4 #include <linux/init.h>
5 #include <linux/sysctl.h>
6 #include <linux/poll.h>
7 #include <linux/proc_fs.h>
8 #include <linux/printk.h>
9 #include <linux/security.h>
10 #include <linux/sched.h>
11 #include <linux/namei.h>
13 #include <linux/module.h>
16 static const struct dentry_operations proc_sys_dentry_operations;
17 static const struct file_operations proc_sys_file_operations;
18 static const struct inode_operations proc_sys_inode_operations;
19 static const struct file_operations proc_sys_dir_file_operations;
20 static const struct inode_operations proc_sys_dir_operations;
22 /* Support for permanently empty directories */
24 struct ctl_table sysctl_mount_point[] = {
28 static bool is_empty_dir(struct ctl_table_header *head)
30 return head->ctl_table[0].child == sysctl_mount_point;
33 static void set_empty_dir(struct ctl_dir *dir)
35 dir->header.ctl_table[0].child = sysctl_mount_point;
38 static void clear_empty_dir(struct ctl_dir *dir)
41 dir->header.ctl_table[0].child = NULL;
44 void proc_sys_poll_notify(struct ctl_table_poll *poll)
49 atomic_inc(&poll->event);
50 wake_up_interruptible(&poll->wait);
53 static struct ctl_table root_table[] = {
56 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
60 static struct ctl_table_root sysctl_table_root = {
61 .default_set.dir.header = {
64 .ctl_table = root_table }},
65 .ctl_table_arg = root_table,
66 .root = &sysctl_table_root,
67 .set = &sysctl_table_root.default_set,
71 static DEFINE_SPINLOCK(sysctl_lock);
73 static void drop_sysctl_table(struct ctl_table_header *header);
74 static int sysctl_follow_link(struct ctl_table_header **phead,
75 struct ctl_table **pentry);
76 static int insert_links(struct ctl_table_header *head);
77 static void put_links(struct ctl_table_header *header);
79 static void sysctl_print_dir(struct ctl_dir *dir)
81 if (dir->header.parent)
82 sysctl_print_dir(dir->header.parent);
83 pr_cont("%s/", dir->header.ctl_table[0].procname);
86 static int namecmp(const char *name1, int len1, const char *name2, int len2)
95 cmp = memcmp(name1, name2, minlen);
101 /* Called under sysctl_lock */
102 static struct ctl_table *find_entry(struct ctl_table_header **phead,
103 struct ctl_dir *dir, const char *name, int namelen)
105 struct ctl_table_header *head;
106 struct ctl_table *entry;
107 struct rb_node *node = dir->root.rb_node;
111 struct ctl_node *ctl_node;
112 const char *procname;
115 ctl_node = rb_entry(node, struct ctl_node, node);
116 head = ctl_node->header;
117 entry = &head->ctl_table[ctl_node - head->node];
118 procname = entry->procname;
120 cmp = namecmp(name, namelen, procname, strlen(procname));
122 node = node->rb_left;
124 node = node->rb_right;
133 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
135 struct rb_node *node = &head->node[entry - head->ctl_table].node;
136 struct rb_node **p = &head->parent->root.rb_node;
137 struct rb_node *parent = NULL;
138 const char *name = entry->procname;
139 int namelen = strlen(name);
142 struct ctl_table_header *parent_head;
143 struct ctl_table *parent_entry;
144 struct ctl_node *parent_node;
145 const char *parent_name;
149 parent_node = rb_entry(parent, struct ctl_node, node);
150 parent_head = parent_node->header;
151 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
152 parent_name = parent_entry->procname;
154 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
160 pr_err("sysctl duplicate entry: ");
161 sysctl_print_dir(head->parent);
162 pr_cont("/%s\n", entry->procname);
167 rb_link_node(node, parent, p);
168 rb_insert_color(node, &head->parent->root);
172 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
174 struct rb_node *node = &head->node[entry - head->ctl_table].node;
176 rb_erase(node, &head->parent->root);
179 static void init_header(struct ctl_table_header *head,
180 struct ctl_table_root *root, struct ctl_table_set *set,
181 struct ctl_node *node, struct ctl_table *table)
183 head->ctl_table = table;
184 head->ctl_table_arg = table;
188 head->unregistering = NULL;
194 struct ctl_table *entry;
195 for (entry = table; entry->procname; entry++, node++)
200 static void erase_header(struct ctl_table_header *head)
202 struct ctl_table *entry;
203 for (entry = head->ctl_table; entry->procname; entry++)
204 erase_entry(head, entry);
207 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
209 struct ctl_table *entry;
212 /* Is this a permanently empty directory? */
213 if (is_empty_dir(&dir->header))
216 /* Am I creating a permanently empty directory? */
217 if (header->ctl_table == sysctl_mount_point) {
218 if (!RB_EMPTY_ROOT(&dir->root))
224 header->parent = dir;
225 err = insert_links(header);
228 for (entry = header->ctl_table; entry->procname; entry++) {
229 err = insert_entry(header, entry);
235 erase_header(header);
238 if (header->ctl_table == sysctl_mount_point)
239 clear_empty_dir(dir);
240 header->parent = NULL;
241 drop_sysctl_table(&dir->header);
245 /* called under sysctl_lock */
246 static int use_table(struct ctl_table_header *p)
248 if (unlikely(p->unregistering))
254 /* called under sysctl_lock */
255 static void unuse_table(struct ctl_table_header *p)
258 if (unlikely(p->unregistering))
259 complete(p->unregistering);
262 /* called under sysctl_lock, will reacquire if has to wait */
263 static void start_unregistering(struct ctl_table_header *p)
266 * if p->used is 0, nobody will ever touch that entry again;
267 * we'll eliminate all paths to it before dropping sysctl_lock
269 if (unlikely(p->used)) {
270 struct completion wait;
271 init_completion(&wait);
272 p->unregistering = &wait;
273 spin_unlock(&sysctl_lock);
274 wait_for_completion(&wait);
275 spin_lock(&sysctl_lock);
277 /* anything non-NULL; we'll never dereference it */
278 p->unregistering = ERR_PTR(-EINVAL);
281 * do not remove from the list until nobody holds it; walking the
282 * list in do_sysctl() relies on that.
287 static void sysctl_head_get(struct ctl_table_header *head)
289 spin_lock(&sysctl_lock);
291 spin_unlock(&sysctl_lock);
294 void sysctl_head_put(struct ctl_table_header *head)
296 spin_lock(&sysctl_lock);
298 kfree_rcu(head, rcu);
299 spin_unlock(&sysctl_lock);
302 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
305 spin_lock(&sysctl_lock);
306 if (!use_table(head))
307 head = ERR_PTR(-ENOENT);
308 spin_unlock(&sysctl_lock);
312 static void sysctl_head_finish(struct ctl_table_header *head)
316 spin_lock(&sysctl_lock);
318 spin_unlock(&sysctl_lock);
321 static struct ctl_table_set *
322 lookup_header_set(struct ctl_table_root *root)
324 struct ctl_table_set *set = &root->default_set;
326 set = root->lookup(root);
330 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
332 const char *name, int namelen)
334 struct ctl_table_header *head;
335 struct ctl_table *entry;
337 spin_lock(&sysctl_lock);
338 entry = find_entry(&head, dir, name, namelen);
339 if (entry && use_table(head))
343 spin_unlock(&sysctl_lock);
347 static struct ctl_node *first_usable_entry(struct rb_node *node)
349 struct ctl_node *ctl_node;
351 for (;node; node = rb_next(node)) {
352 ctl_node = rb_entry(node, struct ctl_node, node);
353 if (use_table(ctl_node->header))
359 static void first_entry(struct ctl_dir *dir,
360 struct ctl_table_header **phead, struct ctl_table **pentry)
362 struct ctl_table_header *head = NULL;
363 struct ctl_table *entry = NULL;
364 struct ctl_node *ctl_node;
366 spin_lock(&sysctl_lock);
367 ctl_node = first_usable_entry(rb_first(&dir->root));
368 spin_unlock(&sysctl_lock);
370 head = ctl_node->header;
371 entry = &head->ctl_table[ctl_node - head->node];
377 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
379 struct ctl_table_header *head = *phead;
380 struct ctl_table *entry = *pentry;
381 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
383 spin_lock(&sysctl_lock);
386 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
387 spin_unlock(&sysctl_lock);
390 head = ctl_node->header;
391 entry = &head->ctl_table[ctl_node - head->node];
397 void register_sysctl_root(struct ctl_table_root *root)
402 * sysctl_perm does NOT grant the superuser all rights automatically, because
403 * some sysctl variables are readonly even to root.
406 static int test_perm(int mode, int op)
408 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
410 else if (in_egroup_p(GLOBAL_ROOT_GID))
412 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
417 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
419 struct ctl_table_root *root = head->root;
422 if (root->permissions)
423 mode = root->permissions(head, table);
427 return test_perm(mode, op);
430 static struct inode *proc_sys_make_inode(struct super_block *sb,
431 struct ctl_table_header *head, struct ctl_table *table)
433 struct ctl_table_root *root = head->root;
435 struct proc_inode *ei;
437 inode = new_inode(sb);
441 inode->i_ino = get_next_ino();
443 sysctl_head_get(head);
446 ei->sysctl_entry = table;
448 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
449 inode->i_mode = table->mode;
450 if (!S_ISDIR(table->mode)) {
451 inode->i_mode |= S_IFREG;
452 inode->i_op = &proc_sys_inode_operations;
453 inode->i_fop = &proc_sys_file_operations;
455 inode->i_mode |= S_IFDIR;
456 inode->i_op = &proc_sys_dir_operations;
457 inode->i_fop = &proc_sys_dir_file_operations;
458 if (is_empty_dir(head))
459 make_empty_dir_inode(inode);
462 if (root->set_ownership)
463 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
469 static struct ctl_table_header *grab_header(struct inode *inode)
471 struct ctl_table_header *head = PROC_I(inode)->sysctl;
473 head = &sysctl_table_root.default_set.dir.header;
474 return sysctl_head_grab(head);
477 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
480 struct ctl_table_header *head = grab_header(dir);
481 struct ctl_table_header *h = NULL;
482 const struct qstr *name = &dentry->d_name;
485 struct dentry *err = ERR_PTR(-ENOENT);
486 struct ctl_dir *ctl_dir;
490 return ERR_CAST(head);
492 ctl_dir = container_of(head, struct ctl_dir, header);
494 p = lookup_entry(&h, ctl_dir, name->name, name->len);
498 if (S_ISLNK(p->mode)) {
499 ret = sysctl_follow_link(&h, &p);
505 err = ERR_PTR(-ENOMEM);
506 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
511 d_set_d_op(dentry, &proc_sys_dentry_operations);
512 d_add(dentry, inode);
516 sysctl_head_finish(h);
517 sysctl_head_finish(head);
521 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
522 size_t count, loff_t *ppos, int write)
524 struct inode *inode = file_inode(filp);
525 struct ctl_table_header *head = grab_header(inode);
526 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
531 return PTR_ERR(head);
534 * At this point we know that the sysctl was not unregistered
535 * and won't be until we finish.
538 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
541 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
543 if (!table->proc_handler)
546 /* careful: calling conventions are nasty here */
548 error = table->proc_handler(table, write, buf, &res, ppos);
552 sysctl_head_finish(head);
557 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
558 size_t count, loff_t *ppos)
560 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
563 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
564 size_t count, loff_t *ppos)
566 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
569 static int proc_sys_open(struct inode *inode, struct file *filp)
571 struct ctl_table_header *head = grab_header(inode);
572 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
574 /* sysctl was unregistered */
576 return PTR_ERR(head);
579 filp->private_data = proc_sys_poll_event(table->poll);
581 sysctl_head_finish(head);
586 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
588 struct inode *inode = file_inode(filp);
589 struct ctl_table_header *head = grab_header(inode);
590 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
591 unsigned int ret = DEFAULT_POLLMASK;
594 /* sysctl was unregistered */
596 return POLLERR | POLLHUP;
598 if (!table->proc_handler)
604 event = (unsigned long)filp->private_data;
605 poll_wait(filp, &table->poll->wait, wait);
607 if (event != atomic_read(&table->poll->event)) {
608 filp->private_data = proc_sys_poll_event(table->poll);
609 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
613 sysctl_head_finish(head);
618 static bool proc_sys_fill_cache(struct file *file,
619 struct dir_context *ctx,
620 struct ctl_table_header *head,
621 struct ctl_table *table)
623 struct dentry *child, *dir = file->f_path.dentry;
627 unsigned type = DT_UNKNOWN;
629 qname.name = table->procname;
630 qname.len = strlen(table->procname);
631 qname.hash = full_name_hash(dir, qname.name, qname.len);
633 child = d_lookup(dir, &qname);
635 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
636 child = d_alloc_parallel(dir, &qname, &wq);
639 if (d_in_lookup(child)) {
640 inode = proc_sys_make_inode(dir->d_sb, head, table);
642 d_lookup_done(child);
646 d_set_d_op(child, &proc_sys_dentry_operations);
650 inode = d_inode(child);
652 type = inode->i_mode >> 12;
654 return dir_emit(ctx, qname.name, qname.len, ino, type);
657 static bool proc_sys_link_fill_cache(struct file *file,
658 struct dir_context *ctx,
659 struct ctl_table_header *head,
660 struct ctl_table *table)
663 head = sysctl_head_grab(head);
665 if (S_ISLNK(table->mode)) {
666 /* It is not an error if we can not follow the link ignore it */
667 int err = sysctl_follow_link(&head, &table);
672 ret = proc_sys_fill_cache(file, ctx, head, table);
674 sysctl_head_finish(head);
678 static int scan(struct ctl_table_header *head, struct ctl_table *table,
679 unsigned long *pos, struct file *file,
680 struct dir_context *ctx)
684 if ((*pos)++ < ctx->pos)
687 if (unlikely(S_ISLNK(table->mode)))
688 res = proc_sys_link_fill_cache(file, ctx, head, table);
690 res = proc_sys_fill_cache(file, ctx, head, table);
698 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
700 struct ctl_table_header *head = grab_header(file_inode(file));
701 struct ctl_table_header *h = NULL;
702 struct ctl_table *entry;
703 struct ctl_dir *ctl_dir;
707 return PTR_ERR(head);
709 ctl_dir = container_of(head, struct ctl_dir, header);
711 if (!dir_emit_dots(file, ctx))
716 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
717 if (!scan(h, entry, &pos, file, ctx)) {
718 sysctl_head_finish(h);
722 sysctl_head_finish(head);
726 static int proc_sys_permission(struct inode *inode, int mask)
729 * sysctl entries that are not writeable,
730 * are _NOT_ writeable, capabilities or not.
732 struct ctl_table_header *head;
733 struct ctl_table *table;
736 /* Executable files are not allowed under /proc/sys/ */
737 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
740 head = grab_header(inode);
742 return PTR_ERR(head);
744 table = PROC_I(inode)->sysctl_entry;
745 if (!table) /* global root - r-xr-xr-x */
746 error = mask & MAY_WRITE ? -EACCES : 0;
747 else /* Use the permissions on the sysctl table entry */
748 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
750 sysctl_head_finish(head);
754 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
756 struct inode *inode = d_inode(dentry);
759 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
762 error = inode_change_ok(inode, attr);
766 setattr_copy(inode, attr);
767 mark_inode_dirty(inode);
771 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
773 struct inode *inode = d_inode(dentry);
774 struct ctl_table_header *head = grab_header(inode);
775 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
778 return PTR_ERR(head);
780 generic_fillattr(inode, stat);
782 stat->mode = (stat->mode & S_IFMT) | table->mode;
784 sysctl_head_finish(head);
788 static const struct file_operations proc_sys_file_operations = {
789 .open = proc_sys_open,
790 .poll = proc_sys_poll,
791 .read = proc_sys_read,
792 .write = proc_sys_write,
793 .llseek = default_llseek,
796 static const struct file_operations proc_sys_dir_file_operations = {
797 .read = generic_read_dir,
798 .iterate_shared = proc_sys_readdir,
799 .llseek = generic_file_llseek,
802 static const struct inode_operations proc_sys_inode_operations = {
803 .permission = proc_sys_permission,
804 .setattr = proc_sys_setattr,
805 .getattr = proc_sys_getattr,
808 static const struct inode_operations proc_sys_dir_operations = {
809 .lookup = proc_sys_lookup,
810 .permission = proc_sys_permission,
811 .setattr = proc_sys_setattr,
812 .getattr = proc_sys_getattr,
815 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
817 if (flags & LOOKUP_RCU)
819 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
822 static int proc_sys_delete(const struct dentry *dentry)
824 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
827 static int sysctl_is_seen(struct ctl_table_header *p)
829 struct ctl_table_set *set = p->set;
831 spin_lock(&sysctl_lock);
832 if (p->unregistering)
834 else if (!set->is_seen)
837 res = set->is_seen(set);
838 spin_unlock(&sysctl_lock);
842 static int proc_sys_compare(const struct dentry *dentry,
843 unsigned int len, const char *str, const struct qstr *name)
845 struct ctl_table_header *head;
848 /* Although proc doesn't have negative dentries, rcu-walk means
849 * that inode here can be NULL */
850 /* AV: can it, indeed? */
851 inode = d_inode_rcu(dentry);
854 if (name->len != len)
856 if (memcmp(name->name, str, len))
858 head = rcu_dereference(PROC_I(inode)->sysctl);
859 return !head || !sysctl_is_seen(head);
862 static const struct dentry_operations proc_sys_dentry_operations = {
863 .d_revalidate = proc_sys_revalidate,
864 .d_delete = proc_sys_delete,
865 .d_compare = proc_sys_compare,
868 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
869 const char *name, int namelen)
871 struct ctl_table_header *head;
872 struct ctl_table *entry;
874 entry = find_entry(&head, dir, name, namelen);
876 return ERR_PTR(-ENOENT);
877 if (!S_ISDIR(entry->mode))
878 return ERR_PTR(-ENOTDIR);
879 return container_of(head, struct ctl_dir, header);
882 static struct ctl_dir *new_dir(struct ctl_table_set *set,
883 const char *name, int namelen)
885 struct ctl_table *table;
887 struct ctl_node *node;
890 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
891 sizeof(struct ctl_table)*2 + namelen + 1,
896 node = (struct ctl_node *)(new + 1);
897 table = (struct ctl_table *)(node + 1);
898 new_name = (char *)(table + 2);
899 memcpy(new_name, name, namelen);
900 new_name[namelen] = '\0';
901 table[0].procname = new_name;
902 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
903 init_header(&new->header, set->dir.header.root, set, node, table);
909 * get_subdir - find or create a subdir with the specified name.
910 * @dir: Directory to create the subdirectory in
911 * @name: The name of the subdirectory to find or create
912 * @namelen: The length of name
914 * Takes a directory with an elevated reference count so we know that
915 * if we drop the lock the directory will not go away. Upon success
916 * the reference is moved from @dir to the returned subdirectory.
917 * Upon error an error code is returned and the reference on @dir is
920 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
921 const char *name, int namelen)
923 struct ctl_table_set *set = dir->header.set;
924 struct ctl_dir *subdir, *new = NULL;
927 spin_lock(&sysctl_lock);
928 subdir = find_subdir(dir, name, namelen);
931 if (PTR_ERR(subdir) != -ENOENT)
934 spin_unlock(&sysctl_lock);
935 new = new_dir(set, name, namelen);
936 spin_lock(&sysctl_lock);
937 subdir = ERR_PTR(-ENOMEM);
941 /* Was the subdir added while we dropped the lock? */
942 subdir = find_subdir(dir, name, namelen);
945 if (PTR_ERR(subdir) != -ENOENT)
948 /* Nope. Use the our freshly made directory entry. */
949 err = insert_header(dir, &new->header);
950 subdir = ERR_PTR(err);
955 subdir->header.nreg++;
957 if (IS_ERR(subdir)) {
958 pr_err("sysctl could not get directory: ");
959 sysctl_print_dir(dir);
960 pr_cont("/%*.*s %ld\n",
961 namelen, namelen, name, PTR_ERR(subdir));
963 drop_sysctl_table(&dir->header);
965 drop_sysctl_table(&new->header);
966 spin_unlock(&sysctl_lock);
970 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
972 struct ctl_dir *parent;
973 const char *procname;
974 if (!dir->header.parent)
976 parent = xlate_dir(set, dir->header.parent);
979 procname = dir->header.ctl_table[0].procname;
980 return find_subdir(parent, procname, strlen(procname));
983 static int sysctl_follow_link(struct ctl_table_header **phead,
984 struct ctl_table **pentry)
986 struct ctl_table_header *head;
987 struct ctl_table_root *root;
988 struct ctl_table_set *set;
989 struct ctl_table *entry;
994 spin_lock(&sysctl_lock);
995 root = (*pentry)->data;
996 set = lookup_header_set(root);
997 dir = xlate_dir(set, (*phead)->parent);
1001 const char *procname = (*pentry)->procname;
1003 entry = find_entry(&head, dir, procname, strlen(procname));
1005 if (entry && use_table(head)) {
1006 unuse_table(*phead);
1013 spin_unlock(&sysctl_lock);
1017 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1019 struct va_format vaf;
1022 va_start(args, fmt);
1026 pr_err("sysctl table check failed: %s/%s %pV\n",
1027 path, table->procname, &vaf);
1033 static int sysctl_check_table(const char *path, struct ctl_table *table)
1036 for (; table->procname; table++) {
1038 err = sysctl_err(path, table, "Not a file");
1040 if ((table->proc_handler == proc_dostring) ||
1041 (table->proc_handler == proc_dointvec) ||
1042 (table->proc_handler == proc_dointvec_minmax) ||
1043 (table->proc_handler == proc_dointvec_jiffies) ||
1044 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1045 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1046 (table->proc_handler == proc_doulongvec_minmax) ||
1047 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1049 err = sysctl_err(path, table, "No data");
1051 err = sysctl_err(path, table, "No maxlen");
1053 if (!table->proc_handler)
1054 err = sysctl_err(path, table, "No proc_handler");
1056 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1057 err = sysctl_err(path, table, "bogus .mode 0%o",
1063 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1064 struct ctl_table_root *link_root)
1066 struct ctl_table *link_table, *entry, *link;
1067 struct ctl_table_header *links;
1068 struct ctl_node *node;
1070 int nr_entries, name_bytes;
1074 for (entry = table; entry->procname; entry++) {
1076 name_bytes += strlen(entry->procname) + 1;
1079 links = kzalloc(sizeof(struct ctl_table_header) +
1080 sizeof(struct ctl_node)*nr_entries +
1081 sizeof(struct ctl_table)*(nr_entries + 1) +
1088 node = (struct ctl_node *)(links + 1);
1089 link_table = (struct ctl_table *)(node + nr_entries);
1090 link_name = (char *)&link_table[nr_entries + 1];
1092 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1093 int len = strlen(entry->procname) + 1;
1094 memcpy(link_name, entry->procname, len);
1095 link->procname = link_name;
1096 link->mode = S_IFLNK|S_IRWXUGO;
1097 link->data = link_root;
1100 init_header(links, dir->header.root, dir->header.set, node, link_table);
1101 links->nreg = nr_entries;
1106 static bool get_links(struct ctl_dir *dir,
1107 struct ctl_table *table, struct ctl_table_root *link_root)
1109 struct ctl_table_header *head;
1110 struct ctl_table *entry, *link;
1112 /* Are there links available for every entry in table? */
1113 for (entry = table; entry->procname; entry++) {
1114 const char *procname = entry->procname;
1115 link = find_entry(&head, dir, procname, strlen(procname));
1118 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1120 if (S_ISLNK(link->mode) && (link->data == link_root))
1125 /* The checks passed. Increase the registration count on the links */
1126 for (entry = table; entry->procname; entry++) {
1127 const char *procname = entry->procname;
1128 link = find_entry(&head, dir, procname, strlen(procname));
1134 static int insert_links(struct ctl_table_header *head)
1136 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1137 struct ctl_dir *core_parent = NULL;
1138 struct ctl_table_header *links;
1141 if (head->set == root_set)
1144 core_parent = xlate_dir(root_set, head->parent);
1145 if (IS_ERR(core_parent))
1148 if (get_links(core_parent, head->ctl_table, head->root))
1151 core_parent->header.nreg++;
1152 spin_unlock(&sysctl_lock);
1154 links = new_links(core_parent, head->ctl_table, head->root);
1156 spin_lock(&sysctl_lock);
1162 if (get_links(core_parent, head->ctl_table, head->root)) {
1167 err = insert_header(core_parent, links);
1171 drop_sysctl_table(&core_parent->header);
1176 * __register_sysctl_table - register a leaf sysctl table
1177 * @set: Sysctl tree to register on
1178 * @path: The path to the directory the sysctl table is in.
1179 * @table: the top-level table structure
1181 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1182 * array. A completely 0 filled entry terminates the table.
1184 * The members of the &struct ctl_table structure are used as follows:
1186 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1187 * enter a sysctl file
1189 * data - a pointer to data for use by proc_handler
1191 * maxlen - the maximum size in bytes of the data
1193 * mode - the file permissions for the /proc/sys file
1195 * child - must be %NULL.
1197 * proc_handler - the text handler routine (described below)
1199 * extra1, extra2 - extra pointers usable by the proc handler routines
1201 * Leaf nodes in the sysctl tree will be represented by a single file
1202 * under /proc; non-leaf nodes will be represented by directories.
1204 * There must be a proc_handler routine for any terminal nodes.
1205 * Several default handlers are available to cover common cases -
1207 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1208 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1209 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1211 * It is the handler's job to read the input buffer from user memory
1212 * and process it. The handler should return 0 on success.
1214 * This routine returns %NULL on a failure to register, and a pointer
1215 * to the table header on success.
1217 struct ctl_table_header *__register_sysctl_table(
1218 struct ctl_table_set *set,
1219 const char *path, struct ctl_table *table)
1221 struct ctl_table_root *root = set->dir.header.root;
1222 struct ctl_table_header *header;
1223 const char *name, *nextname;
1224 struct ctl_dir *dir;
1225 struct ctl_table *entry;
1226 struct ctl_node *node;
1229 for (entry = table; entry->procname; entry++)
1232 header = kzalloc(sizeof(struct ctl_table_header) +
1233 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1237 node = (struct ctl_node *)(header + 1);
1238 init_header(header, root, set, node, table);
1239 if (sysctl_check_table(path, table))
1242 spin_lock(&sysctl_lock);
1244 /* Reference moved down the diretory tree get_subdir */
1246 spin_unlock(&sysctl_lock);
1248 /* Find the directory for the ctl_table */
1249 for (name = path; name; name = nextname) {
1251 nextname = strchr(name, '/');
1253 namelen = nextname - name;
1256 namelen = strlen(name);
1261 dir = get_subdir(dir, name, namelen);
1266 spin_lock(&sysctl_lock);
1267 if (insert_header(dir, header))
1268 goto fail_put_dir_locked;
1270 drop_sysctl_table(&dir->header);
1271 spin_unlock(&sysctl_lock);
1275 fail_put_dir_locked:
1276 drop_sysctl_table(&dir->header);
1277 spin_unlock(&sysctl_lock);
1285 * register_sysctl - register a sysctl table
1286 * @path: The path to the directory the sysctl table is in.
1287 * @table: the table structure
1289 * Register a sysctl table. @table should be a filled in ctl_table
1290 * array. A completely 0 filled entry terminates the table.
1292 * See __register_sysctl_table for more details.
1294 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1296 return __register_sysctl_table(&sysctl_table_root.default_set,
1299 EXPORT_SYMBOL(register_sysctl);
1301 static char *append_path(const char *path, char *pos, const char *name)
1304 namelen = strlen(name);
1305 if (((pos - path) + namelen + 2) >= PATH_MAX)
1307 memcpy(pos, name, namelen);
1309 pos[namelen + 1] = '\0';
1314 static int count_subheaders(struct ctl_table *table)
1317 int nr_subheaders = 0;
1318 struct ctl_table *entry;
1320 /* special case: no directory and empty directory */
1321 if (!table || !table->procname)
1324 for (entry = table; entry->procname; entry++) {
1326 nr_subheaders += count_subheaders(entry->child);
1330 return nr_subheaders + has_files;
1333 static int register_leaf_sysctl_tables(const char *path, char *pos,
1334 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1335 struct ctl_table *table)
1337 struct ctl_table *ctl_table_arg = NULL;
1338 struct ctl_table *entry, *files;
1343 for (entry = table; entry->procname; entry++) {
1351 /* If there are mixed files and directories we need a new table */
1352 if (nr_dirs && nr_files) {
1353 struct ctl_table *new;
1354 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1359 ctl_table_arg = files;
1360 for (new = files, entry = table; entry->procname; entry++) {
1368 /* Register everything except a directory full of subdirectories */
1369 if (nr_files || !nr_dirs) {
1370 struct ctl_table_header *header;
1371 header = __register_sysctl_table(set, path, files);
1373 kfree(ctl_table_arg);
1377 /* Remember if we need to free the file table */
1378 header->ctl_table_arg = ctl_table_arg;
1379 **subheader = header;
1383 /* Recurse into the subdirectories. */
1384 for (entry = table; entry->procname; entry++) {
1390 err = -ENAMETOOLONG;
1391 child_pos = append_path(path, pos, entry->procname);
1395 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1403 /* On failure our caller will unregister all registered subheaders */
1408 * __register_sysctl_paths - register a sysctl table hierarchy
1409 * @set: Sysctl tree to register on
1410 * @path: The path to the directory the sysctl table is in.
1411 * @table: the top-level table structure
1413 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1414 * array. A completely 0 filled entry terminates the table.
1416 * See __register_sysctl_table for more details.
1418 struct ctl_table_header *__register_sysctl_paths(
1419 struct ctl_table_set *set,
1420 const struct ctl_path *path, struct ctl_table *table)
1422 struct ctl_table *ctl_table_arg = table;
1423 int nr_subheaders = count_subheaders(table);
1424 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1425 const struct ctl_path *component;
1426 char *new_path, *pos;
1428 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1433 for (component = path; component->procname; component++) {
1434 pos = append_path(new_path, pos, component->procname);
1438 while (table->procname && table->child && !table[1].procname) {
1439 pos = append_path(new_path, pos, table->procname);
1442 table = table->child;
1444 if (nr_subheaders == 1) {
1445 header = __register_sysctl_table(set, new_path, table);
1447 header->ctl_table_arg = ctl_table_arg;
1449 header = kzalloc(sizeof(*header) +
1450 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1454 subheaders = (struct ctl_table_header **) (header + 1);
1455 subheader = subheaders;
1456 header->ctl_table_arg = ctl_table_arg;
1458 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1460 goto err_register_leaves;
1467 err_register_leaves:
1468 while (subheader > subheaders) {
1469 struct ctl_table_header *subh = *(--subheader);
1470 struct ctl_table *table = subh->ctl_table_arg;
1471 unregister_sysctl_table(subh);
1480 * register_sysctl_table_path - register a sysctl table hierarchy
1481 * @path: The path to the directory the sysctl table is in.
1482 * @table: the top-level table structure
1484 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1485 * array. A completely 0 filled entry terminates the table.
1487 * See __register_sysctl_paths for more details.
1489 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1490 struct ctl_table *table)
1492 return __register_sysctl_paths(&sysctl_table_root.default_set,
1495 EXPORT_SYMBOL(register_sysctl_paths);
1498 * register_sysctl_table - register a sysctl table hierarchy
1499 * @table: the top-level table structure
1501 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1502 * array. A completely 0 filled entry terminates the table.
1504 * See register_sysctl_paths for more details.
1506 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1508 static const struct ctl_path null_path[] = { {} };
1510 return register_sysctl_paths(null_path, table);
1512 EXPORT_SYMBOL(register_sysctl_table);
1514 static void put_links(struct ctl_table_header *header)
1516 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1517 struct ctl_table_root *root = header->root;
1518 struct ctl_dir *parent = header->parent;
1519 struct ctl_dir *core_parent;
1520 struct ctl_table *entry;
1522 if (header->set == root_set)
1525 core_parent = xlate_dir(root_set, parent);
1526 if (IS_ERR(core_parent))
1529 for (entry = header->ctl_table; entry->procname; entry++) {
1530 struct ctl_table_header *link_head;
1531 struct ctl_table *link;
1532 const char *name = entry->procname;
1534 link = find_entry(&link_head, core_parent, name, strlen(name));
1536 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1537 (S_ISLNK(link->mode) && (link->data == root)))) {
1538 drop_sysctl_table(link_head);
1541 pr_err("sysctl link missing during unregister: ");
1542 sysctl_print_dir(parent);
1543 pr_cont("/%s\n", name);
1548 static void drop_sysctl_table(struct ctl_table_header *header)
1550 struct ctl_dir *parent = header->parent;
1556 start_unregistering(header);
1557 if (!--header->count)
1558 kfree_rcu(header, rcu);
1561 drop_sysctl_table(&parent->header);
1565 * unregister_sysctl_table - unregister a sysctl table hierarchy
1566 * @header: the header returned from register_sysctl_table
1568 * Unregisters the sysctl table and all children. proc entries may not
1569 * actually be removed until they are no longer used by anyone.
1571 void unregister_sysctl_table(struct ctl_table_header * header)
1579 nr_subheaders = count_subheaders(header->ctl_table_arg);
1580 if (unlikely(nr_subheaders > 1)) {
1581 struct ctl_table_header **subheaders;
1584 subheaders = (struct ctl_table_header **)(header + 1);
1585 for (i = nr_subheaders -1; i >= 0; i--) {
1586 struct ctl_table_header *subh = subheaders[i];
1587 struct ctl_table *table = subh->ctl_table_arg;
1588 unregister_sysctl_table(subh);
1595 spin_lock(&sysctl_lock);
1596 drop_sysctl_table(header);
1597 spin_unlock(&sysctl_lock);
1599 EXPORT_SYMBOL(unregister_sysctl_table);
1601 void setup_sysctl_set(struct ctl_table_set *set,
1602 struct ctl_table_root *root,
1603 int (*is_seen)(struct ctl_table_set *))
1605 memset(set, 0, sizeof(*set));
1606 set->is_seen = is_seen;
1607 init_header(&set->dir.header, root, set, NULL, root_table);
1610 void retire_sysctl_set(struct ctl_table_set *set)
1612 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1615 int __init proc_sys_init(void)
1617 struct proc_dir_entry *proc_sys_root;
1619 proc_sys_root = proc_mkdir("sys", NULL);
1620 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1621 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1622 proc_sys_root->nlink = 0;
1624 return sysctl_init();