = proc_get_inode(s, &proc_root);
126 if (!root_inode) {
127 pr_err("proc_fill_super: get root inode failed\n");
128 return -ENOMEM;
129 }
130
131 s->s_root = d_make_root(root_inode);
132 if (!s->s_root) {
133 pr_err("proc_fill_super: allocate dentry failed\n");
134 return -ENOMEM;
135 }
136
137 ret = proc_setup_self(s);
138 if (ret) {
139 return ret;
140 }
141 return proc_setup_thread_self(s);
142 }
143
proc_reconfigure(struct fs_context * fc)144 static int proc_reconfigure(struct fs_context *fc)
145 {
146 struct super_block *sb = fc->root->d_sb;
147 struct pid_namespace *pid = sb->s_fs_info;
148
149 sync_filesystem(sb);
150
151 proc_apply_options(fc, pid, current_user_ns());
152 return 0;
153 }
154
proc_get_tree(struct fs_context * fc)155 static int proc_get_tree(struct fs_context *fc)
156 {
157 struct proc_fs_context *ctx = fc->fs_private;
158
159 proc_apply_options(fc, ctx->pid_ns, current_user_ns());
160 return get_tree_keyed(fc, proc_fill_super, ctx->pid_ns);
161 }
162
proc_fs_context_free(struct fs_context * fc)163 static 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
171 static 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
proc_init_fs_context(struct fs_context * fc)178 static 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
proc_kill_sb(struct super_block * sb)194 static 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
207 static 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
proc_root_init(void)215 void __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_uid_init();
225 proc_mkdir("fs", NULL);
226 proc_mkdir("driver", NULL);
227 proc_create_mount_point("fs/nfsd"); /* somewhere for the nfsd filesystem to be mounted */
228 #if defined(CONFIG_SUN_OPENPROMFS) || defined(CONFIG_SUN_OPENPROMFS_MODULE)
229 /* just give it a mountpoint */
230 proc_create_mount_point("openprom");
231 #endif
232 proc_tty_init();
233 proc_mkdir("bus", NULL);
234 proc_sys_init();
235
236 register_filesystem(&proc_fs_type);
237 }
238
proc_root_getattr(const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)239 static int proc_root_getattr(const struct path *path, struct kstat *stat,
240 u32 request_mask, unsigned int query_flags)
241 {
242 generic_fillattr(d_inode(path->dentry), stat);
243 stat->nlink = proc_root.nlink + nr_processes();
244 return 0;
245 }
246
proc_root_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)247 static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentry, unsigned int flags)
248 {
249 if (!proc_pid_lookup(dentry, flags))
250 return NULL;
251
252 return proc_lookup(dir, dentry, flags);
253 }
254
proc_root_readdir(struct file * file,struct dir_context * ctx)255 static int proc_root_readdir(struct file *file, struct dir_context *ctx)
256 {
257 if (ctx->pos < FIRST_PROCESS_ENTRY) {
258 int error = proc_readdir(file, ctx);
259 if (unlikely(error <= 0))
260 return error;
261 ctx->pos = FIRST_PROCESS_ENTRY;
262 }
263
264 return proc_pid_readdir(file, ctx);
265 }
266
267 /*
268 * The root /proc directory is special, as it has the
269 * <pid> directories. Thus we don't use the generic
270 * directory handling functions for that..
271 */
272 static const struct file_operations proc_root_operations = {
273 .read = generic_read_dir,
274 .iterate_shared = proc_root_readdir,
275 .llseek = generic_file_llseek,
276 };
277
278 /*
279 * proc root can do almost nothing..
280 */
281 static const struct inode_operations proc_root_inode_operations = {
282 .lookup = proc_root_lookup,
283 .getattr = proc_root_getattr,
284 };
285
286 /*
287 * This is the root "inode" in the /proc tree..
288 */
289 struct proc_dir_entry proc_root = {
290 .low_ino = PROC_ROOT_INO,
291 .namelen = 5,
292 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
293 .nlink = 2,
294 .refcnt = REFCOUNT_INIT(1),
295 .proc_iops = &proc_root_inode_operations,
296 .proc_fops = &proc_root_operations,
297 .parent = &proc_root,
298 .subdir = RB_ROOT,
299 .name = "/proc",
300 };
301
pid_ns_prepare_proc(struct pid_namespace * ns)302 int pid_ns_prepare_proc(struct pid_namespace *ns)
303 {
304 struct proc_fs_context *ctx;
305 struct fs_context *fc;
306 struct vfsmount *mnt;
307
308 fc = fs_context_for_mount(&proc_fs_type, SB_KERNMOUNT);
309 if (IS_ERR(fc))
310 return PTR_ERR(fc);
311
312 if (fc->user_ns != ns->user_ns) {
313 put_user_ns(fc->user_ns);
314 fc->user_ns = get_user_ns(ns->user_ns);
315 }
316
317 ctx = fc->fs_private;
318 if (ctx->pid_ns != ns) {
319 put_pid_ns(ctx->pid_ns);
320 get_pid_ns(ns);
321 ctx->pid_ns = ns;
322 }
323
324 mnt = fc_mount(fc);
325 put_fs_context(fc);
326 if (IS_ERR(mnt))
327 return PTR_ERR(mnt);
328
329 ns->proc_mnt = mnt;
330 return 0;
331 }
332
pid_ns_release_proc(struct pid_namespace * ns)333 void pid_ns_release_proc(struct pid_namespace *ns)
334 {
335 kern_unmount(ns->proc_mnt);
336 }
337