4 #include <linux/init.h>
5 #include <linux/sysctl.h>
6 #include <linux/poll.h>
7 #include <linux/proc_fs.h>
8 #include <linux/security.h>
9 #include <linux/sched.h>
10 #include <linux/namei.h>
12 #include <linux/module.h>
15 static const struct dentry_operations proc_sys_dentry_operations;
16 static const struct file_operations proc_sys_file_operations;
17 static const struct inode_operations proc_sys_inode_operations;
18 static const struct file_operations proc_sys_dir_file_operations;
19 static const struct inode_operations proc_sys_dir_operations;
21 void proc_sys_poll_notify(struct ctl_table_poll *poll)
26 atomic_inc(&poll->event);
27 wake_up_interruptible(&poll->wait);
30 static struct ctl_table root_table[] = {
33 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
37 static struct ctl_table_root sysctl_table_root = {
38 .default_set.dir.header = {
41 .ctl_table = root_table }},
42 .ctl_table_arg = root_table,
43 .root = &sysctl_table_root,
44 .set = &sysctl_table_root.default_set,
48 static DEFINE_SPINLOCK(sysctl_lock);
50 static void drop_sysctl_table(struct ctl_table_header *header);
51 static int sysctl_follow_link(struct ctl_table_header **phead,
52 struct ctl_table **pentry, struct nsproxy *namespaces);
53 static int insert_links(struct ctl_table_header *head);
54 static void put_links(struct ctl_table_header *header);
56 static void sysctl_print_dir(struct ctl_dir *dir)
58 if (dir->header.parent)
59 sysctl_print_dir(dir->header.parent);
60 printk(KERN_CONT "%s/", dir->header.ctl_table[0].procname);
63 static int namecmp(const char *name1, int len1, const char *name2, int len2)
72 cmp = memcmp(name1, name2, minlen);
78 /* Called under sysctl_lock */
79 static struct ctl_table *find_entry(struct ctl_table_header **phead,
80 struct ctl_dir *dir, const char *name, int namelen)
82 struct ctl_table_header *head;
83 struct ctl_table *entry;
84 struct rb_node *node = dir->root.rb_node;
88 struct ctl_node *ctl_node;
92 ctl_node = rb_entry(node, struct ctl_node, node);
93 head = ctl_node->header;
94 entry = &head->ctl_table[ctl_node - head->node];
95 procname = entry->procname;
97 cmp = namecmp(name, namelen, procname, strlen(procname));
101 node = node->rb_right;
110 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
112 struct rb_node *node = &head->node[entry - head->ctl_table].node;
113 struct rb_node **p = &head->parent->root.rb_node;
114 struct rb_node *parent = NULL;
115 const char *name = entry->procname;
116 int namelen = strlen(name);
119 struct ctl_table_header *parent_head;
120 struct ctl_table *parent_entry;
121 struct ctl_node *parent_node;
122 const char *parent_name;
126 parent_node = rb_entry(parent, struct ctl_node, node);
127 parent_head = parent_node->header;
128 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
129 parent_name = parent_entry->procname;
131 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
137 printk(KERN_ERR "sysctl duplicate entry: ");
138 sysctl_print_dir(head->parent);
139 printk(KERN_CONT "/%s\n", entry->procname);
144 rb_link_node(node, parent, p);
145 rb_insert_color(node, &head->parent->root);
149 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
151 struct rb_node *node = &head->node[entry - head->ctl_table].node;
153 rb_erase(node, &head->parent->root);
156 static void init_header(struct ctl_table_header *head,
157 struct ctl_table_root *root, struct ctl_table_set *set,
158 struct ctl_node *node, struct ctl_table *table)
160 head->ctl_table = table;
161 head->ctl_table_arg = table;
165 head->unregistering = NULL;
171 struct ctl_table *entry;
172 for (entry = table; entry->procname; entry++, node++)
177 static void erase_header(struct ctl_table_header *head)
179 struct ctl_table *entry;
180 for (entry = head->ctl_table; entry->procname; entry++)
181 erase_entry(head, entry);
184 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
186 struct ctl_table *entry;
190 header->parent = dir;
191 err = insert_links(header);
194 for (entry = header->ctl_table; entry->procname; entry++) {
195 err = insert_entry(header, entry);
201 erase_header(header);
204 header->parent = NULL;
205 drop_sysctl_table(&dir->header);
209 /* called under sysctl_lock */
210 static int use_table(struct ctl_table_header *p)
212 if (unlikely(p->unregistering))
218 /* called under sysctl_lock */
219 static void unuse_table(struct ctl_table_header *p)
222 if (unlikely(p->unregistering))
223 complete(p->unregistering);
226 /* called under sysctl_lock, will reacquire if has to wait */
227 static void start_unregistering(struct ctl_table_header *p)
230 * if p->used is 0, nobody will ever touch that entry again;
231 * we'll eliminate all paths to it before dropping sysctl_lock
233 if (unlikely(p->used)) {
234 struct completion wait;
235 init_completion(&wait);
236 p->unregistering = &wait;
237 spin_unlock(&sysctl_lock);
238 wait_for_completion(&wait);
239 spin_lock(&sysctl_lock);
241 /* anything non-NULL; we'll never dereference it */
242 p->unregistering = ERR_PTR(-EINVAL);
245 * do not remove from the list until nobody holds it; walking the
246 * list in do_sysctl() relies on that.
251 static void sysctl_head_get(struct ctl_table_header *head)
253 spin_lock(&sysctl_lock);
255 spin_unlock(&sysctl_lock);
258 void sysctl_head_put(struct ctl_table_header *head)
260 spin_lock(&sysctl_lock);
262 kfree_rcu(head, rcu);
263 spin_unlock(&sysctl_lock);
266 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
269 spin_lock(&sysctl_lock);
270 if (!use_table(head))
271 head = ERR_PTR(-ENOENT);
272 spin_unlock(&sysctl_lock);
276 static void sysctl_head_finish(struct ctl_table_header *head)
280 spin_lock(&sysctl_lock);
282 spin_unlock(&sysctl_lock);
285 static struct ctl_table_set *
286 lookup_header_set(struct ctl_table_root *root, struct nsproxy *namespaces)
288 struct ctl_table_set *set = &root->default_set;
290 set = root->lookup(root, namespaces);
294 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
296 const char *name, int namelen)
298 struct ctl_table_header *head;
299 struct ctl_table *entry;
301 spin_lock(&sysctl_lock);
302 entry = find_entry(&head, dir, name, namelen);
303 if (entry && use_table(head))
307 spin_unlock(&sysctl_lock);
311 static struct ctl_node *first_usable_entry(struct rb_node *node)
313 struct ctl_node *ctl_node;
315 for (;node; node = rb_next(node)) {
316 ctl_node = rb_entry(node, struct ctl_node, node);
317 if (use_table(ctl_node->header))
323 static void first_entry(struct ctl_dir *dir,
324 struct ctl_table_header **phead, struct ctl_table **pentry)
326 struct ctl_table_header *head = NULL;
327 struct ctl_table *entry = NULL;
328 struct ctl_node *ctl_node;
330 spin_lock(&sysctl_lock);
331 ctl_node = first_usable_entry(rb_first(&dir->root));
332 spin_unlock(&sysctl_lock);
334 head = ctl_node->header;
335 entry = &head->ctl_table[ctl_node - head->node];
341 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
343 struct ctl_table_header *head = *phead;
344 struct ctl_table *entry = *pentry;
345 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
347 spin_lock(&sysctl_lock);
350 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
351 spin_unlock(&sysctl_lock);
354 head = ctl_node->header;
355 entry = &head->ctl_table[ctl_node - head->node];
361 void register_sysctl_root(struct ctl_table_root *root)
366 * sysctl_perm does NOT grant the superuser all rights automatically, because
367 * some sysctl variables are readonly even to root.
370 static int test_perm(int mode, int op)
372 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
374 else if (in_egroup_p(GLOBAL_ROOT_GID))
376 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
381 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
383 struct ctl_table_root *root = head->root;
386 if (root->permissions)
387 mode = root->permissions(head, table);
391 return test_perm(mode, op);
394 static struct inode *proc_sys_make_inode(struct super_block *sb,
395 struct ctl_table_header *head, struct ctl_table *table)
398 struct proc_inode *ei;
400 inode = new_inode(sb);
404 inode->i_ino = get_next_ino();
406 sysctl_head_get(head);
409 ei->sysctl_entry = table;
411 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
412 inode->i_mode = table->mode;
413 if (!S_ISDIR(table->mode)) {
414 inode->i_mode |= S_IFREG;
415 inode->i_op = &proc_sys_inode_operations;
416 inode->i_fop = &proc_sys_file_operations;
418 inode->i_mode |= S_IFDIR;
419 inode->i_op = &proc_sys_dir_operations;
420 inode->i_fop = &proc_sys_dir_file_operations;
426 static struct ctl_table_header *grab_header(struct inode *inode)
428 struct ctl_table_header *head = PROC_I(inode)->sysctl;
430 head = &sysctl_table_root.default_set.dir.header;
431 return sysctl_head_grab(head);
434 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
437 struct ctl_table_header *head = grab_header(dir);
438 struct ctl_table_header *h = NULL;
439 struct qstr *name = &dentry->d_name;
442 struct dentry *err = ERR_PTR(-ENOENT);
443 struct ctl_dir *ctl_dir;
447 return ERR_CAST(head);
449 ctl_dir = container_of(head, struct ctl_dir, header);
451 p = lookup_entry(&h, ctl_dir, name->name, name->len);
455 if (S_ISLNK(p->mode)) {
456 ret = sysctl_follow_link(&h, &p, current->nsproxy);
462 err = ERR_PTR(-ENOMEM);
463 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
468 d_set_d_op(dentry, &proc_sys_dentry_operations);
469 d_add(dentry, inode);
473 sysctl_head_finish(h);
474 sysctl_head_finish(head);
478 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
479 size_t count, loff_t *ppos, int write)
481 struct inode *inode = filp->f_path.dentry->d_inode;
482 struct ctl_table_header *head = grab_header(inode);
483 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
488 return PTR_ERR(head);
491 * At this point we know that the sysctl was not unregistered
492 * and won't be until we finish.
495 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
498 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
500 if (!table->proc_handler)
503 /* careful: calling conventions are nasty here */
505 error = table->proc_handler(table, write, buf, &res, ppos);
509 sysctl_head_finish(head);
514 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
515 size_t count, loff_t *ppos)
517 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
520 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
521 size_t count, loff_t *ppos)
523 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
526 static int proc_sys_open(struct inode *inode, struct file *filp)
528 struct ctl_table_header *head = grab_header(inode);
529 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
531 /* sysctl was unregistered */
533 return PTR_ERR(head);
536 filp->private_data = proc_sys_poll_event(table->poll);
538 sysctl_head_finish(head);
543 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
545 struct inode *inode = filp->f_path.dentry->d_inode;
546 struct ctl_table_header *head = grab_header(inode);
547 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
548 unsigned int ret = DEFAULT_POLLMASK;
551 /* sysctl was unregistered */
553 return POLLERR | POLLHUP;
555 if (!table->proc_handler)
561 event = (unsigned long)filp->private_data;
562 poll_wait(filp, &table->poll->wait, wait);
564 if (event != atomic_read(&table->poll->event)) {
565 filp->private_data = proc_sys_poll_event(table->poll);
566 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
570 sysctl_head_finish(head);
575 static int proc_sys_fill_cache(struct file *filp, void *dirent,
577 struct ctl_table_header *head,
578 struct ctl_table *table)
580 struct dentry *child, *dir = filp->f_path.dentry;
584 unsigned type = DT_UNKNOWN;
586 qname.name = table->procname;
587 qname.len = strlen(table->procname);
588 qname.hash = full_name_hash(qname.name, qname.len);
590 child = d_lookup(dir, &qname);
592 child = d_alloc(dir, &qname);
594 inode = proc_sys_make_inode(dir->d_sb, head, table);
599 d_set_d_op(child, &proc_sys_dentry_operations);
606 inode = child->d_inode;
608 type = inode->i_mode >> 12;
610 return !!filldir(dirent, qname.name, qname.len, filp->f_pos, ino, type);
613 static int proc_sys_link_fill_cache(struct file *filp, void *dirent,
615 struct ctl_table_header *head,
616 struct ctl_table *table)
619 head = sysctl_head_grab(head);
621 if (S_ISLNK(table->mode)) {
622 /* It is not an error if we can not follow the link ignore it */
623 err = sysctl_follow_link(&head, &table, current->nsproxy);
628 ret = proc_sys_fill_cache(filp, dirent, filldir, head, table);
630 sysctl_head_finish(head);
634 static int scan(struct ctl_table_header *head, ctl_table *table,
635 unsigned long *pos, struct file *file,
636 void *dirent, filldir_t filldir)
640 if ((*pos)++ < file->f_pos)
643 if (unlikely(S_ISLNK(table->mode)))
644 res = proc_sys_link_fill_cache(file, dirent, filldir, head, table);
646 res = proc_sys_fill_cache(file, dirent, filldir, head, table);
654 static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir)
656 struct dentry *dentry = filp->f_path.dentry;
657 struct inode *inode = dentry->d_inode;
658 struct ctl_table_header *head = grab_header(inode);
659 struct ctl_table_header *h = NULL;
660 struct ctl_table *entry;
661 struct ctl_dir *ctl_dir;
666 return PTR_ERR(head);
668 ctl_dir = container_of(head, struct ctl_dir, header);
671 /* Avoid a switch here: arm builds fail with missing __cmpdi2 */
672 if (filp->f_pos == 0) {
673 if (filldir(dirent, ".", 1, filp->f_pos,
674 inode->i_ino, DT_DIR) < 0)
678 if (filp->f_pos == 1) {
679 if (filldir(dirent, "..", 2, filp->f_pos,
680 parent_ino(dentry), DT_DIR) < 0)
686 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
687 ret = scan(h, entry, &pos, filp, dirent, filldir);
689 sysctl_head_finish(h);
695 sysctl_head_finish(head);
699 static int proc_sys_permission(struct inode *inode, int mask)
702 * sysctl entries that are not writeable,
703 * are _NOT_ writeable, capabilities or not.
705 struct ctl_table_header *head;
706 struct ctl_table *table;
709 /* Executable files are not allowed under /proc/sys/ */
710 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
713 head = grab_header(inode);
715 return PTR_ERR(head);
717 table = PROC_I(inode)->sysctl_entry;
718 if (!table) /* global root - r-xr-xr-x */
719 error = mask & MAY_WRITE ? -EACCES : 0;
720 else /* Use the permissions on the sysctl table entry */
721 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
723 sysctl_head_finish(head);
727 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
729 struct inode *inode = dentry->d_inode;
732 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
735 error = inode_change_ok(inode, attr);
739 setattr_copy(inode, attr);
740 mark_inode_dirty(inode);
744 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
746 struct inode *inode = dentry->d_inode;
747 struct ctl_table_header *head = grab_header(inode);
748 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
751 return PTR_ERR(head);
753 generic_fillattr(inode, stat);
755 stat->mode = (stat->mode & S_IFMT) | table->mode;
757 sysctl_head_finish(head);
761 static const struct file_operations proc_sys_file_operations = {
762 .open = proc_sys_open,
763 .poll = proc_sys_poll,
764 .read = proc_sys_read,
765 .write = proc_sys_write,
766 .llseek = default_llseek,
769 static const struct file_operations proc_sys_dir_file_operations = {
770 .read = generic_read_dir,
771 .readdir = proc_sys_readdir,
772 .llseek = generic_file_llseek,
775 static const struct inode_operations proc_sys_inode_operations = {
776 .permission = proc_sys_permission,
777 .setattr = proc_sys_setattr,
778 .getattr = proc_sys_getattr,
781 static const struct inode_operations proc_sys_dir_operations = {
782 .lookup = proc_sys_lookup,
783 .permission = proc_sys_permission,
784 .setattr = proc_sys_setattr,
785 .getattr = proc_sys_getattr,
788 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
790 if (flags & LOOKUP_RCU)
792 return !PROC_I(dentry->d_inode)->sysctl->unregistering;
795 static int proc_sys_delete(const struct dentry *dentry)
797 return !!PROC_I(dentry->d_inode)->sysctl->unregistering;
800 static int sysctl_is_seen(struct ctl_table_header *p)
802 struct ctl_table_set *set = p->set;
804 spin_lock(&sysctl_lock);
805 if (p->unregistering)
807 else if (!set->is_seen)
810 res = set->is_seen(set);
811 spin_unlock(&sysctl_lock);
815 static int proc_sys_compare(const struct dentry *parent,
816 const struct inode *pinode,
817 const struct dentry *dentry, const struct inode *inode,
818 unsigned int len, const char *str, const struct qstr *name)
820 struct ctl_table_header *head;
821 /* Although proc doesn't have negative dentries, rcu-walk means
822 * that inode here can be NULL */
823 /* AV: can it, indeed? */
826 if (name->len != len)
828 if (memcmp(name->name, str, len))
830 head = rcu_dereference(PROC_I(inode)->sysctl);
831 return !head || !sysctl_is_seen(head);
834 static const struct dentry_operations proc_sys_dentry_operations = {
835 .d_revalidate = proc_sys_revalidate,
836 .d_delete = proc_sys_delete,
837 .d_compare = proc_sys_compare,
840 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
841 const char *name, int namelen)
843 struct ctl_table_header *head;
844 struct ctl_table *entry;
846 entry = find_entry(&head, dir, name, namelen);
848 return ERR_PTR(-ENOENT);
849 if (!S_ISDIR(entry->mode))
850 return ERR_PTR(-ENOTDIR);
851 return container_of(head, struct ctl_dir, header);
854 static struct ctl_dir *new_dir(struct ctl_table_set *set,
855 const char *name, int namelen)
857 struct ctl_table *table;
859 struct ctl_node *node;
862 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
863 sizeof(struct ctl_table)*2 + namelen + 1,
868 node = (struct ctl_node *)(new + 1);
869 table = (struct ctl_table *)(node + 1);
870 new_name = (char *)(table + 2);
871 memcpy(new_name, name, namelen);
872 new_name[namelen] = '\0';
873 table[0].procname = new_name;
874 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
875 init_header(&new->header, set->dir.header.root, set, node, table);
881 * get_subdir - find or create a subdir with the specified name.
882 * @dir: Directory to create the subdirectory in
883 * @name: The name of the subdirectory to find or create
884 * @namelen: The length of name
886 * Takes a directory with an elevated reference count so we know that
887 * if we drop the lock the directory will not go away. Upon success
888 * the reference is moved from @dir to the returned subdirectory.
889 * Upon error an error code is returned and the reference on @dir is
892 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
893 const char *name, int namelen)
895 struct ctl_table_set *set = dir->header.set;
896 struct ctl_dir *subdir, *new = NULL;
899 spin_lock(&sysctl_lock);
900 subdir = find_subdir(dir, name, namelen);
903 if (PTR_ERR(subdir) != -ENOENT)
906 spin_unlock(&sysctl_lock);
907 new = new_dir(set, name, namelen);
908 spin_lock(&sysctl_lock);
909 subdir = ERR_PTR(-ENOMEM);
913 /* Was the subdir added while we dropped the lock? */
914 subdir = find_subdir(dir, name, namelen);
917 if (PTR_ERR(subdir) != -ENOENT)
920 /* Nope. Use the our freshly made directory entry. */
921 err = insert_header(dir, &new->header);
922 subdir = ERR_PTR(err);
927 subdir->header.nreg++;
929 if (unlikely(IS_ERR(subdir))) {
930 printk(KERN_ERR "sysctl could not get directory: ");
931 sysctl_print_dir(dir);
932 printk(KERN_CONT "/%*.*s %ld\n",
933 namelen, namelen, name, PTR_ERR(subdir));
935 drop_sysctl_table(&dir->header);
937 drop_sysctl_table(&new->header);
938 spin_unlock(&sysctl_lock);
942 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
944 struct ctl_dir *parent;
945 const char *procname;
946 if (!dir->header.parent)
948 parent = xlate_dir(set, dir->header.parent);
951 procname = dir->header.ctl_table[0].procname;
952 return find_subdir(parent, procname, strlen(procname));
955 static int sysctl_follow_link(struct ctl_table_header **phead,
956 struct ctl_table **pentry, struct nsproxy *namespaces)
958 struct ctl_table_header *head;
959 struct ctl_table_root *root;
960 struct ctl_table_set *set;
961 struct ctl_table *entry;
966 spin_lock(&sysctl_lock);
967 root = (*pentry)->data;
968 set = lookup_header_set(root, namespaces);
969 dir = xlate_dir(set, (*phead)->parent);
973 const char *procname = (*pentry)->procname;
975 entry = find_entry(&head, dir, procname, strlen(procname));
977 if (entry && use_table(head)) {
985 spin_unlock(&sysctl_lock);
989 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
991 struct va_format vaf;
998 printk(KERN_ERR "sysctl table check failed: %s/%s %pV\n",
999 path, table->procname, &vaf);
1005 static int sysctl_check_table(const char *path, struct ctl_table *table)
1008 for (; table->procname; table++) {
1010 err = sysctl_err(path, table, "Not a file");
1012 if ((table->proc_handler == proc_dostring) ||
1013 (table->proc_handler == proc_dointvec) ||
1014 (table->proc_handler == proc_dointvec_minmax) ||
1015 (table->proc_handler == proc_dointvec_jiffies) ||
1016 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1017 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1018 (table->proc_handler == proc_doulongvec_minmax) ||
1019 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1021 err = sysctl_err(path, table, "No data");
1023 err = sysctl_err(path, table, "No maxlen");
1025 if (!table->proc_handler)
1026 err = sysctl_err(path, table, "No proc_handler");
1028 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1029 err = sysctl_err(path, table, "bogus .mode 0%o",
1035 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1036 struct ctl_table_root *link_root)
1038 struct ctl_table *link_table, *entry, *link;
1039 struct ctl_table_header *links;
1040 struct ctl_node *node;
1042 int nr_entries, name_bytes;
1046 for (entry = table; entry->procname; entry++) {
1048 name_bytes += strlen(entry->procname) + 1;
1051 links = kzalloc(sizeof(struct ctl_table_header) +
1052 sizeof(struct ctl_node)*nr_entries +
1053 sizeof(struct ctl_table)*(nr_entries + 1) +
1060 node = (struct ctl_node *)(links + 1);
1061 link_table = (struct ctl_table *)(node + nr_entries);
1062 link_name = (char *)&link_table[nr_entries + 1];
1064 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1065 int len = strlen(entry->procname) + 1;
1066 memcpy(link_name, entry->procname, len);
1067 link->procname = link_name;
1068 link->mode = S_IFLNK|S_IRWXUGO;
1069 link->data = link_root;
1072 init_header(links, dir->header.root, dir->header.set, node, link_table);
1073 links->nreg = nr_entries;
1078 static bool get_links(struct ctl_dir *dir,
1079 struct ctl_table *table, struct ctl_table_root *link_root)
1081 struct ctl_table_header *head;
1082 struct ctl_table *entry, *link;
1084 /* Are there links available for every entry in table? */
1085 for (entry = table; entry->procname; entry++) {
1086 const char *procname = entry->procname;
1087 link = find_entry(&head, dir, procname, strlen(procname));
1090 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1092 if (S_ISLNK(link->mode) && (link->data == link_root))
1097 /* The checks passed. Increase the registration count on the links */
1098 for (entry = table; entry->procname; entry++) {
1099 const char *procname = entry->procname;
1100 link = find_entry(&head, dir, procname, strlen(procname));
1106 static int insert_links(struct ctl_table_header *head)
1108 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1109 struct ctl_dir *core_parent = NULL;
1110 struct ctl_table_header *links;
1113 if (head->set == root_set)
1116 core_parent = xlate_dir(root_set, head->parent);
1117 if (IS_ERR(core_parent))
1120 if (get_links(core_parent, head->ctl_table, head->root))
1123 core_parent->header.nreg++;
1124 spin_unlock(&sysctl_lock);
1126 links = new_links(core_parent, head->ctl_table, head->root);
1128 spin_lock(&sysctl_lock);
1134 if (get_links(core_parent, head->ctl_table, head->root)) {
1139 err = insert_header(core_parent, links);
1143 drop_sysctl_table(&core_parent->header);
1148 * __register_sysctl_table - register a leaf sysctl table
1149 * @set: Sysctl tree to register on
1150 * @path: The path to the directory the sysctl table is in.
1151 * @table: the top-level table structure
1153 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1154 * array. A completely 0 filled entry terminates the table.
1156 * The members of the &struct ctl_table structure are used as follows:
1158 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1159 * enter a sysctl file
1161 * data - a pointer to data for use by proc_handler
1163 * maxlen - the maximum size in bytes of the data
1165 * mode - the file permissions for the /proc/sys file
1167 * child - must be %NULL.
1169 * proc_handler - the text handler routine (described below)
1171 * extra1, extra2 - extra pointers usable by the proc handler routines
1173 * Leaf nodes in the sysctl tree will be represented by a single file
1174 * under /proc; non-leaf nodes will be represented by directories.
1176 * There must be a proc_handler routine for any terminal nodes.
1177 * Several default handlers are available to cover common cases -
1179 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1180 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1181 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1183 * It is the handler's job to read the input buffer from user memory
1184 * and process it. The handler should return 0 on success.
1186 * This routine returns %NULL on a failure to register, and a pointer
1187 * to the table header on success.
1189 struct ctl_table_header *__register_sysctl_table(
1190 struct ctl_table_set *set,
1191 const char *path, struct ctl_table *table)
1193 struct ctl_table_root *root = set->dir.header.root;
1194 struct ctl_table_header *header;
1195 const char *name, *nextname;
1196 struct ctl_dir *dir;
1197 struct ctl_table *entry;
1198 struct ctl_node *node;
1201 for (entry = table; entry->procname; entry++)
1204 header = kzalloc(sizeof(struct ctl_table_header) +
1205 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1209 node = (struct ctl_node *)(header + 1);
1210 init_header(header, root, set, node, table);
1211 if (sysctl_check_table(path, table))
1214 spin_lock(&sysctl_lock);
1216 /* Reference moved down the diretory tree get_subdir */
1218 spin_unlock(&sysctl_lock);
1220 /* Find the directory for the ctl_table */
1221 for (name = path; name; name = nextname) {
1223 nextname = strchr(name, '/');
1225 namelen = nextname - name;
1228 namelen = strlen(name);
1233 dir = get_subdir(dir, name, namelen);
1238 spin_lock(&sysctl_lock);
1239 if (insert_header(dir, header))
1240 goto fail_put_dir_locked;
1242 drop_sysctl_table(&dir->header);
1243 spin_unlock(&sysctl_lock);
1247 fail_put_dir_locked:
1248 drop_sysctl_table(&dir->header);
1249 spin_unlock(&sysctl_lock);
1257 * register_sysctl - register a sysctl table
1258 * @path: The path to the directory the sysctl table is in.
1259 * @table: the table structure
1261 * Register a sysctl table. @table should be a filled in ctl_table
1262 * array. A completely 0 filled entry terminates the table.
1264 * See __register_sysctl_table for more details.
1266 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1268 return __register_sysctl_table(&sysctl_table_root.default_set,
1271 EXPORT_SYMBOL(register_sysctl);
1273 static char *append_path(const char *path, char *pos, const char *name)
1276 namelen = strlen(name);
1277 if (((pos - path) + namelen + 2) >= PATH_MAX)
1279 memcpy(pos, name, namelen);
1281 pos[namelen + 1] = '\0';
1286 static int count_subheaders(struct ctl_table *table)
1289 int nr_subheaders = 0;
1290 struct ctl_table *entry;
1292 /* special case: no directory and empty directory */
1293 if (!table || !table->procname)
1296 for (entry = table; entry->procname; entry++) {
1298 nr_subheaders += count_subheaders(entry->child);
1302 return nr_subheaders + has_files;
1305 static int register_leaf_sysctl_tables(const char *path, char *pos,
1306 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1307 struct ctl_table *table)
1309 struct ctl_table *ctl_table_arg = NULL;
1310 struct ctl_table *entry, *files;
1315 for (entry = table; entry->procname; entry++) {
1323 /* If there are mixed files and directories we need a new table */
1324 if (nr_dirs && nr_files) {
1325 struct ctl_table *new;
1326 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1331 ctl_table_arg = files;
1332 for (new = files, entry = table; entry->procname; entry++) {
1340 /* Register everything except a directory full of subdirectories */
1341 if (nr_files || !nr_dirs) {
1342 struct ctl_table_header *header;
1343 header = __register_sysctl_table(set, path, files);
1345 kfree(ctl_table_arg);
1349 /* Remember if we need to free the file table */
1350 header->ctl_table_arg = ctl_table_arg;
1351 **subheader = header;
1355 /* Recurse into the subdirectories. */
1356 for (entry = table; entry->procname; entry++) {
1362 err = -ENAMETOOLONG;
1363 child_pos = append_path(path, pos, entry->procname);
1367 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1375 /* On failure our caller will unregister all registered subheaders */
1380 * __register_sysctl_paths - register a sysctl table hierarchy
1381 * @set: Sysctl tree to register on
1382 * @path: The path to the directory the sysctl table is in.
1383 * @table: the top-level table structure
1385 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1386 * array. A completely 0 filled entry terminates the table.
1388 * See __register_sysctl_table for more details.
1390 struct ctl_table_header *__register_sysctl_paths(
1391 struct ctl_table_set *set,
1392 const struct ctl_path *path, struct ctl_table *table)
1394 struct ctl_table *ctl_table_arg = table;
1395 int nr_subheaders = count_subheaders(table);
1396 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1397 const struct ctl_path *component;
1398 char *new_path, *pos;
1400 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1405 for (component = path; component->procname; component++) {
1406 pos = append_path(new_path, pos, component->procname);
1410 while (table->procname && table->child && !table[1].procname) {
1411 pos = append_path(new_path, pos, table->procname);
1414 table = table->child;
1416 if (nr_subheaders == 1) {
1417 header = __register_sysctl_table(set, new_path, table);
1419 header->ctl_table_arg = ctl_table_arg;
1421 header = kzalloc(sizeof(*header) +
1422 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1426 subheaders = (struct ctl_table_header **) (header + 1);
1427 subheader = subheaders;
1428 header->ctl_table_arg = ctl_table_arg;
1430 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1432 goto err_register_leaves;
1439 err_register_leaves:
1440 while (subheader > subheaders) {
1441 struct ctl_table_header *subh = *(--subheader);
1442 struct ctl_table *table = subh->ctl_table_arg;
1443 unregister_sysctl_table(subh);
1452 * register_sysctl_table_path - register a sysctl table hierarchy
1453 * @path: The path to the directory the sysctl table is in.
1454 * @table: the top-level table structure
1456 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1457 * array. A completely 0 filled entry terminates the table.
1459 * See __register_sysctl_paths for more details.
1461 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1462 struct ctl_table *table)
1464 return __register_sysctl_paths(&sysctl_table_root.default_set,
1467 EXPORT_SYMBOL(register_sysctl_paths);
1470 * register_sysctl_table - register a sysctl table hierarchy
1471 * @table: the top-level table structure
1473 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1474 * array. A completely 0 filled entry terminates the table.
1476 * See register_sysctl_paths for more details.
1478 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1480 static const struct ctl_path null_path[] = { {} };
1482 return register_sysctl_paths(null_path, table);
1484 EXPORT_SYMBOL(register_sysctl_table);
1486 static void put_links(struct ctl_table_header *header)
1488 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1489 struct ctl_table_root *root = header->root;
1490 struct ctl_dir *parent = header->parent;
1491 struct ctl_dir *core_parent;
1492 struct ctl_table *entry;
1494 if (header->set == root_set)
1497 core_parent = xlate_dir(root_set, parent);
1498 if (IS_ERR(core_parent))
1501 for (entry = header->ctl_table; entry->procname; entry++) {
1502 struct ctl_table_header *link_head;
1503 struct ctl_table *link;
1504 const char *name = entry->procname;
1506 link = find_entry(&link_head, core_parent, name, strlen(name));
1508 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1509 (S_ISLNK(link->mode) && (link->data == root)))) {
1510 drop_sysctl_table(link_head);
1513 printk(KERN_ERR "sysctl link missing during unregister: ");
1514 sysctl_print_dir(parent);
1515 printk(KERN_CONT "/%s\n", name);
1520 static void drop_sysctl_table(struct ctl_table_header *header)
1522 struct ctl_dir *parent = header->parent;
1528 start_unregistering(header);
1529 if (!--header->count)
1530 kfree_rcu(header, rcu);
1533 drop_sysctl_table(&parent->header);
1537 * unregister_sysctl_table - unregister a sysctl table hierarchy
1538 * @header: the header returned from register_sysctl_table
1540 * Unregisters the sysctl table and all children. proc entries may not
1541 * actually be removed until they are no longer used by anyone.
1543 void unregister_sysctl_table(struct ctl_table_header * header)
1551 nr_subheaders = count_subheaders(header->ctl_table_arg);
1552 if (unlikely(nr_subheaders > 1)) {
1553 struct ctl_table_header **subheaders;
1556 subheaders = (struct ctl_table_header **)(header + 1);
1557 for (i = nr_subheaders -1; i >= 0; i--) {
1558 struct ctl_table_header *subh = subheaders[i];
1559 struct ctl_table *table = subh->ctl_table_arg;
1560 unregister_sysctl_table(subh);
1567 spin_lock(&sysctl_lock);
1568 drop_sysctl_table(header);
1569 spin_unlock(&sysctl_lock);
1571 EXPORT_SYMBOL(unregister_sysctl_table);
1573 void setup_sysctl_set(struct ctl_table_set *set,
1574 struct ctl_table_root *root,
1575 int (*is_seen)(struct ctl_table_set *))
1577 memset(set, 0, sizeof(*set));
1578 set->is_seen = is_seen;
1579 init_header(&set->dir.header, root, set, NULL, root_table);
1582 void retire_sysctl_set(struct ctl_table_set *set)
1584 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1587 int __init proc_sys_init(void)
1589 struct proc_dir_entry *proc_sys_root;
1591 proc_sys_root = proc_mkdir("sys", NULL);
1592 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1593 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1594 proc_sys_root->nlink = 0;
1596 return sysctl_init();