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;
193 INIT_HLIST_HEAD(&head->inodes);
195 struct ctl_table *entry;
196 for (entry = table; entry->procname; entry++, node++)
201 static void erase_header(struct ctl_table_header *head)
203 struct ctl_table *entry;
204 for (entry = head->ctl_table; entry->procname; entry++)
205 erase_entry(head, entry);
208 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
210 struct ctl_table *entry;
213 /* Is this a permanently empty directory? */
214 if (is_empty_dir(&dir->header))
217 /* Am I creating a permanently empty directory? */
218 if (header->ctl_table == sysctl_mount_point) {
219 if (!RB_EMPTY_ROOT(&dir->root))
225 header->parent = dir;
226 err = insert_links(header);
229 for (entry = header->ctl_table; entry->procname; entry++) {
230 err = insert_entry(header, entry);
236 erase_header(header);
239 if (header->ctl_table == sysctl_mount_point)
240 clear_empty_dir(dir);
241 header->parent = NULL;
242 drop_sysctl_table(&dir->header);
246 /* called under sysctl_lock */
247 static int use_table(struct ctl_table_header *p)
249 if (unlikely(p->unregistering))
255 /* called under sysctl_lock */
256 static void unuse_table(struct ctl_table_header *p)
259 if (unlikely(p->unregistering))
260 complete(p->unregistering);
263 static void proc_sys_prune_dcache(struct ctl_table_header *head)
266 struct proc_inode *ei;
267 struct hlist_node *node;
268 struct super_block *sb;
272 node = hlist_first_rcu(&head->inodes);
275 ei = hlist_entry(node, struct proc_inode, sysctl_inodes);
276 spin_lock(&sysctl_lock);
277 hlist_del_init_rcu(&ei->sysctl_inodes);
278 spin_unlock(&sysctl_lock);
280 inode = &ei->vfs_inode;
282 if (!atomic_inc_not_zero(&sb->s_active))
284 inode = igrab(inode);
286 if (unlikely(!inode)) {
287 deactivate_super(sb);
292 d_prune_aliases(inode);
294 deactivate_super(sb);
301 /* called under sysctl_lock, will reacquire if has to wait */
302 static void start_unregistering(struct ctl_table_header *p)
305 * if p->used is 0, nobody will ever touch that entry again;
306 * we'll eliminate all paths to it before dropping sysctl_lock
308 if (unlikely(p->used)) {
309 struct completion wait;
310 init_completion(&wait);
311 p->unregistering = &wait;
312 spin_unlock(&sysctl_lock);
313 wait_for_completion(&wait);
315 /* anything non-NULL; we'll never dereference it */
316 p->unregistering = ERR_PTR(-EINVAL);
317 spin_unlock(&sysctl_lock);
320 * Prune dentries for unregistered sysctls: namespaced sysctls
321 * can have duplicate names and contaminate dcache very badly.
323 proc_sys_prune_dcache(p);
325 * do not remove from the list until nobody holds it; walking the
326 * list in do_sysctl() relies on that.
328 spin_lock(&sysctl_lock);
332 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
335 spin_lock(&sysctl_lock);
336 if (!use_table(head))
337 head = ERR_PTR(-ENOENT);
338 spin_unlock(&sysctl_lock);
342 static void sysctl_head_finish(struct ctl_table_header *head)
346 spin_lock(&sysctl_lock);
348 spin_unlock(&sysctl_lock);
351 static struct ctl_table_set *
352 lookup_header_set(struct ctl_table_root *root)
354 struct ctl_table_set *set = &root->default_set;
356 set = root->lookup(root);
360 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
362 const char *name, int namelen)
364 struct ctl_table_header *head;
365 struct ctl_table *entry;
367 spin_lock(&sysctl_lock);
368 entry = find_entry(&head, dir, name, namelen);
369 if (entry && use_table(head))
373 spin_unlock(&sysctl_lock);
377 static struct ctl_node *first_usable_entry(struct rb_node *node)
379 struct ctl_node *ctl_node;
381 for (;node; node = rb_next(node)) {
382 ctl_node = rb_entry(node, struct ctl_node, node);
383 if (use_table(ctl_node->header))
389 static void first_entry(struct ctl_dir *dir,
390 struct ctl_table_header **phead, struct ctl_table **pentry)
392 struct ctl_table_header *head = NULL;
393 struct ctl_table *entry = NULL;
394 struct ctl_node *ctl_node;
396 spin_lock(&sysctl_lock);
397 ctl_node = first_usable_entry(rb_first(&dir->root));
398 spin_unlock(&sysctl_lock);
400 head = ctl_node->header;
401 entry = &head->ctl_table[ctl_node - head->node];
407 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
409 struct ctl_table_header *head = *phead;
410 struct ctl_table *entry = *pentry;
411 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
413 spin_lock(&sysctl_lock);
416 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
417 spin_unlock(&sysctl_lock);
420 head = ctl_node->header;
421 entry = &head->ctl_table[ctl_node - head->node];
427 void register_sysctl_root(struct ctl_table_root *root)
432 * sysctl_perm does NOT grant the superuser all rights automatically, because
433 * some sysctl variables are readonly even to root.
436 static int test_perm(int mode, int op)
438 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
440 else if (in_egroup_p(GLOBAL_ROOT_GID))
442 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
447 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
449 struct ctl_table_root *root = head->root;
452 if (root->permissions)
453 mode = root->permissions(head, table);
457 return test_perm(mode, op);
460 static struct inode *proc_sys_make_inode(struct super_block *sb,
461 struct ctl_table_header *head, struct ctl_table *table)
463 struct ctl_table_root *root = head->root;
465 struct proc_inode *ei;
467 inode = new_inode(sb);
469 return ERR_PTR(-ENOMEM);
471 inode->i_ino = get_next_ino();
475 spin_lock(&sysctl_lock);
476 if (unlikely(head->unregistering)) {
477 spin_unlock(&sysctl_lock);
479 return ERR_PTR(-ENOENT);
482 ei->sysctl_entry = table;
483 hlist_add_head_rcu(&ei->sysctl_inodes, &head->inodes);
485 spin_unlock(&sysctl_lock);
487 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
488 inode->i_mode = table->mode;
489 if (!S_ISDIR(table->mode)) {
490 inode->i_mode |= S_IFREG;
491 inode->i_op = &proc_sys_inode_operations;
492 inode->i_fop = &proc_sys_file_operations;
494 inode->i_mode |= S_IFDIR;
495 inode->i_op = &proc_sys_dir_operations;
496 inode->i_fop = &proc_sys_dir_file_operations;
497 if (is_empty_dir(head))
498 make_empty_dir_inode(inode);
501 if (root->set_ownership)
502 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
504 inode->i_uid = GLOBAL_ROOT_UID;
505 inode->i_gid = GLOBAL_ROOT_GID;
511 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
513 spin_lock(&sysctl_lock);
514 hlist_del_init_rcu(&PROC_I(inode)->sysctl_inodes);
516 kfree_rcu(head, rcu);
517 spin_unlock(&sysctl_lock);
520 static struct ctl_table_header *grab_header(struct inode *inode)
522 struct ctl_table_header *head = PROC_I(inode)->sysctl;
524 head = &sysctl_table_root.default_set.dir.header;
525 return sysctl_head_grab(head);
528 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
531 struct ctl_table_header *head = grab_header(dir);
532 struct ctl_table_header *h = NULL;
533 const struct qstr *name = &dentry->d_name;
536 struct dentry *err = ERR_PTR(-ENOENT);
537 struct ctl_dir *ctl_dir;
541 return ERR_CAST(head);
543 ctl_dir = container_of(head, struct ctl_dir, header);
545 p = lookup_entry(&h, ctl_dir, name->name, name->len);
549 if (S_ISLNK(p->mode)) {
550 ret = sysctl_follow_link(&h, &p);
556 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
558 err = ERR_CAST(inode);
563 d_set_d_op(dentry, &proc_sys_dentry_operations);
564 d_add(dentry, inode);
568 sysctl_head_finish(h);
569 sysctl_head_finish(head);
573 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
574 size_t count, loff_t *ppos, int write)
576 struct inode *inode = file_inode(filp);
577 struct ctl_table_header *head = grab_header(inode);
578 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
583 return PTR_ERR(head);
586 * At this point we know that the sysctl was not unregistered
587 * and won't be until we finish.
590 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
593 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
595 if (!table->proc_handler)
598 /* careful: calling conventions are nasty here */
600 error = table->proc_handler(table, write, buf, &res, ppos);
604 sysctl_head_finish(head);
609 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
610 size_t count, loff_t *ppos)
612 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
615 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
616 size_t count, loff_t *ppos)
618 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
621 static int proc_sys_open(struct inode *inode, struct file *filp)
623 struct ctl_table_header *head = grab_header(inode);
624 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
626 /* sysctl was unregistered */
628 return PTR_ERR(head);
631 filp->private_data = proc_sys_poll_event(table->poll);
633 sysctl_head_finish(head);
638 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
640 struct inode *inode = file_inode(filp);
641 struct ctl_table_header *head = grab_header(inode);
642 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
643 unsigned int ret = DEFAULT_POLLMASK;
646 /* sysctl was unregistered */
648 return POLLERR | POLLHUP;
650 if (!table->proc_handler)
656 event = (unsigned long)filp->private_data;
657 poll_wait(filp, &table->poll->wait, wait);
659 if (event != atomic_read(&table->poll->event)) {
660 filp->private_data = proc_sys_poll_event(table->poll);
661 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
665 sysctl_head_finish(head);
670 static bool proc_sys_fill_cache(struct file *file,
671 struct dir_context *ctx,
672 struct ctl_table_header *head,
673 struct ctl_table *table)
675 struct dentry *child, *dir = file->f_path.dentry;
679 unsigned type = DT_UNKNOWN;
681 qname.name = table->procname;
682 qname.len = strlen(table->procname);
683 qname.hash = full_name_hash(dir, qname.name, qname.len);
685 child = d_lookup(dir, &qname);
687 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
688 child = d_alloc_parallel(dir, &qname, &wq);
691 if (d_in_lookup(child)) {
692 inode = proc_sys_make_inode(dir->d_sb, head, table);
694 d_lookup_done(child);
698 d_set_d_op(child, &proc_sys_dentry_operations);
702 inode = d_inode(child);
704 type = inode->i_mode >> 12;
706 return dir_emit(ctx, qname.name, qname.len, ino, type);
709 static bool proc_sys_link_fill_cache(struct file *file,
710 struct dir_context *ctx,
711 struct ctl_table_header *head,
712 struct ctl_table *table)
716 head = sysctl_head_grab(head);
720 if (S_ISLNK(table->mode)) {
721 /* It is not an error if we can not follow the link ignore it */
722 int err = sysctl_follow_link(&head, &table);
727 ret = proc_sys_fill_cache(file, ctx, head, table);
729 sysctl_head_finish(head);
733 static int scan(struct ctl_table_header *head, struct ctl_table *table,
734 unsigned long *pos, struct file *file,
735 struct dir_context *ctx)
739 if ((*pos)++ < ctx->pos)
742 if (unlikely(S_ISLNK(table->mode)))
743 res = proc_sys_link_fill_cache(file, ctx, head, table);
745 res = proc_sys_fill_cache(file, ctx, head, table);
753 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
755 struct ctl_table_header *head = grab_header(file_inode(file));
756 struct ctl_table_header *h = NULL;
757 struct ctl_table *entry;
758 struct ctl_dir *ctl_dir;
762 return PTR_ERR(head);
764 ctl_dir = container_of(head, struct ctl_dir, header);
766 if (!dir_emit_dots(file, ctx))
771 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
772 if (!scan(h, entry, &pos, file, ctx)) {
773 sysctl_head_finish(h);
778 sysctl_head_finish(head);
782 static int proc_sys_permission(struct inode *inode, int mask)
785 * sysctl entries that are not writeable,
786 * are _NOT_ writeable, capabilities or not.
788 struct ctl_table_header *head;
789 struct ctl_table *table;
792 /* Executable files are not allowed under /proc/sys/ */
793 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
796 head = grab_header(inode);
798 return PTR_ERR(head);
800 table = PROC_I(inode)->sysctl_entry;
801 if (!table) /* global root - r-xr-xr-x */
802 error = mask & MAY_WRITE ? -EACCES : 0;
803 else /* Use the permissions on the sysctl table entry */
804 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
806 sysctl_head_finish(head);
810 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
812 struct inode *inode = d_inode(dentry);
815 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
818 error = setattr_prepare(dentry, attr);
822 setattr_copy(inode, attr);
823 mark_inode_dirty(inode);
827 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
829 struct inode *inode = d_inode(dentry);
830 struct ctl_table_header *head = grab_header(inode);
831 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
834 return PTR_ERR(head);
836 generic_fillattr(inode, stat);
838 stat->mode = (stat->mode & S_IFMT) | table->mode;
840 sysctl_head_finish(head);
844 static const struct file_operations proc_sys_file_operations = {
845 .open = proc_sys_open,
846 .poll = proc_sys_poll,
847 .read = proc_sys_read,
848 .write = proc_sys_write,
849 .llseek = default_llseek,
852 static const struct file_operations proc_sys_dir_file_operations = {
853 .read = generic_read_dir,
854 .iterate_shared = proc_sys_readdir,
855 .llseek = generic_file_llseek,
858 static const struct inode_operations proc_sys_inode_operations = {
859 .permission = proc_sys_permission,
860 .setattr = proc_sys_setattr,
861 .getattr = proc_sys_getattr,
864 static const struct inode_operations proc_sys_dir_operations = {
865 .lookup = proc_sys_lookup,
866 .permission = proc_sys_permission,
867 .setattr = proc_sys_setattr,
868 .getattr = proc_sys_getattr,
871 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
873 if (flags & LOOKUP_RCU)
875 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
878 static int proc_sys_delete(const struct dentry *dentry)
880 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
883 static int sysctl_is_seen(struct ctl_table_header *p)
885 struct ctl_table_set *set = p->set;
887 spin_lock(&sysctl_lock);
888 if (p->unregistering)
890 else if (!set->is_seen)
893 res = set->is_seen(set);
894 spin_unlock(&sysctl_lock);
898 static int proc_sys_compare(const struct dentry *dentry,
899 unsigned int len, const char *str, const struct qstr *name)
901 struct ctl_table_header *head;
904 /* Although proc doesn't have negative dentries, rcu-walk means
905 * that inode here can be NULL */
906 /* AV: can it, indeed? */
907 inode = d_inode_rcu(dentry);
910 if (name->len != len)
912 if (memcmp(name->name, str, len))
914 head = rcu_dereference(PROC_I(inode)->sysctl);
915 return !head || !sysctl_is_seen(head);
918 static const struct dentry_operations proc_sys_dentry_operations = {
919 .d_revalidate = proc_sys_revalidate,
920 .d_delete = proc_sys_delete,
921 .d_compare = proc_sys_compare,
924 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
925 const char *name, int namelen)
927 struct ctl_table_header *head;
928 struct ctl_table *entry;
930 entry = find_entry(&head, dir, name, namelen);
932 return ERR_PTR(-ENOENT);
933 if (!S_ISDIR(entry->mode))
934 return ERR_PTR(-ENOTDIR);
935 return container_of(head, struct ctl_dir, header);
938 static struct ctl_dir *new_dir(struct ctl_table_set *set,
939 const char *name, int namelen)
941 struct ctl_table *table;
943 struct ctl_node *node;
946 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
947 sizeof(struct ctl_table)*2 + namelen + 1,
952 node = (struct ctl_node *)(new + 1);
953 table = (struct ctl_table *)(node + 1);
954 new_name = (char *)(table + 2);
955 memcpy(new_name, name, namelen);
956 new_name[namelen] = '\0';
957 table[0].procname = new_name;
958 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
959 init_header(&new->header, set->dir.header.root, set, node, table);
965 * get_subdir - find or create a subdir with the specified name.
966 * @dir: Directory to create the subdirectory in
967 * @name: The name of the subdirectory to find or create
968 * @namelen: The length of name
970 * Takes a directory with an elevated reference count so we know that
971 * if we drop the lock the directory will not go away. Upon success
972 * the reference is moved from @dir to the returned subdirectory.
973 * Upon error an error code is returned and the reference on @dir is
976 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
977 const char *name, int namelen)
979 struct ctl_table_set *set = dir->header.set;
980 struct ctl_dir *subdir, *new = NULL;
983 spin_lock(&sysctl_lock);
984 subdir = find_subdir(dir, name, namelen);
987 if (PTR_ERR(subdir) != -ENOENT)
990 spin_unlock(&sysctl_lock);
991 new = new_dir(set, name, namelen);
992 spin_lock(&sysctl_lock);
993 subdir = ERR_PTR(-ENOMEM);
997 /* Was the subdir added while we dropped the lock? */
998 subdir = find_subdir(dir, name, namelen);
1001 if (PTR_ERR(subdir) != -ENOENT)
1004 /* Nope. Use the our freshly made directory entry. */
1005 err = insert_header(dir, &new->header);
1006 subdir = ERR_PTR(err);
1011 subdir->header.nreg++;
1013 if (IS_ERR(subdir)) {
1014 pr_err("sysctl could not get directory: ");
1015 sysctl_print_dir(dir);
1016 pr_cont("/%*.*s %ld\n",
1017 namelen, namelen, name, PTR_ERR(subdir));
1019 drop_sysctl_table(&dir->header);
1021 drop_sysctl_table(&new->header);
1022 spin_unlock(&sysctl_lock);
1026 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1028 struct ctl_dir *parent;
1029 const char *procname;
1030 if (!dir->header.parent)
1032 parent = xlate_dir(set, dir->header.parent);
1035 procname = dir->header.ctl_table[0].procname;
1036 return find_subdir(parent, procname, strlen(procname));
1039 static int sysctl_follow_link(struct ctl_table_header **phead,
1040 struct ctl_table **pentry)
1042 struct ctl_table_header *head;
1043 struct ctl_table_root *root;
1044 struct ctl_table_set *set;
1045 struct ctl_table *entry;
1046 struct ctl_dir *dir;
1050 spin_lock(&sysctl_lock);
1051 root = (*pentry)->data;
1052 set = lookup_header_set(root);
1053 dir = xlate_dir(set, (*phead)->parent);
1057 const char *procname = (*pentry)->procname;
1059 entry = find_entry(&head, dir, procname, strlen(procname));
1061 if (entry && use_table(head)) {
1062 unuse_table(*phead);
1069 spin_unlock(&sysctl_lock);
1073 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1075 struct va_format vaf;
1078 va_start(args, fmt);
1082 pr_err("sysctl table check failed: %s/%s %pV\n",
1083 path, table->procname, &vaf);
1089 static int sysctl_check_table(const char *path, struct ctl_table *table)
1092 for (; table->procname; table++) {
1094 err = sysctl_err(path, table, "Not a file");
1096 if ((table->proc_handler == proc_dostring) ||
1097 (table->proc_handler == proc_dointvec) ||
1098 (table->proc_handler == proc_dointvec_minmax) ||
1099 (table->proc_handler == proc_dointvec_jiffies) ||
1100 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1101 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1102 (table->proc_handler == proc_doulongvec_minmax) ||
1103 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1105 err = sysctl_err(path, table, "No data");
1107 err = sysctl_err(path, table, "No maxlen");
1109 if (!table->proc_handler)
1110 err = sysctl_err(path, table, "No proc_handler");
1112 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1113 err = sysctl_err(path, table, "bogus .mode 0%o",
1119 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1120 struct ctl_table_root *link_root)
1122 struct ctl_table *link_table, *entry, *link;
1123 struct ctl_table_header *links;
1124 struct ctl_node *node;
1126 int nr_entries, name_bytes;
1130 for (entry = table; entry->procname; entry++) {
1132 name_bytes += strlen(entry->procname) + 1;
1135 links = kzalloc(sizeof(struct ctl_table_header) +
1136 sizeof(struct ctl_node)*nr_entries +
1137 sizeof(struct ctl_table)*(nr_entries + 1) +
1144 node = (struct ctl_node *)(links + 1);
1145 link_table = (struct ctl_table *)(node + nr_entries);
1146 link_name = (char *)&link_table[nr_entries + 1];
1148 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1149 int len = strlen(entry->procname) + 1;
1150 memcpy(link_name, entry->procname, len);
1151 link->procname = link_name;
1152 link->mode = S_IFLNK|S_IRWXUGO;
1153 link->data = link_root;
1156 init_header(links, dir->header.root, dir->header.set, node, link_table);
1157 links->nreg = nr_entries;
1162 static bool get_links(struct ctl_dir *dir,
1163 struct ctl_table *table, struct ctl_table_root *link_root)
1165 struct ctl_table_header *head;
1166 struct ctl_table *entry, *link;
1168 /* Are there links available for every entry in table? */
1169 for (entry = table; entry->procname; entry++) {
1170 const char *procname = entry->procname;
1171 link = find_entry(&head, dir, procname, strlen(procname));
1174 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1176 if (S_ISLNK(link->mode) && (link->data == link_root))
1181 /* The checks passed. Increase the registration count on the links */
1182 for (entry = table; entry->procname; entry++) {
1183 const char *procname = entry->procname;
1184 link = find_entry(&head, dir, procname, strlen(procname));
1190 static int insert_links(struct ctl_table_header *head)
1192 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1193 struct ctl_dir *core_parent = NULL;
1194 struct ctl_table_header *links;
1197 if (head->set == root_set)
1200 core_parent = xlate_dir(root_set, head->parent);
1201 if (IS_ERR(core_parent))
1204 if (get_links(core_parent, head->ctl_table, head->root))
1207 core_parent->header.nreg++;
1208 spin_unlock(&sysctl_lock);
1210 links = new_links(core_parent, head->ctl_table, head->root);
1212 spin_lock(&sysctl_lock);
1218 if (get_links(core_parent, head->ctl_table, head->root)) {
1223 err = insert_header(core_parent, links);
1227 drop_sysctl_table(&core_parent->header);
1232 * __register_sysctl_table - register a leaf sysctl table
1233 * @set: Sysctl tree to register on
1234 * @path: The path to the directory the sysctl table is in.
1235 * @table: the top-level table structure
1237 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1238 * array. A completely 0 filled entry terminates the table.
1240 * The members of the &struct ctl_table structure are used as follows:
1242 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1243 * enter a sysctl file
1245 * data - a pointer to data for use by proc_handler
1247 * maxlen - the maximum size in bytes of the data
1249 * mode - the file permissions for the /proc/sys file
1251 * child - must be %NULL.
1253 * proc_handler - the text handler routine (described below)
1255 * extra1, extra2 - extra pointers usable by the proc handler routines
1257 * Leaf nodes in the sysctl tree will be represented by a single file
1258 * under /proc; non-leaf nodes will be represented by directories.
1260 * There must be a proc_handler routine for any terminal nodes.
1261 * Several default handlers are available to cover common cases -
1263 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1264 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1265 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1267 * It is the handler's job to read the input buffer from user memory
1268 * and process it. The handler should return 0 on success.
1270 * This routine returns %NULL on a failure to register, and a pointer
1271 * to the table header on success.
1273 struct ctl_table_header *__register_sysctl_table(
1274 struct ctl_table_set *set,
1275 const char *path, struct ctl_table *table)
1277 struct ctl_table_root *root = set->dir.header.root;
1278 struct ctl_table_header *header;
1279 const char *name, *nextname;
1280 struct ctl_dir *dir;
1281 struct ctl_table *entry;
1282 struct ctl_node *node;
1285 for (entry = table; entry->procname; entry++)
1288 header = kzalloc(sizeof(struct ctl_table_header) +
1289 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1293 node = (struct ctl_node *)(header + 1);
1294 init_header(header, root, set, node, table);
1295 if (sysctl_check_table(path, table))
1298 spin_lock(&sysctl_lock);
1300 /* Reference moved down the diretory tree get_subdir */
1302 spin_unlock(&sysctl_lock);
1304 /* Find the directory for the ctl_table */
1305 for (name = path; name; name = nextname) {
1307 nextname = strchr(name, '/');
1309 namelen = nextname - name;
1312 namelen = strlen(name);
1317 dir = get_subdir(dir, name, namelen);
1322 spin_lock(&sysctl_lock);
1323 if (insert_header(dir, header))
1324 goto fail_put_dir_locked;
1326 drop_sysctl_table(&dir->header);
1327 spin_unlock(&sysctl_lock);
1331 fail_put_dir_locked:
1332 drop_sysctl_table(&dir->header);
1333 spin_unlock(&sysctl_lock);
1341 * register_sysctl - register a sysctl table
1342 * @path: The path to the directory the sysctl table is in.
1343 * @table: the table structure
1345 * Register a sysctl table. @table should be a filled in ctl_table
1346 * array. A completely 0 filled entry terminates the table.
1348 * See __register_sysctl_table for more details.
1350 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1352 return __register_sysctl_table(&sysctl_table_root.default_set,
1355 EXPORT_SYMBOL(register_sysctl);
1357 static char *append_path(const char *path, char *pos, const char *name)
1360 namelen = strlen(name);
1361 if (((pos - path) + namelen + 2) >= PATH_MAX)
1363 memcpy(pos, name, namelen);
1365 pos[namelen + 1] = '\0';
1370 static int count_subheaders(struct ctl_table *table)
1373 int nr_subheaders = 0;
1374 struct ctl_table *entry;
1376 /* special case: no directory and empty directory */
1377 if (!table || !table->procname)
1380 for (entry = table; entry->procname; entry++) {
1382 nr_subheaders += count_subheaders(entry->child);
1386 return nr_subheaders + has_files;
1389 static int register_leaf_sysctl_tables(const char *path, char *pos,
1390 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1391 struct ctl_table *table)
1393 struct ctl_table *ctl_table_arg = NULL;
1394 struct ctl_table *entry, *files;
1399 for (entry = table; entry->procname; entry++) {
1407 /* If there are mixed files and directories we need a new table */
1408 if (nr_dirs && nr_files) {
1409 struct ctl_table *new;
1410 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1415 ctl_table_arg = files;
1416 for (new = files, entry = table; entry->procname; entry++) {
1424 /* Register everything except a directory full of subdirectories */
1425 if (nr_files || !nr_dirs) {
1426 struct ctl_table_header *header;
1427 header = __register_sysctl_table(set, path, files);
1429 kfree(ctl_table_arg);
1433 /* Remember if we need to free the file table */
1434 header->ctl_table_arg = ctl_table_arg;
1435 **subheader = header;
1439 /* Recurse into the subdirectories. */
1440 for (entry = table; entry->procname; entry++) {
1446 err = -ENAMETOOLONG;
1447 child_pos = append_path(path, pos, entry->procname);
1451 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1459 /* On failure our caller will unregister all registered subheaders */
1464 * __register_sysctl_paths - register a sysctl table hierarchy
1465 * @set: Sysctl tree to register on
1466 * @path: The path to the directory the sysctl table is in.
1467 * @table: the top-level table structure
1469 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1470 * array. A completely 0 filled entry terminates the table.
1472 * See __register_sysctl_table for more details.
1474 struct ctl_table_header *__register_sysctl_paths(
1475 struct ctl_table_set *set,
1476 const struct ctl_path *path, struct ctl_table *table)
1478 struct ctl_table *ctl_table_arg = table;
1479 int nr_subheaders = count_subheaders(table);
1480 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1481 const struct ctl_path *component;
1482 char *new_path, *pos;
1484 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1489 for (component = path; component->procname; component++) {
1490 pos = append_path(new_path, pos, component->procname);
1494 while (table->procname && table->child && !table[1].procname) {
1495 pos = append_path(new_path, pos, table->procname);
1498 table = table->child;
1500 if (nr_subheaders == 1) {
1501 header = __register_sysctl_table(set, new_path, table);
1503 header->ctl_table_arg = ctl_table_arg;
1505 header = kzalloc(sizeof(*header) +
1506 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1510 subheaders = (struct ctl_table_header **) (header + 1);
1511 subheader = subheaders;
1512 header->ctl_table_arg = ctl_table_arg;
1514 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1516 goto err_register_leaves;
1523 err_register_leaves:
1524 while (subheader > subheaders) {
1525 struct ctl_table_header *subh = *(--subheader);
1526 struct ctl_table *table = subh->ctl_table_arg;
1527 unregister_sysctl_table(subh);
1536 * register_sysctl_table_path - register a sysctl table hierarchy
1537 * @path: The path to the directory the sysctl table is in.
1538 * @table: the top-level table structure
1540 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1541 * array. A completely 0 filled entry terminates the table.
1543 * See __register_sysctl_paths for more details.
1545 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1546 struct ctl_table *table)
1548 return __register_sysctl_paths(&sysctl_table_root.default_set,
1551 EXPORT_SYMBOL(register_sysctl_paths);
1554 * register_sysctl_table - register a sysctl table hierarchy
1555 * @table: the top-level table structure
1557 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1558 * array. A completely 0 filled entry terminates the table.
1560 * See register_sysctl_paths for more details.
1562 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1564 static const struct ctl_path null_path[] = { {} };
1566 return register_sysctl_paths(null_path, table);
1568 EXPORT_SYMBOL(register_sysctl_table);
1570 static void put_links(struct ctl_table_header *header)
1572 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1573 struct ctl_table_root *root = header->root;
1574 struct ctl_dir *parent = header->parent;
1575 struct ctl_dir *core_parent;
1576 struct ctl_table *entry;
1578 if (header->set == root_set)
1581 core_parent = xlate_dir(root_set, parent);
1582 if (IS_ERR(core_parent))
1585 for (entry = header->ctl_table; entry->procname; entry++) {
1586 struct ctl_table_header *link_head;
1587 struct ctl_table *link;
1588 const char *name = entry->procname;
1590 link = find_entry(&link_head, core_parent, name, strlen(name));
1592 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1593 (S_ISLNK(link->mode) && (link->data == root)))) {
1594 drop_sysctl_table(link_head);
1597 pr_err("sysctl link missing during unregister: ");
1598 sysctl_print_dir(parent);
1599 pr_cont("/%s\n", name);
1604 static void drop_sysctl_table(struct ctl_table_header *header)
1606 struct ctl_dir *parent = header->parent;
1613 start_unregistering(header);
1616 if (!--header->count)
1617 kfree_rcu(header, rcu);
1620 drop_sysctl_table(&parent->header);
1624 * unregister_sysctl_table - unregister a sysctl table hierarchy
1625 * @header: the header returned from register_sysctl_table
1627 * Unregisters the sysctl table and all children. proc entries may not
1628 * actually be removed until they are no longer used by anyone.
1630 void unregister_sysctl_table(struct ctl_table_header * header)
1638 nr_subheaders = count_subheaders(header->ctl_table_arg);
1639 if (unlikely(nr_subheaders > 1)) {
1640 struct ctl_table_header **subheaders;
1643 subheaders = (struct ctl_table_header **)(header + 1);
1644 for (i = nr_subheaders -1; i >= 0; i--) {
1645 struct ctl_table_header *subh = subheaders[i];
1646 struct ctl_table *table = subh->ctl_table_arg;
1647 unregister_sysctl_table(subh);
1654 spin_lock(&sysctl_lock);
1655 drop_sysctl_table(header);
1656 spin_unlock(&sysctl_lock);
1658 EXPORT_SYMBOL(unregister_sysctl_table);
1660 void setup_sysctl_set(struct ctl_table_set *set,
1661 struct ctl_table_root *root,
1662 int (*is_seen)(struct ctl_table_set *))
1664 memset(set, 0, sizeof(*set));
1665 set->is_seen = is_seen;
1666 init_header(&set->dir.header, root, set, NULL, root_table);
1669 void retire_sysctl_set(struct ctl_table_set *set)
1671 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1674 int __init proc_sys_init(void)
1676 struct proc_dir_entry *proc_sys_root;
1678 proc_sys_root = proc_mkdir("sys", NULL);
1679 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1680 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1681 proc_sys_root->nlink = 0;
1683 return sysctl_init();