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>
21 static const struct dentry_operations proc_sys_dentry_operations;
22 static const struct file_operations proc_sys_file_operations;
23 static const struct inode_operations proc_sys_inode_operations;
24 static const struct file_operations proc_sys_dir_file_operations;
25 static const struct inode_operations proc_sys_dir_operations;
27 /* shared constants to be used in various sysctls */
28 const int sysctl_vals[] = { 0, 1, INT_MAX };
29 EXPORT_SYMBOL(sysctl_vals);
31 /* Support for permanently empty directories */
33 struct ctl_table sysctl_mount_point[] = {
37 static bool is_empty_dir(struct ctl_table_header *head)
39 return head->ctl_table[0].child == sysctl_mount_point;
42 static void set_empty_dir(struct ctl_dir *dir)
44 dir->header.ctl_table[0].child = sysctl_mount_point;
47 static void clear_empty_dir(struct ctl_dir *dir)
50 dir->header.ctl_table[0].child = NULL;
53 void proc_sys_poll_notify(struct ctl_table_poll *poll)
58 atomic_inc(&poll->event);
59 wake_up_interruptible(&poll->wait);
62 static struct ctl_table root_table[] = {
65 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
69 static struct ctl_table_root sysctl_table_root = {
70 .default_set.dir.header = {
73 .ctl_table = root_table }},
74 .ctl_table_arg = root_table,
75 .root = &sysctl_table_root,
76 .set = &sysctl_table_root.default_set,
80 static DEFINE_SPINLOCK(sysctl_lock);
82 static void drop_sysctl_table(struct ctl_table_header *header);
83 static int sysctl_follow_link(struct ctl_table_header **phead,
84 struct ctl_table **pentry);
85 static int insert_links(struct ctl_table_header *head);
86 static void put_links(struct ctl_table_header *header);
88 static void sysctl_print_dir(struct ctl_dir *dir)
90 if (dir->header.parent)
91 sysctl_print_dir(dir->header.parent);
92 pr_cont("%s/", dir->header.ctl_table[0].procname);
95 static int namecmp(const char *name1, int len1, const char *name2, int len2)
104 cmp = memcmp(name1, name2, minlen);
110 /* Called under sysctl_lock */
111 static struct ctl_table *find_entry(struct ctl_table_header **phead,
112 struct ctl_dir *dir, const char *name, int namelen)
114 struct ctl_table_header *head;
115 struct ctl_table *entry;
116 struct rb_node *node = dir->root.rb_node;
120 struct ctl_node *ctl_node;
121 const char *procname;
124 ctl_node = rb_entry(node, struct ctl_node, node);
125 head = ctl_node->header;
126 entry = &head->ctl_table[ctl_node - head->node];
127 procname = entry->procname;
129 cmp = namecmp(name, namelen, procname, strlen(procname));
131 node = node->rb_left;
133 node = node->rb_right;
142 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
144 struct rb_node *node = &head->node[entry - head->ctl_table].node;
145 struct rb_node **p = &head->parent->root.rb_node;
146 struct rb_node *parent = NULL;
147 const char *name = entry->procname;
148 int namelen = strlen(name);
151 struct ctl_table_header *parent_head;
152 struct ctl_table *parent_entry;
153 struct ctl_node *parent_node;
154 const char *parent_name;
158 parent_node = rb_entry(parent, struct ctl_node, node);
159 parent_head = parent_node->header;
160 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
161 parent_name = parent_entry->procname;
163 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
169 pr_err("sysctl duplicate entry: ");
170 sysctl_print_dir(head->parent);
171 pr_cont("/%s\n", entry->procname);
176 rb_link_node(node, parent, p);
177 rb_insert_color(node, &head->parent->root);
181 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
183 struct rb_node *node = &head->node[entry - head->ctl_table].node;
185 rb_erase(node, &head->parent->root);
188 static void init_header(struct ctl_table_header *head,
189 struct ctl_table_root *root, struct ctl_table_set *set,
190 struct ctl_node *node, struct ctl_table *table)
192 head->ctl_table = table;
193 head->ctl_table_arg = table;
197 head->unregistering = NULL;
202 INIT_HLIST_HEAD(&head->inodes);
204 struct ctl_table *entry;
205 for (entry = table; entry->procname; entry++, node++)
210 static void erase_header(struct ctl_table_header *head)
212 struct ctl_table *entry;
213 for (entry = head->ctl_table; entry->procname; entry++)
214 erase_entry(head, entry);
217 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
219 struct ctl_table *entry;
222 /* Is this a permanently empty directory? */
223 if (is_empty_dir(&dir->header))
226 /* Am I creating a permanently empty directory? */
227 if (header->ctl_table == sysctl_mount_point) {
228 if (!RB_EMPTY_ROOT(&dir->root))
234 header->parent = dir;
235 err = insert_links(header);
238 for (entry = header->ctl_table; entry->procname; entry++) {
239 err = insert_entry(header, entry);
245 erase_header(header);
248 if (header->ctl_table == sysctl_mount_point)
249 clear_empty_dir(dir);
250 header->parent = NULL;
251 drop_sysctl_table(&dir->header);
255 /* called under sysctl_lock */
256 static int use_table(struct ctl_table_header *p)
258 if (unlikely(p->unregistering))
264 /* called under sysctl_lock */
265 static void unuse_table(struct ctl_table_header *p)
268 if (unlikely(p->unregistering))
269 complete(p->unregistering);
272 static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
274 proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
277 /* called under sysctl_lock, will reacquire if has to wait */
278 static void start_unregistering(struct ctl_table_header *p)
281 * if p->used is 0, nobody will ever touch that entry again;
282 * we'll eliminate all paths to it before dropping sysctl_lock
284 if (unlikely(p->used)) {
285 struct completion wait;
286 init_completion(&wait);
287 p->unregistering = &wait;
288 spin_unlock(&sysctl_lock);
289 wait_for_completion(&wait);
291 /* anything non-NULL; we'll never dereference it */
292 p->unregistering = ERR_PTR(-EINVAL);
293 spin_unlock(&sysctl_lock);
296 * Invalidate dentries for unregistered sysctls: namespaced sysctls
297 * can have duplicate names and contaminate dcache very badly.
299 proc_sys_invalidate_dcache(p);
301 * do not remove from the list until nobody holds it; walking the
302 * list in do_sysctl() relies on that.
304 spin_lock(&sysctl_lock);
308 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
311 spin_lock(&sysctl_lock);
312 if (!use_table(head))
313 head = ERR_PTR(-ENOENT);
314 spin_unlock(&sysctl_lock);
318 static void sysctl_head_finish(struct ctl_table_header *head)
322 spin_lock(&sysctl_lock);
324 spin_unlock(&sysctl_lock);
327 static struct ctl_table_set *
328 lookup_header_set(struct ctl_table_root *root)
330 struct ctl_table_set *set = &root->default_set;
332 set = root->lookup(root);
336 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
338 const char *name, int namelen)
340 struct ctl_table_header *head;
341 struct ctl_table *entry;
343 spin_lock(&sysctl_lock);
344 entry = find_entry(&head, dir, name, namelen);
345 if (entry && use_table(head))
349 spin_unlock(&sysctl_lock);
353 static struct ctl_node *first_usable_entry(struct rb_node *node)
355 struct ctl_node *ctl_node;
357 for (;node; node = rb_next(node)) {
358 ctl_node = rb_entry(node, struct ctl_node, node);
359 if (use_table(ctl_node->header))
365 static void first_entry(struct ctl_dir *dir,
366 struct ctl_table_header **phead, struct ctl_table **pentry)
368 struct ctl_table_header *head = NULL;
369 struct ctl_table *entry = NULL;
370 struct ctl_node *ctl_node;
372 spin_lock(&sysctl_lock);
373 ctl_node = first_usable_entry(rb_first(&dir->root));
374 spin_unlock(&sysctl_lock);
376 head = ctl_node->header;
377 entry = &head->ctl_table[ctl_node - head->node];
383 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
385 struct ctl_table_header *head = *phead;
386 struct ctl_table *entry = *pentry;
387 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
389 spin_lock(&sysctl_lock);
392 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
393 spin_unlock(&sysctl_lock);
396 head = ctl_node->header;
397 entry = &head->ctl_table[ctl_node - head->node];
404 * sysctl_perm does NOT grant the superuser all rights automatically, because
405 * some sysctl variables are readonly even to root.
408 static int test_perm(int mode, int op)
410 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
412 else if (in_egroup_p(GLOBAL_ROOT_GID))
414 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
419 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
421 struct ctl_table_root *root = head->root;
424 if (root->permissions)
425 mode = root->permissions(head, table);
429 return test_perm(mode, op);
432 static struct inode *proc_sys_make_inode(struct super_block *sb,
433 struct ctl_table_header *head, struct ctl_table *table)
435 struct ctl_table_root *root = head->root;
437 struct proc_inode *ei;
439 inode = new_inode(sb);
441 return ERR_PTR(-ENOMEM);
443 inode->i_ino = get_next_ino();
447 spin_lock(&sysctl_lock);
448 if (unlikely(head->unregistering)) {
449 spin_unlock(&sysctl_lock);
451 return ERR_PTR(-ENOENT);
454 ei->sysctl_entry = table;
455 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
457 spin_unlock(&sysctl_lock);
459 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
460 inode->i_mode = table->mode;
461 if (!S_ISDIR(table->mode)) {
462 inode->i_mode |= S_IFREG;
463 inode->i_op = &proc_sys_inode_operations;
464 inode->i_fop = &proc_sys_file_operations;
466 inode->i_mode |= S_IFDIR;
467 inode->i_op = &proc_sys_dir_operations;
468 inode->i_fop = &proc_sys_dir_file_operations;
469 if (is_empty_dir(head))
470 make_empty_dir_inode(inode);
473 if (root->set_ownership)
474 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
476 inode->i_uid = GLOBAL_ROOT_UID;
477 inode->i_gid = GLOBAL_ROOT_GID;
483 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
485 spin_lock(&sysctl_lock);
486 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
488 kfree_rcu(head, rcu);
489 spin_unlock(&sysctl_lock);
492 static struct ctl_table_header *grab_header(struct inode *inode)
494 struct ctl_table_header *head = PROC_I(inode)->sysctl;
496 head = &sysctl_table_root.default_set.dir.header;
497 return sysctl_head_grab(head);
500 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
503 struct ctl_table_header *head = grab_header(dir);
504 struct ctl_table_header *h = NULL;
505 const struct qstr *name = &dentry->d_name;
508 struct dentry *err = ERR_PTR(-ENOENT);
509 struct ctl_dir *ctl_dir;
513 return ERR_CAST(head);
515 ctl_dir = container_of(head, struct ctl_dir, header);
517 p = lookup_entry(&h, ctl_dir, name->name, name->len);
521 if (S_ISLNK(p->mode)) {
522 ret = sysctl_follow_link(&h, &p);
528 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
530 err = ERR_CAST(inode);
534 d_set_d_op(dentry, &proc_sys_dentry_operations);
535 err = d_splice_alias(inode, dentry);
539 sysctl_head_finish(h);
540 sysctl_head_finish(head);
544 static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
547 struct inode *inode = file_inode(iocb->ki_filp);
548 struct ctl_table_header *head = grab_header(inode);
549 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
550 size_t count = iov_iter_count(iter);
555 return PTR_ERR(head);
558 * At this point we know that the sysctl was not unregistered
559 * and won't be until we finish.
562 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
565 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
567 if (!table->proc_handler)
570 /* don't even try if the size is too large */
572 if (count >= KMALLOC_MAX_SIZE)
574 kbuf = kzalloc(count + 1, GFP_KERNEL);
580 if (!copy_from_iter_full(kbuf, count, iter))
585 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
590 /* careful: calling conventions are nasty here */
591 error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
597 if (copy_to_iter(kbuf, count, iter) < count)
605 sysctl_head_finish(head);
610 static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
612 return proc_sys_call_handler(iocb, iter, 0);
615 static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
617 return proc_sys_call_handler(iocb, iter, 1);
620 static int proc_sys_open(struct inode *inode, struct file *filp)
622 struct ctl_table_header *head = grab_header(inode);
623 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
625 /* sysctl was unregistered */
627 return PTR_ERR(head);
630 filp->private_data = proc_sys_poll_event(table->poll);
632 sysctl_head_finish(head);
637 static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
639 struct inode *inode = file_inode(filp);
640 struct ctl_table_header *head = grab_header(inode);
641 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
642 __poll_t ret = DEFAULT_POLLMASK;
645 /* sysctl was unregistered */
647 return EPOLLERR | EPOLLHUP;
649 if (!table->proc_handler)
655 event = (unsigned long)filp->private_data;
656 poll_wait(filp, &table->poll->wait, wait);
658 if (event != atomic_read(&table->poll->event)) {
659 filp->private_data = proc_sys_poll_event(table->poll);
660 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
664 sysctl_head_finish(head);
669 static bool proc_sys_fill_cache(struct file *file,
670 struct dir_context *ctx,
671 struct ctl_table_header *head,
672 struct ctl_table *table)
674 struct dentry *child, *dir = file->f_path.dentry;
678 unsigned type = DT_UNKNOWN;
680 qname.name = table->procname;
681 qname.len = strlen(table->procname);
682 qname.hash = full_name_hash(dir, qname.name, qname.len);
684 child = d_lookup(dir, &qname);
686 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
687 child = d_alloc_parallel(dir, &qname, &wq);
690 if (d_in_lookup(child)) {
692 inode = proc_sys_make_inode(dir->d_sb, head, table);
694 d_lookup_done(child);
698 d_set_d_op(child, &proc_sys_dentry_operations);
699 res = d_splice_alias(inode, child);
700 d_lookup_done(child);
711 inode = d_inode(child);
713 type = inode->i_mode >> 12;
715 return dir_emit(ctx, qname.name, qname.len, ino, type);
718 static bool proc_sys_link_fill_cache(struct file *file,
719 struct dir_context *ctx,
720 struct ctl_table_header *head,
721 struct ctl_table *table)
725 head = sysctl_head_grab(head);
729 /* It is not an error if we can not follow the link ignore it */
730 if (sysctl_follow_link(&head, &table))
733 ret = proc_sys_fill_cache(file, ctx, head, table);
735 sysctl_head_finish(head);
739 static int scan(struct ctl_table_header *head, struct ctl_table *table,
740 unsigned long *pos, struct file *file,
741 struct dir_context *ctx)
745 if ((*pos)++ < ctx->pos)
748 if (unlikely(S_ISLNK(table->mode)))
749 res = proc_sys_link_fill_cache(file, ctx, head, table);
751 res = proc_sys_fill_cache(file, ctx, head, table);
759 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
761 struct ctl_table_header *head = grab_header(file_inode(file));
762 struct ctl_table_header *h = NULL;
763 struct ctl_table *entry;
764 struct ctl_dir *ctl_dir;
768 return PTR_ERR(head);
770 ctl_dir = container_of(head, struct ctl_dir, header);
772 if (!dir_emit_dots(file, ctx))
777 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
778 if (!scan(h, entry, &pos, file, ctx)) {
779 sysctl_head_finish(h);
784 sysctl_head_finish(head);
788 static int proc_sys_permission(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 dentry *dentry, struct iattr *attr)
818 struct inode *inode = d_inode(dentry);
821 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
824 error = setattr_prepare(dentry, attr);
828 setattr_copy(inode, attr);
829 mark_inode_dirty(inode);
833 static int proc_sys_getattr(const struct path *path, struct kstat *stat,
834 u32 request_mask, unsigned int query_flags)
836 struct inode *inode = d_inode(path->dentry);
837 struct ctl_table_header *head = grab_header(inode);
838 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
841 return PTR_ERR(head);
843 generic_fillattr(inode, stat);
845 stat->mode = (stat->mode & S_IFMT) | table->mode;
847 sysctl_head_finish(head);
851 static const struct file_operations proc_sys_file_operations = {
852 .open = proc_sys_open,
853 .poll = proc_sys_poll,
854 .read_iter = proc_sys_read,
855 .write_iter = proc_sys_write,
856 .splice_read = generic_file_splice_read,
857 .splice_write = iter_file_splice_write,
858 .llseek = default_llseek,
861 static const struct file_operations proc_sys_dir_file_operations = {
862 .read = generic_read_dir,
863 .iterate_shared = proc_sys_readdir,
864 .llseek = generic_file_llseek,
867 static const struct inode_operations proc_sys_inode_operations = {
868 .permission = proc_sys_permission,
869 .setattr = proc_sys_setattr,
870 .getattr = proc_sys_getattr,
873 static const struct inode_operations proc_sys_dir_operations = {
874 .lookup = proc_sys_lookup,
875 .permission = proc_sys_permission,
876 .setattr = proc_sys_setattr,
877 .getattr = proc_sys_getattr,
880 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
882 if (flags & LOOKUP_RCU)
884 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
887 static int proc_sys_delete(const struct dentry *dentry)
889 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
892 static int sysctl_is_seen(struct ctl_table_header *p)
894 struct ctl_table_set *set = p->set;
896 spin_lock(&sysctl_lock);
897 if (p->unregistering)
899 else if (!set->is_seen)
902 res = set->is_seen(set);
903 spin_unlock(&sysctl_lock);
907 static int proc_sys_compare(const struct dentry *dentry,
908 unsigned int len, const char *str, const struct qstr *name)
910 struct ctl_table_header *head;
913 /* Although proc doesn't have negative dentries, rcu-walk means
914 * that inode here can be NULL */
915 /* AV: can it, indeed? */
916 inode = d_inode_rcu(dentry);
919 if (name->len != len)
921 if (memcmp(name->name, str, len))
923 head = rcu_dereference(PROC_I(inode)->sysctl);
924 return !head || !sysctl_is_seen(head);
927 static const struct dentry_operations proc_sys_dentry_operations = {
928 .d_revalidate = proc_sys_revalidate,
929 .d_delete = proc_sys_delete,
930 .d_compare = proc_sys_compare,
933 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
934 const char *name, int namelen)
936 struct ctl_table_header *head;
937 struct ctl_table *entry;
939 entry = find_entry(&head, dir, name, namelen);
941 return ERR_PTR(-ENOENT);
942 if (!S_ISDIR(entry->mode))
943 return ERR_PTR(-ENOTDIR);
944 return container_of(head, struct ctl_dir, header);
947 static struct ctl_dir *new_dir(struct ctl_table_set *set,
948 const char *name, int namelen)
950 struct ctl_table *table;
952 struct ctl_node *node;
955 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
956 sizeof(struct ctl_table)*2 + namelen + 1,
961 node = (struct ctl_node *)(new + 1);
962 table = (struct ctl_table *)(node + 1);
963 new_name = (char *)(table + 2);
964 memcpy(new_name, name, namelen);
965 new_name[namelen] = '\0';
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);
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",
1026 namelen, namelen, name, PTR_ERR(subdir));
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;
1059 spin_lock(&sysctl_lock);
1060 root = (*pentry)->data;
1061 set = lookup_header_set(root);
1062 dir = xlate_dir(set, (*phead)->parent);
1066 const char *procname = (*pentry)->procname;
1068 entry = find_entry(&head, dir, procname, strlen(procname));
1070 if (entry && use_table(head)) {
1071 unuse_table(*phead);
1078 spin_unlock(&sysctl_lock);
1082 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1084 struct va_format vaf;
1087 va_start(args, fmt);
1091 pr_err("sysctl table check failed: %s/%s %pV\n",
1092 path, table->procname, &vaf);
1098 static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1102 if ((table->proc_handler == proc_douintvec) ||
1103 (table->proc_handler == proc_douintvec_minmax)) {
1104 if (table->maxlen != sizeof(unsigned int))
1105 err |= sysctl_err(path, table, "array not allowed");
1111 static int sysctl_check_table(const char *path, struct ctl_table *table)
1114 for (; table->procname; table++) {
1116 err |= sysctl_err(path, table, "Not a file");
1118 if ((table->proc_handler == proc_dostring) ||
1119 (table->proc_handler == proc_dointvec) ||
1120 (table->proc_handler == proc_douintvec) ||
1121 (table->proc_handler == proc_douintvec_minmax) ||
1122 (table->proc_handler == proc_dointvec_minmax) ||
1123 (table->proc_handler == proc_dointvec_jiffies) ||
1124 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1125 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1126 (table->proc_handler == proc_doulongvec_minmax) ||
1127 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1129 err |= sysctl_err(path, table, "No data");
1131 err |= sysctl_err(path, table, "No maxlen");
1133 err |= sysctl_check_table_array(path, table);
1135 if (!table->proc_handler)
1136 err |= sysctl_err(path, table, "No proc_handler");
1138 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1139 err |= sysctl_err(path, table, "bogus .mode 0%o",
1145 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1146 struct ctl_table_root *link_root)
1148 struct ctl_table *link_table, *entry, *link;
1149 struct ctl_table_header *links;
1150 struct ctl_node *node;
1152 int nr_entries, name_bytes;
1156 for (entry = table; entry->procname; entry++) {
1158 name_bytes += strlen(entry->procname) + 1;
1161 links = kzalloc(sizeof(struct ctl_table_header) +
1162 sizeof(struct ctl_node)*nr_entries +
1163 sizeof(struct ctl_table)*(nr_entries + 1) +
1170 node = (struct ctl_node *)(links + 1);
1171 link_table = (struct ctl_table *)(node + nr_entries);
1172 link_name = (char *)&link_table[nr_entries + 1];
1174 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1175 int len = strlen(entry->procname) + 1;
1176 memcpy(link_name, entry->procname, len);
1177 link->procname = link_name;
1178 link->mode = S_IFLNK|S_IRWXUGO;
1179 link->data = link_root;
1182 init_header(links, dir->header.root, dir->header.set, node, link_table);
1183 links->nreg = nr_entries;
1188 static bool get_links(struct ctl_dir *dir,
1189 struct ctl_table *table, struct ctl_table_root *link_root)
1191 struct ctl_table_header *head;
1192 struct ctl_table *entry, *link;
1194 /* Are there links available for every entry in table? */
1195 for (entry = table; entry->procname; entry++) {
1196 const char *procname = entry->procname;
1197 link = find_entry(&head, dir, procname, strlen(procname));
1200 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1202 if (S_ISLNK(link->mode) && (link->data == link_root))
1207 /* The checks passed. Increase the registration count on the links */
1208 for (entry = table; entry->procname; entry++) {
1209 const char *procname = entry->procname;
1210 link = find_entry(&head, dir, procname, strlen(procname));
1216 static int insert_links(struct ctl_table_header *head)
1218 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1219 struct ctl_dir *core_parent = NULL;
1220 struct ctl_table_header *links;
1223 if (head->set == root_set)
1226 core_parent = xlate_dir(root_set, head->parent);
1227 if (IS_ERR(core_parent))
1230 if (get_links(core_parent, head->ctl_table, head->root))
1233 core_parent->header.nreg++;
1234 spin_unlock(&sysctl_lock);
1236 links = new_links(core_parent, head->ctl_table, head->root);
1238 spin_lock(&sysctl_lock);
1244 if (get_links(core_parent, head->ctl_table, head->root)) {
1249 err = insert_header(core_parent, links);
1253 drop_sysctl_table(&core_parent->header);
1258 * __register_sysctl_table - register a leaf sysctl table
1259 * @set: Sysctl tree to register on
1260 * @path: The path to the directory the sysctl table is in.
1261 * @table: the top-level table structure
1263 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1264 * array. A completely 0 filled entry terminates the table.
1266 * The members of the &struct ctl_table structure are used as follows:
1268 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1269 * enter a sysctl file
1271 * data - a pointer to data for use by proc_handler
1273 * maxlen - the maximum size in bytes of the data
1275 * mode - the file permissions for the /proc/sys file
1277 * child - must be %NULL.
1279 * proc_handler - the text handler routine (described below)
1281 * extra1, extra2 - extra pointers usable by the proc handler routines
1283 * Leaf nodes in the sysctl tree will be represented by a single file
1284 * under /proc; non-leaf nodes will be represented by directories.
1286 * There must be a proc_handler routine for any terminal nodes.
1287 * Several default handlers are available to cover common cases -
1289 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1290 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1291 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1293 * It is the handler's job to read the input buffer from user memory
1294 * and process it. The handler should return 0 on success.
1296 * This routine returns %NULL on a failure to register, and a pointer
1297 * to the table header on success.
1299 struct ctl_table_header *__register_sysctl_table(
1300 struct ctl_table_set *set,
1301 const char *path, struct ctl_table *table)
1303 struct ctl_table_root *root = set->dir.header.root;
1304 struct ctl_table_header *header;
1305 const char *name, *nextname;
1306 struct ctl_dir *dir;
1307 struct ctl_table *entry;
1308 struct ctl_node *node;
1311 for (entry = table; entry->procname; entry++)
1314 header = kzalloc(sizeof(struct ctl_table_header) +
1315 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1319 node = (struct ctl_node *)(header + 1);
1320 init_header(header, root, set, node, table);
1321 if (sysctl_check_table(path, table))
1324 spin_lock(&sysctl_lock);
1326 /* Reference moved down the diretory tree get_subdir */
1328 spin_unlock(&sysctl_lock);
1330 /* Find the directory for the ctl_table */
1331 for (name = path; name; name = nextname) {
1333 nextname = strchr(name, '/');
1335 namelen = nextname - name;
1338 namelen = strlen(name);
1343 dir = get_subdir(dir, name, namelen);
1348 spin_lock(&sysctl_lock);
1349 if (insert_header(dir, header))
1350 goto fail_put_dir_locked;
1352 drop_sysctl_table(&dir->header);
1353 spin_unlock(&sysctl_lock);
1357 fail_put_dir_locked:
1358 drop_sysctl_table(&dir->header);
1359 spin_unlock(&sysctl_lock);
1367 * register_sysctl - register a sysctl table
1368 * @path: The path to the directory the sysctl table is in.
1369 * @table: the table structure
1371 * Register a sysctl table. @table should be a filled in ctl_table
1372 * array. A completely 0 filled entry terminates the table.
1374 * See __register_sysctl_table for more details.
1376 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1378 return __register_sysctl_table(&sysctl_table_root.default_set,
1381 EXPORT_SYMBOL(register_sysctl);
1383 static char *append_path(const char *path, char *pos, const char *name)
1386 namelen = strlen(name);
1387 if (((pos - path) + namelen + 2) >= PATH_MAX)
1389 memcpy(pos, name, namelen);
1391 pos[namelen + 1] = '\0';
1396 static int count_subheaders(struct ctl_table *table)
1399 int nr_subheaders = 0;
1400 struct ctl_table *entry;
1402 /* special case: no directory and empty directory */
1403 if (!table || !table->procname)
1406 for (entry = table; entry->procname; entry++) {
1408 nr_subheaders += count_subheaders(entry->child);
1412 return nr_subheaders + has_files;
1415 static int register_leaf_sysctl_tables(const char *path, char *pos,
1416 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1417 struct ctl_table *table)
1419 struct ctl_table *ctl_table_arg = NULL;
1420 struct ctl_table *entry, *files;
1425 for (entry = table; entry->procname; entry++) {
1433 /* If there are mixed files and directories we need a new table */
1434 if (nr_dirs && nr_files) {
1435 struct ctl_table *new;
1436 files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1441 ctl_table_arg = files;
1442 for (new = files, entry = table; entry->procname; entry++) {
1450 /* Register everything except a directory full of subdirectories */
1451 if (nr_files || !nr_dirs) {
1452 struct ctl_table_header *header;
1453 header = __register_sysctl_table(set, path, files);
1455 kfree(ctl_table_arg);
1459 /* Remember if we need to free the file table */
1460 header->ctl_table_arg = ctl_table_arg;
1461 **subheader = header;
1465 /* Recurse into the subdirectories. */
1466 for (entry = table; entry->procname; entry++) {
1472 err = -ENAMETOOLONG;
1473 child_pos = append_path(path, pos, entry->procname);
1477 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1485 /* On failure our caller will unregister all registered subheaders */
1490 * __register_sysctl_paths - register a sysctl table hierarchy
1491 * @set: Sysctl tree to register on
1492 * @path: The path to the directory the sysctl table is in.
1493 * @table: the top-level table structure
1495 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1496 * array. A completely 0 filled entry terminates the table.
1498 * See __register_sysctl_table for more details.
1500 struct ctl_table_header *__register_sysctl_paths(
1501 struct ctl_table_set *set,
1502 const struct ctl_path *path, struct ctl_table *table)
1504 struct ctl_table *ctl_table_arg = table;
1505 int nr_subheaders = count_subheaders(table);
1506 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1507 const struct ctl_path *component;
1508 char *new_path, *pos;
1510 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1515 for (component = path; component->procname; component++) {
1516 pos = append_path(new_path, pos, component->procname);
1520 while (table->procname && table->child && !table[1].procname) {
1521 pos = append_path(new_path, pos, table->procname);
1524 table = table->child;
1526 if (nr_subheaders == 1) {
1527 header = __register_sysctl_table(set, new_path, table);
1529 header->ctl_table_arg = ctl_table_arg;
1531 header = kzalloc(sizeof(*header) +
1532 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1536 subheaders = (struct ctl_table_header **) (header + 1);
1537 subheader = subheaders;
1538 header->ctl_table_arg = ctl_table_arg;
1540 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1542 goto err_register_leaves;
1549 err_register_leaves:
1550 while (subheader > subheaders) {
1551 struct ctl_table_header *subh = *(--subheader);
1552 struct ctl_table *table = subh->ctl_table_arg;
1553 unregister_sysctl_table(subh);
1562 * register_sysctl_table_path - register a sysctl table hierarchy
1563 * @path: The path to the directory the sysctl table is in.
1564 * @table: the top-level table structure
1566 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1567 * array. A completely 0 filled entry terminates the table.
1569 * See __register_sysctl_paths for more details.
1571 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1572 struct ctl_table *table)
1574 return __register_sysctl_paths(&sysctl_table_root.default_set,
1577 EXPORT_SYMBOL(register_sysctl_paths);
1580 * register_sysctl_table - register a sysctl table hierarchy
1581 * @table: the top-level table structure
1583 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1584 * array. A completely 0 filled entry terminates the table.
1586 * See register_sysctl_paths for more details.
1588 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1590 static const struct ctl_path null_path[] = { {} };
1592 return register_sysctl_paths(null_path, table);
1594 EXPORT_SYMBOL(register_sysctl_table);
1596 static void put_links(struct ctl_table_header *header)
1598 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1599 struct ctl_table_root *root = header->root;
1600 struct ctl_dir *parent = header->parent;
1601 struct ctl_dir *core_parent;
1602 struct ctl_table *entry;
1604 if (header->set == root_set)
1607 core_parent = xlate_dir(root_set, parent);
1608 if (IS_ERR(core_parent))
1611 for (entry = header->ctl_table; entry->procname; entry++) {
1612 struct ctl_table_header *link_head;
1613 struct ctl_table *link;
1614 const char *name = entry->procname;
1616 link = find_entry(&link_head, core_parent, name, strlen(name));
1618 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1619 (S_ISLNK(link->mode) && (link->data == root)))) {
1620 drop_sysctl_table(link_head);
1623 pr_err("sysctl link missing during unregister: ");
1624 sysctl_print_dir(parent);
1625 pr_cont("/%s\n", name);
1630 static void drop_sysctl_table(struct ctl_table_header *header)
1632 struct ctl_dir *parent = header->parent;
1639 start_unregistering(header);
1642 if (!--header->count)
1643 kfree_rcu(header, rcu);
1646 drop_sysctl_table(&parent->header);
1650 * unregister_sysctl_table - unregister a sysctl table hierarchy
1651 * @header: the header returned from register_sysctl_table
1653 * Unregisters the sysctl table and all children. proc entries may not
1654 * actually be removed until they are no longer used by anyone.
1656 void unregister_sysctl_table(struct ctl_table_header * header)
1664 nr_subheaders = count_subheaders(header->ctl_table_arg);
1665 if (unlikely(nr_subheaders > 1)) {
1666 struct ctl_table_header **subheaders;
1669 subheaders = (struct ctl_table_header **)(header + 1);
1670 for (i = nr_subheaders -1; i >= 0; i--) {
1671 struct ctl_table_header *subh = subheaders[i];
1672 struct ctl_table *table = subh->ctl_table_arg;
1673 unregister_sysctl_table(subh);
1680 spin_lock(&sysctl_lock);
1681 drop_sysctl_table(header);
1682 spin_unlock(&sysctl_lock);
1684 EXPORT_SYMBOL(unregister_sysctl_table);
1686 void setup_sysctl_set(struct ctl_table_set *set,
1687 struct ctl_table_root *root,
1688 int (*is_seen)(struct ctl_table_set *))
1690 memset(set, 0, sizeof(*set));
1691 set->is_seen = is_seen;
1692 init_header(&set->dir.header, root, set, NULL, root_table);
1695 void retire_sysctl_set(struct ctl_table_set *set)
1697 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1700 int __init proc_sys_init(void)
1702 struct proc_dir_entry *proc_sys_root;
1704 proc_sys_root = proc_mkdir("sys", NULL);
1705 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1706 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1707 proc_sys_root->nlink = 0;
1709 return sysctl_init();
1712 struct sysctl_alias {
1713 const char *kernel_param;
1714 const char *sysctl_param;
1718 * Historically some settings had both sysctl and a command line parameter.
1719 * With the generic sysctl. parameter support, we can handle them at a single
1720 * place and only keep the historical name for compatibility. This is not meant
1721 * to add brand new aliases. When adding existing aliases, consider whether
1722 * the possibly different moment of changing the value (e.g. from early_param
1723 * to the moment do_sysctl_args() is called) is an issue for the specific
1726 static const struct sysctl_alias sysctl_aliases[] = {
1727 {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1728 {"hung_task_panic", "kernel.hung_task_panic" },
1729 {"numa_zonelist_order", "vm.numa_zonelist_order" },
1730 {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1731 {"softlockup_panic", "kernel.softlockup_panic" },
1735 static const char *sysctl_find_alias(char *param)
1737 const struct sysctl_alias *alias;
1739 for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1740 if (strcmp(alias->kernel_param, param) == 0)
1741 return alias->sysctl_param;
1747 /* Set sysctl value passed on kernel command line. */
1748 static int process_sysctl_arg(char *param, char *val,
1749 const char *unused, void *arg)
1752 struct vfsmount **proc_mnt = arg;
1753 struct file_system_type *proc_fs_type;
1760 if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1761 param += sizeof("sysctl") - 1;
1763 if (param[0] != '/' && param[0] != '.')
1768 param = (char *) sysctl_find_alias(param);
1780 * To set sysctl options, we use a temporary mount of proc, look up the
1781 * respective sys/ file and write to it. To avoid mounting it when no
1782 * options were given, we mount it only when the first sysctl option is
1783 * found. Why not a persistent mount? There are problems with a
1784 * persistent mount of proc in that it forces userspace not to use any
1785 * proc mount options.
1788 proc_fs_type = get_fs_type("proc");
1789 if (!proc_fs_type) {
1790 pr_err("Failed to find procfs to set sysctl from command line\n");
1793 *proc_mnt = kern_mount(proc_fs_type);
1794 put_filesystem(proc_fs_type);
1795 if (IS_ERR(*proc_mnt)) {
1796 pr_err("Failed to mount procfs to set sysctl from command line\n");
1801 path = kasprintf(GFP_KERNEL, "sys/%s", param);
1803 panic("%s: Failed to allocate path for %s\n", __func__, param);
1804 strreplace(path, '.', '/');
1806 file = file_open_root((*proc_mnt)->mnt_root, *proc_mnt, path, O_WRONLY, 0);
1808 err = PTR_ERR(file);
1810 pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1812 else if (err == -EACCES)
1813 pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1816 pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1820 wret = kernel_write(file, val, len, &pos);
1824 pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1827 pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1828 ERR_PTR(err), param, val);
1829 } else if (wret != len) {
1830 pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1831 wret, len, path, param, val);
1834 err = filp_close(file, NULL);
1836 pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1837 ERR_PTR(err), param, val);
1843 void do_sysctl_args(void)
1846 struct vfsmount *proc_mnt = NULL;
1848 command_line = kstrdup(saved_command_line, GFP_KERNEL);
1850 panic("%s: Failed to allocate copy of command line\n", __func__);
1852 parse_args("Setting sysctl args", command_line,
1853 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1856 kern_unmount(proc_mnt);
1858 kfree(command_line);