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/cred.h>
12 #include <linux/namei.h>
14 #include <linux/module.h>
17 static const struct dentry_operations proc_sys_dentry_operations;
18 static const struct file_operations proc_sys_file_operations;
19 static const struct inode_operations proc_sys_inode_operations;
20 static const struct file_operations proc_sys_dir_file_operations;
21 static const struct inode_operations proc_sys_dir_operations;
23 /* Support for permanently empty directories */
25 struct ctl_table sysctl_mount_point[] = {
29 static bool is_empty_dir(struct ctl_table_header *head)
31 return head->ctl_table[0].child == sysctl_mount_point;
34 static void set_empty_dir(struct ctl_dir *dir)
36 dir->header.ctl_table[0].child = sysctl_mount_point;
39 static void clear_empty_dir(struct ctl_dir *dir)
42 dir->header.ctl_table[0].child = NULL;
45 void proc_sys_poll_notify(struct ctl_table_poll *poll)
50 atomic_inc(&poll->event);
51 wake_up_interruptible(&poll->wait);
54 static struct ctl_table root_table[] = {
57 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
61 static struct ctl_table_root sysctl_table_root = {
62 .default_set.dir.header = {
65 .ctl_table = root_table }},
66 .ctl_table_arg = root_table,
67 .root = &sysctl_table_root,
68 .set = &sysctl_table_root.default_set,
72 static DEFINE_SPINLOCK(sysctl_lock);
74 static void drop_sysctl_table(struct ctl_table_header *header);
75 static int sysctl_follow_link(struct ctl_table_header **phead,
76 struct ctl_table **pentry);
77 static int insert_links(struct ctl_table_header *head);
78 static void put_links(struct ctl_table_header *header);
80 static void sysctl_print_dir(struct ctl_dir *dir)
82 if (dir->header.parent)
83 sysctl_print_dir(dir->header.parent);
84 pr_cont("%s/", dir->header.ctl_table[0].procname);
87 static int namecmp(const char *name1, int len1, const char *name2, int len2)
96 cmp = memcmp(name1, name2, minlen);
102 /* Called under sysctl_lock */
103 static struct ctl_table *find_entry(struct ctl_table_header **phead,
104 struct ctl_dir *dir, const char *name, int namelen)
106 struct ctl_table_header *head;
107 struct ctl_table *entry;
108 struct rb_node *node = dir->root.rb_node;
112 struct ctl_node *ctl_node;
113 const char *procname;
116 ctl_node = rb_entry(node, struct ctl_node, node);
117 head = ctl_node->header;
118 entry = &head->ctl_table[ctl_node - head->node];
119 procname = entry->procname;
121 cmp = namecmp(name, namelen, procname, strlen(procname));
123 node = node->rb_left;
125 node = node->rb_right;
134 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
136 struct rb_node *node = &head->node[entry - head->ctl_table].node;
137 struct rb_node **p = &head->parent->root.rb_node;
138 struct rb_node *parent = NULL;
139 const char *name = entry->procname;
140 int namelen = strlen(name);
143 struct ctl_table_header *parent_head;
144 struct ctl_table *parent_entry;
145 struct ctl_node *parent_node;
146 const char *parent_name;
150 parent_node = rb_entry(parent, struct ctl_node, node);
151 parent_head = parent_node->header;
152 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
153 parent_name = parent_entry->procname;
155 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
161 pr_err("sysctl duplicate entry: ");
162 sysctl_print_dir(head->parent);
163 pr_cont("/%s\n", entry->procname);
168 rb_link_node(node, parent, p);
169 rb_insert_color(node, &head->parent->root);
173 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
175 struct rb_node *node = &head->node[entry - head->ctl_table].node;
177 rb_erase(node, &head->parent->root);
180 static void init_header(struct ctl_table_header *head,
181 struct ctl_table_root *root, struct ctl_table_set *set,
182 struct ctl_node *node, struct ctl_table *table)
184 head->ctl_table = table;
185 head->ctl_table_arg = table;
189 head->unregistering = NULL;
194 INIT_LIST_HEAD(&head->inodes);
196 struct ctl_table *entry;
197 for (entry = table; entry->procname; entry++, node++)
202 static void erase_header(struct ctl_table_header *head)
204 struct ctl_table *entry;
205 for (entry = head->ctl_table; entry->procname; entry++)
206 erase_entry(head, entry);
209 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
211 struct ctl_table *entry;
214 /* Is this a permanently empty directory? */
215 if (is_empty_dir(&dir->header))
218 /* Am I creating a permanently empty directory? */
219 if (header->ctl_table == sysctl_mount_point) {
220 if (!RB_EMPTY_ROOT(&dir->root))
226 header->parent = dir;
227 err = insert_links(header);
230 for (entry = header->ctl_table; entry->procname; entry++) {
231 err = insert_entry(header, entry);
237 erase_header(header);
240 if (header->ctl_table == sysctl_mount_point)
241 clear_empty_dir(dir);
242 header->parent = NULL;
243 drop_sysctl_table(&dir->header);
247 /* called under sysctl_lock */
248 static int use_table(struct ctl_table_header *p)
250 if (unlikely(p->unregistering))
256 /* called under sysctl_lock */
257 static void unuse_table(struct ctl_table_header *p)
260 if (unlikely(p->unregistering))
261 complete(p->unregistering);
264 /* called under sysctl_lock */
265 static void proc_sys_prune_dcache(struct ctl_table_header *head)
267 struct inode *inode, *prev = NULL;
268 struct proc_inode *ei;
271 list_for_each_entry_rcu(ei, &head->inodes, sysctl_inodes) {
272 inode = igrab(&ei->vfs_inode);
277 d_prune_aliases(inode);
285 /* called under sysctl_lock, will reacquire if has to wait */
286 static void start_unregistering(struct ctl_table_header *p)
289 * if p->used is 0, nobody will ever touch that entry again;
290 * we'll eliminate all paths to it before dropping sysctl_lock
292 if (unlikely(p->used)) {
293 struct completion wait;
294 init_completion(&wait);
295 p->unregistering = &wait;
296 spin_unlock(&sysctl_lock);
297 wait_for_completion(&wait);
299 /* anything non-NULL; we'll never dereference it */
300 p->unregistering = ERR_PTR(-EINVAL);
301 spin_unlock(&sysctl_lock);
304 * Prune dentries for unregistered sysctls: namespaced sysctls
305 * can have duplicate names and contaminate dcache very badly.
307 proc_sys_prune_dcache(p);
309 * do not remove from the list until nobody holds it; walking the
310 * list in do_sysctl() relies on that.
312 spin_lock(&sysctl_lock);
316 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
319 spin_lock(&sysctl_lock);
320 if (!use_table(head))
321 head = ERR_PTR(-ENOENT);
322 spin_unlock(&sysctl_lock);
326 static void sysctl_head_finish(struct ctl_table_header *head)
330 spin_lock(&sysctl_lock);
332 spin_unlock(&sysctl_lock);
335 static struct ctl_table_set *
336 lookup_header_set(struct ctl_table_root *root)
338 struct ctl_table_set *set = &root->default_set;
340 set = root->lookup(root);
344 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
346 const char *name, int namelen)
348 struct ctl_table_header *head;
349 struct ctl_table *entry;
351 spin_lock(&sysctl_lock);
352 entry = find_entry(&head, dir, name, namelen);
353 if (entry && use_table(head))
357 spin_unlock(&sysctl_lock);
361 static struct ctl_node *first_usable_entry(struct rb_node *node)
363 struct ctl_node *ctl_node;
365 for (;node; node = rb_next(node)) {
366 ctl_node = rb_entry(node, struct ctl_node, node);
367 if (use_table(ctl_node->header))
373 static void first_entry(struct ctl_dir *dir,
374 struct ctl_table_header **phead, struct ctl_table **pentry)
376 struct ctl_table_header *head = NULL;
377 struct ctl_table *entry = NULL;
378 struct ctl_node *ctl_node;
380 spin_lock(&sysctl_lock);
381 ctl_node = first_usable_entry(rb_first(&dir->root));
382 spin_unlock(&sysctl_lock);
384 head = ctl_node->header;
385 entry = &head->ctl_table[ctl_node - head->node];
391 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
393 struct ctl_table_header *head = *phead;
394 struct ctl_table *entry = *pentry;
395 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
397 spin_lock(&sysctl_lock);
400 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
401 spin_unlock(&sysctl_lock);
404 head = ctl_node->header;
405 entry = &head->ctl_table[ctl_node - head->node];
412 * sysctl_perm does NOT grant the superuser all rights automatically, because
413 * some sysctl variables are readonly even to root.
416 static int test_perm(int mode, int op)
418 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
420 else if (in_egroup_p(GLOBAL_ROOT_GID))
422 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
427 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
429 struct ctl_table_root *root = head->root;
432 if (root->permissions)
433 mode = root->permissions(head, table);
437 return test_perm(mode, op);
440 static struct inode *proc_sys_make_inode(struct super_block *sb,
441 struct ctl_table_header *head, struct ctl_table *table)
443 struct ctl_table_root *root = head->root;
445 struct proc_inode *ei;
447 inode = new_inode(sb);
451 inode->i_ino = get_next_ino();
455 spin_lock(&sysctl_lock);
456 if (unlikely(head->unregistering)) {
457 spin_unlock(&sysctl_lock);
463 ei->sysctl_entry = table;
464 list_add_rcu(&ei->sysctl_inodes, &head->inodes);
466 spin_unlock(&sysctl_lock);
468 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
469 inode->i_mode = table->mode;
470 if (!S_ISDIR(table->mode)) {
471 inode->i_mode |= S_IFREG;
472 inode->i_op = &proc_sys_inode_operations;
473 inode->i_fop = &proc_sys_file_operations;
475 inode->i_mode |= S_IFDIR;
476 inode->i_op = &proc_sys_dir_operations;
477 inode->i_fop = &proc_sys_dir_file_operations;
478 if (is_empty_dir(head))
479 make_empty_dir_inode(inode);
482 if (root->set_ownership)
483 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
489 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
491 spin_lock(&sysctl_lock);
492 list_del_rcu(&PROC_I(inode)->sysctl_inodes);
494 kfree_rcu(head, rcu);
495 spin_unlock(&sysctl_lock);
498 static struct ctl_table_header *grab_header(struct inode *inode)
500 struct ctl_table_header *head = PROC_I(inode)->sysctl;
502 head = &sysctl_table_root.default_set.dir.header;
503 return sysctl_head_grab(head);
506 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
509 struct ctl_table_header *head = grab_header(dir);
510 struct ctl_table_header *h = NULL;
511 const struct qstr *name = &dentry->d_name;
514 struct dentry *err = ERR_PTR(-ENOENT);
515 struct ctl_dir *ctl_dir;
519 return ERR_CAST(head);
521 ctl_dir = container_of(head, struct ctl_dir, header);
523 p = lookup_entry(&h, ctl_dir, name->name, name->len);
527 if (S_ISLNK(p->mode)) {
528 ret = sysctl_follow_link(&h, &p);
534 err = ERR_PTR(-ENOMEM);
535 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
540 d_set_d_op(dentry, &proc_sys_dentry_operations);
541 d_add(dentry, inode);
545 sysctl_head_finish(h);
546 sysctl_head_finish(head);
550 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
551 size_t count, loff_t *ppos, int write)
553 struct inode *inode = file_inode(filp);
554 struct ctl_table_header *head = grab_header(inode);
555 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
560 return PTR_ERR(head);
563 * At this point we know that the sysctl was not unregistered
564 * and won't be until we finish.
567 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
570 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
572 if (!table->proc_handler)
575 /* careful: calling conventions are nasty here */
577 error = table->proc_handler(table, write, buf, &res, ppos);
581 sysctl_head_finish(head);
586 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
587 size_t count, loff_t *ppos)
589 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
592 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
593 size_t count, loff_t *ppos)
595 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
598 static int proc_sys_open(struct inode *inode, struct file *filp)
600 struct ctl_table_header *head = grab_header(inode);
601 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
603 /* sysctl was unregistered */
605 return PTR_ERR(head);
608 filp->private_data = proc_sys_poll_event(table->poll);
610 sysctl_head_finish(head);
615 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
617 struct inode *inode = file_inode(filp);
618 struct ctl_table_header *head = grab_header(inode);
619 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
620 unsigned int ret = DEFAULT_POLLMASK;
623 /* sysctl was unregistered */
625 return POLLERR | POLLHUP;
627 if (!table->proc_handler)
633 event = (unsigned long)filp->private_data;
634 poll_wait(filp, &table->poll->wait, wait);
636 if (event != atomic_read(&table->poll->event)) {
637 filp->private_data = proc_sys_poll_event(table->poll);
638 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
642 sysctl_head_finish(head);
647 static bool proc_sys_fill_cache(struct file *file,
648 struct dir_context *ctx,
649 struct ctl_table_header *head,
650 struct ctl_table *table)
652 struct dentry *child, *dir = file->f_path.dentry;
656 unsigned type = DT_UNKNOWN;
658 qname.name = table->procname;
659 qname.len = strlen(table->procname);
660 qname.hash = full_name_hash(dir, qname.name, qname.len);
662 child = d_lookup(dir, &qname);
664 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
665 child = d_alloc_parallel(dir, &qname, &wq);
668 if (d_in_lookup(child)) {
669 inode = proc_sys_make_inode(dir->d_sb, head, table);
671 d_lookup_done(child);
675 d_set_d_op(child, &proc_sys_dentry_operations);
679 inode = d_inode(child);
681 type = inode->i_mode >> 12;
683 return dir_emit(ctx, qname.name, qname.len, ino, type);
686 static bool proc_sys_link_fill_cache(struct file *file,
687 struct dir_context *ctx,
688 struct ctl_table_header *head,
689 struct ctl_table *table)
692 head = sysctl_head_grab(head);
694 if (S_ISLNK(table->mode)) {
695 /* It is not an error if we can not follow the link ignore it */
696 int err = sysctl_follow_link(&head, &table);
701 ret = proc_sys_fill_cache(file, ctx, head, table);
703 sysctl_head_finish(head);
707 static int scan(struct ctl_table_header *head, struct ctl_table *table,
708 unsigned long *pos, struct file *file,
709 struct dir_context *ctx)
713 if ((*pos)++ < ctx->pos)
716 if (unlikely(S_ISLNK(table->mode)))
717 res = proc_sys_link_fill_cache(file, ctx, head, table);
719 res = proc_sys_fill_cache(file, ctx, head, table);
727 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
729 struct ctl_table_header *head = grab_header(file_inode(file));
730 struct ctl_table_header *h = NULL;
731 struct ctl_table *entry;
732 struct ctl_dir *ctl_dir;
736 return PTR_ERR(head);
738 ctl_dir = container_of(head, struct ctl_dir, header);
740 if (!dir_emit_dots(file, ctx))
745 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
746 if (!scan(h, entry, &pos, file, ctx)) {
747 sysctl_head_finish(h);
752 sysctl_head_finish(head);
756 static int proc_sys_permission(struct inode *inode, int mask)
759 * sysctl entries that are not writeable,
760 * are _NOT_ writeable, capabilities or not.
762 struct ctl_table_header *head;
763 struct ctl_table *table;
766 /* Executable files are not allowed under /proc/sys/ */
767 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
770 head = grab_header(inode);
772 return PTR_ERR(head);
774 table = PROC_I(inode)->sysctl_entry;
775 if (!table) /* global root - r-xr-xr-x */
776 error = mask & MAY_WRITE ? -EACCES : 0;
777 else /* Use the permissions on the sysctl table entry */
778 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
780 sysctl_head_finish(head);
784 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
786 struct inode *inode = d_inode(dentry);
789 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
792 error = setattr_prepare(dentry, attr);
796 setattr_copy(inode, attr);
797 mark_inode_dirty(inode);
801 static int proc_sys_getattr(const struct path *path, struct kstat *stat,
802 u32 request_mask, unsigned int query_flags)
804 struct inode *inode = d_inode(path->dentry);
805 struct ctl_table_header *head = grab_header(inode);
806 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
809 return PTR_ERR(head);
811 generic_fillattr(inode, stat);
813 stat->mode = (stat->mode & S_IFMT) | table->mode;
815 sysctl_head_finish(head);
819 static const struct file_operations proc_sys_file_operations = {
820 .open = proc_sys_open,
821 .poll = proc_sys_poll,
822 .read = proc_sys_read,
823 .write = proc_sys_write,
824 .llseek = default_llseek,
827 static const struct file_operations proc_sys_dir_file_operations = {
828 .read = generic_read_dir,
829 .iterate_shared = proc_sys_readdir,
830 .llseek = generic_file_llseek,
833 static const struct inode_operations proc_sys_inode_operations = {
834 .permission = proc_sys_permission,
835 .setattr = proc_sys_setattr,
836 .getattr = proc_sys_getattr,
839 static const struct inode_operations proc_sys_dir_operations = {
840 .lookup = proc_sys_lookup,
841 .permission = proc_sys_permission,
842 .setattr = proc_sys_setattr,
843 .getattr = proc_sys_getattr,
846 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
848 if (flags & LOOKUP_RCU)
850 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
853 static int proc_sys_delete(const struct dentry *dentry)
855 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
858 static int sysctl_is_seen(struct ctl_table_header *p)
860 struct ctl_table_set *set = p->set;
862 spin_lock(&sysctl_lock);
863 if (p->unregistering)
865 else if (!set->is_seen)
868 res = set->is_seen(set);
869 spin_unlock(&sysctl_lock);
873 static int proc_sys_compare(const struct dentry *dentry,
874 unsigned int len, const char *str, const struct qstr *name)
876 struct ctl_table_header *head;
879 /* Although proc doesn't have negative dentries, rcu-walk means
880 * that inode here can be NULL */
881 /* AV: can it, indeed? */
882 inode = d_inode_rcu(dentry);
885 if (name->len != len)
887 if (memcmp(name->name, str, len))
889 head = rcu_dereference(PROC_I(inode)->sysctl);
890 return !head || !sysctl_is_seen(head);
893 static const struct dentry_operations proc_sys_dentry_operations = {
894 .d_revalidate = proc_sys_revalidate,
895 .d_delete = proc_sys_delete,
896 .d_compare = proc_sys_compare,
899 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
900 const char *name, int namelen)
902 struct ctl_table_header *head;
903 struct ctl_table *entry;
905 entry = find_entry(&head, dir, name, namelen);
907 return ERR_PTR(-ENOENT);
908 if (!S_ISDIR(entry->mode))
909 return ERR_PTR(-ENOTDIR);
910 return container_of(head, struct ctl_dir, header);
913 static struct ctl_dir *new_dir(struct ctl_table_set *set,
914 const char *name, int namelen)
916 struct ctl_table *table;
918 struct ctl_node *node;
921 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
922 sizeof(struct ctl_table)*2 + namelen + 1,
927 node = (struct ctl_node *)(new + 1);
928 table = (struct ctl_table *)(node + 1);
929 new_name = (char *)(table + 2);
930 memcpy(new_name, name, namelen);
931 new_name[namelen] = '\0';
932 table[0].procname = new_name;
933 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
934 init_header(&new->header, set->dir.header.root, set, node, table);
940 * get_subdir - find or create a subdir with the specified name.
941 * @dir: Directory to create the subdirectory in
942 * @name: The name of the subdirectory to find or create
943 * @namelen: The length of name
945 * Takes a directory with an elevated reference count so we know that
946 * if we drop the lock the directory will not go away. Upon success
947 * the reference is moved from @dir to the returned subdirectory.
948 * Upon error an error code is returned and the reference on @dir is
951 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
952 const char *name, int namelen)
954 struct ctl_table_set *set = dir->header.set;
955 struct ctl_dir *subdir, *new = NULL;
958 spin_lock(&sysctl_lock);
959 subdir = find_subdir(dir, name, namelen);
962 if (PTR_ERR(subdir) != -ENOENT)
965 spin_unlock(&sysctl_lock);
966 new = new_dir(set, name, namelen);
967 spin_lock(&sysctl_lock);
968 subdir = ERR_PTR(-ENOMEM);
972 /* Was the subdir added while we dropped the lock? */
973 subdir = find_subdir(dir, name, namelen);
976 if (PTR_ERR(subdir) != -ENOENT)
979 /* Nope. Use the our freshly made directory entry. */
980 err = insert_header(dir, &new->header);
981 subdir = ERR_PTR(err);
986 subdir->header.nreg++;
988 if (IS_ERR(subdir)) {
989 pr_err("sysctl could not get directory: ");
990 sysctl_print_dir(dir);
991 pr_cont("/%*.*s %ld\n",
992 namelen, namelen, name, PTR_ERR(subdir));
994 drop_sysctl_table(&dir->header);
996 drop_sysctl_table(&new->header);
997 spin_unlock(&sysctl_lock);
1001 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1003 struct ctl_dir *parent;
1004 const char *procname;
1005 if (!dir->header.parent)
1007 parent = xlate_dir(set, dir->header.parent);
1010 procname = dir->header.ctl_table[0].procname;
1011 return find_subdir(parent, procname, strlen(procname));
1014 static int sysctl_follow_link(struct ctl_table_header **phead,
1015 struct ctl_table **pentry)
1017 struct ctl_table_header *head;
1018 struct ctl_table_root *root;
1019 struct ctl_table_set *set;
1020 struct ctl_table *entry;
1021 struct ctl_dir *dir;
1025 spin_lock(&sysctl_lock);
1026 root = (*pentry)->data;
1027 set = lookup_header_set(root);
1028 dir = xlate_dir(set, (*phead)->parent);
1032 const char *procname = (*pentry)->procname;
1034 entry = find_entry(&head, dir, procname, strlen(procname));
1036 if (entry && use_table(head)) {
1037 unuse_table(*phead);
1044 spin_unlock(&sysctl_lock);
1048 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1050 struct va_format vaf;
1053 va_start(args, fmt);
1057 pr_err("sysctl table check failed: %s/%s %pV\n",
1058 path, table->procname, &vaf);
1064 static int sysctl_check_table(const char *path, struct ctl_table *table)
1067 for (; table->procname; table++) {
1069 err = sysctl_err(path, table, "Not a file");
1071 if ((table->proc_handler == proc_dostring) ||
1072 (table->proc_handler == proc_dointvec) ||
1073 (table->proc_handler == proc_douintvec) ||
1074 (table->proc_handler == proc_dointvec_minmax) ||
1075 (table->proc_handler == proc_dointvec_jiffies) ||
1076 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1077 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1078 (table->proc_handler == proc_doulongvec_minmax) ||
1079 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1081 err = sysctl_err(path, table, "No data");
1083 err = sysctl_err(path, table, "No maxlen");
1085 if (!table->proc_handler)
1086 err = sysctl_err(path, table, "No proc_handler");
1088 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1089 err = sysctl_err(path, table, "bogus .mode 0%o",
1095 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1096 struct ctl_table_root *link_root)
1098 struct ctl_table *link_table, *entry, *link;
1099 struct ctl_table_header *links;
1100 struct ctl_node *node;
1102 int nr_entries, name_bytes;
1106 for (entry = table; entry->procname; entry++) {
1108 name_bytes += strlen(entry->procname) + 1;
1111 links = kzalloc(sizeof(struct ctl_table_header) +
1112 sizeof(struct ctl_node)*nr_entries +
1113 sizeof(struct ctl_table)*(nr_entries + 1) +
1120 node = (struct ctl_node *)(links + 1);
1121 link_table = (struct ctl_table *)(node + nr_entries);
1122 link_name = (char *)&link_table[nr_entries + 1];
1124 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1125 int len = strlen(entry->procname) + 1;
1126 memcpy(link_name, entry->procname, len);
1127 link->procname = link_name;
1128 link->mode = S_IFLNK|S_IRWXUGO;
1129 link->data = link_root;
1132 init_header(links, dir->header.root, dir->header.set, node, link_table);
1133 links->nreg = nr_entries;
1138 static bool get_links(struct ctl_dir *dir,
1139 struct ctl_table *table, struct ctl_table_root *link_root)
1141 struct ctl_table_header *head;
1142 struct ctl_table *entry, *link;
1144 /* Are there links available for every entry in table? */
1145 for (entry = table; entry->procname; entry++) {
1146 const char *procname = entry->procname;
1147 link = find_entry(&head, dir, procname, strlen(procname));
1150 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1152 if (S_ISLNK(link->mode) && (link->data == link_root))
1157 /* The checks passed. Increase the registration count on the links */
1158 for (entry = table; entry->procname; entry++) {
1159 const char *procname = entry->procname;
1160 link = find_entry(&head, dir, procname, strlen(procname));
1166 static int insert_links(struct ctl_table_header *head)
1168 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1169 struct ctl_dir *core_parent = NULL;
1170 struct ctl_table_header *links;
1173 if (head->set == root_set)
1176 core_parent = xlate_dir(root_set, head->parent);
1177 if (IS_ERR(core_parent))
1180 if (get_links(core_parent, head->ctl_table, head->root))
1183 core_parent->header.nreg++;
1184 spin_unlock(&sysctl_lock);
1186 links = new_links(core_parent, head->ctl_table, head->root);
1188 spin_lock(&sysctl_lock);
1194 if (get_links(core_parent, head->ctl_table, head->root)) {
1199 err = insert_header(core_parent, links);
1203 drop_sysctl_table(&core_parent->header);
1208 * __register_sysctl_table - register a leaf sysctl table
1209 * @set: Sysctl tree to register on
1210 * @path: The path to the directory the sysctl table is in.
1211 * @table: the top-level table structure
1213 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1214 * array. A completely 0 filled entry terminates the table.
1216 * The members of the &struct ctl_table structure are used as follows:
1218 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1219 * enter a sysctl file
1221 * data - a pointer to data for use by proc_handler
1223 * maxlen - the maximum size in bytes of the data
1225 * mode - the file permissions for the /proc/sys file
1227 * child - must be %NULL.
1229 * proc_handler - the text handler routine (described below)
1231 * extra1, extra2 - extra pointers usable by the proc handler routines
1233 * Leaf nodes in the sysctl tree will be represented by a single file
1234 * under /proc; non-leaf nodes will be represented by directories.
1236 * There must be a proc_handler routine for any terminal nodes.
1237 * Several default handlers are available to cover common cases -
1239 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1240 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1241 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1243 * It is the handler's job to read the input buffer from user memory
1244 * and process it. The handler should return 0 on success.
1246 * This routine returns %NULL on a failure to register, and a pointer
1247 * to the table header on success.
1249 struct ctl_table_header *__register_sysctl_table(
1250 struct ctl_table_set *set,
1251 const char *path, struct ctl_table *table)
1253 struct ctl_table_root *root = set->dir.header.root;
1254 struct ctl_table_header *header;
1255 const char *name, *nextname;
1256 struct ctl_dir *dir;
1257 struct ctl_table *entry;
1258 struct ctl_node *node;
1261 for (entry = table; entry->procname; entry++)
1264 header = kzalloc(sizeof(struct ctl_table_header) +
1265 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1269 node = (struct ctl_node *)(header + 1);
1270 init_header(header, root, set, node, table);
1271 if (sysctl_check_table(path, table))
1274 spin_lock(&sysctl_lock);
1276 /* Reference moved down the diretory tree get_subdir */
1278 spin_unlock(&sysctl_lock);
1280 /* Find the directory for the ctl_table */
1281 for (name = path; name; name = nextname) {
1283 nextname = strchr(name, '/');
1285 namelen = nextname - name;
1288 namelen = strlen(name);
1293 dir = get_subdir(dir, name, namelen);
1298 spin_lock(&sysctl_lock);
1299 if (insert_header(dir, header))
1300 goto fail_put_dir_locked;
1302 drop_sysctl_table(&dir->header);
1303 spin_unlock(&sysctl_lock);
1307 fail_put_dir_locked:
1308 drop_sysctl_table(&dir->header);
1309 spin_unlock(&sysctl_lock);
1317 * register_sysctl - register a sysctl table
1318 * @path: The path to the directory the sysctl table is in.
1319 * @table: the table structure
1321 * Register a sysctl table. @table should be a filled in ctl_table
1322 * array. A completely 0 filled entry terminates the table.
1324 * See __register_sysctl_table for more details.
1326 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1328 return __register_sysctl_table(&sysctl_table_root.default_set,
1331 EXPORT_SYMBOL(register_sysctl);
1333 static char *append_path(const char *path, char *pos, const char *name)
1336 namelen = strlen(name);
1337 if (((pos - path) + namelen + 2) >= PATH_MAX)
1339 memcpy(pos, name, namelen);
1341 pos[namelen + 1] = '\0';
1346 static int count_subheaders(struct ctl_table *table)
1349 int nr_subheaders = 0;
1350 struct ctl_table *entry;
1352 /* special case: no directory and empty directory */
1353 if (!table || !table->procname)
1356 for (entry = table; entry->procname; entry++) {
1358 nr_subheaders += count_subheaders(entry->child);
1362 return nr_subheaders + has_files;
1365 static int register_leaf_sysctl_tables(const char *path, char *pos,
1366 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1367 struct ctl_table *table)
1369 struct ctl_table *ctl_table_arg = NULL;
1370 struct ctl_table *entry, *files;
1375 for (entry = table; entry->procname; entry++) {
1383 /* If there are mixed files and directories we need a new table */
1384 if (nr_dirs && nr_files) {
1385 struct ctl_table *new;
1386 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1391 ctl_table_arg = files;
1392 for (new = files, entry = table; entry->procname; entry++) {
1400 /* Register everything except a directory full of subdirectories */
1401 if (nr_files || !nr_dirs) {
1402 struct ctl_table_header *header;
1403 header = __register_sysctl_table(set, path, files);
1405 kfree(ctl_table_arg);
1409 /* Remember if we need to free the file table */
1410 header->ctl_table_arg = ctl_table_arg;
1411 **subheader = header;
1415 /* Recurse into the subdirectories. */
1416 for (entry = table; entry->procname; entry++) {
1422 err = -ENAMETOOLONG;
1423 child_pos = append_path(path, pos, entry->procname);
1427 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1435 /* On failure our caller will unregister all registered subheaders */
1440 * __register_sysctl_paths - register a sysctl table hierarchy
1441 * @set: Sysctl tree to register on
1442 * @path: The path to the directory the sysctl table is in.
1443 * @table: the top-level table structure
1445 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1446 * array. A completely 0 filled entry terminates the table.
1448 * See __register_sysctl_table for more details.
1450 struct ctl_table_header *__register_sysctl_paths(
1451 struct ctl_table_set *set,
1452 const struct ctl_path *path, struct ctl_table *table)
1454 struct ctl_table *ctl_table_arg = table;
1455 int nr_subheaders = count_subheaders(table);
1456 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1457 const struct ctl_path *component;
1458 char *new_path, *pos;
1460 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1465 for (component = path; component->procname; component++) {
1466 pos = append_path(new_path, pos, component->procname);
1470 while (table->procname && table->child && !table[1].procname) {
1471 pos = append_path(new_path, pos, table->procname);
1474 table = table->child;
1476 if (nr_subheaders == 1) {
1477 header = __register_sysctl_table(set, new_path, table);
1479 header->ctl_table_arg = ctl_table_arg;
1481 header = kzalloc(sizeof(*header) +
1482 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1486 subheaders = (struct ctl_table_header **) (header + 1);
1487 subheader = subheaders;
1488 header->ctl_table_arg = ctl_table_arg;
1490 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1492 goto err_register_leaves;
1499 err_register_leaves:
1500 while (subheader > subheaders) {
1501 struct ctl_table_header *subh = *(--subheader);
1502 struct ctl_table *table = subh->ctl_table_arg;
1503 unregister_sysctl_table(subh);
1512 * register_sysctl_table_path - register a sysctl table hierarchy
1513 * @path: The path to the directory the sysctl table is in.
1514 * @table: the top-level table structure
1516 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1517 * array. A completely 0 filled entry terminates the table.
1519 * See __register_sysctl_paths for more details.
1521 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1522 struct ctl_table *table)
1524 return __register_sysctl_paths(&sysctl_table_root.default_set,
1527 EXPORT_SYMBOL(register_sysctl_paths);
1530 * register_sysctl_table - register a sysctl table hierarchy
1531 * @table: the top-level table structure
1533 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1534 * array. A completely 0 filled entry terminates the table.
1536 * See register_sysctl_paths for more details.
1538 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1540 static const struct ctl_path null_path[] = { {} };
1542 return register_sysctl_paths(null_path, table);
1544 EXPORT_SYMBOL(register_sysctl_table);
1546 static void put_links(struct ctl_table_header *header)
1548 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1549 struct ctl_table_root *root = header->root;
1550 struct ctl_dir *parent = header->parent;
1551 struct ctl_dir *core_parent;
1552 struct ctl_table *entry;
1554 if (header->set == root_set)
1557 core_parent = xlate_dir(root_set, parent);
1558 if (IS_ERR(core_parent))
1561 for (entry = header->ctl_table; entry->procname; entry++) {
1562 struct ctl_table_header *link_head;
1563 struct ctl_table *link;
1564 const char *name = entry->procname;
1566 link = find_entry(&link_head, core_parent, name, strlen(name));
1568 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1569 (S_ISLNK(link->mode) && (link->data == root)))) {
1570 drop_sysctl_table(link_head);
1573 pr_err("sysctl link missing during unregister: ");
1574 sysctl_print_dir(parent);
1575 pr_cont("/%s\n", name);
1580 static void drop_sysctl_table(struct ctl_table_header *header)
1582 struct ctl_dir *parent = header->parent;
1588 start_unregistering(header);
1589 if (!--header->count)
1590 kfree_rcu(header, rcu);
1593 drop_sysctl_table(&parent->header);
1597 * unregister_sysctl_table - unregister a sysctl table hierarchy
1598 * @header: the header returned from register_sysctl_table
1600 * Unregisters the sysctl table and all children. proc entries may not
1601 * actually be removed until they are no longer used by anyone.
1603 void unregister_sysctl_table(struct ctl_table_header * header)
1611 nr_subheaders = count_subheaders(header->ctl_table_arg);
1612 if (unlikely(nr_subheaders > 1)) {
1613 struct ctl_table_header **subheaders;
1616 subheaders = (struct ctl_table_header **)(header + 1);
1617 for (i = nr_subheaders -1; i >= 0; i--) {
1618 struct ctl_table_header *subh = subheaders[i];
1619 struct ctl_table *table = subh->ctl_table_arg;
1620 unregister_sysctl_table(subh);
1627 spin_lock(&sysctl_lock);
1628 drop_sysctl_table(header);
1629 spin_unlock(&sysctl_lock);
1631 EXPORT_SYMBOL(unregister_sysctl_table);
1633 void setup_sysctl_set(struct ctl_table_set *set,
1634 struct ctl_table_root *root,
1635 int (*is_seen)(struct ctl_table_set *))
1637 memset(set, 0, sizeof(*set));
1638 set->is_seen = is_seen;
1639 init_header(&set->dir.header, root, set, NULL, root_table);
1642 void retire_sysctl_set(struct ctl_table_set *set)
1644 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1647 int __init proc_sys_init(void)
1649 struct proc_dir_entry *proc_sys_root;
1651 proc_sys_root = proc_mkdir("sys", NULL);
1652 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1653 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1654 proc_sys_root->nlink = 0;
1656 return sysctl_init();