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/uio.h>
16 #include <linux/module.h>
17 #include <linux/bpf-cgroup.h>
18 #include <linux/mount.h>
19 #include <linux/kmemleak.h>
22 #define list_for_each_table_entry(entry, header) \
23 entry = header->ctl_table; \
24 for (size_t i = 0 ; i < header->ctl_table_size && entry->procname; ++i, entry++)
26 static const struct dentry_operations proc_sys_dentry_operations;
27 static const struct file_operations proc_sys_file_operations;
28 static const struct inode_operations proc_sys_inode_operations;
29 static const struct file_operations proc_sys_dir_file_operations;
30 static const struct inode_operations proc_sys_dir_operations;
32 /* Support for permanently empty directories */
33 static struct ctl_table sysctl_mount_point[] = {
34 {.type = SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY }
38 * register_sysctl_mount_point() - registers a sysctl mount point
39 * @path: path for the mount point
41 * Used to create a permanently empty directory to serve as mount point.
42 * There are some subtle but important permission checks this allows in the
43 * case of unprivileged mounts.
45 struct ctl_table_header *register_sysctl_mount_point(const char *path)
47 return register_sysctl(path, sysctl_mount_point);
49 EXPORT_SYMBOL(register_sysctl_mount_point);
51 #define sysctl_is_perm_empty_ctl_table(tptr) \
52 (tptr[0].type == SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
53 #define sysctl_is_perm_empty_ctl_header(hptr) \
54 (sysctl_is_perm_empty_ctl_table(hptr->ctl_table))
55 #define sysctl_set_perm_empty_ctl_header(hptr) \
56 (hptr->ctl_table[0].type = SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
57 #define sysctl_clear_perm_empty_ctl_header(hptr) \
58 (hptr->ctl_table[0].type = SYSCTL_TABLE_TYPE_DEFAULT)
60 void proc_sys_poll_notify(struct ctl_table_poll *poll)
65 atomic_inc(&poll->event);
66 wake_up_interruptible(&poll->wait);
69 static struct ctl_table root_table[] = {
72 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
75 static struct ctl_table_root sysctl_table_root = {
76 .default_set.dir.header = {
79 .ctl_table = root_table }},
80 .ctl_table_arg = root_table,
81 .root = &sysctl_table_root,
82 .set = &sysctl_table_root.default_set,
86 static DEFINE_SPINLOCK(sysctl_lock);
88 static void drop_sysctl_table(struct ctl_table_header *header);
89 static int sysctl_follow_link(struct ctl_table_header **phead,
90 struct ctl_table **pentry);
91 static int insert_links(struct ctl_table_header *head);
92 static void put_links(struct ctl_table_header *header);
94 static void sysctl_print_dir(struct ctl_dir *dir)
96 if (dir->header.parent)
97 sysctl_print_dir(dir->header.parent);
98 pr_cont("%s/", dir->header.ctl_table[0].procname);
101 static int namecmp(const char *name1, int len1, const char *name2, int len2)
105 cmp = memcmp(name1, name2, min(len1, len2));
111 /* Called under sysctl_lock */
112 static struct ctl_table *find_entry(struct ctl_table_header **phead,
113 struct ctl_dir *dir, const char *name, int namelen)
115 struct ctl_table_header *head;
116 struct ctl_table *entry;
117 struct rb_node *node = dir->root.rb_node;
121 struct ctl_node *ctl_node;
122 const char *procname;
125 ctl_node = rb_entry(node, struct ctl_node, node);
126 head = ctl_node->header;
127 entry = &head->ctl_table[ctl_node - head->node];
128 procname = entry->procname;
130 cmp = namecmp(name, namelen, procname, strlen(procname));
132 node = node->rb_left;
134 node = node->rb_right;
143 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
145 struct rb_node *node = &head->node[entry - head->ctl_table].node;
146 struct rb_node **p = &head->parent->root.rb_node;
147 struct rb_node *parent = NULL;
148 const char *name = entry->procname;
149 int namelen = strlen(name);
152 struct ctl_table_header *parent_head;
153 struct ctl_table *parent_entry;
154 struct ctl_node *parent_node;
155 const char *parent_name;
159 parent_node = rb_entry(parent, struct ctl_node, node);
160 parent_head = parent_node->header;
161 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
162 parent_name = parent_entry->procname;
164 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
170 pr_err("sysctl duplicate entry: ");
171 sysctl_print_dir(head->parent);
172 pr_cont("%s\n", entry->procname);
177 rb_link_node(node, parent, p);
178 rb_insert_color(node, &head->parent->root);
182 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
184 struct rb_node *node = &head->node[entry - head->ctl_table].node;
186 rb_erase(node, &head->parent->root);
189 static void init_header(struct ctl_table_header *head,
190 struct ctl_table_root *root, struct ctl_table_set *set,
191 struct ctl_node *node, struct ctl_table *table, size_t table_size)
193 head->ctl_table = table;
194 head->ctl_table_size = table_size;
195 head->ctl_table_arg = table;
199 head->unregistering = NULL;
204 INIT_HLIST_HEAD(&head->inodes);
206 struct ctl_table *entry;
208 list_for_each_table_entry(entry, head) {
215 static void erase_header(struct ctl_table_header *head)
217 struct ctl_table *entry;
219 list_for_each_table_entry(entry, head)
220 erase_entry(head, entry);
223 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
225 struct ctl_table *entry;
226 struct ctl_table_header *dir_h = &dir->header;
230 /* Is this a permanently empty directory? */
231 if (sysctl_is_perm_empty_ctl_header(dir_h))
234 /* Am I creating a permanently empty directory? */
235 if (header->ctl_table_size > 0 &&
236 sysctl_is_perm_empty_ctl_table(header->ctl_table)) {
237 if (!RB_EMPTY_ROOT(&dir->root))
239 sysctl_set_perm_empty_ctl_header(dir_h);
243 header->parent = dir;
244 err = insert_links(header);
247 list_for_each_table_entry(entry, header) {
248 err = insert_entry(header, entry);
254 erase_header(header);
257 if (header->ctl_table == sysctl_mount_point)
258 sysctl_clear_perm_empty_ctl_header(dir_h);
259 header->parent = NULL;
260 drop_sysctl_table(dir_h);
264 /* called under sysctl_lock */
265 static int use_table(struct ctl_table_header *p)
267 if (unlikely(p->unregistering))
273 /* called under sysctl_lock */
274 static void unuse_table(struct ctl_table_header *p)
277 if (unlikely(p->unregistering))
278 complete(p->unregistering);
281 static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
283 proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
286 /* called under sysctl_lock, will reacquire if has to wait */
287 static void start_unregistering(struct ctl_table_header *p)
290 * if p->used is 0, nobody will ever touch that entry again;
291 * we'll eliminate all paths to it before dropping sysctl_lock
293 if (unlikely(p->used)) {
294 struct completion wait;
295 init_completion(&wait);
296 p->unregistering = &wait;
297 spin_unlock(&sysctl_lock);
298 wait_for_completion(&wait);
300 /* anything non-NULL; we'll never dereference it */
301 p->unregistering = ERR_PTR(-EINVAL);
302 spin_unlock(&sysctl_lock);
305 * Invalidate dentries for unregistered sysctls: namespaced sysctls
306 * can have duplicate names and contaminate dcache very badly.
308 proc_sys_invalidate_dcache(p);
310 * do not remove from the list until nobody holds it; walking the
311 * list in do_sysctl() relies on that.
313 spin_lock(&sysctl_lock);
317 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
320 spin_lock(&sysctl_lock);
321 if (!use_table(head))
322 head = ERR_PTR(-ENOENT);
323 spin_unlock(&sysctl_lock);
327 static void sysctl_head_finish(struct ctl_table_header *head)
331 spin_lock(&sysctl_lock);
333 spin_unlock(&sysctl_lock);
336 static struct ctl_table_set *
337 lookup_header_set(struct ctl_table_root *root)
339 struct ctl_table_set *set = &root->default_set;
341 set = root->lookup(root);
345 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
347 const char *name, int namelen)
349 struct ctl_table_header *head;
350 struct ctl_table *entry;
352 spin_lock(&sysctl_lock);
353 entry = find_entry(&head, dir, name, namelen);
354 if (entry && use_table(head))
358 spin_unlock(&sysctl_lock);
362 static struct ctl_node *first_usable_entry(struct rb_node *node)
364 struct ctl_node *ctl_node;
366 for (;node; node = rb_next(node)) {
367 ctl_node = rb_entry(node, struct ctl_node, node);
368 if (use_table(ctl_node->header))
374 static void first_entry(struct ctl_dir *dir,
375 struct ctl_table_header **phead, struct ctl_table **pentry)
377 struct ctl_table_header *head = NULL;
378 struct ctl_table *entry = NULL;
379 struct ctl_node *ctl_node;
381 spin_lock(&sysctl_lock);
382 ctl_node = first_usable_entry(rb_first(&dir->root));
383 spin_unlock(&sysctl_lock);
385 head = ctl_node->header;
386 entry = &head->ctl_table[ctl_node - head->node];
392 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
394 struct ctl_table_header *head = *phead;
395 struct ctl_table *entry = *pentry;
396 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
398 spin_lock(&sysctl_lock);
401 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
402 spin_unlock(&sysctl_lock);
405 head = ctl_node->header;
406 entry = &head->ctl_table[ctl_node - head->node];
413 * sysctl_perm does NOT grant the superuser all rights automatically, because
414 * some sysctl variables are readonly even to root.
417 static int test_perm(int mode, int op)
419 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
421 else if (in_egroup_p(GLOBAL_ROOT_GID))
423 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
428 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
430 struct ctl_table_root *root = head->root;
433 if (root->permissions)
434 mode = root->permissions(head, table);
438 return test_perm(mode, op);
441 static struct inode *proc_sys_make_inode(struct super_block *sb,
442 struct ctl_table_header *head, struct ctl_table *table)
444 struct ctl_table_root *root = head->root;
446 struct proc_inode *ei;
448 inode = new_inode(sb);
450 return ERR_PTR(-ENOMEM);
452 inode->i_ino = get_next_ino();
456 spin_lock(&sysctl_lock);
457 if (unlikely(head->unregistering)) {
458 spin_unlock(&sysctl_lock);
460 return ERR_PTR(-ENOENT);
463 ei->sysctl_entry = table;
464 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
466 spin_unlock(&sysctl_lock);
468 simple_inode_init_ts(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 (sysctl_is_perm_empty_ctl_header(head))
479 make_empty_dir_inode(inode);
482 if (root->set_ownership)
483 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
485 inode->i_uid = GLOBAL_ROOT_UID;
486 inode->i_gid = GLOBAL_ROOT_GID;
492 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
494 spin_lock(&sysctl_lock);
495 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
497 kfree_rcu(head, rcu);
498 spin_unlock(&sysctl_lock);
501 static struct ctl_table_header *grab_header(struct inode *inode)
503 struct ctl_table_header *head = PROC_I(inode)->sysctl;
505 head = &sysctl_table_root.default_set.dir.header;
506 return sysctl_head_grab(head);
509 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
512 struct ctl_table_header *head = grab_header(dir);
513 struct ctl_table_header *h = NULL;
514 const struct qstr *name = &dentry->d_name;
517 struct dentry *err = ERR_PTR(-ENOENT);
518 struct ctl_dir *ctl_dir;
522 return ERR_CAST(head);
524 ctl_dir = container_of(head, struct ctl_dir, header);
526 p = lookup_entry(&h, ctl_dir, name->name, name->len);
530 if (S_ISLNK(p->mode)) {
531 ret = sysctl_follow_link(&h, &p);
537 d_set_d_op(dentry, &proc_sys_dentry_operations);
538 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
539 err = d_splice_alias(inode, dentry);
543 sysctl_head_finish(h);
544 sysctl_head_finish(head);
548 static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
551 struct inode *inode = file_inode(iocb->ki_filp);
552 struct ctl_table_header *head = grab_header(inode);
553 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
554 size_t count = iov_iter_count(iter);
559 return PTR_ERR(head);
562 * At this point we know that the sysctl was not unregistered
563 * and won't be until we finish.
566 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
569 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
571 if (!table->proc_handler)
574 /* don't even try if the size is too large */
576 if (count >= KMALLOC_MAX_SIZE)
578 kbuf = kvzalloc(count + 1, GFP_KERNEL);
584 if (!copy_from_iter_full(kbuf, count, iter))
589 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
594 /* careful: calling conventions are nasty here */
595 error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
601 if (copy_to_iter(kbuf, count, iter) < count)
609 sysctl_head_finish(head);
614 static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
616 return proc_sys_call_handler(iocb, iter, 0);
619 static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
621 return proc_sys_call_handler(iocb, iter, 1);
624 static int proc_sys_open(struct inode *inode, struct file *filp)
626 struct ctl_table_header *head = grab_header(inode);
627 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
629 /* sysctl was unregistered */
631 return PTR_ERR(head);
634 filp->private_data = proc_sys_poll_event(table->poll);
636 sysctl_head_finish(head);
641 static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
643 struct inode *inode = file_inode(filp);
644 struct ctl_table_header *head = grab_header(inode);
645 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
646 __poll_t ret = DEFAULT_POLLMASK;
649 /* sysctl was unregistered */
651 return EPOLLERR | EPOLLHUP;
653 if (!table->proc_handler)
659 event = (unsigned long)filp->private_data;
660 poll_wait(filp, &table->poll->wait, wait);
662 if (event != atomic_read(&table->poll->event)) {
663 filp->private_data = proc_sys_poll_event(table->poll);
664 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
668 sysctl_head_finish(head);
673 static bool proc_sys_fill_cache(struct file *file,
674 struct dir_context *ctx,
675 struct ctl_table_header *head,
676 struct ctl_table *table)
678 struct dentry *child, *dir = file->f_path.dentry;
682 unsigned type = DT_UNKNOWN;
684 qname.name = table->procname;
685 qname.len = strlen(table->procname);
686 qname.hash = full_name_hash(dir, qname.name, qname.len);
688 child = d_lookup(dir, &qname);
690 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
691 child = d_alloc_parallel(dir, &qname, &wq);
694 if (d_in_lookup(child)) {
696 d_set_d_op(child, &proc_sys_dentry_operations);
697 inode = proc_sys_make_inode(dir->d_sb, head, table);
698 res = d_splice_alias(inode, child);
699 d_lookup_done(child);
710 inode = d_inode(child);
712 type = inode->i_mode >> 12;
714 return dir_emit(ctx, qname.name, qname.len, ino, type);
717 static bool proc_sys_link_fill_cache(struct file *file,
718 struct dir_context *ctx,
719 struct ctl_table_header *head,
720 struct ctl_table *table)
724 head = sysctl_head_grab(head);
728 /* It is not an error if we can not follow the link ignore it */
729 if (sysctl_follow_link(&head, &table))
732 ret = proc_sys_fill_cache(file, ctx, head, table);
734 sysctl_head_finish(head);
738 static int scan(struct ctl_table_header *head, struct ctl_table *table,
739 unsigned long *pos, struct file *file,
740 struct dir_context *ctx)
744 if ((*pos)++ < ctx->pos)
747 if (unlikely(S_ISLNK(table->mode)))
748 res = proc_sys_link_fill_cache(file, ctx, head, table);
750 res = proc_sys_fill_cache(file, ctx, head, table);
758 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
760 struct ctl_table_header *head = grab_header(file_inode(file));
761 struct ctl_table_header *h = NULL;
762 struct ctl_table *entry;
763 struct ctl_dir *ctl_dir;
767 return PTR_ERR(head);
769 ctl_dir = container_of(head, struct ctl_dir, header);
771 if (!dir_emit_dots(file, ctx))
776 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
777 if (!scan(h, entry, &pos, file, ctx)) {
778 sysctl_head_finish(h);
783 sysctl_head_finish(head);
787 static int proc_sys_permission(struct mnt_idmap *idmap,
788 struct inode *inode, int mask)
791 * sysctl entries that are not writeable,
792 * are _NOT_ writeable, capabilities or not.
794 struct ctl_table_header *head;
795 struct ctl_table *table;
798 /* Executable files are not allowed under /proc/sys/ */
799 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
802 head = grab_header(inode);
804 return PTR_ERR(head);
806 table = PROC_I(inode)->sysctl_entry;
807 if (!table) /* global root - r-xr-xr-x */
808 error = mask & MAY_WRITE ? -EACCES : 0;
809 else /* Use the permissions on the sysctl table entry */
810 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
812 sysctl_head_finish(head);
816 static int proc_sys_setattr(struct mnt_idmap *idmap,
817 struct dentry *dentry, struct iattr *attr)
819 struct inode *inode = d_inode(dentry);
822 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
825 error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
829 setattr_copy(&nop_mnt_idmap, inode, attr);
833 static int proc_sys_getattr(struct mnt_idmap *idmap,
834 const struct path *path, struct kstat *stat,
835 u32 request_mask, unsigned int query_flags)
837 struct inode *inode = d_inode(path->dentry);
838 struct ctl_table_header *head = grab_header(inode);
839 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
842 return PTR_ERR(head);
844 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
846 stat->mode = (stat->mode & S_IFMT) | table->mode;
848 sysctl_head_finish(head);
852 static const struct file_operations proc_sys_file_operations = {
853 .open = proc_sys_open,
854 .poll = proc_sys_poll,
855 .read_iter = proc_sys_read,
856 .write_iter = proc_sys_write,
857 .splice_read = copy_splice_read,
858 .splice_write = iter_file_splice_write,
859 .llseek = default_llseek,
862 static const struct file_operations proc_sys_dir_file_operations = {
863 .read = generic_read_dir,
864 .iterate_shared = proc_sys_readdir,
865 .llseek = generic_file_llseek,
868 static const struct inode_operations proc_sys_inode_operations = {
869 .permission = proc_sys_permission,
870 .setattr = proc_sys_setattr,
871 .getattr = proc_sys_getattr,
874 static const struct inode_operations proc_sys_dir_operations = {
875 .lookup = proc_sys_lookup,
876 .permission = proc_sys_permission,
877 .setattr = proc_sys_setattr,
878 .getattr = proc_sys_getattr,
881 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
883 if (flags & LOOKUP_RCU)
885 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
888 static int proc_sys_delete(const struct dentry *dentry)
890 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
893 static int sysctl_is_seen(struct ctl_table_header *p)
895 struct ctl_table_set *set = p->set;
897 spin_lock(&sysctl_lock);
898 if (p->unregistering)
900 else if (!set->is_seen)
903 res = set->is_seen(set);
904 spin_unlock(&sysctl_lock);
908 static int proc_sys_compare(const struct dentry *dentry,
909 unsigned int len, const char *str, const struct qstr *name)
911 struct ctl_table_header *head;
914 /* Although proc doesn't have negative dentries, rcu-walk means
915 * that inode here can be NULL */
916 /* AV: can it, indeed? */
917 inode = d_inode_rcu(dentry);
920 if (name->len != len)
922 if (memcmp(name->name, str, len))
924 head = rcu_dereference(PROC_I(inode)->sysctl);
925 return !head || !sysctl_is_seen(head);
928 static const struct dentry_operations proc_sys_dentry_operations = {
929 .d_revalidate = proc_sys_revalidate,
930 .d_delete = proc_sys_delete,
931 .d_compare = proc_sys_compare,
934 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
935 const char *name, int namelen)
937 struct ctl_table_header *head;
938 struct ctl_table *entry;
940 entry = find_entry(&head, dir, name, namelen);
942 return ERR_PTR(-ENOENT);
943 if (!S_ISDIR(entry->mode))
944 return ERR_PTR(-ENOTDIR);
945 return container_of(head, struct ctl_dir, header);
948 static struct ctl_dir *new_dir(struct ctl_table_set *set,
949 const char *name, int namelen)
951 struct ctl_table *table;
953 struct ctl_node *node;
956 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
957 sizeof(struct ctl_table)*2 + namelen + 1,
962 node = (struct ctl_node *)(new + 1);
963 table = (struct ctl_table *)(node + 1);
964 new_name = (char *)(table + 2);
965 memcpy(new_name, name, namelen);
966 table[0].procname = new_name;
967 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
968 init_header(&new->header, set->dir.header.root, set, node, table, 1);
974 * get_subdir - find or create a subdir with the specified name.
975 * @dir: Directory to create the subdirectory in
976 * @name: The name of the subdirectory to find or create
977 * @namelen: The length of name
979 * Takes a directory with an elevated reference count so we know that
980 * if we drop the lock the directory will not go away. Upon success
981 * the reference is moved from @dir to the returned subdirectory.
982 * Upon error an error code is returned and the reference on @dir is
985 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
986 const char *name, int namelen)
988 struct ctl_table_set *set = dir->header.set;
989 struct ctl_dir *subdir, *new = NULL;
992 spin_lock(&sysctl_lock);
993 subdir = find_subdir(dir, name, namelen);
996 if (PTR_ERR(subdir) != -ENOENT)
999 spin_unlock(&sysctl_lock);
1000 new = new_dir(set, name, namelen);
1001 spin_lock(&sysctl_lock);
1002 subdir = ERR_PTR(-ENOMEM);
1006 /* Was the subdir added while we dropped the lock? */
1007 subdir = find_subdir(dir, name, namelen);
1008 if (!IS_ERR(subdir))
1010 if (PTR_ERR(subdir) != -ENOENT)
1013 /* Nope. Use the our freshly made directory entry. */
1014 err = insert_header(dir, &new->header);
1015 subdir = ERR_PTR(err);
1020 subdir->header.nreg++;
1022 if (IS_ERR(subdir)) {
1023 pr_err("sysctl could not get directory: ");
1024 sysctl_print_dir(dir);
1025 pr_cont("%*.*s %ld\n", namelen, namelen, name,
1028 drop_sysctl_table(&dir->header);
1030 drop_sysctl_table(&new->header);
1031 spin_unlock(&sysctl_lock);
1035 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1037 struct ctl_dir *parent;
1038 const char *procname;
1039 if (!dir->header.parent)
1041 parent = xlate_dir(set, dir->header.parent);
1044 procname = dir->header.ctl_table[0].procname;
1045 return find_subdir(parent, procname, strlen(procname));
1048 static int sysctl_follow_link(struct ctl_table_header **phead,
1049 struct ctl_table **pentry)
1051 struct ctl_table_header *head;
1052 struct ctl_table_root *root;
1053 struct ctl_table_set *set;
1054 struct ctl_table *entry;
1055 struct ctl_dir *dir;
1058 spin_lock(&sysctl_lock);
1059 root = (*pentry)->data;
1060 set = lookup_header_set(root);
1061 dir = xlate_dir(set, (*phead)->parent);
1065 const char *procname = (*pentry)->procname;
1067 entry = find_entry(&head, dir, procname, strlen(procname));
1069 if (entry && use_table(head)) {
1070 unuse_table(*phead);
1077 spin_unlock(&sysctl_lock);
1081 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1083 struct va_format vaf;
1086 va_start(args, fmt);
1090 pr_err("sysctl table check failed: %s/%s %pV\n",
1091 path, table->procname, &vaf);
1097 static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1101 if ((table->proc_handler == proc_douintvec) ||
1102 (table->proc_handler == proc_douintvec_minmax)) {
1103 if (table->maxlen != sizeof(unsigned int))
1104 err |= sysctl_err(path, table, "array not allowed");
1107 if (table->proc_handler == proc_dou8vec_minmax) {
1108 if (table->maxlen != sizeof(u8))
1109 err |= sysctl_err(path, table, "array not allowed");
1112 if (table->proc_handler == proc_dobool) {
1113 if (table->maxlen != sizeof(bool))
1114 err |= sysctl_err(path, table, "array not allowed");
1120 static int sysctl_check_table(const char *path, struct ctl_table_header *header)
1122 struct ctl_table *entry;
1124 list_for_each_table_entry(entry, header) {
1125 if ((entry->proc_handler == proc_dostring) ||
1126 (entry->proc_handler == proc_dobool) ||
1127 (entry->proc_handler == proc_dointvec) ||
1128 (entry->proc_handler == proc_douintvec) ||
1129 (entry->proc_handler == proc_douintvec_minmax) ||
1130 (entry->proc_handler == proc_dointvec_minmax) ||
1131 (entry->proc_handler == proc_dou8vec_minmax) ||
1132 (entry->proc_handler == proc_dointvec_jiffies) ||
1133 (entry->proc_handler == proc_dointvec_userhz_jiffies) ||
1134 (entry->proc_handler == proc_dointvec_ms_jiffies) ||
1135 (entry->proc_handler == proc_doulongvec_minmax) ||
1136 (entry->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1138 err |= sysctl_err(path, entry, "No data");
1140 err |= sysctl_err(path, entry, "No maxlen");
1142 err |= sysctl_check_table_array(path, entry);
1144 if (!entry->proc_handler)
1145 err |= sysctl_err(path, entry, "No proc_handler");
1147 if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode)
1148 err |= sysctl_err(path, entry, "bogus .mode 0%o",
1154 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table_header *head)
1156 struct ctl_table *link_table, *entry, *link;
1157 struct ctl_table_header *links;
1158 struct ctl_node *node;
1160 int nr_entries, name_bytes;
1164 list_for_each_table_entry(entry, head) {
1166 name_bytes += strlen(entry->procname) + 1;
1169 links = kzalloc(sizeof(struct ctl_table_header) +
1170 sizeof(struct ctl_node)*nr_entries +
1171 sizeof(struct ctl_table)*(nr_entries + 1) +
1178 node = (struct ctl_node *)(links + 1);
1179 link_table = (struct ctl_table *)(node + nr_entries);
1180 link_name = (char *)&link_table[nr_entries + 1];
1183 list_for_each_table_entry(entry, head) {
1184 int len = strlen(entry->procname) + 1;
1185 memcpy(link_name, entry->procname, len);
1186 link->procname = link_name;
1187 link->mode = S_IFLNK|S_IRWXUGO;
1188 link->data = head->root;
1192 init_header(links, dir->header.root, dir->header.set, node, link_table,
1193 head->ctl_table_size);
1194 links->nreg = nr_entries;
1199 static bool get_links(struct ctl_dir *dir,
1200 struct ctl_table_header *header,
1201 struct ctl_table_root *link_root)
1203 struct ctl_table_header *tmp_head;
1204 struct ctl_table *entry, *link;
1206 if (header->ctl_table_size == 0 ||
1207 sysctl_is_perm_empty_ctl_table(header->ctl_table))
1210 /* Are there links available for every entry in table? */
1211 list_for_each_table_entry(entry, header) {
1212 const char *procname = entry->procname;
1213 link = find_entry(&tmp_head, dir, procname, strlen(procname));
1216 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1218 if (S_ISLNK(link->mode) && (link->data == link_root))
1223 /* The checks passed. Increase the registration count on the links */
1224 list_for_each_table_entry(entry, header) {
1225 const char *procname = entry->procname;
1226 link = find_entry(&tmp_head, dir, procname, strlen(procname));
1232 static int insert_links(struct ctl_table_header *head)
1234 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1235 struct ctl_dir *core_parent;
1236 struct ctl_table_header *links;
1239 if (head->set == root_set)
1242 core_parent = xlate_dir(root_set, head->parent);
1243 if (IS_ERR(core_parent))
1246 if (get_links(core_parent, head, head->root))
1249 core_parent->header.nreg++;
1250 spin_unlock(&sysctl_lock);
1252 links = new_links(core_parent, head);
1254 spin_lock(&sysctl_lock);
1260 if (get_links(core_parent, head, head->root)) {
1265 err = insert_header(core_parent, links);
1269 drop_sysctl_table(&core_parent->header);
1273 /* Find the directory for the ctl_table. If one is not found create it. */
1274 static struct ctl_dir *sysctl_mkdir_p(struct ctl_dir *dir, const char *path)
1276 const char *name, *nextname;
1278 for (name = path; name; name = nextname) {
1280 nextname = strchr(name, '/');
1282 namelen = nextname - name;
1285 namelen = strlen(name);
1291 * namelen ensures if name is "foo/bar/yay" only foo is
1292 * registered first. We traverse as if using mkdir -p and
1293 * return a ctl_dir for the last directory entry.
1295 dir = get_subdir(dir, name, namelen);
1303 * __register_sysctl_table - register a leaf sysctl table
1304 * @set: Sysctl tree to register on
1305 * @path: The path to the directory the sysctl table is in.
1306 * @table: the top-level table structure without any child. This table
1307 * should not be free'd after registration. So it should not be
1308 * used on stack. It can either be a global or dynamically allocated
1309 * by the caller and free'd later after sysctl unregistration.
1310 * @table_size : The number of elements in table
1312 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1313 * array. A completely 0 filled entry terminates the table.
1315 * The members of the &struct ctl_table structure are used as follows:
1317 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1318 * enter a sysctl file
1320 * data - a pointer to data for use by proc_handler
1322 * maxlen - the maximum size in bytes of the data
1324 * mode - the file permissions for the /proc/sys file
1326 * child - must be %NULL.
1328 * proc_handler - the text handler routine (described below)
1330 * extra1, extra2 - extra pointers usable by the proc handler routines
1331 * XXX: we should eventually modify these to use long min / max [0]
1332 * [0] https://lkml.kernel.org/87zgpte9o4.fsf@email.froward.int.ebiederm.org
1334 * Leaf nodes in the sysctl tree will be represented by a single file
1335 * under /proc; non-leaf nodes (where child is not NULL) are not allowed,
1336 * sysctl_check_table() verifies this.
1338 * There must be a proc_handler routine for any terminal nodes.
1339 * Several default handlers are available to cover common cases -
1341 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1342 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1343 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1345 * It is the handler's job to read the input buffer from user memory
1346 * and process it. The handler should return 0 on success.
1348 * This routine returns %NULL on a failure to register, and a pointer
1349 * to the table header on success.
1351 struct ctl_table_header *__register_sysctl_table(
1352 struct ctl_table_set *set,
1353 const char *path, struct ctl_table *table, size_t table_size)
1355 struct ctl_table_root *root = set->dir.header.root;
1356 struct ctl_table_header *header;
1357 struct ctl_dir *dir;
1358 struct ctl_node *node;
1360 header = kzalloc(sizeof(struct ctl_table_header) +
1361 sizeof(struct ctl_node)*table_size, GFP_KERNEL_ACCOUNT);
1365 node = (struct ctl_node *)(header + 1);
1366 init_header(header, root, set, node, table, table_size);
1367 if (sysctl_check_table(path, header))
1370 spin_lock(&sysctl_lock);
1372 /* Reference moved down the directory tree get_subdir */
1374 spin_unlock(&sysctl_lock);
1376 dir = sysctl_mkdir_p(dir, path);
1379 spin_lock(&sysctl_lock);
1380 if (insert_header(dir, header))
1381 goto fail_put_dir_locked;
1383 drop_sysctl_table(&dir->header);
1384 spin_unlock(&sysctl_lock);
1388 fail_put_dir_locked:
1389 drop_sysctl_table(&dir->header);
1390 spin_unlock(&sysctl_lock);
1397 * register_sysctl_sz - register a sysctl table
1398 * @path: The path to the directory the sysctl table is in. If the path
1399 * doesn't exist we will create it for you.
1400 * @table: the table structure. The calller must ensure the life of the @table
1401 * will be kept during the lifetime use of the syctl. It must not be freed
1402 * until unregister_sysctl_table() is called with the given returned table
1403 * with this registration. If your code is non modular then you don't need
1404 * to call unregister_sysctl_table() and can instead use something like
1405 * register_sysctl_init() which does not care for the result of the syctl
1407 * @table_size: The number of elements in table.
1409 * Register a sysctl table. @table should be a filled in ctl_table
1410 * array. A completely 0 filled entry terminates the table.
1412 * See __register_sysctl_table for more details.
1414 struct ctl_table_header *register_sysctl_sz(const char *path, struct ctl_table *table,
1417 return __register_sysctl_table(&sysctl_table_root.default_set,
1418 path, table, table_size);
1420 EXPORT_SYMBOL(register_sysctl_sz);
1423 * __register_sysctl_init() - register sysctl table to path
1424 * @path: path name for sysctl base. If that path doesn't exist we will create
1426 * @table: This is the sysctl table that needs to be registered to the path.
1427 * The caller must ensure the life of the @table will be kept during the
1428 * lifetime use of the sysctl.
1429 * @table_name: The name of sysctl table, only used for log printing when
1430 * registration fails
1431 * @table_size: The number of elements in table
1433 * The sysctl interface is used by userspace to query or modify at runtime
1434 * a predefined value set on a variable. These variables however have default
1435 * values pre-set. Code which depends on these variables will always work even
1436 * if register_sysctl() fails. If register_sysctl() fails you'd just loose the
1437 * ability to query or modify the sysctls dynamically at run time. Chances of
1438 * register_sysctl() failing on init are extremely low, and so for both reasons
1439 * this function does not return any error as it is used by initialization code.
1441 * Context: if your base directory does not exist it will be created for you.
1443 void __init __register_sysctl_init(const char *path, struct ctl_table *table,
1444 const char *table_name, size_t table_size)
1446 struct ctl_table_header *hdr = register_sysctl_sz(path, table, table_size);
1448 if (unlikely(!hdr)) {
1449 pr_err("failed when register_sysctl_sz %s to %s\n", table_name, path);
1452 kmemleak_not_leak(hdr);
1455 static void put_links(struct ctl_table_header *header)
1457 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1458 struct ctl_table_root *root = header->root;
1459 struct ctl_dir *parent = header->parent;
1460 struct ctl_dir *core_parent;
1461 struct ctl_table *entry;
1463 if (header->set == root_set)
1466 core_parent = xlate_dir(root_set, parent);
1467 if (IS_ERR(core_parent))
1470 list_for_each_table_entry(entry, header) {
1471 struct ctl_table_header *link_head;
1472 struct ctl_table *link;
1473 const char *name = entry->procname;
1475 link = find_entry(&link_head, core_parent, name, strlen(name));
1477 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1478 (S_ISLNK(link->mode) && (link->data == root)))) {
1479 drop_sysctl_table(link_head);
1482 pr_err("sysctl link missing during unregister: ");
1483 sysctl_print_dir(parent);
1484 pr_cont("%s\n", name);
1489 static void drop_sysctl_table(struct ctl_table_header *header)
1491 struct ctl_dir *parent = header->parent;
1498 start_unregistering(header);
1501 if (!--header->count)
1502 kfree_rcu(header, rcu);
1505 drop_sysctl_table(&parent->header);
1509 * unregister_sysctl_table - unregister a sysctl table hierarchy
1510 * @header: the header returned from register_sysctl or __register_sysctl_table
1512 * Unregisters the sysctl table and all children. proc entries may not
1513 * actually be removed until they are no longer used by anyone.
1515 void unregister_sysctl_table(struct ctl_table_header * header)
1522 spin_lock(&sysctl_lock);
1523 drop_sysctl_table(header);
1524 spin_unlock(&sysctl_lock);
1526 EXPORT_SYMBOL(unregister_sysctl_table);
1528 void setup_sysctl_set(struct ctl_table_set *set,
1529 struct ctl_table_root *root,
1530 int (*is_seen)(struct ctl_table_set *))
1532 memset(set, 0, sizeof(*set));
1533 set->is_seen = is_seen;
1534 init_header(&set->dir.header, root, set, NULL, root_table, 1);
1537 void retire_sysctl_set(struct ctl_table_set *set)
1539 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1542 int __init proc_sys_init(void)
1544 struct proc_dir_entry *proc_sys_root;
1546 proc_sys_root = proc_mkdir("sys", NULL);
1547 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1548 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1549 proc_sys_root->nlink = 0;
1551 return sysctl_init_bases();
1554 struct sysctl_alias {
1555 const char *kernel_param;
1556 const char *sysctl_param;
1560 * Historically some settings had both sysctl and a command line parameter.
1561 * With the generic sysctl. parameter support, we can handle them at a single
1562 * place and only keep the historical name for compatibility. This is not meant
1563 * to add brand new aliases. When adding existing aliases, consider whether
1564 * the possibly different moment of changing the value (e.g. from early_param
1565 * to the moment do_sysctl_args() is called) is an issue for the specific
1568 static const struct sysctl_alias sysctl_aliases[] = {
1569 {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1570 {"hung_task_panic", "kernel.hung_task_panic" },
1571 {"numa_zonelist_order", "vm.numa_zonelist_order" },
1572 {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1576 static const char *sysctl_find_alias(char *param)
1578 const struct sysctl_alias *alias;
1580 for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1581 if (strcmp(alias->kernel_param, param) == 0)
1582 return alias->sysctl_param;
1588 bool sysctl_is_alias(char *param)
1590 const char *alias = sysctl_find_alias(param);
1592 return alias != NULL;
1595 /* Set sysctl value passed on kernel command line. */
1596 static int process_sysctl_arg(char *param, char *val,
1597 const char *unused, void *arg)
1600 struct vfsmount **proc_mnt = arg;
1601 struct file_system_type *proc_fs_type;
1608 if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1609 param += sizeof("sysctl") - 1;
1611 if (param[0] != '/' && param[0] != '.')
1616 param = (char *) sysctl_find_alias(param);
1628 * To set sysctl options, we use a temporary mount of proc, look up the
1629 * respective sys/ file and write to it. To avoid mounting it when no
1630 * options were given, we mount it only when the first sysctl option is
1631 * found. Why not a persistent mount? There are problems with a
1632 * persistent mount of proc in that it forces userspace not to use any
1633 * proc mount options.
1636 proc_fs_type = get_fs_type("proc");
1637 if (!proc_fs_type) {
1638 pr_err("Failed to find procfs to set sysctl from command line\n");
1641 *proc_mnt = kern_mount(proc_fs_type);
1642 put_filesystem(proc_fs_type);
1643 if (IS_ERR(*proc_mnt)) {
1644 pr_err("Failed to mount procfs to set sysctl from command line\n");
1649 path = kasprintf(GFP_KERNEL, "sys/%s", param);
1651 panic("%s: Failed to allocate path for %s\n", __func__, param);
1652 strreplace(path, '.', '/');
1654 file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1656 err = PTR_ERR(file);
1658 pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1660 else if (err == -EACCES)
1661 pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1664 pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1668 wret = kernel_write(file, val, len, &pos);
1672 pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1675 pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1676 ERR_PTR(err), param, val);
1677 } else if (wret != len) {
1678 pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1679 wret, len, path, param, val);
1682 err = filp_close(file, NULL);
1684 pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1685 ERR_PTR(err), param, val);
1691 void do_sysctl_args(void)
1694 struct vfsmount *proc_mnt = NULL;
1696 command_line = kstrdup(saved_command_line, GFP_KERNEL);
1698 panic("%s: Failed to allocate copy of command line\n", __func__);
1700 parse_args("Setting sysctl args", command_line,
1701 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1704 kern_unmount(proc_mnt);
1706 kfree(command_line);