1 // SPDX-License-Identifier: GPL-2.0
5 #include <linux/init.h>
6 #include <linux/sysctl.h>
7 #include <linux/poll.h>
8 #include <linux/proc_fs.h>
9 #include <linux/printk.h>
10 #include <linux/security.h>
11 #include <linux/sched.h>
12 #include <linux/cred.h>
13 #include <linux/namei.h>
15 #include <linux/module.h>
18 static const struct dentry_operations proc_sys_dentry_operations;
19 static const struct file_operations proc_sys_file_operations;
20 static const struct inode_operations proc_sys_inode_operations;
21 static const struct file_operations proc_sys_dir_file_operations;
22 static const struct inode_operations proc_sys_dir_operations;
24 /* Support for permanently empty directories */
26 struct ctl_table sysctl_mount_point[] = {
30 static bool is_empty_dir(struct ctl_table_header *head)
32 return head->ctl_table[0].child == sysctl_mount_point;
35 static void set_empty_dir(struct ctl_dir *dir)
37 dir->header.ctl_table[0].child = sysctl_mount_point;
40 static void clear_empty_dir(struct ctl_dir *dir)
43 dir->header.ctl_table[0].child = NULL;
46 void proc_sys_poll_notify(struct ctl_table_poll *poll)
51 atomic_inc(&poll->event);
52 wake_up_interruptible(&poll->wait);
55 static struct ctl_table root_table[] = {
58 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
62 static struct ctl_table_root sysctl_table_root = {
63 .default_set.dir.header = {
66 .ctl_table = root_table }},
67 .ctl_table_arg = root_table,
68 .root = &sysctl_table_root,
69 .set = &sysctl_table_root.default_set,
73 static DEFINE_SPINLOCK(sysctl_lock);
75 static void drop_sysctl_table(struct ctl_table_header *header);
76 static int sysctl_follow_link(struct ctl_table_header **phead,
77 struct ctl_table **pentry);
78 static int insert_links(struct ctl_table_header *head);
79 static void put_links(struct ctl_table_header *header);
81 static void sysctl_print_dir(struct ctl_dir *dir)
83 if (dir->header.parent)
84 sysctl_print_dir(dir->header.parent);
85 pr_cont("%s/", dir->header.ctl_table[0].procname);
88 static int namecmp(const char *name1, int len1, const char *name2, int len2)
97 cmp = memcmp(name1, name2, minlen);
103 /* Called under sysctl_lock */
104 static struct ctl_table *find_entry(struct ctl_table_header **phead,
105 struct ctl_dir *dir, const char *name, int namelen)
107 struct ctl_table_header *head;
108 struct ctl_table *entry;
109 struct rb_node *node = dir->root.rb_node;
113 struct ctl_node *ctl_node;
114 const char *procname;
117 ctl_node = rb_entry(node, struct ctl_node, node);
118 head = ctl_node->header;
119 entry = &head->ctl_table[ctl_node - head->node];
120 procname = entry->procname;
122 cmp = namecmp(name, namelen, procname, strlen(procname));
124 node = node->rb_left;
126 node = node->rb_right;
135 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
137 struct rb_node *node = &head->node[entry - head->ctl_table].node;
138 struct rb_node **p = &head->parent->root.rb_node;
139 struct rb_node *parent = NULL;
140 const char *name = entry->procname;
141 int namelen = strlen(name);
144 struct ctl_table_header *parent_head;
145 struct ctl_table *parent_entry;
146 struct ctl_node *parent_node;
147 const char *parent_name;
151 parent_node = rb_entry(parent, struct ctl_node, node);
152 parent_head = parent_node->header;
153 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
154 parent_name = parent_entry->procname;
156 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
162 pr_err("sysctl duplicate entry: ");
163 sysctl_print_dir(head->parent);
164 pr_cont("/%s\n", entry->procname);
169 rb_link_node(node, parent, p);
170 rb_insert_color(node, &head->parent->root);
174 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
176 struct rb_node *node = &head->node[entry - head->ctl_table].node;
178 rb_erase(node, &head->parent->root);
181 static void init_header(struct ctl_table_header *head,
182 struct ctl_table_root *root, struct ctl_table_set *set,
183 struct ctl_node *node, struct ctl_table *table)
185 head->ctl_table = table;
186 head->ctl_table_arg = table;
190 head->unregistering = NULL;
195 INIT_HLIST_HEAD(&head->inodes);
197 struct ctl_table *entry;
198 for (entry = table; entry->procname; entry++, node++)
203 static void erase_header(struct ctl_table_header *head)
205 struct ctl_table *entry;
206 for (entry = head->ctl_table; entry->procname; entry++)
207 erase_entry(head, entry);
210 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
212 struct ctl_table *entry;
215 /* Is this a permanently empty directory? */
216 if (is_empty_dir(&dir->header))
219 /* Am I creating a permanently empty directory? */
220 if (header->ctl_table == sysctl_mount_point) {
221 if (!RB_EMPTY_ROOT(&dir->root))
227 header->parent = dir;
228 err = insert_links(header);
231 for (entry = header->ctl_table; entry->procname; entry++) {
232 err = insert_entry(header, entry);
238 erase_header(header);
241 if (header->ctl_table == sysctl_mount_point)
242 clear_empty_dir(dir);
243 header->parent = NULL;
244 drop_sysctl_table(&dir->header);
248 /* called under sysctl_lock */
249 static int use_table(struct ctl_table_header *p)
251 if (unlikely(p->unregistering))
257 /* called under sysctl_lock */
258 static void unuse_table(struct ctl_table_header *p)
261 if (unlikely(p->unregistering))
262 complete(p->unregistering);
265 static void proc_sys_prune_dcache(struct ctl_table_header *head)
268 struct proc_inode *ei;
269 struct hlist_node *node;
270 struct super_block *sb;
274 node = hlist_first_rcu(&head->inodes);
277 ei = hlist_entry(node, struct proc_inode, sysctl_inodes);
278 spin_lock(&sysctl_lock);
279 hlist_del_init_rcu(&ei->sysctl_inodes);
280 spin_unlock(&sysctl_lock);
282 inode = &ei->vfs_inode;
284 if (!atomic_inc_not_zero(&sb->s_active))
286 inode = igrab(inode);
288 if (unlikely(!inode)) {
289 deactivate_super(sb);
294 d_prune_aliases(inode);
296 deactivate_super(sb);
303 /* called under sysctl_lock, will reacquire if has to wait */
304 static void start_unregistering(struct ctl_table_header *p)
307 * if p->used is 0, nobody will ever touch that entry again;
308 * we'll eliminate all paths to it before dropping sysctl_lock
310 if (unlikely(p->used)) {
311 struct completion wait;
312 init_completion(&wait);
313 p->unregistering = &wait;
314 spin_unlock(&sysctl_lock);
315 wait_for_completion(&wait);
317 /* anything non-NULL; we'll never dereference it */
318 p->unregistering = ERR_PTR(-EINVAL);
319 spin_unlock(&sysctl_lock);
322 * Prune dentries for unregistered sysctls: namespaced sysctls
323 * can have duplicate names and contaminate dcache very badly.
325 proc_sys_prune_dcache(p);
327 * do not remove from the list until nobody holds it; walking the
328 * list in do_sysctl() relies on that.
330 spin_lock(&sysctl_lock);
334 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
337 spin_lock(&sysctl_lock);
338 if (!use_table(head))
339 head = ERR_PTR(-ENOENT);
340 spin_unlock(&sysctl_lock);
344 static void sysctl_head_finish(struct ctl_table_header *head)
348 spin_lock(&sysctl_lock);
350 spin_unlock(&sysctl_lock);
353 static struct ctl_table_set *
354 lookup_header_set(struct ctl_table_root *root)
356 struct ctl_table_set *set = &root->default_set;
358 set = root->lookup(root);
362 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
364 const char *name, int namelen)
366 struct ctl_table_header *head;
367 struct ctl_table *entry;
369 spin_lock(&sysctl_lock);
370 entry = find_entry(&head, dir, name, namelen);
371 if (entry && use_table(head))
375 spin_unlock(&sysctl_lock);
379 static struct ctl_node *first_usable_entry(struct rb_node *node)
381 struct ctl_node *ctl_node;
383 for (;node; node = rb_next(node)) {
384 ctl_node = rb_entry(node, struct ctl_node, node);
385 if (use_table(ctl_node->header))
391 static void first_entry(struct ctl_dir *dir,
392 struct ctl_table_header **phead, struct ctl_table **pentry)
394 struct ctl_table_header *head = NULL;
395 struct ctl_table *entry = NULL;
396 struct ctl_node *ctl_node;
398 spin_lock(&sysctl_lock);
399 ctl_node = first_usable_entry(rb_first(&dir->root));
400 spin_unlock(&sysctl_lock);
402 head = ctl_node->header;
403 entry = &head->ctl_table[ctl_node - head->node];
409 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
411 struct ctl_table_header *head = *phead;
412 struct ctl_table *entry = *pentry;
413 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
415 spin_lock(&sysctl_lock);
418 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
419 spin_unlock(&sysctl_lock);
422 head = ctl_node->header;
423 entry = &head->ctl_table[ctl_node - head->node];
430 * sysctl_perm does NOT grant the superuser all rights automatically, because
431 * some sysctl variables are readonly even to root.
434 static int test_perm(int mode, int op)
436 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
438 else if (in_egroup_p(GLOBAL_ROOT_GID))
440 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
445 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
447 struct ctl_table_root *root = head->root;
450 if (root->permissions)
451 mode = root->permissions(head, table);
455 return test_perm(mode, op);
458 static struct inode *proc_sys_make_inode(struct super_block *sb,
459 struct ctl_table_header *head, struct ctl_table *table)
461 struct ctl_table_root *root = head->root;
463 struct proc_inode *ei;
465 inode = new_inode(sb);
467 return ERR_PTR(-ENOMEM);
469 inode->i_ino = get_next_ino();
473 spin_lock(&sysctl_lock);
474 if (unlikely(head->unregistering)) {
475 spin_unlock(&sysctl_lock);
477 return ERR_PTR(-ENOENT);
480 ei->sysctl_entry = table;
481 hlist_add_head_rcu(&ei->sysctl_inodes, &head->inodes);
483 spin_unlock(&sysctl_lock);
485 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
486 inode->i_mode = table->mode;
487 if (!S_ISDIR(table->mode)) {
488 inode->i_mode |= S_IFREG;
489 inode->i_op = &proc_sys_inode_operations;
490 inode->i_fop = &proc_sys_file_operations;
492 inode->i_mode |= S_IFDIR;
493 inode->i_op = &proc_sys_dir_operations;
494 inode->i_fop = &proc_sys_dir_file_operations;
495 if (is_empty_dir(head))
496 make_empty_dir_inode(inode);
499 if (root->set_ownership)
500 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
502 inode->i_uid = GLOBAL_ROOT_UID;
503 inode->i_gid = GLOBAL_ROOT_GID;
509 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
511 spin_lock(&sysctl_lock);
512 hlist_del_init_rcu(&PROC_I(inode)->sysctl_inodes);
514 kfree_rcu(head, rcu);
515 spin_unlock(&sysctl_lock);
518 static struct ctl_table_header *grab_header(struct inode *inode)
520 struct ctl_table_header *head = PROC_I(inode)->sysctl;
522 head = &sysctl_table_root.default_set.dir.header;
523 return sysctl_head_grab(head);
526 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
529 struct ctl_table_header *head = grab_header(dir);
530 struct ctl_table_header *h = NULL;
531 const struct qstr *name = &dentry->d_name;
534 struct dentry *err = ERR_PTR(-ENOENT);
535 struct ctl_dir *ctl_dir;
539 return ERR_CAST(head);
541 ctl_dir = container_of(head, struct ctl_dir, header);
543 p = lookup_entry(&h, ctl_dir, name->name, name->len);
547 if (S_ISLNK(p->mode)) {
548 ret = sysctl_follow_link(&h, &p);
554 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
556 err = ERR_CAST(inode);
561 d_set_d_op(dentry, &proc_sys_dentry_operations);
562 d_add(dentry, inode);
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;
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 /* careful: calling conventions are nasty here */
598 error = table->proc_handler(table, write, buf, &res, ppos);
602 sysctl_head_finish(head);
607 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
608 size_t count, loff_t *ppos)
610 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
613 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
614 size_t count, loff_t *ppos)
616 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
619 static int proc_sys_open(struct inode *inode, struct file *filp)
621 struct ctl_table_header *head = grab_header(inode);
622 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
624 /* sysctl was unregistered */
626 return PTR_ERR(head);
629 filp->private_data = proc_sys_poll_event(table->poll);
631 sysctl_head_finish(head);
636 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
638 struct inode *inode = file_inode(filp);
639 struct ctl_table_header *head = grab_header(inode);
640 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
641 unsigned int ret = DEFAULT_POLLMASK;
644 /* sysctl was unregistered */
646 return POLLERR | POLLHUP;
648 if (!table->proc_handler)
654 event = (unsigned long)filp->private_data;
655 poll_wait(filp, &table->poll->wait, wait);
657 if (event != atomic_read(&table->poll->event)) {
658 filp->private_data = proc_sys_poll_event(table->poll);
659 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
663 sysctl_head_finish(head);
668 static bool proc_sys_fill_cache(struct file *file,
669 struct dir_context *ctx,
670 struct ctl_table_header *head,
671 struct ctl_table *table)
673 struct dentry *child, *dir = file->f_path.dentry;
677 unsigned type = DT_UNKNOWN;
679 qname.name = table->procname;
680 qname.len = strlen(table->procname);
681 qname.hash = full_name_hash(dir, qname.name, qname.len);
683 child = d_lookup(dir, &qname);
685 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
686 child = d_alloc_parallel(dir, &qname, &wq);
689 if (d_in_lookup(child)) {
690 inode = proc_sys_make_inode(dir->d_sb, head, table);
692 d_lookup_done(child);
696 d_set_d_op(child, &proc_sys_dentry_operations);
700 inode = d_inode(child);
702 type = inode->i_mode >> 12;
704 return dir_emit(ctx, qname.name, qname.len, ino, type);
707 static bool proc_sys_link_fill_cache(struct file *file,
708 struct dir_context *ctx,
709 struct ctl_table_header *head,
710 struct ctl_table *table)
714 head = sysctl_head_grab(head);
718 if (S_ISLNK(table->mode)) {
719 /* It is not an error if we can not follow the link ignore it */
720 int err = sysctl_follow_link(&head, &table);
725 ret = proc_sys_fill_cache(file, ctx, head, table);
727 sysctl_head_finish(head);
731 static int scan(struct ctl_table_header *head, struct ctl_table *table,
732 unsigned long *pos, struct file *file,
733 struct dir_context *ctx)
737 if ((*pos)++ < ctx->pos)
740 if (unlikely(S_ISLNK(table->mode)))
741 res = proc_sys_link_fill_cache(file, ctx, head, table);
743 res = proc_sys_fill_cache(file, ctx, head, table);
751 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
753 struct ctl_table_header *head = grab_header(file_inode(file));
754 struct ctl_table_header *h = NULL;
755 struct ctl_table *entry;
756 struct ctl_dir *ctl_dir;
760 return PTR_ERR(head);
762 ctl_dir = container_of(head, struct ctl_dir, header);
764 if (!dir_emit_dots(file, ctx))
769 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
770 if (!scan(h, entry, &pos, file, ctx)) {
771 sysctl_head_finish(h);
776 sysctl_head_finish(head);
780 static int proc_sys_permission(struct inode *inode, int mask)
783 * sysctl entries that are not writeable,
784 * are _NOT_ writeable, capabilities or not.
786 struct ctl_table_header *head;
787 struct ctl_table *table;
790 /* Executable files are not allowed under /proc/sys/ */
791 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
794 head = grab_header(inode);
796 return PTR_ERR(head);
798 table = PROC_I(inode)->sysctl_entry;
799 if (!table) /* global root - r-xr-xr-x */
800 error = mask & MAY_WRITE ? -EACCES : 0;
801 else /* Use the permissions on the sysctl table entry */
802 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
804 sysctl_head_finish(head);
808 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
810 struct inode *inode = d_inode(dentry);
813 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
816 error = setattr_prepare(dentry, attr);
820 setattr_copy(inode, attr);
821 mark_inode_dirty(inode);
825 static int proc_sys_getattr(const struct path *path, struct kstat *stat,
826 u32 request_mask, unsigned int query_flags)
828 struct inode *inode = d_inode(path->dentry);
829 struct ctl_table_header *head = grab_header(inode);
830 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
833 return PTR_ERR(head);
835 generic_fillattr(inode, stat);
837 stat->mode = (stat->mode & S_IFMT) | table->mode;
839 sysctl_head_finish(head);
843 static const struct file_operations proc_sys_file_operations = {
844 .open = proc_sys_open,
845 .poll = proc_sys_poll,
846 .read = proc_sys_read,
847 .write = proc_sys_write,
848 .llseek = default_llseek,
851 static const struct file_operations proc_sys_dir_file_operations = {
852 .read = generic_read_dir,
853 .iterate_shared = proc_sys_readdir,
854 .llseek = generic_file_llseek,
857 static const struct inode_operations proc_sys_inode_operations = {
858 .permission = proc_sys_permission,
859 .setattr = proc_sys_setattr,
860 .getattr = proc_sys_getattr,
863 static const struct inode_operations proc_sys_dir_operations = {
864 .lookup = proc_sys_lookup,
865 .permission = proc_sys_permission,
866 .setattr = proc_sys_setattr,
867 .getattr = proc_sys_getattr,
870 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
872 if (flags & LOOKUP_RCU)
874 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
877 static int proc_sys_delete(const struct dentry *dentry)
879 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
882 static int sysctl_is_seen(struct ctl_table_header *p)
884 struct ctl_table_set *set = p->set;
886 spin_lock(&sysctl_lock);
887 if (p->unregistering)
889 else if (!set->is_seen)
892 res = set->is_seen(set);
893 spin_unlock(&sysctl_lock);
897 static int proc_sys_compare(const struct dentry *dentry,
898 unsigned int len, const char *str, const struct qstr *name)
900 struct ctl_table_header *head;
903 /* Although proc doesn't have negative dentries, rcu-walk means
904 * that inode here can be NULL */
905 /* AV: can it, indeed? */
906 inode = d_inode_rcu(dentry);
909 if (name->len != len)
911 if (memcmp(name->name, str, len))
913 head = rcu_dereference(PROC_I(inode)->sysctl);
914 return !head || !sysctl_is_seen(head);
917 static const struct dentry_operations proc_sys_dentry_operations = {
918 .d_revalidate = proc_sys_revalidate,
919 .d_delete = proc_sys_delete,
920 .d_compare = proc_sys_compare,
923 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
924 const char *name, int namelen)
926 struct ctl_table_header *head;
927 struct ctl_table *entry;
929 entry = find_entry(&head, dir, name, namelen);
931 return ERR_PTR(-ENOENT);
932 if (!S_ISDIR(entry->mode))
933 return ERR_PTR(-ENOTDIR);
934 return container_of(head, struct ctl_dir, header);
937 static struct ctl_dir *new_dir(struct ctl_table_set *set,
938 const char *name, int namelen)
940 struct ctl_table *table;
942 struct ctl_node *node;
945 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
946 sizeof(struct ctl_table)*2 + namelen + 1,
951 node = (struct ctl_node *)(new + 1);
952 table = (struct ctl_table *)(node + 1);
953 new_name = (char *)(table + 2);
954 memcpy(new_name, name, namelen);
955 new_name[namelen] = '\0';
956 table[0].procname = new_name;
957 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
958 init_header(&new->header, set->dir.header.root, set, node, table);
964 * get_subdir - find or create a subdir with the specified name.
965 * @dir: Directory to create the subdirectory in
966 * @name: The name of the subdirectory to find or create
967 * @namelen: The length of name
969 * Takes a directory with an elevated reference count so we know that
970 * if we drop the lock the directory will not go away. Upon success
971 * the reference is moved from @dir to the returned subdirectory.
972 * Upon error an error code is returned and the reference on @dir is
975 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
976 const char *name, int namelen)
978 struct ctl_table_set *set = dir->header.set;
979 struct ctl_dir *subdir, *new = NULL;
982 spin_lock(&sysctl_lock);
983 subdir = find_subdir(dir, name, namelen);
986 if (PTR_ERR(subdir) != -ENOENT)
989 spin_unlock(&sysctl_lock);
990 new = new_dir(set, name, namelen);
991 spin_lock(&sysctl_lock);
992 subdir = ERR_PTR(-ENOMEM);
996 /* Was the subdir added while we dropped the lock? */
997 subdir = find_subdir(dir, name, namelen);
1000 if (PTR_ERR(subdir) != -ENOENT)
1003 /* Nope. Use the our freshly made directory entry. */
1004 err = insert_header(dir, &new->header);
1005 subdir = ERR_PTR(err);
1010 subdir->header.nreg++;
1012 if (IS_ERR(subdir)) {
1013 pr_err("sysctl could not get directory: ");
1014 sysctl_print_dir(dir);
1015 pr_cont("/%*.*s %ld\n",
1016 namelen, namelen, name, PTR_ERR(subdir));
1018 drop_sysctl_table(&dir->header);
1020 drop_sysctl_table(&new->header);
1021 spin_unlock(&sysctl_lock);
1025 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1027 struct ctl_dir *parent;
1028 const char *procname;
1029 if (!dir->header.parent)
1031 parent = xlate_dir(set, dir->header.parent);
1034 procname = dir->header.ctl_table[0].procname;
1035 return find_subdir(parent, procname, strlen(procname));
1038 static int sysctl_follow_link(struct ctl_table_header **phead,
1039 struct ctl_table **pentry)
1041 struct ctl_table_header *head;
1042 struct ctl_table_root *root;
1043 struct ctl_table_set *set;
1044 struct ctl_table *entry;
1045 struct ctl_dir *dir;
1049 spin_lock(&sysctl_lock);
1050 root = (*pentry)->data;
1051 set = lookup_header_set(root);
1052 dir = xlate_dir(set, (*phead)->parent);
1056 const char *procname = (*pentry)->procname;
1058 entry = find_entry(&head, dir, procname, strlen(procname));
1060 if (entry && use_table(head)) {
1061 unuse_table(*phead);
1068 spin_unlock(&sysctl_lock);
1072 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1074 struct va_format vaf;
1077 va_start(args, fmt);
1081 pr_err("sysctl table check failed: %s/%s %pV\n",
1082 path, table->procname, &vaf);
1088 static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1092 if ((table->proc_handler == proc_douintvec) ||
1093 (table->proc_handler == proc_douintvec_minmax)) {
1094 if (table->maxlen != sizeof(unsigned int))
1095 err |= sysctl_err(path, table, "array now allowed");
1101 static int sysctl_check_table(const char *path, struct ctl_table *table)
1104 for (; table->procname; table++) {
1106 err |= sysctl_err(path, table, "Not a file");
1108 if ((table->proc_handler == proc_dostring) ||
1109 (table->proc_handler == proc_dointvec) ||
1110 (table->proc_handler == proc_douintvec) ||
1111 (table->proc_handler == proc_douintvec_minmax) ||
1112 (table->proc_handler == proc_dointvec_minmax) ||
1113 (table->proc_handler == proc_dointvec_jiffies) ||
1114 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1115 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1116 (table->proc_handler == proc_doulongvec_minmax) ||
1117 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1119 err |= sysctl_err(path, table, "No data");
1121 err |= sysctl_err(path, table, "No maxlen");
1123 err |= sysctl_check_table_array(path, table);
1125 if (!table->proc_handler)
1126 err |= sysctl_err(path, table, "No proc_handler");
1128 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1129 err |= sysctl_err(path, table, "bogus .mode 0%o",
1135 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1136 struct ctl_table_root *link_root)
1138 struct ctl_table *link_table, *entry, *link;
1139 struct ctl_table_header *links;
1140 struct ctl_node *node;
1142 int nr_entries, name_bytes;
1146 for (entry = table; entry->procname; entry++) {
1148 name_bytes += strlen(entry->procname) + 1;
1151 links = kzalloc(sizeof(struct ctl_table_header) +
1152 sizeof(struct ctl_node)*nr_entries +
1153 sizeof(struct ctl_table)*(nr_entries + 1) +
1160 node = (struct ctl_node *)(links + 1);
1161 link_table = (struct ctl_table *)(node + nr_entries);
1162 link_name = (char *)&link_table[nr_entries + 1];
1164 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1165 int len = strlen(entry->procname) + 1;
1166 memcpy(link_name, entry->procname, len);
1167 link->procname = link_name;
1168 link->mode = S_IFLNK|S_IRWXUGO;
1169 link->data = link_root;
1172 init_header(links, dir->header.root, dir->header.set, node, link_table);
1173 links->nreg = nr_entries;
1178 static bool get_links(struct ctl_dir *dir,
1179 struct ctl_table *table, struct ctl_table_root *link_root)
1181 struct ctl_table_header *head;
1182 struct ctl_table *entry, *link;
1184 /* Are there links available for every entry in table? */
1185 for (entry = table; entry->procname; entry++) {
1186 const char *procname = entry->procname;
1187 link = find_entry(&head, dir, procname, strlen(procname));
1190 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1192 if (S_ISLNK(link->mode) && (link->data == link_root))
1197 /* The checks passed. Increase the registration count on the links */
1198 for (entry = table; entry->procname; entry++) {
1199 const char *procname = entry->procname;
1200 link = find_entry(&head, dir, procname, strlen(procname));
1206 static int insert_links(struct ctl_table_header *head)
1208 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1209 struct ctl_dir *core_parent = NULL;
1210 struct ctl_table_header *links;
1213 if (head->set == root_set)
1216 core_parent = xlate_dir(root_set, head->parent);
1217 if (IS_ERR(core_parent))
1220 if (get_links(core_parent, head->ctl_table, head->root))
1223 core_parent->header.nreg++;
1224 spin_unlock(&sysctl_lock);
1226 links = new_links(core_parent, head->ctl_table, head->root);
1228 spin_lock(&sysctl_lock);
1234 if (get_links(core_parent, head->ctl_table, head->root)) {
1239 err = insert_header(core_parent, links);
1243 drop_sysctl_table(&core_parent->header);
1248 * __register_sysctl_table - register a leaf sysctl table
1249 * @set: Sysctl tree to register on
1250 * @path: The path to the directory the sysctl table is in.
1251 * @table: the top-level table structure
1253 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1254 * array. A completely 0 filled entry terminates the table.
1256 * The members of the &struct ctl_table structure are used as follows:
1258 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1259 * enter a sysctl file
1261 * data - a pointer to data for use by proc_handler
1263 * maxlen - the maximum size in bytes of the data
1265 * mode - the file permissions for the /proc/sys file
1267 * child - must be %NULL.
1269 * proc_handler - the text handler routine (described below)
1271 * extra1, extra2 - extra pointers usable by the proc handler routines
1273 * Leaf nodes in the sysctl tree will be represented by a single file
1274 * under /proc; non-leaf nodes will be represented by directories.
1276 * There must be a proc_handler routine for any terminal nodes.
1277 * Several default handlers are available to cover common cases -
1279 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1280 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1281 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1283 * It is the handler's job to read the input buffer from user memory
1284 * and process it. The handler should return 0 on success.
1286 * This routine returns %NULL on a failure to register, and a pointer
1287 * to the table header on success.
1289 struct ctl_table_header *__register_sysctl_table(
1290 struct ctl_table_set *set,
1291 const char *path, struct ctl_table *table)
1293 struct ctl_table_root *root = set->dir.header.root;
1294 struct ctl_table_header *header;
1295 const char *name, *nextname;
1296 struct ctl_dir *dir;
1297 struct ctl_table *entry;
1298 struct ctl_node *node;
1301 for (entry = table; entry->procname; entry++)
1304 header = kzalloc(sizeof(struct ctl_table_header) +
1305 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1309 node = (struct ctl_node *)(header + 1);
1310 init_header(header, root, set, node, table);
1311 if (sysctl_check_table(path, table))
1314 spin_lock(&sysctl_lock);
1316 /* Reference moved down the diretory tree get_subdir */
1318 spin_unlock(&sysctl_lock);
1320 /* Find the directory for the ctl_table */
1321 for (name = path; name; name = nextname) {
1323 nextname = strchr(name, '/');
1325 namelen = nextname - name;
1328 namelen = strlen(name);
1333 dir = get_subdir(dir, name, namelen);
1338 spin_lock(&sysctl_lock);
1339 if (insert_header(dir, header))
1340 goto fail_put_dir_locked;
1342 drop_sysctl_table(&dir->header);
1343 spin_unlock(&sysctl_lock);
1347 fail_put_dir_locked:
1348 drop_sysctl_table(&dir->header);
1349 spin_unlock(&sysctl_lock);
1357 * register_sysctl - register a sysctl table
1358 * @path: The path to the directory the sysctl table is in.
1359 * @table: the table structure
1361 * Register a sysctl table. @table should be a filled in ctl_table
1362 * array. A completely 0 filled entry terminates the table.
1364 * See __register_sysctl_table for more details.
1366 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1368 return __register_sysctl_table(&sysctl_table_root.default_set,
1371 EXPORT_SYMBOL(register_sysctl);
1373 static char *append_path(const char *path, char *pos, const char *name)
1376 namelen = strlen(name);
1377 if (((pos - path) + namelen + 2) >= PATH_MAX)
1379 memcpy(pos, name, namelen);
1381 pos[namelen + 1] = '\0';
1386 static int count_subheaders(struct ctl_table *table)
1389 int nr_subheaders = 0;
1390 struct ctl_table *entry;
1392 /* special case: no directory and empty directory */
1393 if (!table || !table->procname)
1396 for (entry = table; entry->procname; entry++) {
1398 nr_subheaders += count_subheaders(entry->child);
1402 return nr_subheaders + has_files;
1405 static int register_leaf_sysctl_tables(const char *path, char *pos,
1406 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1407 struct ctl_table *table)
1409 struct ctl_table *ctl_table_arg = NULL;
1410 struct ctl_table *entry, *files;
1415 for (entry = table; entry->procname; entry++) {
1423 /* If there are mixed files and directories we need a new table */
1424 if (nr_dirs && nr_files) {
1425 struct ctl_table *new;
1426 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1431 ctl_table_arg = files;
1432 for (new = files, entry = table; entry->procname; entry++) {
1440 /* Register everything except a directory full of subdirectories */
1441 if (nr_files || !nr_dirs) {
1442 struct ctl_table_header *header;
1443 header = __register_sysctl_table(set, path, files);
1445 kfree(ctl_table_arg);
1449 /* Remember if we need to free the file table */
1450 header->ctl_table_arg = ctl_table_arg;
1451 **subheader = header;
1455 /* Recurse into the subdirectories. */
1456 for (entry = table; entry->procname; entry++) {
1462 err = -ENAMETOOLONG;
1463 child_pos = append_path(path, pos, entry->procname);
1467 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1475 /* On failure our caller will unregister all registered subheaders */
1480 * __register_sysctl_paths - register a sysctl table hierarchy
1481 * @set: Sysctl tree to register on
1482 * @path: The path to the directory the sysctl table is in.
1483 * @table: the top-level table structure
1485 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1486 * array. A completely 0 filled entry terminates the table.
1488 * See __register_sysctl_table for more details.
1490 struct ctl_table_header *__register_sysctl_paths(
1491 struct ctl_table_set *set,
1492 const struct ctl_path *path, struct ctl_table *table)
1494 struct ctl_table *ctl_table_arg = table;
1495 int nr_subheaders = count_subheaders(table);
1496 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1497 const struct ctl_path *component;
1498 char *new_path, *pos;
1500 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1505 for (component = path; component->procname; component++) {
1506 pos = append_path(new_path, pos, component->procname);
1510 while (table->procname && table->child && !table[1].procname) {
1511 pos = append_path(new_path, pos, table->procname);
1514 table = table->child;
1516 if (nr_subheaders == 1) {
1517 header = __register_sysctl_table(set, new_path, table);
1519 header->ctl_table_arg = ctl_table_arg;
1521 header = kzalloc(sizeof(*header) +
1522 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1526 subheaders = (struct ctl_table_header **) (header + 1);
1527 subheader = subheaders;
1528 header->ctl_table_arg = ctl_table_arg;
1530 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1532 goto err_register_leaves;
1539 err_register_leaves:
1540 while (subheader > subheaders) {
1541 struct ctl_table_header *subh = *(--subheader);
1542 struct ctl_table *table = subh->ctl_table_arg;
1543 unregister_sysctl_table(subh);
1552 * register_sysctl_table_path - register a sysctl table hierarchy
1553 * @path: The path to the directory the sysctl table is in.
1554 * @table: the top-level table structure
1556 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1557 * array. A completely 0 filled entry terminates the table.
1559 * See __register_sysctl_paths for more details.
1561 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1562 struct ctl_table *table)
1564 return __register_sysctl_paths(&sysctl_table_root.default_set,
1567 EXPORT_SYMBOL(register_sysctl_paths);
1570 * register_sysctl_table - register a sysctl table hierarchy
1571 * @table: the top-level table structure
1573 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1574 * array. A completely 0 filled entry terminates the table.
1576 * See register_sysctl_paths for more details.
1578 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1580 static const struct ctl_path null_path[] = { {} };
1582 return register_sysctl_paths(null_path, table);
1584 EXPORT_SYMBOL(register_sysctl_table);
1586 static void put_links(struct ctl_table_header *header)
1588 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1589 struct ctl_table_root *root = header->root;
1590 struct ctl_dir *parent = header->parent;
1591 struct ctl_dir *core_parent;
1592 struct ctl_table *entry;
1594 if (header->set == root_set)
1597 core_parent = xlate_dir(root_set, parent);
1598 if (IS_ERR(core_parent))
1601 for (entry = header->ctl_table; entry->procname; entry++) {
1602 struct ctl_table_header *link_head;
1603 struct ctl_table *link;
1604 const char *name = entry->procname;
1606 link = find_entry(&link_head, core_parent, name, strlen(name));
1608 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1609 (S_ISLNK(link->mode) && (link->data == root)))) {
1610 drop_sysctl_table(link_head);
1613 pr_err("sysctl link missing during unregister: ");
1614 sysctl_print_dir(parent);
1615 pr_cont("/%s\n", name);
1620 static void drop_sysctl_table(struct ctl_table_header *header)
1622 struct ctl_dir *parent = header->parent;
1629 start_unregistering(header);
1632 if (!--header->count)
1633 kfree_rcu(header, rcu);
1636 drop_sysctl_table(&parent->header);
1640 * unregister_sysctl_table - unregister a sysctl table hierarchy
1641 * @header: the header returned from register_sysctl_table
1643 * Unregisters the sysctl table and all children. proc entries may not
1644 * actually be removed until they are no longer used by anyone.
1646 void unregister_sysctl_table(struct ctl_table_header * header)
1654 nr_subheaders = count_subheaders(header->ctl_table_arg);
1655 if (unlikely(nr_subheaders > 1)) {
1656 struct ctl_table_header **subheaders;
1659 subheaders = (struct ctl_table_header **)(header + 1);
1660 for (i = nr_subheaders -1; i >= 0; i--) {
1661 struct ctl_table_header *subh = subheaders[i];
1662 struct ctl_table *table = subh->ctl_table_arg;
1663 unregister_sysctl_table(subh);
1670 spin_lock(&sysctl_lock);
1671 drop_sysctl_table(header);
1672 spin_unlock(&sysctl_lock);
1674 EXPORT_SYMBOL(unregister_sysctl_table);
1676 void setup_sysctl_set(struct ctl_table_set *set,
1677 struct ctl_table_root *root,
1678 int (*is_seen)(struct ctl_table_set *))
1680 memset(set, 0, sizeof(*set));
1681 set->is_seen = is_seen;
1682 init_header(&set->dir.header, root, set, NULL, root_table);
1685 void retire_sysctl_set(struct ctl_table_set *set)
1687 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1690 int __init proc_sys_init(void)
1692 struct proc_dir_entry *proc_sys_root;
1694 proc_sys_root = proc_mkdir("sys", NULL);
1695 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1696 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1697 proc_sys_root->nlink = 0;
1699 return sysctl_init();