Loading...
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
11static const struct dentry_operations proc_sys_dentry_operations;
12static const struct file_operations proc_sys_file_operations;
13static const struct inode_operations proc_sys_inode_operations;
14static const struct file_operations proc_sys_dir_file_operations;
15static const struct inode_operations proc_sys_dir_operations;
16
17static 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 }
46out:
47 return inode;
48}
49
50static 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
71static 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
79static 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
126out:
127 sysctl_head_finish(head);
128 return err;
129}
130
131static 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;
161out:
162 sysctl_head_finish(head);
163
164 return error;
165}
166
167static 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
173static 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
180static 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
218static 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
242static 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;
292out:
293 sysctl_head_finish(head);
294 return ret;
295}
296
297static 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
325static 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
349static 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
366static const struct file_operations proc_sys_file_operations = {
367 .read = proc_sys_read,
368 .write = proc_sys_write,
369 .llseek = default_llseek,
370};
371
372static const struct file_operations proc_sys_dir_file_operations = {
373 .readdir = proc_sys_readdir,
374 .llseek = generic_file_llseek,
375};
376
377static 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
383static 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
390static 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
397static int proc_sys_delete(const struct dentry *dentry)
398{
399 return !!PROC_I(dentry->d_inode)->sysctl->unregistering;
400}
401
402static 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
421static 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
427int __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}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * /proc/sys support
4 */
5#include <linux/init.h>
6#include <linux/sysctl.h>
7#include <linux/poll.h>
8#include <linux/proc_fs.h>
9#include <linux/printk.h>
10#include <linux/security.h>
11#include <linux/sched.h>
12#include <linux/cred.h>
13#include <linux/namei.h>
14#include <linux/mm.h>
15#include <linux/module.h>
16#include "internal.h"
17
18static const struct dentry_operations proc_sys_dentry_operations;
19static const struct file_operations proc_sys_file_operations;
20static const struct inode_operations proc_sys_inode_operations;
21static const struct file_operations proc_sys_dir_file_operations;
22static const struct inode_operations proc_sys_dir_operations;
23
24/* Support for permanently empty directories */
25
26struct ctl_table sysctl_mount_point[] = {
27 { }
28};
29
30static bool is_empty_dir(struct ctl_table_header *head)
31{
32 return head->ctl_table[0].child == sysctl_mount_point;
33}
34
35static void set_empty_dir(struct ctl_dir *dir)
36{
37 dir->header.ctl_table[0].child = sysctl_mount_point;
38}
39
40static void clear_empty_dir(struct ctl_dir *dir)
41
42{
43 dir->header.ctl_table[0].child = NULL;
44}
45
46void proc_sys_poll_notify(struct ctl_table_poll *poll)
47{
48 if (!poll)
49 return;
50
51 atomic_inc(&poll->event);
52 wake_up_interruptible(&poll->wait);
53}
54
55static struct ctl_table root_table[] = {
56 {
57 .procname = "",
58 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
59 },
60 { }
61};
62static struct ctl_table_root sysctl_table_root = {
63 .default_set.dir.header = {
64 {{.count = 1,
65 .nreg = 1,
66 .ctl_table = root_table }},
67 .ctl_table_arg = root_table,
68 .root = &sysctl_table_root,
69 .set = &sysctl_table_root.default_set,
70 },
71};
72
73static DEFINE_SPINLOCK(sysctl_lock);
74
75static void drop_sysctl_table(struct ctl_table_header *header);
76static int sysctl_follow_link(struct ctl_table_header **phead,
77 struct ctl_table **pentry);
78static int insert_links(struct ctl_table_header *head);
79static void put_links(struct ctl_table_header *header);
80
81static void sysctl_print_dir(struct ctl_dir *dir)
82{
83 if (dir->header.parent)
84 sysctl_print_dir(dir->header.parent);
85 pr_cont("%s/", dir->header.ctl_table[0].procname);
86}
87
88static int namecmp(const char *name1, int len1, const char *name2, int len2)
89{
90 int minlen;
91 int cmp;
92
93 minlen = len1;
94 if (minlen > len2)
95 minlen = len2;
96
97 cmp = memcmp(name1, name2, minlen);
98 if (cmp == 0)
99 cmp = len1 - len2;
100 return cmp;
101}
102
103/* Called under sysctl_lock */
104static struct ctl_table *find_entry(struct ctl_table_header **phead,
105 struct ctl_dir *dir, const char *name, int namelen)
106{
107 struct ctl_table_header *head;
108 struct ctl_table *entry;
109 struct rb_node *node = dir->root.rb_node;
110
111 while (node)
112 {
113 struct ctl_node *ctl_node;
114 const char *procname;
115 int cmp;
116
117 ctl_node = rb_entry(node, struct ctl_node, node);
118 head = ctl_node->header;
119 entry = &head->ctl_table[ctl_node - head->node];
120 procname = entry->procname;
121
122 cmp = namecmp(name, namelen, procname, strlen(procname));
123 if (cmp < 0)
124 node = node->rb_left;
125 else if (cmp > 0)
126 node = node->rb_right;
127 else {
128 *phead = head;
129 return entry;
130 }
131 }
132 return NULL;
133}
134
135static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
136{
137 struct rb_node *node = &head->node[entry - head->ctl_table].node;
138 struct rb_node **p = &head->parent->root.rb_node;
139 struct rb_node *parent = NULL;
140 const char *name = entry->procname;
141 int namelen = strlen(name);
142
143 while (*p) {
144 struct ctl_table_header *parent_head;
145 struct ctl_table *parent_entry;
146 struct ctl_node *parent_node;
147 const char *parent_name;
148 int cmp;
149
150 parent = *p;
151 parent_node = rb_entry(parent, struct ctl_node, node);
152 parent_head = parent_node->header;
153 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
154 parent_name = parent_entry->procname;
155
156 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
157 if (cmp < 0)
158 p = &(*p)->rb_left;
159 else if (cmp > 0)
160 p = &(*p)->rb_right;
161 else {
162 pr_err("sysctl duplicate entry: ");
163 sysctl_print_dir(head->parent);
164 pr_cont("/%s\n", entry->procname);
165 return -EEXIST;
166 }
167 }
168
169 rb_link_node(node, parent, p);
170 rb_insert_color(node, &head->parent->root);
171 return 0;
172}
173
174static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
175{
176 struct rb_node *node = &head->node[entry - head->ctl_table].node;
177
178 rb_erase(node, &head->parent->root);
179}
180
181static void init_header(struct ctl_table_header *head,
182 struct ctl_table_root *root, struct ctl_table_set *set,
183 struct ctl_node *node, struct ctl_table *table)
184{
185 head->ctl_table = table;
186 head->ctl_table_arg = table;
187 head->used = 0;
188 head->count = 1;
189 head->nreg = 1;
190 head->unregistering = NULL;
191 head->root = root;
192 head->set = set;
193 head->parent = NULL;
194 head->node = node;
195 INIT_HLIST_HEAD(&head->inodes);
196 if (node) {
197 struct ctl_table *entry;
198 for (entry = table; entry->procname; entry++, node++)
199 node->header = head;
200 }
201}
202
203static void erase_header(struct ctl_table_header *head)
204{
205 struct ctl_table *entry;
206 for (entry = head->ctl_table; entry->procname; entry++)
207 erase_entry(head, entry);
208}
209
210static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
211{
212 struct ctl_table *entry;
213 int err;
214
215 /* Is this a permanently empty directory? */
216 if (is_empty_dir(&dir->header))
217 return -EROFS;
218
219 /* Am I creating a permanently empty directory? */
220 if (header->ctl_table == sysctl_mount_point) {
221 if (!RB_EMPTY_ROOT(&dir->root))
222 return -EINVAL;
223 set_empty_dir(dir);
224 }
225
226 dir->header.nreg++;
227 header->parent = dir;
228 err = insert_links(header);
229 if (err)
230 goto fail_links;
231 for (entry = header->ctl_table; entry->procname; entry++) {
232 err = insert_entry(header, entry);
233 if (err)
234 goto fail;
235 }
236 return 0;
237fail:
238 erase_header(header);
239 put_links(header);
240fail_links:
241 if (header->ctl_table == sysctl_mount_point)
242 clear_empty_dir(dir);
243 header->parent = NULL;
244 drop_sysctl_table(&dir->header);
245 return err;
246}
247
248/* called under sysctl_lock */
249static int use_table(struct ctl_table_header *p)
250{
251 if (unlikely(p->unregistering))
252 return 0;
253 p->used++;
254 return 1;
255}
256
257/* called under sysctl_lock */
258static void unuse_table(struct ctl_table_header *p)
259{
260 if (!--p->used)
261 if (unlikely(p->unregistering))
262 complete(p->unregistering);
263}
264
265static void proc_sys_prune_dcache(struct ctl_table_header *head)
266{
267 struct inode *inode;
268 struct proc_inode *ei;
269 struct hlist_node *node;
270 struct super_block *sb;
271
272 rcu_read_lock();
273 for (;;) {
274 node = hlist_first_rcu(&head->inodes);
275 if (!node)
276 break;
277 ei = hlist_entry(node, struct proc_inode, sysctl_inodes);
278 spin_lock(&sysctl_lock);
279 hlist_del_init_rcu(&ei->sysctl_inodes);
280 spin_unlock(&sysctl_lock);
281
282 inode = &ei->vfs_inode;
283 sb = inode->i_sb;
284 if (!atomic_inc_not_zero(&sb->s_active))
285 continue;
286 inode = igrab(inode);
287 rcu_read_unlock();
288 if (unlikely(!inode)) {
289 deactivate_super(sb);
290 rcu_read_lock();
291 continue;
292 }
293
294 d_prune_aliases(inode);
295 iput(inode);
296 deactivate_super(sb);
297
298 rcu_read_lock();
299 }
300 rcu_read_unlock();
301}
302
303/* called under sysctl_lock, will reacquire if has to wait */
304static void start_unregistering(struct ctl_table_header *p)
305{
306 /*
307 * if p->used is 0, nobody will ever touch that entry again;
308 * we'll eliminate all paths to it before dropping sysctl_lock
309 */
310 if (unlikely(p->used)) {
311 struct completion wait;
312 init_completion(&wait);
313 p->unregistering = &wait;
314 spin_unlock(&sysctl_lock);
315 wait_for_completion(&wait);
316 } else {
317 /* anything non-NULL; we'll never dereference it */
318 p->unregistering = ERR_PTR(-EINVAL);
319 spin_unlock(&sysctl_lock);
320 }
321 /*
322 * Prune dentries for unregistered sysctls: namespaced sysctls
323 * can have duplicate names and contaminate dcache very badly.
324 */
325 proc_sys_prune_dcache(p);
326 /*
327 * do not remove from the list until nobody holds it; walking the
328 * list in do_sysctl() relies on that.
329 */
330 spin_lock(&sysctl_lock);
331 erase_header(p);
332}
333
334static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
335{
336 BUG_ON(!head);
337 spin_lock(&sysctl_lock);
338 if (!use_table(head))
339 head = ERR_PTR(-ENOENT);
340 spin_unlock(&sysctl_lock);
341 return head;
342}
343
344static void sysctl_head_finish(struct ctl_table_header *head)
345{
346 if (!head)
347 return;
348 spin_lock(&sysctl_lock);
349 unuse_table(head);
350 spin_unlock(&sysctl_lock);
351}
352
353static struct ctl_table_set *
354lookup_header_set(struct ctl_table_root *root)
355{
356 struct ctl_table_set *set = &root->default_set;
357 if (root->lookup)
358 set = root->lookup(root);
359 return set;
360}
361
362static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
363 struct ctl_dir *dir,
364 const char *name, int namelen)
365{
366 struct ctl_table_header *head;
367 struct ctl_table *entry;
368
369 spin_lock(&sysctl_lock);
370 entry = find_entry(&head, dir, name, namelen);
371 if (entry && use_table(head))
372 *phead = head;
373 else
374 entry = NULL;
375 spin_unlock(&sysctl_lock);
376 return entry;
377}
378
379static struct ctl_node *first_usable_entry(struct rb_node *node)
380{
381 struct ctl_node *ctl_node;
382
383 for (;node; node = rb_next(node)) {
384 ctl_node = rb_entry(node, struct ctl_node, node);
385 if (use_table(ctl_node->header))
386 return ctl_node;
387 }
388 return NULL;
389}
390
391static void first_entry(struct ctl_dir *dir,
392 struct ctl_table_header **phead, struct ctl_table **pentry)
393{
394 struct ctl_table_header *head = NULL;
395 struct ctl_table *entry = NULL;
396 struct ctl_node *ctl_node;
397
398 spin_lock(&sysctl_lock);
399 ctl_node = first_usable_entry(rb_first(&dir->root));
400 spin_unlock(&sysctl_lock);
401 if (ctl_node) {
402 head = ctl_node->header;
403 entry = &head->ctl_table[ctl_node - head->node];
404 }
405 *phead = head;
406 *pentry = entry;
407}
408
409static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
410{
411 struct ctl_table_header *head = *phead;
412 struct ctl_table *entry = *pentry;
413 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
414
415 spin_lock(&sysctl_lock);
416 unuse_table(head);
417
418 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
419 spin_unlock(&sysctl_lock);
420 head = NULL;
421 if (ctl_node) {
422 head = ctl_node->header;
423 entry = &head->ctl_table[ctl_node - head->node];
424 }
425 *phead = head;
426 *pentry = entry;
427}
428
429/*
430 * sysctl_perm does NOT grant the superuser all rights automatically, because
431 * some sysctl variables are readonly even to root.
432 */
433
434static int test_perm(int mode, int op)
435{
436 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
437 mode >>= 6;
438 else if (in_egroup_p(GLOBAL_ROOT_GID))
439 mode >>= 3;
440 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
441 return 0;
442 return -EACCES;
443}
444
445static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
446{
447 struct ctl_table_root *root = head->root;
448 int mode;
449
450 if (root->permissions)
451 mode = root->permissions(head, table);
452 else
453 mode = table->mode;
454
455 return test_perm(mode, op);
456}
457
458static struct inode *proc_sys_make_inode(struct super_block *sb,
459 struct ctl_table_header *head, struct ctl_table *table)
460{
461 struct ctl_table_root *root = head->root;
462 struct inode *inode;
463 struct proc_inode *ei;
464
465 inode = new_inode(sb);
466 if (!inode)
467 goto out;
468
469 inode->i_ino = get_next_ino();
470
471 ei = PROC_I(inode);
472
473 spin_lock(&sysctl_lock);
474 if (unlikely(head->unregistering)) {
475 spin_unlock(&sysctl_lock);
476 iput(inode);
477 inode = NULL;
478 goto out;
479 }
480 ei->sysctl = head;
481 ei->sysctl_entry = table;
482 hlist_add_head_rcu(&ei->sysctl_inodes, &head->inodes);
483 head->count++;
484 spin_unlock(&sysctl_lock);
485
486 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
487 inode->i_mode = table->mode;
488 if (!S_ISDIR(table->mode)) {
489 inode->i_mode |= S_IFREG;
490 inode->i_op = &proc_sys_inode_operations;
491 inode->i_fop = &proc_sys_file_operations;
492 } else {
493 inode->i_mode |= S_IFDIR;
494 inode->i_op = &proc_sys_dir_operations;
495 inode->i_fop = &proc_sys_dir_file_operations;
496 if (is_empty_dir(head))
497 make_empty_dir_inode(inode);
498 }
499
500 if (root->set_ownership)
501 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
502
503out:
504 return inode;
505}
506
507void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
508{
509 spin_lock(&sysctl_lock);
510 hlist_del_init_rcu(&PROC_I(inode)->sysctl_inodes);
511 if (!--head->count)
512 kfree_rcu(head, rcu);
513 spin_unlock(&sysctl_lock);
514}
515
516static struct ctl_table_header *grab_header(struct inode *inode)
517{
518 struct ctl_table_header *head = PROC_I(inode)->sysctl;
519 if (!head)
520 head = &sysctl_table_root.default_set.dir.header;
521 return sysctl_head_grab(head);
522}
523
524static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
525 unsigned int flags)
526{
527 struct ctl_table_header *head = grab_header(dir);
528 struct ctl_table_header *h = NULL;
529 const struct qstr *name = &dentry->d_name;
530 struct ctl_table *p;
531 struct inode *inode;
532 struct dentry *err = ERR_PTR(-ENOENT);
533 struct ctl_dir *ctl_dir;
534 int ret;
535
536 if (IS_ERR(head))
537 return ERR_CAST(head);
538
539 ctl_dir = container_of(head, struct ctl_dir, header);
540
541 p = lookup_entry(&h, ctl_dir, name->name, name->len);
542 if (!p)
543 goto out;
544
545 if (S_ISLNK(p->mode)) {
546 ret = sysctl_follow_link(&h, &p);
547 err = ERR_PTR(ret);
548 if (ret)
549 goto out;
550 }
551
552 err = ERR_PTR(-ENOMEM);
553 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
554 if (!inode)
555 goto out;
556
557 err = NULL;
558 d_set_d_op(dentry, &proc_sys_dentry_operations);
559 d_add(dentry, inode);
560
561out:
562 if (h)
563 sysctl_head_finish(h);
564 sysctl_head_finish(head);
565 return err;
566}
567
568static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
569 size_t count, loff_t *ppos, int write)
570{
571 struct inode *inode = file_inode(filp);
572 struct ctl_table_header *head = grab_header(inode);
573 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
574 ssize_t error;
575 size_t res;
576
577 if (IS_ERR(head))
578 return PTR_ERR(head);
579
580 /*
581 * At this point we know that the sysctl was not unregistered
582 * and won't be until we finish.
583 */
584 error = -EPERM;
585 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
586 goto out;
587
588 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
589 error = -EINVAL;
590 if (!table->proc_handler)
591 goto out;
592
593 /* careful: calling conventions are nasty here */
594 res = count;
595 error = table->proc_handler(table, write, buf, &res, ppos);
596 if (!error)
597 error = res;
598out:
599 sysctl_head_finish(head);
600
601 return error;
602}
603
604static ssize_t proc_sys_read(struct file *filp, char __user *buf,
605 size_t count, loff_t *ppos)
606{
607 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
608}
609
610static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
611 size_t count, loff_t *ppos)
612{
613 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
614}
615
616static int proc_sys_open(struct inode *inode, struct file *filp)
617{
618 struct ctl_table_header *head = grab_header(inode);
619 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
620
621 /* sysctl was unregistered */
622 if (IS_ERR(head))
623 return PTR_ERR(head);
624
625 if (table->poll)
626 filp->private_data = proc_sys_poll_event(table->poll);
627
628 sysctl_head_finish(head);
629
630 return 0;
631}
632
633static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
634{
635 struct inode *inode = file_inode(filp);
636 struct ctl_table_header *head = grab_header(inode);
637 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
638 __poll_t ret = DEFAULT_POLLMASK;
639 unsigned long event;
640
641 /* sysctl was unregistered */
642 if (IS_ERR(head))
643 return EPOLLERR | EPOLLHUP;
644
645 if (!table->proc_handler)
646 goto out;
647
648 if (!table->poll)
649 goto out;
650
651 event = (unsigned long)filp->private_data;
652 poll_wait(filp, &table->poll->wait, wait);
653
654 if (event != atomic_read(&table->poll->event)) {
655 filp->private_data = proc_sys_poll_event(table->poll);
656 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
657 }
658
659out:
660 sysctl_head_finish(head);
661
662 return ret;
663}
664
665static bool proc_sys_fill_cache(struct file *file,
666 struct dir_context *ctx,
667 struct ctl_table_header *head,
668 struct ctl_table *table)
669{
670 struct dentry *child, *dir = file->f_path.dentry;
671 struct inode *inode;
672 struct qstr qname;
673 ino_t ino = 0;
674 unsigned type = DT_UNKNOWN;
675
676 qname.name = table->procname;
677 qname.len = strlen(table->procname);
678 qname.hash = full_name_hash(dir, qname.name, qname.len);
679
680 child = d_lookup(dir, &qname);
681 if (!child) {
682 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
683 child = d_alloc_parallel(dir, &qname, &wq);
684 if (IS_ERR(child))
685 return false;
686 if (d_in_lookup(child)) {
687 inode = proc_sys_make_inode(dir->d_sb, head, table);
688 if (!inode) {
689 d_lookup_done(child);
690 dput(child);
691 return false;
692 }
693 d_set_d_op(child, &proc_sys_dentry_operations);
694 d_add(child, inode);
695 }
696 }
697 inode = d_inode(child);
698 ino = inode->i_ino;
699 type = inode->i_mode >> 12;
700 dput(child);
701 return dir_emit(ctx, qname.name, qname.len, ino, type);
702}
703
704static bool proc_sys_link_fill_cache(struct file *file,
705 struct dir_context *ctx,
706 struct ctl_table_header *head,
707 struct ctl_table *table)
708{
709 bool ret = true;
710
711 head = sysctl_head_grab(head);
712 if (IS_ERR(head))
713 return false;
714
715 /* It is not an error if we can not follow the link ignore it */
716 if (sysctl_follow_link(&head, &table))
717 goto out;
718
719 ret = proc_sys_fill_cache(file, ctx, head, table);
720out:
721 sysctl_head_finish(head);
722 return ret;
723}
724
725static int scan(struct ctl_table_header *head, struct ctl_table *table,
726 unsigned long *pos, struct file *file,
727 struct dir_context *ctx)
728{
729 bool res;
730
731 if ((*pos)++ < ctx->pos)
732 return true;
733
734 if (unlikely(S_ISLNK(table->mode)))
735 res = proc_sys_link_fill_cache(file, ctx, head, table);
736 else
737 res = proc_sys_fill_cache(file, ctx, head, table);
738
739 if (res)
740 ctx->pos = *pos;
741
742 return res;
743}
744
745static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
746{
747 struct ctl_table_header *head = grab_header(file_inode(file));
748 struct ctl_table_header *h = NULL;
749 struct ctl_table *entry;
750 struct ctl_dir *ctl_dir;
751 unsigned long pos;
752
753 if (IS_ERR(head))
754 return PTR_ERR(head);
755
756 ctl_dir = container_of(head, struct ctl_dir, header);
757
758 if (!dir_emit_dots(file, ctx))
759 goto out;
760
761 pos = 2;
762
763 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
764 if (!scan(h, entry, &pos, file, ctx)) {
765 sysctl_head_finish(h);
766 break;
767 }
768 }
769out:
770 sysctl_head_finish(head);
771 return 0;
772}
773
774static int proc_sys_permission(struct inode *inode, int mask)
775{
776 /*
777 * sysctl entries that are not writeable,
778 * are _NOT_ writeable, capabilities or not.
779 */
780 struct ctl_table_header *head;
781 struct ctl_table *table;
782 int error;
783
784 /* Executable files are not allowed under /proc/sys/ */
785 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
786 return -EACCES;
787
788 head = grab_header(inode);
789 if (IS_ERR(head))
790 return PTR_ERR(head);
791
792 table = PROC_I(inode)->sysctl_entry;
793 if (!table) /* global root - r-xr-xr-x */
794 error = mask & MAY_WRITE ? -EACCES : 0;
795 else /* Use the permissions on the sysctl table entry */
796 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
797
798 sysctl_head_finish(head);
799 return error;
800}
801
802static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
803{
804 struct inode *inode = d_inode(dentry);
805 int error;
806
807 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
808 return -EPERM;
809
810 error = setattr_prepare(dentry, attr);
811 if (error)
812 return error;
813
814 setattr_copy(inode, attr);
815 mark_inode_dirty(inode);
816 return 0;
817}
818
819static int proc_sys_getattr(const struct path *path, struct kstat *stat,
820 u32 request_mask, unsigned int query_flags)
821{
822 struct inode *inode = d_inode(path->dentry);
823 struct ctl_table_header *head = grab_header(inode);
824 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
825
826 if (IS_ERR(head))
827 return PTR_ERR(head);
828
829 generic_fillattr(inode, stat);
830 if (table)
831 stat->mode = (stat->mode & S_IFMT) | table->mode;
832
833 sysctl_head_finish(head);
834 return 0;
835}
836
837static const struct file_operations proc_sys_file_operations = {
838 .open = proc_sys_open,
839 .poll = proc_sys_poll,
840 .read = proc_sys_read,
841 .write = proc_sys_write,
842 .llseek = default_llseek,
843};
844
845static const struct file_operations proc_sys_dir_file_operations = {
846 .read = generic_read_dir,
847 .iterate_shared = proc_sys_readdir,
848 .llseek = generic_file_llseek,
849};
850
851static const struct inode_operations proc_sys_inode_operations = {
852 .permission = proc_sys_permission,
853 .setattr = proc_sys_setattr,
854 .getattr = proc_sys_getattr,
855};
856
857static const struct inode_operations proc_sys_dir_operations = {
858 .lookup = proc_sys_lookup,
859 .permission = proc_sys_permission,
860 .setattr = proc_sys_setattr,
861 .getattr = proc_sys_getattr,
862};
863
864static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
865{
866 if (flags & LOOKUP_RCU)
867 return -ECHILD;
868 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
869}
870
871static int proc_sys_delete(const struct dentry *dentry)
872{
873 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
874}
875
876static int sysctl_is_seen(struct ctl_table_header *p)
877{
878 struct ctl_table_set *set = p->set;
879 int res;
880 spin_lock(&sysctl_lock);
881 if (p->unregistering)
882 res = 0;
883 else if (!set->is_seen)
884 res = 1;
885 else
886 res = set->is_seen(set);
887 spin_unlock(&sysctl_lock);
888 return res;
889}
890
891static int proc_sys_compare(const struct dentry *dentry,
892 unsigned int len, const char *str, const struct qstr *name)
893{
894 struct ctl_table_header *head;
895 struct inode *inode;
896
897 /* Although proc doesn't have negative dentries, rcu-walk means
898 * that inode here can be NULL */
899 /* AV: can it, indeed? */
900 inode = d_inode_rcu(dentry);
901 if (!inode)
902 return 1;
903 if (name->len != len)
904 return 1;
905 if (memcmp(name->name, str, len))
906 return 1;
907 head = rcu_dereference(PROC_I(inode)->sysctl);
908 return !head || !sysctl_is_seen(head);
909}
910
911static const struct dentry_operations proc_sys_dentry_operations = {
912 .d_revalidate = proc_sys_revalidate,
913 .d_delete = proc_sys_delete,
914 .d_compare = proc_sys_compare,
915};
916
917static struct ctl_dir *find_subdir(struct ctl_dir *dir,
918 const char *name, int namelen)
919{
920 struct ctl_table_header *head;
921 struct ctl_table *entry;
922
923 entry = find_entry(&head, dir, name, namelen);
924 if (!entry)
925 return ERR_PTR(-ENOENT);
926 if (!S_ISDIR(entry->mode))
927 return ERR_PTR(-ENOTDIR);
928 return container_of(head, struct ctl_dir, header);
929}
930
931static struct ctl_dir *new_dir(struct ctl_table_set *set,
932 const char *name, int namelen)
933{
934 struct ctl_table *table;
935 struct ctl_dir *new;
936 struct ctl_node *node;
937 char *new_name;
938
939 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
940 sizeof(struct ctl_table)*2 + namelen + 1,
941 GFP_KERNEL);
942 if (!new)
943 return NULL;
944
945 node = (struct ctl_node *)(new + 1);
946 table = (struct ctl_table *)(node + 1);
947 new_name = (char *)(table + 2);
948 memcpy(new_name, name, namelen);
949 new_name[namelen] = '\0';
950 table[0].procname = new_name;
951 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
952 init_header(&new->header, set->dir.header.root, set, node, table);
953
954 return new;
955}
956
957/**
958 * get_subdir - find or create a subdir with the specified name.
959 * @dir: Directory to create the subdirectory in
960 * @name: The name of the subdirectory to find or create
961 * @namelen: The length of name
962 *
963 * Takes a directory with an elevated reference count so we know that
964 * if we drop the lock the directory will not go away. Upon success
965 * the reference is moved from @dir to the returned subdirectory.
966 * Upon error an error code is returned and the reference on @dir is
967 * simply dropped.
968 */
969static struct ctl_dir *get_subdir(struct ctl_dir *dir,
970 const char *name, int namelen)
971{
972 struct ctl_table_set *set = dir->header.set;
973 struct ctl_dir *subdir, *new = NULL;
974 int err;
975
976 spin_lock(&sysctl_lock);
977 subdir = find_subdir(dir, name, namelen);
978 if (!IS_ERR(subdir))
979 goto found;
980 if (PTR_ERR(subdir) != -ENOENT)
981 goto failed;
982
983 spin_unlock(&sysctl_lock);
984 new = new_dir(set, name, namelen);
985 spin_lock(&sysctl_lock);
986 subdir = ERR_PTR(-ENOMEM);
987 if (!new)
988 goto failed;
989
990 /* Was the subdir added while we dropped the lock? */
991 subdir = find_subdir(dir, name, namelen);
992 if (!IS_ERR(subdir))
993 goto found;
994 if (PTR_ERR(subdir) != -ENOENT)
995 goto failed;
996
997 /* Nope. Use the our freshly made directory entry. */
998 err = insert_header(dir, &new->header);
999 subdir = ERR_PTR(err);
1000 if (err)
1001 goto failed;
1002 subdir = new;
1003found:
1004 subdir->header.nreg++;
1005failed:
1006 if (IS_ERR(subdir)) {
1007 pr_err("sysctl could not get directory: ");
1008 sysctl_print_dir(dir);
1009 pr_cont("/%*.*s %ld\n",
1010 namelen, namelen, name, PTR_ERR(subdir));
1011 }
1012 drop_sysctl_table(&dir->header);
1013 if (new)
1014 drop_sysctl_table(&new->header);
1015 spin_unlock(&sysctl_lock);
1016 return subdir;
1017}
1018
1019static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1020{
1021 struct ctl_dir *parent;
1022 const char *procname;
1023 if (!dir->header.parent)
1024 return &set->dir;
1025 parent = xlate_dir(set, dir->header.parent);
1026 if (IS_ERR(parent))
1027 return parent;
1028 procname = dir->header.ctl_table[0].procname;
1029 return find_subdir(parent, procname, strlen(procname));
1030}
1031
1032static int sysctl_follow_link(struct ctl_table_header **phead,
1033 struct ctl_table **pentry)
1034{
1035 struct ctl_table_header *head;
1036 struct ctl_table_root *root;
1037 struct ctl_table_set *set;
1038 struct ctl_table *entry;
1039 struct ctl_dir *dir;
1040 int ret;
1041
1042 ret = 0;
1043 spin_lock(&sysctl_lock);
1044 root = (*pentry)->data;
1045 set = lookup_header_set(root);
1046 dir = xlate_dir(set, (*phead)->parent);
1047 if (IS_ERR(dir))
1048 ret = PTR_ERR(dir);
1049 else {
1050 const char *procname = (*pentry)->procname;
1051 head = NULL;
1052 entry = find_entry(&head, dir, procname, strlen(procname));
1053 ret = -ENOENT;
1054 if (entry && use_table(head)) {
1055 unuse_table(*phead);
1056 *phead = head;
1057 *pentry = entry;
1058 ret = 0;
1059 }
1060 }
1061
1062 spin_unlock(&sysctl_lock);
1063 return ret;
1064}
1065
1066static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1067{
1068 struct va_format vaf;
1069 va_list args;
1070
1071 va_start(args, fmt);
1072 vaf.fmt = fmt;
1073 vaf.va = &args;
1074
1075 pr_err("sysctl table check failed: %s/%s %pV\n",
1076 path, table->procname, &vaf);
1077
1078 va_end(args);
1079 return -EINVAL;
1080}
1081
1082static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1083{
1084 int err = 0;
1085
1086 if ((table->proc_handler == proc_douintvec) ||
1087 (table->proc_handler == proc_douintvec_minmax)) {
1088 if (table->maxlen != sizeof(unsigned int))
1089 err |= sysctl_err(path, table, "array not allowed");
1090 }
1091
1092 return err;
1093}
1094
1095static int sysctl_check_table(const char *path, struct ctl_table *table)
1096{
1097 int err = 0;
1098 for (; table->procname; table++) {
1099 if (table->child)
1100 err |= sysctl_err(path, table, "Not a file");
1101
1102 if ((table->proc_handler == proc_dostring) ||
1103 (table->proc_handler == proc_dointvec) ||
1104 (table->proc_handler == proc_douintvec) ||
1105 (table->proc_handler == proc_douintvec_minmax) ||
1106 (table->proc_handler == proc_dointvec_minmax) ||
1107 (table->proc_handler == proc_dointvec_jiffies) ||
1108 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1109 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1110 (table->proc_handler == proc_doulongvec_minmax) ||
1111 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1112 if (!table->data)
1113 err |= sysctl_err(path, table, "No data");
1114 if (!table->maxlen)
1115 err |= sysctl_err(path, table, "No maxlen");
1116 else
1117 err |= sysctl_check_table_array(path, table);
1118 }
1119 if (!table->proc_handler)
1120 err |= sysctl_err(path, table, "No proc_handler");
1121
1122 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1123 err |= sysctl_err(path, table, "bogus .mode 0%o",
1124 table->mode);
1125 }
1126 return err;
1127}
1128
1129static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1130 struct ctl_table_root *link_root)
1131{
1132 struct ctl_table *link_table, *entry, *link;
1133 struct ctl_table_header *links;
1134 struct ctl_node *node;
1135 char *link_name;
1136 int nr_entries, name_bytes;
1137
1138 name_bytes = 0;
1139 nr_entries = 0;
1140 for (entry = table; entry->procname; entry++) {
1141 nr_entries++;
1142 name_bytes += strlen(entry->procname) + 1;
1143 }
1144
1145 links = kzalloc(sizeof(struct ctl_table_header) +
1146 sizeof(struct ctl_node)*nr_entries +
1147 sizeof(struct ctl_table)*(nr_entries + 1) +
1148 name_bytes,
1149 GFP_KERNEL);
1150
1151 if (!links)
1152 return NULL;
1153
1154 node = (struct ctl_node *)(links + 1);
1155 link_table = (struct ctl_table *)(node + nr_entries);
1156 link_name = (char *)&link_table[nr_entries + 1];
1157
1158 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1159 int len = strlen(entry->procname) + 1;
1160 memcpy(link_name, entry->procname, len);
1161 link->procname = link_name;
1162 link->mode = S_IFLNK|S_IRWXUGO;
1163 link->data = link_root;
1164 link_name += len;
1165 }
1166 init_header(links, dir->header.root, dir->header.set, node, link_table);
1167 links->nreg = nr_entries;
1168
1169 return links;
1170}
1171
1172static bool get_links(struct ctl_dir *dir,
1173 struct ctl_table *table, struct ctl_table_root *link_root)
1174{
1175 struct ctl_table_header *head;
1176 struct ctl_table *entry, *link;
1177
1178 /* Are there links available for every entry in table? */
1179 for (entry = table; entry->procname; entry++) {
1180 const char *procname = entry->procname;
1181 link = find_entry(&head, dir, procname, strlen(procname));
1182 if (!link)
1183 return false;
1184 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1185 continue;
1186 if (S_ISLNK(link->mode) && (link->data == link_root))
1187 continue;
1188 return false;
1189 }
1190
1191 /* The checks passed. Increase the registration count on the links */
1192 for (entry = table; entry->procname; entry++) {
1193 const char *procname = entry->procname;
1194 link = find_entry(&head, dir, procname, strlen(procname));
1195 head->nreg++;
1196 }
1197 return true;
1198}
1199
1200static int insert_links(struct ctl_table_header *head)
1201{
1202 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1203 struct ctl_dir *core_parent = NULL;
1204 struct ctl_table_header *links;
1205 int err;
1206
1207 if (head->set == root_set)
1208 return 0;
1209
1210 core_parent = xlate_dir(root_set, head->parent);
1211 if (IS_ERR(core_parent))
1212 return 0;
1213
1214 if (get_links(core_parent, head->ctl_table, head->root))
1215 return 0;
1216
1217 core_parent->header.nreg++;
1218 spin_unlock(&sysctl_lock);
1219
1220 links = new_links(core_parent, head->ctl_table, head->root);
1221
1222 spin_lock(&sysctl_lock);
1223 err = -ENOMEM;
1224 if (!links)
1225 goto out;
1226
1227 err = 0;
1228 if (get_links(core_parent, head->ctl_table, head->root)) {
1229 kfree(links);
1230 goto out;
1231 }
1232
1233 err = insert_header(core_parent, links);
1234 if (err)
1235 kfree(links);
1236out:
1237 drop_sysctl_table(&core_parent->header);
1238 return err;
1239}
1240
1241/**
1242 * __register_sysctl_table - register a leaf sysctl table
1243 * @set: Sysctl tree to register on
1244 * @path: The path to the directory the sysctl table is in.
1245 * @table: the top-level table structure
1246 *
1247 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1248 * array. A completely 0 filled entry terminates the table.
1249 *
1250 * The members of the &struct ctl_table structure are used as follows:
1251 *
1252 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1253 * enter a sysctl file
1254 *
1255 * data - a pointer to data for use by proc_handler
1256 *
1257 * maxlen - the maximum size in bytes of the data
1258 *
1259 * mode - the file permissions for the /proc/sys file
1260 *
1261 * child - must be %NULL.
1262 *
1263 * proc_handler - the text handler routine (described below)
1264 *
1265 * extra1, extra2 - extra pointers usable by the proc handler routines
1266 *
1267 * Leaf nodes in the sysctl tree will be represented by a single file
1268 * under /proc; non-leaf nodes will be represented by directories.
1269 *
1270 * There must be a proc_handler routine for any terminal nodes.
1271 * Several default handlers are available to cover common cases -
1272 *
1273 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1274 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1275 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1276 *
1277 * It is the handler's job to read the input buffer from user memory
1278 * and process it. The handler should return 0 on success.
1279 *
1280 * This routine returns %NULL on a failure to register, and a pointer
1281 * to the table header on success.
1282 */
1283struct ctl_table_header *__register_sysctl_table(
1284 struct ctl_table_set *set,
1285 const char *path, struct ctl_table *table)
1286{
1287 struct ctl_table_root *root = set->dir.header.root;
1288 struct ctl_table_header *header;
1289 const char *name, *nextname;
1290 struct ctl_dir *dir;
1291 struct ctl_table *entry;
1292 struct ctl_node *node;
1293 int nr_entries = 0;
1294
1295 for (entry = table; entry->procname; entry++)
1296 nr_entries++;
1297
1298 header = kzalloc(sizeof(struct ctl_table_header) +
1299 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1300 if (!header)
1301 return NULL;
1302
1303 node = (struct ctl_node *)(header + 1);
1304 init_header(header, root, set, node, table);
1305 if (sysctl_check_table(path, table))
1306 goto fail;
1307
1308 spin_lock(&sysctl_lock);
1309 dir = &set->dir;
1310 /* Reference moved down the diretory tree get_subdir */
1311 dir->header.nreg++;
1312 spin_unlock(&sysctl_lock);
1313
1314 /* Find the directory for the ctl_table */
1315 for (name = path; name; name = nextname) {
1316 int namelen;
1317 nextname = strchr(name, '/');
1318 if (nextname) {
1319 namelen = nextname - name;
1320 nextname++;
1321 } else {
1322 namelen = strlen(name);
1323 }
1324 if (namelen == 0)
1325 continue;
1326
1327 dir = get_subdir(dir, name, namelen);
1328 if (IS_ERR(dir))
1329 goto fail;
1330 }
1331
1332 spin_lock(&sysctl_lock);
1333 if (insert_header(dir, header))
1334 goto fail_put_dir_locked;
1335
1336 drop_sysctl_table(&dir->header);
1337 spin_unlock(&sysctl_lock);
1338
1339 return header;
1340
1341fail_put_dir_locked:
1342 drop_sysctl_table(&dir->header);
1343 spin_unlock(&sysctl_lock);
1344fail:
1345 kfree(header);
1346 dump_stack();
1347 return NULL;
1348}
1349
1350/**
1351 * register_sysctl - register a sysctl table
1352 * @path: The path to the directory the sysctl table is in.
1353 * @table: the table structure
1354 *
1355 * Register a sysctl table. @table should be a filled in ctl_table
1356 * array. A completely 0 filled entry terminates the table.
1357 *
1358 * See __register_sysctl_table for more details.
1359 */
1360struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1361{
1362 return __register_sysctl_table(&sysctl_table_root.default_set,
1363 path, table);
1364}
1365EXPORT_SYMBOL(register_sysctl);
1366
1367static char *append_path(const char *path, char *pos, const char *name)
1368{
1369 int namelen;
1370 namelen = strlen(name);
1371 if (((pos - path) + namelen + 2) >= PATH_MAX)
1372 return NULL;
1373 memcpy(pos, name, namelen);
1374 pos[namelen] = '/';
1375 pos[namelen + 1] = '\0';
1376 pos += namelen + 1;
1377 return pos;
1378}
1379
1380static int count_subheaders(struct ctl_table *table)
1381{
1382 int has_files = 0;
1383 int nr_subheaders = 0;
1384 struct ctl_table *entry;
1385
1386 /* special case: no directory and empty directory */
1387 if (!table || !table->procname)
1388 return 1;
1389
1390 for (entry = table; entry->procname; entry++) {
1391 if (entry->child)
1392 nr_subheaders += count_subheaders(entry->child);
1393 else
1394 has_files = 1;
1395 }
1396 return nr_subheaders + has_files;
1397}
1398
1399static int register_leaf_sysctl_tables(const char *path, char *pos,
1400 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1401 struct ctl_table *table)
1402{
1403 struct ctl_table *ctl_table_arg = NULL;
1404 struct ctl_table *entry, *files;
1405 int nr_files = 0;
1406 int nr_dirs = 0;
1407 int err = -ENOMEM;
1408
1409 for (entry = table; entry->procname; entry++) {
1410 if (entry->child)
1411 nr_dirs++;
1412 else
1413 nr_files++;
1414 }
1415
1416 files = table;
1417 /* If there are mixed files and directories we need a new table */
1418 if (nr_dirs && nr_files) {
1419 struct ctl_table *new;
1420 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1421 GFP_KERNEL);
1422 if (!files)
1423 goto out;
1424
1425 ctl_table_arg = files;
1426 for (new = files, entry = table; entry->procname; entry++) {
1427 if (entry->child)
1428 continue;
1429 *new = *entry;
1430 new++;
1431 }
1432 }
1433
1434 /* Register everything except a directory full of subdirectories */
1435 if (nr_files || !nr_dirs) {
1436 struct ctl_table_header *header;
1437 header = __register_sysctl_table(set, path, files);
1438 if (!header) {
1439 kfree(ctl_table_arg);
1440 goto out;
1441 }
1442
1443 /* Remember if we need to free the file table */
1444 header->ctl_table_arg = ctl_table_arg;
1445 **subheader = header;
1446 (*subheader)++;
1447 }
1448
1449 /* Recurse into the subdirectories. */
1450 for (entry = table; entry->procname; entry++) {
1451 char *child_pos;
1452
1453 if (!entry->child)
1454 continue;
1455
1456 err = -ENAMETOOLONG;
1457 child_pos = append_path(path, pos, entry->procname);
1458 if (!child_pos)
1459 goto out;
1460
1461 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1462 set, entry->child);
1463 pos[0] = '\0';
1464 if (err)
1465 goto out;
1466 }
1467 err = 0;
1468out:
1469 /* On failure our caller will unregister all registered subheaders */
1470 return err;
1471}
1472
1473/**
1474 * __register_sysctl_paths - register a sysctl table hierarchy
1475 * @set: Sysctl tree to register on
1476 * @path: The path to the directory the sysctl table is in.
1477 * @table: the top-level table structure
1478 *
1479 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1480 * array. A completely 0 filled entry terminates the table.
1481 *
1482 * See __register_sysctl_table for more details.
1483 */
1484struct ctl_table_header *__register_sysctl_paths(
1485 struct ctl_table_set *set,
1486 const struct ctl_path *path, struct ctl_table *table)
1487{
1488 struct ctl_table *ctl_table_arg = table;
1489 int nr_subheaders = count_subheaders(table);
1490 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1491 const struct ctl_path *component;
1492 char *new_path, *pos;
1493
1494 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1495 if (!new_path)
1496 return NULL;
1497
1498 pos[0] = '\0';
1499 for (component = path; component->procname; component++) {
1500 pos = append_path(new_path, pos, component->procname);
1501 if (!pos)
1502 goto out;
1503 }
1504 while (table->procname && table->child && !table[1].procname) {
1505 pos = append_path(new_path, pos, table->procname);
1506 if (!pos)
1507 goto out;
1508 table = table->child;
1509 }
1510 if (nr_subheaders == 1) {
1511 header = __register_sysctl_table(set, new_path, table);
1512 if (header)
1513 header->ctl_table_arg = ctl_table_arg;
1514 } else {
1515 header = kzalloc(sizeof(*header) +
1516 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1517 if (!header)
1518 goto out;
1519
1520 subheaders = (struct ctl_table_header **) (header + 1);
1521 subheader = subheaders;
1522 header->ctl_table_arg = ctl_table_arg;
1523
1524 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1525 set, table))
1526 goto err_register_leaves;
1527 }
1528
1529out:
1530 kfree(new_path);
1531 return header;
1532
1533err_register_leaves:
1534 while (subheader > subheaders) {
1535 struct ctl_table_header *subh = *(--subheader);
1536 struct ctl_table *table = subh->ctl_table_arg;
1537 unregister_sysctl_table(subh);
1538 kfree(table);
1539 }
1540 kfree(header);
1541 header = NULL;
1542 goto out;
1543}
1544
1545/**
1546 * register_sysctl_table_path - register a sysctl table hierarchy
1547 * @path: The path to the directory the sysctl table is in.
1548 * @table: the top-level table structure
1549 *
1550 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1551 * array. A completely 0 filled entry terminates the table.
1552 *
1553 * See __register_sysctl_paths for more details.
1554 */
1555struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1556 struct ctl_table *table)
1557{
1558 return __register_sysctl_paths(&sysctl_table_root.default_set,
1559 path, table);
1560}
1561EXPORT_SYMBOL(register_sysctl_paths);
1562
1563/**
1564 * register_sysctl_table - register a sysctl table hierarchy
1565 * @table: the top-level table structure
1566 *
1567 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1568 * array. A completely 0 filled entry terminates the table.
1569 *
1570 * See register_sysctl_paths for more details.
1571 */
1572struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1573{
1574 static const struct ctl_path null_path[] = { {} };
1575
1576 return register_sysctl_paths(null_path, table);
1577}
1578EXPORT_SYMBOL(register_sysctl_table);
1579
1580static void put_links(struct ctl_table_header *header)
1581{
1582 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1583 struct ctl_table_root *root = header->root;
1584 struct ctl_dir *parent = header->parent;
1585 struct ctl_dir *core_parent;
1586 struct ctl_table *entry;
1587
1588 if (header->set == root_set)
1589 return;
1590
1591 core_parent = xlate_dir(root_set, parent);
1592 if (IS_ERR(core_parent))
1593 return;
1594
1595 for (entry = header->ctl_table; entry->procname; entry++) {
1596 struct ctl_table_header *link_head;
1597 struct ctl_table *link;
1598 const char *name = entry->procname;
1599
1600 link = find_entry(&link_head, core_parent, name, strlen(name));
1601 if (link &&
1602 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1603 (S_ISLNK(link->mode) && (link->data == root)))) {
1604 drop_sysctl_table(link_head);
1605 }
1606 else {
1607 pr_err("sysctl link missing during unregister: ");
1608 sysctl_print_dir(parent);
1609 pr_cont("/%s\n", name);
1610 }
1611 }
1612}
1613
1614static void drop_sysctl_table(struct ctl_table_header *header)
1615{
1616 struct ctl_dir *parent = header->parent;
1617
1618 if (--header->nreg)
1619 return;
1620
1621 put_links(header);
1622 start_unregistering(header);
1623 if (!--header->count)
1624 kfree_rcu(header, rcu);
1625
1626 if (parent)
1627 drop_sysctl_table(&parent->header);
1628}
1629
1630/**
1631 * unregister_sysctl_table - unregister a sysctl table hierarchy
1632 * @header: the header returned from register_sysctl_table
1633 *
1634 * Unregisters the sysctl table and all children. proc entries may not
1635 * actually be removed until they are no longer used by anyone.
1636 */
1637void unregister_sysctl_table(struct ctl_table_header * header)
1638{
1639 int nr_subheaders;
1640 might_sleep();
1641
1642 if (header == NULL)
1643 return;
1644
1645 nr_subheaders = count_subheaders(header->ctl_table_arg);
1646 if (unlikely(nr_subheaders > 1)) {
1647 struct ctl_table_header **subheaders;
1648 int i;
1649
1650 subheaders = (struct ctl_table_header **)(header + 1);
1651 for (i = nr_subheaders -1; i >= 0; i--) {
1652 struct ctl_table_header *subh = subheaders[i];
1653 struct ctl_table *table = subh->ctl_table_arg;
1654 unregister_sysctl_table(subh);
1655 kfree(table);
1656 }
1657 kfree(header);
1658 return;
1659 }
1660
1661 spin_lock(&sysctl_lock);
1662 drop_sysctl_table(header);
1663 spin_unlock(&sysctl_lock);
1664}
1665EXPORT_SYMBOL(unregister_sysctl_table);
1666
1667void setup_sysctl_set(struct ctl_table_set *set,
1668 struct ctl_table_root *root,
1669 int (*is_seen)(struct ctl_table_set *))
1670{
1671 memset(set, 0, sizeof(*set));
1672 set->is_seen = is_seen;
1673 init_header(&set->dir.header, root, set, NULL, root_table);
1674}
1675
1676void retire_sysctl_set(struct ctl_table_set *set)
1677{
1678 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1679}
1680
1681int __init proc_sys_init(void)
1682{
1683 struct proc_dir_entry *proc_sys_root;
1684
1685 proc_sys_root = proc_mkdir("sys", NULL);
1686 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1687 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1688 proc_sys_root->nlink = 0;
1689
1690 return sysctl_init();
1691}