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, struct nsproxy *namespaces);
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, struct nsproxy *namespaces)
324 struct ctl_table_set *set = &root->default_set;
326 set = root->lookup(root, namespaces);
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)
434 struct proc_inode *ei;
436 inode = new_inode(sb);
440 inode->i_ino = get_next_ino();
442 sysctl_head_get(head);
445 ei->sysctl_entry = table;
447 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
448 inode->i_mode = table->mode;
449 if (!S_ISDIR(table->mode)) {
450 inode->i_mode |= S_IFREG;
451 inode->i_op = &proc_sys_inode_operations;
452 inode->i_fop = &proc_sys_file_operations;
454 inode->i_mode |= S_IFDIR;
455 inode->i_op = &proc_sys_dir_operations;
456 inode->i_fop = &proc_sys_dir_file_operations;
457 if (is_empty_dir(head))
458 make_empty_dir_inode(inode);
464 static struct ctl_table_header *grab_header(struct inode *inode)
466 struct ctl_table_header *head = PROC_I(inode)->sysctl;
468 head = &sysctl_table_root.default_set.dir.header;
469 return sysctl_head_grab(head);
472 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
475 struct ctl_table_header *head = grab_header(dir);
476 struct ctl_table_header *h = NULL;
477 struct qstr *name = &dentry->d_name;
480 struct dentry *err = ERR_PTR(-ENOENT);
481 struct ctl_dir *ctl_dir;
485 return ERR_CAST(head);
487 ctl_dir = container_of(head, struct ctl_dir, header);
489 p = lookup_entry(&h, ctl_dir, name->name, name->len);
493 if (S_ISLNK(p->mode)) {
494 ret = sysctl_follow_link(&h, &p, current->nsproxy);
500 err = ERR_PTR(-ENOMEM);
501 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
506 d_set_d_op(dentry, &proc_sys_dentry_operations);
507 d_add(dentry, inode);
511 sysctl_head_finish(h);
512 sysctl_head_finish(head);
516 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
517 size_t count, loff_t *ppos, int write)
519 struct inode *inode = file_inode(filp);
520 struct ctl_table_header *head = grab_header(inode);
521 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
526 return PTR_ERR(head);
529 * At this point we know that the sysctl was not unregistered
530 * and won't be until we finish.
533 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
536 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
538 if (!table->proc_handler)
541 /* careful: calling conventions are nasty here */
543 error = table->proc_handler(table, write, buf, &res, ppos);
547 sysctl_head_finish(head);
552 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
553 size_t count, loff_t *ppos)
555 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
558 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
559 size_t count, loff_t *ppos)
561 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
564 static int proc_sys_open(struct inode *inode, struct file *filp)
566 struct ctl_table_header *head = grab_header(inode);
567 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
569 /* sysctl was unregistered */
571 return PTR_ERR(head);
574 filp->private_data = proc_sys_poll_event(table->poll);
576 sysctl_head_finish(head);
581 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
583 struct inode *inode = file_inode(filp);
584 struct ctl_table_header *head = grab_header(inode);
585 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
586 unsigned int ret = DEFAULT_POLLMASK;
589 /* sysctl was unregistered */
591 return POLLERR | POLLHUP;
593 if (!table->proc_handler)
599 event = (unsigned long)filp->private_data;
600 poll_wait(filp, &table->poll->wait, wait);
602 if (event != atomic_read(&table->poll->event)) {
603 filp->private_data = proc_sys_poll_event(table->poll);
604 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
608 sysctl_head_finish(head);
613 static bool proc_sys_fill_cache(struct file *file,
614 struct dir_context *ctx,
615 struct ctl_table_header *head,
616 struct ctl_table *table)
618 struct dentry *child, *dir = file->f_path.dentry;
622 unsigned type = DT_UNKNOWN;
624 qname.name = table->procname;
625 qname.len = strlen(table->procname);
626 qname.hash = full_name_hash(qname.name, qname.len);
628 child = d_lookup(dir, &qname);
630 child = d_alloc(dir, &qname);
632 inode = proc_sys_make_inode(dir->d_sb, head, table);
637 d_set_d_op(child, &proc_sys_dentry_operations);
644 inode = d_inode(child);
646 type = inode->i_mode >> 12;
648 return dir_emit(ctx, qname.name, qname.len, ino, type);
651 static bool proc_sys_link_fill_cache(struct file *file,
652 struct dir_context *ctx,
653 struct ctl_table_header *head,
654 struct ctl_table *table)
658 head = sysctl_head_grab(head);
662 if (S_ISLNK(table->mode)) {
663 /* It is not an error if we can not follow the link ignore it */
664 int err = sysctl_follow_link(&head, &table, current->nsproxy);
669 ret = proc_sys_fill_cache(file, ctx, head, table);
671 sysctl_head_finish(head);
675 static int scan(struct ctl_table_header *head, struct ctl_table *table,
676 unsigned long *pos, struct file *file,
677 struct dir_context *ctx)
681 if ((*pos)++ < ctx->pos)
684 if (unlikely(S_ISLNK(table->mode)))
685 res = proc_sys_link_fill_cache(file, ctx, head, table);
687 res = proc_sys_fill_cache(file, ctx, head, table);
695 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
697 struct ctl_table_header *head = grab_header(file_inode(file));
698 struct ctl_table_header *h = NULL;
699 struct ctl_table *entry;
700 struct ctl_dir *ctl_dir;
704 return PTR_ERR(head);
706 ctl_dir = container_of(head, struct ctl_dir, header);
708 if (!dir_emit_dots(file, ctx))
713 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
714 if (!scan(h, entry, &pos, file, ctx)) {
715 sysctl_head_finish(h);
720 sysctl_head_finish(head);
724 static int proc_sys_permission(struct inode *inode, int mask)
727 * sysctl entries that are not writeable,
728 * are _NOT_ writeable, capabilities or not.
730 struct ctl_table_header *head;
731 struct ctl_table *table;
734 /* Executable files are not allowed under /proc/sys/ */
735 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
738 head = grab_header(inode);
740 return PTR_ERR(head);
742 table = PROC_I(inode)->sysctl_entry;
743 if (!table) /* global root - r-xr-xr-x */
744 error = mask & MAY_WRITE ? -EACCES : 0;
745 else /* Use the permissions on the sysctl table entry */
746 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
748 sysctl_head_finish(head);
752 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
754 struct inode *inode = d_inode(dentry);
757 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
760 error = inode_change_ok(inode, attr);
764 setattr_copy(inode, attr);
765 mark_inode_dirty(inode);
769 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
771 struct inode *inode = d_inode(dentry);
772 struct ctl_table_header *head = grab_header(inode);
773 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
776 return PTR_ERR(head);
778 generic_fillattr(inode, stat);
780 stat->mode = (stat->mode & S_IFMT) | table->mode;
782 sysctl_head_finish(head);
786 static const struct file_operations proc_sys_file_operations = {
787 .open = proc_sys_open,
788 .poll = proc_sys_poll,
789 .read = proc_sys_read,
790 .write = proc_sys_write,
791 .llseek = default_llseek,
794 static const struct file_operations proc_sys_dir_file_operations = {
795 .read = generic_read_dir,
796 .iterate = proc_sys_readdir,
797 .llseek = generic_file_llseek,
800 static const struct inode_operations proc_sys_inode_operations = {
801 .permission = proc_sys_permission,
802 .setattr = proc_sys_setattr,
803 .getattr = proc_sys_getattr,
806 static const struct inode_operations proc_sys_dir_operations = {
807 .lookup = proc_sys_lookup,
808 .permission = proc_sys_permission,
809 .setattr = proc_sys_setattr,
810 .getattr = proc_sys_getattr,
813 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
815 if (flags & LOOKUP_RCU)
817 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
820 static int proc_sys_delete(const struct dentry *dentry)
822 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
825 static int sysctl_is_seen(struct ctl_table_header *p)
827 struct ctl_table_set *set = p->set;
829 spin_lock(&sysctl_lock);
830 if (p->unregistering)
832 else if (!set->is_seen)
835 res = set->is_seen(set);
836 spin_unlock(&sysctl_lock);
840 static int proc_sys_compare(const struct dentry *parent, const struct dentry *dentry,
841 unsigned int len, const char *str, const struct qstr *name)
843 struct ctl_table_header *head;
846 /* Although proc doesn't have negative dentries, rcu-walk means
847 * that inode here can be NULL */
848 /* AV: can it, indeed? */
849 inode = d_inode_rcu(dentry);
852 if (name->len != len)
854 if (memcmp(name->name, str, len))
856 head = rcu_dereference(PROC_I(inode)->sysctl);
857 return !head || !sysctl_is_seen(head);
860 static const struct dentry_operations proc_sys_dentry_operations = {
861 .d_revalidate = proc_sys_revalidate,
862 .d_delete = proc_sys_delete,
863 .d_compare = proc_sys_compare,
866 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
867 const char *name, int namelen)
869 struct ctl_table_header *head;
870 struct ctl_table *entry;
872 entry = find_entry(&head, dir, name, namelen);
874 return ERR_PTR(-ENOENT);
875 if (!S_ISDIR(entry->mode))
876 return ERR_PTR(-ENOTDIR);
877 return container_of(head, struct ctl_dir, header);
880 static struct ctl_dir *new_dir(struct ctl_table_set *set,
881 const char *name, int namelen)
883 struct ctl_table *table;
885 struct ctl_node *node;
888 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
889 sizeof(struct ctl_table)*2 + namelen + 1,
894 node = (struct ctl_node *)(new + 1);
895 table = (struct ctl_table *)(node + 1);
896 new_name = (char *)(table + 2);
897 memcpy(new_name, name, namelen);
898 new_name[namelen] = '\0';
899 table[0].procname = new_name;
900 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
901 init_header(&new->header, set->dir.header.root, set, node, table);
907 * get_subdir - find or create a subdir with the specified name.
908 * @dir: Directory to create the subdirectory in
909 * @name: The name of the subdirectory to find or create
910 * @namelen: The length of name
912 * Takes a directory with an elevated reference count so we know that
913 * if we drop the lock the directory will not go away. Upon success
914 * the reference is moved from @dir to the returned subdirectory.
915 * Upon error an error code is returned and the reference on @dir is
918 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
919 const char *name, int namelen)
921 struct ctl_table_set *set = dir->header.set;
922 struct ctl_dir *subdir, *new = NULL;
925 spin_lock(&sysctl_lock);
926 subdir = find_subdir(dir, name, namelen);
929 if (PTR_ERR(subdir) != -ENOENT)
932 spin_unlock(&sysctl_lock);
933 new = new_dir(set, name, namelen);
934 spin_lock(&sysctl_lock);
935 subdir = ERR_PTR(-ENOMEM);
939 /* Was the subdir added while we dropped the lock? */
940 subdir = find_subdir(dir, name, namelen);
943 if (PTR_ERR(subdir) != -ENOENT)
946 /* Nope. Use the our freshly made directory entry. */
947 err = insert_header(dir, &new->header);
948 subdir = ERR_PTR(err);
953 subdir->header.nreg++;
955 if (IS_ERR(subdir)) {
956 pr_err("sysctl could not get directory: ");
957 sysctl_print_dir(dir);
958 pr_cont("/%*.*s %ld\n",
959 namelen, namelen, name, PTR_ERR(subdir));
961 drop_sysctl_table(&dir->header);
963 drop_sysctl_table(&new->header);
964 spin_unlock(&sysctl_lock);
968 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
970 struct ctl_dir *parent;
971 const char *procname;
972 if (!dir->header.parent)
974 parent = xlate_dir(set, dir->header.parent);
977 procname = dir->header.ctl_table[0].procname;
978 return find_subdir(parent, procname, strlen(procname));
981 static int sysctl_follow_link(struct ctl_table_header **phead,
982 struct ctl_table **pentry, struct nsproxy *namespaces)
984 struct ctl_table_header *head;
985 struct ctl_table_root *root;
986 struct ctl_table_set *set;
987 struct ctl_table *entry;
992 spin_lock(&sysctl_lock);
993 root = (*pentry)->data;
994 set = lookup_header_set(root, namespaces);
995 dir = xlate_dir(set, (*phead)->parent);
999 const char *procname = (*pentry)->procname;
1001 entry = find_entry(&head, dir, procname, strlen(procname));
1003 if (entry && use_table(head)) {
1004 unuse_table(*phead);
1011 spin_unlock(&sysctl_lock);
1015 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1017 struct va_format vaf;
1020 va_start(args, fmt);
1024 pr_err("sysctl table check failed: %s/%s %pV\n",
1025 path, table->procname, &vaf);
1031 static int sysctl_check_table(const char *path, struct ctl_table *table)
1034 for (; table->procname; table++) {
1036 err = sysctl_err(path, table, "Not a file");
1038 if ((table->proc_handler == proc_dostring) ||
1039 (table->proc_handler == proc_dointvec) ||
1040 (table->proc_handler == proc_dointvec_minmax) ||
1041 (table->proc_handler == proc_dointvec_jiffies) ||
1042 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1043 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1044 (table->proc_handler == proc_doulongvec_minmax) ||
1045 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1047 err = sysctl_err(path, table, "No data");
1049 err = sysctl_err(path, table, "No maxlen");
1051 if (!table->proc_handler)
1052 err = sysctl_err(path, table, "No proc_handler");
1054 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1055 err = sysctl_err(path, table, "bogus .mode 0%o",
1061 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1062 struct ctl_table_root *link_root)
1064 struct ctl_table *link_table, *entry, *link;
1065 struct ctl_table_header *links;
1066 struct ctl_node *node;
1068 int nr_entries, name_bytes;
1072 for (entry = table; entry->procname; entry++) {
1074 name_bytes += strlen(entry->procname) + 1;
1077 links = kzalloc(sizeof(struct ctl_table_header) +
1078 sizeof(struct ctl_node)*nr_entries +
1079 sizeof(struct ctl_table)*(nr_entries + 1) +
1086 node = (struct ctl_node *)(links + 1);
1087 link_table = (struct ctl_table *)(node + nr_entries);
1088 link_name = (char *)&link_table[nr_entries + 1];
1090 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1091 int len = strlen(entry->procname) + 1;
1092 memcpy(link_name, entry->procname, len);
1093 link->procname = link_name;
1094 link->mode = S_IFLNK|S_IRWXUGO;
1095 link->data = link_root;
1098 init_header(links, dir->header.root, dir->header.set, node, link_table);
1099 links->nreg = nr_entries;
1104 static bool get_links(struct ctl_dir *dir,
1105 struct ctl_table *table, struct ctl_table_root *link_root)
1107 struct ctl_table_header *head;
1108 struct ctl_table *entry, *link;
1110 /* Are there links available for every entry in table? */
1111 for (entry = table; entry->procname; entry++) {
1112 const char *procname = entry->procname;
1113 link = find_entry(&head, dir, procname, strlen(procname));
1116 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1118 if (S_ISLNK(link->mode) && (link->data == link_root))
1123 /* The checks passed. Increase the registration count on the links */
1124 for (entry = table; entry->procname; entry++) {
1125 const char *procname = entry->procname;
1126 link = find_entry(&head, dir, procname, strlen(procname));
1132 static int insert_links(struct ctl_table_header *head)
1134 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1135 struct ctl_dir *core_parent = NULL;
1136 struct ctl_table_header *links;
1139 if (head->set == root_set)
1142 core_parent = xlate_dir(root_set, head->parent);
1143 if (IS_ERR(core_parent))
1146 if (get_links(core_parent, head->ctl_table, head->root))
1149 core_parent->header.nreg++;
1150 spin_unlock(&sysctl_lock);
1152 links = new_links(core_parent, head->ctl_table, head->root);
1154 spin_lock(&sysctl_lock);
1160 if (get_links(core_parent, head->ctl_table, head->root)) {
1165 err = insert_header(core_parent, links);
1169 drop_sysctl_table(&core_parent->header);
1174 * __register_sysctl_table - register a leaf sysctl table
1175 * @set: Sysctl tree to register on
1176 * @path: The path to the directory the sysctl table is in.
1177 * @table: the top-level table structure
1179 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1180 * array. A completely 0 filled entry terminates the table.
1182 * The members of the &struct ctl_table structure are used as follows:
1184 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1185 * enter a sysctl file
1187 * data - a pointer to data for use by proc_handler
1189 * maxlen - the maximum size in bytes of the data
1191 * mode - the file permissions for the /proc/sys file
1193 * child - must be %NULL.
1195 * proc_handler - the text handler routine (described below)
1197 * extra1, extra2 - extra pointers usable by the proc handler routines
1199 * Leaf nodes in the sysctl tree will be represented by a single file
1200 * under /proc; non-leaf nodes will be represented by directories.
1202 * There must be a proc_handler routine for any terminal nodes.
1203 * Several default handlers are available to cover common cases -
1205 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1206 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1207 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1209 * It is the handler's job to read the input buffer from user memory
1210 * and process it. The handler should return 0 on success.
1212 * This routine returns %NULL on a failure to register, and a pointer
1213 * to the table header on success.
1215 struct ctl_table_header *__register_sysctl_table(
1216 struct ctl_table_set *set,
1217 const char *path, struct ctl_table *table)
1219 struct ctl_table_root *root = set->dir.header.root;
1220 struct ctl_table_header *header;
1221 const char *name, *nextname;
1222 struct ctl_dir *dir;
1223 struct ctl_table *entry;
1224 struct ctl_node *node;
1227 for (entry = table; entry->procname; entry++)
1230 header = kzalloc(sizeof(struct ctl_table_header) +
1231 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1235 node = (struct ctl_node *)(header + 1);
1236 init_header(header, root, set, node, table);
1237 if (sysctl_check_table(path, table))
1240 spin_lock(&sysctl_lock);
1242 /* Reference moved down the diretory tree get_subdir */
1244 spin_unlock(&sysctl_lock);
1246 /* Find the directory for the ctl_table */
1247 for (name = path; name; name = nextname) {
1249 nextname = strchr(name, '/');
1251 namelen = nextname - name;
1254 namelen = strlen(name);
1259 dir = get_subdir(dir, name, namelen);
1264 spin_lock(&sysctl_lock);
1265 if (insert_header(dir, header))
1266 goto fail_put_dir_locked;
1268 drop_sysctl_table(&dir->header);
1269 spin_unlock(&sysctl_lock);
1273 fail_put_dir_locked:
1274 drop_sysctl_table(&dir->header);
1275 spin_unlock(&sysctl_lock);
1283 * register_sysctl - register a sysctl table
1284 * @path: The path to the directory the sysctl table is in.
1285 * @table: the table structure
1287 * Register a sysctl table. @table should be a filled in ctl_table
1288 * array. A completely 0 filled entry terminates the table.
1290 * See __register_sysctl_table for more details.
1292 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1294 return __register_sysctl_table(&sysctl_table_root.default_set,
1297 EXPORT_SYMBOL(register_sysctl);
1299 static char *append_path(const char *path, char *pos, const char *name)
1302 namelen = strlen(name);
1303 if (((pos - path) + namelen + 2) >= PATH_MAX)
1305 memcpy(pos, name, namelen);
1307 pos[namelen + 1] = '\0';
1312 static int count_subheaders(struct ctl_table *table)
1315 int nr_subheaders = 0;
1316 struct ctl_table *entry;
1318 /* special case: no directory and empty directory */
1319 if (!table || !table->procname)
1322 for (entry = table; entry->procname; entry++) {
1324 nr_subheaders += count_subheaders(entry->child);
1328 return nr_subheaders + has_files;
1331 static int register_leaf_sysctl_tables(const char *path, char *pos,
1332 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1333 struct ctl_table *table)
1335 struct ctl_table *ctl_table_arg = NULL;
1336 struct ctl_table *entry, *files;
1341 for (entry = table; entry->procname; entry++) {
1349 /* If there are mixed files and directories we need a new table */
1350 if (nr_dirs && nr_files) {
1351 struct ctl_table *new;
1352 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1357 ctl_table_arg = files;
1358 for (new = files, entry = table; entry->procname; entry++) {
1366 /* Register everything except a directory full of subdirectories */
1367 if (nr_files || !nr_dirs) {
1368 struct ctl_table_header *header;
1369 header = __register_sysctl_table(set, path, files);
1371 kfree(ctl_table_arg);
1375 /* Remember if we need to free the file table */
1376 header->ctl_table_arg = ctl_table_arg;
1377 **subheader = header;
1381 /* Recurse into the subdirectories. */
1382 for (entry = table; entry->procname; entry++) {
1388 err = -ENAMETOOLONG;
1389 child_pos = append_path(path, pos, entry->procname);
1393 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1401 /* On failure our caller will unregister all registered subheaders */
1406 * __register_sysctl_paths - register a sysctl table hierarchy
1407 * @set: Sysctl tree to register on
1408 * @path: The path to the directory the sysctl table is in.
1409 * @table: the top-level table structure
1411 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1412 * array. A completely 0 filled entry terminates the table.
1414 * See __register_sysctl_table for more details.
1416 struct ctl_table_header *__register_sysctl_paths(
1417 struct ctl_table_set *set,
1418 const struct ctl_path *path, struct ctl_table *table)
1420 struct ctl_table *ctl_table_arg = table;
1421 int nr_subheaders = count_subheaders(table);
1422 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1423 const struct ctl_path *component;
1424 char *new_path, *pos;
1426 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1431 for (component = path; component->procname; component++) {
1432 pos = append_path(new_path, pos, component->procname);
1436 while (table->procname && table->child && !table[1].procname) {
1437 pos = append_path(new_path, pos, table->procname);
1440 table = table->child;
1442 if (nr_subheaders == 1) {
1443 header = __register_sysctl_table(set, new_path, table);
1445 header->ctl_table_arg = ctl_table_arg;
1447 header = kzalloc(sizeof(*header) +
1448 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1452 subheaders = (struct ctl_table_header **) (header + 1);
1453 subheader = subheaders;
1454 header->ctl_table_arg = ctl_table_arg;
1456 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1458 goto err_register_leaves;
1465 err_register_leaves:
1466 while (subheader > subheaders) {
1467 struct ctl_table_header *subh = *(--subheader);
1468 struct ctl_table *table = subh->ctl_table_arg;
1469 unregister_sysctl_table(subh);
1478 * register_sysctl_table_path - register a sysctl table hierarchy
1479 * @path: The path to the directory the sysctl table is in.
1480 * @table: the top-level table structure
1482 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1483 * array. A completely 0 filled entry terminates the table.
1485 * See __register_sysctl_paths for more details.
1487 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1488 struct ctl_table *table)
1490 return __register_sysctl_paths(&sysctl_table_root.default_set,
1493 EXPORT_SYMBOL(register_sysctl_paths);
1496 * register_sysctl_table - register a sysctl table hierarchy
1497 * @table: the top-level table structure
1499 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1500 * array. A completely 0 filled entry terminates the table.
1502 * See register_sysctl_paths for more details.
1504 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1506 static const struct ctl_path null_path[] = { {} };
1508 return register_sysctl_paths(null_path, table);
1510 EXPORT_SYMBOL(register_sysctl_table);
1512 static void put_links(struct ctl_table_header *header)
1514 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1515 struct ctl_table_root *root = header->root;
1516 struct ctl_dir *parent = header->parent;
1517 struct ctl_dir *core_parent;
1518 struct ctl_table *entry;
1520 if (header->set == root_set)
1523 core_parent = xlate_dir(root_set, parent);
1524 if (IS_ERR(core_parent))
1527 for (entry = header->ctl_table; entry->procname; entry++) {
1528 struct ctl_table_header *link_head;
1529 struct ctl_table *link;
1530 const char *name = entry->procname;
1532 link = find_entry(&link_head, core_parent, name, strlen(name));
1534 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1535 (S_ISLNK(link->mode) && (link->data == root)))) {
1536 drop_sysctl_table(link_head);
1539 pr_err("sysctl link missing during unregister: ");
1540 sysctl_print_dir(parent);
1541 pr_cont("/%s\n", name);
1546 static void drop_sysctl_table(struct ctl_table_header *header)
1548 struct ctl_dir *parent = header->parent;
1555 start_unregistering(header);
1558 if (!--header->count)
1559 kfree_rcu(header, rcu);
1562 drop_sysctl_table(&parent->header);
1566 * unregister_sysctl_table - unregister a sysctl table hierarchy
1567 * @header: the header returned from register_sysctl_table
1569 * Unregisters the sysctl table and all children. proc entries may not
1570 * actually be removed until they are no longer used by anyone.
1572 void unregister_sysctl_table(struct ctl_table_header * header)
1580 nr_subheaders = count_subheaders(header->ctl_table_arg);
1581 if (unlikely(nr_subheaders > 1)) {
1582 struct ctl_table_header **subheaders;
1585 subheaders = (struct ctl_table_header **)(header + 1);
1586 for (i = nr_subheaders -1; i >= 0; i--) {
1587 struct ctl_table_header *subh = subheaders[i];
1588 struct ctl_table *table = subh->ctl_table_arg;
1589 unregister_sysctl_table(subh);
1596 spin_lock(&sysctl_lock);
1597 drop_sysctl_table(header);
1598 spin_unlock(&sysctl_lock);
1600 EXPORT_SYMBOL(unregister_sysctl_table);
1602 void setup_sysctl_set(struct ctl_table_set *set,
1603 struct ctl_table_root *root,
1604 int (*is_seen)(struct ctl_table_set *))
1606 memset(set, 0, sizeof(*set));
1607 set->is_seen = is_seen;
1608 init_header(&set->dir.header, root, set, NULL, root_table);
1611 void retire_sysctl_set(struct ctl_table_set *set)
1613 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1616 int __init proc_sys_init(void)
1618 struct proc_dir_entry *proc_sys_root;
1620 proc_sys_root = proc_mkdir("sys", NULL);
1621 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1622 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1623 proc_sys_root->nlink = 0;
1625 return sysctl_init();