1 // SPDX-License-Identifier: GPL-2.0
5 #include <linux/init.h>
6 #include <linux/sysctl.h>
7 #include <linux/poll.h>
8 #include <linux/proc_fs.h>
9 #include <linux/printk.h>
10 #include <linux/security.h>
11 #include <linux/sched.h>
12 #include <linux/cred.h>
13 #include <linux/namei.h>
15 #include <linux/module.h>
16 #include <linux/bpf-cgroup.h>
17 #include <linux/mount.h>
20 static const struct dentry_operations proc_sys_dentry_operations;
21 static const struct file_operations proc_sys_file_operations;
22 static const struct inode_operations proc_sys_inode_operations;
23 static const struct file_operations proc_sys_dir_file_operations;
24 static const struct inode_operations proc_sys_dir_operations;
26 /* shared constants to be used in various sysctls */
27 const int sysctl_vals[] = { 0, 1, INT_MAX };
28 EXPORT_SYMBOL(sysctl_vals);
30 /* Support for permanently empty directories */
32 struct ctl_table sysctl_mount_point[] = {
36 static bool is_empty_dir(struct ctl_table_header *head)
38 return head->ctl_table[0].child == sysctl_mount_point;
41 static void set_empty_dir(struct ctl_dir *dir)
43 dir->header.ctl_table[0].child = sysctl_mount_point;
46 static void clear_empty_dir(struct ctl_dir *dir)
49 dir->header.ctl_table[0].child = NULL;
52 void proc_sys_poll_notify(struct ctl_table_poll *poll)
57 atomic_inc(&poll->event);
58 wake_up_interruptible(&poll->wait);
61 static struct ctl_table root_table[] = {
64 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
68 static struct ctl_table_root sysctl_table_root = {
69 .default_set.dir.header = {
72 .ctl_table = root_table }},
73 .ctl_table_arg = root_table,
74 .root = &sysctl_table_root,
75 .set = &sysctl_table_root.default_set,
79 static DEFINE_SPINLOCK(sysctl_lock);
81 static void drop_sysctl_table(struct ctl_table_header *header);
82 static int sysctl_follow_link(struct ctl_table_header **phead,
83 struct ctl_table **pentry);
84 static int insert_links(struct ctl_table_header *head);
85 static void put_links(struct ctl_table_header *header);
87 static void sysctl_print_dir(struct ctl_dir *dir)
89 if (dir->header.parent)
90 sysctl_print_dir(dir->header.parent);
91 pr_cont("%s/", dir->header.ctl_table[0].procname);
94 static int namecmp(const char *name1, int len1, const char *name2, int len2)
103 cmp = memcmp(name1, name2, minlen);
109 /* Called under sysctl_lock */
110 static struct ctl_table *find_entry(struct ctl_table_header **phead,
111 struct ctl_dir *dir, const char *name, int namelen)
113 struct ctl_table_header *head;
114 struct ctl_table *entry;
115 struct rb_node *node = dir->root.rb_node;
119 struct ctl_node *ctl_node;
120 const char *procname;
123 ctl_node = rb_entry(node, struct ctl_node, node);
124 head = ctl_node->header;
125 entry = &head->ctl_table[ctl_node - head->node];
126 procname = entry->procname;
128 cmp = namecmp(name, namelen, procname, strlen(procname));
130 node = node->rb_left;
132 node = node->rb_right;
141 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
143 struct rb_node *node = &head->node[entry - head->ctl_table].node;
144 struct rb_node **p = &head->parent->root.rb_node;
145 struct rb_node *parent = NULL;
146 const char *name = entry->procname;
147 int namelen = strlen(name);
150 struct ctl_table_header *parent_head;
151 struct ctl_table *parent_entry;
152 struct ctl_node *parent_node;
153 const char *parent_name;
157 parent_node = rb_entry(parent, struct ctl_node, node);
158 parent_head = parent_node->header;
159 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
160 parent_name = parent_entry->procname;
162 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
168 pr_err("sysctl duplicate entry: ");
169 sysctl_print_dir(head->parent);
170 pr_cont("/%s\n", entry->procname);
175 rb_link_node(node, parent, p);
176 rb_insert_color(node, &head->parent->root);
180 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
182 struct rb_node *node = &head->node[entry - head->ctl_table].node;
184 rb_erase(node, &head->parent->root);
187 static void init_header(struct ctl_table_header *head,
188 struct ctl_table_root *root, struct ctl_table_set *set,
189 struct ctl_node *node, struct ctl_table *table)
191 head->ctl_table = table;
192 head->ctl_table_arg = table;
196 head->unregistering = NULL;
201 INIT_HLIST_HEAD(&head->inodes);
203 struct ctl_table *entry;
204 for (entry = table; entry->procname; entry++, node++)
209 static void erase_header(struct ctl_table_header *head)
211 struct ctl_table *entry;
212 for (entry = head->ctl_table; entry->procname; entry++)
213 erase_entry(head, entry);
216 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
218 struct ctl_table *entry;
221 /* Is this a permanently empty directory? */
222 if (is_empty_dir(&dir->header))
225 /* Am I creating a permanently empty directory? */
226 if (header->ctl_table == sysctl_mount_point) {
227 if (!RB_EMPTY_ROOT(&dir->root))
233 header->parent = dir;
234 err = insert_links(header);
237 for (entry = header->ctl_table; entry->procname; entry++) {
238 err = insert_entry(header, entry);
244 erase_header(header);
247 if (header->ctl_table == sysctl_mount_point)
248 clear_empty_dir(dir);
249 header->parent = NULL;
250 drop_sysctl_table(&dir->header);
254 /* called under sysctl_lock */
255 static int use_table(struct ctl_table_header *p)
257 if (unlikely(p->unregistering))
263 /* called under sysctl_lock */
264 static void unuse_table(struct ctl_table_header *p)
267 if (unlikely(p->unregistering))
268 complete(p->unregistering);
271 static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
273 proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
276 /* called under sysctl_lock, will reacquire if has to wait */
277 static void start_unregistering(struct ctl_table_header *p)
280 * if p->used is 0, nobody will ever touch that entry again;
281 * we'll eliminate all paths to it before dropping sysctl_lock
283 if (unlikely(p->used)) {
284 struct completion wait;
285 init_completion(&wait);
286 p->unregistering = &wait;
287 spin_unlock(&sysctl_lock);
288 wait_for_completion(&wait);
290 /* anything non-NULL; we'll never dereference it */
291 p->unregistering = ERR_PTR(-EINVAL);
292 spin_unlock(&sysctl_lock);
295 * Invalidate dentries for unregistered sysctls: namespaced sysctls
296 * can have duplicate names and contaminate dcache very badly.
298 proc_sys_invalidate_dcache(p);
300 * do not remove from the list until nobody holds it; walking the
301 * list in do_sysctl() relies on that.
303 spin_lock(&sysctl_lock);
307 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
310 spin_lock(&sysctl_lock);
311 if (!use_table(head))
312 head = ERR_PTR(-ENOENT);
313 spin_unlock(&sysctl_lock);
317 static void sysctl_head_finish(struct ctl_table_header *head)
321 spin_lock(&sysctl_lock);
323 spin_unlock(&sysctl_lock);
326 static struct ctl_table_set *
327 lookup_header_set(struct ctl_table_root *root)
329 struct ctl_table_set *set = &root->default_set;
331 set = root->lookup(root);
335 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
337 const char *name, int namelen)
339 struct ctl_table_header *head;
340 struct ctl_table *entry;
342 spin_lock(&sysctl_lock);
343 entry = find_entry(&head, dir, name, namelen);
344 if (entry && use_table(head))
348 spin_unlock(&sysctl_lock);
352 static struct ctl_node *first_usable_entry(struct rb_node *node)
354 struct ctl_node *ctl_node;
356 for (;node; node = rb_next(node)) {
357 ctl_node = rb_entry(node, struct ctl_node, node);
358 if (use_table(ctl_node->header))
364 static void first_entry(struct ctl_dir *dir,
365 struct ctl_table_header **phead, struct ctl_table **pentry)
367 struct ctl_table_header *head = NULL;
368 struct ctl_table *entry = NULL;
369 struct ctl_node *ctl_node;
371 spin_lock(&sysctl_lock);
372 ctl_node = first_usable_entry(rb_first(&dir->root));
373 spin_unlock(&sysctl_lock);
375 head = ctl_node->header;
376 entry = &head->ctl_table[ctl_node - head->node];
382 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
384 struct ctl_table_header *head = *phead;
385 struct ctl_table *entry = *pentry;
386 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
388 spin_lock(&sysctl_lock);
391 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
392 spin_unlock(&sysctl_lock);
395 head = ctl_node->header;
396 entry = &head->ctl_table[ctl_node - head->node];
403 * sysctl_perm does NOT grant the superuser all rights automatically, because
404 * some sysctl variables are readonly even to root.
407 static int test_perm(int mode, int op)
409 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
411 else if (in_egroup_p(GLOBAL_ROOT_GID))
413 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
418 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
420 struct ctl_table_root *root = head->root;
423 if (root->permissions)
424 mode = root->permissions(head, table);
428 return test_perm(mode, op);
431 static struct inode *proc_sys_make_inode(struct super_block *sb,
432 struct ctl_table_header *head, struct ctl_table *table)
434 struct ctl_table_root *root = head->root;
436 struct proc_inode *ei;
438 inode = new_inode(sb);
440 return ERR_PTR(-ENOMEM);
442 inode->i_ino = get_next_ino();
446 spin_lock(&sysctl_lock);
447 if (unlikely(head->unregistering)) {
448 spin_unlock(&sysctl_lock);
450 return ERR_PTR(-ENOENT);
453 ei->sysctl_entry = table;
454 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
456 spin_unlock(&sysctl_lock);
458 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
459 inode->i_mode = table->mode;
460 if (!S_ISDIR(table->mode)) {
461 inode->i_mode |= S_IFREG;
462 inode->i_op = &proc_sys_inode_operations;
463 inode->i_fop = &proc_sys_file_operations;
465 inode->i_mode |= S_IFDIR;
466 inode->i_op = &proc_sys_dir_operations;
467 inode->i_fop = &proc_sys_dir_file_operations;
468 if (is_empty_dir(head))
469 make_empty_dir_inode(inode);
472 if (root->set_ownership)
473 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
475 inode->i_uid = GLOBAL_ROOT_UID;
476 inode->i_gid = GLOBAL_ROOT_GID;
482 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
484 spin_lock(&sysctl_lock);
485 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
487 kfree_rcu(head, rcu);
488 spin_unlock(&sysctl_lock);
491 static struct ctl_table_header *grab_header(struct inode *inode)
493 struct ctl_table_header *head = PROC_I(inode)->sysctl;
495 head = &sysctl_table_root.default_set.dir.header;
496 return sysctl_head_grab(head);
499 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
502 struct ctl_table_header *head = grab_header(dir);
503 struct ctl_table_header *h = NULL;
504 const struct qstr *name = &dentry->d_name;
507 struct dentry *err = ERR_PTR(-ENOENT);
508 struct ctl_dir *ctl_dir;
512 return ERR_CAST(head);
514 ctl_dir = container_of(head, struct ctl_dir, header);
516 p = lookup_entry(&h, ctl_dir, name->name, name->len);
520 if (S_ISLNK(p->mode)) {
521 ret = sysctl_follow_link(&h, &p);
527 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
529 err = ERR_CAST(inode);
533 d_set_d_op(dentry, &proc_sys_dentry_operations);
534 err = d_splice_alias(inode, dentry);
538 sysctl_head_finish(h);
539 sysctl_head_finish(head);
543 static ssize_t proc_sys_call_handler(struct file *filp, void __user *ubuf,
544 size_t count, loff_t *ppos, int write)
546 struct inode *inode = file_inode(filp);
547 struct ctl_table_header *head = grab_header(inode);
548 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
553 return PTR_ERR(head);
556 * At this point we know that the sysctl was not unregistered
557 * and won't be until we finish.
560 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
563 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
565 if (!table->proc_handler)
569 kbuf = memdup_user_nul(ubuf, count);
571 error = PTR_ERR(kbuf);
576 kbuf = kzalloc(count, GFP_KERNEL);
581 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
586 /* careful: calling conventions are nasty here */
587 error = table->proc_handler(table, write, kbuf, &count, ppos);
593 if (copy_to_user(ubuf, kbuf, count))
601 sysctl_head_finish(head);
606 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
607 size_t count, loff_t *ppos)
609 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
612 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
613 size_t count, loff_t *ppos)
615 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
618 static int proc_sys_open(struct inode *inode, struct file *filp)
620 struct ctl_table_header *head = grab_header(inode);
621 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
623 /* sysctl was unregistered */
625 return PTR_ERR(head);
628 filp->private_data = proc_sys_poll_event(table->poll);
630 sysctl_head_finish(head);
635 static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
637 struct inode *inode = file_inode(filp);
638 struct ctl_table_header *head = grab_header(inode);
639 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
640 __poll_t ret = DEFAULT_POLLMASK;
643 /* sysctl was unregistered */
645 return EPOLLERR | EPOLLHUP;
647 if (!table->proc_handler)
653 event = (unsigned long)filp->private_data;
654 poll_wait(filp, &table->poll->wait, wait);
656 if (event != atomic_read(&table->poll->event)) {
657 filp->private_data = proc_sys_poll_event(table->poll);
658 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
662 sysctl_head_finish(head);
667 static bool proc_sys_fill_cache(struct file *file,
668 struct dir_context *ctx,
669 struct ctl_table_header *head,
670 struct ctl_table *table)
672 struct dentry *child, *dir = file->f_path.dentry;
676 unsigned type = DT_UNKNOWN;
678 qname.name = table->procname;
679 qname.len = strlen(table->procname);
680 qname.hash = full_name_hash(dir, qname.name, qname.len);
682 child = d_lookup(dir, &qname);
684 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
685 child = d_alloc_parallel(dir, &qname, &wq);
688 if (d_in_lookup(child)) {
690 inode = proc_sys_make_inode(dir->d_sb, head, table);
692 d_lookup_done(child);
696 d_set_d_op(child, &proc_sys_dentry_operations);
697 res = d_splice_alias(inode, child);
698 d_lookup_done(child);
709 inode = d_inode(child);
711 type = inode->i_mode >> 12;
713 return dir_emit(ctx, qname.name, qname.len, ino, type);
716 static bool proc_sys_link_fill_cache(struct file *file,
717 struct dir_context *ctx,
718 struct ctl_table_header *head,
719 struct ctl_table *table)
723 head = sysctl_head_grab(head);
727 /* It is not an error if we can not follow the link ignore it */
728 if (sysctl_follow_link(&head, &table))
731 ret = proc_sys_fill_cache(file, ctx, head, table);
733 sysctl_head_finish(head);
737 static int scan(struct ctl_table_header *head, struct ctl_table *table,
738 unsigned long *pos, struct file *file,
739 struct dir_context *ctx)
743 if ((*pos)++ < ctx->pos)
746 if (unlikely(S_ISLNK(table->mode)))
747 res = proc_sys_link_fill_cache(file, ctx, head, table);
749 res = proc_sys_fill_cache(file, ctx, head, table);
757 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
759 struct ctl_table_header *head = grab_header(file_inode(file));
760 struct ctl_table_header *h = NULL;
761 struct ctl_table *entry;
762 struct ctl_dir *ctl_dir;
766 return PTR_ERR(head);
768 ctl_dir = container_of(head, struct ctl_dir, header);
770 if (!dir_emit_dots(file, ctx))
775 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
776 if (!scan(h, entry, &pos, file, ctx)) {
777 sysctl_head_finish(h);
782 sysctl_head_finish(head);
786 static int proc_sys_permission(struct inode *inode, int mask)
789 * sysctl entries that are not writeable,
790 * are _NOT_ writeable, capabilities or not.
792 struct ctl_table_header *head;
793 struct ctl_table *table;
796 /* Executable files are not allowed under /proc/sys/ */
797 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
800 head = grab_header(inode);
802 return PTR_ERR(head);
804 table = PROC_I(inode)->sysctl_entry;
805 if (!table) /* global root - r-xr-xr-x */
806 error = mask & MAY_WRITE ? -EACCES : 0;
807 else /* Use the permissions on the sysctl table entry */
808 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
810 sysctl_head_finish(head);
814 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
816 struct inode *inode = d_inode(dentry);
819 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
822 error = setattr_prepare(dentry, attr);
826 setattr_copy(inode, attr);
827 mark_inode_dirty(inode);
831 static int proc_sys_getattr(const struct path *path, struct kstat *stat,
832 u32 request_mask, unsigned int query_flags)
834 struct inode *inode = d_inode(path->dentry);
835 struct ctl_table_header *head = grab_header(inode);
836 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
839 return PTR_ERR(head);
841 generic_fillattr(inode, stat);
843 stat->mode = (stat->mode & S_IFMT) | table->mode;
845 sysctl_head_finish(head);
849 static const struct file_operations proc_sys_file_operations = {
850 .open = proc_sys_open,
851 .poll = proc_sys_poll,
852 .read = proc_sys_read,
853 .write = proc_sys_write,
854 .llseek = default_llseek,
857 static const struct file_operations proc_sys_dir_file_operations = {
858 .read = generic_read_dir,
859 .iterate_shared = proc_sys_readdir,
860 .llseek = generic_file_llseek,
863 static const struct inode_operations proc_sys_inode_operations = {
864 .permission = proc_sys_permission,
865 .setattr = proc_sys_setattr,
866 .getattr = proc_sys_getattr,
869 static const struct inode_operations proc_sys_dir_operations = {
870 .lookup = proc_sys_lookup,
871 .permission = proc_sys_permission,
872 .setattr = proc_sys_setattr,
873 .getattr = proc_sys_getattr,
876 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
878 if (flags & LOOKUP_RCU)
880 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
883 static int proc_sys_delete(const struct dentry *dentry)
885 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
888 static int sysctl_is_seen(struct ctl_table_header *p)
890 struct ctl_table_set *set = p->set;
892 spin_lock(&sysctl_lock);
893 if (p->unregistering)
895 else if (!set->is_seen)
898 res = set->is_seen(set);
899 spin_unlock(&sysctl_lock);
903 static int proc_sys_compare(const struct dentry *dentry,
904 unsigned int len, const char *str, const struct qstr *name)
906 struct ctl_table_header *head;
909 /* Although proc doesn't have negative dentries, rcu-walk means
910 * that inode here can be NULL */
911 /* AV: can it, indeed? */
912 inode = d_inode_rcu(dentry);
915 if (name->len != len)
917 if (memcmp(name->name, str, len))
919 head = rcu_dereference(PROC_I(inode)->sysctl);
920 return !head || !sysctl_is_seen(head);
923 static const struct dentry_operations proc_sys_dentry_operations = {
924 .d_revalidate = proc_sys_revalidate,
925 .d_delete = proc_sys_delete,
926 .d_compare = proc_sys_compare,
929 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
930 const char *name, int namelen)
932 struct ctl_table_header *head;
933 struct ctl_table *entry;
935 entry = find_entry(&head, dir, name, namelen);
937 return ERR_PTR(-ENOENT);
938 if (!S_ISDIR(entry->mode))
939 return ERR_PTR(-ENOTDIR);
940 return container_of(head, struct ctl_dir, header);
943 static struct ctl_dir *new_dir(struct ctl_table_set *set,
944 const char *name, int namelen)
946 struct ctl_table *table;
948 struct ctl_node *node;
951 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
952 sizeof(struct ctl_table)*2 + namelen + 1,
957 node = (struct ctl_node *)(new + 1);
958 table = (struct ctl_table *)(node + 1);
959 new_name = (char *)(table + 2);
960 memcpy(new_name, name, namelen);
961 new_name[namelen] = '\0';
962 table[0].procname = new_name;
963 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
964 init_header(&new->header, set->dir.header.root, set, node, table);
970 * get_subdir - find or create a subdir with the specified name.
971 * @dir: Directory to create the subdirectory in
972 * @name: The name of the subdirectory to find or create
973 * @namelen: The length of name
975 * Takes a directory with an elevated reference count so we know that
976 * if we drop the lock the directory will not go away. Upon success
977 * the reference is moved from @dir to the returned subdirectory.
978 * Upon error an error code is returned and the reference on @dir is
981 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
982 const char *name, int namelen)
984 struct ctl_table_set *set = dir->header.set;
985 struct ctl_dir *subdir, *new = NULL;
988 spin_lock(&sysctl_lock);
989 subdir = find_subdir(dir, name, namelen);
992 if (PTR_ERR(subdir) != -ENOENT)
995 spin_unlock(&sysctl_lock);
996 new = new_dir(set, name, namelen);
997 spin_lock(&sysctl_lock);
998 subdir = ERR_PTR(-ENOMEM);
1002 /* Was the subdir added while we dropped the lock? */
1003 subdir = find_subdir(dir, name, namelen);
1004 if (!IS_ERR(subdir))
1006 if (PTR_ERR(subdir) != -ENOENT)
1009 /* Nope. Use the our freshly made directory entry. */
1010 err = insert_header(dir, &new->header);
1011 subdir = ERR_PTR(err);
1016 subdir->header.nreg++;
1018 if (IS_ERR(subdir)) {
1019 pr_err("sysctl could not get directory: ");
1020 sysctl_print_dir(dir);
1021 pr_cont("/%*.*s %ld\n",
1022 namelen, namelen, name, PTR_ERR(subdir));
1024 drop_sysctl_table(&dir->header);
1026 drop_sysctl_table(&new->header);
1027 spin_unlock(&sysctl_lock);
1031 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1033 struct ctl_dir *parent;
1034 const char *procname;
1035 if (!dir->header.parent)
1037 parent = xlate_dir(set, dir->header.parent);
1040 procname = dir->header.ctl_table[0].procname;
1041 return find_subdir(parent, procname, strlen(procname));
1044 static int sysctl_follow_link(struct ctl_table_header **phead,
1045 struct ctl_table **pentry)
1047 struct ctl_table_header *head;
1048 struct ctl_table_root *root;
1049 struct ctl_table_set *set;
1050 struct ctl_table *entry;
1051 struct ctl_dir *dir;
1055 spin_lock(&sysctl_lock);
1056 root = (*pentry)->data;
1057 set = lookup_header_set(root);
1058 dir = xlate_dir(set, (*phead)->parent);
1062 const char *procname = (*pentry)->procname;
1064 entry = find_entry(&head, dir, procname, strlen(procname));
1066 if (entry && use_table(head)) {
1067 unuse_table(*phead);
1074 spin_unlock(&sysctl_lock);
1078 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1080 struct va_format vaf;
1083 va_start(args, fmt);
1087 pr_err("sysctl table check failed: %s/%s %pV\n",
1088 path, table->procname, &vaf);
1094 static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1098 if ((table->proc_handler == proc_douintvec) ||
1099 (table->proc_handler == proc_douintvec_minmax)) {
1100 if (table->maxlen != sizeof(unsigned int))
1101 err |= sysctl_err(path, table, "array not allowed");
1107 static int sysctl_check_table(const char *path, struct ctl_table *table)
1110 for (; table->procname; table++) {
1112 err |= sysctl_err(path, table, "Not a file");
1114 if ((table->proc_handler == proc_dostring) ||
1115 (table->proc_handler == proc_dointvec) ||
1116 (table->proc_handler == proc_douintvec) ||
1117 (table->proc_handler == proc_douintvec_minmax) ||
1118 (table->proc_handler == proc_dointvec_minmax) ||
1119 (table->proc_handler == proc_dointvec_jiffies) ||
1120 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1121 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1122 (table->proc_handler == proc_doulongvec_minmax) ||
1123 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1125 err |= sysctl_err(path, table, "No data");
1127 err |= sysctl_err(path, table, "No maxlen");
1129 err |= sysctl_check_table_array(path, table);
1131 if (!table->proc_handler)
1132 err |= sysctl_err(path, table, "No proc_handler");
1134 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1135 err |= sysctl_err(path, table, "bogus .mode 0%o",
1141 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1142 struct ctl_table_root *link_root)
1144 struct ctl_table *link_table, *entry, *link;
1145 struct ctl_table_header *links;
1146 struct ctl_node *node;
1148 int nr_entries, name_bytes;
1152 for (entry = table; entry->procname; entry++) {
1154 name_bytes += strlen(entry->procname) + 1;
1157 links = kzalloc(sizeof(struct ctl_table_header) +
1158 sizeof(struct ctl_node)*nr_entries +
1159 sizeof(struct ctl_table)*(nr_entries + 1) +
1166 node = (struct ctl_node *)(links + 1);
1167 link_table = (struct ctl_table *)(node + nr_entries);
1168 link_name = (char *)&link_table[nr_entries + 1];
1170 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1171 int len = strlen(entry->procname) + 1;
1172 memcpy(link_name, entry->procname, len);
1173 link->procname = link_name;
1174 link->mode = S_IFLNK|S_IRWXUGO;
1175 link->data = link_root;
1178 init_header(links, dir->header.root, dir->header.set, node, link_table);
1179 links->nreg = nr_entries;
1184 static bool get_links(struct ctl_dir *dir,
1185 struct ctl_table *table, struct ctl_table_root *link_root)
1187 struct ctl_table_header *head;
1188 struct ctl_table *entry, *link;
1190 /* Are there links available for every entry in table? */
1191 for (entry = table; entry->procname; entry++) {
1192 const char *procname = entry->procname;
1193 link = find_entry(&head, dir, procname, strlen(procname));
1196 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1198 if (S_ISLNK(link->mode) && (link->data == link_root))
1203 /* The checks passed. Increase the registration count on the links */
1204 for (entry = table; entry->procname; entry++) {
1205 const char *procname = entry->procname;
1206 link = find_entry(&head, dir, procname, strlen(procname));
1212 static int insert_links(struct ctl_table_header *head)
1214 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1215 struct ctl_dir *core_parent = NULL;
1216 struct ctl_table_header *links;
1219 if (head->set == root_set)
1222 core_parent = xlate_dir(root_set, head->parent);
1223 if (IS_ERR(core_parent))
1226 if (get_links(core_parent, head->ctl_table, head->root))
1229 core_parent->header.nreg++;
1230 spin_unlock(&sysctl_lock);
1232 links = new_links(core_parent, head->ctl_table, head->root);
1234 spin_lock(&sysctl_lock);
1240 if (get_links(core_parent, head->ctl_table, head->root)) {
1245 err = insert_header(core_parent, links);
1249 drop_sysctl_table(&core_parent->header);
1254 * __register_sysctl_table - register a leaf sysctl table
1255 * @set: Sysctl tree to register on
1256 * @path: The path to the directory the sysctl table is in.
1257 * @table: the top-level table structure
1259 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1260 * array. A completely 0 filled entry terminates the table.
1262 * The members of the &struct ctl_table structure are used as follows:
1264 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1265 * enter a sysctl file
1267 * data - a pointer to data for use by proc_handler
1269 * maxlen - the maximum size in bytes of the data
1271 * mode - the file permissions for the /proc/sys file
1273 * child - must be %NULL.
1275 * proc_handler - the text handler routine (described below)
1277 * extra1, extra2 - extra pointers usable by the proc handler routines
1279 * Leaf nodes in the sysctl tree will be represented by a single file
1280 * under /proc; non-leaf nodes will be represented by directories.
1282 * There must be a proc_handler routine for any terminal nodes.
1283 * Several default handlers are available to cover common cases -
1285 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1286 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1287 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1289 * It is the handler's job to read the input buffer from user memory
1290 * and process it. The handler should return 0 on success.
1292 * This routine returns %NULL on a failure to register, and a pointer
1293 * to the table header on success.
1295 struct ctl_table_header *__register_sysctl_table(
1296 struct ctl_table_set *set,
1297 const char *path, struct ctl_table *table)
1299 struct ctl_table_root *root = set->dir.header.root;
1300 struct ctl_table_header *header;
1301 const char *name, *nextname;
1302 struct ctl_dir *dir;
1303 struct ctl_table *entry;
1304 struct ctl_node *node;
1307 for (entry = table; entry->procname; entry++)
1310 header = kzalloc(sizeof(struct ctl_table_header) +
1311 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1315 node = (struct ctl_node *)(header + 1);
1316 init_header(header, root, set, node, table);
1317 if (sysctl_check_table(path, table))
1320 spin_lock(&sysctl_lock);
1322 /* Reference moved down the diretory tree get_subdir */
1324 spin_unlock(&sysctl_lock);
1326 /* Find the directory for the ctl_table */
1327 for (name = path; name; name = nextname) {
1329 nextname = strchr(name, '/');
1331 namelen = nextname - name;
1334 namelen = strlen(name);
1339 dir = get_subdir(dir, name, namelen);
1344 spin_lock(&sysctl_lock);
1345 if (insert_header(dir, header))
1346 goto fail_put_dir_locked;
1348 drop_sysctl_table(&dir->header);
1349 spin_unlock(&sysctl_lock);
1353 fail_put_dir_locked:
1354 drop_sysctl_table(&dir->header);
1355 spin_unlock(&sysctl_lock);
1363 * register_sysctl - register a sysctl table
1364 * @path: The path to the directory the sysctl table is in.
1365 * @table: the table structure
1367 * Register a sysctl table. @table should be a filled in ctl_table
1368 * array. A completely 0 filled entry terminates the table.
1370 * See __register_sysctl_table for more details.
1372 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1374 return __register_sysctl_table(&sysctl_table_root.default_set,
1377 EXPORT_SYMBOL(register_sysctl);
1379 static char *append_path(const char *path, char *pos, const char *name)
1382 namelen = strlen(name);
1383 if (((pos - path) + namelen + 2) >= PATH_MAX)
1385 memcpy(pos, name, namelen);
1387 pos[namelen + 1] = '\0';
1392 static int count_subheaders(struct ctl_table *table)
1395 int nr_subheaders = 0;
1396 struct ctl_table *entry;
1398 /* special case: no directory and empty directory */
1399 if (!table || !table->procname)
1402 for (entry = table; entry->procname; entry++) {
1404 nr_subheaders += count_subheaders(entry->child);
1408 return nr_subheaders + has_files;
1411 static int register_leaf_sysctl_tables(const char *path, char *pos,
1412 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1413 struct ctl_table *table)
1415 struct ctl_table *ctl_table_arg = NULL;
1416 struct ctl_table *entry, *files;
1421 for (entry = table; entry->procname; entry++) {
1429 /* If there are mixed files and directories we need a new table */
1430 if (nr_dirs && nr_files) {
1431 struct ctl_table *new;
1432 files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1437 ctl_table_arg = files;
1438 for (new = files, entry = table; entry->procname; entry++) {
1446 /* Register everything except a directory full of subdirectories */
1447 if (nr_files || !nr_dirs) {
1448 struct ctl_table_header *header;
1449 header = __register_sysctl_table(set, path, files);
1451 kfree(ctl_table_arg);
1455 /* Remember if we need to free the file table */
1456 header->ctl_table_arg = ctl_table_arg;
1457 **subheader = header;
1461 /* Recurse into the subdirectories. */
1462 for (entry = table; entry->procname; entry++) {
1468 err = -ENAMETOOLONG;
1469 child_pos = append_path(path, pos, entry->procname);
1473 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1481 /* On failure our caller will unregister all registered subheaders */
1486 * __register_sysctl_paths - register a sysctl table hierarchy
1487 * @set: Sysctl tree to register on
1488 * @path: The path to the directory the sysctl table is in.
1489 * @table: the top-level table structure
1491 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1492 * array. A completely 0 filled entry terminates the table.
1494 * See __register_sysctl_table for more details.
1496 struct ctl_table_header *__register_sysctl_paths(
1497 struct ctl_table_set *set,
1498 const struct ctl_path *path, struct ctl_table *table)
1500 struct ctl_table *ctl_table_arg = table;
1501 int nr_subheaders = count_subheaders(table);
1502 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1503 const struct ctl_path *component;
1504 char *new_path, *pos;
1506 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1511 for (component = path; component->procname; component++) {
1512 pos = append_path(new_path, pos, component->procname);
1516 while (table->procname && table->child && !table[1].procname) {
1517 pos = append_path(new_path, pos, table->procname);
1520 table = table->child;
1522 if (nr_subheaders == 1) {
1523 header = __register_sysctl_table(set, new_path, table);
1525 header->ctl_table_arg = ctl_table_arg;
1527 header = kzalloc(sizeof(*header) +
1528 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1532 subheaders = (struct ctl_table_header **) (header + 1);
1533 subheader = subheaders;
1534 header->ctl_table_arg = ctl_table_arg;
1536 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1538 goto err_register_leaves;
1545 err_register_leaves:
1546 while (subheader > subheaders) {
1547 struct ctl_table_header *subh = *(--subheader);
1548 struct ctl_table *table = subh->ctl_table_arg;
1549 unregister_sysctl_table(subh);
1558 * register_sysctl_table_path - register a sysctl table hierarchy
1559 * @path: The path to the directory the sysctl table is in.
1560 * @table: the top-level table structure
1562 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1563 * array. A completely 0 filled entry terminates the table.
1565 * See __register_sysctl_paths for more details.
1567 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1568 struct ctl_table *table)
1570 return __register_sysctl_paths(&sysctl_table_root.default_set,
1573 EXPORT_SYMBOL(register_sysctl_paths);
1576 * register_sysctl_table - register a sysctl table hierarchy
1577 * @table: the top-level table structure
1579 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1580 * array. A completely 0 filled entry terminates the table.
1582 * See register_sysctl_paths for more details.
1584 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1586 static const struct ctl_path null_path[] = { {} };
1588 return register_sysctl_paths(null_path, table);
1590 EXPORT_SYMBOL(register_sysctl_table);
1592 static void put_links(struct ctl_table_header *header)
1594 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1595 struct ctl_table_root *root = header->root;
1596 struct ctl_dir *parent = header->parent;
1597 struct ctl_dir *core_parent;
1598 struct ctl_table *entry;
1600 if (header->set == root_set)
1603 core_parent = xlate_dir(root_set, parent);
1604 if (IS_ERR(core_parent))
1607 for (entry = header->ctl_table; entry->procname; entry++) {
1608 struct ctl_table_header *link_head;
1609 struct ctl_table *link;
1610 const char *name = entry->procname;
1612 link = find_entry(&link_head, core_parent, name, strlen(name));
1614 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1615 (S_ISLNK(link->mode) && (link->data == root)))) {
1616 drop_sysctl_table(link_head);
1619 pr_err("sysctl link missing during unregister: ");
1620 sysctl_print_dir(parent);
1621 pr_cont("/%s\n", name);
1626 static void drop_sysctl_table(struct ctl_table_header *header)
1628 struct ctl_dir *parent = header->parent;
1635 start_unregistering(header);
1638 if (!--header->count)
1639 kfree_rcu(header, rcu);
1642 drop_sysctl_table(&parent->header);
1646 * unregister_sysctl_table - unregister a sysctl table hierarchy
1647 * @header: the header returned from register_sysctl_table
1649 * Unregisters the sysctl table and all children. proc entries may not
1650 * actually be removed until they are no longer used by anyone.
1652 void unregister_sysctl_table(struct ctl_table_header * header)
1660 nr_subheaders = count_subheaders(header->ctl_table_arg);
1661 if (unlikely(nr_subheaders > 1)) {
1662 struct ctl_table_header **subheaders;
1665 subheaders = (struct ctl_table_header **)(header + 1);
1666 for (i = nr_subheaders -1; i >= 0; i--) {
1667 struct ctl_table_header *subh = subheaders[i];
1668 struct ctl_table *table = subh->ctl_table_arg;
1669 unregister_sysctl_table(subh);
1676 spin_lock(&sysctl_lock);
1677 drop_sysctl_table(header);
1678 spin_unlock(&sysctl_lock);
1680 EXPORT_SYMBOL(unregister_sysctl_table);
1682 void setup_sysctl_set(struct ctl_table_set *set,
1683 struct ctl_table_root *root,
1684 int (*is_seen)(struct ctl_table_set *))
1686 memset(set, 0, sizeof(*set));
1687 set->is_seen = is_seen;
1688 init_header(&set->dir.header, root, set, NULL, root_table);
1691 void retire_sysctl_set(struct ctl_table_set *set)
1693 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1696 int __init proc_sys_init(void)
1698 struct proc_dir_entry *proc_sys_root;
1700 proc_sys_root = proc_mkdir("sys", NULL);
1701 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1702 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1703 proc_sys_root->nlink = 0;
1705 return sysctl_init();
1708 struct sysctl_alias {
1709 const char *kernel_param;
1710 const char *sysctl_param;
1714 * Historically some settings had both sysctl and a command line parameter.
1715 * With the generic sysctl. parameter support, we can handle them at a single
1716 * place and only keep the historical name for compatibility. This is not meant
1717 * to add brand new aliases. When adding existing aliases, consider whether
1718 * the possibly different moment of changing the value (e.g. from early_param
1719 * to the moment do_sysctl_args() is called) is an issue for the specific
1722 static const struct sysctl_alias sysctl_aliases[] = {
1723 {"numa_zonelist_order", "vm.numa_zonelist_order" },
1727 static const char *sysctl_find_alias(char *param)
1729 const struct sysctl_alias *alias;
1731 for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1732 if (strcmp(alias->kernel_param, param) == 0)
1733 return alias->sysctl_param;
1739 /* Set sysctl value passed on kernel command line. */
1740 static int process_sysctl_arg(char *param, char *val,
1741 const char *unused, void *arg)
1744 struct vfsmount **proc_mnt = arg;
1745 struct file_system_type *proc_fs_type;
1752 if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1753 param += sizeof("sysctl") - 1;
1755 if (param[0] != '/' && param[0] != '.')
1760 param = (char *) sysctl_find_alias(param);
1766 * To set sysctl options, we use a temporary mount of proc, look up the
1767 * respective sys/ file and write to it. To avoid mounting it when no
1768 * options were given, we mount it only when the first sysctl option is
1769 * found. Why not a persistent mount? There are problems with a
1770 * persistent mount of proc in that it forces userspace not to use any
1771 * proc mount options.
1774 proc_fs_type = get_fs_type("proc");
1775 if (!proc_fs_type) {
1776 pr_err("Failed to find procfs to set sysctl from command line\n");
1779 *proc_mnt = kern_mount(proc_fs_type);
1780 put_filesystem(proc_fs_type);
1781 if (IS_ERR(*proc_mnt)) {
1782 pr_err("Failed to mount procfs to set sysctl from command line\n");
1787 path = kasprintf(GFP_KERNEL, "sys/%s", param);
1789 panic("%s: Failed to allocate path for %s\n", __func__, param);
1790 strreplace(path, '.', '/');
1792 file = file_open_root((*proc_mnt)->mnt_root, *proc_mnt, path, O_WRONLY, 0);
1794 err = PTR_ERR(file);
1796 pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1798 else if (err == -EACCES)
1799 pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1802 pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1807 wret = kernel_write(file, val, len, &pos);
1811 pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1814 pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1815 ERR_PTR(err), param, val);
1816 } else if (wret != len) {
1817 pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1818 wret, len, path, param, val);
1821 err = filp_close(file, NULL);
1823 pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1824 ERR_PTR(err), param, val);
1830 void do_sysctl_args(void)
1833 struct vfsmount *proc_mnt = NULL;
1835 command_line = kstrdup(saved_command_line, GFP_KERNEL);
1837 panic("%s: Failed to allocate copy of command line\n", __func__);
1839 parse_args("Setting sysctl args", command_line,
1840 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1843 kern_unmount(proc_mnt);
1845 kfree(command_line);