blob: 614c34b6d1c22214ffcdeb46a5f9e9eb63d69678 [file] [log] [blame]
Eric W. Biederman77b14db2007-02-14 00:34:12 -08001/*
2 * /proc/sys support
3 */
4
5#include <linux/sysctl.h>
6#include <linux/proc_fs.h>
7#include <linux/security.h>
8#include "internal.h"
9
10static struct dentry_operations proc_sys_dentry_operations;
11static const struct file_operations proc_sys_file_operations;
Jan Engelhardt03a44822008-02-08 04:21:19 -080012static const struct inode_operations proc_sys_inode_operations;
Eric W. Biederman77b14db2007-02-14 00:34:12 -080013
14static void proc_sys_refresh_inode(struct inode *inode, struct ctl_table *table)
15{
16 /* Refresh the cached information bits in the inode */
17 if (table) {
18 inode->i_uid = 0;
19 inode->i_gid = 0;
20 inode->i_mode = table->mode;
21 if (table->proc_handler) {
22 inode->i_mode |= S_IFREG;
23 inode->i_nlink = 1;
24 } else {
25 inode->i_mode |= S_IFDIR;
26 inode->i_nlink = 0; /* It is too hard to figure out */
27 }
28 }
29}
30
31static struct inode *proc_sys_make_inode(struct inode *dir, struct ctl_table *table)
32{
33 struct inode *inode;
34 struct proc_inode *dir_ei, *ei;
35 int depth;
36
37 inode = new_inode(dir->i_sb);
38 if (!inode)
39 goto out;
40
41 /* A directory is always one deeper than it's parent */
42 dir_ei = PROC_I(dir);
43 depth = dir_ei->fd + 1;
44
45 ei = PROC_I(inode);
46 ei->fd = depth;
47 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
48 inode->i_op = &proc_sys_inode_operations;
49 inode->i_fop = &proc_sys_file_operations;
Eric W. Biederman86a71dbd2007-02-14 00:34:16 -080050 inode->i_flags |= S_PRIVATE; /* tell selinux to ignore this inode */
Eric W. Biederman77b14db2007-02-14 00:34:12 -080051 proc_sys_refresh_inode(inode, table);
52out:
53 return inode;
54}
55
56static struct dentry *proc_sys_ancestor(struct dentry *dentry, int depth)
57{
58 for (;;) {
59 struct proc_inode *ei;
60
61 ei = PROC_I(dentry->d_inode);
62 if (ei->fd == depth)
63 break; /* found */
64
65 dentry = dentry->d_parent;
66 }
67 return dentry;
68}
69
70static struct ctl_table *proc_sys_lookup_table_one(struct ctl_table *table,
71 struct qstr *name)
72{
73 int len;
74 for ( ; table->ctl_name || table->procname; table++) {
75
76 if (!table->procname)
77 continue;
78
79 len = strlen(table->procname);
80 if (len != name->len)
81 continue;
82
83 if (memcmp(table->procname, name->name, len) != 0)
84 continue;
85
86 /* I have a match */
87 return table;
88 }
89 return NULL;
90}
91
92static struct ctl_table *proc_sys_lookup_table(struct dentry *dentry,
93 struct ctl_table *table)
94{
95 struct dentry *ancestor;
96 struct proc_inode *ei;
97 int depth, i;
98
99 ei = PROC_I(dentry->d_inode);
100 depth = ei->fd;
101
102 if (depth == 0)
103 return table;
104
105 for (i = 1; table && (i <= depth); i++) {
106 ancestor = proc_sys_ancestor(dentry, i);
107 table = proc_sys_lookup_table_one(table, &ancestor->d_name);
108 if (table)
109 table = table->child;
110 }
111 return table;
112
113}
114static struct ctl_table *proc_sys_lookup_entry(struct dentry *dparent,
115 struct qstr *name,
116 struct ctl_table *table)
117{
118 table = proc_sys_lookup_table(dparent, table);
119 if (table)
120 table = proc_sys_lookup_table_one(table, name);
121 return table;
122}
123
124static struct ctl_table *do_proc_sys_lookup(struct dentry *parent,
125 struct qstr *name,
126 struct ctl_table_header **ptr)
127{
128 struct ctl_table_header *head;
129 struct ctl_table *table = NULL;
130
131 for (head = sysctl_head_next(NULL); head;
132 head = sysctl_head_next(head)) {
133 table = proc_sys_lookup_entry(parent, name, head->ctl_table);
134 if (table)
135 break;
136 }
137 *ptr = head;
138 return table;
139}
140
141static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
142 struct nameidata *nd)
143{
144 struct ctl_table_header *head;
145 struct inode *inode;
146 struct dentry *err;
147 struct ctl_table *table;
148
149 err = ERR_PTR(-ENOENT);
150 table = do_proc_sys_lookup(dentry->d_parent, &dentry->d_name, &head);
151 if (!table)
152 goto out;
153
154 err = ERR_PTR(-ENOMEM);
155 inode = proc_sys_make_inode(dir, table);
156 if (!inode)
157 goto out;
158
159 err = NULL;
160 dentry->d_op = &proc_sys_dentry_operations;
161 d_add(dentry, inode);
162
163out:
164 sysctl_head_finish(head);
165 return err;
166}
167
168static ssize_t proc_sys_read(struct file *filp, char __user *buf,
169 size_t count, loff_t *ppos)
170{
171 struct dentry *dentry = filp->f_dentry;
172 struct ctl_table_header *head;
173 struct ctl_table *table;
David Howells2a2da532007-10-25 15:27:40 +0100174 ssize_t error;
175 size_t res;
Eric W. Biederman77b14db2007-02-14 00:34:12 -0800176
177 table = do_proc_sys_lookup(dentry->d_parent, &dentry->d_name, &head);
178 /* Has the sysctl entry disappeared on us? */
179 error = -ENOENT;
180 if (!table)
181 goto out;
182
183 /* Has the sysctl entry been replaced by a directory? */
184 error = -EISDIR;
185 if (!table->proc_handler)
186 goto out;
187
188 /*
189 * At this point we know that the sysctl was not unregistered
190 * and won't be until we finish.
191 */
192 error = -EPERM;
193 if (sysctl_perm(table, MAY_READ))
194 goto out;
195
196 /* careful: calling conventions are nasty here */
197 res = count;
198 error = table->proc_handler(table, 0, filp, buf, &res, ppos);
199 if (!error)
200 error = res;
201out:
202 sysctl_head_finish(head);
203
204 return error;
205}
206
207static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
208 size_t count, loff_t *ppos)
209{
210 struct dentry *dentry = filp->f_dentry;
211 struct ctl_table_header *head;
212 struct ctl_table *table;
David Howells2a2da532007-10-25 15:27:40 +0100213 ssize_t error;
214 size_t res;
Eric W. Biederman77b14db2007-02-14 00:34:12 -0800215
216 table = do_proc_sys_lookup(dentry->d_parent, &dentry->d_name, &head);
217 /* Has the sysctl entry disappeared on us? */
218 error = -ENOENT;
219 if (!table)
220 goto out;
221
222 /* Has the sysctl entry been replaced by a directory? */
223 error = -EISDIR;
224 if (!table->proc_handler)
225 goto out;
226
227 /*
228 * At this point we know that the sysctl was not unregistered
229 * and won't be until we finish.
230 */
231 error = -EPERM;
232 if (sysctl_perm(table, MAY_WRITE))
233 goto out;
234
235 /* careful: calling conventions are nasty here */
236 res = count;
237 error = table->proc_handler(table, 1, filp, (char __user *)buf,
238 &res, ppos);
239 if (!error)
240 error = res;
241out:
242 sysctl_head_finish(head);
243
244 return error;
245}
246
247
248static int proc_sys_fill_cache(struct file *filp, void *dirent,
249 filldir_t filldir, struct ctl_table *table)
250{
251 struct ctl_table_header *head;
252 struct ctl_table *child_table = NULL;
253 struct dentry *child, *dir = filp->f_path.dentry;
254 struct inode *inode;
255 struct qstr qname;
256 ino_t ino = 0;
257 unsigned type = DT_UNKNOWN;
258 int ret;
259
260 qname.name = table->procname;
261 qname.len = strlen(table->procname);
262 qname.hash = full_name_hash(qname.name, qname.len);
263
264 /* Suppress duplicates.
265 * Only fill a directory entry if it is the value that
266 * an ordinary lookup of that name returns. Hide all
267 * others.
268 *
269 * If we ever cache this translation in the dcache
270 * I should do a dcache lookup first. But for now
271 * it is just simpler not to.
272 */
273 ret = 0;
274 child_table = do_proc_sys_lookup(dir, &qname, &head);
275 sysctl_head_finish(head);
276 if (child_table != table)
277 return 0;
278
279 child = d_lookup(dir, &qname);
280 if (!child) {
281 struct dentry *new;
282 new = d_alloc(dir, &qname);
283 if (new) {
284 inode = proc_sys_make_inode(dir->d_inode, table);
285 if (!inode)
286 child = ERR_PTR(-ENOMEM);
287 else {
288 new->d_op = &proc_sys_dentry_operations;
289 d_add(new, inode);
290 }
291 if (child)
292 dput(new);
293 else
294 child = new;
295 }
296 }
297 if (!child || IS_ERR(child) || !child->d_inode)
298 goto end_instantiate;
299 inode = child->d_inode;
300 if (inode) {
301 ino = inode->i_ino;
302 type = inode->i_mode >> 12;
303 }
304 dput(child);
305end_instantiate:
306 if (!ino)
307 ino= find_inode_number(dir, &qname);
308 if (!ino)
309 ino = 1;
310 return filldir(dirent, qname.name, qname.len, filp->f_pos, ino, type);
311}
312
313static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir)
314{
315 struct dentry *dentry = filp->f_dentry;
316 struct inode *inode = dentry->d_inode;
317 struct ctl_table_header *head = NULL;
318 struct ctl_table *table;
319 unsigned long pos;
320 int ret;
321
322 ret = -ENOTDIR;
323 if (!S_ISDIR(inode->i_mode))
324 goto out;
325
326 ret = 0;
327 /* Avoid a switch here: arm builds fail with missing __cmpdi2 */
328 if (filp->f_pos == 0) {
329 if (filldir(dirent, ".", 1, filp->f_pos,
330 inode->i_ino, DT_DIR) < 0)
331 goto out;
332 filp->f_pos++;
333 }
334 if (filp->f_pos == 1) {
335 if (filldir(dirent, "..", 2, filp->f_pos,
336 parent_ino(dentry), DT_DIR) < 0)
337 goto out;
338 filp->f_pos++;
339 }
340 pos = 2;
341
342 /* - Find each instance of the directory
343 * - Read all entries in each instance
344 * - Before returning an entry to user space lookup the entry
345 * by name and if I find a different entry don't return
346 * this one because it means it is a buried dup.
347 * For sysctl this should only happen for directory entries.
348 */
349 for (head = sysctl_head_next(NULL); head; head = sysctl_head_next(head)) {
350 table = proc_sys_lookup_table(dentry, head->ctl_table);
351
352 if (!table)
353 continue;
354
355 for (; table->ctl_name || table->procname; table++, pos++) {
356 /* Can't do anything without a proc name */
357 if (!table->procname)
358 continue;
359
360 if (pos < filp->f_pos)
361 continue;
362
363 if (proc_sys_fill_cache(filp, dirent, filldir, table) < 0)
364 goto out;
365 filp->f_pos = pos + 1;
366 }
367 }
368 ret = 1;
369out:
370 sysctl_head_finish(head);
371 return ret;
372}
373
374static int proc_sys_permission(struct inode *inode, int mask, struct nameidata *nd)
375{
376 /*
377 * sysctl entries that are not writeable,
378 * are _NOT_ writeable, capabilities or not.
379 */
380 struct ctl_table_header *head;
381 struct ctl_table *table;
382 struct dentry *dentry;
383 int mode;
384 int depth;
385 int error;
386
387 head = NULL;
388 depth = PROC_I(inode)->fd;
389
390 /* First check the cached permissions, in case we don't have
391 * enough information to lookup the sysctl table entry.
392 */
393 error = -EACCES;
394 mode = inode->i_mode;
395
396 if (current->euid == 0)
397 mode >>= 6;
398 else if (in_group_p(0))
399 mode >>= 3;
400
401 if ((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)
402 error = 0;
403
404 /* If we can't get a sysctl table entry the permission
405 * checks on the cached mode will have to be enough.
406 */
407 if (!nd || !depth)
408 goto out;
409
Jan Blunck4ac91372008-02-14 19:34:32 -0800410 dentry = nd->path.dentry;
Eric W. Biederman77b14db2007-02-14 00:34:12 -0800411 table = do_proc_sys_lookup(dentry->d_parent, &dentry->d_name, &head);
412
413 /* If the entry does not exist deny permission */
414 error = -EACCES;
415 if (!table)
416 goto out;
417
418 /* Use the permissions on the sysctl table entry */
419 error = sysctl_perm(table, mask);
420out:
421 sysctl_head_finish(head);
422 return error;
423}
424
425static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
426{
427 struct inode *inode = dentry->d_inode;
428 int error;
429
430 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
431 return -EPERM;
432
433 error = inode_change_ok(inode, attr);
John Johansen9d0633c2007-05-08 00:29:44 -0700434 if (!error)
435 error = inode_setattr(inode, attr);
Eric W. Biederman77b14db2007-02-14 00:34:12 -0800436
437 return error;
438}
439
440/* I'm lazy and don't distinguish between files and directories,
441 * until access time.
442 */
443static const struct file_operations proc_sys_file_operations = {
444 .read = proc_sys_read,
445 .write = proc_sys_write,
446 .readdir = proc_sys_readdir,
447};
448
Jan Engelhardt03a44822008-02-08 04:21:19 -0800449static const struct inode_operations proc_sys_inode_operations = {
Eric W. Biederman77b14db2007-02-14 00:34:12 -0800450 .lookup = proc_sys_lookup,
451 .permission = proc_sys_permission,
452 .setattr = proc_sys_setattr,
453};
454
455static int proc_sys_revalidate(struct dentry *dentry, struct nameidata *nd)
456{
457 struct ctl_table_header *head;
458 struct ctl_table *table;
459 table = do_proc_sys_lookup(dentry->d_parent, &dentry->d_name, &head);
460 proc_sys_refresh_inode(dentry->d_inode, table);
461 sysctl_head_finish(head);
462 return !!table;
463}
464
465static struct dentry_operations proc_sys_dentry_operations = {
466 .d_revalidate = proc_sys_revalidate,
467};
468
469static struct proc_dir_entry *proc_sys_root;
470
471int proc_sys_init(void)
472{
473 proc_sys_root = proc_mkdir("sys", NULL);
474 proc_sys_root->proc_iops = &proc_sys_inode_operations;
475 proc_sys_root->proc_fops = &proc_sys_file_operations;
476 proc_sys_root->nlink = 0;
477 return 0;
478}