4 #include <linux/init.h>
5 #include <linux/sysctl.h>
6 #include <linux/poll.h>
7 #include <linux/proc_fs.h>
8 #include <linux/printk.h>
9 #include <linux/security.h>
10 #include <linux/sched.h>
11 #include <linux/namei.h>
13 #include <linux/module.h>
16 static const struct dentry_operations proc_sys_dentry_operations;
17 static const struct file_operations proc_sys_file_operations;
18 static const struct inode_operations proc_sys_inode_operations;
19 static const struct file_operations proc_sys_dir_file_operations;
20 static const struct inode_operations proc_sys_dir_operations;
22 /* Support for permanently empty directories */
24 struct ctl_table sysctl_mount_point[] = {
28 static bool is_empty_dir(struct ctl_table_header *head)
30 return head->ctl_table[0].child == sysctl_mount_point;
33 static void set_empty_dir(struct ctl_dir *dir)
35 dir->header.ctl_table[0].child = sysctl_mount_point;
38 static void clear_empty_dir(struct ctl_dir *dir)
41 dir->header.ctl_table[0].child = NULL;
44 void proc_sys_poll_notify(struct ctl_table_poll *poll)
49 atomic_inc(&poll->event);
50 wake_up_interruptible(&poll->wait);
53 static struct ctl_table root_table[] = {
56 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
60 static struct ctl_table_root sysctl_table_root = {
61 .default_set.dir.header = {
64 .ctl_table = root_table }},
65 .ctl_table_arg = root_table,
66 .root = &sysctl_table_root,
67 .set = &sysctl_table_root.default_set,
71 static DEFINE_SPINLOCK(sysctl_lock);
73 static void drop_sysctl_table(struct ctl_table_header *header);
74 static int sysctl_follow_link(struct ctl_table_header **phead,
75 struct ctl_table **pentry);
76 static int insert_links(struct ctl_table_header *head);
77 static void put_links(struct ctl_table_header *header);
79 static void sysctl_print_dir(struct ctl_dir *dir)
81 if (dir->header.parent)
82 sysctl_print_dir(dir->header.parent);
83 pr_cont("%s/", dir->header.ctl_table[0].procname);
86 static int namecmp(const char *name1, int len1, const char *name2, int len2)
95 cmp = memcmp(name1, name2, minlen);
101 /* Called under sysctl_lock */
102 static struct ctl_table *find_entry(struct ctl_table_header **phead,
103 struct ctl_dir *dir, const char *name, int namelen)
105 struct ctl_table_header *head;
106 struct ctl_table *entry;
107 struct rb_node *node = dir->root.rb_node;
111 struct ctl_node *ctl_node;
112 const char *procname;
115 ctl_node = rb_entry(node, struct ctl_node, node);
116 head = ctl_node->header;
117 entry = &head->ctl_table[ctl_node - head->node];
118 procname = entry->procname;
120 cmp = namecmp(name, namelen, procname, strlen(procname));
122 node = node->rb_left;
124 node = node->rb_right;
133 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
135 struct rb_node *node = &head->node[entry - head->ctl_table].node;
136 struct rb_node **p = &head->parent->root.rb_node;
137 struct rb_node *parent = NULL;
138 const char *name = entry->procname;
139 int namelen = strlen(name);
142 struct ctl_table_header *parent_head;
143 struct ctl_table *parent_entry;
144 struct ctl_node *parent_node;
145 const char *parent_name;
149 parent_node = rb_entry(parent, struct ctl_node, node);
150 parent_head = parent_node->header;
151 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
152 parent_name = parent_entry->procname;
154 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
160 pr_err("sysctl duplicate entry: ");
161 sysctl_print_dir(head->parent);
162 pr_cont("/%s\n", entry->procname);
167 rb_link_node(node, parent, p);
168 rb_insert_color(node, &head->parent->root);
172 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
174 struct rb_node *node = &head->node[entry - head->ctl_table].node;
176 rb_erase(node, &head->parent->root);
179 static void init_header(struct ctl_table_header *head,
180 struct ctl_table_root *root, struct ctl_table_set *set,
181 struct ctl_node *node, struct ctl_table *table)
183 head->ctl_table = table;
184 head->ctl_table_arg = table;
188 head->unregistering = NULL;
194 struct ctl_table *entry;
195 for (entry = table; entry->procname; entry++, node++)
200 static void erase_header(struct ctl_table_header *head)
202 struct ctl_table *entry;
203 for (entry = head->ctl_table; entry->procname; entry++)
204 erase_entry(head, entry);
207 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
209 struct ctl_table *entry;
212 /* Is this a permanently empty directory? */
213 if (is_empty_dir(&dir->header))
216 /* Am I creating a permanently empty directory? */
217 if (header->ctl_table == sysctl_mount_point) {
218 if (!RB_EMPTY_ROOT(&dir->root))
224 header->parent = dir;
225 err = insert_links(header);
228 for (entry = header->ctl_table; entry->procname; entry++) {
229 err = insert_entry(header, entry);
235 erase_header(header);
238 if (header->ctl_table == sysctl_mount_point)
239 clear_empty_dir(dir);
240 header->parent = NULL;
241 drop_sysctl_table(&dir->header);
245 /* called under sysctl_lock */
246 static int use_table(struct ctl_table_header *p)
248 if (unlikely(p->unregistering))
254 /* called under sysctl_lock */
255 static void unuse_table(struct ctl_table_header *p)
258 if (unlikely(p->unregistering))
259 complete(p->unregistering);
262 /* called under sysctl_lock, will reacquire if has to wait */
263 static void start_unregistering(struct ctl_table_header *p)
266 * if p->used is 0, nobody will ever touch that entry again;
267 * we'll eliminate all paths to it before dropping sysctl_lock
269 if (unlikely(p->used)) {
270 struct completion wait;
271 init_completion(&wait);
272 p->unregistering = &wait;
273 spin_unlock(&sysctl_lock);
274 wait_for_completion(&wait);
275 spin_lock(&sysctl_lock);
277 /* anything non-NULL; we'll never dereference it */
278 p->unregistering = ERR_PTR(-EINVAL);
281 * do not remove from the list until nobody holds it; walking the
282 * list in do_sysctl() relies on that.
287 static void sysctl_head_get(struct ctl_table_header *head)
289 spin_lock(&sysctl_lock);
291 spin_unlock(&sysctl_lock);
294 void sysctl_head_put(struct ctl_table_header *head)
296 spin_lock(&sysctl_lock);
298 kfree_rcu(head, rcu);
299 spin_unlock(&sysctl_lock);
302 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
305 spin_lock(&sysctl_lock);
306 if (!use_table(head))
307 head = ERR_PTR(-ENOENT);
308 spin_unlock(&sysctl_lock);
312 static void sysctl_head_finish(struct ctl_table_header *head)
316 spin_lock(&sysctl_lock);
318 spin_unlock(&sysctl_lock);
321 static struct ctl_table_set *
322 lookup_header_set(struct ctl_table_root *root)
324 struct ctl_table_set *set = &root->default_set;
326 set = root->lookup(root);
330 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
332 const char *name, int namelen)
334 struct ctl_table_header *head;
335 struct ctl_table *entry;
337 spin_lock(&sysctl_lock);
338 entry = find_entry(&head, dir, name, namelen);
339 if (entry && use_table(head))
343 spin_unlock(&sysctl_lock);
347 static struct ctl_node *first_usable_entry(struct rb_node *node)
349 struct ctl_node *ctl_node;
351 for (;node; node = rb_next(node)) {
352 ctl_node = rb_entry(node, struct ctl_node, node);
353 if (use_table(ctl_node->header))
359 static void first_entry(struct ctl_dir *dir,
360 struct ctl_table_header **phead, struct ctl_table **pentry)
362 struct ctl_table_header *head = NULL;
363 struct ctl_table *entry = NULL;
364 struct ctl_node *ctl_node;
366 spin_lock(&sysctl_lock);
367 ctl_node = first_usable_entry(rb_first(&dir->root));
368 spin_unlock(&sysctl_lock);
370 head = ctl_node->header;
371 entry = &head->ctl_table[ctl_node - head->node];
377 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
379 struct ctl_table_header *head = *phead;
380 struct ctl_table *entry = *pentry;
381 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
383 spin_lock(&sysctl_lock);
386 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
387 spin_unlock(&sysctl_lock);
390 head = ctl_node->header;
391 entry = &head->ctl_table[ctl_node - head->node];
397 void register_sysctl_root(struct ctl_table_root *root)
402 * sysctl_perm does NOT grant the superuser all rights automatically, because
403 * some sysctl variables are readonly even to root.
406 static int test_perm(int mode, int op)
408 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
410 else if (in_egroup_p(GLOBAL_ROOT_GID))
412 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
417 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
419 struct ctl_table_root *root = head->root;
422 if (root->permissions)
423 mode = root->permissions(head, table);
427 return test_perm(mode, op);
430 static struct inode *proc_sys_make_inode(struct super_block *sb,
431 struct ctl_table_header *head, struct ctl_table *table)
433 struct ctl_table_root *root = head->root;
435 struct proc_inode *ei;
437 inode = new_inode(sb);
441 inode->i_ino = get_next_ino();
443 sysctl_head_get(head);
446 ei->sysctl_entry = table;
448 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
449 inode->i_mode = table->mode;
450 if (!S_ISDIR(table->mode)) {
451 inode->i_mode |= S_IFREG;
452 inode->i_op = &proc_sys_inode_operations;
453 inode->i_fop = &proc_sys_file_operations;
455 inode->i_mode |= S_IFDIR;
456 inode->i_op = &proc_sys_dir_operations;
457 inode->i_fop = &proc_sys_dir_file_operations;
458 if (is_empty_dir(head))
459 make_empty_dir_inode(inode);
462 if (root->set_ownership)
463 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
469 static struct ctl_table_header *grab_header(struct inode *inode)
471 struct ctl_table_header *head = PROC_I(inode)->sysctl;
473 head = &sysctl_table_root.default_set.dir.header;
474 return sysctl_head_grab(head);
477 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
480 struct ctl_table_header *head = grab_header(dir);
481 struct ctl_table_header *h = NULL;
482 const struct qstr *name = &dentry->d_name;
485 struct dentry *err = ERR_PTR(-ENOENT);
486 struct ctl_dir *ctl_dir;
490 return ERR_CAST(head);
492 ctl_dir = container_of(head, struct ctl_dir, header);
494 p = lookup_entry(&h, ctl_dir, name->name, name->len);
498 if (S_ISLNK(p->mode)) {
499 ret = sysctl_follow_link(&h, &p);
505 err = ERR_PTR(-ENOMEM);
506 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
511 d_set_d_op(dentry, &proc_sys_dentry_operations);
512 d_add(dentry, inode);
516 sysctl_head_finish(h);
517 sysctl_head_finish(head);
521 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
522 size_t count, loff_t *ppos, int write)
524 struct inode *inode = file_inode(filp);
525 struct ctl_table_header *head = grab_header(inode);
526 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
531 return PTR_ERR(head);
534 * At this point we know that the sysctl was not unregistered
535 * and won't be until we finish.
538 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
541 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
543 if (!table->proc_handler)
546 /* careful: calling conventions are nasty here */
548 error = table->proc_handler(table, write, buf, &res, ppos);
552 sysctl_head_finish(head);
557 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
558 size_t count, loff_t *ppos)
560 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
563 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
564 size_t count, loff_t *ppos)
566 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
569 static int proc_sys_open(struct inode *inode, struct file *filp)
571 struct ctl_table_header *head = grab_header(inode);
572 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
574 /* sysctl was unregistered */
576 return PTR_ERR(head);
579 filp->private_data = proc_sys_poll_event(table->poll);
581 sysctl_head_finish(head);
586 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
588 struct inode *inode = file_inode(filp);
589 struct ctl_table_header *head = grab_header(inode);
590 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
591 unsigned int ret = DEFAULT_POLLMASK;
594 /* sysctl was unregistered */
596 return POLLERR | POLLHUP;
598 if (!table->proc_handler)
604 event = (unsigned long)filp->private_data;
605 poll_wait(filp, &table->poll->wait, wait);
607 if (event != atomic_read(&table->poll->event)) {
608 filp->private_data = proc_sys_poll_event(table->poll);
609 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
613 sysctl_head_finish(head);
618 static bool proc_sys_fill_cache(struct file *file,
619 struct dir_context *ctx,
620 struct ctl_table_header *head,
621 struct ctl_table *table)
623 struct dentry *child, *dir = file->f_path.dentry;
627 unsigned type = DT_UNKNOWN;
629 qname.name = table->procname;
630 qname.len = strlen(table->procname);
631 qname.hash = full_name_hash(dir, qname.name, qname.len);
633 child = d_lookup(dir, &qname);
635 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
636 child = d_alloc_parallel(dir, &qname, &wq);
639 if (d_in_lookup(child)) {
640 inode = proc_sys_make_inode(dir->d_sb, head, table);
642 d_lookup_done(child);
646 d_set_d_op(child, &proc_sys_dentry_operations);
650 inode = d_inode(child);
652 type = inode->i_mode >> 12;
654 return dir_emit(ctx, qname.name, qname.len, ino, type);
657 static bool proc_sys_link_fill_cache(struct file *file,
658 struct dir_context *ctx,
659 struct ctl_table_header *head,
660 struct ctl_table *table)
663 head = sysctl_head_grab(head);
665 if (S_ISLNK(table->mode)) {
666 /* It is not an error if we can not follow the link ignore it */
667 int err = sysctl_follow_link(&head, &table);
672 ret = proc_sys_fill_cache(file, ctx, head, table);
674 sysctl_head_finish(head);
678 static int scan(struct ctl_table_header *head, struct ctl_table *table,
679 unsigned long *pos, struct file *file,
680 struct dir_context *ctx)
684 if ((*pos)++ < ctx->pos)
687 if (unlikely(S_ISLNK(table->mode)))
688 res = proc_sys_link_fill_cache(file, ctx, head, table);
690 res = proc_sys_fill_cache(file, ctx, head, table);
698 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
700 struct ctl_table_header *head = grab_header(file_inode(file));
701 struct ctl_table_header *h = NULL;
702 struct ctl_table *entry;
703 struct ctl_dir *ctl_dir;
707 return PTR_ERR(head);
709 ctl_dir = container_of(head, struct ctl_dir, header);
711 if (!dir_emit_dots(file, ctx))
716 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
717 if (!scan(h, entry, &pos, file, ctx)) {
718 sysctl_head_finish(h);
723 sysctl_head_finish(head);
727 static int proc_sys_permission(struct inode *inode, int mask)
730 * sysctl entries that are not writeable,
731 * are _NOT_ writeable, capabilities or not.
733 struct ctl_table_header *head;
734 struct ctl_table *table;
737 /* Executable files are not allowed under /proc/sys/ */
738 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
741 head = grab_header(inode);
743 return PTR_ERR(head);
745 table = PROC_I(inode)->sysctl_entry;
746 if (!table) /* global root - r-xr-xr-x */
747 error = mask & MAY_WRITE ? -EACCES : 0;
748 else /* Use the permissions on the sysctl table entry */
749 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
751 sysctl_head_finish(head);
755 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
757 struct inode *inode = d_inode(dentry);
760 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
763 error = setattr_prepare(dentry, attr);
767 setattr_copy(inode, attr);
768 mark_inode_dirty(inode);
772 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
774 struct inode *inode = d_inode(dentry);
775 struct ctl_table_header *head = grab_header(inode);
776 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
779 return PTR_ERR(head);
781 generic_fillattr(inode, stat);
783 stat->mode = (stat->mode & S_IFMT) | table->mode;
785 sysctl_head_finish(head);
789 static const struct file_operations proc_sys_file_operations = {
790 .open = proc_sys_open,
791 .poll = proc_sys_poll,
792 .read = proc_sys_read,
793 .write = proc_sys_write,
794 .llseek = default_llseek,
797 static const struct file_operations proc_sys_dir_file_operations = {
798 .read = generic_read_dir,
799 .iterate_shared = proc_sys_readdir,
800 .llseek = generic_file_llseek,
803 static const struct inode_operations proc_sys_inode_operations = {
804 .permission = proc_sys_permission,
805 .setattr = proc_sys_setattr,
806 .getattr = proc_sys_getattr,
809 static const struct inode_operations proc_sys_dir_operations = {
810 .lookup = proc_sys_lookup,
811 .permission = proc_sys_permission,
812 .setattr = proc_sys_setattr,
813 .getattr = proc_sys_getattr,
816 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
818 if (flags & LOOKUP_RCU)
820 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
823 static int proc_sys_delete(const struct dentry *dentry)
825 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
828 static int sysctl_is_seen(struct ctl_table_header *p)
830 struct ctl_table_set *set = p->set;
832 spin_lock(&sysctl_lock);
833 if (p->unregistering)
835 else if (!set->is_seen)
838 res = set->is_seen(set);
839 spin_unlock(&sysctl_lock);
843 static int proc_sys_compare(const struct dentry *dentry,
844 unsigned int len, const char *str, const struct qstr *name)
846 struct ctl_table_header *head;
849 /* Although proc doesn't have negative dentries, rcu-walk means
850 * that inode here can be NULL */
851 /* AV: can it, indeed? */
852 inode = d_inode_rcu(dentry);
855 if (name->len != len)
857 if (memcmp(name->name, str, len))
859 head = rcu_dereference(PROC_I(inode)->sysctl);
860 return !head || !sysctl_is_seen(head);
863 static const struct dentry_operations proc_sys_dentry_operations = {
864 .d_revalidate = proc_sys_revalidate,
865 .d_delete = proc_sys_delete,
866 .d_compare = proc_sys_compare,
869 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
870 const char *name, int namelen)
872 struct ctl_table_header *head;
873 struct ctl_table *entry;
875 entry = find_entry(&head, dir, name, namelen);
877 return ERR_PTR(-ENOENT);
878 if (!S_ISDIR(entry->mode))
879 return ERR_PTR(-ENOTDIR);
880 return container_of(head, struct ctl_dir, header);
883 static struct ctl_dir *new_dir(struct ctl_table_set *set,
884 const char *name, int namelen)
886 struct ctl_table *table;
888 struct ctl_node *node;
891 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
892 sizeof(struct ctl_table)*2 + namelen + 1,
897 node = (struct ctl_node *)(new + 1);
898 table = (struct ctl_table *)(node + 1);
899 new_name = (char *)(table + 2);
900 memcpy(new_name, name, namelen);
901 new_name[namelen] = '\0';
902 table[0].procname = new_name;
903 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
904 init_header(&new->header, set->dir.header.root, set, node, table);
910 * get_subdir - find or create a subdir with the specified name.
911 * @dir: Directory to create the subdirectory in
912 * @name: The name of the subdirectory to find or create
913 * @namelen: The length of name
915 * Takes a directory with an elevated reference count so we know that
916 * if we drop the lock the directory will not go away. Upon success
917 * the reference is moved from @dir to the returned subdirectory.
918 * Upon error an error code is returned and the reference on @dir is
921 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
922 const char *name, int namelen)
924 struct ctl_table_set *set = dir->header.set;
925 struct ctl_dir *subdir, *new = NULL;
928 spin_lock(&sysctl_lock);
929 subdir = find_subdir(dir, name, namelen);
932 if (PTR_ERR(subdir) != -ENOENT)
935 spin_unlock(&sysctl_lock);
936 new = new_dir(set, name, namelen);
937 spin_lock(&sysctl_lock);
938 subdir = ERR_PTR(-ENOMEM);
942 /* Was the subdir added while we dropped the lock? */
943 subdir = find_subdir(dir, name, namelen);
946 if (PTR_ERR(subdir) != -ENOENT)
949 /* Nope. Use the our freshly made directory entry. */
950 err = insert_header(dir, &new->header);
951 subdir = ERR_PTR(err);
956 subdir->header.nreg++;
958 if (IS_ERR(subdir)) {
959 pr_err("sysctl could not get directory: ");
960 sysctl_print_dir(dir);
961 pr_cont("/%*.*s %ld\n",
962 namelen, namelen, name, PTR_ERR(subdir));
964 drop_sysctl_table(&dir->header);
966 drop_sysctl_table(&new->header);
967 spin_unlock(&sysctl_lock);
971 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
973 struct ctl_dir *parent;
974 const char *procname;
975 if (!dir->header.parent)
977 parent = xlate_dir(set, dir->header.parent);
980 procname = dir->header.ctl_table[0].procname;
981 return find_subdir(parent, procname, strlen(procname));
984 static int sysctl_follow_link(struct ctl_table_header **phead,
985 struct ctl_table **pentry)
987 struct ctl_table_header *head;
988 struct ctl_table_root *root;
989 struct ctl_table_set *set;
990 struct ctl_table *entry;
995 spin_lock(&sysctl_lock);
996 root = (*pentry)->data;
997 set = lookup_header_set(root);
998 dir = xlate_dir(set, (*phead)->parent);
1002 const char *procname = (*pentry)->procname;
1004 entry = find_entry(&head, dir, procname, strlen(procname));
1006 if (entry && use_table(head)) {
1007 unuse_table(*phead);
1014 spin_unlock(&sysctl_lock);
1018 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1020 struct va_format vaf;
1023 va_start(args, fmt);
1027 pr_err("sysctl table check failed: %s/%s %pV\n",
1028 path, table->procname, &vaf);
1034 static int sysctl_check_table(const char *path, struct ctl_table *table)
1037 for (; table->procname; table++) {
1039 err = sysctl_err(path, table, "Not a file");
1041 if ((table->proc_handler == proc_dostring) ||
1042 (table->proc_handler == proc_dointvec) ||
1043 (table->proc_handler == proc_dointvec_minmax) ||
1044 (table->proc_handler == proc_dointvec_jiffies) ||
1045 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1046 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1047 (table->proc_handler == proc_doulongvec_minmax) ||
1048 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1050 err = sysctl_err(path, table, "No data");
1052 err = sysctl_err(path, table, "No maxlen");
1054 if (!table->proc_handler)
1055 err = sysctl_err(path, table, "No proc_handler");
1057 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1058 err = sysctl_err(path, table, "bogus .mode 0%o",
1064 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1065 struct ctl_table_root *link_root)
1067 struct ctl_table *link_table, *entry, *link;
1068 struct ctl_table_header *links;
1069 struct ctl_node *node;
1071 int nr_entries, name_bytes;
1075 for (entry = table; entry->procname; entry++) {
1077 name_bytes += strlen(entry->procname) + 1;
1080 links = kzalloc(sizeof(struct ctl_table_header) +
1081 sizeof(struct ctl_node)*nr_entries +
1082 sizeof(struct ctl_table)*(nr_entries + 1) +
1089 node = (struct ctl_node *)(links + 1);
1090 link_table = (struct ctl_table *)(node + nr_entries);
1091 link_name = (char *)&link_table[nr_entries + 1];
1093 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1094 int len = strlen(entry->procname) + 1;
1095 memcpy(link_name, entry->procname, len);
1096 link->procname = link_name;
1097 link->mode = S_IFLNK|S_IRWXUGO;
1098 link->data = link_root;
1101 init_header(links, dir->header.root, dir->header.set, node, link_table);
1102 links->nreg = nr_entries;
1107 static bool get_links(struct ctl_dir *dir,
1108 struct ctl_table *table, struct ctl_table_root *link_root)
1110 struct ctl_table_header *head;
1111 struct ctl_table *entry, *link;
1113 /* Are there links available for every entry in table? */
1114 for (entry = table; entry->procname; entry++) {
1115 const char *procname = entry->procname;
1116 link = find_entry(&head, dir, procname, strlen(procname));
1119 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1121 if (S_ISLNK(link->mode) && (link->data == link_root))
1126 /* The checks passed. Increase the registration count on the links */
1127 for (entry = table; entry->procname; entry++) {
1128 const char *procname = entry->procname;
1129 link = find_entry(&head, dir, procname, strlen(procname));
1135 static int insert_links(struct ctl_table_header *head)
1137 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1138 struct ctl_dir *core_parent = NULL;
1139 struct ctl_table_header *links;
1142 if (head->set == root_set)
1145 core_parent = xlate_dir(root_set, head->parent);
1146 if (IS_ERR(core_parent))
1149 if (get_links(core_parent, head->ctl_table, head->root))
1152 core_parent->header.nreg++;
1153 spin_unlock(&sysctl_lock);
1155 links = new_links(core_parent, head->ctl_table, head->root);
1157 spin_lock(&sysctl_lock);
1163 if (get_links(core_parent, head->ctl_table, head->root)) {
1168 err = insert_header(core_parent, links);
1172 drop_sysctl_table(&core_parent->header);
1177 * __register_sysctl_table - register a leaf sysctl table
1178 * @set: Sysctl tree to register on
1179 * @path: The path to the directory the sysctl table is in.
1180 * @table: the top-level table structure
1182 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1183 * array. A completely 0 filled entry terminates the table.
1185 * The members of the &struct ctl_table structure are used as follows:
1187 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1188 * enter a sysctl file
1190 * data - a pointer to data for use by proc_handler
1192 * maxlen - the maximum size in bytes of the data
1194 * mode - the file permissions for the /proc/sys file
1196 * child - must be %NULL.
1198 * proc_handler - the text handler routine (described below)
1200 * extra1, extra2 - extra pointers usable by the proc handler routines
1202 * Leaf nodes in the sysctl tree will be represented by a single file
1203 * under /proc; non-leaf nodes will be represented by directories.
1205 * There must be a proc_handler routine for any terminal nodes.
1206 * Several default handlers are available to cover common cases -
1208 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1209 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1210 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1212 * It is the handler's job to read the input buffer from user memory
1213 * and process it. The handler should return 0 on success.
1215 * This routine returns %NULL on a failure to register, and a pointer
1216 * to the table header on success.
1218 struct ctl_table_header *__register_sysctl_table(
1219 struct ctl_table_set *set,
1220 const char *path, struct ctl_table *table)
1222 struct ctl_table_root *root = set->dir.header.root;
1223 struct ctl_table_header *header;
1224 const char *name, *nextname;
1225 struct ctl_dir *dir;
1226 struct ctl_table *entry;
1227 struct ctl_node *node;
1230 for (entry = table; entry->procname; entry++)
1233 header = kzalloc(sizeof(struct ctl_table_header) +
1234 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1238 node = (struct ctl_node *)(header + 1);
1239 init_header(header, root, set, node, table);
1240 if (sysctl_check_table(path, table))
1243 spin_lock(&sysctl_lock);
1245 /* Reference moved down the diretory tree get_subdir */
1247 spin_unlock(&sysctl_lock);
1249 /* Find the directory for the ctl_table */
1250 for (name = path; name; name = nextname) {
1252 nextname = strchr(name, '/');
1254 namelen = nextname - name;
1257 namelen = strlen(name);
1262 dir = get_subdir(dir, name, namelen);
1267 spin_lock(&sysctl_lock);
1268 if (insert_header(dir, header))
1269 goto fail_put_dir_locked;
1271 drop_sysctl_table(&dir->header);
1272 spin_unlock(&sysctl_lock);
1276 fail_put_dir_locked:
1277 drop_sysctl_table(&dir->header);
1278 spin_unlock(&sysctl_lock);
1286 * register_sysctl - register a sysctl table
1287 * @path: The path to the directory the sysctl table is in.
1288 * @table: the table structure
1290 * Register a sysctl table. @table should be a filled in ctl_table
1291 * array. A completely 0 filled entry terminates the table.
1293 * See __register_sysctl_table for more details.
1295 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1297 return __register_sysctl_table(&sysctl_table_root.default_set,
1300 EXPORT_SYMBOL(register_sysctl);
1302 static char *append_path(const char *path, char *pos, const char *name)
1305 namelen = strlen(name);
1306 if (((pos - path) + namelen + 2) >= PATH_MAX)
1308 memcpy(pos, name, namelen);
1310 pos[namelen + 1] = '\0';
1315 static int count_subheaders(struct ctl_table *table)
1318 int nr_subheaders = 0;
1319 struct ctl_table *entry;
1321 /* special case: no directory and empty directory */
1322 if (!table || !table->procname)
1325 for (entry = table; entry->procname; entry++) {
1327 nr_subheaders += count_subheaders(entry->child);
1331 return nr_subheaders + has_files;
1334 static int register_leaf_sysctl_tables(const char *path, char *pos,
1335 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1336 struct ctl_table *table)
1338 struct ctl_table *ctl_table_arg = NULL;
1339 struct ctl_table *entry, *files;
1344 for (entry = table; entry->procname; entry++) {
1352 /* If there are mixed files and directories we need a new table */
1353 if (nr_dirs && nr_files) {
1354 struct ctl_table *new;
1355 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1360 ctl_table_arg = files;
1361 for (new = files, entry = table; entry->procname; entry++) {
1369 /* Register everything except a directory full of subdirectories */
1370 if (nr_files || !nr_dirs) {
1371 struct ctl_table_header *header;
1372 header = __register_sysctl_table(set, path, files);
1374 kfree(ctl_table_arg);
1378 /* Remember if we need to free the file table */
1379 header->ctl_table_arg = ctl_table_arg;
1380 **subheader = header;
1384 /* Recurse into the subdirectories. */
1385 for (entry = table; entry->procname; entry++) {
1391 err = -ENAMETOOLONG;
1392 child_pos = append_path(path, pos, entry->procname);
1396 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1404 /* On failure our caller will unregister all registered subheaders */
1409 * __register_sysctl_paths - register a sysctl table hierarchy
1410 * @set: Sysctl tree to register on
1411 * @path: The path to the directory the sysctl table is in.
1412 * @table: the top-level table structure
1414 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1415 * array. A completely 0 filled entry terminates the table.
1417 * See __register_sysctl_table for more details.
1419 struct ctl_table_header *__register_sysctl_paths(
1420 struct ctl_table_set *set,
1421 const struct ctl_path *path, struct ctl_table *table)
1423 struct ctl_table *ctl_table_arg = table;
1424 int nr_subheaders = count_subheaders(table);
1425 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1426 const struct ctl_path *component;
1427 char *new_path, *pos;
1429 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1434 for (component = path; component->procname; component++) {
1435 pos = append_path(new_path, pos, component->procname);
1439 while (table->procname && table->child && !table[1].procname) {
1440 pos = append_path(new_path, pos, table->procname);
1443 table = table->child;
1445 if (nr_subheaders == 1) {
1446 header = __register_sysctl_table(set, new_path, table);
1448 header->ctl_table_arg = ctl_table_arg;
1450 header = kzalloc(sizeof(*header) +
1451 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1455 subheaders = (struct ctl_table_header **) (header + 1);
1456 subheader = subheaders;
1457 header->ctl_table_arg = ctl_table_arg;
1459 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1461 goto err_register_leaves;
1468 err_register_leaves:
1469 while (subheader > subheaders) {
1470 struct ctl_table_header *subh = *(--subheader);
1471 struct ctl_table *table = subh->ctl_table_arg;
1472 unregister_sysctl_table(subh);
1481 * register_sysctl_table_path - register a sysctl table hierarchy
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_paths for more details.
1490 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1491 struct ctl_table *table)
1493 return __register_sysctl_paths(&sysctl_table_root.default_set,
1496 EXPORT_SYMBOL(register_sysctl_paths);
1499 * register_sysctl_table - register a sysctl table hierarchy
1500 * @table: the top-level table structure
1502 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1503 * array. A completely 0 filled entry terminates the table.
1505 * See register_sysctl_paths for more details.
1507 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1509 static const struct ctl_path null_path[] = { {} };
1511 return register_sysctl_paths(null_path, table);
1513 EXPORT_SYMBOL(register_sysctl_table);
1515 static void put_links(struct ctl_table_header *header)
1517 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1518 struct ctl_table_root *root = header->root;
1519 struct ctl_dir *parent = header->parent;
1520 struct ctl_dir *core_parent;
1521 struct ctl_table *entry;
1523 if (header->set == root_set)
1526 core_parent = xlate_dir(root_set, parent);
1527 if (IS_ERR(core_parent))
1530 for (entry = header->ctl_table; entry->procname; entry++) {
1531 struct ctl_table_header *link_head;
1532 struct ctl_table *link;
1533 const char *name = entry->procname;
1535 link = find_entry(&link_head, core_parent, name, strlen(name));
1537 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1538 (S_ISLNK(link->mode) && (link->data == root)))) {
1539 drop_sysctl_table(link_head);
1542 pr_err("sysctl link missing during unregister: ");
1543 sysctl_print_dir(parent);
1544 pr_cont("/%s\n", name);
1549 static void drop_sysctl_table(struct ctl_table_header *header)
1551 struct ctl_dir *parent = header->parent;
1557 start_unregistering(header);
1558 if (!--header->count)
1559 kfree_rcu(header, rcu);
1562 drop_sysctl_table(&parent->header);
1566 * unregister_sysctl_table - unregister a sysctl table hierarchy
1567 * @header: the header returned from register_sysctl_table
1569 * Unregisters the sysctl table and all children. proc entries may not
1570 * actually be removed until they are no longer used by anyone.
1572 void unregister_sysctl_table(struct ctl_table_header * header)
1580 nr_subheaders = count_subheaders(header->ctl_table_arg);
1581 if (unlikely(nr_subheaders > 1)) {
1582 struct ctl_table_header **subheaders;
1585 subheaders = (struct ctl_table_header **)(header + 1);
1586 for (i = nr_subheaders -1; i >= 0; i--) {
1587 struct ctl_table_header *subh = subheaders[i];
1588 struct ctl_table *table = subh->ctl_table_arg;
1589 unregister_sysctl_table(subh);
1596 spin_lock(&sysctl_lock);
1597 drop_sysctl_table(header);
1598 spin_unlock(&sysctl_lock);
1600 EXPORT_SYMBOL(unregister_sysctl_table);
1602 void setup_sysctl_set(struct ctl_table_set *set,
1603 struct ctl_table_root *root,
1604 int (*is_seen)(struct ctl_table_set *))
1606 memset(set, 0, sizeof(*set));
1607 set->is_seen = is_seen;
1608 init_header(&set->dir.header, root, set, NULL, root_table);
1611 void retire_sysctl_set(struct ctl_table_set *set)
1613 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1616 int __init proc_sys_init(void)
1618 struct proc_dir_entry *proc_sys_root;
1620 proc_sys_root = proc_mkdir("sys", NULL);
1621 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1622 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1623 proc_sys_root->nlink = 0;
1625 return sysctl_init();