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, table) \
23 for ((entry) = (table); (entry)->procname; (entry)++)
25 static const struct dentry_operations proc_sys_dentry_operations;
26 static const struct file_operations proc_sys_file_operations;
27 static const struct inode_operations proc_sys_inode_operations;
28 static const struct file_operations proc_sys_dir_file_operations;
29 static const struct inode_operations proc_sys_dir_operations;
31 /* Support for permanently empty directories */
33 struct ctl_table sysctl_mount_point[] = {
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 static bool is_empty_dir(struct ctl_table_header *head)
53 return head->ctl_table[0].child == sysctl_mount_point;
56 static void set_empty_dir(struct ctl_dir *dir)
58 dir->header.ctl_table[0].child = sysctl_mount_point;
61 static void clear_empty_dir(struct ctl_dir *dir)
64 dir->header.ctl_table[0].child = NULL;
67 void proc_sys_poll_notify(struct ctl_table_poll *poll)
72 atomic_inc(&poll->event);
73 wake_up_interruptible(&poll->wait);
76 static struct ctl_table root_table[] = {
79 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
83 static struct ctl_table_root sysctl_table_root = {
84 .default_set.dir.header = {
87 .ctl_table = root_table }},
88 .ctl_table_arg = root_table,
89 .root = &sysctl_table_root,
90 .set = &sysctl_table_root.default_set,
94 static DEFINE_SPINLOCK(sysctl_lock);
96 static void drop_sysctl_table(struct ctl_table_header *header);
97 static int sysctl_follow_link(struct ctl_table_header **phead,
98 struct ctl_table **pentry);
99 static int insert_links(struct ctl_table_header *head);
100 static void put_links(struct ctl_table_header *header);
102 static void sysctl_print_dir(struct ctl_dir *dir)
104 if (dir->header.parent)
105 sysctl_print_dir(dir->header.parent);
106 pr_cont("%s/", dir->header.ctl_table[0].procname);
109 static int namecmp(const char *name1, int len1, const char *name2, int len2)
113 cmp = memcmp(name1, name2, min(len1, len2));
119 /* Called under sysctl_lock */
120 static struct ctl_table *find_entry(struct ctl_table_header **phead,
121 struct ctl_dir *dir, const char *name, int namelen)
123 struct ctl_table_header *head;
124 struct ctl_table *entry;
125 struct rb_node *node = dir->root.rb_node;
129 struct ctl_node *ctl_node;
130 const char *procname;
133 ctl_node = rb_entry(node, struct ctl_node, node);
134 head = ctl_node->header;
135 entry = &head->ctl_table[ctl_node - head->node];
136 procname = entry->procname;
138 cmp = namecmp(name, namelen, procname, strlen(procname));
140 node = node->rb_left;
142 node = node->rb_right;
151 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
153 struct rb_node *node = &head->node[entry - head->ctl_table].node;
154 struct rb_node **p = &head->parent->root.rb_node;
155 struct rb_node *parent = NULL;
156 const char *name = entry->procname;
157 int namelen = strlen(name);
160 struct ctl_table_header *parent_head;
161 struct ctl_table *parent_entry;
162 struct ctl_node *parent_node;
163 const char *parent_name;
167 parent_node = rb_entry(parent, struct ctl_node, node);
168 parent_head = parent_node->header;
169 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
170 parent_name = parent_entry->procname;
172 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
178 pr_err("sysctl duplicate entry: ");
179 sysctl_print_dir(head->parent);
180 pr_cont("%s\n", entry->procname);
185 rb_link_node(node, parent, p);
186 rb_insert_color(node, &head->parent->root);
190 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
192 struct rb_node *node = &head->node[entry - head->ctl_table].node;
194 rb_erase(node, &head->parent->root);
197 static void init_header(struct ctl_table_header *head,
198 struct ctl_table_root *root, struct ctl_table_set *set,
199 struct ctl_node *node, struct ctl_table *table)
201 head->ctl_table = table;
202 head->ctl_table_arg = table;
206 head->unregistering = NULL;
211 INIT_HLIST_HEAD(&head->inodes);
213 struct ctl_table *entry;
215 list_for_each_table_entry(entry, table) {
222 static void erase_header(struct ctl_table_header *head)
224 struct ctl_table *entry;
226 list_for_each_table_entry(entry, head->ctl_table)
227 erase_entry(head, entry);
230 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
232 struct ctl_table *entry;
235 /* Is this a permanently empty directory? */
236 if (is_empty_dir(&dir->header))
239 /* Am I creating a permanently empty directory? */
240 if (header->ctl_table == sysctl_mount_point) {
241 if (!RB_EMPTY_ROOT(&dir->root))
247 header->parent = dir;
248 err = insert_links(header);
251 list_for_each_table_entry(entry, header->ctl_table) {
252 err = insert_entry(header, entry);
258 erase_header(header);
261 if (header->ctl_table == sysctl_mount_point)
262 clear_empty_dir(dir);
263 header->parent = NULL;
264 drop_sysctl_table(&dir->header);
268 /* called under sysctl_lock */
269 static int use_table(struct ctl_table_header *p)
271 if (unlikely(p->unregistering))
277 /* called under sysctl_lock */
278 static void unuse_table(struct ctl_table_header *p)
281 if (unlikely(p->unregistering))
282 complete(p->unregistering);
285 static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
287 proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
290 /* called under sysctl_lock, will reacquire if has to wait */
291 static void start_unregistering(struct ctl_table_header *p)
294 * if p->used is 0, nobody will ever touch that entry again;
295 * we'll eliminate all paths to it before dropping sysctl_lock
297 if (unlikely(p->used)) {
298 struct completion wait;
299 init_completion(&wait);
300 p->unregistering = &wait;
301 spin_unlock(&sysctl_lock);
302 wait_for_completion(&wait);
304 /* anything non-NULL; we'll never dereference it */
305 p->unregistering = ERR_PTR(-EINVAL);
306 spin_unlock(&sysctl_lock);
309 * Invalidate dentries for unregistered sysctls: namespaced sysctls
310 * can have duplicate names and contaminate dcache very badly.
312 proc_sys_invalidate_dcache(p);
314 * do not remove from the list until nobody holds it; walking the
315 * list in do_sysctl() relies on that.
317 spin_lock(&sysctl_lock);
321 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
324 spin_lock(&sysctl_lock);
325 if (!use_table(head))
326 head = ERR_PTR(-ENOENT);
327 spin_unlock(&sysctl_lock);
331 static void sysctl_head_finish(struct ctl_table_header *head)
335 spin_lock(&sysctl_lock);
337 spin_unlock(&sysctl_lock);
340 static struct ctl_table_set *
341 lookup_header_set(struct ctl_table_root *root)
343 struct ctl_table_set *set = &root->default_set;
345 set = root->lookup(root);
349 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
351 const char *name, int namelen)
353 struct ctl_table_header *head;
354 struct ctl_table *entry;
356 spin_lock(&sysctl_lock);
357 entry = find_entry(&head, dir, name, namelen);
358 if (entry && use_table(head))
362 spin_unlock(&sysctl_lock);
366 static struct ctl_node *first_usable_entry(struct rb_node *node)
368 struct ctl_node *ctl_node;
370 for (;node; node = rb_next(node)) {
371 ctl_node = rb_entry(node, struct ctl_node, node);
372 if (use_table(ctl_node->header))
378 static void first_entry(struct ctl_dir *dir,
379 struct ctl_table_header **phead, struct ctl_table **pentry)
381 struct ctl_table_header *head = NULL;
382 struct ctl_table *entry = NULL;
383 struct ctl_node *ctl_node;
385 spin_lock(&sysctl_lock);
386 ctl_node = first_usable_entry(rb_first(&dir->root));
387 spin_unlock(&sysctl_lock);
389 head = ctl_node->header;
390 entry = &head->ctl_table[ctl_node - head->node];
396 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
398 struct ctl_table_header *head = *phead;
399 struct ctl_table *entry = *pentry;
400 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
402 spin_lock(&sysctl_lock);
405 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
406 spin_unlock(&sysctl_lock);
409 head = ctl_node->header;
410 entry = &head->ctl_table[ctl_node - head->node];
417 * sysctl_perm does NOT grant the superuser all rights automatically, because
418 * some sysctl variables are readonly even to root.
421 static int test_perm(int mode, int op)
423 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
425 else if (in_egroup_p(GLOBAL_ROOT_GID))
427 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
432 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
434 struct ctl_table_root *root = head->root;
437 if (root->permissions)
438 mode = root->permissions(head, table);
442 return test_perm(mode, op);
445 static struct inode *proc_sys_make_inode(struct super_block *sb,
446 struct ctl_table_header *head, struct ctl_table *table)
448 struct ctl_table_root *root = head->root;
450 struct proc_inode *ei;
452 inode = new_inode(sb);
454 return ERR_PTR(-ENOMEM);
456 inode->i_ino = get_next_ino();
460 spin_lock(&sysctl_lock);
461 if (unlikely(head->unregistering)) {
462 spin_unlock(&sysctl_lock);
464 return ERR_PTR(-ENOENT);
467 ei->sysctl_entry = table;
468 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
470 spin_unlock(&sysctl_lock);
472 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
473 inode->i_mode = table->mode;
474 if (!S_ISDIR(table->mode)) {
475 inode->i_mode |= S_IFREG;
476 inode->i_op = &proc_sys_inode_operations;
477 inode->i_fop = &proc_sys_file_operations;
479 inode->i_mode |= S_IFDIR;
480 inode->i_op = &proc_sys_dir_operations;
481 inode->i_fop = &proc_sys_dir_file_operations;
482 if (is_empty_dir(head))
483 make_empty_dir_inode(inode);
486 if (root->set_ownership)
487 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
489 inode->i_uid = GLOBAL_ROOT_UID;
490 inode->i_gid = GLOBAL_ROOT_GID;
496 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
498 spin_lock(&sysctl_lock);
499 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
501 kfree_rcu(head, rcu);
502 spin_unlock(&sysctl_lock);
505 static struct ctl_table_header *grab_header(struct inode *inode)
507 struct ctl_table_header *head = PROC_I(inode)->sysctl;
509 head = &sysctl_table_root.default_set.dir.header;
510 return sysctl_head_grab(head);
513 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
516 struct ctl_table_header *head = grab_header(dir);
517 struct ctl_table_header *h = NULL;
518 const struct qstr *name = &dentry->d_name;
521 struct dentry *err = ERR_PTR(-ENOENT);
522 struct ctl_dir *ctl_dir;
526 return ERR_CAST(head);
528 ctl_dir = container_of(head, struct ctl_dir, header);
530 p = lookup_entry(&h, ctl_dir, name->name, name->len);
534 if (S_ISLNK(p->mode)) {
535 ret = sysctl_follow_link(&h, &p);
541 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
543 err = ERR_CAST(inode);
547 d_set_d_op(dentry, &proc_sys_dentry_operations);
548 err = d_splice_alias(inode, dentry);
552 sysctl_head_finish(h);
553 sysctl_head_finish(head);
557 static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
560 struct inode *inode = file_inode(iocb->ki_filp);
561 struct ctl_table_header *head = grab_header(inode);
562 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
563 size_t count = iov_iter_count(iter);
568 return PTR_ERR(head);
571 * At this point we know that the sysctl was not unregistered
572 * and won't be until we finish.
575 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
578 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
580 if (!table->proc_handler)
583 /* don't even try if the size is too large */
585 if (count >= KMALLOC_MAX_SIZE)
587 kbuf = kvzalloc(count + 1, GFP_KERNEL);
593 if (!copy_from_iter_full(kbuf, count, iter))
598 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
603 /* careful: calling conventions are nasty here */
604 error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
610 if (copy_to_iter(kbuf, count, iter) < count)
618 sysctl_head_finish(head);
623 static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
625 return proc_sys_call_handler(iocb, iter, 0);
628 static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
630 return proc_sys_call_handler(iocb, iter, 1);
633 static int proc_sys_open(struct inode *inode, struct file *filp)
635 struct ctl_table_header *head = grab_header(inode);
636 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
638 /* sysctl was unregistered */
640 return PTR_ERR(head);
643 filp->private_data = proc_sys_poll_event(table->poll);
645 sysctl_head_finish(head);
650 static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
652 struct inode *inode = file_inode(filp);
653 struct ctl_table_header *head = grab_header(inode);
654 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
655 __poll_t ret = DEFAULT_POLLMASK;
658 /* sysctl was unregistered */
660 return EPOLLERR | EPOLLHUP;
662 if (!table->proc_handler)
668 event = (unsigned long)filp->private_data;
669 poll_wait(filp, &table->poll->wait, wait);
671 if (event != atomic_read(&table->poll->event)) {
672 filp->private_data = proc_sys_poll_event(table->poll);
673 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
677 sysctl_head_finish(head);
682 static bool proc_sys_fill_cache(struct file *file,
683 struct dir_context *ctx,
684 struct ctl_table_header *head,
685 struct ctl_table *table)
687 struct dentry *child, *dir = file->f_path.dentry;
691 unsigned type = DT_UNKNOWN;
693 qname.name = table->procname;
694 qname.len = strlen(table->procname);
695 qname.hash = full_name_hash(dir, qname.name, qname.len);
697 child = d_lookup(dir, &qname);
699 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
700 child = d_alloc_parallel(dir, &qname, &wq);
703 if (d_in_lookup(child)) {
705 inode = proc_sys_make_inode(dir->d_sb, head, table);
707 d_lookup_done(child);
711 d_set_d_op(child, &proc_sys_dentry_operations);
712 res = d_splice_alias(inode, child);
713 d_lookup_done(child);
724 inode = d_inode(child);
726 type = inode->i_mode >> 12;
728 return dir_emit(ctx, qname.name, qname.len, ino, type);
731 static bool proc_sys_link_fill_cache(struct file *file,
732 struct dir_context *ctx,
733 struct ctl_table_header *head,
734 struct ctl_table *table)
738 head = sysctl_head_grab(head);
742 /* It is not an error if we can not follow the link ignore it */
743 if (sysctl_follow_link(&head, &table))
746 ret = proc_sys_fill_cache(file, ctx, head, table);
748 sysctl_head_finish(head);
752 static int scan(struct ctl_table_header *head, struct ctl_table *table,
753 unsigned long *pos, struct file *file,
754 struct dir_context *ctx)
758 if ((*pos)++ < ctx->pos)
761 if (unlikely(S_ISLNK(table->mode)))
762 res = proc_sys_link_fill_cache(file, ctx, head, table);
764 res = proc_sys_fill_cache(file, ctx, head, table);
772 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
774 struct ctl_table_header *head = grab_header(file_inode(file));
775 struct ctl_table_header *h = NULL;
776 struct ctl_table *entry;
777 struct ctl_dir *ctl_dir;
781 return PTR_ERR(head);
783 ctl_dir = container_of(head, struct ctl_dir, header);
785 if (!dir_emit_dots(file, ctx))
790 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
791 if (!scan(h, entry, &pos, file, ctx)) {
792 sysctl_head_finish(h);
797 sysctl_head_finish(head);
801 static int proc_sys_permission(struct mnt_idmap *idmap,
802 struct inode *inode, int mask)
805 * sysctl entries that are not writeable,
806 * are _NOT_ writeable, capabilities or not.
808 struct ctl_table_header *head;
809 struct ctl_table *table;
812 /* Executable files are not allowed under /proc/sys/ */
813 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
816 head = grab_header(inode);
818 return PTR_ERR(head);
820 table = PROC_I(inode)->sysctl_entry;
821 if (!table) /* global root - r-xr-xr-x */
822 error = mask & MAY_WRITE ? -EACCES : 0;
823 else /* Use the permissions on the sysctl table entry */
824 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
826 sysctl_head_finish(head);
830 static int proc_sys_setattr(struct mnt_idmap *idmap,
831 struct dentry *dentry, struct iattr *attr)
833 struct inode *inode = d_inode(dentry);
836 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
839 error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
843 setattr_copy(&nop_mnt_idmap, inode, attr);
844 mark_inode_dirty(inode);
848 static int proc_sys_getattr(struct mnt_idmap *idmap,
849 const struct path *path, struct kstat *stat,
850 u32 request_mask, unsigned int query_flags)
852 struct inode *inode = d_inode(path->dentry);
853 struct ctl_table_header *head = grab_header(inode);
854 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
857 return PTR_ERR(head);
859 generic_fillattr(&nop_mnt_idmap, inode, stat);
861 stat->mode = (stat->mode & S_IFMT) | table->mode;
863 sysctl_head_finish(head);
867 static const struct file_operations proc_sys_file_operations = {
868 .open = proc_sys_open,
869 .poll = proc_sys_poll,
870 .read_iter = proc_sys_read,
871 .write_iter = proc_sys_write,
872 .splice_read = generic_file_splice_read,
873 .splice_write = iter_file_splice_write,
874 .llseek = default_llseek,
877 static const struct file_operations proc_sys_dir_file_operations = {
878 .read = generic_read_dir,
879 .iterate_shared = proc_sys_readdir,
880 .llseek = generic_file_llseek,
883 static const struct inode_operations proc_sys_inode_operations = {
884 .permission = proc_sys_permission,
885 .setattr = proc_sys_setattr,
886 .getattr = proc_sys_getattr,
889 static const struct inode_operations proc_sys_dir_operations = {
890 .lookup = proc_sys_lookup,
891 .permission = proc_sys_permission,
892 .setattr = proc_sys_setattr,
893 .getattr = proc_sys_getattr,
896 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
898 if (flags & LOOKUP_RCU)
900 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
903 static int proc_sys_delete(const struct dentry *dentry)
905 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
908 static int sysctl_is_seen(struct ctl_table_header *p)
910 struct ctl_table_set *set = p->set;
912 spin_lock(&sysctl_lock);
913 if (p->unregistering)
915 else if (!set->is_seen)
918 res = set->is_seen(set);
919 spin_unlock(&sysctl_lock);
923 static int proc_sys_compare(const struct dentry *dentry,
924 unsigned int len, const char *str, const struct qstr *name)
926 struct ctl_table_header *head;
929 /* Although proc doesn't have negative dentries, rcu-walk means
930 * that inode here can be NULL */
931 /* AV: can it, indeed? */
932 inode = d_inode_rcu(dentry);
935 if (name->len != len)
937 if (memcmp(name->name, str, len))
939 head = rcu_dereference(PROC_I(inode)->sysctl);
940 return !head || !sysctl_is_seen(head);
943 static const struct dentry_operations proc_sys_dentry_operations = {
944 .d_revalidate = proc_sys_revalidate,
945 .d_delete = proc_sys_delete,
946 .d_compare = proc_sys_compare,
949 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
950 const char *name, int namelen)
952 struct ctl_table_header *head;
953 struct ctl_table *entry;
955 entry = find_entry(&head, dir, name, namelen);
957 return ERR_PTR(-ENOENT);
958 if (!S_ISDIR(entry->mode))
959 return ERR_PTR(-ENOTDIR);
960 return container_of(head, struct ctl_dir, header);
963 static struct ctl_dir *new_dir(struct ctl_table_set *set,
964 const char *name, int namelen)
966 struct ctl_table *table;
968 struct ctl_node *node;
971 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
972 sizeof(struct ctl_table)*2 + namelen + 1,
977 node = (struct ctl_node *)(new + 1);
978 table = (struct ctl_table *)(node + 1);
979 new_name = (char *)(table + 2);
980 memcpy(new_name, name, namelen);
981 table[0].procname = new_name;
982 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
983 init_header(&new->header, set->dir.header.root, set, node, table);
989 * get_subdir - find or create a subdir with the specified name.
990 * @dir: Directory to create the subdirectory in
991 * @name: The name of the subdirectory to find or create
992 * @namelen: The length of name
994 * Takes a directory with an elevated reference count so we know that
995 * if we drop the lock the directory will not go away. Upon success
996 * the reference is moved from @dir to the returned subdirectory.
997 * Upon error an error code is returned and the reference on @dir is
1000 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
1001 const char *name, int namelen)
1003 struct ctl_table_set *set = dir->header.set;
1004 struct ctl_dir *subdir, *new = NULL;
1007 spin_lock(&sysctl_lock);
1008 subdir = find_subdir(dir, name, namelen);
1009 if (!IS_ERR(subdir))
1011 if (PTR_ERR(subdir) != -ENOENT)
1014 spin_unlock(&sysctl_lock);
1015 new = new_dir(set, name, namelen);
1016 spin_lock(&sysctl_lock);
1017 subdir = ERR_PTR(-ENOMEM);
1021 /* Was the subdir added while we dropped the lock? */
1022 subdir = find_subdir(dir, name, namelen);
1023 if (!IS_ERR(subdir))
1025 if (PTR_ERR(subdir) != -ENOENT)
1028 /* Nope. Use the our freshly made directory entry. */
1029 err = insert_header(dir, &new->header);
1030 subdir = ERR_PTR(err);
1035 subdir->header.nreg++;
1037 if (IS_ERR(subdir)) {
1038 pr_err("sysctl could not get directory: ");
1039 sysctl_print_dir(dir);
1040 pr_cont("%*.*s %ld\n", namelen, namelen, name,
1043 drop_sysctl_table(&dir->header);
1045 drop_sysctl_table(&new->header);
1046 spin_unlock(&sysctl_lock);
1050 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1052 struct ctl_dir *parent;
1053 const char *procname;
1054 if (!dir->header.parent)
1056 parent = xlate_dir(set, dir->header.parent);
1059 procname = dir->header.ctl_table[0].procname;
1060 return find_subdir(parent, procname, strlen(procname));
1063 static int sysctl_follow_link(struct ctl_table_header **phead,
1064 struct ctl_table **pentry)
1066 struct ctl_table_header *head;
1067 struct ctl_table_root *root;
1068 struct ctl_table_set *set;
1069 struct ctl_table *entry;
1070 struct ctl_dir *dir;
1073 spin_lock(&sysctl_lock);
1074 root = (*pentry)->data;
1075 set = lookup_header_set(root);
1076 dir = xlate_dir(set, (*phead)->parent);
1080 const char *procname = (*pentry)->procname;
1082 entry = find_entry(&head, dir, procname, strlen(procname));
1084 if (entry && use_table(head)) {
1085 unuse_table(*phead);
1092 spin_unlock(&sysctl_lock);
1096 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1098 struct va_format vaf;
1101 va_start(args, fmt);
1105 pr_err("sysctl table check failed: %s/%s %pV\n",
1106 path, table->procname, &vaf);
1112 static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1116 if ((table->proc_handler == proc_douintvec) ||
1117 (table->proc_handler == proc_douintvec_minmax)) {
1118 if (table->maxlen != sizeof(unsigned int))
1119 err |= sysctl_err(path, table, "array not allowed");
1122 if (table->proc_handler == proc_dou8vec_minmax) {
1123 if (table->maxlen != sizeof(u8))
1124 err |= sysctl_err(path, table, "array not allowed");
1127 if (table->proc_handler == proc_dobool) {
1128 if (table->maxlen != sizeof(bool))
1129 err |= sysctl_err(path, table, "array not allowed");
1135 static int sysctl_check_table(const char *path, struct ctl_table *table)
1137 struct ctl_table *entry;
1139 list_for_each_table_entry(entry, table) {
1141 err |= sysctl_err(path, entry, "Not a file");
1143 if ((entry->proc_handler == proc_dostring) ||
1144 (entry->proc_handler == proc_dobool) ||
1145 (entry->proc_handler == proc_dointvec) ||
1146 (entry->proc_handler == proc_douintvec) ||
1147 (entry->proc_handler == proc_douintvec_minmax) ||
1148 (entry->proc_handler == proc_dointvec_minmax) ||
1149 (entry->proc_handler == proc_dou8vec_minmax) ||
1150 (entry->proc_handler == proc_dointvec_jiffies) ||
1151 (entry->proc_handler == proc_dointvec_userhz_jiffies) ||
1152 (entry->proc_handler == proc_dointvec_ms_jiffies) ||
1153 (entry->proc_handler == proc_doulongvec_minmax) ||
1154 (entry->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1156 err |= sysctl_err(path, entry, "No data");
1158 err |= sysctl_err(path, entry, "No maxlen");
1160 err |= sysctl_check_table_array(path, entry);
1162 if (!entry->proc_handler)
1163 err |= sysctl_err(path, entry, "No proc_handler");
1165 if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode)
1166 err |= sysctl_err(path, entry, "bogus .mode 0%o",
1172 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1173 struct ctl_table_root *link_root)
1175 struct ctl_table *link_table, *entry, *link;
1176 struct ctl_table_header *links;
1177 struct ctl_node *node;
1179 int nr_entries, name_bytes;
1183 list_for_each_table_entry(entry, table) {
1185 name_bytes += strlen(entry->procname) + 1;
1188 links = kzalloc(sizeof(struct ctl_table_header) +
1189 sizeof(struct ctl_node)*nr_entries +
1190 sizeof(struct ctl_table)*(nr_entries + 1) +
1197 node = (struct ctl_node *)(links + 1);
1198 link_table = (struct ctl_table *)(node + nr_entries);
1199 link_name = (char *)&link_table[nr_entries + 1];
1202 list_for_each_table_entry(entry, table) {
1203 int len = strlen(entry->procname) + 1;
1204 memcpy(link_name, entry->procname, len);
1205 link->procname = link_name;
1206 link->mode = S_IFLNK|S_IRWXUGO;
1207 link->data = link_root;
1211 init_header(links, dir->header.root, dir->header.set, node, link_table);
1212 links->nreg = nr_entries;
1217 static bool get_links(struct ctl_dir *dir,
1218 struct ctl_table *table, struct ctl_table_root *link_root)
1220 struct ctl_table_header *head;
1221 struct ctl_table *entry, *link;
1223 /* Are there links available for every entry in table? */
1224 list_for_each_table_entry(entry, table) {
1225 const char *procname = entry->procname;
1226 link = find_entry(&head, dir, procname, strlen(procname));
1229 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1231 if (S_ISLNK(link->mode) && (link->data == link_root))
1236 /* The checks passed. Increase the registration count on the links */
1237 list_for_each_table_entry(entry, table) {
1238 const char *procname = entry->procname;
1239 link = find_entry(&head, dir, procname, strlen(procname));
1245 static int insert_links(struct ctl_table_header *head)
1247 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1248 struct ctl_dir *core_parent;
1249 struct ctl_table_header *links;
1252 if (head->set == root_set)
1255 core_parent = xlate_dir(root_set, head->parent);
1256 if (IS_ERR(core_parent))
1259 if (get_links(core_parent, head->ctl_table, head->root))
1262 core_parent->header.nreg++;
1263 spin_unlock(&sysctl_lock);
1265 links = new_links(core_parent, head->ctl_table, head->root);
1267 spin_lock(&sysctl_lock);
1273 if (get_links(core_parent, head->ctl_table, head->root)) {
1278 err = insert_header(core_parent, links);
1282 drop_sysctl_table(&core_parent->header);
1287 * __register_sysctl_table - register a leaf sysctl table
1288 * @set: Sysctl tree to register on
1289 * @path: The path to the directory the sysctl table is in.
1290 * @table: the top-level table structure
1292 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1293 * array. A completely 0 filled entry terminates the table.
1295 * The members of the &struct ctl_table structure are used as follows:
1297 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1298 * enter a sysctl file
1300 * data - a pointer to data for use by proc_handler
1302 * maxlen - the maximum size in bytes of the data
1304 * mode - the file permissions for the /proc/sys file
1306 * child - must be %NULL.
1308 * proc_handler - the text handler routine (described below)
1310 * extra1, extra2 - extra pointers usable by the proc handler routines
1312 * Leaf nodes in the sysctl tree will be represented by a single file
1313 * under /proc; non-leaf nodes will be represented by directories.
1315 * There must be a proc_handler routine for any terminal nodes.
1316 * Several default handlers are available to cover common cases -
1318 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1319 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1320 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1322 * It is the handler's job to read the input buffer from user memory
1323 * and process it. The handler should return 0 on success.
1325 * This routine returns %NULL on a failure to register, and a pointer
1326 * to the table header on success.
1328 struct ctl_table_header *__register_sysctl_table(
1329 struct ctl_table_set *set,
1330 const char *path, struct ctl_table *table)
1332 struct ctl_table_root *root = set->dir.header.root;
1333 struct ctl_table_header *header;
1334 const char *name, *nextname;
1335 struct ctl_dir *dir;
1336 struct ctl_table *entry;
1337 struct ctl_node *node;
1340 list_for_each_table_entry(entry, table)
1343 header = kzalloc(sizeof(struct ctl_table_header) +
1344 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL_ACCOUNT);
1348 node = (struct ctl_node *)(header + 1);
1349 init_header(header, root, set, node, table);
1350 if (sysctl_check_table(path, table))
1353 spin_lock(&sysctl_lock);
1355 /* Reference moved down the diretory tree get_subdir */
1357 spin_unlock(&sysctl_lock);
1359 /* Find the directory for the ctl_table */
1360 for (name = path; name; name = nextname) {
1362 nextname = strchr(name, '/');
1364 namelen = nextname - name;
1367 namelen = strlen(name);
1372 dir = get_subdir(dir, name, namelen);
1377 spin_lock(&sysctl_lock);
1378 if (insert_header(dir, header))
1379 goto fail_put_dir_locked;
1381 drop_sysctl_table(&dir->header);
1382 spin_unlock(&sysctl_lock);
1386 fail_put_dir_locked:
1387 drop_sysctl_table(&dir->header);
1388 spin_unlock(&sysctl_lock);
1396 * register_sysctl - register a sysctl table
1397 * @path: The path to the directory the sysctl table is in.
1398 * @table: the table structure
1400 * Register a sysctl table. @table should be a filled in ctl_table
1401 * array. A completely 0 filled entry terminates the table.
1403 * See __register_sysctl_table for more details.
1405 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1407 return __register_sysctl_table(&sysctl_table_root.default_set,
1410 EXPORT_SYMBOL(register_sysctl);
1413 * __register_sysctl_init() - register sysctl table to path
1414 * @path: path name for sysctl base
1415 * @table: This is the sysctl table that needs to be registered to the path
1416 * @table_name: The name of sysctl table, only used for log printing when
1417 * registration fails
1419 * The sysctl interface is used by userspace to query or modify at runtime
1420 * a predefined value set on a variable. These variables however have default
1421 * values pre-set. Code which depends on these variables will always work even
1422 * if register_sysctl() fails. If register_sysctl() fails you'd just loose the
1423 * ability to query or modify the sysctls dynamically at run time. Chances of
1424 * register_sysctl() failing on init are extremely low, and so for both reasons
1425 * this function does not return any error as it is used by initialization code.
1427 * Context: Can only be called after your respective sysctl base path has been
1428 * registered. So for instance, most base directories are registered early on
1429 * init before init levels are processed through proc_sys_init() and
1430 * sysctl_init_bases().
1432 void __init __register_sysctl_init(const char *path, struct ctl_table *table,
1433 const char *table_name)
1435 struct ctl_table_header *hdr = register_sysctl(path, table);
1437 if (unlikely(!hdr)) {
1438 pr_err("failed when register_sysctl %s to %s\n", table_name, path);
1441 kmemleak_not_leak(hdr);
1444 static char *append_path(const char *path, char *pos, const char *name)
1447 namelen = strlen(name);
1448 if (((pos - path) + namelen + 2) >= PATH_MAX)
1450 memcpy(pos, name, namelen);
1452 pos[namelen + 1] = '\0';
1457 static int count_subheaders(struct ctl_table *table)
1460 int nr_subheaders = 0;
1461 struct ctl_table *entry;
1463 /* special case: no directory and empty directory */
1464 if (!table || !table->procname)
1467 list_for_each_table_entry(entry, table) {
1469 nr_subheaders += count_subheaders(entry->child);
1473 return nr_subheaders + has_files;
1476 static int register_leaf_sysctl_tables(const char *path, char *pos,
1477 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1478 struct ctl_table *table)
1480 struct ctl_table *ctl_table_arg = NULL;
1481 struct ctl_table *entry, *files;
1486 list_for_each_table_entry(entry, table) {
1494 /* If there are mixed files and directories we need a new table */
1495 if (nr_dirs && nr_files) {
1496 struct ctl_table *new;
1497 files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1502 ctl_table_arg = files;
1505 list_for_each_table_entry(entry, table) {
1513 /* Register everything except a directory full of subdirectories */
1514 if (nr_files || !nr_dirs) {
1515 struct ctl_table_header *header;
1516 header = __register_sysctl_table(set, path, files);
1518 kfree(ctl_table_arg);
1522 /* Remember if we need to free the file table */
1523 header->ctl_table_arg = ctl_table_arg;
1524 **subheader = header;
1528 /* Recurse into the subdirectories. */
1529 list_for_each_table_entry(entry, table) {
1535 err = -ENAMETOOLONG;
1536 child_pos = append_path(path, pos, entry->procname);
1540 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1548 /* On failure our caller will unregister all registered subheaders */
1553 * __register_sysctl_paths - register a sysctl table hierarchy
1554 * @set: Sysctl tree to register on
1555 * @path: The path to the directory the sysctl table is in.
1556 * @table: the top-level table structure
1558 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1559 * array. A completely 0 filled entry terminates the table.
1561 * See __register_sysctl_table for more details.
1563 struct ctl_table_header *__register_sysctl_paths(
1564 struct ctl_table_set *set,
1565 const struct ctl_path *path, struct ctl_table *table)
1567 struct ctl_table *ctl_table_arg = table;
1568 int nr_subheaders = count_subheaders(table);
1569 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1570 const struct ctl_path *component;
1571 char *new_path, *pos;
1573 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1578 for (component = path; component->procname; component++) {
1579 pos = append_path(new_path, pos, component->procname);
1583 while (table->procname && table->child && !table[1].procname) {
1584 pos = append_path(new_path, pos, table->procname);
1587 table = table->child;
1589 if (nr_subheaders == 1) {
1590 header = __register_sysctl_table(set, new_path, table);
1592 header->ctl_table_arg = ctl_table_arg;
1594 header = kzalloc(sizeof(*header) +
1595 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1599 subheaders = (struct ctl_table_header **) (header + 1);
1600 subheader = subheaders;
1601 header->ctl_table_arg = ctl_table_arg;
1603 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1605 goto err_register_leaves;
1612 err_register_leaves:
1613 while (subheader > subheaders) {
1614 struct ctl_table_header *subh = *(--subheader);
1615 struct ctl_table *table = subh->ctl_table_arg;
1616 unregister_sysctl_table(subh);
1625 * register_sysctl_paths - register a sysctl table hierarchy
1626 * @path: The path to the directory the sysctl table is in.
1627 * @table: the top-level table structure
1629 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1630 * array. A completely 0 filled entry terminates the table.
1632 * See __register_sysctl_paths for more details.
1634 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1635 struct ctl_table *table)
1637 return __register_sysctl_paths(&sysctl_table_root.default_set,
1640 EXPORT_SYMBOL(register_sysctl_paths);
1643 * register_sysctl_table - register a sysctl table hierarchy
1644 * @table: the top-level table structure
1646 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1647 * array. A completely 0 filled entry terminates the table.
1649 * See register_sysctl_paths for more details.
1651 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1653 static const struct ctl_path null_path[] = { {} };
1655 return register_sysctl_paths(null_path, table);
1657 EXPORT_SYMBOL(register_sysctl_table);
1659 int __register_sysctl_base(struct ctl_table *base_table)
1661 struct ctl_table_header *hdr;
1663 hdr = register_sysctl_table(base_table);
1664 kmemleak_not_leak(hdr);
1668 static void put_links(struct ctl_table_header *header)
1670 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1671 struct ctl_table_root *root = header->root;
1672 struct ctl_dir *parent = header->parent;
1673 struct ctl_dir *core_parent;
1674 struct ctl_table *entry;
1676 if (header->set == root_set)
1679 core_parent = xlate_dir(root_set, parent);
1680 if (IS_ERR(core_parent))
1683 list_for_each_table_entry(entry, header->ctl_table) {
1684 struct ctl_table_header *link_head;
1685 struct ctl_table *link;
1686 const char *name = entry->procname;
1688 link = find_entry(&link_head, core_parent, name, strlen(name));
1690 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1691 (S_ISLNK(link->mode) && (link->data == root)))) {
1692 drop_sysctl_table(link_head);
1695 pr_err("sysctl link missing during unregister: ");
1696 sysctl_print_dir(parent);
1697 pr_cont("%s\n", name);
1702 static void drop_sysctl_table(struct ctl_table_header *header)
1704 struct ctl_dir *parent = header->parent;
1711 start_unregistering(header);
1714 if (!--header->count)
1715 kfree_rcu(header, rcu);
1718 drop_sysctl_table(&parent->header);
1722 * unregister_sysctl_table - unregister a sysctl table hierarchy
1723 * @header: the header returned from register_sysctl_table
1725 * Unregisters the sysctl table and all children. proc entries may not
1726 * actually be removed until they are no longer used by anyone.
1728 void unregister_sysctl_table(struct ctl_table_header * header)
1736 nr_subheaders = count_subheaders(header->ctl_table_arg);
1737 if (unlikely(nr_subheaders > 1)) {
1738 struct ctl_table_header **subheaders;
1741 subheaders = (struct ctl_table_header **)(header + 1);
1742 for (i = nr_subheaders -1; i >= 0; i--) {
1743 struct ctl_table_header *subh = subheaders[i];
1744 struct ctl_table *table = subh->ctl_table_arg;
1745 unregister_sysctl_table(subh);
1752 spin_lock(&sysctl_lock);
1753 drop_sysctl_table(header);
1754 spin_unlock(&sysctl_lock);
1756 EXPORT_SYMBOL(unregister_sysctl_table);
1758 void setup_sysctl_set(struct ctl_table_set *set,
1759 struct ctl_table_root *root,
1760 int (*is_seen)(struct ctl_table_set *))
1762 memset(set, 0, sizeof(*set));
1763 set->is_seen = is_seen;
1764 init_header(&set->dir.header, root, set, NULL, root_table);
1767 void retire_sysctl_set(struct ctl_table_set *set)
1769 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1772 int __init proc_sys_init(void)
1774 struct proc_dir_entry *proc_sys_root;
1776 proc_sys_root = proc_mkdir("sys", NULL);
1777 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1778 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1779 proc_sys_root->nlink = 0;
1781 return sysctl_init_bases();
1784 struct sysctl_alias {
1785 const char *kernel_param;
1786 const char *sysctl_param;
1790 * Historically some settings had both sysctl and a command line parameter.
1791 * With the generic sysctl. parameter support, we can handle them at a single
1792 * place and only keep the historical name for compatibility. This is not meant
1793 * to add brand new aliases. When adding existing aliases, consider whether
1794 * the possibly different moment of changing the value (e.g. from early_param
1795 * to the moment do_sysctl_args() is called) is an issue for the specific
1798 static const struct sysctl_alias sysctl_aliases[] = {
1799 {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1800 {"hung_task_panic", "kernel.hung_task_panic" },
1801 {"numa_zonelist_order", "vm.numa_zonelist_order" },
1802 {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1803 {"softlockup_panic", "kernel.softlockup_panic" },
1807 static const char *sysctl_find_alias(char *param)
1809 const struct sysctl_alias *alias;
1811 for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1812 if (strcmp(alias->kernel_param, param) == 0)
1813 return alias->sysctl_param;
1819 /* Set sysctl value passed on kernel command line. */
1820 static int process_sysctl_arg(char *param, char *val,
1821 const char *unused, void *arg)
1824 struct vfsmount **proc_mnt = arg;
1825 struct file_system_type *proc_fs_type;
1832 if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1833 param += sizeof("sysctl") - 1;
1835 if (param[0] != '/' && param[0] != '.')
1840 param = (char *) sysctl_find_alias(param);
1852 * To set sysctl options, we use a temporary mount of proc, look up the
1853 * respective sys/ file and write to it. To avoid mounting it when no
1854 * options were given, we mount it only when the first sysctl option is
1855 * found. Why not a persistent mount? There are problems with a
1856 * persistent mount of proc in that it forces userspace not to use any
1857 * proc mount options.
1860 proc_fs_type = get_fs_type("proc");
1861 if (!proc_fs_type) {
1862 pr_err("Failed to find procfs to set sysctl from command line\n");
1865 *proc_mnt = kern_mount(proc_fs_type);
1866 put_filesystem(proc_fs_type);
1867 if (IS_ERR(*proc_mnt)) {
1868 pr_err("Failed to mount procfs to set sysctl from command line\n");
1873 path = kasprintf(GFP_KERNEL, "sys/%s", param);
1875 panic("%s: Failed to allocate path for %s\n", __func__, param);
1876 strreplace(path, '.', '/');
1878 file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1880 err = PTR_ERR(file);
1882 pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1884 else if (err == -EACCES)
1885 pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1888 pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1892 wret = kernel_write(file, val, len, &pos);
1896 pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1899 pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1900 ERR_PTR(err), param, val);
1901 } else if (wret != len) {
1902 pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1903 wret, len, path, param, val);
1906 err = filp_close(file, NULL);
1908 pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1909 ERR_PTR(err), param, val);
1915 void do_sysctl_args(void)
1918 struct vfsmount *proc_mnt = NULL;
1920 command_line = kstrdup(saved_command_line, GFP_KERNEL);
1922 panic("%s: Failed to allocate copy of command line\n", __func__);
1924 parse_args("Setting sysctl args", command_line,
1925 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1928 kern_unmount(proc_mnt);
1930 kfree(command_line);