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 <linux/bpf-cgroup.h>
17#include <linux/mount.h>
18#include "internal.h"
19
20static const struct dentry_operations proc_sys_dentry_operations;
21static const struct file_operations proc_sys_file_operations;
22static const struct inode_operations proc_sys_inode_operations;
23static const struct file_operations proc_sys_dir_file_operations;
24static const struct inode_operations proc_sys_dir_operations;
25
26/* shared constants to be used in various sysctls */
27const int sysctl_vals[] = { 0, 1, INT_MAX };
28EXPORT_SYMBOL(sysctl_vals);
29
30/* Support for permanently empty directories */
31
32struct ctl_table sysctl_mount_point[] = {
33 { }
34};
35
36static bool is_empty_dir(struct ctl_table_header *head)
37{
38 return head->ctl_table[0].child == sysctl_mount_point;
39}
40
41static void set_empty_dir(struct ctl_dir *dir)
42{
43 dir->header.ctl_table[0].child = sysctl_mount_point;
44}
45
46static void clear_empty_dir(struct ctl_dir *dir)
47
48{
49 dir->header.ctl_table[0].child = NULL;
50}
51
52void proc_sys_poll_notify(struct ctl_table_poll *poll)
53{
54 if (!poll)
55 return;
56
57 atomic_inc(&poll->event);
58 wake_up_interruptible(&poll->wait);
59}
60
61static struct ctl_table root_table[] = {
62 {
63 .procname = "",
64 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
65 },
66 { }
67};
68static struct ctl_table_root sysctl_table_root = {
69 .default_set.dir.header = {
70 {{.count = 1,
71 .nreg = 1,
72 .ctl_table = root_table }},
73 .ctl_table_arg = root_table,
74 .root = &sysctl_table_root,
75 .set = &sysctl_table_root.default_set,
76 },
77};
78
79static DEFINE_SPINLOCK(sysctl_lock);
80
81static void drop_sysctl_table(struct ctl_table_header *header);
82static int sysctl_follow_link(struct ctl_table_header **phead,
83 struct ctl_table **pentry);
84static int insert_links(struct ctl_table_header *head);
85static void put_links(struct ctl_table_header *header);
86
87static void sysctl_print_dir(struct ctl_dir *dir)
88{
89 if (dir->header.parent)
90 sysctl_print_dir(dir->header.parent);
91 pr_cont("%s/", dir->header.ctl_table[0].procname);
92}
93
94static int namecmp(const char *name1, int len1, const char *name2, int len2)
95{
96 int minlen;
97 int cmp;
98
99 minlen = len1;
100 if (minlen > len2)
101 minlen = len2;
102
103 cmp = memcmp(name1, name2, minlen);
104 if (cmp == 0)
105 cmp = len1 - len2;
106 return cmp;
107}
108
109/* Called under sysctl_lock */
110static struct ctl_table *find_entry(struct ctl_table_header **phead,
111 struct ctl_dir *dir, const char *name, int namelen)
112{
113 struct ctl_table_header *head;
114 struct ctl_table *entry;
115 struct rb_node *node = dir->root.rb_node;
116
117 while (node)
118 {
119 struct ctl_node *ctl_node;
120 const char *procname;
121 int cmp;
122
123 ctl_node = rb_entry(node, struct ctl_node, node);
124 head = ctl_node->header;
125 entry = &head->ctl_table[ctl_node - head->node];
126 procname = entry->procname;
127
128 cmp = namecmp(name, namelen, procname, strlen(procname));
129 if (cmp < 0)
130 node = node->rb_left;
131 else if (cmp > 0)
132 node = node->rb_right;
133 else {
134 *phead = head;
135 return entry;
136 }
137 }
138 return NULL;
139}
140
141static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
142{
143 struct rb_node *node = &head->node[entry - head->ctl_table].node;
144 struct rb_node **p = &head->parent->root.rb_node;
145 struct rb_node *parent = NULL;
146 const char *name = entry->procname;
147 int namelen = strlen(name);
148
149 while (*p) {
150 struct ctl_table_header *parent_head;
151 struct ctl_table *parent_entry;
152 struct ctl_node *parent_node;
153 const char *parent_name;
154 int cmp;
155
156 parent = *p;
157 parent_node = rb_entry(parent, struct ctl_node, node);
158 parent_head = parent_node->header;
159 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
160 parent_name = parent_entry->procname;
161
162 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
163 if (cmp < 0)
164 p = &(*p)->rb_left;
165 else if (cmp > 0)
166 p = &(*p)->rb_right;
167 else {
168 pr_err("sysctl duplicate entry: ");
169 sysctl_print_dir(head->parent);
170 pr_cont("/%s\n", entry->procname);
171 return -EEXIST;
172 }
173 }
174
175 rb_link_node(node, parent, p);
176 rb_insert_color(node, &head->parent->root);
177 return 0;
178}
179
180static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
181{
182 struct rb_node *node = &head->node[entry - head->ctl_table].node;
183
184 rb_erase(node, &head->parent->root);
185}
186
187static void init_header(struct ctl_table_header *head,
188 struct ctl_table_root *root, struct ctl_table_set *set,
189 struct ctl_node *node, struct ctl_table *table)
190{
191 head->ctl_table = table;
192 head->ctl_table_arg = table;
193 head->used = 0;
194 head->count = 1;
195 head->nreg = 1;
196 head->unregistering = NULL;
197 head->root = root;
198 head->set = set;
199 head->parent = NULL;
200 head->node = node;
201 INIT_HLIST_HEAD(&head->inodes);
202 if (node) {
203 struct ctl_table *entry;
204 for (entry = table; entry->procname; entry++, node++)
205 node->header = head;
206 }
207}
208
209static void erase_header(struct ctl_table_header *head)
210{
211 struct ctl_table *entry;
212 for (entry = head->ctl_table; entry->procname; entry++)
213 erase_entry(head, entry);
214}
215
216static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
217{
218 struct ctl_table *entry;
219 int err;
220
221 /* Is this a permanently empty directory? */
222 if (is_empty_dir(&dir->header))
223 return -EROFS;
224
225 /* Am I creating a permanently empty directory? */
226 if (header->ctl_table == sysctl_mount_point) {
227 if (!RB_EMPTY_ROOT(&dir->root))
228 return -EINVAL;
229 set_empty_dir(dir);
230 }
231
232 dir->header.nreg++;
233 header->parent = dir;
234 err = insert_links(header);
235 if (err)
236 goto fail_links;
237 for (entry = header->ctl_table; entry->procname; entry++) {
238 err = insert_entry(header, entry);
239 if (err)
240 goto fail;
241 }
242 return 0;
243fail:
244 erase_header(header);
245 put_links(header);
246fail_links:
247 if (header->ctl_table == sysctl_mount_point)
248 clear_empty_dir(dir);
249 header->parent = NULL;
250 drop_sysctl_table(&dir->header);
251 return err;
252}
253
254/* called under sysctl_lock */
255static int use_table(struct ctl_table_header *p)
256{
257 if (unlikely(p->unregistering))
258 return 0;
259 p->used++;
260 return 1;
261}
262
263/* called under sysctl_lock */
264static void unuse_table(struct ctl_table_header *p)
265{
266 if (!--p->used)
267 if (unlikely(p->unregistering))
268 complete(p->unregistering);
269}
270
271static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
272{
273 proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
274}
275
276/* called under sysctl_lock, will reacquire if has to wait */
277static void start_unregistering(struct ctl_table_header *p)
278{
279 /*
280 * if p->used is 0, nobody will ever touch that entry again;
281 * we'll eliminate all paths to it before dropping sysctl_lock
282 */
283 if (unlikely(p->used)) {
284 struct completion wait;
285 init_completion(&wait);
286 p->unregistering = &wait;
287 spin_unlock(&sysctl_lock);
288 wait_for_completion(&wait);
289 } else {
290 /* anything non-NULL; we'll never dereference it */
291 p->unregistering = ERR_PTR(-EINVAL);
292 spin_unlock(&sysctl_lock);
293 }
294 /*
295 * Invalidate dentries for unregistered sysctls: namespaced sysctls
296 * can have duplicate names and contaminate dcache very badly.
297 */
298 proc_sys_invalidate_dcache(p);
299 /*
300 * do not remove from the list until nobody holds it; walking the
301 * list in do_sysctl() relies on that.
302 */
303 spin_lock(&sysctl_lock);
304 erase_header(p);
305}
306
307static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
308{
309 BUG_ON(!head);
310 spin_lock(&sysctl_lock);
311 if (!use_table(head))
312 head = ERR_PTR(-ENOENT);
313 spin_unlock(&sysctl_lock);
314 return head;
315}
316
317static void sysctl_head_finish(struct ctl_table_header *head)
318{
319 if (!head)
320 return;
321 spin_lock(&sysctl_lock);
322 unuse_table(head);
323 spin_unlock(&sysctl_lock);
324}
325
326static struct ctl_table_set *
327lookup_header_set(struct ctl_table_root *root)
328{
329 struct ctl_table_set *set = &root->default_set;
330 if (root->lookup)
331 set = root->lookup(root);
332 return set;
333}
334
335static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
336 struct ctl_dir *dir,
337 const char *name, int namelen)
338{
339 struct ctl_table_header *head;
340 struct ctl_table *entry;
341
342 spin_lock(&sysctl_lock);
343 entry = find_entry(&head, dir, name, namelen);
344 if (entry && use_table(head))
345 *phead = head;
346 else
347 entry = NULL;
348 spin_unlock(&sysctl_lock);
349 return entry;
350}
351
352static struct ctl_node *first_usable_entry(struct rb_node *node)
353{
354 struct ctl_node *ctl_node;
355
356 for (;node; node = rb_next(node)) {
357 ctl_node = rb_entry(node, struct ctl_node, node);
358 if (use_table(ctl_node->header))
359 return ctl_node;
360 }
361 return NULL;
362}
363
364static void first_entry(struct ctl_dir *dir,
365 struct ctl_table_header **phead, struct ctl_table **pentry)
366{
367 struct ctl_table_header *head = NULL;
368 struct ctl_table *entry = NULL;
369 struct ctl_node *ctl_node;
370
371 spin_lock(&sysctl_lock);
372 ctl_node = first_usable_entry(rb_first(&dir->root));
373 spin_unlock(&sysctl_lock);
374 if (ctl_node) {
375 head = ctl_node->header;
376 entry = &head->ctl_table[ctl_node - head->node];
377 }
378 *phead = head;
379 *pentry = entry;
380}
381
382static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
383{
384 struct ctl_table_header *head = *phead;
385 struct ctl_table *entry = *pentry;
386 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
387
388 spin_lock(&sysctl_lock);
389 unuse_table(head);
390
391 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
392 spin_unlock(&sysctl_lock);
393 head = NULL;
394 if (ctl_node) {
395 head = ctl_node->header;
396 entry = &head->ctl_table[ctl_node - head->node];
397 }
398 *phead = head;
399 *pentry = entry;
400}
401
402/*
403 * sysctl_perm does NOT grant the superuser all rights automatically, because
404 * some sysctl variables are readonly even to root.
405 */
406
407static int test_perm(int mode, int op)
408{
409 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
410 mode >>= 6;
411 else if (in_egroup_p(GLOBAL_ROOT_GID))
412 mode >>= 3;
413 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
414 return 0;
415 return -EACCES;
416}
417
418static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
419{
420 struct ctl_table_root *root = head->root;
421 int mode;
422
423 if (root->permissions)
424 mode = root->permissions(head, table);
425 else
426 mode = table->mode;
427
428 return test_perm(mode, op);
429}
430
431static struct inode *proc_sys_make_inode(struct super_block *sb,
432 struct ctl_table_header *head, struct ctl_table *table)
433{
434 struct ctl_table_root *root = head->root;
435 struct inode *inode;
436 struct proc_inode *ei;
437
438 inode = new_inode(sb);
439 if (!inode)
440 return ERR_PTR(-ENOMEM);
441
442 inode->i_ino = get_next_ino();
443
444 ei = PROC_I(inode);
445
446 spin_lock(&sysctl_lock);
447 if (unlikely(head->unregistering)) {
448 spin_unlock(&sysctl_lock);
449 iput(inode);
450 return ERR_PTR(-ENOENT);
451 }
452 ei->sysctl = head;
453 ei->sysctl_entry = table;
454 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
455 head->count++;
456 spin_unlock(&sysctl_lock);
457
458 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
459 inode->i_mode = table->mode;
460 if (!S_ISDIR(table->mode)) {
461 inode->i_mode |= S_IFREG;
462 inode->i_op = &proc_sys_inode_operations;
463 inode->i_fop = &proc_sys_file_operations;
464 } else {
465 inode->i_mode |= S_IFDIR;
466 inode->i_op = &proc_sys_dir_operations;
467 inode->i_fop = &proc_sys_dir_file_operations;
468 if (is_empty_dir(head))
469 make_empty_dir_inode(inode);
470 }
471
472 if (root->set_ownership)
473 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
474 else {
475 inode->i_uid = GLOBAL_ROOT_UID;
476 inode->i_gid = GLOBAL_ROOT_GID;
477 }
478
479 return inode;
480}
481
482void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
483{
484 spin_lock(&sysctl_lock);
485 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
486 if (!--head->count)
487 kfree_rcu(head, rcu);
488 spin_unlock(&sysctl_lock);
489}
490
491static struct ctl_table_header *grab_header(struct inode *inode)
492{
493 struct ctl_table_header *head = PROC_I(inode)->sysctl;
494 if (!head)
495 head = &sysctl_table_root.default_set.dir.header;
496 return sysctl_head_grab(head);
497}
498
499static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
500 unsigned int flags)
501{
502 struct ctl_table_header *head = grab_header(dir);
503 struct ctl_table_header *h = NULL;
504 const struct qstr *name = &dentry->d_name;
505 struct ctl_table *p;
506 struct inode *inode;
507 struct dentry *err = ERR_PTR(-ENOENT);
508 struct ctl_dir *ctl_dir;
509 int ret;
510
511 if (IS_ERR(head))
512 return ERR_CAST(head);
513
514 ctl_dir = container_of(head, struct ctl_dir, header);
515
516 p = lookup_entry(&h, ctl_dir, name->name, name->len);
517 if (!p)
518 goto out;
519
520 if (S_ISLNK(p->mode)) {
521 ret = sysctl_follow_link(&h, &p);
522 err = ERR_PTR(ret);
523 if (ret)
524 goto out;
525 }
526
527 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
528 if (IS_ERR(inode)) {
529 err = ERR_CAST(inode);
530 goto out;
531 }
532
533 d_set_d_op(dentry, &proc_sys_dentry_operations);
534 err = d_splice_alias(inode, dentry);
535
536out:
537 if (h)
538 sysctl_head_finish(h);
539 sysctl_head_finish(head);
540 return err;
541}
542
543static ssize_t proc_sys_call_handler(struct file *filp, void __user *ubuf,
544 size_t count, loff_t *ppos, int write)
545{
546 struct inode *inode = file_inode(filp);
547 struct ctl_table_header *head = grab_header(inode);
548 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
549 void *kbuf;
550 ssize_t error;
551
552 if (IS_ERR(head))
553 return PTR_ERR(head);
554
555 /*
556 * At this point we know that the sysctl was not unregistered
557 * and won't be until we finish.
558 */
559 error = -EPERM;
560 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
561 goto out;
562
563 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
564 error = -EINVAL;
565 if (!table->proc_handler)
566 goto out;
567
568 /* don't even try if the size is too large */
569 error = -ENOMEM;
570 if (count >= KMALLOC_MAX_SIZE)
571 goto out;
572
573 if (write) {
574 kbuf = memdup_user_nul(ubuf, count);
575 if (IS_ERR(kbuf)) {
576 error = PTR_ERR(kbuf);
577 goto out;
578 }
579 } else {
580 kbuf = kzalloc(count, GFP_KERNEL);
581 if (!kbuf)
582 goto out;
583 }
584
585 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
586 ppos);
587 if (error)
588 goto out_free_buf;
589
590 /* careful: calling conventions are nasty here */
591 error = table->proc_handler(table, write, kbuf, &count, ppos);
592 if (error)
593 goto out_free_buf;
594
595 if (!write) {
596 error = -EFAULT;
597 if (copy_to_user(ubuf, kbuf, count))
598 goto out_free_buf;
599 }
600
601 error = count;
602out_free_buf:
603 kfree(kbuf);
604out:
605 sysctl_head_finish(head);
606
607 return error;
608}
609
610static ssize_t proc_sys_read(struct file *filp, char __user *buf,
611 size_t count, loff_t *ppos)
612{
613 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
614}
615
616static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
617 size_t count, loff_t *ppos)
618{
619 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
620}
621
622static int proc_sys_open(struct inode *inode, struct file *filp)
623{
624 struct ctl_table_header *head = grab_header(inode);
625 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
626
627 /* sysctl was unregistered */
628 if (IS_ERR(head))
629 return PTR_ERR(head);
630
631 if (table->poll)
632 filp->private_data = proc_sys_poll_event(table->poll);
633
634 sysctl_head_finish(head);
635
636 return 0;
637}
638
639static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
640{
641 struct inode *inode = file_inode(filp);
642 struct ctl_table_header *head = grab_header(inode);
643 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
644 __poll_t ret = DEFAULT_POLLMASK;
645 unsigned long event;
646
647 /* sysctl was unregistered */
648 if (IS_ERR(head))
649 return EPOLLERR | EPOLLHUP;
650
651 if (!table->proc_handler)
652 goto out;
653
654 if (!table->poll)
655 goto out;
656
657 event = (unsigned long)filp->private_data;
658 poll_wait(filp, &table->poll->wait, wait);
659
660 if (event != atomic_read(&table->poll->event)) {
661 filp->private_data = proc_sys_poll_event(table->poll);
662 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
663 }
664
665out:
666 sysctl_head_finish(head);
667
668 return ret;
669}
670
671static bool proc_sys_fill_cache(struct file *file,
672 struct dir_context *ctx,
673 struct ctl_table_header *head,
674 struct ctl_table *table)
675{
676 struct dentry *child, *dir = file->f_path.dentry;
677 struct inode *inode;
678 struct qstr qname;
679 ino_t ino = 0;
680 unsigned type = DT_UNKNOWN;
681
682 qname.name = table->procname;
683 qname.len = strlen(table->procname);
684 qname.hash = full_name_hash(dir, qname.name, qname.len);
685
686 child = d_lookup(dir, &qname);
687 if (!child) {
688 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
689 child = d_alloc_parallel(dir, &qname, &wq);
690 if (IS_ERR(child))
691 return false;
692 if (d_in_lookup(child)) {
693 struct dentry *res;
694 inode = proc_sys_make_inode(dir->d_sb, head, table);
695 if (IS_ERR(inode)) {
696 d_lookup_done(child);
697 dput(child);
698 return false;
699 }
700 d_set_d_op(child, &proc_sys_dentry_operations);
701 res = d_splice_alias(inode, child);
702 d_lookup_done(child);
703 if (unlikely(res)) {
704 if (IS_ERR(res)) {
705 dput(child);
706 return false;
707 }
708 dput(child);
709 child = res;
710 }
711 }
712 }
713 inode = d_inode(child);
714 ino = inode->i_ino;
715 type = inode->i_mode >> 12;
716 dput(child);
717 return dir_emit(ctx, qname.name, qname.len, ino, type);
718}
719
720static bool proc_sys_link_fill_cache(struct file *file,
721 struct dir_context *ctx,
722 struct ctl_table_header *head,
723 struct ctl_table *table)
724{
725 bool ret = true;
726
727 head = sysctl_head_grab(head);
728 if (IS_ERR(head))
729 return false;
730
731 /* It is not an error if we can not follow the link ignore it */
732 if (sysctl_follow_link(&head, &table))
733 goto out;
734
735 ret = proc_sys_fill_cache(file, ctx, head, table);
736out:
737 sysctl_head_finish(head);
738 return ret;
739}
740
741static int scan(struct ctl_table_header *head, struct ctl_table *table,
742 unsigned long *pos, struct file *file,
743 struct dir_context *ctx)
744{
745 bool res;
746
747 if ((*pos)++ < ctx->pos)
748 return true;
749
750 if (unlikely(S_ISLNK(table->mode)))
751 res = proc_sys_link_fill_cache(file, ctx, head, table);
752 else
753 res = proc_sys_fill_cache(file, ctx, head, table);
754
755 if (res)
756 ctx->pos = *pos;
757
758 return res;
759}
760
761static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
762{
763 struct ctl_table_header *head = grab_header(file_inode(file));
764 struct ctl_table_header *h = NULL;
765 struct ctl_table *entry;
766 struct ctl_dir *ctl_dir;
767 unsigned long pos;
768
769 if (IS_ERR(head))
770 return PTR_ERR(head);
771
772 ctl_dir = container_of(head, struct ctl_dir, header);
773
774 if (!dir_emit_dots(file, ctx))
775 goto out;
776
777 pos = 2;
778
779 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
780 if (!scan(h, entry, &pos, file, ctx)) {
781 sysctl_head_finish(h);
782 break;
783 }
784 }
785out:
786 sysctl_head_finish(head);
787 return 0;
788}
789
790static int proc_sys_permission(struct inode *inode, int mask)
791{
792 /*
793 * sysctl entries that are not writeable,
794 * are _NOT_ writeable, capabilities or not.
795 */
796 struct ctl_table_header *head;
797 struct ctl_table *table;
798 int error;
799
800 /* Executable files are not allowed under /proc/sys/ */
801 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
802 return -EACCES;
803
804 head = grab_header(inode);
805 if (IS_ERR(head))
806 return PTR_ERR(head);
807
808 table = PROC_I(inode)->sysctl_entry;
809 if (!table) /* global root - r-xr-xr-x */
810 error = mask & MAY_WRITE ? -EACCES : 0;
811 else /* Use the permissions on the sysctl table entry */
812 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
813
814 sysctl_head_finish(head);
815 return error;
816}
817
818static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
819{
820 struct inode *inode = d_inode(dentry);
821 int error;
822
823 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
824 return -EPERM;
825
826 error = setattr_prepare(dentry, attr);
827 if (error)
828 return error;
829
830 setattr_copy(inode, attr);
831 mark_inode_dirty(inode);
832 return 0;
833}
834
835static int proc_sys_getattr(const struct path *path, struct kstat *stat,
836 u32 request_mask, unsigned int query_flags)
837{
838 struct inode *inode = d_inode(path->dentry);
839 struct ctl_table_header *head = grab_header(inode);
840 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
841
842 if (IS_ERR(head))
843 return PTR_ERR(head);
844
845 generic_fillattr(inode, stat);
846 if (table)
847 stat->mode = (stat->mode & S_IFMT) | table->mode;
848
849 sysctl_head_finish(head);
850 return 0;
851}
852
853static const struct file_operations proc_sys_file_operations = {
854 .open = proc_sys_open,
855 .poll = proc_sys_poll,
856 .read = proc_sys_read,
857 .write = proc_sys_write,
858 .llseek = default_llseek,
859};
860
861static const struct file_operations proc_sys_dir_file_operations = {
862 .read = generic_read_dir,
863 .iterate_shared = proc_sys_readdir,
864 .llseek = generic_file_llseek,
865};
866
867static const struct inode_operations proc_sys_inode_operations = {
868 .permission = proc_sys_permission,
869 .setattr = proc_sys_setattr,
870 .getattr = proc_sys_getattr,
871};
872
873static const struct inode_operations proc_sys_dir_operations = {
874 .lookup = proc_sys_lookup,
875 .permission = proc_sys_permission,
876 .setattr = proc_sys_setattr,
877 .getattr = proc_sys_getattr,
878};
879
880static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
881{
882 if (flags & LOOKUP_RCU)
883 return -ECHILD;
884 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
885}
886
887static int proc_sys_delete(const struct dentry *dentry)
888{
889 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
890}
891
892static int sysctl_is_seen(struct ctl_table_header *p)
893{
894 struct ctl_table_set *set = p->set;
895 int res;
896 spin_lock(&sysctl_lock);
897 if (p->unregistering)
898 res = 0;
899 else if (!set->is_seen)
900 res = 1;
901 else
902 res = set->is_seen(set);
903 spin_unlock(&sysctl_lock);
904 return res;
905}
906
907static int proc_sys_compare(const struct dentry *dentry,
908 unsigned int len, const char *str, const struct qstr *name)
909{
910 struct ctl_table_header *head;
911 struct inode *inode;
912
913 /* Although proc doesn't have negative dentries, rcu-walk means
914 * that inode here can be NULL */
915 /* AV: can it, indeed? */
916 inode = d_inode_rcu(dentry);
917 if (!inode)
918 return 1;
919 if (name->len != len)
920 return 1;
921 if (memcmp(name->name, str, len))
922 return 1;
923 head = rcu_dereference(PROC_I(inode)->sysctl);
924 return !head || !sysctl_is_seen(head);
925}
926
927static const struct dentry_operations proc_sys_dentry_operations = {
928 .d_revalidate = proc_sys_revalidate,
929 .d_delete = proc_sys_delete,
930 .d_compare = proc_sys_compare,
931};
932
933static struct ctl_dir *find_subdir(struct ctl_dir *dir,
934 const char *name, int namelen)
935{
936 struct ctl_table_header *head;
937 struct ctl_table *entry;
938
939 entry = find_entry(&head, dir, name, namelen);
940 if (!entry)
941 return ERR_PTR(-ENOENT);
942 if (!S_ISDIR(entry->mode))
943 return ERR_PTR(-ENOTDIR);
944 return container_of(head, struct ctl_dir, header);
945}
946
947static struct ctl_dir *new_dir(struct ctl_table_set *set,
948 const char *name, int namelen)
949{
950 struct ctl_table *table;
951 struct ctl_dir *new;
952 struct ctl_node *node;
953 char *new_name;
954
955 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
956 sizeof(struct ctl_table)*2 + namelen + 1,
957 GFP_KERNEL);
958 if (!new)
959 return NULL;
960
961 node = (struct ctl_node *)(new + 1);
962 table = (struct ctl_table *)(node + 1);
963 new_name = (char *)(table + 2);
964 memcpy(new_name, name, namelen);
965 new_name[namelen] = '\0';
966 table[0].procname = new_name;
967 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
968 init_header(&new->header, set->dir.header.root, set, node, table);
969
970 return new;
971}
972
973/**
974 * get_subdir - find or create a subdir with the specified name.
975 * @dir: Directory to create the subdirectory in
976 * @name: The name of the subdirectory to find or create
977 * @namelen: The length of name
978 *
979 * Takes a directory with an elevated reference count so we know that
980 * if we drop the lock the directory will not go away. Upon success
981 * the reference is moved from @dir to the returned subdirectory.
982 * Upon error an error code is returned and the reference on @dir is
983 * simply dropped.
984 */
985static struct ctl_dir *get_subdir(struct ctl_dir *dir,
986 const char *name, int namelen)
987{
988 struct ctl_table_set *set = dir->header.set;
989 struct ctl_dir *subdir, *new = NULL;
990 int err;
991
992 spin_lock(&sysctl_lock);
993 subdir = find_subdir(dir, name, namelen);
994 if (!IS_ERR(subdir))
995 goto found;
996 if (PTR_ERR(subdir) != -ENOENT)
997 goto failed;
998
999 spin_unlock(&sysctl_lock);
1000 new = new_dir(set, name, namelen);
1001 spin_lock(&sysctl_lock);
1002 subdir = ERR_PTR(-ENOMEM);
1003 if (!new)
1004 goto failed;
1005
1006 /* Was the subdir added while we dropped the lock? */
1007 subdir = find_subdir(dir, name, namelen);
1008 if (!IS_ERR(subdir))
1009 goto found;
1010 if (PTR_ERR(subdir) != -ENOENT)
1011 goto failed;
1012
1013 /* Nope. Use the our freshly made directory entry. */
1014 err = insert_header(dir, &new->header);
1015 subdir = ERR_PTR(err);
1016 if (err)
1017 goto failed;
1018 subdir = new;
1019found:
1020 subdir->header.nreg++;
1021failed:
1022 if (IS_ERR(subdir)) {
1023 pr_err("sysctl could not get directory: ");
1024 sysctl_print_dir(dir);
1025 pr_cont("/%*.*s %ld\n",
1026 namelen, namelen, name, PTR_ERR(subdir));
1027 }
1028 drop_sysctl_table(&dir->header);
1029 if (new)
1030 drop_sysctl_table(&new->header);
1031 spin_unlock(&sysctl_lock);
1032 return subdir;
1033}
1034
1035static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1036{
1037 struct ctl_dir *parent;
1038 const char *procname;
1039 if (!dir->header.parent)
1040 return &set->dir;
1041 parent = xlate_dir(set, dir->header.parent);
1042 if (IS_ERR(parent))
1043 return parent;
1044 procname = dir->header.ctl_table[0].procname;
1045 return find_subdir(parent, procname, strlen(procname));
1046}
1047
1048static int sysctl_follow_link(struct ctl_table_header **phead,
1049 struct ctl_table **pentry)
1050{
1051 struct ctl_table_header *head;
1052 struct ctl_table_root *root;
1053 struct ctl_table_set *set;
1054 struct ctl_table *entry;
1055 struct ctl_dir *dir;
1056 int ret;
1057
1058 ret = 0;
1059 spin_lock(&sysctl_lock);
1060 root = (*pentry)->data;
1061 set = lookup_header_set(root);
1062 dir = xlate_dir(set, (*phead)->parent);
1063 if (IS_ERR(dir))
1064 ret = PTR_ERR(dir);
1065 else {
1066 const char *procname = (*pentry)->procname;
1067 head = NULL;
1068 entry = find_entry(&head, dir, procname, strlen(procname));
1069 ret = -ENOENT;
1070 if (entry && use_table(head)) {
1071 unuse_table(*phead);
1072 *phead = head;
1073 *pentry = entry;
1074 ret = 0;
1075 }
1076 }
1077
1078 spin_unlock(&sysctl_lock);
1079 return ret;
1080}
1081
1082static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1083{
1084 struct va_format vaf;
1085 va_list args;
1086
1087 va_start(args, fmt);
1088 vaf.fmt = fmt;
1089 vaf.va = &args;
1090
1091 pr_err("sysctl table check failed: %s/%s %pV\n",
1092 path, table->procname, &vaf);
1093
1094 va_end(args);
1095 return -EINVAL;
1096}
1097
1098static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1099{
1100 int err = 0;
1101
1102 if ((table->proc_handler == proc_douintvec) ||
1103 (table->proc_handler == proc_douintvec_minmax)) {
1104 if (table->maxlen != sizeof(unsigned int))
1105 err |= sysctl_err(path, table, "array not allowed");
1106 }
1107
1108 return err;
1109}
1110
1111static int sysctl_check_table(const char *path, struct ctl_table *table)
1112{
1113 int err = 0;
1114 for (; table->procname; table++) {
1115 if (table->child)
1116 err |= sysctl_err(path, table, "Not a file");
1117
1118 if ((table->proc_handler == proc_dostring) ||
1119 (table->proc_handler == proc_dointvec) ||
1120 (table->proc_handler == proc_douintvec) ||
1121 (table->proc_handler == proc_douintvec_minmax) ||
1122 (table->proc_handler == proc_dointvec_minmax) ||
1123 (table->proc_handler == proc_dointvec_jiffies) ||
1124 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1125 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1126 (table->proc_handler == proc_doulongvec_minmax) ||
1127 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1128 if (!table->data)
1129 err |= sysctl_err(path, table, "No data");
1130 if (!table->maxlen)
1131 err |= sysctl_err(path, table, "No maxlen");
1132 else
1133 err |= sysctl_check_table_array(path, table);
1134 }
1135 if (!table->proc_handler)
1136 err |= sysctl_err(path, table, "No proc_handler");
1137
1138 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1139 err |= sysctl_err(path, table, "bogus .mode 0%o",
1140 table->mode);
1141 }
1142 return err;
1143}
1144
1145static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1146 struct ctl_table_root *link_root)
1147{
1148 struct ctl_table *link_table, *entry, *link;
1149 struct ctl_table_header *links;
1150 struct ctl_node *node;
1151 char *link_name;
1152 int nr_entries, name_bytes;
1153
1154 name_bytes = 0;
1155 nr_entries = 0;
1156 for (entry = table; entry->procname; entry++) {
1157 nr_entries++;
1158 name_bytes += strlen(entry->procname) + 1;
1159 }
1160
1161 links = kzalloc(sizeof(struct ctl_table_header) +
1162 sizeof(struct ctl_node)*nr_entries +
1163 sizeof(struct ctl_table)*(nr_entries + 1) +
1164 name_bytes,
1165 GFP_KERNEL);
1166
1167 if (!links)
1168 return NULL;
1169
1170 node = (struct ctl_node *)(links + 1);
1171 link_table = (struct ctl_table *)(node + nr_entries);
1172 link_name = (char *)&link_table[nr_entries + 1];
1173
1174 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1175 int len = strlen(entry->procname) + 1;
1176 memcpy(link_name, entry->procname, len);
1177 link->procname = link_name;
1178 link->mode = S_IFLNK|S_IRWXUGO;
1179 link->data = link_root;
1180 link_name += len;
1181 }
1182 init_header(links, dir->header.root, dir->header.set, node, link_table);
1183 links->nreg = nr_entries;
1184
1185 return links;
1186}
1187
1188static bool get_links(struct ctl_dir *dir,
1189 struct ctl_table *table, struct ctl_table_root *link_root)
1190{
1191 struct ctl_table_header *head;
1192 struct ctl_table *entry, *link;
1193
1194 /* Are there links available for every entry in table? */
1195 for (entry = table; entry->procname; entry++) {
1196 const char *procname = entry->procname;
1197 link = find_entry(&head, dir, procname, strlen(procname));
1198 if (!link)
1199 return false;
1200 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1201 continue;
1202 if (S_ISLNK(link->mode) && (link->data == link_root))
1203 continue;
1204 return false;
1205 }
1206
1207 /* The checks passed. Increase the registration count on the links */
1208 for (entry = table; entry->procname; entry++) {
1209 const char *procname = entry->procname;
1210 link = find_entry(&head, dir, procname, strlen(procname));
1211 head->nreg++;
1212 }
1213 return true;
1214}
1215
1216static int insert_links(struct ctl_table_header *head)
1217{
1218 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1219 struct ctl_dir *core_parent = NULL;
1220 struct ctl_table_header *links;
1221 int err;
1222
1223 if (head->set == root_set)
1224 return 0;
1225
1226 core_parent = xlate_dir(root_set, head->parent);
1227 if (IS_ERR(core_parent))
1228 return 0;
1229
1230 if (get_links(core_parent, head->ctl_table, head->root))
1231 return 0;
1232
1233 core_parent->header.nreg++;
1234 spin_unlock(&sysctl_lock);
1235
1236 links = new_links(core_parent, head->ctl_table, head->root);
1237
1238 spin_lock(&sysctl_lock);
1239 err = -ENOMEM;
1240 if (!links)
1241 goto out;
1242
1243 err = 0;
1244 if (get_links(core_parent, head->ctl_table, head->root)) {
1245 kfree(links);
1246 goto out;
1247 }
1248
1249 err = insert_header(core_parent, links);
1250 if (err)
1251 kfree(links);
1252out:
1253 drop_sysctl_table(&core_parent->header);
1254 return err;
1255}
1256
1257/**
1258 * __register_sysctl_table - register a leaf sysctl table
1259 * @set: Sysctl tree to register on
1260 * @path: The path to the directory the sysctl table is in.
1261 * @table: the top-level table structure
1262 *
1263 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1264 * array. A completely 0 filled entry terminates the table.
1265 *
1266 * The members of the &struct ctl_table structure are used as follows:
1267 *
1268 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1269 * enter a sysctl file
1270 *
1271 * data - a pointer to data for use by proc_handler
1272 *
1273 * maxlen - the maximum size in bytes of the data
1274 *
1275 * mode - the file permissions for the /proc/sys file
1276 *
1277 * child - must be %NULL.
1278 *
1279 * proc_handler - the text handler routine (described below)
1280 *
1281 * extra1, extra2 - extra pointers usable by the proc handler routines
1282 *
1283 * Leaf nodes in the sysctl tree will be represented by a single file
1284 * under /proc; non-leaf nodes will be represented by directories.
1285 *
1286 * There must be a proc_handler routine for any terminal nodes.
1287 * Several default handlers are available to cover common cases -
1288 *
1289 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1290 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1291 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1292 *
1293 * It is the handler's job to read the input buffer from user memory
1294 * and process it. The handler should return 0 on success.
1295 *
1296 * This routine returns %NULL on a failure to register, and a pointer
1297 * to the table header on success.
1298 */
1299struct ctl_table_header *__register_sysctl_table(
1300 struct ctl_table_set *set,
1301 const char *path, struct ctl_table *table)
1302{
1303 struct ctl_table_root *root = set->dir.header.root;
1304 struct ctl_table_header *header;
1305 const char *name, *nextname;
1306 struct ctl_dir *dir;
1307 struct ctl_table *entry;
1308 struct ctl_node *node;
1309 int nr_entries = 0;
1310
1311 for (entry = table; entry->procname; entry++)
1312 nr_entries++;
1313
1314 header = kzalloc(sizeof(struct ctl_table_header) +
1315 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1316 if (!header)
1317 return NULL;
1318
1319 node = (struct ctl_node *)(header + 1);
1320 init_header(header, root, set, node, table);
1321 if (sysctl_check_table(path, table))
1322 goto fail;
1323
1324 spin_lock(&sysctl_lock);
1325 dir = &set->dir;
1326 /* Reference moved down the diretory tree get_subdir */
1327 dir->header.nreg++;
1328 spin_unlock(&sysctl_lock);
1329
1330 /* Find the directory for the ctl_table */
1331 for (name = path; name; name = nextname) {
1332 int namelen;
1333 nextname = strchr(name, '/');
1334 if (nextname) {
1335 namelen = nextname - name;
1336 nextname++;
1337 } else {
1338 namelen = strlen(name);
1339 }
1340 if (namelen == 0)
1341 continue;
1342
1343 dir = get_subdir(dir, name, namelen);
1344 if (IS_ERR(dir))
1345 goto fail;
1346 }
1347
1348 spin_lock(&sysctl_lock);
1349 if (insert_header(dir, header))
1350 goto fail_put_dir_locked;
1351
1352 drop_sysctl_table(&dir->header);
1353 spin_unlock(&sysctl_lock);
1354
1355 return header;
1356
1357fail_put_dir_locked:
1358 drop_sysctl_table(&dir->header);
1359 spin_unlock(&sysctl_lock);
1360fail:
1361 kfree(header);
1362 dump_stack();
1363 return NULL;
1364}
1365
1366/**
1367 * register_sysctl - register a sysctl table
1368 * @path: The path to the directory the sysctl table is in.
1369 * @table: the table structure
1370 *
1371 * Register a sysctl table. @table should be a filled in ctl_table
1372 * array. A completely 0 filled entry terminates the table.
1373 *
1374 * See __register_sysctl_table for more details.
1375 */
1376struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1377{
1378 return __register_sysctl_table(&sysctl_table_root.default_set,
1379 path, table);
1380}
1381EXPORT_SYMBOL(register_sysctl);
1382
1383static char *append_path(const char *path, char *pos, const char *name)
1384{
1385 int namelen;
1386 namelen = strlen(name);
1387 if (((pos - path) + namelen + 2) >= PATH_MAX)
1388 return NULL;
1389 memcpy(pos, name, namelen);
1390 pos[namelen] = '/';
1391 pos[namelen + 1] = '\0';
1392 pos += namelen + 1;
1393 return pos;
1394}
1395
1396static int count_subheaders(struct ctl_table *table)
1397{
1398 int has_files = 0;
1399 int nr_subheaders = 0;
1400 struct ctl_table *entry;
1401
1402 /* special case: no directory and empty directory */
1403 if (!table || !table->procname)
1404 return 1;
1405
1406 for (entry = table; entry->procname; entry++) {
1407 if (entry->child)
1408 nr_subheaders += count_subheaders(entry->child);
1409 else
1410 has_files = 1;
1411 }
1412 return nr_subheaders + has_files;
1413}
1414
1415static int register_leaf_sysctl_tables(const char *path, char *pos,
1416 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1417 struct ctl_table *table)
1418{
1419 struct ctl_table *ctl_table_arg = NULL;
1420 struct ctl_table *entry, *files;
1421 int nr_files = 0;
1422 int nr_dirs = 0;
1423 int err = -ENOMEM;
1424
1425 for (entry = table; entry->procname; entry++) {
1426 if (entry->child)
1427 nr_dirs++;
1428 else
1429 nr_files++;
1430 }
1431
1432 files = table;
1433 /* If there are mixed files and directories we need a new table */
1434 if (nr_dirs && nr_files) {
1435 struct ctl_table *new;
1436 files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1437 GFP_KERNEL);
1438 if (!files)
1439 goto out;
1440
1441 ctl_table_arg = files;
1442 for (new = files, entry = table; entry->procname; entry++) {
1443 if (entry->child)
1444 continue;
1445 *new = *entry;
1446 new++;
1447 }
1448 }
1449
1450 /* Register everything except a directory full of subdirectories */
1451 if (nr_files || !nr_dirs) {
1452 struct ctl_table_header *header;
1453 header = __register_sysctl_table(set, path, files);
1454 if (!header) {
1455 kfree(ctl_table_arg);
1456 goto out;
1457 }
1458
1459 /* Remember if we need to free the file table */
1460 header->ctl_table_arg = ctl_table_arg;
1461 **subheader = header;
1462 (*subheader)++;
1463 }
1464
1465 /* Recurse into the subdirectories. */
1466 for (entry = table; entry->procname; entry++) {
1467 char *child_pos;
1468
1469 if (!entry->child)
1470 continue;
1471
1472 err = -ENAMETOOLONG;
1473 child_pos = append_path(path, pos, entry->procname);
1474 if (!child_pos)
1475 goto out;
1476
1477 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1478 set, entry->child);
1479 pos[0] = '\0';
1480 if (err)
1481 goto out;
1482 }
1483 err = 0;
1484out:
1485 /* On failure our caller will unregister all registered subheaders */
1486 return err;
1487}
1488
1489/**
1490 * __register_sysctl_paths - register a sysctl table hierarchy
1491 * @set: Sysctl tree to register on
1492 * @path: The path to the directory the sysctl table is in.
1493 * @table: the top-level table structure
1494 *
1495 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1496 * array. A completely 0 filled entry terminates the table.
1497 *
1498 * See __register_sysctl_table for more details.
1499 */
1500struct ctl_table_header *__register_sysctl_paths(
1501 struct ctl_table_set *set,
1502 const struct ctl_path *path, struct ctl_table *table)
1503{
1504 struct ctl_table *ctl_table_arg = table;
1505 int nr_subheaders = count_subheaders(table);
1506 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1507 const struct ctl_path *component;
1508 char *new_path, *pos;
1509
1510 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1511 if (!new_path)
1512 return NULL;
1513
1514 pos[0] = '\0';
1515 for (component = path; component->procname; component++) {
1516 pos = append_path(new_path, pos, component->procname);
1517 if (!pos)
1518 goto out;
1519 }
1520 while (table->procname && table->child && !table[1].procname) {
1521 pos = append_path(new_path, pos, table->procname);
1522 if (!pos)
1523 goto out;
1524 table = table->child;
1525 }
1526 if (nr_subheaders == 1) {
1527 header = __register_sysctl_table(set, new_path, table);
1528 if (header)
1529 header->ctl_table_arg = ctl_table_arg;
1530 } else {
1531 header = kzalloc(sizeof(*header) +
1532 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1533 if (!header)
1534 goto out;
1535
1536 subheaders = (struct ctl_table_header **) (header + 1);
1537 subheader = subheaders;
1538 header->ctl_table_arg = ctl_table_arg;
1539
1540 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1541 set, table))
1542 goto err_register_leaves;
1543 }
1544
1545out:
1546 kfree(new_path);
1547 return header;
1548
1549err_register_leaves:
1550 while (subheader > subheaders) {
1551 struct ctl_table_header *subh = *(--subheader);
1552 struct ctl_table *table = subh->ctl_table_arg;
1553 unregister_sysctl_table(subh);
1554 kfree(table);
1555 }
1556 kfree(header);
1557 header = NULL;
1558 goto out;
1559}
1560
1561/**
1562 * register_sysctl_table_path - register a sysctl table hierarchy
1563 * @path: The path to the directory the sysctl table is in.
1564 * @table: the top-level table structure
1565 *
1566 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1567 * array. A completely 0 filled entry terminates the table.
1568 *
1569 * See __register_sysctl_paths for more details.
1570 */
1571struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1572 struct ctl_table *table)
1573{
1574 return __register_sysctl_paths(&sysctl_table_root.default_set,
1575 path, table);
1576}
1577EXPORT_SYMBOL(register_sysctl_paths);
1578
1579/**
1580 * register_sysctl_table - register a sysctl table hierarchy
1581 * @table: the top-level table structure
1582 *
1583 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1584 * array. A completely 0 filled entry terminates the table.
1585 *
1586 * See register_sysctl_paths for more details.
1587 */
1588struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1589{
1590 static const struct ctl_path null_path[] = { {} };
1591
1592 return register_sysctl_paths(null_path, table);
1593}
1594EXPORT_SYMBOL(register_sysctl_table);
1595
1596static void put_links(struct ctl_table_header *header)
1597{
1598 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1599 struct ctl_table_root *root = header->root;
1600 struct ctl_dir *parent = header->parent;
1601 struct ctl_dir *core_parent;
1602 struct ctl_table *entry;
1603
1604 if (header->set == root_set)
1605 return;
1606
1607 core_parent = xlate_dir(root_set, parent);
1608 if (IS_ERR(core_parent))
1609 return;
1610
1611 for (entry = header->ctl_table; entry->procname; entry++) {
1612 struct ctl_table_header *link_head;
1613 struct ctl_table *link;
1614 const char *name = entry->procname;
1615
1616 link = find_entry(&link_head, core_parent, name, strlen(name));
1617 if (link &&
1618 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1619 (S_ISLNK(link->mode) && (link->data == root)))) {
1620 drop_sysctl_table(link_head);
1621 }
1622 else {
1623 pr_err("sysctl link missing during unregister: ");
1624 sysctl_print_dir(parent);
1625 pr_cont("/%s\n", name);
1626 }
1627 }
1628}
1629
1630static void drop_sysctl_table(struct ctl_table_header *header)
1631{
1632 struct ctl_dir *parent = header->parent;
1633
1634 if (--header->nreg)
1635 return;
1636
1637 if (parent) {
1638 put_links(header);
1639 start_unregistering(header);
1640 }
1641
1642 if (!--header->count)
1643 kfree_rcu(header, rcu);
1644
1645 if (parent)
1646 drop_sysctl_table(&parent->header);
1647}
1648
1649/**
1650 * unregister_sysctl_table - unregister a sysctl table hierarchy
1651 * @header: the header returned from register_sysctl_table
1652 *
1653 * Unregisters the sysctl table and all children. proc entries may not
1654 * actually be removed until they are no longer used by anyone.
1655 */
1656void unregister_sysctl_table(struct ctl_table_header * header)
1657{
1658 int nr_subheaders;
1659 might_sleep();
1660
1661 if (header == NULL)
1662 return;
1663
1664 nr_subheaders = count_subheaders(header->ctl_table_arg);
1665 if (unlikely(nr_subheaders > 1)) {
1666 struct ctl_table_header **subheaders;
1667 int i;
1668
1669 subheaders = (struct ctl_table_header **)(header + 1);
1670 for (i = nr_subheaders -1; i >= 0; i--) {
1671 struct ctl_table_header *subh = subheaders[i];
1672 struct ctl_table *table = subh->ctl_table_arg;
1673 unregister_sysctl_table(subh);
1674 kfree(table);
1675 }
1676 kfree(header);
1677 return;
1678 }
1679
1680 spin_lock(&sysctl_lock);
1681 drop_sysctl_table(header);
1682 spin_unlock(&sysctl_lock);
1683}
1684EXPORT_SYMBOL(unregister_sysctl_table);
1685
1686void setup_sysctl_set(struct ctl_table_set *set,
1687 struct ctl_table_root *root,
1688 int (*is_seen)(struct ctl_table_set *))
1689{
1690 memset(set, 0, sizeof(*set));
1691 set->is_seen = is_seen;
1692 init_header(&set->dir.header, root, set, NULL, root_table);
1693}
1694
1695void retire_sysctl_set(struct ctl_table_set *set)
1696{
1697 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1698}
1699
1700int __init proc_sys_init(void)
1701{
1702 struct proc_dir_entry *proc_sys_root;
1703
1704 proc_sys_root = proc_mkdir("sys", NULL);
1705 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1706 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1707 proc_sys_root->nlink = 0;
1708
1709 return sysctl_init();
1710}
1711
1712struct sysctl_alias {
1713 const char *kernel_param;
1714 const char *sysctl_param;
1715};
1716
1717/*
1718 * Historically some settings had both sysctl and a command line parameter.
1719 * With the generic sysctl. parameter support, we can handle them at a single
1720 * place and only keep the historical name for compatibility. This is not meant
1721 * to add brand new aliases. When adding existing aliases, consider whether
1722 * the possibly different moment of changing the value (e.g. from early_param
1723 * to the moment do_sysctl_args() is called) is an issue for the specific
1724 * parameter.
1725 */
1726static const struct sysctl_alias sysctl_aliases[] = {
1727 {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1728 {"hung_task_panic", "kernel.hung_task_panic" },
1729 {"numa_zonelist_order", "vm.numa_zonelist_order" },
1730 {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1731 {"softlockup_panic", "kernel.softlockup_panic" },
1732 { }
1733};
1734
1735static const char *sysctl_find_alias(char *param)
1736{
1737 const struct sysctl_alias *alias;
1738
1739 for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1740 if (strcmp(alias->kernel_param, param) == 0)
1741 return alias->sysctl_param;
1742 }
1743
1744 return NULL;
1745}
1746
1747/* Set sysctl value passed on kernel command line. */
1748static int process_sysctl_arg(char *param, char *val,
1749 const char *unused, void *arg)
1750{
1751 char *path;
1752 struct vfsmount **proc_mnt = arg;
1753 struct file_system_type *proc_fs_type;
1754 struct file *file;
1755 int len;
1756 int err;
1757 loff_t pos = 0;
1758 ssize_t wret;
1759
1760 if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1761 param += sizeof("sysctl") - 1;
1762
1763 if (param[0] != '/' && param[0] != '.')
1764 return 0;
1765
1766 param++;
1767 } else {
1768 param = (char *) sysctl_find_alias(param);
1769 if (!param)
1770 return 0;
1771 }
1772
1773 /*
1774 * To set sysctl options, we use a temporary mount of proc, look up the
1775 * respective sys/ file and write to it. To avoid mounting it when no
1776 * options were given, we mount it only when the first sysctl option is
1777 * found. Why not a persistent mount? There are problems with a
1778 * persistent mount of proc in that it forces userspace not to use any
1779 * proc mount options.
1780 */
1781 if (!*proc_mnt) {
1782 proc_fs_type = get_fs_type("proc");
1783 if (!proc_fs_type) {
1784 pr_err("Failed to find procfs to set sysctl from command line\n");
1785 return 0;
1786 }
1787 *proc_mnt = kern_mount(proc_fs_type);
1788 put_filesystem(proc_fs_type);
1789 if (IS_ERR(*proc_mnt)) {
1790 pr_err("Failed to mount procfs to set sysctl from command line\n");
1791 return 0;
1792 }
1793 }
1794
1795 path = kasprintf(GFP_KERNEL, "sys/%s", param);
1796 if (!path)
1797 panic("%s: Failed to allocate path for %s\n", __func__, param);
1798 strreplace(path, '.', '/');
1799
1800 file = file_open_root((*proc_mnt)->mnt_root, *proc_mnt, path, O_WRONLY, 0);
1801 if (IS_ERR(file)) {
1802 err = PTR_ERR(file);
1803 if (err == -ENOENT)
1804 pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1805 param, val);
1806 else if (err == -EACCES)
1807 pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1808 param, val);
1809 else
1810 pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1811 file, param, val);
1812 goto out;
1813 }
1814 len = strlen(val);
1815 wret = kernel_write(file, val, len, &pos);
1816 if (wret < 0) {
1817 err = wret;
1818 if (err == -EINVAL)
1819 pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1820 param, val);
1821 else
1822 pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1823 ERR_PTR(err), param, val);
1824 } else if (wret != len) {
1825 pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1826 wret, len, path, param, val);
1827 }
1828
1829 err = filp_close(file, NULL);
1830 if (err)
1831 pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1832 ERR_PTR(err), param, val);
1833out:
1834 kfree(path);
1835 return 0;
1836}
1837
1838void do_sysctl_args(void)
1839{
1840 char *command_line;
1841 struct vfsmount *proc_mnt = NULL;
1842
1843 command_line = kstrdup(saved_command_line, GFP_KERNEL);
1844 if (!command_line)
1845 panic("%s: Failed to allocate copy of command line\n", __func__);
1846
1847 parse_args("Setting sysctl args", command_line,
1848 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1849
1850 if (proc_mnt)
1851 kern_unmount(proc_mnt);
1852
1853 kfree(command_line);
1854}