1 // SPDX-License-Identifier: GPL-2.0
5 #include <linux/init.h>
6 #include <linux/sysctl.h>
7 #include <linux/poll.h>
8 #include <linux/proc_fs.h>
9 #include <linux/printk.h>
10 #include <linux/security.h>
11 #include <linux/sched.h>
12 #include <linux/cred.h>
13 #include <linux/namei.h>
15 #include <linux/module.h>
16 #include <linux/bpf-cgroup.h>
19 static const struct dentry_operations proc_sys_dentry_operations;
20 static const struct file_operations proc_sys_file_operations;
21 static const struct inode_operations proc_sys_inode_operations;
22 static const struct file_operations proc_sys_dir_file_operations;
23 static const struct inode_operations proc_sys_dir_operations;
25 /* Support for permanently empty directories */
27 struct ctl_table sysctl_mount_point[] = {
31 static bool is_empty_dir(struct ctl_table_header *head)
33 return head->ctl_table[0].child == sysctl_mount_point;
36 static void set_empty_dir(struct ctl_dir *dir)
38 dir->header.ctl_table[0].child = sysctl_mount_point;
41 static void clear_empty_dir(struct ctl_dir *dir)
44 dir->header.ctl_table[0].child = NULL;
47 void proc_sys_poll_notify(struct ctl_table_poll *poll)
52 atomic_inc(&poll->event);
53 wake_up_interruptible(&poll->wait);
56 static struct ctl_table root_table[] = {
59 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
63 static struct ctl_table_root sysctl_table_root = {
64 .default_set.dir.header = {
67 .ctl_table = root_table }},
68 .ctl_table_arg = root_table,
69 .root = &sysctl_table_root,
70 .set = &sysctl_table_root.default_set,
74 static DEFINE_SPINLOCK(sysctl_lock);
76 static void drop_sysctl_table(struct ctl_table_header *header);
77 static int sysctl_follow_link(struct ctl_table_header **phead,
78 struct ctl_table **pentry);
79 static int insert_links(struct ctl_table_header *head);
80 static void put_links(struct ctl_table_header *header);
82 static void sysctl_print_dir(struct ctl_dir *dir)
84 if (dir->header.parent)
85 sysctl_print_dir(dir->header.parent);
86 pr_cont("%s/", dir->header.ctl_table[0].procname);
89 static int namecmp(const char *name1, int len1, const char *name2, int len2)
98 cmp = memcmp(name1, name2, minlen);
104 /* Called under sysctl_lock */
105 static struct ctl_table *find_entry(struct ctl_table_header **phead,
106 struct ctl_dir *dir, const char *name, int namelen)
108 struct ctl_table_header *head;
109 struct ctl_table *entry;
110 struct rb_node *node = dir->root.rb_node;
114 struct ctl_node *ctl_node;
115 const char *procname;
118 ctl_node = rb_entry(node, struct ctl_node, node);
119 head = ctl_node->header;
120 entry = &head->ctl_table[ctl_node - head->node];
121 procname = entry->procname;
123 cmp = namecmp(name, namelen, procname, strlen(procname));
125 node = node->rb_left;
127 node = node->rb_right;
136 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
138 struct rb_node *node = &head->node[entry - head->ctl_table].node;
139 struct rb_node **p = &head->parent->root.rb_node;
140 struct rb_node *parent = NULL;
141 const char *name = entry->procname;
142 int namelen = strlen(name);
145 struct ctl_table_header *parent_head;
146 struct ctl_table *parent_entry;
147 struct ctl_node *parent_node;
148 const char *parent_name;
152 parent_node = rb_entry(parent, struct ctl_node, node);
153 parent_head = parent_node->header;
154 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
155 parent_name = parent_entry->procname;
157 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
163 pr_err("sysctl duplicate entry: ");
164 sysctl_print_dir(head->parent);
165 pr_cont("/%s\n", entry->procname);
170 rb_link_node(node, parent, p);
171 rb_insert_color(node, &head->parent->root);
175 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
177 struct rb_node *node = &head->node[entry - head->ctl_table].node;
179 rb_erase(node, &head->parent->root);
182 static void init_header(struct ctl_table_header *head,
183 struct ctl_table_root *root, struct ctl_table_set *set,
184 struct ctl_node *node, struct ctl_table *table)
186 head->ctl_table = table;
187 head->ctl_table_arg = table;
191 head->unregistering = NULL;
196 INIT_HLIST_HEAD(&head->inodes);
198 struct ctl_table *entry;
199 for (entry = table; entry->procname; entry++, node++)
204 static void erase_header(struct ctl_table_header *head)
206 struct ctl_table *entry;
207 for (entry = head->ctl_table; entry->procname; entry++)
208 erase_entry(head, entry);
211 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
213 struct ctl_table *entry;
216 /* Is this a permanently empty directory? */
217 if (is_empty_dir(&dir->header))
220 /* Am I creating a permanently empty directory? */
221 if (header->ctl_table == sysctl_mount_point) {
222 if (!RB_EMPTY_ROOT(&dir->root))
228 header->parent = dir;
229 err = insert_links(header);
232 for (entry = header->ctl_table; entry->procname; entry++) {
233 err = insert_entry(header, entry);
239 erase_header(header);
242 if (header->ctl_table == sysctl_mount_point)
243 clear_empty_dir(dir);
244 header->parent = NULL;
245 drop_sysctl_table(&dir->header);
249 /* called under sysctl_lock */
250 static int use_table(struct ctl_table_header *p)
252 if (unlikely(p->unregistering))
258 /* called under sysctl_lock */
259 static void unuse_table(struct ctl_table_header *p)
262 if (unlikely(p->unregistering))
263 complete(p->unregistering);
266 static void proc_sys_prune_dcache(struct ctl_table_header *head)
269 struct proc_inode *ei;
270 struct hlist_node *node;
271 struct super_block *sb;
275 node = hlist_first_rcu(&head->inodes);
278 ei = hlist_entry(node, struct proc_inode, sysctl_inodes);
279 spin_lock(&sysctl_lock);
280 hlist_del_init_rcu(&ei->sysctl_inodes);
281 spin_unlock(&sysctl_lock);
283 inode = &ei->vfs_inode;
285 if (!atomic_inc_not_zero(&sb->s_active))
287 inode = igrab(inode);
289 if (unlikely(!inode)) {
290 deactivate_super(sb);
295 d_prune_aliases(inode);
297 deactivate_super(sb);
304 /* called under sysctl_lock, will reacquire if has to wait */
305 static void start_unregistering(struct ctl_table_header *p)
308 * if p->used is 0, nobody will ever touch that entry again;
309 * we'll eliminate all paths to it before dropping sysctl_lock
311 if (unlikely(p->used)) {
312 struct completion wait;
313 init_completion(&wait);
314 p->unregistering = &wait;
315 spin_unlock(&sysctl_lock);
316 wait_for_completion(&wait);
318 /* anything non-NULL; we'll never dereference it */
319 p->unregistering = ERR_PTR(-EINVAL);
320 spin_unlock(&sysctl_lock);
323 * Prune dentries for unregistered sysctls: namespaced sysctls
324 * can have duplicate names and contaminate dcache very badly.
326 proc_sys_prune_dcache(p);
328 * do not remove from the list until nobody holds it; walking the
329 * list in do_sysctl() relies on that.
331 spin_lock(&sysctl_lock);
335 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
338 spin_lock(&sysctl_lock);
339 if (!use_table(head))
340 head = ERR_PTR(-ENOENT);
341 spin_unlock(&sysctl_lock);
345 static void sysctl_head_finish(struct ctl_table_header *head)
349 spin_lock(&sysctl_lock);
351 spin_unlock(&sysctl_lock);
354 static struct ctl_table_set *
355 lookup_header_set(struct ctl_table_root *root)
357 struct ctl_table_set *set = &root->default_set;
359 set = root->lookup(root);
363 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
365 const char *name, int namelen)
367 struct ctl_table_header *head;
368 struct ctl_table *entry;
370 spin_lock(&sysctl_lock);
371 entry = find_entry(&head, dir, name, namelen);
372 if (entry && use_table(head))
376 spin_unlock(&sysctl_lock);
380 static struct ctl_node *first_usable_entry(struct rb_node *node)
382 struct ctl_node *ctl_node;
384 for (;node; node = rb_next(node)) {
385 ctl_node = rb_entry(node, struct ctl_node, node);
386 if (use_table(ctl_node->header))
392 static void first_entry(struct ctl_dir *dir,
393 struct ctl_table_header **phead, struct ctl_table **pentry)
395 struct ctl_table_header *head = NULL;
396 struct ctl_table *entry = NULL;
397 struct ctl_node *ctl_node;
399 spin_lock(&sysctl_lock);
400 ctl_node = first_usable_entry(rb_first(&dir->root));
401 spin_unlock(&sysctl_lock);
403 head = ctl_node->header;
404 entry = &head->ctl_table[ctl_node - head->node];
410 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
412 struct ctl_table_header *head = *phead;
413 struct ctl_table *entry = *pentry;
414 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
416 spin_lock(&sysctl_lock);
419 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
420 spin_unlock(&sysctl_lock);
423 head = ctl_node->header;
424 entry = &head->ctl_table[ctl_node - head->node];
431 * sysctl_perm does NOT grant the superuser all rights automatically, because
432 * some sysctl variables are readonly even to root.
435 static int test_perm(int mode, int op)
437 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
439 else if (in_egroup_p(GLOBAL_ROOT_GID))
441 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
446 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
448 struct ctl_table_root *root = head->root;
451 if (root->permissions)
452 mode = root->permissions(head, table);
456 return test_perm(mode, op);
459 static struct inode *proc_sys_make_inode(struct super_block *sb,
460 struct ctl_table_header *head, struct ctl_table *table)
462 struct ctl_table_root *root = head->root;
464 struct proc_inode *ei;
466 inode = new_inode(sb);
468 return ERR_PTR(-ENOMEM);
470 inode->i_ino = get_next_ino();
474 spin_lock(&sysctl_lock);
475 if (unlikely(head->unregistering)) {
476 spin_unlock(&sysctl_lock);
478 return ERR_PTR(-ENOENT);
481 ei->sysctl_entry = table;
482 hlist_add_head_rcu(&ei->sysctl_inodes, &head->inodes);
484 spin_unlock(&sysctl_lock);
486 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
487 inode->i_mode = table->mode;
488 if (!S_ISDIR(table->mode)) {
489 inode->i_mode |= S_IFREG;
490 inode->i_op = &proc_sys_inode_operations;
491 inode->i_fop = &proc_sys_file_operations;
493 inode->i_mode |= S_IFDIR;
494 inode->i_op = &proc_sys_dir_operations;
495 inode->i_fop = &proc_sys_dir_file_operations;
496 if (is_empty_dir(head))
497 make_empty_dir_inode(inode);
500 if (root->set_ownership)
501 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
503 inode->i_uid = GLOBAL_ROOT_UID;
504 inode->i_gid = GLOBAL_ROOT_GID;
510 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
512 spin_lock(&sysctl_lock);
513 hlist_del_init_rcu(&PROC_I(inode)->sysctl_inodes);
515 kfree_rcu(head, rcu);
516 spin_unlock(&sysctl_lock);
519 static struct ctl_table_header *grab_header(struct inode *inode)
521 struct ctl_table_header *head = PROC_I(inode)->sysctl;
523 head = &sysctl_table_root.default_set.dir.header;
524 return sysctl_head_grab(head);
527 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
530 struct ctl_table_header *head = grab_header(dir);
531 struct ctl_table_header *h = NULL;
532 const struct qstr *name = &dentry->d_name;
535 struct dentry *err = ERR_PTR(-ENOENT);
536 struct ctl_dir *ctl_dir;
540 return ERR_CAST(head);
542 ctl_dir = container_of(head, struct ctl_dir, header);
544 p = lookup_entry(&h, ctl_dir, name->name, name->len);
548 if (S_ISLNK(p->mode)) {
549 ret = sysctl_follow_link(&h, &p);
555 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
557 err = ERR_CAST(inode);
561 d_set_d_op(dentry, &proc_sys_dentry_operations);
562 err = d_splice_alias(inode, dentry);
566 sysctl_head_finish(h);
567 sysctl_head_finish(head);
571 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
572 size_t count, loff_t *ppos, int write)
574 struct inode *inode = file_inode(filp);
575 struct ctl_table_header *head = grab_header(inode);
576 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
577 void *new_buf = NULL;
581 return PTR_ERR(head);
584 * At this point we know that the sysctl was not unregistered
585 * and won't be until we finish.
588 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
591 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
593 if (!table->proc_handler)
596 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, buf, &count,
601 /* careful: calling conventions are nasty here */
607 error = table->proc_handler(table, write, (void __user *)new_buf,
612 error = table->proc_handler(table, write, buf, &count, ppos);
618 sysctl_head_finish(head);
623 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
624 size_t count, loff_t *ppos)
626 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
629 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
630 size_t count, loff_t *ppos)
632 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
635 static int proc_sys_open(struct inode *inode, struct file *filp)
637 struct ctl_table_header *head = grab_header(inode);
638 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
640 /* sysctl was unregistered */
642 return PTR_ERR(head);
645 filp->private_data = proc_sys_poll_event(table->poll);
647 sysctl_head_finish(head);
652 static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
654 struct inode *inode = file_inode(filp);
655 struct ctl_table_header *head = grab_header(inode);
656 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
657 __poll_t ret = DEFAULT_POLLMASK;
660 /* sysctl was unregistered */
662 return EPOLLERR | EPOLLHUP;
664 if (!table->proc_handler)
670 event = (unsigned long)filp->private_data;
671 poll_wait(filp, &table->poll->wait, wait);
673 if (event != atomic_read(&table->poll->event)) {
674 filp->private_data = proc_sys_poll_event(table->poll);
675 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
679 sysctl_head_finish(head);
684 static bool proc_sys_fill_cache(struct file *file,
685 struct dir_context *ctx,
686 struct ctl_table_header *head,
687 struct ctl_table *table)
689 struct dentry *child, *dir = file->f_path.dentry;
693 unsigned type = DT_UNKNOWN;
695 qname.name = table->procname;
696 qname.len = strlen(table->procname);
697 qname.hash = full_name_hash(dir, qname.name, qname.len);
699 child = d_lookup(dir, &qname);
701 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
702 child = d_alloc_parallel(dir, &qname, &wq);
705 if (d_in_lookup(child)) {
707 inode = proc_sys_make_inode(dir->d_sb, head, table);
709 d_lookup_done(child);
713 d_set_d_op(child, &proc_sys_dentry_operations);
714 res = d_splice_alias(inode, child);
715 d_lookup_done(child);
726 inode = d_inode(child);
728 type = inode->i_mode >> 12;
730 return dir_emit(ctx, qname.name, qname.len, ino, type);
733 static bool proc_sys_link_fill_cache(struct file *file,
734 struct dir_context *ctx,
735 struct ctl_table_header *head,
736 struct ctl_table *table)
740 head = sysctl_head_grab(head);
744 /* It is not an error if we can not follow the link ignore it */
745 if (sysctl_follow_link(&head, &table))
748 ret = proc_sys_fill_cache(file, ctx, head, table);
750 sysctl_head_finish(head);
754 static int scan(struct ctl_table_header *head, struct ctl_table *table,
755 unsigned long *pos, struct file *file,
756 struct dir_context *ctx)
760 if ((*pos)++ < ctx->pos)
763 if (unlikely(S_ISLNK(table->mode)))
764 res = proc_sys_link_fill_cache(file, ctx, head, table);
766 res = proc_sys_fill_cache(file, ctx, head, table);
774 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
776 struct ctl_table_header *head = grab_header(file_inode(file));
777 struct ctl_table_header *h = NULL;
778 struct ctl_table *entry;
779 struct ctl_dir *ctl_dir;
783 return PTR_ERR(head);
785 ctl_dir = container_of(head, struct ctl_dir, header);
787 if (!dir_emit_dots(file, ctx))
792 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
793 if (!scan(h, entry, &pos, file, ctx)) {
794 sysctl_head_finish(h);
799 sysctl_head_finish(head);
803 static int proc_sys_permission(struct inode *inode, int mask)
806 * sysctl entries that are not writeable,
807 * are _NOT_ writeable, capabilities or not.
809 struct ctl_table_header *head;
810 struct ctl_table *table;
813 /* Executable files are not allowed under /proc/sys/ */
814 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
817 head = grab_header(inode);
819 return PTR_ERR(head);
821 table = PROC_I(inode)->sysctl_entry;
822 if (!table) /* global root - r-xr-xr-x */
823 error = mask & MAY_WRITE ? -EACCES : 0;
824 else /* Use the permissions on the sysctl table entry */
825 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
827 sysctl_head_finish(head);
831 static int proc_sys_setattr(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(dentry, attr);
843 setattr_copy(inode, attr);
844 mark_inode_dirty(inode);
848 static int proc_sys_getattr(const struct path *path, struct kstat *stat,
849 u32 request_mask, unsigned int query_flags)
851 struct inode *inode = d_inode(path->dentry);
852 struct ctl_table_header *head = grab_header(inode);
853 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
856 return PTR_ERR(head);
858 generic_fillattr(inode, stat);
860 stat->mode = (stat->mode & S_IFMT) | table->mode;
862 sysctl_head_finish(head);
866 static const struct file_operations proc_sys_file_operations = {
867 .open = proc_sys_open,
868 .poll = proc_sys_poll,
869 .read = proc_sys_read,
870 .write = proc_sys_write,
871 .llseek = default_llseek,
874 static const struct file_operations proc_sys_dir_file_operations = {
875 .read = generic_read_dir,
876 .iterate_shared = proc_sys_readdir,
877 .llseek = generic_file_llseek,
880 static const struct inode_operations proc_sys_inode_operations = {
881 .permission = proc_sys_permission,
882 .setattr = proc_sys_setattr,
883 .getattr = proc_sys_getattr,
886 static const struct inode_operations proc_sys_dir_operations = {
887 .lookup = proc_sys_lookup,
888 .permission = proc_sys_permission,
889 .setattr = proc_sys_setattr,
890 .getattr = proc_sys_getattr,
893 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
895 if (flags & LOOKUP_RCU)
897 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
900 static int proc_sys_delete(const struct dentry *dentry)
902 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
905 static int sysctl_is_seen(struct ctl_table_header *p)
907 struct ctl_table_set *set = p->set;
909 spin_lock(&sysctl_lock);
910 if (p->unregistering)
912 else if (!set->is_seen)
915 res = set->is_seen(set);
916 spin_unlock(&sysctl_lock);
920 static int proc_sys_compare(const struct dentry *dentry,
921 unsigned int len, const char *str, const struct qstr *name)
923 struct ctl_table_header *head;
926 /* Although proc doesn't have negative dentries, rcu-walk means
927 * that inode here can be NULL */
928 /* AV: can it, indeed? */
929 inode = d_inode_rcu(dentry);
932 if (name->len != len)
934 if (memcmp(name->name, str, len))
936 head = rcu_dereference(PROC_I(inode)->sysctl);
937 return !head || !sysctl_is_seen(head);
940 static const struct dentry_operations proc_sys_dentry_operations = {
941 .d_revalidate = proc_sys_revalidate,
942 .d_delete = proc_sys_delete,
943 .d_compare = proc_sys_compare,
946 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
947 const char *name, int namelen)
949 struct ctl_table_header *head;
950 struct ctl_table *entry;
952 entry = find_entry(&head, dir, name, namelen);
954 return ERR_PTR(-ENOENT);
955 if (!S_ISDIR(entry->mode))
956 return ERR_PTR(-ENOTDIR);
957 return container_of(head, struct ctl_dir, header);
960 static struct ctl_dir *new_dir(struct ctl_table_set *set,
961 const char *name, int namelen)
963 struct ctl_table *table;
965 struct ctl_node *node;
968 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
969 sizeof(struct ctl_table)*2 + namelen + 1,
974 node = (struct ctl_node *)(new + 1);
975 table = (struct ctl_table *)(node + 1);
976 new_name = (char *)(table + 2);
977 memcpy(new_name, name, namelen);
978 new_name[namelen] = '\0';
979 table[0].procname = new_name;
980 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
981 init_header(&new->header, set->dir.header.root, set, node, table);
987 * get_subdir - find or create a subdir with the specified name.
988 * @dir: Directory to create the subdirectory in
989 * @name: The name of the subdirectory to find or create
990 * @namelen: The length of name
992 * Takes a directory with an elevated reference count so we know that
993 * if we drop the lock the directory will not go away. Upon success
994 * the reference is moved from @dir to the returned subdirectory.
995 * Upon error an error code is returned and the reference on @dir is
998 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
999 const char *name, int namelen)
1001 struct ctl_table_set *set = dir->header.set;
1002 struct ctl_dir *subdir, *new = NULL;
1005 spin_lock(&sysctl_lock);
1006 subdir = find_subdir(dir, name, namelen);
1007 if (!IS_ERR(subdir))
1009 if (PTR_ERR(subdir) != -ENOENT)
1012 spin_unlock(&sysctl_lock);
1013 new = new_dir(set, name, namelen);
1014 spin_lock(&sysctl_lock);
1015 subdir = ERR_PTR(-ENOMEM);
1019 /* Was the subdir added while we dropped the lock? */
1020 subdir = find_subdir(dir, name, namelen);
1021 if (!IS_ERR(subdir))
1023 if (PTR_ERR(subdir) != -ENOENT)
1026 /* Nope. Use the our freshly made directory entry. */
1027 err = insert_header(dir, &new->header);
1028 subdir = ERR_PTR(err);
1033 subdir->header.nreg++;
1035 if (IS_ERR(subdir)) {
1036 pr_err("sysctl could not get directory: ");
1037 sysctl_print_dir(dir);
1038 pr_cont("/%*.*s %ld\n",
1039 namelen, namelen, name, PTR_ERR(subdir));
1041 drop_sysctl_table(&dir->header);
1043 drop_sysctl_table(&new->header);
1044 spin_unlock(&sysctl_lock);
1048 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1050 struct ctl_dir *parent;
1051 const char *procname;
1052 if (!dir->header.parent)
1054 parent = xlate_dir(set, dir->header.parent);
1057 procname = dir->header.ctl_table[0].procname;
1058 return find_subdir(parent, procname, strlen(procname));
1061 static int sysctl_follow_link(struct ctl_table_header **phead,
1062 struct ctl_table **pentry)
1064 struct ctl_table_header *head;
1065 struct ctl_table_root *root;
1066 struct ctl_table_set *set;
1067 struct ctl_table *entry;
1068 struct ctl_dir *dir;
1072 spin_lock(&sysctl_lock);
1073 root = (*pentry)->data;
1074 set = lookup_header_set(root);
1075 dir = xlate_dir(set, (*phead)->parent);
1079 const char *procname = (*pentry)->procname;
1081 entry = find_entry(&head, dir, procname, strlen(procname));
1083 if (entry && use_table(head)) {
1084 unuse_table(*phead);
1091 spin_unlock(&sysctl_lock);
1095 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1097 struct va_format vaf;
1100 va_start(args, fmt);
1104 pr_err("sysctl table check failed: %s/%s %pV\n",
1105 path, table->procname, &vaf);
1111 static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1115 if ((table->proc_handler == proc_douintvec) ||
1116 (table->proc_handler == proc_douintvec_minmax)) {
1117 if (table->maxlen != sizeof(unsigned int))
1118 err |= sysctl_err(path, table, "array not allowed");
1124 static int sysctl_check_table(const char *path, struct ctl_table *table)
1127 for (; table->procname; table++) {
1129 err |= sysctl_err(path, table, "Not a file");
1131 if ((table->proc_handler == proc_dostring) ||
1132 (table->proc_handler == proc_dointvec) ||
1133 (table->proc_handler == proc_douintvec) ||
1134 (table->proc_handler == proc_douintvec_minmax) ||
1135 (table->proc_handler == proc_dointvec_minmax) ||
1136 (table->proc_handler == proc_dointvec_jiffies) ||
1137 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1138 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1139 (table->proc_handler == proc_doulongvec_minmax) ||
1140 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1142 err |= sysctl_err(path, table, "No data");
1144 err |= sysctl_err(path, table, "No maxlen");
1146 err |= sysctl_check_table_array(path, table);
1148 if (!table->proc_handler)
1149 err |= sysctl_err(path, table, "No proc_handler");
1151 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1152 err |= sysctl_err(path, table, "bogus .mode 0%o",
1158 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1159 struct ctl_table_root *link_root)
1161 struct ctl_table *link_table, *entry, *link;
1162 struct ctl_table_header *links;
1163 struct ctl_node *node;
1165 int nr_entries, name_bytes;
1169 for (entry = table; entry->procname; entry++) {
1171 name_bytes += strlen(entry->procname) + 1;
1174 links = kzalloc(sizeof(struct ctl_table_header) +
1175 sizeof(struct ctl_node)*nr_entries +
1176 sizeof(struct ctl_table)*(nr_entries + 1) +
1183 node = (struct ctl_node *)(links + 1);
1184 link_table = (struct ctl_table *)(node + nr_entries);
1185 link_name = (char *)&link_table[nr_entries + 1];
1187 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1188 int len = strlen(entry->procname) + 1;
1189 memcpy(link_name, entry->procname, len);
1190 link->procname = link_name;
1191 link->mode = S_IFLNK|S_IRWXUGO;
1192 link->data = link_root;
1195 init_header(links, dir->header.root, dir->header.set, node, link_table);
1196 links->nreg = nr_entries;
1201 static bool get_links(struct ctl_dir *dir,
1202 struct ctl_table *table, struct ctl_table_root *link_root)
1204 struct ctl_table_header *head;
1205 struct ctl_table *entry, *link;
1207 /* Are there links available for every entry in table? */
1208 for (entry = table; entry->procname; entry++) {
1209 const char *procname = entry->procname;
1210 link = find_entry(&head, dir, procname, strlen(procname));
1213 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1215 if (S_ISLNK(link->mode) && (link->data == link_root))
1220 /* The checks passed. Increase the registration count on the links */
1221 for (entry = table; entry->procname; entry++) {
1222 const char *procname = entry->procname;
1223 link = find_entry(&head, dir, procname, strlen(procname));
1229 static int insert_links(struct ctl_table_header *head)
1231 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1232 struct ctl_dir *core_parent = NULL;
1233 struct ctl_table_header *links;
1236 if (head->set == root_set)
1239 core_parent = xlate_dir(root_set, head->parent);
1240 if (IS_ERR(core_parent))
1243 if (get_links(core_parent, head->ctl_table, head->root))
1246 core_parent->header.nreg++;
1247 spin_unlock(&sysctl_lock);
1249 links = new_links(core_parent, head->ctl_table, head->root);
1251 spin_lock(&sysctl_lock);
1257 if (get_links(core_parent, head->ctl_table, head->root)) {
1262 err = insert_header(core_parent, links);
1266 drop_sysctl_table(&core_parent->header);
1271 * __register_sysctl_table - register a leaf sysctl table
1272 * @set: Sysctl tree to register on
1273 * @path: The path to the directory the sysctl table is in.
1274 * @table: the top-level table structure
1276 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1277 * array. A completely 0 filled entry terminates the table.
1279 * The members of the &struct ctl_table structure are used as follows:
1281 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1282 * enter a sysctl file
1284 * data - a pointer to data for use by proc_handler
1286 * maxlen - the maximum size in bytes of the data
1288 * mode - the file permissions for the /proc/sys file
1290 * child - must be %NULL.
1292 * proc_handler - the text handler routine (described below)
1294 * extra1, extra2 - extra pointers usable by the proc handler routines
1296 * Leaf nodes in the sysctl tree will be represented by a single file
1297 * under /proc; non-leaf nodes will be represented by directories.
1299 * There must be a proc_handler routine for any terminal nodes.
1300 * Several default handlers are available to cover common cases -
1302 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1303 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1304 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1306 * It is the handler's job to read the input buffer from user memory
1307 * and process it. The handler should return 0 on success.
1309 * This routine returns %NULL on a failure to register, and a pointer
1310 * to the table header on success.
1312 struct ctl_table_header *__register_sysctl_table(
1313 struct ctl_table_set *set,
1314 const char *path, struct ctl_table *table)
1316 struct ctl_table_root *root = set->dir.header.root;
1317 struct ctl_table_header *header;
1318 const char *name, *nextname;
1319 struct ctl_dir *dir;
1320 struct ctl_table *entry;
1321 struct ctl_node *node;
1324 for (entry = table; entry->procname; entry++)
1327 header = kzalloc(sizeof(struct ctl_table_header) +
1328 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1332 node = (struct ctl_node *)(header + 1);
1333 init_header(header, root, set, node, table);
1334 if (sysctl_check_table(path, table))
1337 spin_lock(&sysctl_lock);
1339 /* Reference moved down the diretory tree get_subdir */
1341 spin_unlock(&sysctl_lock);
1343 /* Find the directory for the ctl_table */
1344 for (name = path; name; name = nextname) {
1346 nextname = strchr(name, '/');
1348 namelen = nextname - name;
1351 namelen = strlen(name);
1356 dir = get_subdir(dir, name, namelen);
1361 spin_lock(&sysctl_lock);
1362 if (insert_header(dir, header))
1363 goto fail_put_dir_locked;
1365 drop_sysctl_table(&dir->header);
1366 spin_unlock(&sysctl_lock);
1370 fail_put_dir_locked:
1371 drop_sysctl_table(&dir->header);
1372 spin_unlock(&sysctl_lock);
1380 * register_sysctl - register a sysctl table
1381 * @path: The path to the directory the sysctl table is in.
1382 * @table: the table structure
1384 * Register a sysctl table. @table should be a filled in ctl_table
1385 * array. A completely 0 filled entry terminates the table.
1387 * See __register_sysctl_table for more details.
1389 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1391 return __register_sysctl_table(&sysctl_table_root.default_set,
1394 EXPORT_SYMBOL(register_sysctl);
1396 static char *append_path(const char *path, char *pos, const char *name)
1399 namelen = strlen(name);
1400 if (((pos - path) + namelen + 2) >= PATH_MAX)
1402 memcpy(pos, name, namelen);
1404 pos[namelen + 1] = '\0';
1409 static int count_subheaders(struct ctl_table *table)
1412 int nr_subheaders = 0;
1413 struct ctl_table *entry;
1415 /* special case: no directory and empty directory */
1416 if (!table || !table->procname)
1419 for (entry = table; entry->procname; entry++) {
1421 nr_subheaders += count_subheaders(entry->child);
1425 return nr_subheaders + has_files;
1428 static int register_leaf_sysctl_tables(const char *path, char *pos,
1429 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1430 struct ctl_table *table)
1432 struct ctl_table *ctl_table_arg = NULL;
1433 struct ctl_table *entry, *files;
1438 for (entry = table; entry->procname; entry++) {
1446 /* If there are mixed files and directories we need a new table */
1447 if (nr_dirs && nr_files) {
1448 struct ctl_table *new;
1449 files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1454 ctl_table_arg = files;
1455 for (new = files, entry = table; entry->procname; entry++) {
1463 /* Register everything except a directory full of subdirectories */
1464 if (nr_files || !nr_dirs) {
1465 struct ctl_table_header *header;
1466 header = __register_sysctl_table(set, path, files);
1468 kfree(ctl_table_arg);
1472 /* Remember if we need to free the file table */
1473 header->ctl_table_arg = ctl_table_arg;
1474 **subheader = header;
1478 /* Recurse into the subdirectories. */
1479 for (entry = table; entry->procname; entry++) {
1485 err = -ENAMETOOLONG;
1486 child_pos = append_path(path, pos, entry->procname);
1490 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1498 /* On failure our caller will unregister all registered subheaders */
1503 * __register_sysctl_paths - register a sysctl table hierarchy
1504 * @set: Sysctl tree to register on
1505 * @path: The path to the directory the sysctl table is in.
1506 * @table: the top-level table structure
1508 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1509 * array. A completely 0 filled entry terminates the table.
1511 * See __register_sysctl_table for more details.
1513 struct ctl_table_header *__register_sysctl_paths(
1514 struct ctl_table_set *set,
1515 const struct ctl_path *path, struct ctl_table *table)
1517 struct ctl_table *ctl_table_arg = table;
1518 int nr_subheaders = count_subheaders(table);
1519 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1520 const struct ctl_path *component;
1521 char *new_path, *pos;
1523 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1528 for (component = path; component->procname; component++) {
1529 pos = append_path(new_path, pos, component->procname);
1533 while (table->procname && table->child && !table[1].procname) {
1534 pos = append_path(new_path, pos, table->procname);
1537 table = table->child;
1539 if (nr_subheaders == 1) {
1540 header = __register_sysctl_table(set, new_path, table);
1542 header->ctl_table_arg = ctl_table_arg;
1544 header = kzalloc(sizeof(*header) +
1545 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1549 subheaders = (struct ctl_table_header **) (header + 1);
1550 subheader = subheaders;
1551 header->ctl_table_arg = ctl_table_arg;
1553 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1555 goto err_register_leaves;
1562 err_register_leaves:
1563 while (subheader > subheaders) {
1564 struct ctl_table_header *subh = *(--subheader);
1565 struct ctl_table *table = subh->ctl_table_arg;
1566 unregister_sysctl_table(subh);
1575 * register_sysctl_table_path - register a sysctl table hierarchy
1576 * @path: The path to the directory the sysctl table is in.
1577 * @table: the top-level table structure
1579 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1580 * array. A completely 0 filled entry terminates the table.
1582 * See __register_sysctl_paths for more details.
1584 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1585 struct ctl_table *table)
1587 return __register_sysctl_paths(&sysctl_table_root.default_set,
1590 EXPORT_SYMBOL(register_sysctl_paths);
1593 * register_sysctl_table - register a sysctl table hierarchy
1594 * @table: the top-level table structure
1596 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1597 * array. A completely 0 filled entry terminates the table.
1599 * See register_sysctl_paths for more details.
1601 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1603 static const struct ctl_path null_path[] = { {} };
1605 return register_sysctl_paths(null_path, table);
1607 EXPORT_SYMBOL(register_sysctl_table);
1609 static void put_links(struct ctl_table_header *header)
1611 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1612 struct ctl_table_root *root = header->root;
1613 struct ctl_dir *parent = header->parent;
1614 struct ctl_dir *core_parent;
1615 struct ctl_table *entry;
1617 if (header->set == root_set)
1620 core_parent = xlate_dir(root_set, parent);
1621 if (IS_ERR(core_parent))
1624 for (entry = header->ctl_table; entry->procname; entry++) {
1625 struct ctl_table_header *link_head;
1626 struct ctl_table *link;
1627 const char *name = entry->procname;
1629 link = find_entry(&link_head, core_parent, name, strlen(name));
1631 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1632 (S_ISLNK(link->mode) && (link->data == root)))) {
1633 drop_sysctl_table(link_head);
1636 pr_err("sysctl link missing during unregister: ");
1637 sysctl_print_dir(parent);
1638 pr_cont("/%s\n", name);
1643 static void drop_sysctl_table(struct ctl_table_header *header)
1645 struct ctl_dir *parent = header->parent;
1652 start_unregistering(header);
1655 if (!--header->count)
1656 kfree_rcu(header, rcu);
1659 drop_sysctl_table(&parent->header);
1663 * unregister_sysctl_table - unregister a sysctl table hierarchy
1664 * @header: the header returned from register_sysctl_table
1666 * Unregisters the sysctl table and all children. proc entries may not
1667 * actually be removed until they are no longer used by anyone.
1669 void unregister_sysctl_table(struct ctl_table_header * header)
1677 nr_subheaders = count_subheaders(header->ctl_table_arg);
1678 if (unlikely(nr_subheaders > 1)) {
1679 struct ctl_table_header **subheaders;
1682 subheaders = (struct ctl_table_header **)(header + 1);
1683 for (i = nr_subheaders -1; i >= 0; i--) {
1684 struct ctl_table_header *subh = subheaders[i];
1685 struct ctl_table *table = subh->ctl_table_arg;
1686 unregister_sysctl_table(subh);
1693 spin_lock(&sysctl_lock);
1694 drop_sysctl_table(header);
1695 spin_unlock(&sysctl_lock);
1697 EXPORT_SYMBOL(unregister_sysctl_table);
1699 void setup_sysctl_set(struct ctl_table_set *set,
1700 struct ctl_table_root *root,
1701 int (*is_seen)(struct ctl_table_set *))
1703 memset(set, 0, sizeof(*set));
1704 set->is_seen = is_seen;
1705 init_header(&set->dir.header, root, set, NULL, root_table);
1708 void retire_sysctl_set(struct ctl_table_set *set)
1710 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1713 int __init proc_sys_init(void)
1715 struct proc_dir_entry *proc_sys_root;
1717 proc_sys_root = proc_mkdir("sys", NULL);
1718 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1719 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1720 proc_sys_root->nlink = 0;
1722 return sysctl_init();