xref: /linux/fs/proc/proc_sysctl.c (revision 93d546399c2b7d66a54d5fbd5eee17de19246bf6)
1 /*
2  * /proc/sys support
3  */
4 #include <linux/init.h>
5 #include <linux/sysctl.h>
6 #include <linux/proc_fs.h>
7 #include <linux/security.h>
8 #include "internal.h"
9 
10 static struct dentry_operations proc_sys_dentry_operations;
11 static const struct file_operations proc_sys_file_operations;
12 static const struct inode_operations proc_sys_inode_operations;
13 static const struct file_operations proc_sys_dir_file_operations;
14 static const struct inode_operations proc_sys_dir_operations;
15 
16 static struct inode *proc_sys_make_inode(struct super_block *sb,
17 		struct ctl_table_header *head, struct ctl_table *table)
18 {
19 	struct inode *inode;
20 	struct proc_inode *ei;
21 
22 	inode = new_inode(sb);
23 	if (!inode)
24 		goto out;
25 
26 	sysctl_head_get(head);
27 	ei = PROC_I(inode);
28 	ei->sysctl = head;
29 	ei->sysctl_entry = table;
30 
31 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
32 	inode->i_flags |= S_PRIVATE; /* tell selinux to ignore this inode */
33 	inode->i_mode = table->mode;
34 	inode->i_uid = inode->i_gid = 0;
35 	if (!table->child) {
36 		inode->i_mode |= S_IFREG;
37 		inode->i_op = &proc_sys_inode_operations;
38 		inode->i_fop = &proc_sys_file_operations;
39 	} else {
40 		inode->i_mode |= S_IFDIR;
41 		inode->i_nlink = 0;
42 		inode->i_op = &proc_sys_dir_operations;
43 		inode->i_fop = &proc_sys_dir_file_operations;
44 	}
45 out:
46 	return inode;
47 }
48 
49 static struct ctl_table *find_in_table(struct ctl_table *p, struct qstr *name)
50 {
51 	int len;
52 	for ( ; p->ctl_name || p->procname; p++) {
53 
54 		if (!p->procname)
55 			continue;
56 
57 		len = strlen(p->procname);
58 		if (len != name->len)
59 			continue;
60 
61 		if (memcmp(p->procname, name->name, len) != 0)
62 			continue;
63 
64 		/* I have a match */
65 		return p;
66 	}
67 	return NULL;
68 }
69 
70 static struct ctl_table_header *grab_header(struct inode *inode)
71 {
72 	if (PROC_I(inode)->sysctl)
73 		return sysctl_head_grab(PROC_I(inode)->sysctl);
74 	else
75 		return sysctl_head_next(NULL);
76 }
77 
78 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
79 					struct nameidata *nd)
80 {
81 	struct ctl_table_header *head = grab_header(dir);
82 	struct ctl_table *table = PROC_I(dir)->sysctl_entry;
83 	struct ctl_table_header *h = NULL;
84 	struct qstr *name = &dentry->d_name;
85 	struct ctl_table *p;
86 	struct inode *inode;
87 	struct dentry *err = ERR_PTR(-ENOENT);
88 
89 	if (IS_ERR(head))
90 		return ERR_CAST(head);
91 
92 	if (table && !table->child) {
93 		WARN_ON(1);
94 		goto out;
95 	}
96 
97 	table = table ? table->child : head->ctl_table;
98 
99 	p = find_in_table(table, name);
100 	if (!p) {
101 		for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
102 			if (h->attached_to != table)
103 				continue;
104 			p = find_in_table(h->attached_by, name);
105 			if (p)
106 				break;
107 		}
108 	}
109 
110 	if (!p)
111 		goto out;
112 
113 	err = ERR_PTR(-ENOMEM);
114 	inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
115 	if (h)
116 		sysctl_head_finish(h);
117 
118 	if (!inode)
119 		goto out;
120 
121 	err = NULL;
122 	dentry->d_op = &proc_sys_dentry_operations;
123 	d_add(dentry, inode);
124 
125 out:
126 	sysctl_head_finish(head);
127 	return err;
128 }
129 
130 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
131 		size_t count, loff_t *ppos, int write)
132 {
133 	struct inode *inode = filp->f_path.dentry->d_inode;
134 	struct ctl_table_header *head = grab_header(inode);
135 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
136 	ssize_t error;
137 	size_t res;
138 
139 	if (IS_ERR(head))
140 		return PTR_ERR(head);
141 
142 	/*
143 	 * At this point we know that the sysctl was not unregistered
144 	 * and won't be until we finish.
145 	 */
146 	error = -EPERM;
147 	if (sysctl_perm(head->root, table, write ? MAY_WRITE : MAY_READ))
148 		goto out;
149 
150 	/* if that can happen at all, it should be -EINVAL, not -EISDIR */
151 	error = -EINVAL;
152 	if (!table->proc_handler)
153 		goto out;
154 
155 	/* careful: calling conventions are nasty here */
156 	res = count;
157 	error = table->proc_handler(table, write, filp, buf, &res, ppos);
158 	if (!error)
159 		error = res;
160 out:
161 	sysctl_head_finish(head);
162 
163 	return error;
164 }
165 
166 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
167 				size_t count, loff_t *ppos)
168 {
169 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
170 }
171 
172 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
173 				size_t count, loff_t *ppos)
174 {
175 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
176 }
177 
178 
179 static int proc_sys_fill_cache(struct file *filp, void *dirent,
180 				filldir_t filldir,
181 				struct ctl_table_header *head,
182 				struct ctl_table *table)
183 {
184 	struct dentry *child, *dir = filp->f_path.dentry;
185 	struct inode *inode;
186 	struct qstr qname;
187 	ino_t ino = 0;
188 	unsigned type = DT_UNKNOWN;
189 
190 	qname.name = table->procname;
191 	qname.len  = strlen(table->procname);
192 	qname.hash = full_name_hash(qname.name, qname.len);
193 
194 	child = d_lookup(dir, &qname);
195 	if (!child) {
196 		child = d_alloc(dir, &qname);
197 		if (child) {
198 			inode = proc_sys_make_inode(dir->d_sb, head, table);
199 			if (!inode) {
200 				dput(child);
201 				return -ENOMEM;
202 			} else {
203 				child->d_op = &proc_sys_dentry_operations;
204 				d_add(child, inode);
205 			}
206 		} else {
207 			return -ENOMEM;
208 		}
209 	}
210 	inode = child->d_inode;
211 	ino  = inode->i_ino;
212 	type = inode->i_mode >> 12;
213 	dput(child);
214 	return !!filldir(dirent, qname.name, qname.len, filp->f_pos, ino, type);
215 }
216 
217 static int scan(struct ctl_table_header *head, ctl_table *table,
218 		unsigned long *pos, struct file *file,
219 		void *dirent, filldir_t filldir)
220 {
221 
222 	for (; table->ctl_name || table->procname; table++, (*pos)++) {
223 		int res;
224 
225 		/* Can't do anything without a proc name */
226 		if (!table->procname)
227 			continue;
228 
229 		if (*pos < file->f_pos)
230 			continue;
231 
232 		res = proc_sys_fill_cache(file, dirent, filldir, head, table);
233 		if (res)
234 			return res;
235 
236 		file->f_pos = *pos + 1;
237 	}
238 	return 0;
239 }
240 
241 static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir)
242 {
243 	struct dentry *dentry = filp->f_path.dentry;
244 	struct inode *inode = dentry->d_inode;
245 	struct ctl_table_header *head = grab_header(inode);
246 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
247 	struct ctl_table_header *h = NULL;
248 	unsigned long pos;
249 	int ret = -EINVAL;
250 
251 	if (IS_ERR(head))
252 		return PTR_ERR(head);
253 
254 	if (table && !table->child) {
255 		WARN_ON(1);
256 		goto out;
257 	}
258 
259 	table = table ? table->child : head->ctl_table;
260 
261 	ret = 0;
262 	/* Avoid a switch here: arm builds fail with missing __cmpdi2 */
263 	if (filp->f_pos == 0) {
264 		if (filldir(dirent, ".", 1, filp->f_pos,
265 				inode->i_ino, DT_DIR) < 0)
266 			goto out;
267 		filp->f_pos++;
268 	}
269 	if (filp->f_pos == 1) {
270 		if (filldir(dirent, "..", 2, filp->f_pos,
271 				parent_ino(dentry), DT_DIR) < 0)
272 			goto out;
273 		filp->f_pos++;
274 	}
275 	pos = 2;
276 
277 	ret = scan(head, table, &pos, filp, dirent, filldir);
278 	if (ret)
279 		goto out;
280 
281 	for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
282 		if (h->attached_to != table)
283 			continue;
284 		ret = scan(h, h->attached_by, &pos, filp, dirent, filldir);
285 		if (ret) {
286 			sysctl_head_finish(h);
287 			break;
288 		}
289 	}
290 	ret = 1;
291 out:
292 	sysctl_head_finish(head);
293 	return ret;
294 }
295 
296 static int proc_sys_permission(struct inode *inode, int mask)
297 {
298 	/*
299 	 * sysctl entries that are not writeable,
300 	 * are _NOT_ writeable, capabilities or not.
301 	 */
302 	struct ctl_table_header *head;
303 	struct ctl_table *table;
304 	int error;
305 
306 	/* Executable files are not allowed under /proc/sys/ */
307 	if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
308 		return -EACCES;
309 
310 	head = grab_header(inode);
311 	if (IS_ERR(head))
312 		return PTR_ERR(head);
313 
314 	table = PROC_I(inode)->sysctl_entry;
315 	if (!table) /* global root - r-xr-xr-x */
316 		error = mask & MAY_WRITE ? -EACCES : 0;
317 	else /* Use the permissions on the sysctl table entry */
318 		error = sysctl_perm(head->root, table, mask);
319 
320 	sysctl_head_finish(head);
321 	return error;
322 }
323 
324 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
325 {
326 	struct inode *inode = dentry->d_inode;
327 	int error;
328 
329 	if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
330 		return -EPERM;
331 
332 	error = inode_change_ok(inode, attr);
333 	if (!error)
334 		error = inode_setattr(inode, attr);
335 
336 	return error;
337 }
338 
339 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
340 {
341 	struct inode *inode = dentry->d_inode;
342 	struct ctl_table_header *head = grab_header(inode);
343 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
344 
345 	if (IS_ERR(head))
346 		return PTR_ERR(head);
347 
348 	generic_fillattr(inode, stat);
349 	if (table)
350 		stat->mode = (stat->mode & S_IFMT) | table->mode;
351 
352 	sysctl_head_finish(head);
353 	return 0;
354 }
355 
356 static const struct file_operations proc_sys_file_operations = {
357 	.read		= proc_sys_read,
358 	.write		= proc_sys_write,
359 };
360 
361 static const struct file_operations proc_sys_dir_file_operations = {
362 	.readdir	= proc_sys_readdir,
363 	.llseek		= generic_file_llseek,
364 };
365 
366 static const struct inode_operations proc_sys_inode_operations = {
367 	.permission	= proc_sys_permission,
368 	.setattr	= proc_sys_setattr,
369 	.getattr	= proc_sys_getattr,
370 };
371 
372 static const struct inode_operations proc_sys_dir_operations = {
373 	.lookup		= proc_sys_lookup,
374 	.permission	= proc_sys_permission,
375 	.setattr	= proc_sys_setattr,
376 	.getattr	= proc_sys_getattr,
377 };
378 
379 static int proc_sys_revalidate(struct dentry *dentry, struct nameidata *nd)
380 {
381 	return !PROC_I(dentry->d_inode)->sysctl->unregistering;
382 }
383 
384 static int proc_sys_delete(struct dentry *dentry)
385 {
386 	return !!PROC_I(dentry->d_inode)->sysctl->unregistering;
387 }
388 
389 static int proc_sys_compare(struct dentry *dir, struct qstr *qstr,
390 			    struct qstr *name)
391 {
392 	struct dentry *dentry = container_of(qstr, struct dentry, d_name);
393 	if (qstr->len != name->len)
394 		return 1;
395 	if (memcmp(qstr->name, name->name, name->len))
396 		return 1;
397 	return !sysctl_is_seen(PROC_I(dentry->d_inode)->sysctl);
398 }
399 
400 static struct dentry_operations proc_sys_dentry_operations = {
401 	.d_revalidate	= proc_sys_revalidate,
402 	.d_delete	= proc_sys_delete,
403 	.d_compare	= proc_sys_compare,
404 };
405 
406 int __init proc_sys_init(void)
407 {
408 	struct proc_dir_entry *proc_sys_root;
409 
410 	proc_sys_root = proc_mkdir("sys", NULL);
411 	proc_sys_root->proc_iops = &proc_sys_dir_operations;
412 	proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
413 	proc_sys_root->nlink = 0;
414 	return 0;
415 }
416