1 // SPDX-License-Identifier: GPL-2.0
5 #include <linux/init.h>
6 #include <linux/sysctl.h>
7 #include <linux/poll.h>
8 #include <linux/proc_fs.h>
9 #include <linux/printk.h>
10 #include <linux/security.h>
11 #include <linux/sched.h>
12 #include <linux/cred.h>
13 #include <linux/namei.h>
15 #include <linux/module.h>
18 static const struct dentry_operations proc_sys_dentry_operations;
19 static const struct file_operations proc_sys_file_operations;
20 static const struct inode_operations proc_sys_inode_operations;
21 static const struct file_operations proc_sys_dir_file_operations;
22 static const struct inode_operations proc_sys_dir_operations;
24 /* Support for permanently empty directories */
26 struct ctl_table sysctl_mount_point[] = {
30 static bool is_empty_dir(struct ctl_table_header *head)
32 return head->ctl_table[0].child == sysctl_mount_point;
35 static void set_empty_dir(struct ctl_dir *dir)
37 dir->header.ctl_table[0].child = sysctl_mount_point;
40 static void clear_empty_dir(struct ctl_dir *dir)
43 dir->header.ctl_table[0].child = NULL;
46 void proc_sys_poll_notify(struct ctl_table_poll *poll)
51 atomic_inc(&poll->event);
52 wake_up_interruptible(&poll->wait);
55 static struct ctl_table root_table[] = {
58 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
62 static struct ctl_table_root sysctl_table_root = {
63 .default_set.dir.header = {
66 .ctl_table = root_table }},
67 .ctl_table_arg = root_table,
68 .root = &sysctl_table_root,
69 .set = &sysctl_table_root.default_set,
73 static DEFINE_SPINLOCK(sysctl_lock);
75 static void drop_sysctl_table(struct ctl_table_header *header);
76 static int sysctl_follow_link(struct ctl_table_header **phead,
77 struct ctl_table **pentry);
78 static int insert_links(struct ctl_table_header *head);
79 static void put_links(struct ctl_table_header *header);
81 static void sysctl_print_dir(struct ctl_dir *dir)
83 if (dir->header.parent)
84 sysctl_print_dir(dir->header.parent);
85 pr_cont("%s/", dir->header.ctl_table[0].procname);
88 static int namecmp(const char *name1, int len1, const char *name2, int len2)
97 cmp = memcmp(name1, name2, minlen);
103 /* Called under sysctl_lock */
104 static struct ctl_table *find_entry(struct ctl_table_header **phead,
105 struct ctl_dir *dir, const char *name, int namelen)
107 struct ctl_table_header *head;
108 struct ctl_table *entry;
109 struct rb_node *node = dir->root.rb_node;
113 struct ctl_node *ctl_node;
114 const char *procname;
117 ctl_node = rb_entry(node, struct ctl_node, node);
118 head = ctl_node->header;
119 entry = &head->ctl_table[ctl_node - head->node];
120 procname = entry->procname;
122 cmp = namecmp(name, namelen, procname, strlen(procname));
124 node = node->rb_left;
126 node = node->rb_right;
135 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
137 struct rb_node *node = &head->node[entry - head->ctl_table].node;
138 struct rb_node **p = &head->parent->root.rb_node;
139 struct rb_node *parent = NULL;
140 const char *name = entry->procname;
141 int namelen = strlen(name);
144 struct ctl_table_header *parent_head;
145 struct ctl_table *parent_entry;
146 struct ctl_node *parent_node;
147 const char *parent_name;
151 parent_node = rb_entry(parent, struct ctl_node, node);
152 parent_head = parent_node->header;
153 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
154 parent_name = parent_entry->procname;
156 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
162 pr_err("sysctl duplicate entry: ");
163 sysctl_print_dir(head->parent);
164 pr_cont("/%s\n", entry->procname);
169 rb_link_node(node, parent, p);
170 rb_insert_color(node, &head->parent->root);
174 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
176 struct rb_node *node = &head->node[entry - head->ctl_table].node;
178 rb_erase(node, &head->parent->root);
181 static void init_header(struct ctl_table_header *head,
182 struct ctl_table_root *root, struct ctl_table_set *set,
183 struct ctl_node *node, struct ctl_table *table)
185 head->ctl_table = table;
186 head->ctl_table_arg = table;
190 head->unregistering = NULL;
195 INIT_HLIST_HEAD(&head->inodes);
197 struct ctl_table *entry;
198 for (entry = table; entry->procname; entry++, node++)
203 static void erase_header(struct ctl_table_header *head)
205 struct ctl_table *entry;
206 for (entry = head->ctl_table; entry->procname; entry++)
207 erase_entry(head, entry);
210 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
212 struct ctl_table *entry;
215 /* Is this a permanently empty directory? */
216 if (is_empty_dir(&dir->header))
219 /* Am I creating a permanently empty directory? */
220 if (header->ctl_table == sysctl_mount_point) {
221 if (!RB_EMPTY_ROOT(&dir->root))
227 header->parent = dir;
228 err = insert_links(header);
231 for (entry = header->ctl_table; entry->procname; entry++) {
232 err = insert_entry(header, entry);
238 erase_header(header);
241 if (header->ctl_table == sysctl_mount_point)
242 clear_empty_dir(dir);
243 header->parent = NULL;
244 drop_sysctl_table(&dir->header);
248 /* called under sysctl_lock */
249 static int use_table(struct ctl_table_header *p)
251 if (unlikely(p->unregistering))
257 /* called under sysctl_lock */
258 static void unuse_table(struct ctl_table_header *p)
261 if (unlikely(p->unregistering))
262 complete(p->unregistering);
265 static void proc_sys_prune_dcache(struct ctl_table_header *head)
268 struct proc_inode *ei;
269 struct hlist_node *node;
270 struct super_block *sb;
274 node = hlist_first_rcu(&head->inodes);
277 ei = hlist_entry(node, struct proc_inode, sysctl_inodes);
278 spin_lock(&sysctl_lock);
279 hlist_del_init_rcu(&ei->sysctl_inodes);
280 spin_unlock(&sysctl_lock);
282 inode = &ei->vfs_inode;
284 if (!atomic_inc_not_zero(&sb->s_active))
286 inode = igrab(inode);
288 if (unlikely(!inode)) {
289 deactivate_super(sb);
294 d_prune_aliases(inode);
296 deactivate_super(sb);
303 /* called under sysctl_lock, will reacquire if has to wait */
304 static void start_unregistering(struct ctl_table_header *p)
307 * if p->used is 0, nobody will ever touch that entry again;
308 * we'll eliminate all paths to it before dropping sysctl_lock
310 if (unlikely(p->used)) {
311 struct completion wait;
312 init_completion(&wait);
313 p->unregistering = &wait;
314 spin_unlock(&sysctl_lock);
315 wait_for_completion(&wait);
317 /* anything non-NULL; we'll never dereference it */
318 p->unregistering = ERR_PTR(-EINVAL);
319 spin_unlock(&sysctl_lock);
322 * Prune dentries for unregistered sysctls: namespaced sysctls
323 * can have duplicate names and contaminate dcache very badly.
325 proc_sys_prune_dcache(p);
327 * do not remove from the list until nobody holds it; walking the
328 * list in do_sysctl() relies on that.
330 spin_lock(&sysctl_lock);
334 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
337 spin_lock(&sysctl_lock);
338 if (!use_table(head))
339 head = ERR_PTR(-ENOENT);
340 spin_unlock(&sysctl_lock);
344 static void sysctl_head_finish(struct ctl_table_header *head)
348 spin_lock(&sysctl_lock);
350 spin_unlock(&sysctl_lock);
353 static struct ctl_table_set *
354 lookup_header_set(struct ctl_table_root *root)
356 struct ctl_table_set *set = &root->default_set;
358 set = root->lookup(root);
362 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
364 const char *name, int namelen)
366 struct ctl_table_header *head;
367 struct ctl_table *entry;
369 spin_lock(&sysctl_lock);
370 entry = find_entry(&head, dir, name, namelen);
371 if (entry && use_table(head))
375 spin_unlock(&sysctl_lock);
379 static struct ctl_node *first_usable_entry(struct rb_node *node)
381 struct ctl_node *ctl_node;
383 for (;node; node = rb_next(node)) {
384 ctl_node = rb_entry(node, struct ctl_node, node);
385 if (use_table(ctl_node->header))
391 static void first_entry(struct ctl_dir *dir,
392 struct ctl_table_header **phead, struct ctl_table **pentry)
394 struct ctl_table_header *head = NULL;
395 struct ctl_table *entry = NULL;
396 struct ctl_node *ctl_node;
398 spin_lock(&sysctl_lock);
399 ctl_node = first_usable_entry(rb_first(&dir->root));
400 spin_unlock(&sysctl_lock);
402 head = ctl_node->header;
403 entry = &head->ctl_table[ctl_node - head->node];
409 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
411 struct ctl_table_header *head = *phead;
412 struct ctl_table *entry = *pentry;
413 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
415 spin_lock(&sysctl_lock);
418 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
419 spin_unlock(&sysctl_lock);
422 head = ctl_node->header;
423 entry = &head->ctl_table[ctl_node - head->node];
430 * sysctl_perm does NOT grant the superuser all rights automatically, because
431 * some sysctl variables are readonly even to root.
434 static int test_perm(int mode, int op)
436 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
438 else if (in_egroup_p(GLOBAL_ROOT_GID))
440 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
445 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
447 struct ctl_table_root *root = head->root;
450 if (root->permissions)
451 mode = root->permissions(head, table);
455 return test_perm(mode, op);
458 static struct inode *proc_sys_make_inode(struct super_block *sb,
459 struct ctl_table_header *head, struct ctl_table *table)
461 struct ctl_table_root *root = head->root;
463 struct proc_inode *ei;
465 inode = new_inode(sb);
469 inode->i_ino = get_next_ino();
473 spin_lock(&sysctl_lock);
474 if (unlikely(head->unregistering)) {
475 spin_unlock(&sysctl_lock);
481 ei->sysctl_entry = table;
482 hlist_add_head_rcu(&ei->sysctl_inodes, &head->inodes);
484 spin_unlock(&sysctl_lock);
486 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
487 inode->i_mode = table->mode;
488 if (!S_ISDIR(table->mode)) {
489 inode->i_mode |= S_IFREG;
490 inode->i_op = &proc_sys_inode_operations;
491 inode->i_fop = &proc_sys_file_operations;
493 inode->i_mode |= S_IFDIR;
494 inode->i_op = &proc_sys_dir_operations;
495 inode->i_fop = &proc_sys_dir_file_operations;
496 if (is_empty_dir(head))
497 make_empty_dir_inode(inode);
500 if (root->set_ownership)
501 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
507 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
509 spin_lock(&sysctl_lock);
510 hlist_del_init_rcu(&PROC_I(inode)->sysctl_inodes);
512 kfree_rcu(head, rcu);
513 spin_unlock(&sysctl_lock);
516 static struct ctl_table_header *grab_header(struct inode *inode)
518 struct ctl_table_header *head = PROC_I(inode)->sysctl;
520 head = &sysctl_table_root.default_set.dir.header;
521 return sysctl_head_grab(head);
524 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
527 struct ctl_table_header *head = grab_header(dir);
528 struct ctl_table_header *h = NULL;
529 const struct qstr *name = &dentry->d_name;
532 struct dentry *err = ERR_PTR(-ENOENT);
533 struct ctl_dir *ctl_dir;
537 return ERR_CAST(head);
539 ctl_dir = container_of(head, struct ctl_dir, header);
541 p = lookup_entry(&h, ctl_dir, name->name, name->len);
545 if (S_ISLNK(p->mode)) {
546 ret = sysctl_follow_link(&h, &p);
552 err = ERR_PTR(-ENOMEM);
553 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
557 d_set_d_op(dentry, &proc_sys_dentry_operations);
558 err = d_splice_alias(inode, dentry);
562 sysctl_head_finish(h);
563 sysctl_head_finish(head);
567 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
568 size_t count, loff_t *ppos, int write)
570 struct inode *inode = file_inode(filp);
571 struct ctl_table_header *head = grab_header(inode);
572 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
577 return PTR_ERR(head);
580 * At this point we know that the sysctl was not unregistered
581 * and won't be until we finish.
584 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
587 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
589 if (!table->proc_handler)
592 /* careful: calling conventions are nasty here */
594 error = table->proc_handler(table, write, buf, &res, ppos);
598 sysctl_head_finish(head);
603 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
604 size_t count, loff_t *ppos)
606 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
609 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
610 size_t count, loff_t *ppos)
612 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
615 static int proc_sys_open(struct inode *inode, struct file *filp)
617 struct ctl_table_header *head = grab_header(inode);
618 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
620 /* sysctl was unregistered */
622 return PTR_ERR(head);
625 filp->private_data = proc_sys_poll_event(table->poll);
627 sysctl_head_finish(head);
632 static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
634 struct inode *inode = file_inode(filp);
635 struct ctl_table_header *head = grab_header(inode);
636 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
637 __poll_t ret = DEFAULT_POLLMASK;
640 /* sysctl was unregistered */
642 return EPOLLERR | EPOLLHUP;
644 if (!table->proc_handler)
650 event = (unsigned long)filp->private_data;
651 poll_wait(filp, &table->poll->wait, wait);
653 if (event != atomic_read(&table->poll->event)) {
654 filp->private_data = proc_sys_poll_event(table->poll);
655 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
659 sysctl_head_finish(head);
664 static bool proc_sys_fill_cache(struct file *file,
665 struct dir_context *ctx,
666 struct ctl_table_header *head,
667 struct ctl_table *table)
669 struct dentry *child, *dir = file->f_path.dentry;
673 unsigned type = DT_UNKNOWN;
675 qname.name = table->procname;
676 qname.len = strlen(table->procname);
677 qname.hash = full_name_hash(dir, qname.name, qname.len);
679 child = d_lookup(dir, &qname);
681 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
682 child = d_alloc_parallel(dir, &qname, &wq);
685 if (d_in_lookup(child)) {
687 inode = proc_sys_make_inode(dir->d_sb, head, table);
689 d_lookup_done(child);
693 d_set_d_op(child, &proc_sys_dentry_operations);
694 res = d_splice_alias(inode, child);
695 d_lookup_done(child);
706 inode = d_inode(child);
708 type = inode->i_mode >> 12;
710 return dir_emit(ctx, qname.name, qname.len, ino, type);
713 static bool proc_sys_link_fill_cache(struct file *file,
714 struct dir_context *ctx,
715 struct ctl_table_header *head,
716 struct ctl_table *table)
720 head = sysctl_head_grab(head);
724 /* It is not an error if we can not follow the link ignore it */
725 if (sysctl_follow_link(&head, &table))
728 ret = proc_sys_fill_cache(file, ctx, head, table);
730 sysctl_head_finish(head);
734 static int scan(struct ctl_table_header *head, struct ctl_table *table,
735 unsigned long *pos, struct file *file,
736 struct dir_context *ctx)
740 if ((*pos)++ < ctx->pos)
743 if (unlikely(S_ISLNK(table->mode)))
744 res = proc_sys_link_fill_cache(file, ctx, head, table);
746 res = proc_sys_fill_cache(file, ctx, head, table);
754 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
756 struct ctl_table_header *head = grab_header(file_inode(file));
757 struct ctl_table_header *h = NULL;
758 struct ctl_table *entry;
759 struct ctl_dir *ctl_dir;
763 return PTR_ERR(head);
765 ctl_dir = container_of(head, struct ctl_dir, header);
767 if (!dir_emit_dots(file, ctx))
772 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
773 if (!scan(h, entry, &pos, file, ctx)) {
774 sysctl_head_finish(h);
779 sysctl_head_finish(head);
783 static int proc_sys_permission(struct inode *inode, int mask)
786 * sysctl entries that are not writeable,
787 * are _NOT_ writeable, capabilities or not.
789 struct ctl_table_header *head;
790 struct ctl_table *table;
793 /* Executable files are not allowed under /proc/sys/ */
794 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
797 head = grab_header(inode);
799 return PTR_ERR(head);
801 table = PROC_I(inode)->sysctl_entry;
802 if (!table) /* global root - r-xr-xr-x */
803 error = mask & MAY_WRITE ? -EACCES : 0;
804 else /* Use the permissions on the sysctl table entry */
805 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
807 sysctl_head_finish(head);
811 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
813 struct inode *inode = d_inode(dentry);
816 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
819 error = setattr_prepare(dentry, attr);
823 setattr_copy(inode, attr);
824 mark_inode_dirty(inode);
828 static int proc_sys_getattr(const struct path *path, struct kstat *stat,
829 u32 request_mask, unsigned int query_flags)
831 struct inode *inode = d_inode(path->dentry);
832 struct ctl_table_header *head = grab_header(inode);
833 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
836 return PTR_ERR(head);
838 generic_fillattr(inode, stat);
840 stat->mode = (stat->mode & S_IFMT) | table->mode;
842 sysctl_head_finish(head);
846 static const struct file_operations proc_sys_file_operations = {
847 .open = proc_sys_open,
848 .poll = proc_sys_poll,
849 .read = proc_sys_read,
850 .write = proc_sys_write,
851 .llseek = default_llseek,
854 static const struct file_operations proc_sys_dir_file_operations = {
855 .read = generic_read_dir,
856 .iterate_shared = proc_sys_readdir,
857 .llseek = generic_file_llseek,
860 static const struct inode_operations proc_sys_inode_operations = {
861 .permission = proc_sys_permission,
862 .setattr = proc_sys_setattr,
863 .getattr = proc_sys_getattr,
866 static const struct inode_operations proc_sys_dir_operations = {
867 .lookup = proc_sys_lookup,
868 .permission = proc_sys_permission,
869 .setattr = proc_sys_setattr,
870 .getattr = proc_sys_getattr,
873 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
875 if (flags & LOOKUP_RCU)
877 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
880 static int proc_sys_delete(const struct dentry *dentry)
882 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
885 static int sysctl_is_seen(struct ctl_table_header *p)
887 struct ctl_table_set *set = p->set;
889 spin_lock(&sysctl_lock);
890 if (p->unregistering)
892 else if (!set->is_seen)
895 res = set->is_seen(set);
896 spin_unlock(&sysctl_lock);
900 static int proc_sys_compare(const struct dentry *dentry,
901 unsigned int len, const char *str, const struct qstr *name)
903 struct ctl_table_header *head;
906 /* Although proc doesn't have negative dentries, rcu-walk means
907 * that inode here can be NULL */
908 /* AV: can it, indeed? */
909 inode = d_inode_rcu(dentry);
912 if (name->len != len)
914 if (memcmp(name->name, str, len))
916 head = rcu_dereference(PROC_I(inode)->sysctl);
917 return !head || !sysctl_is_seen(head);
920 static const struct dentry_operations proc_sys_dentry_operations = {
921 .d_revalidate = proc_sys_revalidate,
922 .d_delete = proc_sys_delete,
923 .d_compare = proc_sys_compare,
926 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
927 const char *name, int namelen)
929 struct ctl_table_header *head;
930 struct ctl_table *entry;
932 entry = find_entry(&head, dir, name, namelen);
934 return ERR_PTR(-ENOENT);
935 if (!S_ISDIR(entry->mode))
936 return ERR_PTR(-ENOTDIR);
937 return container_of(head, struct ctl_dir, header);
940 static struct ctl_dir *new_dir(struct ctl_table_set *set,
941 const char *name, int namelen)
943 struct ctl_table *table;
945 struct ctl_node *node;
948 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
949 sizeof(struct ctl_table)*2 + namelen + 1,
954 node = (struct ctl_node *)(new + 1);
955 table = (struct ctl_table *)(node + 1);
956 new_name = (char *)(table + 2);
957 memcpy(new_name, name, namelen);
958 new_name[namelen] = '\0';
959 table[0].procname = new_name;
960 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
961 init_header(&new->header, set->dir.header.root, set, node, table);
967 * get_subdir - find or create a subdir with the specified name.
968 * @dir: Directory to create the subdirectory in
969 * @name: The name of the subdirectory to find or create
970 * @namelen: The length of name
972 * Takes a directory with an elevated reference count so we know that
973 * if we drop the lock the directory will not go away. Upon success
974 * the reference is moved from @dir to the returned subdirectory.
975 * Upon error an error code is returned and the reference on @dir is
978 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
979 const char *name, int namelen)
981 struct ctl_table_set *set = dir->header.set;
982 struct ctl_dir *subdir, *new = NULL;
985 spin_lock(&sysctl_lock);
986 subdir = find_subdir(dir, name, namelen);
989 if (PTR_ERR(subdir) != -ENOENT)
992 spin_unlock(&sysctl_lock);
993 new = new_dir(set, name, namelen);
994 spin_lock(&sysctl_lock);
995 subdir = ERR_PTR(-ENOMEM);
999 /* Was the subdir added while we dropped the lock? */
1000 subdir = find_subdir(dir, name, namelen);
1001 if (!IS_ERR(subdir))
1003 if (PTR_ERR(subdir) != -ENOENT)
1006 /* Nope. Use the our freshly made directory entry. */
1007 err = insert_header(dir, &new->header);
1008 subdir = ERR_PTR(err);
1013 subdir->header.nreg++;
1015 if (IS_ERR(subdir)) {
1016 pr_err("sysctl could not get directory: ");
1017 sysctl_print_dir(dir);
1018 pr_cont("/%*.*s %ld\n",
1019 namelen, namelen, name, PTR_ERR(subdir));
1021 drop_sysctl_table(&dir->header);
1023 drop_sysctl_table(&new->header);
1024 spin_unlock(&sysctl_lock);
1028 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1030 struct ctl_dir *parent;
1031 const char *procname;
1032 if (!dir->header.parent)
1034 parent = xlate_dir(set, dir->header.parent);
1037 procname = dir->header.ctl_table[0].procname;
1038 return find_subdir(parent, procname, strlen(procname));
1041 static int sysctl_follow_link(struct ctl_table_header **phead,
1042 struct ctl_table **pentry)
1044 struct ctl_table_header *head;
1045 struct ctl_table_root *root;
1046 struct ctl_table_set *set;
1047 struct ctl_table *entry;
1048 struct ctl_dir *dir;
1052 spin_lock(&sysctl_lock);
1053 root = (*pentry)->data;
1054 set = lookup_header_set(root);
1055 dir = xlate_dir(set, (*phead)->parent);
1059 const char *procname = (*pentry)->procname;
1061 entry = find_entry(&head, dir, procname, strlen(procname));
1063 if (entry && use_table(head)) {
1064 unuse_table(*phead);
1071 spin_unlock(&sysctl_lock);
1075 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1077 struct va_format vaf;
1080 va_start(args, fmt);
1084 pr_err("sysctl table check failed: %s/%s %pV\n",
1085 path, table->procname, &vaf);
1091 static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1095 if ((table->proc_handler == proc_douintvec) ||
1096 (table->proc_handler == proc_douintvec_minmax)) {
1097 if (table->maxlen != sizeof(unsigned int))
1098 err |= sysctl_err(path, table, "array not allowed");
1104 static int sysctl_check_table(const char *path, struct ctl_table *table)
1107 for (; table->procname; table++) {
1109 err |= sysctl_err(path, table, "Not a file");
1111 if ((table->proc_handler == proc_dostring) ||
1112 (table->proc_handler == proc_dointvec) ||
1113 (table->proc_handler == proc_douintvec) ||
1114 (table->proc_handler == proc_douintvec_minmax) ||
1115 (table->proc_handler == proc_dointvec_minmax) ||
1116 (table->proc_handler == proc_dointvec_jiffies) ||
1117 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1118 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1119 (table->proc_handler == proc_doulongvec_minmax) ||
1120 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1122 err |= sysctl_err(path, table, "No data");
1124 err |= sysctl_err(path, table, "No maxlen");
1126 err |= sysctl_check_table_array(path, table);
1128 if (!table->proc_handler)
1129 err |= sysctl_err(path, table, "No proc_handler");
1131 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1132 err |= sysctl_err(path, table, "bogus .mode 0%o",
1138 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1139 struct ctl_table_root *link_root)
1141 struct ctl_table *link_table, *entry, *link;
1142 struct ctl_table_header *links;
1143 struct ctl_node *node;
1145 int nr_entries, name_bytes;
1149 for (entry = table; entry->procname; entry++) {
1151 name_bytes += strlen(entry->procname) + 1;
1154 links = kzalloc(sizeof(struct ctl_table_header) +
1155 sizeof(struct ctl_node)*nr_entries +
1156 sizeof(struct ctl_table)*(nr_entries + 1) +
1163 node = (struct ctl_node *)(links + 1);
1164 link_table = (struct ctl_table *)(node + nr_entries);
1165 link_name = (char *)&link_table[nr_entries + 1];
1167 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1168 int len = strlen(entry->procname) + 1;
1169 memcpy(link_name, entry->procname, len);
1170 link->procname = link_name;
1171 link->mode = S_IFLNK|S_IRWXUGO;
1172 link->data = link_root;
1175 init_header(links, dir->header.root, dir->header.set, node, link_table);
1176 links->nreg = nr_entries;
1181 static bool get_links(struct ctl_dir *dir,
1182 struct ctl_table *table, struct ctl_table_root *link_root)
1184 struct ctl_table_header *head;
1185 struct ctl_table *entry, *link;
1187 /* Are there links available for every entry in table? */
1188 for (entry = table; entry->procname; entry++) {
1189 const char *procname = entry->procname;
1190 link = find_entry(&head, dir, procname, strlen(procname));
1193 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1195 if (S_ISLNK(link->mode) && (link->data == link_root))
1200 /* The checks passed. Increase the registration count on the links */
1201 for (entry = table; entry->procname; entry++) {
1202 const char *procname = entry->procname;
1203 link = find_entry(&head, dir, procname, strlen(procname));
1209 static int insert_links(struct ctl_table_header *head)
1211 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1212 struct ctl_dir *core_parent = NULL;
1213 struct ctl_table_header *links;
1216 if (head->set == root_set)
1219 core_parent = xlate_dir(root_set, head->parent);
1220 if (IS_ERR(core_parent))
1223 if (get_links(core_parent, head->ctl_table, head->root))
1226 core_parent->header.nreg++;
1227 spin_unlock(&sysctl_lock);
1229 links = new_links(core_parent, head->ctl_table, head->root);
1231 spin_lock(&sysctl_lock);
1237 if (get_links(core_parent, head->ctl_table, head->root)) {
1242 err = insert_header(core_parent, links);
1246 drop_sysctl_table(&core_parent->header);
1251 * __register_sysctl_table - register a leaf sysctl table
1252 * @set: Sysctl tree to register on
1253 * @path: The path to the directory the sysctl table is in.
1254 * @table: the top-level table structure
1256 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1257 * array. A completely 0 filled entry terminates the table.
1259 * The members of the &struct ctl_table structure are used as follows:
1261 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1262 * enter a sysctl file
1264 * data - a pointer to data for use by proc_handler
1266 * maxlen - the maximum size in bytes of the data
1268 * mode - the file permissions for the /proc/sys file
1270 * child - must be %NULL.
1272 * proc_handler - the text handler routine (described below)
1274 * extra1, extra2 - extra pointers usable by the proc handler routines
1276 * Leaf nodes in the sysctl tree will be represented by a single file
1277 * under /proc; non-leaf nodes will be represented by directories.
1279 * There must be a proc_handler routine for any terminal nodes.
1280 * Several default handlers are available to cover common cases -
1282 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1283 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1284 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1286 * It is the handler's job to read the input buffer from user memory
1287 * and process it. The handler should return 0 on success.
1289 * This routine returns %NULL on a failure to register, and a pointer
1290 * to the table header on success.
1292 struct ctl_table_header *__register_sysctl_table(
1293 struct ctl_table_set *set,
1294 const char *path, struct ctl_table *table)
1296 struct ctl_table_root *root = set->dir.header.root;
1297 struct ctl_table_header *header;
1298 const char *name, *nextname;
1299 struct ctl_dir *dir;
1300 struct ctl_table *entry;
1301 struct ctl_node *node;
1304 for (entry = table; entry->procname; entry++)
1307 header = kzalloc(sizeof(struct ctl_table_header) +
1308 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1312 node = (struct ctl_node *)(header + 1);
1313 init_header(header, root, set, node, table);
1314 if (sysctl_check_table(path, table))
1317 spin_lock(&sysctl_lock);
1319 /* Reference moved down the diretory tree get_subdir */
1321 spin_unlock(&sysctl_lock);
1323 /* Find the directory for the ctl_table */
1324 for (name = path; name; name = nextname) {
1326 nextname = strchr(name, '/');
1328 namelen = nextname - name;
1331 namelen = strlen(name);
1336 dir = get_subdir(dir, name, namelen);
1341 spin_lock(&sysctl_lock);
1342 if (insert_header(dir, header))
1343 goto fail_put_dir_locked;
1345 drop_sysctl_table(&dir->header);
1346 spin_unlock(&sysctl_lock);
1350 fail_put_dir_locked:
1351 drop_sysctl_table(&dir->header);
1352 spin_unlock(&sysctl_lock);
1360 * register_sysctl - register a sysctl table
1361 * @path: The path to the directory the sysctl table is in.
1362 * @table: the table structure
1364 * Register a sysctl table. @table should be a filled in ctl_table
1365 * array. A completely 0 filled entry terminates the table.
1367 * See __register_sysctl_table for more details.
1369 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1371 return __register_sysctl_table(&sysctl_table_root.default_set,
1374 EXPORT_SYMBOL(register_sysctl);
1376 static char *append_path(const char *path, char *pos, const char *name)
1379 namelen = strlen(name);
1380 if (((pos - path) + namelen + 2) >= PATH_MAX)
1382 memcpy(pos, name, namelen);
1384 pos[namelen + 1] = '\0';
1389 static int count_subheaders(struct ctl_table *table)
1392 int nr_subheaders = 0;
1393 struct ctl_table *entry;
1395 /* special case: no directory and empty directory */
1396 if (!table || !table->procname)
1399 for (entry = table; entry->procname; entry++) {
1401 nr_subheaders += count_subheaders(entry->child);
1405 return nr_subheaders + has_files;
1408 static int register_leaf_sysctl_tables(const char *path, char *pos,
1409 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1410 struct ctl_table *table)
1412 struct ctl_table *ctl_table_arg = NULL;
1413 struct ctl_table *entry, *files;
1418 for (entry = table; entry->procname; entry++) {
1426 /* If there are mixed files and directories we need a new table */
1427 if (nr_dirs && nr_files) {
1428 struct ctl_table *new;
1429 files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1434 ctl_table_arg = files;
1435 for (new = files, entry = table; entry->procname; entry++) {
1443 /* Register everything except a directory full of subdirectories */
1444 if (nr_files || !nr_dirs) {
1445 struct ctl_table_header *header;
1446 header = __register_sysctl_table(set, path, files);
1448 kfree(ctl_table_arg);
1452 /* Remember if we need to free the file table */
1453 header->ctl_table_arg = ctl_table_arg;
1454 **subheader = header;
1458 /* Recurse into the subdirectories. */
1459 for (entry = table; entry->procname; entry++) {
1465 err = -ENAMETOOLONG;
1466 child_pos = append_path(path, pos, entry->procname);
1470 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1478 /* On failure our caller will unregister all registered subheaders */
1483 * __register_sysctl_paths - register a sysctl table hierarchy
1484 * @set: Sysctl tree to register on
1485 * @path: The path to the directory the sysctl table is in.
1486 * @table: the top-level table structure
1488 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1489 * array. A completely 0 filled entry terminates the table.
1491 * See __register_sysctl_table for more details.
1493 struct ctl_table_header *__register_sysctl_paths(
1494 struct ctl_table_set *set,
1495 const struct ctl_path *path, struct ctl_table *table)
1497 struct ctl_table *ctl_table_arg = table;
1498 int nr_subheaders = count_subheaders(table);
1499 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1500 const struct ctl_path *component;
1501 char *new_path, *pos;
1503 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1508 for (component = path; component->procname; component++) {
1509 pos = append_path(new_path, pos, component->procname);
1513 while (table->procname && table->child && !table[1].procname) {
1514 pos = append_path(new_path, pos, table->procname);
1517 table = table->child;
1519 if (nr_subheaders == 1) {
1520 header = __register_sysctl_table(set, new_path, table);
1522 header->ctl_table_arg = ctl_table_arg;
1524 header = kzalloc(sizeof(*header) +
1525 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1529 subheaders = (struct ctl_table_header **) (header + 1);
1530 subheader = subheaders;
1531 header->ctl_table_arg = ctl_table_arg;
1533 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1535 goto err_register_leaves;
1542 err_register_leaves:
1543 while (subheader > subheaders) {
1544 struct ctl_table_header *subh = *(--subheader);
1545 struct ctl_table *table = subh->ctl_table_arg;
1546 unregister_sysctl_table(subh);
1555 * register_sysctl_table_path - register a sysctl table hierarchy
1556 * @path: The path to the directory the sysctl table is in.
1557 * @table: the top-level table structure
1559 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1560 * array. A completely 0 filled entry terminates the table.
1562 * See __register_sysctl_paths for more details.
1564 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1565 struct ctl_table *table)
1567 return __register_sysctl_paths(&sysctl_table_root.default_set,
1570 EXPORT_SYMBOL(register_sysctl_paths);
1573 * register_sysctl_table - register a sysctl table hierarchy
1574 * @table: the top-level table structure
1576 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1577 * array. A completely 0 filled entry terminates the table.
1579 * See register_sysctl_paths for more details.
1581 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1583 static const struct ctl_path null_path[] = { {} };
1585 return register_sysctl_paths(null_path, table);
1587 EXPORT_SYMBOL(register_sysctl_table);
1589 static void put_links(struct ctl_table_header *header)
1591 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1592 struct ctl_table_root *root = header->root;
1593 struct ctl_dir *parent = header->parent;
1594 struct ctl_dir *core_parent;
1595 struct ctl_table *entry;
1597 if (header->set == root_set)
1600 core_parent = xlate_dir(root_set, parent);
1601 if (IS_ERR(core_parent))
1604 for (entry = header->ctl_table; entry->procname; entry++) {
1605 struct ctl_table_header *link_head;
1606 struct ctl_table *link;
1607 const char *name = entry->procname;
1609 link = find_entry(&link_head, core_parent, name, strlen(name));
1611 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1612 (S_ISLNK(link->mode) && (link->data == root)))) {
1613 drop_sysctl_table(link_head);
1616 pr_err("sysctl link missing during unregister: ");
1617 sysctl_print_dir(parent);
1618 pr_cont("/%s\n", name);
1623 static void drop_sysctl_table(struct ctl_table_header *header)
1625 struct ctl_dir *parent = header->parent;
1631 start_unregistering(header);
1632 if (!--header->count)
1633 kfree_rcu(header, rcu);
1636 drop_sysctl_table(&parent->header);
1640 * unregister_sysctl_table - unregister a sysctl table hierarchy
1641 * @header: the header returned from register_sysctl_table
1643 * Unregisters the sysctl table and all children. proc entries may not
1644 * actually be removed until they are no longer used by anyone.
1646 void unregister_sysctl_table(struct ctl_table_header * header)
1654 nr_subheaders = count_subheaders(header->ctl_table_arg);
1655 if (unlikely(nr_subheaders > 1)) {
1656 struct ctl_table_header **subheaders;
1659 subheaders = (struct ctl_table_header **)(header + 1);
1660 for (i = nr_subheaders -1; i >= 0; i--) {
1661 struct ctl_table_header *subh = subheaders[i];
1662 struct ctl_table *table = subh->ctl_table_arg;
1663 unregister_sysctl_table(subh);
1670 spin_lock(&sysctl_lock);
1671 drop_sysctl_table(header);
1672 spin_unlock(&sysctl_lock);
1674 EXPORT_SYMBOL(unregister_sysctl_table);
1676 void setup_sysctl_set(struct ctl_table_set *set,
1677 struct ctl_table_root *root,
1678 int (*is_seen)(struct ctl_table_set *))
1680 memset(set, 0, sizeof(*set));
1681 set->is_seen = is_seen;
1682 init_header(&set->dir.header, root, set, NULL, root_table);
1685 void retire_sysctl_set(struct ctl_table_set *set)
1687 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1690 int __init proc_sys_init(void)
1692 struct proc_dir_entry *proc_sys_root;
1694 proc_sys_root = proc_mkdir("sys", NULL);
1695 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1696 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1697 proc_sys_root->nlink = 0;
1699 return sysctl_init();