xref: /openbmc/linux/fs/proc/proc_sysctl.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
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 <linux/namei.h>
9 #include "internal.h"
10 
11 static const struct dentry_operations proc_sys_dentry_operations;
12 static const struct file_operations proc_sys_file_operations;
13 static const struct inode_operations proc_sys_inode_operations;
14 static const struct file_operations proc_sys_dir_file_operations;
15 static const struct inode_operations proc_sys_dir_operations;
16 
17 static struct inode *proc_sys_make_inode(struct super_block *sb,
18 		struct ctl_table_header *head, struct ctl_table *table)
19 {
20 	struct inode *inode;
21 	struct proc_inode *ei;
22 
23 	inode = new_inode(sb);
24 	if (!inode)
25 		goto out;
26 
27 	inode->i_ino = get_next_ino();
28 
29 	sysctl_head_get(head);
30 	ei = PROC_I(inode);
31 	ei->sysctl = head;
32 	ei->sysctl_entry = table;
33 
34 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
35 	inode->i_flags |= S_PRIVATE; /* tell selinux to ignore this inode */
36 	inode->i_mode = table->mode;
37 	if (!table->child) {
38 		inode->i_mode |= S_IFREG;
39 		inode->i_op = &proc_sys_inode_operations;
40 		inode->i_fop = &proc_sys_file_operations;
41 	} else {
42 		inode->i_mode |= S_IFDIR;
43 		inode->i_nlink = 0;
44 		inode->i_op = &proc_sys_dir_operations;
45 		inode->i_fop = &proc_sys_dir_file_operations;
46 	}
47 out:
48 	return inode;
49 }
50 
51 static struct ctl_table *find_in_table(struct ctl_table *p, struct qstr *name)
52 {
53 	int len;
54 	for ( ; p->procname; p++) {
55 
56 		if (!p->procname)
57 			continue;
58 
59 		len = strlen(p->procname);
60 		if (len != name->len)
61 			continue;
62 
63 		if (memcmp(p->procname, name->name, len) != 0)
64 			continue;
65 
66 		/* I have a match */
67 		return p;
68 	}
69 	return NULL;
70 }
71 
72 static struct ctl_table_header *grab_header(struct inode *inode)
73 {
74 	if (PROC_I(inode)->sysctl)
75 		return sysctl_head_grab(PROC_I(inode)->sysctl);
76 	else
77 		return sysctl_head_next(NULL);
78 }
79 
80 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
81 					struct nameidata *nd)
82 {
83 	struct ctl_table_header *head = grab_header(dir);
84 	struct ctl_table *table = PROC_I(dir)->sysctl_entry;
85 	struct ctl_table_header *h = NULL;
86 	struct qstr *name = &dentry->d_name;
87 	struct ctl_table *p;
88 	struct inode *inode;
89 	struct dentry *err = ERR_PTR(-ENOENT);
90 
91 	if (IS_ERR(head))
92 		return ERR_CAST(head);
93 
94 	if (table && !table->child) {
95 		WARN_ON(1);
96 		goto out;
97 	}
98 
99 	table = table ? table->child : head->ctl_table;
100 
101 	p = find_in_table(table, name);
102 	if (!p) {
103 		for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
104 			if (h->attached_to != table)
105 				continue;
106 			p = find_in_table(h->attached_by, name);
107 			if (p)
108 				break;
109 		}
110 	}
111 
112 	if (!p)
113 		goto out;
114 
115 	err = ERR_PTR(-ENOMEM);
116 	inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
117 	if (h)
118 		sysctl_head_finish(h);
119 
120 	if (!inode)
121 		goto out;
122 
123 	err = NULL;
124 	d_set_d_op(dentry, &proc_sys_dentry_operations);
125 	d_add(dentry, inode);
126 
127 out:
128 	sysctl_head_finish(head);
129 	return err;
130 }
131 
132 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
133 		size_t count, loff_t *ppos, int write)
134 {
135 	struct inode *inode = filp->f_path.dentry->d_inode;
136 	struct ctl_table_header *head = grab_header(inode);
137 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
138 	ssize_t error;
139 	size_t res;
140 
141 	if (IS_ERR(head))
142 		return PTR_ERR(head);
143 
144 	/*
145 	 * At this point we know that the sysctl was not unregistered
146 	 * and won't be until we finish.
147 	 */
148 	error = -EPERM;
149 	if (sysctl_perm(head->root, table, write ? MAY_WRITE : MAY_READ))
150 		goto out;
151 
152 	/* if that can happen at all, it should be -EINVAL, not -EISDIR */
153 	error = -EINVAL;
154 	if (!table->proc_handler)
155 		goto out;
156 
157 	/* careful: calling conventions are nasty here */
158 	res = count;
159 	error = table->proc_handler(table, write, buf, &res, ppos);
160 	if (!error)
161 		error = res;
162 out:
163 	sysctl_head_finish(head);
164 
165 	return error;
166 }
167 
168 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
169 				size_t count, loff_t *ppos)
170 {
171 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
172 }
173 
174 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
175 				size_t count, loff_t *ppos)
176 {
177 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
178 }
179 
180 
181 static int proc_sys_fill_cache(struct file *filp, void *dirent,
182 				filldir_t filldir,
183 				struct ctl_table_header *head,
184 				struct ctl_table *table)
185 {
186 	struct dentry *child, *dir = filp->f_path.dentry;
187 	struct inode *inode;
188 	struct qstr qname;
189 	ino_t ino = 0;
190 	unsigned type = DT_UNKNOWN;
191 
192 	qname.name = table->procname;
193 	qname.len  = strlen(table->procname);
194 	qname.hash = full_name_hash(qname.name, qname.len);
195 
196 	child = d_lookup(dir, &qname);
197 	if (!child) {
198 		child = d_alloc(dir, &qname);
199 		if (child) {
200 			inode = proc_sys_make_inode(dir->d_sb, head, table);
201 			if (!inode) {
202 				dput(child);
203 				return -ENOMEM;
204 			} else {
205 				d_set_d_op(child, &proc_sys_dentry_operations);
206 				d_add(child, inode);
207 			}
208 		} else {
209 			return -ENOMEM;
210 		}
211 	}
212 	inode = child->d_inode;
213 	ino  = inode->i_ino;
214 	type = inode->i_mode >> 12;
215 	dput(child);
216 	return !!filldir(dirent, qname.name, qname.len, filp->f_pos, ino, type);
217 }
218 
219 static int scan(struct ctl_table_header *head, ctl_table *table,
220 		unsigned long *pos, struct file *file,
221 		void *dirent, filldir_t filldir)
222 {
223 
224 	for (; table->procname; table++, (*pos)++) {
225 		int res;
226 
227 		/* Can't do anything without a proc name */
228 		if (!table->procname)
229 			continue;
230 
231 		if (*pos < file->f_pos)
232 			continue;
233 
234 		res = proc_sys_fill_cache(file, dirent, filldir, head, table);
235 		if (res)
236 			return res;
237 
238 		file->f_pos = *pos + 1;
239 	}
240 	return 0;
241 }
242 
243 static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir)
244 {
245 	struct dentry *dentry = filp->f_path.dentry;
246 	struct inode *inode = dentry->d_inode;
247 	struct ctl_table_header *head = grab_header(inode);
248 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
249 	struct ctl_table_header *h = NULL;
250 	unsigned long pos;
251 	int ret = -EINVAL;
252 
253 	if (IS_ERR(head))
254 		return PTR_ERR(head);
255 
256 	if (table && !table->child) {
257 		WARN_ON(1);
258 		goto out;
259 	}
260 
261 	table = table ? table->child : head->ctl_table;
262 
263 	ret = 0;
264 	/* Avoid a switch here: arm builds fail with missing __cmpdi2 */
265 	if (filp->f_pos == 0) {
266 		if (filldir(dirent, ".", 1, filp->f_pos,
267 				inode->i_ino, DT_DIR) < 0)
268 			goto out;
269 		filp->f_pos++;
270 	}
271 	if (filp->f_pos == 1) {
272 		if (filldir(dirent, "..", 2, filp->f_pos,
273 				parent_ino(dentry), DT_DIR) < 0)
274 			goto out;
275 		filp->f_pos++;
276 	}
277 	pos = 2;
278 
279 	ret = scan(head, table, &pos, filp, dirent, filldir);
280 	if (ret)
281 		goto out;
282 
283 	for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
284 		if (h->attached_to != table)
285 			continue;
286 		ret = scan(h, h->attached_by, &pos, filp, dirent, filldir);
287 		if (ret) {
288 			sysctl_head_finish(h);
289 			break;
290 		}
291 	}
292 	ret = 1;
293 out:
294 	sysctl_head_finish(head);
295 	return ret;
296 }
297 
298 static int proc_sys_permission(struct inode *inode, int mask,unsigned int flags)
299 {
300 	/*
301 	 * sysctl entries that are not writeable,
302 	 * are _NOT_ writeable, capabilities or not.
303 	 */
304 	struct ctl_table_header *head;
305 	struct ctl_table *table;
306 	int error;
307 
308 	if (flags & IPERM_FLAG_RCU)
309 		return -ECHILD;
310 
311 	/* Executable files are not allowed under /proc/sys/ */
312 	if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
313 		return -EACCES;
314 
315 	head = grab_header(inode);
316 	if (IS_ERR(head))
317 		return PTR_ERR(head);
318 
319 	table = PROC_I(inode)->sysctl_entry;
320 	if (!table) /* global root - r-xr-xr-x */
321 		error = mask & MAY_WRITE ? -EACCES : 0;
322 	else /* Use the permissions on the sysctl table entry */
323 		error = sysctl_perm(head->root, table, mask);
324 
325 	sysctl_head_finish(head);
326 	return error;
327 }
328 
329 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
330 {
331 	struct inode *inode = dentry->d_inode;
332 	int error;
333 
334 	if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
335 		return -EPERM;
336 
337 	error = inode_change_ok(inode, attr);
338 	if (error)
339 		return error;
340 
341 	if ((attr->ia_valid & ATTR_SIZE) &&
342 	    attr->ia_size != i_size_read(inode)) {
343 		error = vmtruncate(inode, attr->ia_size);
344 		if (error)
345 			return error;
346 	}
347 
348 	setattr_copy(inode, attr);
349 	mark_inode_dirty(inode);
350 	return 0;
351 }
352 
353 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
354 {
355 	struct inode *inode = dentry->d_inode;
356 	struct ctl_table_header *head = grab_header(inode);
357 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
358 
359 	if (IS_ERR(head))
360 		return PTR_ERR(head);
361 
362 	generic_fillattr(inode, stat);
363 	if (table)
364 		stat->mode = (stat->mode & S_IFMT) | table->mode;
365 
366 	sysctl_head_finish(head);
367 	return 0;
368 }
369 
370 static const struct file_operations proc_sys_file_operations = {
371 	.read		= proc_sys_read,
372 	.write		= proc_sys_write,
373 	.llseek		= default_llseek,
374 };
375 
376 static const struct file_operations proc_sys_dir_file_operations = {
377 	.readdir	= proc_sys_readdir,
378 	.llseek		= generic_file_llseek,
379 };
380 
381 static const struct inode_operations proc_sys_inode_operations = {
382 	.permission	= proc_sys_permission,
383 	.setattr	= proc_sys_setattr,
384 	.getattr	= proc_sys_getattr,
385 };
386 
387 static const struct inode_operations proc_sys_dir_operations = {
388 	.lookup		= proc_sys_lookup,
389 	.permission	= proc_sys_permission,
390 	.setattr	= proc_sys_setattr,
391 	.getattr	= proc_sys_getattr,
392 };
393 
394 static int proc_sys_revalidate(struct dentry *dentry, struct nameidata *nd)
395 {
396 	if (nd->flags & LOOKUP_RCU)
397 		return -ECHILD;
398 	return !PROC_I(dentry->d_inode)->sysctl->unregistering;
399 }
400 
401 static int proc_sys_delete(const struct dentry *dentry)
402 {
403 	return !!PROC_I(dentry->d_inode)->sysctl->unregistering;
404 }
405 
406 static int proc_sys_compare(const struct dentry *parent,
407 		const struct inode *pinode,
408 		const struct dentry *dentry, const struct inode *inode,
409 		unsigned int len, const char *str, const struct qstr *name)
410 {
411 	/* Although proc doesn't have negative dentries, rcu-walk means
412 	 * that inode here can be NULL */
413 	if (!inode)
414 		return 0;
415 	if (name->len != len)
416 		return 1;
417 	if (memcmp(name->name, str, len))
418 		return 1;
419 	return !sysctl_is_seen(PROC_I(inode)->sysctl);
420 }
421 
422 static const struct dentry_operations proc_sys_dentry_operations = {
423 	.d_revalidate	= proc_sys_revalidate,
424 	.d_delete	= proc_sys_delete,
425 	.d_compare	= proc_sys_compare,
426 };
427 
428 int __init proc_sys_init(void)
429 {
430 	struct proc_dir_entry *proc_sys_root;
431 
432 	proc_sys_root = proc_mkdir("sys", NULL);
433 	proc_sys_root->proc_iops = &proc_sys_dir_operations;
434 	proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
435 	proc_sys_root->nlink = 0;
436 	return 0;
437 }
438