xref: /openbmc/linux/fs/proc/proc_sysctl.c (revision 10556cb2)
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_mode = table->mode;
36 	if (!table->child) {
37 		inode->i_mode |= S_IFREG;
38 		inode->i_op = &proc_sys_inode_operations;
39 		inode->i_fop = &proc_sys_file_operations;
40 	} else {
41 		inode->i_mode |= S_IFDIR;
42 		inode->i_nlink = 0;
43 		inode->i_op = &proc_sys_dir_operations;
44 		inode->i_fop = &proc_sys_dir_file_operations;
45 	}
46 out:
47 	return inode;
48 }
49 
50 static struct ctl_table *find_in_table(struct ctl_table *p, struct qstr *name)
51 {
52 	int len;
53 	for ( ; p->procname; p++) {
54 
55 		if (!p->procname)
56 			continue;
57 
58 		len = strlen(p->procname);
59 		if (len != name->len)
60 			continue;
61 
62 		if (memcmp(p->procname, name->name, len) != 0)
63 			continue;
64 
65 		/* I have a match */
66 		return p;
67 	}
68 	return NULL;
69 }
70 
71 static struct ctl_table_header *grab_header(struct inode *inode)
72 {
73 	if (PROC_I(inode)->sysctl)
74 		return sysctl_head_grab(PROC_I(inode)->sysctl);
75 	else
76 		return sysctl_head_next(NULL);
77 }
78 
79 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
80 					struct nameidata *nd)
81 {
82 	struct ctl_table_header *head = grab_header(dir);
83 	struct ctl_table *table = PROC_I(dir)->sysctl_entry;
84 	struct ctl_table_header *h = NULL;
85 	struct qstr *name = &dentry->d_name;
86 	struct ctl_table *p;
87 	struct inode *inode;
88 	struct dentry *err = ERR_PTR(-ENOENT);
89 
90 	if (IS_ERR(head))
91 		return ERR_CAST(head);
92 
93 	if (table && !table->child) {
94 		WARN_ON(1);
95 		goto out;
96 	}
97 
98 	table = table ? table->child : head->ctl_table;
99 
100 	p = find_in_table(table, name);
101 	if (!p) {
102 		for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
103 			if (h->attached_to != table)
104 				continue;
105 			p = find_in_table(h->attached_by, name);
106 			if (p)
107 				break;
108 		}
109 	}
110 
111 	if (!p)
112 		goto out;
113 
114 	err = ERR_PTR(-ENOMEM);
115 	inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
116 	if (h)
117 		sysctl_head_finish(h);
118 
119 	if (!inode)
120 		goto out;
121 
122 	err = NULL;
123 	d_set_d_op(dentry, &proc_sys_dentry_operations);
124 	d_add(dentry, inode);
125 
126 out:
127 	sysctl_head_finish(head);
128 	return err;
129 }
130 
131 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
132 		size_t count, loff_t *ppos, int write)
133 {
134 	struct inode *inode = filp->f_path.dentry->d_inode;
135 	struct ctl_table_header *head = grab_header(inode);
136 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
137 	ssize_t error;
138 	size_t res;
139 
140 	if (IS_ERR(head))
141 		return PTR_ERR(head);
142 
143 	/*
144 	 * At this point we know that the sysctl was not unregistered
145 	 * and won't be until we finish.
146 	 */
147 	error = -EPERM;
148 	if (sysctl_perm(head->root, table, write ? MAY_WRITE : MAY_READ))
149 		goto out;
150 
151 	/* if that can happen at all, it should be -EINVAL, not -EISDIR */
152 	error = -EINVAL;
153 	if (!table->proc_handler)
154 		goto out;
155 
156 	/* careful: calling conventions are nasty here */
157 	res = count;
158 	error = table->proc_handler(table, write, buf, &res, ppos);
159 	if (!error)
160 		error = res;
161 out:
162 	sysctl_head_finish(head);
163 
164 	return error;
165 }
166 
167 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
168 				size_t count, loff_t *ppos)
169 {
170 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
171 }
172 
173 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
174 				size_t count, loff_t *ppos)
175 {
176 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
177 }
178 
179 
180 static int proc_sys_fill_cache(struct file *filp, void *dirent,
181 				filldir_t filldir,
182 				struct ctl_table_header *head,
183 				struct ctl_table *table)
184 {
185 	struct dentry *child, *dir = filp->f_path.dentry;
186 	struct inode *inode;
187 	struct qstr qname;
188 	ino_t ino = 0;
189 	unsigned type = DT_UNKNOWN;
190 
191 	qname.name = table->procname;
192 	qname.len  = strlen(table->procname);
193 	qname.hash = full_name_hash(qname.name, qname.len);
194 
195 	child = d_lookup(dir, &qname);
196 	if (!child) {
197 		child = d_alloc(dir, &qname);
198 		if (child) {
199 			inode = proc_sys_make_inode(dir->d_sb, head, table);
200 			if (!inode) {
201 				dput(child);
202 				return -ENOMEM;
203 			} else {
204 				d_set_d_op(child, &proc_sys_dentry_operations);
205 				d_add(child, inode);
206 			}
207 		} else {
208 			return -ENOMEM;
209 		}
210 	}
211 	inode = child->d_inode;
212 	ino  = inode->i_ino;
213 	type = inode->i_mode >> 12;
214 	dput(child);
215 	return !!filldir(dirent, qname.name, qname.len, filp->f_pos, ino, type);
216 }
217 
218 static int scan(struct ctl_table_header *head, ctl_table *table,
219 		unsigned long *pos, struct file *file,
220 		void *dirent, filldir_t filldir)
221 {
222 
223 	for (; table->procname; table++, (*pos)++) {
224 		int res;
225 
226 		/* Can't do anything without a proc name */
227 		if (!table->procname)
228 			continue;
229 
230 		if (*pos < file->f_pos)
231 			continue;
232 
233 		res = proc_sys_fill_cache(file, dirent, filldir, head, table);
234 		if (res)
235 			return res;
236 
237 		file->f_pos = *pos + 1;
238 	}
239 	return 0;
240 }
241 
242 static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir)
243 {
244 	struct dentry *dentry = filp->f_path.dentry;
245 	struct inode *inode = dentry->d_inode;
246 	struct ctl_table_header *head = grab_header(inode);
247 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
248 	struct ctl_table_header *h = NULL;
249 	unsigned long pos;
250 	int ret = -EINVAL;
251 
252 	if (IS_ERR(head))
253 		return PTR_ERR(head);
254 
255 	if (table && !table->child) {
256 		WARN_ON(1);
257 		goto out;
258 	}
259 
260 	table = table ? table->child : head->ctl_table;
261 
262 	ret = 0;
263 	/* Avoid a switch here: arm builds fail with missing __cmpdi2 */
264 	if (filp->f_pos == 0) {
265 		if (filldir(dirent, ".", 1, filp->f_pos,
266 				inode->i_ino, DT_DIR) < 0)
267 			goto out;
268 		filp->f_pos++;
269 	}
270 	if (filp->f_pos == 1) {
271 		if (filldir(dirent, "..", 2, filp->f_pos,
272 				parent_ino(dentry), DT_DIR) < 0)
273 			goto out;
274 		filp->f_pos++;
275 	}
276 	pos = 2;
277 
278 	ret = scan(head, table, &pos, filp, dirent, filldir);
279 	if (ret)
280 		goto out;
281 
282 	for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
283 		if (h->attached_to != table)
284 			continue;
285 		ret = scan(h, h->attached_by, &pos, filp, dirent, filldir);
286 		if (ret) {
287 			sysctl_head_finish(h);
288 			break;
289 		}
290 	}
291 	ret = 1;
292 out:
293 	sysctl_head_finish(head);
294 	return ret;
295 }
296 
297 static int proc_sys_permission(struct inode *inode, int mask)
298 {
299 	/*
300 	 * sysctl entries that are not writeable,
301 	 * are _NOT_ writeable, capabilities or not.
302 	 */
303 	struct ctl_table_header *head;
304 	struct ctl_table *table;
305 	int error;
306 
307 	/* Executable files are not allowed under /proc/sys/ */
308 	if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
309 		return -EACCES;
310 
311 	head = grab_header(inode);
312 	if (IS_ERR(head))
313 		return PTR_ERR(head);
314 
315 	table = PROC_I(inode)->sysctl_entry;
316 	if (!table) /* global root - r-xr-xr-x */
317 		error = mask & MAY_WRITE ? -EACCES : 0;
318 	else /* Use the permissions on the sysctl table entry */
319 		error = sysctl_perm(head->root, table, mask & ~MAY_NOT_BLOCK);
320 
321 	sysctl_head_finish(head);
322 	return error;
323 }
324 
325 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
326 {
327 	struct inode *inode = dentry->d_inode;
328 	int error;
329 
330 	if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
331 		return -EPERM;
332 
333 	error = inode_change_ok(inode, attr);
334 	if (error)
335 		return error;
336 
337 	if ((attr->ia_valid & ATTR_SIZE) &&
338 	    attr->ia_size != i_size_read(inode)) {
339 		error = vmtruncate(inode, attr->ia_size);
340 		if (error)
341 			return error;
342 	}
343 
344 	setattr_copy(inode, attr);
345 	mark_inode_dirty(inode);
346 	return 0;
347 }
348 
349 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
350 {
351 	struct inode *inode = dentry->d_inode;
352 	struct ctl_table_header *head = grab_header(inode);
353 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
354 
355 	if (IS_ERR(head))
356 		return PTR_ERR(head);
357 
358 	generic_fillattr(inode, stat);
359 	if (table)
360 		stat->mode = (stat->mode & S_IFMT) | table->mode;
361 
362 	sysctl_head_finish(head);
363 	return 0;
364 }
365 
366 static const struct file_operations proc_sys_file_operations = {
367 	.read		= proc_sys_read,
368 	.write		= proc_sys_write,
369 	.llseek		= default_llseek,
370 };
371 
372 static const struct file_operations proc_sys_dir_file_operations = {
373 	.readdir	= proc_sys_readdir,
374 	.llseek		= generic_file_llseek,
375 };
376 
377 static const struct inode_operations proc_sys_inode_operations = {
378 	.permission	= proc_sys_permission,
379 	.setattr	= proc_sys_setattr,
380 	.getattr	= proc_sys_getattr,
381 };
382 
383 static const struct inode_operations proc_sys_dir_operations = {
384 	.lookup		= proc_sys_lookup,
385 	.permission	= proc_sys_permission,
386 	.setattr	= proc_sys_setattr,
387 	.getattr	= proc_sys_getattr,
388 };
389 
390 static int proc_sys_revalidate(struct dentry *dentry, struct nameidata *nd)
391 {
392 	if (nd->flags & LOOKUP_RCU)
393 		return -ECHILD;
394 	return !PROC_I(dentry->d_inode)->sysctl->unregistering;
395 }
396 
397 static int proc_sys_delete(const struct dentry *dentry)
398 {
399 	return !!PROC_I(dentry->d_inode)->sysctl->unregistering;
400 }
401 
402 static int proc_sys_compare(const struct dentry *parent,
403 		const struct inode *pinode,
404 		const struct dentry *dentry, const struct inode *inode,
405 		unsigned int len, const char *str, const struct qstr *name)
406 {
407 	struct ctl_table_header *head;
408 	/* Although proc doesn't have negative dentries, rcu-walk means
409 	 * that inode here can be NULL */
410 	/* AV: can it, indeed? */
411 	if (!inode)
412 		return 1;
413 	if (name->len != len)
414 		return 1;
415 	if (memcmp(name->name, str, len))
416 		return 1;
417 	head = rcu_dereference(PROC_I(inode)->sysctl);
418 	return !head || !sysctl_is_seen(head);
419 }
420 
421 static const struct dentry_operations proc_sys_dentry_operations = {
422 	.d_revalidate	= proc_sys_revalidate,
423 	.d_delete	= proc_sys_delete,
424 	.d_compare	= proc_sys_compare,
425 };
426 
427 int __init proc_sys_init(void)
428 {
429 	struct proc_dir_entry *proc_sys_root;
430 
431 	proc_sys_root = proc_mkdir("sys", NULL);
432 	proc_sys_root->proc_iops = &proc_sys_dir_operations;
433 	proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
434 	proc_sys_root->nlink = 0;
435 	return 0;
436 }
437