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
);
469 inode
->i_ino
= get_next_ino();
473 spin_lock(&sysctl_lock
);
474 if (unlikely(head
->unregistering
)) {
475 spin_unlock(&sysctl_lock
);
481 ei
->sysctl_entry
= table
;
482 hlist_add_head_rcu(&ei
->sysctl_inodes
, &head
->inodes
);
484 spin_unlock(&sysctl_lock
);
486 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= current_time(inode
);
487 inode
->i_mode
= table
->mode
;
488 if (!S_ISDIR(table
->mode
)) {
489 inode
->i_mode
|= S_IFREG
;
490 inode
->i_op
= &proc_sys_inode_operations
;
491 inode
->i_fop
= &proc_sys_file_operations
;
493 inode
->i_mode
|= S_IFDIR
;
494 inode
->i_op
= &proc_sys_dir_operations
;
495 inode
->i_fop
= &proc_sys_dir_file_operations
;
496 if (is_empty_dir(head
))
497 make_empty_dir_inode(inode
);
500 if (root
->set_ownership
)
501 root
->set_ownership(head
, table
, &inode
->i_uid
, &inode
->i_gid
);
507 void proc_sys_evict_inode(struct inode
*inode
, struct ctl_table_header
*head
)
509 spin_lock(&sysctl_lock
);
510 hlist_del_init_rcu(&PROC_I(inode
)->sysctl_inodes
);
512 kfree_rcu(head
, rcu
);
513 spin_unlock(&sysctl_lock
);
516 static struct ctl_table_header
*grab_header(struct inode
*inode
)
518 struct ctl_table_header
*head
= PROC_I(inode
)->sysctl
;
520 head
= &sysctl_table_root
.default_set
.dir
.header
;
521 return sysctl_head_grab(head
);
524 static struct dentry
*proc_sys_lookup(struct inode
*dir
, struct dentry
*dentry
,
527 struct ctl_table_header
*head
= grab_header(dir
);
528 struct ctl_table_header
*h
= NULL
;
529 const struct qstr
*name
= &dentry
->d_name
;
532 struct dentry
*err
= ERR_PTR(-ENOENT
);
533 struct ctl_dir
*ctl_dir
;
537 return ERR_CAST(head
);
539 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
541 p
= lookup_entry(&h
, ctl_dir
, name
->name
, name
->len
);
545 if (S_ISLNK(p
->mode
)) {
546 ret
= sysctl_follow_link(&h
, &p
);
552 err
= ERR_PTR(-ENOMEM
);
553 inode
= proc_sys_make_inode(dir
->i_sb
, h
? h
: head
, p
);
558 d_set_d_op(dentry
, &proc_sys_dentry_operations
);
559 d_add(dentry
, inode
);
563 sysctl_head_finish(h
);
564 sysctl_head_finish(head
);
568 static ssize_t
proc_sys_call_handler(struct file
*filp
, void __user
*buf
,
569 size_t count
, loff_t
*ppos
, int write
)
571 struct inode
*inode
= file_inode(filp
);
572 struct ctl_table_header
*head
= grab_header(inode
);
573 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
578 return PTR_ERR(head
);
581 * At this point we know that the sysctl was not unregistered
582 * and won't be until we finish.
585 if (sysctl_perm(head
, table
, write
? MAY_WRITE
: MAY_READ
))
588 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
590 if (!table
->proc_handler
)
593 /* careful: calling conventions are nasty here */
595 error
= table
->proc_handler(table
, write
, buf
, &res
, ppos
);
599 sysctl_head_finish(head
);
604 static ssize_t
proc_sys_read(struct file
*filp
, char __user
*buf
,
605 size_t count
, loff_t
*ppos
)
607 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 0);
610 static ssize_t
proc_sys_write(struct file
*filp
, const char __user
*buf
,
611 size_t count
, loff_t
*ppos
)
613 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 1);
616 static int proc_sys_open(struct inode
*inode
, struct file
*filp
)
618 struct ctl_table_header
*head
= grab_header(inode
);
619 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
621 /* sysctl was unregistered */
623 return PTR_ERR(head
);
626 filp
->private_data
= proc_sys_poll_event(table
->poll
);
628 sysctl_head_finish(head
);
633 static unsigned int proc_sys_poll(struct file
*filp
, poll_table
*wait
)
635 struct inode
*inode
= file_inode(filp
);
636 struct ctl_table_header
*head
= grab_header(inode
);
637 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
638 unsigned int ret
= DEFAULT_POLLMASK
;
641 /* sysctl was unregistered */
643 return POLLERR
| POLLHUP
;
645 if (!table
->proc_handler
)
651 event
= (unsigned long)filp
->private_data
;
652 poll_wait(filp
, &table
->poll
->wait
, wait
);
654 if (event
!= atomic_read(&table
->poll
->event
)) {
655 filp
->private_data
= proc_sys_poll_event(table
->poll
);
656 ret
= POLLIN
| POLLRDNORM
| POLLERR
| POLLPRI
;
660 sysctl_head_finish(head
);
665 static bool proc_sys_fill_cache(struct file
*file
,
666 struct dir_context
*ctx
,
667 struct ctl_table_header
*head
,
668 struct ctl_table
*table
)
670 struct dentry
*child
, *dir
= file
->f_path
.dentry
;
674 unsigned type
= DT_UNKNOWN
;
676 qname
.name
= table
->procname
;
677 qname
.len
= strlen(table
->procname
);
678 qname
.hash
= full_name_hash(dir
, qname
.name
, qname
.len
);
680 child
= d_lookup(dir
, &qname
);
682 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
683 child
= d_alloc_parallel(dir
, &qname
, &wq
);
686 if (d_in_lookup(child
)) {
687 inode
= proc_sys_make_inode(dir
->d_sb
, head
, table
);
689 d_lookup_done(child
);
693 d_set_d_op(child
, &proc_sys_dentry_operations
);
697 inode
= d_inode(child
);
699 type
= inode
->i_mode
>> 12;
701 return dir_emit(ctx
, qname
.name
, qname
.len
, ino
, type
);
704 static bool proc_sys_link_fill_cache(struct file
*file
,
705 struct dir_context
*ctx
,
706 struct ctl_table_header
*head
,
707 struct ctl_table
*table
)
711 head
= sysctl_head_grab(head
);
715 if (S_ISLNK(table
->mode
)) {
716 /* It is not an error if we can not follow the link ignore it */
717 int err
= sysctl_follow_link(&head
, &table
);
722 ret
= proc_sys_fill_cache(file
, ctx
, head
, table
);
724 sysctl_head_finish(head
);
728 static int scan(struct ctl_table_header
*head
, struct ctl_table
*table
,
729 unsigned long *pos
, struct file
*file
,
730 struct dir_context
*ctx
)
734 if ((*pos
)++ < ctx
->pos
)
737 if (unlikely(S_ISLNK(table
->mode
)))
738 res
= proc_sys_link_fill_cache(file
, ctx
, head
, table
);
740 res
= proc_sys_fill_cache(file
, ctx
, head
, table
);
748 static int proc_sys_readdir(struct file
*file
, struct dir_context
*ctx
)
750 struct ctl_table_header
*head
= grab_header(file_inode(file
));
751 struct ctl_table_header
*h
= NULL
;
752 struct ctl_table
*entry
;
753 struct ctl_dir
*ctl_dir
;
757 return PTR_ERR(head
);
759 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
761 if (!dir_emit_dots(file
, ctx
))
766 for (first_entry(ctl_dir
, &h
, &entry
); h
; next_entry(&h
, &entry
)) {
767 if (!scan(h
, entry
, &pos
, file
, ctx
)) {
768 sysctl_head_finish(h
);
773 sysctl_head_finish(head
);
777 static int proc_sys_permission(struct inode
*inode
, int mask
)
780 * sysctl entries that are not writeable,
781 * are _NOT_ writeable, capabilities or not.
783 struct ctl_table_header
*head
;
784 struct ctl_table
*table
;
787 /* Executable files are not allowed under /proc/sys/ */
788 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
))
791 head
= grab_header(inode
);
793 return PTR_ERR(head
);
795 table
= PROC_I(inode
)->sysctl_entry
;
796 if (!table
) /* global root - r-xr-xr-x */
797 error
= mask
& MAY_WRITE
? -EACCES
: 0;
798 else /* Use the permissions on the sysctl table entry */
799 error
= sysctl_perm(head
, table
, mask
& ~MAY_NOT_BLOCK
);
801 sysctl_head_finish(head
);
805 static int proc_sys_setattr(struct dentry
*dentry
, struct iattr
*attr
)
807 struct inode
*inode
= d_inode(dentry
);
808 struct user_namespace
*s_user_ns
;
811 if (attr
->ia_valid
& (ATTR_MODE
| ATTR_UID
| ATTR_GID
))
814 /* Don't let anyone mess with weird proc files */
815 s_user_ns
= inode
->i_sb
->s_user_ns
;
816 if (!kuid_has_mapping(s_user_ns
, inode
->i_uid
) ||
817 !kgid_has_mapping(s_user_ns
, inode
->i_gid
))
820 error
= setattr_prepare(dentry
, attr
);
824 setattr_copy(inode
, attr
);
825 mark_inode_dirty(inode
);
829 static int proc_sys_getattr(const struct path
*path
, struct kstat
*stat
,
830 u32 request_mask
, unsigned int query_flags
)
832 struct inode
*inode
= d_inode(path
->dentry
);
833 struct ctl_table_header
*head
= grab_header(inode
);
834 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
837 return PTR_ERR(head
);
839 generic_fillattr(inode
, stat
);
841 stat
->mode
= (stat
->mode
& S_IFMT
) | table
->mode
;
843 sysctl_head_finish(head
);
847 static const struct file_operations proc_sys_file_operations
= {
848 .open
= proc_sys_open
,
849 .poll
= proc_sys_poll
,
850 .read
= proc_sys_read
,
851 .write
= proc_sys_write
,
852 .llseek
= default_llseek
,
855 static const struct file_operations proc_sys_dir_file_operations
= {
856 .read
= generic_read_dir
,
857 .iterate_shared
= proc_sys_readdir
,
858 .llseek
= generic_file_llseek
,
861 static const struct inode_operations proc_sys_inode_operations
= {
862 .permission
= proc_sys_permission
,
863 .setattr
= proc_sys_setattr
,
864 .getattr
= proc_sys_getattr
,
867 static const struct inode_operations proc_sys_dir_operations
= {
868 .lookup
= proc_sys_lookup
,
869 .permission
= proc_sys_permission
,
870 .setattr
= proc_sys_setattr
,
871 .getattr
= proc_sys_getattr
,
874 static int proc_sys_revalidate(struct dentry
*dentry
, unsigned int flags
)
876 if (flags
& LOOKUP_RCU
)
878 return !PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
881 static int proc_sys_delete(const struct dentry
*dentry
)
883 return !!PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
886 static int sysctl_is_seen(struct ctl_table_header
*p
)
888 struct ctl_table_set
*set
= p
->set
;
890 spin_lock(&sysctl_lock
);
891 if (p
->unregistering
)
893 else if (!set
->is_seen
)
896 res
= set
->is_seen(set
);
897 spin_unlock(&sysctl_lock
);
901 static int proc_sys_compare(const struct dentry
*dentry
,
902 unsigned int len
, const char *str
, const struct qstr
*name
)
904 struct ctl_table_header
*head
;
907 /* Although proc doesn't have negative dentries, rcu-walk means
908 * that inode here can be NULL */
909 /* AV: can it, indeed? */
910 inode
= d_inode_rcu(dentry
);
913 if (name
->len
!= len
)
915 if (memcmp(name
->name
, str
, len
))
917 head
= rcu_dereference(PROC_I(inode
)->sysctl
);
918 return !head
|| !sysctl_is_seen(head
);
921 static const struct dentry_operations proc_sys_dentry_operations
= {
922 .d_revalidate
= proc_sys_revalidate
,
923 .d_delete
= proc_sys_delete
,
924 .d_compare
= proc_sys_compare
,
927 static struct ctl_dir
*find_subdir(struct ctl_dir
*dir
,
928 const char *name
, int namelen
)
930 struct ctl_table_header
*head
;
931 struct ctl_table
*entry
;
933 entry
= find_entry(&head
, dir
, name
, namelen
);
935 return ERR_PTR(-ENOENT
);
936 if (!S_ISDIR(entry
->mode
))
937 return ERR_PTR(-ENOTDIR
);
938 return container_of(head
, struct ctl_dir
, header
);
941 static struct ctl_dir
*new_dir(struct ctl_table_set
*set
,
942 const char *name
, int namelen
)
944 struct ctl_table
*table
;
946 struct ctl_node
*node
;
949 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node
) +
950 sizeof(struct ctl_table
)*2 + namelen
+ 1,
955 node
= (struct ctl_node
*)(new + 1);
956 table
= (struct ctl_table
*)(node
+ 1);
957 new_name
= (char *)(table
+ 2);
958 memcpy(new_name
, name
, namelen
);
959 new_name
[namelen
] = '\0';
960 table
[0].procname
= new_name
;
961 table
[0].mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
;
962 init_header(&new->header
, set
->dir
.header
.root
, set
, node
, table
);
968 * get_subdir - find or create a subdir with the specified name.
969 * @dir: Directory to create the subdirectory in
970 * @name: The name of the subdirectory to find or create
971 * @namelen: The length of name
973 * Takes a directory with an elevated reference count so we know that
974 * if we drop the lock the directory will not go away. Upon success
975 * the reference is moved from @dir to the returned subdirectory.
976 * Upon error an error code is returned and the reference on @dir is
979 static struct ctl_dir
*get_subdir(struct ctl_dir
*dir
,
980 const char *name
, int namelen
)
982 struct ctl_table_set
*set
= dir
->header
.set
;
983 struct ctl_dir
*subdir
, *new = NULL
;
986 spin_lock(&sysctl_lock
);
987 subdir
= find_subdir(dir
, name
, namelen
);
990 if (PTR_ERR(subdir
) != -ENOENT
)
993 spin_unlock(&sysctl_lock
);
994 new = new_dir(set
, name
, namelen
);
995 spin_lock(&sysctl_lock
);
996 subdir
= ERR_PTR(-ENOMEM
);
1000 /* Was the subdir added while we dropped the lock? */
1001 subdir
= find_subdir(dir
, name
, namelen
);
1002 if (!IS_ERR(subdir
))
1004 if (PTR_ERR(subdir
) != -ENOENT
)
1007 /* Nope. Use the our freshly made directory entry. */
1008 err
= insert_header(dir
, &new->header
);
1009 subdir
= ERR_PTR(err
);
1014 subdir
->header
.nreg
++;
1016 if (IS_ERR(subdir
)) {
1017 pr_err("sysctl could not get directory: ");
1018 sysctl_print_dir(dir
);
1019 pr_cont("/%*.*s %ld\n",
1020 namelen
, namelen
, name
, PTR_ERR(subdir
));
1022 drop_sysctl_table(&dir
->header
);
1024 drop_sysctl_table(&new->header
);
1025 spin_unlock(&sysctl_lock
);
1029 static struct ctl_dir
*xlate_dir(struct ctl_table_set
*set
, struct ctl_dir
*dir
)
1031 struct ctl_dir
*parent
;
1032 const char *procname
;
1033 if (!dir
->header
.parent
)
1035 parent
= xlate_dir(set
, dir
->header
.parent
);
1038 procname
= dir
->header
.ctl_table
[0].procname
;
1039 return find_subdir(parent
, procname
, strlen(procname
));
1042 static int sysctl_follow_link(struct ctl_table_header
**phead
,
1043 struct ctl_table
**pentry
)
1045 struct ctl_table_header
*head
;
1046 struct ctl_table_root
*root
;
1047 struct ctl_table_set
*set
;
1048 struct ctl_table
*entry
;
1049 struct ctl_dir
*dir
;
1053 spin_lock(&sysctl_lock
);
1054 root
= (*pentry
)->data
;
1055 set
= lookup_header_set(root
);
1056 dir
= xlate_dir(set
, (*phead
)->parent
);
1060 const char *procname
= (*pentry
)->procname
;
1062 entry
= find_entry(&head
, dir
, procname
, strlen(procname
));
1064 if (entry
&& use_table(head
)) {
1065 unuse_table(*phead
);
1072 spin_unlock(&sysctl_lock
);
1076 static int sysctl_err(const char *path
, struct ctl_table
*table
, char *fmt
, ...)
1078 struct va_format vaf
;
1081 va_start(args
, fmt
);
1085 pr_err("sysctl table check failed: %s/%s %pV\n",
1086 path
, table
->procname
, &vaf
);
1092 static int sysctl_check_table_array(const char *path
, struct ctl_table
*table
)
1096 if ((table
->proc_handler
== proc_douintvec
) ||
1097 (table
->proc_handler
== proc_douintvec_minmax
)) {
1098 if (table
->maxlen
!= sizeof(unsigned int))
1099 err
|= sysctl_err(path
, table
, "array now allowed");
1105 static int sysctl_check_table(const char *path
, struct ctl_table
*table
)
1108 for (; table
->procname
; table
++) {
1110 err
|= sysctl_err(path
, table
, "Not a file");
1112 if ((table
->proc_handler
== proc_dostring
) ||
1113 (table
->proc_handler
== proc_dointvec
) ||
1114 (table
->proc_handler
== proc_douintvec
) ||
1115 (table
->proc_handler
== proc_douintvec_minmax
) ||
1116 (table
->proc_handler
== proc_dointvec_minmax
) ||
1117 (table
->proc_handler
== proc_dointvec_jiffies
) ||
1118 (table
->proc_handler
== proc_dointvec_userhz_jiffies
) ||
1119 (table
->proc_handler
== proc_dointvec_ms_jiffies
) ||
1120 (table
->proc_handler
== proc_doulongvec_minmax
) ||
1121 (table
->proc_handler
== proc_doulongvec_ms_jiffies_minmax
)) {
1123 err
|= sysctl_err(path
, table
, "No data");
1125 err
|= sysctl_err(path
, table
, "No maxlen");
1127 err
|= sysctl_check_table_array(path
, table
);
1129 if (!table
->proc_handler
)
1130 err
|= sysctl_err(path
, table
, "No proc_handler");
1132 if ((table
->mode
& (S_IRUGO
|S_IWUGO
)) != table
->mode
)
1133 err
|= sysctl_err(path
, table
, "bogus .mode 0%o",
1139 static struct ctl_table_header
*new_links(struct ctl_dir
*dir
, struct ctl_table
*table
,
1140 struct ctl_table_root
*link_root
)
1142 struct ctl_table
*link_table
, *entry
, *link
;
1143 struct ctl_table_header
*links
;
1144 struct ctl_node
*node
;
1146 int nr_entries
, name_bytes
;
1150 for (entry
= table
; entry
->procname
; entry
++) {
1152 name_bytes
+= strlen(entry
->procname
) + 1;
1155 links
= kzalloc(sizeof(struct ctl_table_header
) +
1156 sizeof(struct ctl_node
)*nr_entries
+
1157 sizeof(struct ctl_table
)*(nr_entries
+ 1) +
1164 node
= (struct ctl_node
*)(links
+ 1);
1165 link_table
= (struct ctl_table
*)(node
+ nr_entries
);
1166 link_name
= (char *)&link_table
[nr_entries
+ 1];
1168 for (link
= link_table
, entry
= table
; entry
->procname
; link
++, entry
++) {
1169 int len
= strlen(entry
->procname
) + 1;
1170 memcpy(link_name
, entry
->procname
, len
);
1171 link
->procname
= link_name
;
1172 link
->mode
= S_IFLNK
|S_IRWXUGO
;
1173 link
->data
= link_root
;
1176 init_header(links
, dir
->header
.root
, dir
->header
.set
, node
, link_table
);
1177 links
->nreg
= nr_entries
;
1182 static bool get_links(struct ctl_dir
*dir
,
1183 struct ctl_table
*table
, struct ctl_table_root
*link_root
)
1185 struct ctl_table_header
*head
;
1186 struct ctl_table
*entry
, *link
;
1188 /* Are there links available for every entry in table? */
1189 for (entry
= table
; entry
->procname
; entry
++) {
1190 const char *procname
= entry
->procname
;
1191 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1194 if (S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
))
1196 if (S_ISLNK(link
->mode
) && (link
->data
== link_root
))
1201 /* The checks passed. Increase the registration count on the links */
1202 for (entry
= table
; entry
->procname
; entry
++) {
1203 const char *procname
= entry
->procname
;
1204 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1210 static int insert_links(struct ctl_table_header
*head
)
1212 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1213 struct ctl_dir
*core_parent
= NULL
;
1214 struct ctl_table_header
*links
;
1217 if (head
->set
== root_set
)
1220 core_parent
= xlate_dir(root_set
, head
->parent
);
1221 if (IS_ERR(core_parent
))
1224 if (get_links(core_parent
, head
->ctl_table
, head
->root
))
1227 core_parent
->header
.nreg
++;
1228 spin_unlock(&sysctl_lock
);
1230 links
= new_links(core_parent
, head
->ctl_table
, head
->root
);
1232 spin_lock(&sysctl_lock
);
1238 if (get_links(core_parent
, head
->ctl_table
, head
->root
)) {
1243 err
= insert_header(core_parent
, links
);
1247 drop_sysctl_table(&core_parent
->header
);
1252 * __register_sysctl_table - register a leaf sysctl table
1253 * @set: Sysctl tree to register on
1254 * @path: The path to the directory the sysctl table is in.
1255 * @table: the top-level table structure
1257 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1258 * array. A completely 0 filled entry terminates the table.
1260 * The members of the &struct ctl_table structure are used as follows:
1262 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1263 * enter a sysctl file
1265 * data - a pointer to data for use by proc_handler
1267 * maxlen - the maximum size in bytes of the data
1269 * mode - the file permissions for the /proc/sys file
1271 * child - must be %NULL.
1273 * proc_handler - the text handler routine (described below)
1275 * extra1, extra2 - extra pointers usable by the proc handler routines
1277 * Leaf nodes in the sysctl tree will be represented by a single file
1278 * under /proc; non-leaf nodes will be represented by directories.
1280 * There must be a proc_handler routine for any terminal nodes.
1281 * Several default handlers are available to cover common cases -
1283 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1284 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1285 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1287 * It is the handler's job to read the input buffer from user memory
1288 * and process it. The handler should return 0 on success.
1290 * This routine returns %NULL on a failure to register, and a pointer
1291 * to the table header on success.
1293 struct ctl_table_header
*__register_sysctl_table(
1294 struct ctl_table_set
*set
,
1295 const char *path
, struct ctl_table
*table
)
1297 struct ctl_table_root
*root
= set
->dir
.header
.root
;
1298 struct ctl_table_header
*header
;
1299 const char *name
, *nextname
;
1300 struct ctl_dir
*dir
;
1301 struct ctl_table
*entry
;
1302 struct ctl_node
*node
;
1305 for (entry
= table
; entry
->procname
; entry
++)
1308 header
= kzalloc(sizeof(struct ctl_table_header
) +
1309 sizeof(struct ctl_node
)*nr_entries
, GFP_KERNEL
);
1313 node
= (struct ctl_node
*)(header
+ 1);
1314 init_header(header
, root
, set
, node
, table
);
1315 if (sysctl_check_table(path
, table
))
1318 spin_lock(&sysctl_lock
);
1320 /* Reference moved down the diretory tree get_subdir */
1322 spin_unlock(&sysctl_lock
);
1324 /* Find the directory for the ctl_table */
1325 for (name
= path
; name
; name
= nextname
) {
1327 nextname
= strchr(name
, '/');
1329 namelen
= nextname
- name
;
1332 namelen
= strlen(name
);
1337 dir
= get_subdir(dir
, name
, namelen
);
1342 spin_lock(&sysctl_lock
);
1343 if (insert_header(dir
, header
))
1344 goto fail_put_dir_locked
;
1346 drop_sysctl_table(&dir
->header
);
1347 spin_unlock(&sysctl_lock
);
1351 fail_put_dir_locked
:
1352 drop_sysctl_table(&dir
->header
);
1353 spin_unlock(&sysctl_lock
);
1361 * register_sysctl - register a sysctl table
1362 * @path: The path to the directory the sysctl table is in.
1363 * @table: the table structure
1365 * Register a sysctl table. @table should be a filled in ctl_table
1366 * array. A completely 0 filled entry terminates the table.
1368 * See __register_sysctl_table for more details.
1370 struct ctl_table_header
*register_sysctl(const char *path
, struct ctl_table
*table
)
1372 return __register_sysctl_table(&sysctl_table_root
.default_set
,
1375 EXPORT_SYMBOL(register_sysctl
);
1377 static char *append_path(const char *path
, char *pos
, const char *name
)
1380 namelen
= strlen(name
);
1381 if (((pos
- path
) + namelen
+ 2) >= PATH_MAX
)
1383 memcpy(pos
, name
, namelen
);
1385 pos
[namelen
+ 1] = '\0';
1390 static int count_subheaders(struct ctl_table
*table
)
1393 int nr_subheaders
= 0;
1394 struct ctl_table
*entry
;
1396 /* special case: no directory and empty directory */
1397 if (!table
|| !table
->procname
)
1400 for (entry
= table
; entry
->procname
; entry
++) {
1402 nr_subheaders
+= count_subheaders(entry
->child
);
1406 return nr_subheaders
+ has_files
;
1409 static int register_leaf_sysctl_tables(const char *path
, char *pos
,
1410 struct ctl_table_header
***subheader
, struct ctl_table_set
*set
,
1411 struct ctl_table
*table
)
1413 struct ctl_table
*ctl_table_arg
= NULL
;
1414 struct ctl_table
*entry
, *files
;
1419 for (entry
= table
; entry
->procname
; entry
++) {
1427 /* If there are mixed files and directories we need a new table */
1428 if (nr_dirs
&& nr_files
) {
1429 struct ctl_table
*new;
1430 files
= kzalloc(sizeof(struct ctl_table
) * (nr_files
+ 1),
1435 ctl_table_arg
= files
;
1436 for (new = files
, entry
= table
; entry
->procname
; entry
++) {
1444 /* Register everything except a directory full of subdirectories */
1445 if (nr_files
|| !nr_dirs
) {
1446 struct ctl_table_header
*header
;
1447 header
= __register_sysctl_table(set
, path
, files
);
1449 kfree(ctl_table_arg
);
1453 /* Remember if we need to free the file table */
1454 header
->ctl_table_arg
= ctl_table_arg
;
1455 **subheader
= header
;
1459 /* Recurse into the subdirectories. */
1460 for (entry
= table
; entry
->procname
; entry
++) {
1466 err
= -ENAMETOOLONG
;
1467 child_pos
= append_path(path
, pos
, entry
->procname
);
1471 err
= register_leaf_sysctl_tables(path
, child_pos
, subheader
,
1479 /* On failure our caller will unregister all registered subheaders */
1484 * __register_sysctl_paths - register a sysctl table hierarchy
1485 * @set: Sysctl tree to register on
1486 * @path: The path to the directory the sysctl table is in.
1487 * @table: the top-level table structure
1489 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1490 * array. A completely 0 filled entry terminates the table.
1492 * See __register_sysctl_table for more details.
1494 struct ctl_table_header
*__register_sysctl_paths(
1495 struct ctl_table_set
*set
,
1496 const struct ctl_path
*path
, struct ctl_table
*table
)
1498 struct ctl_table
*ctl_table_arg
= table
;
1499 int nr_subheaders
= count_subheaders(table
);
1500 struct ctl_table_header
*header
= NULL
, **subheaders
, **subheader
;
1501 const struct ctl_path
*component
;
1502 char *new_path
, *pos
;
1504 pos
= new_path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1509 for (component
= path
; component
->procname
; component
++) {
1510 pos
= append_path(new_path
, pos
, component
->procname
);
1514 while (table
->procname
&& table
->child
&& !table
[1].procname
) {
1515 pos
= append_path(new_path
, pos
, table
->procname
);
1518 table
= table
->child
;
1520 if (nr_subheaders
== 1) {
1521 header
= __register_sysctl_table(set
, new_path
, table
);
1523 header
->ctl_table_arg
= ctl_table_arg
;
1525 header
= kzalloc(sizeof(*header
) +
1526 sizeof(*subheaders
)*nr_subheaders
, GFP_KERNEL
);
1530 subheaders
= (struct ctl_table_header
**) (header
+ 1);
1531 subheader
= subheaders
;
1532 header
->ctl_table_arg
= ctl_table_arg
;
1534 if (register_leaf_sysctl_tables(new_path
, pos
, &subheader
,
1536 goto err_register_leaves
;
1543 err_register_leaves
:
1544 while (subheader
> subheaders
) {
1545 struct ctl_table_header
*subh
= *(--subheader
);
1546 struct ctl_table
*table
= subh
->ctl_table_arg
;
1547 unregister_sysctl_table(subh
);
1556 * register_sysctl_table_path - register a sysctl table hierarchy
1557 * @path: The path to the directory the sysctl table is in.
1558 * @table: the top-level table structure
1560 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1561 * array. A completely 0 filled entry terminates the table.
1563 * See __register_sysctl_paths for more details.
1565 struct ctl_table_header
*register_sysctl_paths(const struct ctl_path
*path
,
1566 struct ctl_table
*table
)
1568 return __register_sysctl_paths(&sysctl_table_root
.default_set
,
1571 EXPORT_SYMBOL(register_sysctl_paths
);
1574 * register_sysctl_table - register a sysctl table hierarchy
1575 * @table: the top-level table structure
1577 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1578 * array. A completely 0 filled entry terminates the table.
1580 * See register_sysctl_paths for more details.
1582 struct ctl_table_header
*register_sysctl_table(struct ctl_table
*table
)
1584 static const struct ctl_path null_path
[] = { {} };
1586 return register_sysctl_paths(null_path
, table
);
1588 EXPORT_SYMBOL(register_sysctl_table
);
1590 static void put_links(struct ctl_table_header
*header
)
1592 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1593 struct ctl_table_root
*root
= header
->root
;
1594 struct ctl_dir
*parent
= header
->parent
;
1595 struct ctl_dir
*core_parent
;
1596 struct ctl_table
*entry
;
1598 if (header
->set
== root_set
)
1601 core_parent
= xlate_dir(root_set
, parent
);
1602 if (IS_ERR(core_parent
))
1605 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
1606 struct ctl_table_header
*link_head
;
1607 struct ctl_table
*link
;
1608 const char *name
= entry
->procname
;
1610 link
= find_entry(&link_head
, core_parent
, name
, strlen(name
));
1612 ((S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
)) ||
1613 (S_ISLNK(link
->mode
) && (link
->data
== root
)))) {
1614 drop_sysctl_table(link_head
);
1617 pr_err("sysctl link missing during unregister: ");
1618 sysctl_print_dir(parent
);
1619 pr_cont("/%s\n", name
);
1624 static void drop_sysctl_table(struct ctl_table_header
*header
)
1626 struct ctl_dir
*parent
= header
->parent
;
1632 start_unregistering(header
);
1633 if (!--header
->count
)
1634 kfree_rcu(header
, rcu
);
1637 drop_sysctl_table(&parent
->header
);
1641 * unregister_sysctl_table - unregister a sysctl table hierarchy
1642 * @header: the header returned from register_sysctl_table
1644 * Unregisters the sysctl table and all children. proc entries may not
1645 * actually be removed until they are no longer used by anyone.
1647 void unregister_sysctl_table(struct ctl_table_header
* header
)
1655 nr_subheaders
= count_subheaders(header
->ctl_table_arg
);
1656 if (unlikely(nr_subheaders
> 1)) {
1657 struct ctl_table_header
**subheaders
;
1660 subheaders
= (struct ctl_table_header
**)(header
+ 1);
1661 for (i
= nr_subheaders
-1; i
>= 0; i
--) {
1662 struct ctl_table_header
*subh
= subheaders
[i
];
1663 struct ctl_table
*table
= subh
->ctl_table_arg
;
1664 unregister_sysctl_table(subh
);
1671 spin_lock(&sysctl_lock
);
1672 drop_sysctl_table(header
);
1673 spin_unlock(&sysctl_lock
);
1675 EXPORT_SYMBOL(unregister_sysctl_table
);
1677 void setup_sysctl_set(struct ctl_table_set
*set
,
1678 struct ctl_table_root
*root
,
1679 int (*is_seen
)(struct ctl_table_set
*))
1681 memset(set
, 0, sizeof(*set
));
1682 set
->is_seen
= is_seen
;
1683 init_header(&set
->dir
.header
, root
, set
, NULL
, root_table
);
1686 void retire_sysctl_set(struct ctl_table_set
*set
)
1688 WARN_ON(!RB_EMPTY_ROOT(&set
->dir
.root
));
1691 int __init
proc_sys_init(void)
1693 struct proc_dir_entry
*proc_sys_root
;
1695 proc_sys_root
= proc_mkdir("sys", NULL
);
1696 proc_sys_root
->proc_iops
= &proc_sys_dir_operations
;
1697 proc_sys_root
->proc_fops
= &proc_sys_dir_file_operations
;
1698 proc_sys_root
->nlink
= 0;
1700 return sysctl_init();