xref: /openbmc/linux/fs/proc/proc_sysctl.c (revision b627b4ed)
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 const 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 	if (!table->child) {
35 		inode->i_mode |= S_IFREG;
36 		inode->i_op = &proc_sys_inode_operations;
37 		inode->i_fop = &proc_sys_file_operations;
38 	} else {
39 		inode->i_mode |= S_IFDIR;
40 		inode->i_nlink = 0;
41 		inode->i_op = &proc_sys_dir_operations;
42 		inode->i_fop = &proc_sys_dir_file_operations;
43 	}
44 out:
45 	return inode;
46 }
47 
48 static struct ctl_table *find_in_table(struct ctl_table *p, struct qstr *name)
49 {
50 	int len;
51 	for ( ; p->ctl_name || p->procname; p++) {
52 
53 		if (!p->procname)
54 			continue;
55 
56 		len = strlen(p->procname);
57 		if (len != name->len)
58 			continue;
59 
60 		if (memcmp(p->procname, name->name, len) != 0)
61 			continue;
62 
63 		/* I have a match */
64 		return p;
65 	}
66 	return NULL;
67 }
68 
69 static struct ctl_table_header *grab_header(struct inode *inode)
70 {
71 	if (PROC_I(inode)->sysctl)
72 		return sysctl_head_grab(PROC_I(inode)->sysctl);
73 	else
74 		return sysctl_head_next(NULL);
75 }
76 
77 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
78 					struct nameidata *nd)
79 {
80 	struct ctl_table_header *head = grab_header(dir);
81 	struct ctl_table *table = PROC_I(dir)->sysctl_entry;
82 	struct ctl_table_header *h = NULL;
83 	struct qstr *name = &dentry->d_name;
84 	struct ctl_table *p;
85 	struct inode *inode;
86 	struct dentry *err = ERR_PTR(-ENOENT);
87 
88 	if (IS_ERR(head))
89 		return ERR_CAST(head);
90 
91 	if (table && !table->child) {
92 		WARN_ON(1);
93 		goto out;
94 	}
95 
96 	table = table ? table->child : head->ctl_table;
97 
98 	p = find_in_table(table, name);
99 	if (!p) {
100 		for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
101 			if (h->attached_to != table)
102 				continue;
103 			p = find_in_table(h->attached_by, name);
104 			if (p)
105 				break;
106 		}
107 	}
108 
109 	if (!p)
110 		goto out;
111 
112 	err = ERR_PTR(-ENOMEM);
113 	inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
114 	if (h)
115 		sysctl_head_finish(h);
116 
117 	if (!inode)
118 		goto out;
119 
120 	err = NULL;
121 	dentry->d_op = &proc_sys_dentry_operations;
122 	d_add(dentry, inode);
123 
124 out:
125 	sysctl_head_finish(head);
126 	return err;
127 }
128 
129 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
130 		size_t count, loff_t *ppos, int write)
131 {
132 	struct inode *inode = filp->f_path.dentry->d_inode;
133 	struct ctl_table_header *head = grab_header(inode);
134 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
135 	ssize_t error;
136 	size_t res;
137 
138 	if (IS_ERR(head))
139 		return PTR_ERR(head);
140 
141 	/*
142 	 * At this point we know that the sysctl was not unregistered
143 	 * and won't be until we finish.
144 	 */
145 	error = -EPERM;
146 	if (sysctl_perm(head->root, table, write ? MAY_WRITE : MAY_READ))
147 		goto out;
148 
149 	/* if that can happen at all, it should be -EINVAL, not -EISDIR */
150 	error = -EINVAL;
151 	if (!table->proc_handler)
152 		goto out;
153 
154 	/* careful: calling conventions are nasty here */
155 	res = count;
156 	error = table->proc_handler(table, write, filp, buf, &res, ppos);
157 	if (!error)
158 		error = res;
159 out:
160 	sysctl_head_finish(head);
161 
162 	return error;
163 }
164 
165 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
166 				size_t count, loff_t *ppos)
167 {
168 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
169 }
170 
171 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
172 				size_t count, loff_t *ppos)
173 {
174 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
175 }
176 
177 
178 static int proc_sys_fill_cache(struct file *filp, void *dirent,
179 				filldir_t filldir,
180 				struct ctl_table_header *head,
181 				struct ctl_table *table)
182 {
183 	struct dentry *child, *dir = filp->f_path.dentry;
184 	struct inode *inode;
185 	struct qstr qname;
186 	ino_t ino = 0;
187 	unsigned type = DT_UNKNOWN;
188 
189 	qname.name = table->procname;
190 	qname.len  = strlen(table->procname);
191 	qname.hash = full_name_hash(qname.name, qname.len);
192 
193 	child = d_lookup(dir, &qname);
194 	if (!child) {
195 		child = d_alloc(dir, &qname);
196 		if (child) {
197 			inode = proc_sys_make_inode(dir->d_sb, head, table);
198 			if (!inode) {
199 				dput(child);
200 				return -ENOMEM;
201 			} else {
202 				child->d_op = &proc_sys_dentry_operations;
203 				d_add(child, inode);
204 			}
205 		} else {
206 			return -ENOMEM;
207 		}
208 	}
209 	inode = child->d_inode;
210 	ino  = inode->i_ino;
211 	type = inode->i_mode >> 12;
212 	dput(child);
213 	return !!filldir(dirent, qname.name, qname.len, filp->f_pos, ino, type);
214 }
215 
216 static int scan(struct ctl_table_header *head, ctl_table *table,
217 		unsigned long *pos, struct file *file,
218 		void *dirent, filldir_t filldir)
219 {
220 
221 	for (; table->ctl_name || table->procname; table++, (*pos)++) {
222 		int res;
223 
224 		/* Can't do anything without a proc name */
225 		if (!table->procname)
226 			continue;
227 
228 		if (*pos < file->f_pos)
229 			continue;
230 
231 		res = proc_sys_fill_cache(file, dirent, filldir, head, table);
232 		if (res)
233 			return res;
234 
235 		file->f_pos = *pos + 1;
236 	}
237 	return 0;
238 }
239 
240 static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir)
241 {
242 	struct dentry *dentry = filp->f_path.dentry;
243 	struct inode *inode = dentry->d_inode;
244 	struct ctl_table_header *head = grab_header(inode);
245 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
246 	struct ctl_table_header *h = NULL;
247 	unsigned long pos;
248 	int ret = -EINVAL;
249 
250 	if (IS_ERR(head))
251 		return PTR_ERR(head);
252 
253 	if (table && !table->child) {
254 		WARN_ON(1);
255 		goto out;
256 	}
257 
258 	table = table ? table->child : head->ctl_table;
259 
260 	ret = 0;
261 	/* Avoid a switch here: arm builds fail with missing __cmpdi2 */
262 	if (filp->f_pos == 0) {
263 		if (filldir(dirent, ".", 1, filp->f_pos,
264 				inode->i_ino, DT_DIR) < 0)
265 			goto out;
266 		filp->f_pos++;
267 	}
268 	if (filp->f_pos == 1) {
269 		if (filldir(dirent, "..", 2, filp->f_pos,
270 				parent_ino(dentry), DT_DIR) < 0)
271 			goto out;
272 		filp->f_pos++;
273 	}
274 	pos = 2;
275 
276 	ret = scan(head, table, &pos, filp, dirent, filldir);
277 	if (ret)
278 		goto out;
279 
280 	for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
281 		if (h->attached_to != table)
282 			continue;
283 		ret = scan(h, h->attached_by, &pos, filp, dirent, filldir);
284 		if (ret) {
285 			sysctl_head_finish(h);
286 			break;
287 		}
288 	}
289 	ret = 1;
290 out:
291 	sysctl_head_finish(head);
292 	return ret;
293 }
294 
295 static int proc_sys_permission(struct inode *inode, int mask)
296 {
297 	/*
298 	 * sysctl entries that are not writeable,
299 	 * are _NOT_ writeable, capabilities or not.
300 	 */
301 	struct ctl_table_header *head;
302 	struct ctl_table *table;
303 	int error;
304 
305 	/* Executable files are not allowed under /proc/sys/ */
306 	if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
307 		return -EACCES;
308 
309 	head = grab_header(inode);
310 	if (IS_ERR(head))
311 		return PTR_ERR(head);
312 
313 	table = PROC_I(inode)->sysctl_entry;
314 	if (!table) /* global root - r-xr-xr-x */
315 		error = mask & MAY_WRITE ? -EACCES : 0;
316 	else /* Use the permissions on the sysctl table entry */
317 		error = sysctl_perm(head->root, table, mask);
318 
319 	sysctl_head_finish(head);
320 	return error;
321 }
322 
323 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
324 {
325 	struct inode *inode = dentry->d_inode;
326 	int error;
327 
328 	if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
329 		return -EPERM;
330 
331 	error = inode_change_ok(inode, attr);
332 	if (!error)
333 		error = inode_setattr(inode, attr);
334 
335 	return error;
336 }
337 
338 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
339 {
340 	struct inode *inode = dentry->d_inode;
341 	struct ctl_table_header *head = grab_header(inode);
342 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
343 
344 	if (IS_ERR(head))
345 		return PTR_ERR(head);
346 
347 	generic_fillattr(inode, stat);
348 	if (table)
349 		stat->mode = (stat->mode & S_IFMT) | table->mode;
350 
351 	sysctl_head_finish(head);
352 	return 0;
353 }
354 
355 static const struct file_operations proc_sys_file_operations = {
356 	.read		= proc_sys_read,
357 	.write		= proc_sys_write,
358 };
359 
360 static const struct file_operations proc_sys_dir_file_operations = {
361 	.readdir	= proc_sys_readdir,
362 	.llseek		= generic_file_llseek,
363 };
364 
365 static const struct inode_operations proc_sys_inode_operations = {
366 	.permission	= proc_sys_permission,
367 	.setattr	= proc_sys_setattr,
368 	.getattr	= proc_sys_getattr,
369 };
370 
371 static const struct inode_operations proc_sys_dir_operations = {
372 	.lookup		= proc_sys_lookup,
373 	.permission	= proc_sys_permission,
374 	.setattr	= proc_sys_setattr,
375 	.getattr	= proc_sys_getattr,
376 };
377 
378 static int proc_sys_revalidate(struct dentry *dentry, struct nameidata *nd)
379 {
380 	return !PROC_I(dentry->d_inode)->sysctl->unregistering;
381 }
382 
383 static int proc_sys_delete(struct dentry *dentry)
384 {
385 	return !!PROC_I(dentry->d_inode)->sysctl->unregistering;
386 }
387 
388 static int proc_sys_compare(struct dentry *dir, struct qstr *qstr,
389 			    struct qstr *name)
390 {
391 	struct dentry *dentry = container_of(qstr, struct dentry, d_name);
392 	if (qstr->len != name->len)
393 		return 1;
394 	if (memcmp(qstr->name, name->name, name->len))
395 		return 1;
396 	return !sysctl_is_seen(PROC_I(dentry->d_inode)->sysctl);
397 }
398 
399 static const struct dentry_operations proc_sys_dentry_operations = {
400 	.d_revalidate	= proc_sys_revalidate,
401 	.d_delete	= proc_sys_delete,
402 	.d_compare	= proc_sys_compare,
403 };
404 
405 int __init proc_sys_init(void)
406 {
407 	struct proc_dir_entry *proc_sys_root;
408 
409 	proc_sys_root = proc_mkdir("sys", NULL);
410 	proc_sys_root->proc_iops = &proc_sys_dir_operations;
411 	proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
412 	proc_sys_root->nlink = 0;
413 	return 0;
414 }
415