1 // SPDX-License-Identifier: GPL-2.0
5 #include <linux/init.h>
6 #include <linux/sysctl.h>
7 #include <linux/poll.h>
8 #include <linux/proc_fs.h>
9 #include <linux/printk.h>
10 #include <linux/security.h>
11 #include <linux/sched.h>
12 #include <linux/cred.h>
13 #include <linux/namei.h>
15 #include <linux/module.h>
18 static const struct dentry_operations proc_sys_dentry_operations
;
19 static const struct file_operations proc_sys_file_operations
;
20 static const struct inode_operations proc_sys_inode_operations
;
21 static const struct file_operations proc_sys_dir_file_operations
;
22 static const struct inode_operations proc_sys_dir_operations
;
24 /* Support for permanently empty directories */
26 struct ctl_table sysctl_mount_point
[] = {
30 static bool is_empty_dir(struct ctl_table_header
*head
)
32 return head
->ctl_table
[0].child
== sysctl_mount_point
;
35 static void set_empty_dir(struct ctl_dir
*dir
)
37 dir
->header
.ctl_table
[0].child
= sysctl_mount_point
;
40 static void clear_empty_dir(struct ctl_dir
*dir
)
43 dir
->header
.ctl_table
[0].child
= NULL
;
46 void proc_sys_poll_notify(struct ctl_table_poll
*poll
)
51 atomic_inc(&poll
->event
);
52 wake_up_interruptible(&poll
->wait
);
55 static struct ctl_table root_table
[] = {
58 .mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
,
62 static struct ctl_table_root sysctl_table_root
= {
63 .default_set
.dir
.header
= {
66 .ctl_table
= root_table
}},
67 .ctl_table_arg
= root_table
,
68 .root
= &sysctl_table_root
,
69 .set
= &sysctl_table_root
.default_set
,
73 static DEFINE_SPINLOCK(sysctl_lock
);
75 static void drop_sysctl_table(struct ctl_table_header
*header
);
76 static int sysctl_follow_link(struct ctl_table_header
**phead
,
77 struct ctl_table
**pentry
);
78 static int insert_links(struct ctl_table_header
*head
);
79 static void put_links(struct ctl_table_header
*header
);
81 static void sysctl_print_dir(struct ctl_dir
*dir
)
83 if (dir
->header
.parent
)
84 sysctl_print_dir(dir
->header
.parent
);
85 pr_cont("%s/", dir
->header
.ctl_table
[0].procname
);
88 static int namecmp(const char *name1
, int len1
, const char *name2
, int len2
)
97 cmp
= memcmp(name1
, name2
, minlen
);
103 /* Called under sysctl_lock */
104 static struct ctl_table
*find_entry(struct ctl_table_header
**phead
,
105 struct ctl_dir
*dir
, const char *name
, int namelen
)
107 struct ctl_table_header
*head
;
108 struct ctl_table
*entry
;
109 struct rb_node
*node
= dir
->root
.rb_node
;
113 struct ctl_node
*ctl_node
;
114 const char *procname
;
117 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
118 head
= ctl_node
->header
;
119 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
120 procname
= entry
->procname
;
122 cmp
= namecmp(name
, namelen
, procname
, strlen(procname
));
124 node
= node
->rb_left
;
126 node
= node
->rb_right
;
135 static int insert_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
137 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
138 struct rb_node
**p
= &head
->parent
->root
.rb_node
;
139 struct rb_node
*parent
= NULL
;
140 const char *name
= entry
->procname
;
141 int namelen
= strlen(name
);
144 struct ctl_table_header
*parent_head
;
145 struct ctl_table
*parent_entry
;
146 struct ctl_node
*parent_node
;
147 const char *parent_name
;
151 parent_node
= rb_entry(parent
, struct ctl_node
, node
);
152 parent_head
= parent_node
->header
;
153 parent_entry
= &parent_head
->ctl_table
[parent_node
- parent_head
->node
];
154 parent_name
= parent_entry
->procname
;
156 cmp
= namecmp(name
, namelen
, parent_name
, strlen(parent_name
));
162 pr_err("sysctl duplicate entry: ");
163 sysctl_print_dir(head
->parent
);
164 pr_cont("/%s\n", entry
->procname
);
169 rb_link_node(node
, parent
, p
);
170 rb_insert_color(node
, &head
->parent
->root
);
174 static void erase_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
176 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
178 rb_erase(node
, &head
->parent
->root
);
181 static void init_header(struct ctl_table_header
*head
,
182 struct ctl_table_root
*root
, struct ctl_table_set
*set
,
183 struct ctl_node
*node
, struct ctl_table
*table
)
185 head
->ctl_table
= table
;
186 head
->ctl_table_arg
= table
;
190 head
->unregistering
= NULL
;
195 INIT_HLIST_HEAD(&head
->inodes
);
197 struct ctl_table
*entry
;
198 for (entry
= table
; entry
->procname
; entry
++, node
++)
203 static void erase_header(struct ctl_table_header
*head
)
205 struct ctl_table
*entry
;
206 for (entry
= head
->ctl_table
; entry
->procname
; entry
++)
207 erase_entry(head
, entry
);
210 static int insert_header(struct ctl_dir
*dir
, struct ctl_table_header
*header
)
212 struct ctl_table
*entry
;
215 /* Is this a permanently empty directory? */
216 if (is_empty_dir(&dir
->header
))
219 /* Am I creating a permanently empty directory? */
220 if (header
->ctl_table
== sysctl_mount_point
) {
221 if (!RB_EMPTY_ROOT(&dir
->root
))
227 header
->parent
= dir
;
228 err
= insert_links(header
);
231 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
232 err
= insert_entry(header
, entry
);
238 erase_header(header
);
241 if (header
->ctl_table
== sysctl_mount_point
)
242 clear_empty_dir(dir
);
243 header
->parent
= NULL
;
244 drop_sysctl_table(&dir
->header
);
248 /* called under sysctl_lock */
249 static int use_table(struct ctl_table_header
*p
)
251 if (unlikely(p
->unregistering
))
257 /* called under sysctl_lock */
258 static void unuse_table(struct ctl_table_header
*p
)
261 if (unlikely(p
->unregistering
))
262 complete(p
->unregistering
);
265 static void proc_sys_prune_dcache(struct ctl_table_header
*head
)
268 struct proc_inode
*ei
;
269 struct hlist_node
*node
;
270 struct super_block
*sb
;
274 node
= hlist_first_rcu(&head
->inodes
);
277 ei
= hlist_entry(node
, struct proc_inode
, sysctl_inodes
);
278 spin_lock(&sysctl_lock
);
279 hlist_del_init_rcu(&ei
->sysctl_inodes
);
280 spin_unlock(&sysctl_lock
);
282 inode
= &ei
->vfs_inode
;
284 if (!atomic_inc_not_zero(&sb
->s_active
))
286 inode
= igrab(inode
);
288 if (unlikely(!inode
)) {
289 deactivate_super(sb
);
294 d_prune_aliases(inode
);
296 deactivate_super(sb
);
303 /* called under sysctl_lock, will reacquire if has to wait */
304 static void start_unregistering(struct ctl_table_header
*p
)
307 * if p->used is 0, nobody will ever touch that entry again;
308 * we'll eliminate all paths to it before dropping sysctl_lock
310 if (unlikely(p
->used
)) {
311 struct completion wait
;
312 init_completion(&wait
);
313 p
->unregistering
= &wait
;
314 spin_unlock(&sysctl_lock
);
315 wait_for_completion(&wait
);
317 /* anything non-NULL; we'll never dereference it */
318 p
->unregistering
= ERR_PTR(-EINVAL
);
319 spin_unlock(&sysctl_lock
);
322 * Prune dentries for unregistered sysctls: namespaced sysctls
323 * can have duplicate names and contaminate dcache very badly.
325 proc_sys_prune_dcache(p
);
327 * do not remove from the list until nobody holds it; walking the
328 * list in do_sysctl() relies on that.
330 spin_lock(&sysctl_lock
);
334 static struct ctl_table_header
*sysctl_head_grab(struct ctl_table_header
*head
)
337 spin_lock(&sysctl_lock
);
338 if (!use_table(head
))
339 head
= ERR_PTR(-ENOENT
);
340 spin_unlock(&sysctl_lock
);
344 static void sysctl_head_finish(struct ctl_table_header
*head
)
348 spin_lock(&sysctl_lock
);
350 spin_unlock(&sysctl_lock
);
353 static struct ctl_table_set
*
354 lookup_header_set(struct ctl_table_root
*root
)
356 struct ctl_table_set
*set
= &root
->default_set
;
358 set
= root
->lookup(root
);
362 static struct ctl_table
*lookup_entry(struct ctl_table_header
**phead
,
364 const char *name
, int namelen
)
366 struct ctl_table_header
*head
;
367 struct ctl_table
*entry
;
369 spin_lock(&sysctl_lock
);
370 entry
= find_entry(&head
, dir
, name
, namelen
);
371 if (entry
&& use_table(head
))
375 spin_unlock(&sysctl_lock
);
379 static struct ctl_node
*first_usable_entry(struct rb_node
*node
)
381 struct ctl_node
*ctl_node
;
383 for (;node
; node
= rb_next(node
)) {
384 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
385 if (use_table(ctl_node
->header
))
391 static void first_entry(struct ctl_dir
*dir
,
392 struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
394 struct ctl_table_header
*head
= NULL
;
395 struct ctl_table
*entry
= NULL
;
396 struct ctl_node
*ctl_node
;
398 spin_lock(&sysctl_lock
);
399 ctl_node
= first_usable_entry(rb_first(&dir
->root
));
400 spin_unlock(&sysctl_lock
);
402 head
= ctl_node
->header
;
403 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
409 static void next_entry(struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
411 struct ctl_table_header
*head
= *phead
;
412 struct ctl_table
*entry
= *pentry
;
413 struct ctl_node
*ctl_node
= &head
->node
[entry
- head
->ctl_table
];
415 spin_lock(&sysctl_lock
);
418 ctl_node
= first_usable_entry(rb_next(&ctl_node
->node
));
419 spin_unlock(&sysctl_lock
);
422 head
= ctl_node
->header
;
423 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
430 * sysctl_perm does NOT grant the superuser all rights automatically, because
431 * some sysctl variables are readonly even to root.
434 static int test_perm(int mode
, int op
)
436 if (uid_eq(current_euid(), GLOBAL_ROOT_UID
))
438 else if (in_egroup_p(GLOBAL_ROOT_GID
))
440 if ((op
& ~mode
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
)) == 0)
445 static int sysctl_perm(struct ctl_table_header
*head
, struct ctl_table
*table
, int op
)
447 struct ctl_table_root
*root
= head
->root
;
450 if (root
->permissions
)
451 mode
= root
->permissions(head
, table
);
455 return test_perm(mode
, op
);
458 static struct inode
*proc_sys_make_inode(struct super_block
*sb
,
459 struct ctl_table_header
*head
, struct ctl_table
*table
)
461 struct ctl_table_root
*root
= head
->root
;
463 struct proc_inode
*ei
;
465 inode
= new_inode(sb
);
467 return ERR_PTR(-ENOMEM
);
469 inode
->i_ino
= get_next_ino();
473 spin_lock(&sysctl_lock
);
474 if (unlikely(head
->unregistering
)) {
475 spin_unlock(&sysctl_lock
);
477 return ERR_PTR(-ENOENT
);
480 ei
->sysctl_entry
= table
;
481 hlist_add_head_rcu(&ei
->sysctl_inodes
, &head
->inodes
);
483 spin_unlock(&sysctl_lock
);
485 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= current_time(inode
);
486 inode
->i_mode
= table
->mode
;
487 if (!S_ISDIR(table
->mode
)) {
488 inode
->i_mode
|= S_IFREG
;
489 inode
->i_op
= &proc_sys_inode_operations
;
490 inode
->i_fop
= &proc_sys_file_operations
;
492 inode
->i_mode
|= S_IFDIR
;
493 inode
->i_op
= &proc_sys_dir_operations
;
494 inode
->i_fop
= &proc_sys_dir_file_operations
;
495 if (is_empty_dir(head
))
496 make_empty_dir_inode(inode
);
499 if (root
->set_ownership
)
500 root
->set_ownership(head
, table
, &inode
->i_uid
, &inode
->i_gid
);
505 void proc_sys_evict_inode(struct inode
*inode
, struct ctl_table_header
*head
)
507 spin_lock(&sysctl_lock
);
508 hlist_del_init_rcu(&PROC_I(inode
)->sysctl_inodes
);
510 kfree_rcu(head
, rcu
);
511 spin_unlock(&sysctl_lock
);
514 static struct ctl_table_header
*grab_header(struct inode
*inode
)
516 struct ctl_table_header
*head
= PROC_I(inode
)->sysctl
;
518 head
= &sysctl_table_root
.default_set
.dir
.header
;
519 return sysctl_head_grab(head
);
522 static struct dentry
*proc_sys_lookup(struct inode
*dir
, struct dentry
*dentry
,
525 struct ctl_table_header
*head
= grab_header(dir
);
526 struct ctl_table_header
*h
= NULL
;
527 const struct qstr
*name
= &dentry
->d_name
;
530 struct dentry
*err
= ERR_PTR(-ENOENT
);
531 struct ctl_dir
*ctl_dir
;
535 return ERR_CAST(head
);
537 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
539 p
= lookup_entry(&h
, ctl_dir
, name
->name
, name
->len
);
543 if (S_ISLNK(p
->mode
)) {
544 ret
= sysctl_follow_link(&h
, &p
);
550 inode
= proc_sys_make_inode(dir
->i_sb
, h
? h
: head
, p
);
552 err
= ERR_CAST(inode
);
557 d_set_d_op(dentry
, &proc_sys_dentry_operations
);
558 d_add(dentry
, inode
);
562 sysctl_head_finish(h
);
563 sysctl_head_finish(head
);
567 static ssize_t
proc_sys_call_handler(struct file
*filp
, void __user
*buf
,
568 size_t count
, loff_t
*ppos
, int write
)
570 struct inode
*inode
= file_inode(filp
);
571 struct ctl_table_header
*head
= grab_header(inode
);
572 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
577 return PTR_ERR(head
);
580 * At this point we know that the sysctl was not unregistered
581 * and won't be until we finish.
584 if (sysctl_perm(head
, table
, write
? MAY_WRITE
: MAY_READ
))
587 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
589 if (!table
->proc_handler
)
592 /* careful: calling conventions are nasty here */
594 error
= table
->proc_handler(table
, write
, buf
, &res
, ppos
);
598 sysctl_head_finish(head
);
603 static ssize_t
proc_sys_read(struct file
*filp
, char __user
*buf
,
604 size_t count
, loff_t
*ppos
)
606 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 0);
609 static ssize_t
proc_sys_write(struct file
*filp
, const char __user
*buf
,
610 size_t count
, loff_t
*ppos
)
612 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 1);
615 static int proc_sys_open(struct inode
*inode
, struct file
*filp
)
617 struct ctl_table_header
*head
= grab_header(inode
);
618 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
620 /* sysctl was unregistered */
622 return PTR_ERR(head
);
625 filp
->private_data
= proc_sys_poll_event(table
->poll
);
627 sysctl_head_finish(head
);
632 static unsigned int proc_sys_poll(struct file
*filp
, poll_table
*wait
)
634 struct inode
*inode
= file_inode(filp
);
635 struct ctl_table_header
*head
= grab_header(inode
);
636 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
637 unsigned int ret
= DEFAULT_POLLMASK
;
640 /* sysctl was unregistered */
642 return POLLERR
| POLLHUP
;
644 if (!table
->proc_handler
)
650 event
= (unsigned long)filp
->private_data
;
651 poll_wait(filp
, &table
->poll
->wait
, wait
);
653 if (event
!= atomic_read(&table
->poll
->event
)) {
654 filp
->private_data
= proc_sys_poll_event(table
->poll
);
655 ret
= POLLIN
| POLLRDNORM
| POLLERR
| POLLPRI
;
659 sysctl_head_finish(head
);
664 static bool proc_sys_fill_cache(struct file
*file
,
665 struct dir_context
*ctx
,
666 struct ctl_table_header
*head
,
667 struct ctl_table
*table
)
669 struct dentry
*child
, *dir
= file
->f_path
.dentry
;
673 unsigned type
= DT_UNKNOWN
;
675 qname
.name
= table
->procname
;
676 qname
.len
= strlen(table
->procname
);
677 qname
.hash
= full_name_hash(dir
, qname
.name
, qname
.len
);
679 child
= d_lookup(dir
, &qname
);
681 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
682 child
= d_alloc_parallel(dir
, &qname
, &wq
);
685 if (d_in_lookup(child
)) {
686 inode
= proc_sys_make_inode(dir
->d_sb
, head
, table
);
688 d_lookup_done(child
);
692 d_set_d_op(child
, &proc_sys_dentry_operations
);
696 inode
= d_inode(child
);
698 type
= inode
->i_mode
>> 12;
700 return dir_emit(ctx
, qname
.name
, qname
.len
, ino
, type
);
703 static bool proc_sys_link_fill_cache(struct file
*file
,
704 struct dir_context
*ctx
,
705 struct ctl_table_header
*head
,
706 struct ctl_table
*table
)
710 head
= sysctl_head_grab(head
);
714 if (S_ISLNK(table
->mode
)) {
715 /* It is not an error if we can not follow the link ignore it */
716 int err
= sysctl_follow_link(&head
, &table
);
721 ret
= proc_sys_fill_cache(file
, ctx
, head
, table
);
723 sysctl_head_finish(head
);
727 static int scan(struct ctl_table_header
*head
, struct ctl_table
*table
,
728 unsigned long *pos
, struct file
*file
,
729 struct dir_context
*ctx
)
733 if ((*pos
)++ < ctx
->pos
)
736 if (unlikely(S_ISLNK(table
->mode
)))
737 res
= proc_sys_link_fill_cache(file
, ctx
, head
, table
);
739 res
= proc_sys_fill_cache(file
, ctx
, head
, table
);
747 static int proc_sys_readdir(struct file
*file
, struct dir_context
*ctx
)
749 struct ctl_table_header
*head
= grab_header(file_inode(file
));
750 struct ctl_table_header
*h
= NULL
;
751 struct ctl_table
*entry
;
752 struct ctl_dir
*ctl_dir
;
756 return PTR_ERR(head
);
758 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
760 if (!dir_emit_dots(file
, ctx
))
765 for (first_entry(ctl_dir
, &h
, &entry
); h
; next_entry(&h
, &entry
)) {
766 if (!scan(h
, entry
, &pos
, file
, ctx
)) {
767 sysctl_head_finish(h
);
772 sysctl_head_finish(head
);
776 static int proc_sys_permission(struct inode
*inode
, int mask
)
779 * sysctl entries that are not writeable,
780 * are _NOT_ writeable, capabilities or not.
782 struct ctl_table_header
*head
;
783 struct ctl_table
*table
;
786 /* Executable files are not allowed under /proc/sys/ */
787 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
))
790 head
= grab_header(inode
);
792 return PTR_ERR(head
);
794 table
= PROC_I(inode
)->sysctl_entry
;
795 if (!table
) /* global root - r-xr-xr-x */
796 error
= mask
& MAY_WRITE
? -EACCES
: 0;
797 else /* Use the permissions on the sysctl table entry */
798 error
= sysctl_perm(head
, table
, mask
& ~MAY_NOT_BLOCK
);
800 sysctl_head_finish(head
);
804 static int proc_sys_setattr(struct dentry
*dentry
, struct iattr
*attr
)
806 struct inode
*inode
= d_inode(dentry
);
809 if (attr
->ia_valid
& (ATTR_MODE
| ATTR_UID
| ATTR_GID
))
812 error
= setattr_prepare(dentry
, attr
);
816 setattr_copy(inode
, attr
);
817 mark_inode_dirty(inode
);
821 static int proc_sys_getattr(const struct path
*path
, struct kstat
*stat
,
822 u32 request_mask
, unsigned int query_flags
)
824 struct inode
*inode
= d_inode(path
->dentry
);
825 struct ctl_table_header
*head
= grab_header(inode
);
826 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
829 return PTR_ERR(head
);
831 generic_fillattr(inode
, stat
);
833 stat
->mode
= (stat
->mode
& S_IFMT
) | table
->mode
;
835 sysctl_head_finish(head
);
839 static const struct file_operations proc_sys_file_operations
= {
840 .open
= proc_sys_open
,
841 .poll
= proc_sys_poll
,
842 .read
= proc_sys_read
,
843 .write
= proc_sys_write
,
844 .llseek
= default_llseek
,
847 static const struct file_operations proc_sys_dir_file_operations
= {
848 .read
= generic_read_dir
,
849 .iterate_shared
= proc_sys_readdir
,
850 .llseek
= generic_file_llseek
,
853 static const struct inode_operations proc_sys_inode_operations
= {
854 .permission
= proc_sys_permission
,
855 .setattr
= proc_sys_setattr
,
856 .getattr
= proc_sys_getattr
,
859 static const struct inode_operations proc_sys_dir_operations
= {
860 .lookup
= proc_sys_lookup
,
861 .permission
= proc_sys_permission
,
862 .setattr
= proc_sys_setattr
,
863 .getattr
= proc_sys_getattr
,
866 static int proc_sys_revalidate(struct dentry
*dentry
, unsigned int flags
)
868 if (flags
& LOOKUP_RCU
)
870 return !PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
873 static int proc_sys_delete(const struct dentry
*dentry
)
875 return !!PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
878 static int sysctl_is_seen(struct ctl_table_header
*p
)
880 struct ctl_table_set
*set
= p
->set
;
882 spin_lock(&sysctl_lock
);
883 if (p
->unregistering
)
885 else if (!set
->is_seen
)
888 res
= set
->is_seen(set
);
889 spin_unlock(&sysctl_lock
);
893 static int proc_sys_compare(const struct dentry
*dentry
,
894 unsigned int len
, const char *str
, const struct qstr
*name
)
896 struct ctl_table_header
*head
;
899 /* Although proc doesn't have negative dentries, rcu-walk means
900 * that inode here can be NULL */
901 /* AV: can it, indeed? */
902 inode
= d_inode_rcu(dentry
);
905 if (name
->len
!= len
)
907 if (memcmp(name
->name
, str
, len
))
909 head
= rcu_dereference(PROC_I(inode
)->sysctl
);
910 return !head
|| !sysctl_is_seen(head
);
913 static const struct dentry_operations proc_sys_dentry_operations
= {
914 .d_revalidate
= proc_sys_revalidate
,
915 .d_delete
= proc_sys_delete
,
916 .d_compare
= proc_sys_compare
,
919 static struct ctl_dir
*find_subdir(struct ctl_dir
*dir
,
920 const char *name
, int namelen
)
922 struct ctl_table_header
*head
;
923 struct ctl_table
*entry
;
925 entry
= find_entry(&head
, dir
, name
, namelen
);
927 return ERR_PTR(-ENOENT
);
928 if (!S_ISDIR(entry
->mode
))
929 return ERR_PTR(-ENOTDIR
);
930 return container_of(head
, struct ctl_dir
, header
);
933 static struct ctl_dir
*new_dir(struct ctl_table_set
*set
,
934 const char *name
, int namelen
)
936 struct ctl_table
*table
;
938 struct ctl_node
*node
;
941 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node
) +
942 sizeof(struct ctl_table
)*2 + namelen
+ 1,
947 node
= (struct ctl_node
*)(new + 1);
948 table
= (struct ctl_table
*)(node
+ 1);
949 new_name
= (char *)(table
+ 2);
950 memcpy(new_name
, name
, namelen
);
951 new_name
[namelen
] = '\0';
952 table
[0].procname
= new_name
;
953 table
[0].mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
;
954 init_header(&new->header
, set
->dir
.header
.root
, set
, node
, table
);
960 * get_subdir - find or create a subdir with the specified name.
961 * @dir: Directory to create the subdirectory in
962 * @name: The name of the subdirectory to find or create
963 * @namelen: The length of name
965 * Takes a directory with an elevated reference count so we know that
966 * if we drop the lock the directory will not go away. Upon success
967 * the reference is moved from @dir to the returned subdirectory.
968 * Upon error an error code is returned and the reference on @dir is
971 static struct ctl_dir
*get_subdir(struct ctl_dir
*dir
,
972 const char *name
, int namelen
)
974 struct ctl_table_set
*set
= dir
->header
.set
;
975 struct ctl_dir
*subdir
, *new = NULL
;
978 spin_lock(&sysctl_lock
);
979 subdir
= find_subdir(dir
, name
, namelen
);
982 if (PTR_ERR(subdir
) != -ENOENT
)
985 spin_unlock(&sysctl_lock
);
986 new = new_dir(set
, name
, namelen
);
987 spin_lock(&sysctl_lock
);
988 subdir
= ERR_PTR(-ENOMEM
);
992 /* Was the subdir added while we dropped the lock? */
993 subdir
= find_subdir(dir
, name
, namelen
);
996 if (PTR_ERR(subdir
) != -ENOENT
)
999 /* Nope. Use the our freshly made directory entry. */
1000 err
= insert_header(dir
, &new->header
);
1001 subdir
= ERR_PTR(err
);
1006 subdir
->header
.nreg
++;
1008 if (IS_ERR(subdir
)) {
1009 pr_err("sysctl could not get directory: ");
1010 sysctl_print_dir(dir
);
1011 pr_cont("/%*.*s %ld\n",
1012 namelen
, namelen
, name
, PTR_ERR(subdir
));
1014 drop_sysctl_table(&dir
->header
);
1016 drop_sysctl_table(&new->header
);
1017 spin_unlock(&sysctl_lock
);
1021 static struct ctl_dir
*xlate_dir(struct ctl_table_set
*set
, struct ctl_dir
*dir
)
1023 struct ctl_dir
*parent
;
1024 const char *procname
;
1025 if (!dir
->header
.parent
)
1027 parent
= xlate_dir(set
, dir
->header
.parent
);
1030 procname
= dir
->header
.ctl_table
[0].procname
;
1031 return find_subdir(parent
, procname
, strlen(procname
));
1034 static int sysctl_follow_link(struct ctl_table_header
**phead
,
1035 struct ctl_table
**pentry
)
1037 struct ctl_table_header
*head
;
1038 struct ctl_table_root
*root
;
1039 struct ctl_table_set
*set
;
1040 struct ctl_table
*entry
;
1041 struct ctl_dir
*dir
;
1045 spin_lock(&sysctl_lock
);
1046 root
= (*pentry
)->data
;
1047 set
= lookup_header_set(root
);
1048 dir
= xlate_dir(set
, (*phead
)->parent
);
1052 const char *procname
= (*pentry
)->procname
;
1054 entry
= find_entry(&head
, dir
, procname
, strlen(procname
));
1056 if (entry
&& use_table(head
)) {
1057 unuse_table(*phead
);
1064 spin_unlock(&sysctl_lock
);
1068 static int sysctl_err(const char *path
, struct ctl_table
*table
, char *fmt
, ...)
1070 struct va_format vaf
;
1073 va_start(args
, fmt
);
1077 pr_err("sysctl table check failed: %s/%s %pV\n",
1078 path
, table
->procname
, &vaf
);
1084 static int sysctl_check_table_array(const char *path
, struct ctl_table
*table
)
1088 if ((table
->proc_handler
== proc_douintvec
) ||
1089 (table
->proc_handler
== proc_douintvec_minmax
)) {
1090 if (table
->maxlen
!= sizeof(unsigned int))
1091 err
|= sysctl_err(path
, table
, "array now allowed");
1097 static int sysctl_check_table(const char *path
, struct ctl_table
*table
)
1100 for (; table
->procname
; table
++) {
1102 err
|= sysctl_err(path
, table
, "Not a file");
1104 if ((table
->proc_handler
== proc_dostring
) ||
1105 (table
->proc_handler
== proc_dointvec
) ||
1106 (table
->proc_handler
== proc_douintvec
) ||
1107 (table
->proc_handler
== proc_douintvec_minmax
) ||
1108 (table
->proc_handler
== proc_dointvec_minmax
) ||
1109 (table
->proc_handler
== proc_dointvec_jiffies
) ||
1110 (table
->proc_handler
== proc_dointvec_userhz_jiffies
) ||
1111 (table
->proc_handler
== proc_dointvec_ms_jiffies
) ||
1112 (table
->proc_handler
== proc_doulongvec_minmax
) ||
1113 (table
->proc_handler
== proc_doulongvec_ms_jiffies_minmax
)) {
1115 err
|= sysctl_err(path
, table
, "No data");
1117 err
|= sysctl_err(path
, table
, "No maxlen");
1119 err
|= sysctl_check_table_array(path
, table
);
1121 if (!table
->proc_handler
)
1122 err
|= sysctl_err(path
, table
, "No proc_handler");
1124 if ((table
->mode
& (S_IRUGO
|S_IWUGO
)) != table
->mode
)
1125 err
|= sysctl_err(path
, table
, "bogus .mode 0%o",
1131 static struct ctl_table_header
*new_links(struct ctl_dir
*dir
, struct ctl_table
*table
,
1132 struct ctl_table_root
*link_root
)
1134 struct ctl_table
*link_table
, *entry
, *link
;
1135 struct ctl_table_header
*links
;
1136 struct ctl_node
*node
;
1138 int nr_entries
, name_bytes
;
1142 for (entry
= table
; entry
->procname
; entry
++) {
1144 name_bytes
+= strlen(entry
->procname
) + 1;
1147 links
= kzalloc(sizeof(struct ctl_table_header
) +
1148 sizeof(struct ctl_node
)*nr_entries
+
1149 sizeof(struct ctl_table
)*(nr_entries
+ 1) +
1156 node
= (struct ctl_node
*)(links
+ 1);
1157 link_table
= (struct ctl_table
*)(node
+ nr_entries
);
1158 link_name
= (char *)&link_table
[nr_entries
+ 1];
1160 for (link
= link_table
, entry
= table
; entry
->procname
; link
++, entry
++) {
1161 int len
= strlen(entry
->procname
) + 1;
1162 memcpy(link_name
, entry
->procname
, len
);
1163 link
->procname
= link_name
;
1164 link
->mode
= S_IFLNK
|S_IRWXUGO
;
1165 link
->data
= link_root
;
1168 init_header(links
, dir
->header
.root
, dir
->header
.set
, node
, link_table
);
1169 links
->nreg
= nr_entries
;
1174 static bool get_links(struct ctl_dir
*dir
,
1175 struct ctl_table
*table
, struct ctl_table_root
*link_root
)
1177 struct ctl_table_header
*head
;
1178 struct ctl_table
*entry
, *link
;
1180 /* Are there links available for every entry in table? */
1181 for (entry
= table
; entry
->procname
; entry
++) {
1182 const char *procname
= entry
->procname
;
1183 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1186 if (S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
))
1188 if (S_ISLNK(link
->mode
) && (link
->data
== link_root
))
1193 /* The checks passed. Increase the registration count on the links */
1194 for (entry
= table
; entry
->procname
; entry
++) {
1195 const char *procname
= entry
->procname
;
1196 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1202 static int insert_links(struct ctl_table_header
*head
)
1204 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1205 struct ctl_dir
*core_parent
= NULL
;
1206 struct ctl_table_header
*links
;
1209 if (head
->set
== root_set
)
1212 core_parent
= xlate_dir(root_set
, head
->parent
);
1213 if (IS_ERR(core_parent
))
1216 if (get_links(core_parent
, head
->ctl_table
, head
->root
))
1219 core_parent
->header
.nreg
++;
1220 spin_unlock(&sysctl_lock
);
1222 links
= new_links(core_parent
, head
->ctl_table
, head
->root
);
1224 spin_lock(&sysctl_lock
);
1230 if (get_links(core_parent
, head
->ctl_table
, head
->root
)) {
1235 err
= insert_header(core_parent
, links
);
1239 drop_sysctl_table(&core_parent
->header
);
1244 * __register_sysctl_table - register a leaf sysctl table
1245 * @set: Sysctl tree to register on
1246 * @path: The path to the directory the sysctl table is in.
1247 * @table: the top-level table structure
1249 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1250 * array. A completely 0 filled entry terminates the table.
1252 * The members of the &struct ctl_table structure are used as follows:
1254 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1255 * enter a sysctl file
1257 * data - a pointer to data for use by proc_handler
1259 * maxlen - the maximum size in bytes of the data
1261 * mode - the file permissions for the /proc/sys file
1263 * child - must be %NULL.
1265 * proc_handler - the text handler routine (described below)
1267 * extra1, extra2 - extra pointers usable by the proc handler routines
1269 * Leaf nodes in the sysctl tree will be represented by a single file
1270 * under /proc; non-leaf nodes will be represented by directories.
1272 * There must be a proc_handler routine for any terminal nodes.
1273 * Several default handlers are available to cover common cases -
1275 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1276 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1277 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1279 * It is the handler's job to read the input buffer from user memory
1280 * and process it. The handler should return 0 on success.
1282 * This routine returns %NULL on a failure to register, and a pointer
1283 * to the table header on success.
1285 struct ctl_table_header
*__register_sysctl_table(
1286 struct ctl_table_set
*set
,
1287 const char *path
, struct ctl_table
*table
)
1289 struct ctl_table_root
*root
= set
->dir
.header
.root
;
1290 struct ctl_table_header
*header
;
1291 const char *name
, *nextname
;
1292 struct ctl_dir
*dir
;
1293 struct ctl_table
*entry
;
1294 struct ctl_node
*node
;
1297 for (entry
= table
; entry
->procname
; entry
++)
1300 header
= kzalloc(sizeof(struct ctl_table_header
) +
1301 sizeof(struct ctl_node
)*nr_entries
, GFP_KERNEL
);
1305 node
= (struct ctl_node
*)(header
+ 1);
1306 init_header(header
, root
, set
, node
, table
);
1307 if (sysctl_check_table(path
, table
))
1310 spin_lock(&sysctl_lock
);
1312 /* Reference moved down the diretory tree get_subdir */
1314 spin_unlock(&sysctl_lock
);
1316 /* Find the directory for the ctl_table */
1317 for (name
= path
; name
; name
= nextname
) {
1319 nextname
= strchr(name
, '/');
1321 namelen
= nextname
- name
;
1324 namelen
= strlen(name
);
1329 dir
= get_subdir(dir
, name
, namelen
);
1334 spin_lock(&sysctl_lock
);
1335 if (insert_header(dir
, header
))
1336 goto fail_put_dir_locked
;
1338 drop_sysctl_table(&dir
->header
);
1339 spin_unlock(&sysctl_lock
);
1343 fail_put_dir_locked
:
1344 drop_sysctl_table(&dir
->header
);
1345 spin_unlock(&sysctl_lock
);
1353 * register_sysctl - register a sysctl table
1354 * @path: The path to the directory the sysctl table is in.
1355 * @table: the table structure
1357 * Register a sysctl table. @table should be a filled in ctl_table
1358 * array. A completely 0 filled entry terminates the table.
1360 * See __register_sysctl_table for more details.
1362 struct ctl_table_header
*register_sysctl(const char *path
, struct ctl_table
*table
)
1364 return __register_sysctl_table(&sysctl_table_root
.default_set
,
1367 EXPORT_SYMBOL(register_sysctl
);
1369 static char *append_path(const char *path
, char *pos
, const char *name
)
1372 namelen
= strlen(name
);
1373 if (((pos
- path
) + namelen
+ 2) >= PATH_MAX
)
1375 memcpy(pos
, name
, namelen
);
1377 pos
[namelen
+ 1] = '\0';
1382 static int count_subheaders(struct ctl_table
*table
)
1385 int nr_subheaders
= 0;
1386 struct ctl_table
*entry
;
1388 /* special case: no directory and empty directory */
1389 if (!table
|| !table
->procname
)
1392 for (entry
= table
; entry
->procname
; entry
++) {
1394 nr_subheaders
+= count_subheaders(entry
->child
);
1398 return nr_subheaders
+ has_files
;
1401 static int register_leaf_sysctl_tables(const char *path
, char *pos
,
1402 struct ctl_table_header
***subheader
, struct ctl_table_set
*set
,
1403 struct ctl_table
*table
)
1405 struct ctl_table
*ctl_table_arg
= NULL
;
1406 struct ctl_table
*entry
, *files
;
1411 for (entry
= table
; entry
->procname
; entry
++) {
1419 /* If there are mixed files and directories we need a new table */
1420 if (nr_dirs
&& nr_files
) {
1421 struct ctl_table
*new;
1422 files
= kzalloc(sizeof(struct ctl_table
) * (nr_files
+ 1),
1427 ctl_table_arg
= files
;
1428 for (new = files
, entry
= table
; entry
->procname
; entry
++) {
1436 /* Register everything except a directory full of subdirectories */
1437 if (nr_files
|| !nr_dirs
) {
1438 struct ctl_table_header
*header
;
1439 header
= __register_sysctl_table(set
, path
, files
);
1441 kfree(ctl_table_arg
);
1445 /* Remember if we need to free the file table */
1446 header
->ctl_table_arg
= ctl_table_arg
;
1447 **subheader
= header
;
1451 /* Recurse into the subdirectories. */
1452 for (entry
= table
; entry
->procname
; entry
++) {
1458 err
= -ENAMETOOLONG
;
1459 child_pos
= append_path(path
, pos
, entry
->procname
);
1463 err
= register_leaf_sysctl_tables(path
, child_pos
, subheader
,
1471 /* On failure our caller will unregister all registered subheaders */
1476 * __register_sysctl_paths - register a sysctl table hierarchy
1477 * @set: Sysctl tree to register on
1478 * @path: The path to the directory the sysctl table is in.
1479 * @table: the top-level table structure
1481 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1482 * array. A completely 0 filled entry terminates the table.
1484 * See __register_sysctl_table for more details.
1486 struct ctl_table_header
*__register_sysctl_paths(
1487 struct ctl_table_set
*set
,
1488 const struct ctl_path
*path
, struct ctl_table
*table
)
1490 struct ctl_table
*ctl_table_arg
= table
;
1491 int nr_subheaders
= count_subheaders(table
);
1492 struct ctl_table_header
*header
= NULL
, **subheaders
, **subheader
;
1493 const struct ctl_path
*component
;
1494 char *new_path
, *pos
;
1496 pos
= new_path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1501 for (component
= path
; component
->procname
; component
++) {
1502 pos
= append_path(new_path
, pos
, component
->procname
);
1506 while (table
->procname
&& table
->child
&& !table
[1].procname
) {
1507 pos
= append_path(new_path
, pos
, table
->procname
);
1510 table
= table
->child
;
1512 if (nr_subheaders
== 1) {
1513 header
= __register_sysctl_table(set
, new_path
, table
);
1515 header
->ctl_table_arg
= ctl_table_arg
;
1517 header
= kzalloc(sizeof(*header
) +
1518 sizeof(*subheaders
)*nr_subheaders
, GFP_KERNEL
);
1522 subheaders
= (struct ctl_table_header
**) (header
+ 1);
1523 subheader
= subheaders
;
1524 header
->ctl_table_arg
= ctl_table_arg
;
1526 if (register_leaf_sysctl_tables(new_path
, pos
, &subheader
,
1528 goto err_register_leaves
;
1535 err_register_leaves
:
1536 while (subheader
> subheaders
) {
1537 struct ctl_table_header
*subh
= *(--subheader
);
1538 struct ctl_table
*table
= subh
->ctl_table_arg
;
1539 unregister_sysctl_table(subh
);
1548 * register_sysctl_table_path - register a sysctl table hierarchy
1549 * @path: The path to the directory the sysctl table is in.
1550 * @table: the top-level table structure
1552 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1553 * array. A completely 0 filled entry terminates the table.
1555 * See __register_sysctl_paths for more details.
1557 struct ctl_table_header
*register_sysctl_paths(const struct ctl_path
*path
,
1558 struct ctl_table
*table
)
1560 return __register_sysctl_paths(&sysctl_table_root
.default_set
,
1563 EXPORT_SYMBOL(register_sysctl_paths
);
1566 * register_sysctl_table - register a sysctl table hierarchy
1567 * @table: the top-level table structure
1569 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1570 * array. A completely 0 filled entry terminates the table.
1572 * See register_sysctl_paths for more details.
1574 struct ctl_table_header
*register_sysctl_table(struct ctl_table
*table
)
1576 static const struct ctl_path null_path
[] = { {} };
1578 return register_sysctl_paths(null_path
, table
);
1580 EXPORT_SYMBOL(register_sysctl_table
);
1582 static void put_links(struct ctl_table_header
*header
)
1584 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1585 struct ctl_table_root
*root
= header
->root
;
1586 struct ctl_dir
*parent
= header
->parent
;
1587 struct ctl_dir
*core_parent
;
1588 struct ctl_table
*entry
;
1590 if (header
->set
== root_set
)
1593 core_parent
= xlate_dir(root_set
, parent
);
1594 if (IS_ERR(core_parent
))
1597 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
1598 struct ctl_table_header
*link_head
;
1599 struct ctl_table
*link
;
1600 const char *name
= entry
->procname
;
1602 link
= find_entry(&link_head
, core_parent
, name
, strlen(name
));
1604 ((S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
)) ||
1605 (S_ISLNK(link
->mode
) && (link
->data
== root
)))) {
1606 drop_sysctl_table(link_head
);
1609 pr_err("sysctl link missing during unregister: ");
1610 sysctl_print_dir(parent
);
1611 pr_cont("/%s\n", name
);
1616 static void drop_sysctl_table(struct ctl_table_header
*header
)
1618 struct ctl_dir
*parent
= header
->parent
;
1625 start_unregistering(header
);
1626 if (!--header
->count
)
1627 kfree_rcu(header
, rcu
);
1630 drop_sysctl_table(&parent
->header
);
1634 * unregister_sysctl_table - unregister a sysctl table hierarchy
1635 * @header: the header returned from register_sysctl_table
1637 * Unregisters the sysctl table and all children. proc entries may not
1638 * actually be removed until they are no longer used by anyone.
1640 void unregister_sysctl_table(struct ctl_table_header
* header
)
1648 nr_subheaders
= count_subheaders(header
->ctl_table_arg
);
1649 if (unlikely(nr_subheaders
> 1)) {
1650 struct ctl_table_header
**subheaders
;
1653 subheaders
= (struct ctl_table_header
**)(header
+ 1);
1654 for (i
= nr_subheaders
-1; i
>= 0; i
--) {
1655 struct ctl_table_header
*subh
= subheaders
[i
];
1656 struct ctl_table
*table
= subh
->ctl_table_arg
;
1657 unregister_sysctl_table(subh
);
1664 spin_lock(&sysctl_lock
);
1665 drop_sysctl_table(header
);
1666 spin_unlock(&sysctl_lock
);
1668 EXPORT_SYMBOL(unregister_sysctl_table
);
1670 void setup_sysctl_set(struct ctl_table_set
*set
,
1671 struct ctl_table_root
*root
,
1672 int (*is_seen
)(struct ctl_table_set
*))
1674 memset(set
, 0, sizeof(*set
));
1675 set
->is_seen
= is_seen
;
1676 init_header(&set
->dir
.header
, root
, set
, NULL
, root_table
);
1679 void retire_sysctl_set(struct ctl_table_set
*set
)
1681 WARN_ON(!RB_EMPTY_ROOT(&set
->dir
.root
));
1684 int __init
proc_sys_init(void)
1686 struct proc_dir_entry
*proc_sys_root
;
1688 proc_sys_root
= proc_mkdir("sys", NULL
);
1689 proc_sys_root
->proc_iops
= &proc_sys_dir_operations
;
1690 proc_sys_root
->proc_fops
= &proc_sys_dir_file_operations
;
1691 proc_sys_root
->nlink
= 0;
1693 return sysctl_init();