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/cred.h>
12 #include <linux/namei.h>
14 #include <linux/module.h>
17 static const struct dentry_operations proc_sys_dentry_operations;
18 static const struct file_operations proc_sys_file_operations;
19 static const struct inode_operations proc_sys_inode_operations;
20 static const struct file_operations proc_sys_dir_file_operations;
21 static const struct inode_operations proc_sys_dir_operations;
23 /* Support for permanently empty directories */
25 struct ctl_table sysctl_mount_point[] = {
29 static bool is_empty_dir(struct ctl_table_header *head)
31 return head->ctl_table[0].child == sysctl_mount_point;
34 static void set_empty_dir(struct ctl_dir *dir)
36 dir->header.ctl_table[0].child = sysctl_mount_point;
39 static void clear_empty_dir(struct ctl_dir *dir)
42 dir->header.ctl_table[0].child = NULL;
45 void proc_sys_poll_notify(struct ctl_table_poll *poll)
50 atomic_inc(&poll->event);
51 wake_up_interruptible(&poll->wait);
54 static struct ctl_table root_table[] = {
57 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
61 static struct ctl_table_root sysctl_table_root = {
62 .default_set.dir.header = {
65 .ctl_table = root_table }},
66 .ctl_table_arg = root_table,
67 .root = &sysctl_table_root,
68 .set = &sysctl_table_root.default_set,
72 static DEFINE_SPINLOCK(sysctl_lock);
74 static void drop_sysctl_table(struct ctl_table_header *header);
75 static int sysctl_follow_link(struct ctl_table_header **phead,
76 struct ctl_table **pentry);
77 static int insert_links(struct ctl_table_header *head);
78 static void put_links(struct ctl_table_header *header);
80 static void sysctl_print_dir(struct ctl_dir *dir)
82 if (dir->header.parent)
83 sysctl_print_dir(dir->header.parent);
84 pr_cont("%s/", dir->header.ctl_table[0].procname);
87 static int namecmp(const char *name1, int len1, const char *name2, int len2)
96 cmp = memcmp(name1, name2, minlen);
102 /* Called under sysctl_lock */
103 static struct ctl_table *find_entry(struct ctl_table_header **phead,
104 struct ctl_dir *dir, const char *name, int namelen)
106 struct ctl_table_header *head;
107 struct ctl_table *entry;
108 struct rb_node *node = dir->root.rb_node;
112 struct ctl_node *ctl_node;
113 const char *procname;
116 ctl_node = rb_entry(node, struct ctl_node, node);
117 head = ctl_node->header;
118 entry = &head->ctl_table[ctl_node - head->node];
119 procname = entry->procname;
121 cmp = namecmp(name, namelen, procname, strlen(procname));
123 node = node->rb_left;
125 node = node->rb_right;
134 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
136 struct rb_node *node = &head->node[entry - head->ctl_table].node;
137 struct rb_node **p = &head->parent->root.rb_node;
138 struct rb_node *parent = NULL;
139 const char *name = entry->procname;
140 int namelen = strlen(name);
143 struct ctl_table_header *parent_head;
144 struct ctl_table *parent_entry;
145 struct ctl_node *parent_node;
146 const char *parent_name;
150 parent_node = rb_entry(parent, struct ctl_node, node);
151 parent_head = parent_node->header;
152 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
153 parent_name = parent_entry->procname;
155 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
161 pr_err("sysctl duplicate entry: ");
162 sysctl_print_dir(head->parent);
163 pr_cont("/%s\n", entry->procname);
168 rb_link_node(node, parent, p);
169 rb_insert_color(node, &head->parent->root);
173 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
175 struct rb_node *node = &head->node[entry - head->ctl_table].node;
177 rb_erase(node, &head->parent->root);
180 static void init_header(struct ctl_table_header *head,
181 struct ctl_table_root *root, struct ctl_table_set *set,
182 struct ctl_node *node, struct ctl_table *table)
184 head->ctl_table = table;
185 head->ctl_table_arg = table;
189 head->unregistering = NULL;
194 INIT_LIST_HEAD(&head->inodes);
196 struct ctl_table *entry;
197 for (entry = table; entry->procname; entry++, node++)
202 static void erase_header(struct ctl_table_header *head)
204 struct ctl_table *entry;
205 for (entry = head->ctl_table; entry->procname; entry++)
206 erase_entry(head, entry);
209 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
211 struct ctl_table *entry;
214 /* Is this a permanently empty directory? */
215 if (is_empty_dir(&dir->header))
218 /* Am I creating a permanently empty directory? */
219 if (header->ctl_table == sysctl_mount_point) {
220 if (!RB_EMPTY_ROOT(&dir->root))
226 header->parent = dir;
227 err = insert_links(header);
230 for (entry = header->ctl_table; entry->procname; entry++) {
231 err = insert_entry(header, entry);
237 erase_header(header);
240 if (header->ctl_table == sysctl_mount_point)
241 clear_empty_dir(dir);
242 header->parent = NULL;
243 drop_sysctl_table(&dir->header);
247 /* called under sysctl_lock */
248 static int use_table(struct ctl_table_header *p)
250 if (unlikely(p->unregistering))
256 /* called under sysctl_lock */
257 static void unuse_table(struct ctl_table_header *p)
260 if (unlikely(p->unregistering))
261 complete(p->unregistering);
264 /* called under sysctl_lock */
265 static void proc_sys_prune_dcache(struct ctl_table_header *head)
267 struct inode *inode, *prev = NULL;
268 struct proc_inode *ei;
271 list_for_each_entry_rcu(ei, &head->inodes, sysctl_inodes) {
272 inode = igrab(&ei->vfs_inode);
277 d_prune_aliases(inode);
285 /* called under sysctl_lock, will reacquire if has to wait */
286 static void start_unregistering(struct ctl_table_header *p)
289 * if p->used is 0, nobody will ever touch that entry again;
290 * we'll eliminate all paths to it before dropping sysctl_lock
292 if (unlikely(p->used)) {
293 struct completion wait;
294 init_completion(&wait);
295 p->unregistering = &wait;
296 spin_unlock(&sysctl_lock);
297 wait_for_completion(&wait);
299 /* anything non-NULL; we'll never dereference it */
300 p->unregistering = ERR_PTR(-EINVAL);
301 spin_unlock(&sysctl_lock);
304 * Prune dentries for unregistered sysctls: namespaced sysctls
305 * can have duplicate names and contaminate dcache very badly.
307 proc_sys_prune_dcache(p);
309 * do not remove from the list until nobody holds it; walking the
310 * list in do_sysctl() relies on that.
312 spin_lock(&sysctl_lock);
316 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
319 spin_lock(&sysctl_lock);
320 if (!use_table(head))
321 head = ERR_PTR(-ENOENT);
322 spin_unlock(&sysctl_lock);
326 static void sysctl_head_finish(struct ctl_table_header *head)
330 spin_lock(&sysctl_lock);
332 spin_unlock(&sysctl_lock);
335 static struct ctl_table_set *
336 lookup_header_set(struct ctl_table_root *root)
338 struct ctl_table_set *set = &root->default_set;
340 set = root->lookup(root);
344 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
346 const char *name, int namelen)
348 struct ctl_table_header *head;
349 struct ctl_table *entry;
351 spin_lock(&sysctl_lock);
352 entry = find_entry(&head, dir, name, namelen);
353 if (entry && use_table(head))
357 spin_unlock(&sysctl_lock);
361 static struct ctl_node *first_usable_entry(struct rb_node *node)
363 struct ctl_node *ctl_node;
365 for (;node; node = rb_next(node)) {
366 ctl_node = rb_entry(node, struct ctl_node, node);
367 if (use_table(ctl_node->header))
373 static void first_entry(struct ctl_dir *dir,
374 struct ctl_table_header **phead, struct ctl_table **pentry)
376 struct ctl_table_header *head = NULL;
377 struct ctl_table *entry = NULL;
378 struct ctl_node *ctl_node;
380 spin_lock(&sysctl_lock);
381 ctl_node = first_usable_entry(rb_first(&dir->root));
382 spin_unlock(&sysctl_lock);
384 head = ctl_node->header;
385 entry = &head->ctl_table[ctl_node - head->node];
391 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
393 struct ctl_table_header *head = *phead;
394 struct ctl_table *entry = *pentry;
395 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
397 spin_lock(&sysctl_lock);
400 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
401 spin_unlock(&sysctl_lock);
404 head = ctl_node->header;
405 entry = &head->ctl_table[ctl_node - head->node];
411 void register_sysctl_root(struct ctl_table_root *root)
416 * sysctl_perm does NOT grant the superuser all rights automatically, because
417 * some sysctl variables are readonly even to root.
420 static int test_perm(int mode, int op)
422 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
424 else if (in_egroup_p(GLOBAL_ROOT_GID))
426 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
431 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
433 struct ctl_table_root *root = head->root;
436 if (root->permissions)
437 mode = root->permissions(head, table);
441 return test_perm(mode, op);
444 static struct inode *proc_sys_make_inode(struct super_block *sb,
445 struct ctl_table_header *head, struct ctl_table *table)
447 struct ctl_table_root *root = head->root;
449 struct proc_inode *ei;
451 inode = new_inode(sb);
455 inode->i_ino = get_next_ino();
459 spin_lock(&sysctl_lock);
460 if (unlikely(head->unregistering)) {
461 spin_unlock(&sysctl_lock);
467 ei->sysctl_entry = table;
468 list_add_rcu(&ei->sysctl_inodes, &head->inodes);
470 spin_unlock(&sysctl_lock);
472 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
473 inode->i_mode = table->mode;
474 if (!S_ISDIR(table->mode)) {
475 inode->i_mode |= S_IFREG;
476 inode->i_op = &proc_sys_inode_operations;
477 inode->i_fop = &proc_sys_file_operations;
479 inode->i_mode |= S_IFDIR;
480 inode->i_op = &proc_sys_dir_operations;
481 inode->i_fop = &proc_sys_dir_file_operations;
482 if (is_empty_dir(head))
483 make_empty_dir_inode(inode);
486 if (root->set_ownership)
487 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
493 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
495 spin_lock(&sysctl_lock);
496 list_del_rcu(&PROC_I(inode)->sysctl_inodes);
498 kfree_rcu(head, rcu);
499 spin_unlock(&sysctl_lock);
502 static struct ctl_table_header *grab_header(struct inode *inode)
504 struct ctl_table_header *head = PROC_I(inode)->sysctl;
506 head = &sysctl_table_root.default_set.dir.header;
507 return sysctl_head_grab(head);
510 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
513 struct ctl_table_header *head = grab_header(dir);
514 struct ctl_table_header *h = NULL;
515 const struct qstr *name = &dentry->d_name;
518 struct dentry *err = ERR_PTR(-ENOENT);
519 struct ctl_dir *ctl_dir;
523 return ERR_CAST(head);
525 ctl_dir = container_of(head, struct ctl_dir, header);
527 p = lookup_entry(&h, ctl_dir, name->name, name->len);
531 if (S_ISLNK(p->mode)) {
532 ret = sysctl_follow_link(&h, &p);
538 err = ERR_PTR(-ENOMEM);
539 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
544 d_set_d_op(dentry, &proc_sys_dentry_operations);
545 d_add(dentry, inode);
549 sysctl_head_finish(h);
550 sysctl_head_finish(head);
554 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
555 size_t count, loff_t *ppos, int write)
557 struct inode *inode = file_inode(filp);
558 struct ctl_table_header *head = grab_header(inode);
559 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
564 return PTR_ERR(head);
567 * At this point we know that the sysctl was not unregistered
568 * and won't be until we finish.
571 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
574 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
576 if (!table->proc_handler)
579 /* careful: calling conventions are nasty here */
581 error = table->proc_handler(table, write, buf, &res, ppos);
585 sysctl_head_finish(head);
590 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
591 size_t count, loff_t *ppos)
593 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
596 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
597 size_t count, loff_t *ppos)
599 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
602 static int proc_sys_open(struct inode *inode, struct file *filp)
604 struct ctl_table_header *head = grab_header(inode);
605 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
607 /* sysctl was unregistered */
609 return PTR_ERR(head);
612 filp->private_data = proc_sys_poll_event(table->poll);
614 sysctl_head_finish(head);
619 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
621 struct inode *inode = file_inode(filp);
622 struct ctl_table_header *head = grab_header(inode);
623 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
624 unsigned int ret = DEFAULT_POLLMASK;
627 /* sysctl was unregistered */
629 return POLLERR | POLLHUP;
631 if (!table->proc_handler)
637 event = (unsigned long)filp->private_data;
638 poll_wait(filp, &table->poll->wait, wait);
640 if (event != atomic_read(&table->poll->event)) {
641 filp->private_data = proc_sys_poll_event(table->poll);
642 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
646 sysctl_head_finish(head);
651 static bool proc_sys_fill_cache(struct file *file,
652 struct dir_context *ctx,
653 struct ctl_table_header *head,
654 struct ctl_table *table)
656 struct dentry *child, *dir = file->f_path.dentry;
660 unsigned type = DT_UNKNOWN;
662 qname.name = table->procname;
663 qname.len = strlen(table->procname);
664 qname.hash = full_name_hash(dir, qname.name, qname.len);
666 child = d_lookup(dir, &qname);
668 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
669 child = d_alloc_parallel(dir, &qname, &wq);
672 if (d_in_lookup(child)) {
673 inode = proc_sys_make_inode(dir->d_sb, head, table);
675 d_lookup_done(child);
679 d_set_d_op(child, &proc_sys_dentry_operations);
683 inode = d_inode(child);
685 type = inode->i_mode >> 12;
687 return dir_emit(ctx, qname.name, qname.len, ino, type);
690 static bool proc_sys_link_fill_cache(struct file *file,
691 struct dir_context *ctx,
692 struct ctl_table_header *head,
693 struct ctl_table *table)
696 head = sysctl_head_grab(head);
698 if (S_ISLNK(table->mode)) {
699 /* It is not an error if we can not follow the link ignore it */
700 int err = sysctl_follow_link(&head, &table);
705 ret = proc_sys_fill_cache(file, ctx, head, table);
707 sysctl_head_finish(head);
711 static int scan(struct ctl_table_header *head, struct ctl_table *table,
712 unsigned long *pos, struct file *file,
713 struct dir_context *ctx)
717 if ((*pos)++ < ctx->pos)
720 if (unlikely(S_ISLNK(table->mode)))
721 res = proc_sys_link_fill_cache(file, ctx, head, table);
723 res = proc_sys_fill_cache(file, ctx, head, table);
731 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
733 struct ctl_table_header *head = grab_header(file_inode(file));
734 struct ctl_table_header *h = NULL;
735 struct ctl_table *entry;
736 struct ctl_dir *ctl_dir;
740 return PTR_ERR(head);
742 ctl_dir = container_of(head, struct ctl_dir, header);
744 if (!dir_emit_dots(file, ctx))
749 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
750 if (!scan(h, entry, &pos, file, ctx)) {
751 sysctl_head_finish(h);
756 sysctl_head_finish(head);
760 static int proc_sys_permission(struct inode *inode, int mask)
763 * sysctl entries that are not writeable,
764 * are _NOT_ writeable, capabilities or not.
766 struct ctl_table_header *head;
767 struct ctl_table *table;
770 /* Executable files are not allowed under /proc/sys/ */
771 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
774 head = grab_header(inode);
776 return PTR_ERR(head);
778 table = PROC_I(inode)->sysctl_entry;
779 if (!table) /* global root - r-xr-xr-x */
780 error = mask & MAY_WRITE ? -EACCES : 0;
781 else /* Use the permissions on the sysctl table entry */
782 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
784 sysctl_head_finish(head);
788 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
790 struct inode *inode = d_inode(dentry);
793 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
796 error = setattr_prepare(dentry, attr);
800 setattr_copy(inode, attr);
801 mark_inode_dirty(inode);
805 static int proc_sys_getattr(const struct path *path, struct kstat *stat,
806 u32 request_mask, unsigned int query_flags)
808 struct inode *inode = d_inode(path->dentry);
809 struct ctl_table_header *head = grab_header(inode);
810 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
813 return PTR_ERR(head);
815 generic_fillattr(inode, stat);
817 stat->mode = (stat->mode & S_IFMT) | table->mode;
819 sysctl_head_finish(head);
823 static const struct file_operations proc_sys_file_operations = {
824 .open = proc_sys_open,
825 .poll = proc_sys_poll,
826 .read = proc_sys_read,
827 .write = proc_sys_write,
828 .llseek = default_llseek,
831 static const struct file_operations proc_sys_dir_file_operations = {
832 .read = generic_read_dir,
833 .iterate_shared = proc_sys_readdir,
834 .llseek = generic_file_llseek,
837 static const struct inode_operations proc_sys_inode_operations = {
838 .permission = proc_sys_permission,
839 .setattr = proc_sys_setattr,
840 .getattr = proc_sys_getattr,
843 static const struct inode_operations proc_sys_dir_operations = {
844 .lookup = proc_sys_lookup,
845 .permission = proc_sys_permission,
846 .setattr = proc_sys_setattr,
847 .getattr = proc_sys_getattr,
850 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
852 if (flags & LOOKUP_RCU)
854 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
857 static int proc_sys_delete(const struct dentry *dentry)
859 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
862 static int sysctl_is_seen(struct ctl_table_header *p)
864 struct ctl_table_set *set = p->set;
866 spin_lock(&sysctl_lock);
867 if (p->unregistering)
869 else if (!set->is_seen)
872 res = set->is_seen(set);
873 spin_unlock(&sysctl_lock);
877 static int proc_sys_compare(const struct dentry *dentry,
878 unsigned int len, const char *str, const struct qstr *name)
880 struct ctl_table_header *head;
883 /* Although proc doesn't have negative dentries, rcu-walk means
884 * that inode here can be NULL */
885 /* AV: can it, indeed? */
886 inode = d_inode_rcu(dentry);
889 if (name->len != len)
891 if (memcmp(name->name, str, len))
893 head = rcu_dereference(PROC_I(inode)->sysctl);
894 return !head || !sysctl_is_seen(head);
897 static const struct dentry_operations proc_sys_dentry_operations = {
898 .d_revalidate = proc_sys_revalidate,
899 .d_delete = proc_sys_delete,
900 .d_compare = proc_sys_compare,
903 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
904 const char *name, int namelen)
906 struct ctl_table_header *head;
907 struct ctl_table *entry;
909 entry = find_entry(&head, dir, name, namelen);
911 return ERR_PTR(-ENOENT);
912 if (!S_ISDIR(entry->mode))
913 return ERR_PTR(-ENOTDIR);
914 return container_of(head, struct ctl_dir, header);
917 static struct ctl_dir *new_dir(struct ctl_table_set *set,
918 const char *name, int namelen)
920 struct ctl_table *table;
922 struct ctl_node *node;
925 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
926 sizeof(struct ctl_table)*2 + namelen + 1,
931 node = (struct ctl_node *)(new + 1);
932 table = (struct ctl_table *)(node + 1);
933 new_name = (char *)(table + 2);
934 memcpy(new_name, name, namelen);
935 new_name[namelen] = '\0';
936 table[0].procname = new_name;
937 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
938 init_header(&new->header, set->dir.header.root, set, node, table);
944 * get_subdir - find or create a subdir with the specified name.
945 * @dir: Directory to create the subdirectory in
946 * @name: The name of the subdirectory to find or create
947 * @namelen: The length of name
949 * Takes a directory with an elevated reference count so we know that
950 * if we drop the lock the directory will not go away. Upon success
951 * the reference is moved from @dir to the returned subdirectory.
952 * Upon error an error code is returned and the reference on @dir is
955 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
956 const char *name, int namelen)
958 struct ctl_table_set *set = dir->header.set;
959 struct ctl_dir *subdir, *new = NULL;
962 spin_lock(&sysctl_lock);
963 subdir = find_subdir(dir, name, namelen);
966 if (PTR_ERR(subdir) != -ENOENT)
969 spin_unlock(&sysctl_lock);
970 new = new_dir(set, name, namelen);
971 spin_lock(&sysctl_lock);
972 subdir = ERR_PTR(-ENOMEM);
976 /* Was the subdir added while we dropped the lock? */
977 subdir = find_subdir(dir, name, namelen);
980 if (PTR_ERR(subdir) != -ENOENT)
983 /* Nope. Use the our freshly made directory entry. */
984 err = insert_header(dir, &new->header);
985 subdir = ERR_PTR(err);
990 subdir->header.nreg++;
992 if (IS_ERR(subdir)) {
993 pr_err("sysctl could not get directory: ");
994 sysctl_print_dir(dir);
995 pr_cont("/%*.*s %ld\n",
996 namelen, namelen, name, PTR_ERR(subdir));
998 drop_sysctl_table(&dir->header);
1000 drop_sysctl_table(&new->header);
1001 spin_unlock(&sysctl_lock);
1005 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1007 struct ctl_dir *parent;
1008 const char *procname;
1009 if (!dir->header.parent)
1011 parent = xlate_dir(set, dir->header.parent);
1014 procname = dir->header.ctl_table[0].procname;
1015 return find_subdir(parent, procname, strlen(procname));
1018 static int sysctl_follow_link(struct ctl_table_header **phead,
1019 struct ctl_table **pentry)
1021 struct ctl_table_header *head;
1022 struct ctl_table_root *root;
1023 struct ctl_table_set *set;
1024 struct ctl_table *entry;
1025 struct ctl_dir *dir;
1029 spin_lock(&sysctl_lock);
1030 root = (*pentry)->data;
1031 set = lookup_header_set(root);
1032 dir = xlate_dir(set, (*phead)->parent);
1036 const char *procname = (*pentry)->procname;
1038 entry = find_entry(&head, dir, procname, strlen(procname));
1040 if (entry && use_table(head)) {
1041 unuse_table(*phead);
1048 spin_unlock(&sysctl_lock);
1052 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1054 struct va_format vaf;
1057 va_start(args, fmt);
1061 pr_err("sysctl table check failed: %s/%s %pV\n",
1062 path, table->procname, &vaf);
1068 static int sysctl_check_table(const char *path, struct ctl_table *table)
1071 for (; table->procname; table++) {
1073 err = sysctl_err(path, table, "Not a file");
1075 if ((table->proc_handler == proc_dostring) ||
1076 (table->proc_handler == proc_dointvec) ||
1077 (table->proc_handler == proc_douintvec) ||
1078 (table->proc_handler == proc_dointvec_minmax) ||
1079 (table->proc_handler == proc_dointvec_jiffies) ||
1080 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1081 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1082 (table->proc_handler == proc_doulongvec_minmax) ||
1083 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1085 err = sysctl_err(path, table, "No data");
1087 err = sysctl_err(path, table, "No maxlen");
1089 if (!table->proc_handler)
1090 err = sysctl_err(path, table, "No proc_handler");
1092 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1093 err = sysctl_err(path, table, "bogus .mode 0%o",
1099 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1100 struct ctl_table_root *link_root)
1102 struct ctl_table *link_table, *entry, *link;
1103 struct ctl_table_header *links;
1104 struct ctl_node *node;
1106 int nr_entries, name_bytes;
1110 for (entry = table; entry->procname; entry++) {
1112 name_bytes += strlen(entry->procname) + 1;
1115 links = kzalloc(sizeof(struct ctl_table_header) +
1116 sizeof(struct ctl_node)*nr_entries +
1117 sizeof(struct ctl_table)*(nr_entries + 1) +
1124 node = (struct ctl_node *)(links + 1);
1125 link_table = (struct ctl_table *)(node + nr_entries);
1126 link_name = (char *)&link_table[nr_entries + 1];
1128 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1129 int len = strlen(entry->procname) + 1;
1130 memcpy(link_name, entry->procname, len);
1131 link->procname = link_name;
1132 link->mode = S_IFLNK|S_IRWXUGO;
1133 link->data = link_root;
1136 init_header(links, dir->header.root, dir->header.set, node, link_table);
1137 links->nreg = nr_entries;
1142 static bool get_links(struct ctl_dir *dir,
1143 struct ctl_table *table, struct ctl_table_root *link_root)
1145 struct ctl_table_header *head;
1146 struct ctl_table *entry, *link;
1148 /* Are there links available for every entry in table? */
1149 for (entry = table; entry->procname; entry++) {
1150 const char *procname = entry->procname;
1151 link = find_entry(&head, dir, procname, strlen(procname));
1154 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1156 if (S_ISLNK(link->mode) && (link->data == link_root))
1161 /* The checks passed. Increase the registration count on the links */
1162 for (entry = table; entry->procname; entry++) {
1163 const char *procname = entry->procname;
1164 link = find_entry(&head, dir, procname, strlen(procname));
1170 static int insert_links(struct ctl_table_header *head)
1172 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1173 struct ctl_dir *core_parent = NULL;
1174 struct ctl_table_header *links;
1177 if (head->set == root_set)
1180 core_parent = xlate_dir(root_set, head->parent);
1181 if (IS_ERR(core_parent))
1184 if (get_links(core_parent, head->ctl_table, head->root))
1187 core_parent->header.nreg++;
1188 spin_unlock(&sysctl_lock);
1190 links = new_links(core_parent, head->ctl_table, head->root);
1192 spin_lock(&sysctl_lock);
1198 if (get_links(core_parent, head->ctl_table, head->root)) {
1203 err = insert_header(core_parent, links);
1207 drop_sysctl_table(&core_parent->header);
1212 * __register_sysctl_table - register a leaf sysctl table
1213 * @set: Sysctl tree to register on
1214 * @path: The path to the directory the sysctl table is in.
1215 * @table: the top-level table structure
1217 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1218 * array. A completely 0 filled entry terminates the table.
1220 * The members of the &struct ctl_table structure are used as follows:
1222 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1223 * enter a sysctl file
1225 * data - a pointer to data for use by proc_handler
1227 * maxlen - the maximum size in bytes of the data
1229 * mode - the file permissions for the /proc/sys file
1231 * child - must be %NULL.
1233 * proc_handler - the text handler routine (described below)
1235 * extra1, extra2 - extra pointers usable by the proc handler routines
1237 * Leaf nodes in the sysctl tree will be represented by a single file
1238 * under /proc; non-leaf nodes will be represented by directories.
1240 * There must be a proc_handler routine for any terminal nodes.
1241 * Several default handlers are available to cover common cases -
1243 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1244 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1245 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1247 * It is the handler's job to read the input buffer from user memory
1248 * and process it. The handler should return 0 on success.
1250 * This routine returns %NULL on a failure to register, and a pointer
1251 * to the table header on success.
1253 struct ctl_table_header *__register_sysctl_table(
1254 struct ctl_table_set *set,
1255 const char *path, struct ctl_table *table)
1257 struct ctl_table_root *root = set->dir.header.root;
1258 struct ctl_table_header *header;
1259 const char *name, *nextname;
1260 struct ctl_dir *dir;
1261 struct ctl_table *entry;
1262 struct ctl_node *node;
1265 for (entry = table; entry->procname; entry++)
1268 header = kzalloc(sizeof(struct ctl_table_header) +
1269 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1273 node = (struct ctl_node *)(header + 1);
1274 init_header(header, root, set, node, table);
1275 if (sysctl_check_table(path, table))
1278 spin_lock(&sysctl_lock);
1280 /* Reference moved down the diretory tree get_subdir */
1282 spin_unlock(&sysctl_lock);
1284 /* Find the directory for the ctl_table */
1285 for (name = path; name; name = nextname) {
1287 nextname = strchr(name, '/');
1289 namelen = nextname - name;
1292 namelen = strlen(name);
1297 dir = get_subdir(dir, name, namelen);
1302 spin_lock(&sysctl_lock);
1303 if (insert_header(dir, header))
1304 goto fail_put_dir_locked;
1306 drop_sysctl_table(&dir->header);
1307 spin_unlock(&sysctl_lock);
1311 fail_put_dir_locked:
1312 drop_sysctl_table(&dir->header);
1313 spin_unlock(&sysctl_lock);
1321 * register_sysctl - register a sysctl table
1322 * @path: The path to the directory the sysctl table is in.
1323 * @table: the table structure
1325 * Register a sysctl table. @table should be a filled in ctl_table
1326 * array. A completely 0 filled entry terminates the table.
1328 * See __register_sysctl_table for more details.
1330 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1332 return __register_sysctl_table(&sysctl_table_root.default_set,
1335 EXPORT_SYMBOL(register_sysctl);
1337 static char *append_path(const char *path, char *pos, const char *name)
1340 namelen = strlen(name);
1341 if (((pos - path) + namelen + 2) >= PATH_MAX)
1343 memcpy(pos, name, namelen);
1345 pos[namelen + 1] = '\0';
1350 static int count_subheaders(struct ctl_table *table)
1353 int nr_subheaders = 0;
1354 struct ctl_table *entry;
1356 /* special case: no directory and empty directory */
1357 if (!table || !table->procname)
1360 for (entry = table; entry->procname; entry++) {
1362 nr_subheaders += count_subheaders(entry->child);
1366 return nr_subheaders + has_files;
1369 static int register_leaf_sysctl_tables(const char *path, char *pos,
1370 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1371 struct ctl_table *table)
1373 struct ctl_table *ctl_table_arg = NULL;
1374 struct ctl_table *entry, *files;
1379 for (entry = table; entry->procname; entry++) {
1387 /* If there are mixed files and directories we need a new table */
1388 if (nr_dirs && nr_files) {
1389 struct ctl_table *new;
1390 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1395 ctl_table_arg = files;
1396 for (new = files, entry = table; entry->procname; entry++) {
1404 /* Register everything except a directory full of subdirectories */
1405 if (nr_files || !nr_dirs) {
1406 struct ctl_table_header *header;
1407 header = __register_sysctl_table(set, path, files);
1409 kfree(ctl_table_arg);
1413 /* Remember if we need to free the file table */
1414 header->ctl_table_arg = ctl_table_arg;
1415 **subheader = header;
1419 /* Recurse into the subdirectories. */
1420 for (entry = table; entry->procname; entry++) {
1426 err = -ENAMETOOLONG;
1427 child_pos = append_path(path, pos, entry->procname);
1431 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1439 /* On failure our caller will unregister all registered subheaders */
1444 * __register_sysctl_paths - register a sysctl table hierarchy
1445 * @set: Sysctl tree to register on
1446 * @path: The path to the directory the sysctl table is in.
1447 * @table: the top-level table structure
1449 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1450 * array. A completely 0 filled entry terminates the table.
1452 * See __register_sysctl_table for more details.
1454 struct ctl_table_header *__register_sysctl_paths(
1455 struct ctl_table_set *set,
1456 const struct ctl_path *path, struct ctl_table *table)
1458 struct ctl_table *ctl_table_arg = table;
1459 int nr_subheaders = count_subheaders(table);
1460 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1461 const struct ctl_path *component;
1462 char *new_path, *pos;
1464 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1469 for (component = path; component->procname; component++) {
1470 pos = append_path(new_path, pos, component->procname);
1474 while (table->procname && table->child && !table[1].procname) {
1475 pos = append_path(new_path, pos, table->procname);
1478 table = table->child;
1480 if (nr_subheaders == 1) {
1481 header = __register_sysctl_table(set, new_path, table);
1483 header->ctl_table_arg = ctl_table_arg;
1485 header = kzalloc(sizeof(*header) +
1486 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1490 subheaders = (struct ctl_table_header **) (header + 1);
1491 subheader = subheaders;
1492 header->ctl_table_arg = ctl_table_arg;
1494 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1496 goto err_register_leaves;
1503 err_register_leaves:
1504 while (subheader > subheaders) {
1505 struct ctl_table_header *subh = *(--subheader);
1506 struct ctl_table *table = subh->ctl_table_arg;
1507 unregister_sysctl_table(subh);
1516 * register_sysctl_table_path - register a sysctl table hierarchy
1517 * @path: The path to the directory the sysctl table is in.
1518 * @table: the top-level table structure
1520 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1521 * array. A completely 0 filled entry terminates the table.
1523 * See __register_sysctl_paths for more details.
1525 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1526 struct ctl_table *table)
1528 return __register_sysctl_paths(&sysctl_table_root.default_set,
1531 EXPORT_SYMBOL(register_sysctl_paths);
1534 * register_sysctl_table - register a sysctl table hierarchy
1535 * @table: the top-level table structure
1537 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1538 * array. A completely 0 filled entry terminates the table.
1540 * See register_sysctl_paths for more details.
1542 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1544 static const struct ctl_path null_path[] = { {} };
1546 return register_sysctl_paths(null_path, table);
1548 EXPORT_SYMBOL(register_sysctl_table);
1550 static void put_links(struct ctl_table_header *header)
1552 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1553 struct ctl_table_root *root = header->root;
1554 struct ctl_dir *parent = header->parent;
1555 struct ctl_dir *core_parent;
1556 struct ctl_table *entry;
1558 if (header->set == root_set)
1561 core_parent = xlate_dir(root_set, parent);
1562 if (IS_ERR(core_parent))
1565 for (entry = header->ctl_table; entry->procname; entry++) {
1566 struct ctl_table_header *link_head;
1567 struct ctl_table *link;
1568 const char *name = entry->procname;
1570 link = find_entry(&link_head, core_parent, name, strlen(name));
1572 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1573 (S_ISLNK(link->mode) && (link->data == root)))) {
1574 drop_sysctl_table(link_head);
1577 pr_err("sysctl link missing during unregister: ");
1578 sysctl_print_dir(parent);
1579 pr_cont("/%s\n", name);
1584 static void drop_sysctl_table(struct ctl_table_header *header)
1586 struct ctl_dir *parent = header->parent;
1592 start_unregistering(header);
1593 if (!--header->count)
1594 kfree_rcu(header, rcu);
1597 drop_sysctl_table(&parent->header);
1601 * unregister_sysctl_table - unregister a sysctl table hierarchy
1602 * @header: the header returned from register_sysctl_table
1604 * Unregisters the sysctl table and all children. proc entries may not
1605 * actually be removed until they are no longer used by anyone.
1607 void unregister_sysctl_table(struct ctl_table_header * header)
1615 nr_subheaders = count_subheaders(header->ctl_table_arg);
1616 if (unlikely(nr_subheaders > 1)) {
1617 struct ctl_table_header **subheaders;
1620 subheaders = (struct ctl_table_header **)(header + 1);
1621 for (i = nr_subheaders -1; i >= 0; i--) {
1622 struct ctl_table_header *subh = subheaders[i];
1623 struct ctl_table *table = subh->ctl_table_arg;
1624 unregister_sysctl_table(subh);
1631 spin_lock(&sysctl_lock);
1632 drop_sysctl_table(header);
1633 spin_unlock(&sysctl_lock);
1635 EXPORT_SYMBOL(unregister_sysctl_table);
1637 void setup_sysctl_set(struct ctl_table_set *set,
1638 struct ctl_table_root *root,
1639 int (*is_seen)(struct ctl_table_set *))
1641 memset(set, 0, sizeof(*set));
1642 set->is_seen = is_seen;
1643 init_header(&set->dir.header, root, set, NULL, root_table);
1646 void retire_sysctl_set(struct ctl_table_set *set)
1648 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1651 int __init proc_sys_init(void)
1653 struct proc_dir_entry *proc_sys_root;
1655 proc_sys_root = proc_mkdir("sys", NULL);
1656 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1657 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1658 proc_sys_root->nlink = 0;
1660 return sysctl_init();