4 #include <linux/init.h>
5 #include <linux/sysctl.h>
6 #include <linux/poll.h>
7 #include <linux/proc_fs.h>
8 #include <linux/printk.h>
9 #include <linux/security.h>
10 #include <linux/sched.h>
11 #include <linux/namei.h>
13 #include <linux/module.h>
16 static const struct dentry_operations proc_sys_dentry_operations
;
17 static const struct file_operations proc_sys_file_operations
;
18 static const struct inode_operations proc_sys_inode_operations
;
19 static const struct file_operations proc_sys_dir_file_operations
;
20 static const struct inode_operations proc_sys_dir_operations
;
22 /* Support for permanently empty directories */
24 struct ctl_table sysctl_mount_point
[] = {
28 static bool is_empty_dir(struct ctl_table_header
*head
)
30 return head
->ctl_table
[0].child
== sysctl_mount_point
;
33 static void set_empty_dir(struct ctl_dir
*dir
)
35 dir
->header
.ctl_table
[0].child
= sysctl_mount_point
;
38 static void clear_empty_dir(struct ctl_dir
*dir
)
41 dir
->header
.ctl_table
[0].child
= NULL
;
44 void proc_sys_poll_notify(struct ctl_table_poll
*poll
)
49 atomic_inc(&poll
->event
);
50 wake_up_interruptible(&poll
->wait
);
53 static struct ctl_table root_table
[] = {
56 .mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
,
60 static struct ctl_table_root sysctl_table_root
= {
61 .default_set
.dir
.header
= {
64 .ctl_table
= root_table
}},
65 .ctl_table_arg
= root_table
,
66 .root
= &sysctl_table_root
,
67 .set
= &sysctl_table_root
.default_set
,
71 static DEFINE_SPINLOCK(sysctl_lock
);
73 static void drop_sysctl_table(struct ctl_table_header
*header
);
74 static int sysctl_follow_link(struct ctl_table_header
**phead
,
75 struct ctl_table
**pentry
);
76 static int insert_links(struct ctl_table_header
*head
);
77 static void put_links(struct ctl_table_header
*header
);
79 static void sysctl_print_dir(struct ctl_dir
*dir
)
81 if (dir
->header
.parent
)
82 sysctl_print_dir(dir
->header
.parent
);
83 pr_cont("%s/", dir
->header
.ctl_table
[0].procname
);
86 static int namecmp(const char *name1
, int len1
, const char *name2
, int len2
)
95 cmp
= memcmp(name1
, name2
, minlen
);
101 /* Called under sysctl_lock */
102 static struct ctl_table
*find_entry(struct ctl_table_header
**phead
,
103 struct ctl_dir
*dir
, const char *name
, int namelen
)
105 struct ctl_table_header
*head
;
106 struct ctl_table
*entry
;
107 struct rb_node
*node
= dir
->root
.rb_node
;
111 struct ctl_node
*ctl_node
;
112 const char *procname
;
115 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
116 head
= ctl_node
->header
;
117 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
118 procname
= entry
->procname
;
120 cmp
= namecmp(name
, namelen
, procname
, strlen(procname
));
122 node
= node
->rb_left
;
124 node
= node
->rb_right
;
133 static int insert_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
135 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
136 struct rb_node
**p
= &head
->parent
->root
.rb_node
;
137 struct rb_node
*parent
= NULL
;
138 const char *name
= entry
->procname
;
139 int namelen
= strlen(name
);
142 struct ctl_table_header
*parent_head
;
143 struct ctl_table
*parent_entry
;
144 struct ctl_node
*parent_node
;
145 const char *parent_name
;
149 parent_node
= rb_entry(parent
, struct ctl_node
, node
);
150 parent_head
= parent_node
->header
;
151 parent_entry
= &parent_head
->ctl_table
[parent_node
- parent_head
->node
];
152 parent_name
= parent_entry
->procname
;
154 cmp
= namecmp(name
, namelen
, parent_name
, strlen(parent_name
));
160 pr_err("sysctl duplicate entry: ");
161 sysctl_print_dir(head
->parent
);
162 pr_cont("/%s\n", entry
->procname
);
167 rb_link_node(node
, parent
, p
);
168 rb_insert_color(node
, &head
->parent
->root
);
172 static void erase_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
174 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
176 rb_erase(node
, &head
->parent
->root
);
179 static void init_header(struct ctl_table_header
*head
,
180 struct ctl_table_root
*root
, struct ctl_table_set
*set
,
181 struct ctl_node
*node
, struct ctl_table
*table
)
183 head
->ctl_table
= table
;
184 head
->ctl_table_arg
= table
;
188 head
->unregistering
= NULL
;
194 struct ctl_table
*entry
;
195 for (entry
= table
; entry
->procname
; entry
++, node
++)
200 static void erase_header(struct ctl_table_header
*head
)
202 struct ctl_table
*entry
;
203 for (entry
= head
->ctl_table
; entry
->procname
; entry
++)
204 erase_entry(head
, entry
);
207 static int insert_header(struct ctl_dir
*dir
, struct ctl_table_header
*header
)
209 struct ctl_table
*entry
;
212 /* Is this a permanently empty directory? */
213 if (is_empty_dir(&dir
->header
))
216 /* Am I creating a permanently empty directory? */
217 if (header
->ctl_table
== sysctl_mount_point
) {
218 if (!RB_EMPTY_ROOT(&dir
->root
))
224 header
->parent
= dir
;
225 err
= insert_links(header
);
228 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
229 err
= insert_entry(header
, entry
);
235 erase_header(header
);
238 if (header
->ctl_table
== sysctl_mount_point
)
239 clear_empty_dir(dir
);
240 header
->parent
= NULL
;
241 drop_sysctl_table(&dir
->header
);
245 /* called under sysctl_lock */
246 static int use_table(struct ctl_table_header
*p
)
248 if (unlikely(p
->unregistering
))
254 /* called under sysctl_lock */
255 static void unuse_table(struct ctl_table_header
*p
)
258 if (unlikely(p
->unregistering
))
259 complete(p
->unregistering
);
262 /* called under sysctl_lock, will reacquire if has to wait */
263 static void start_unregistering(struct ctl_table_header
*p
)
266 * if p->used is 0, nobody will ever touch that entry again;
267 * we'll eliminate all paths to it before dropping sysctl_lock
269 if (unlikely(p
->used
)) {
270 struct completion wait
;
271 init_completion(&wait
);
272 p
->unregistering
= &wait
;
273 spin_unlock(&sysctl_lock
);
274 wait_for_completion(&wait
);
275 spin_lock(&sysctl_lock
);
277 /* anything non-NULL; we'll never dereference it */
278 p
->unregistering
= ERR_PTR(-EINVAL
);
281 * do not remove from the list until nobody holds it; walking the
282 * list in do_sysctl() relies on that.
287 static void sysctl_head_get(struct ctl_table_header
*head
)
289 spin_lock(&sysctl_lock
);
291 spin_unlock(&sysctl_lock
);
294 void sysctl_head_put(struct ctl_table_header
*head
)
296 spin_lock(&sysctl_lock
);
298 kfree_rcu(head
, rcu
);
299 spin_unlock(&sysctl_lock
);
302 static struct ctl_table_header
*sysctl_head_grab(struct ctl_table_header
*head
)
305 spin_lock(&sysctl_lock
);
306 if (!use_table(head
))
307 head
= ERR_PTR(-ENOENT
);
308 spin_unlock(&sysctl_lock
);
312 static void sysctl_head_finish(struct ctl_table_header
*head
)
316 spin_lock(&sysctl_lock
);
318 spin_unlock(&sysctl_lock
);
321 static struct ctl_table_set
*
322 lookup_header_set(struct ctl_table_root
*root
)
324 struct ctl_table_set
*set
= &root
->default_set
;
326 set
= root
->lookup(root
);
330 static struct ctl_table
*lookup_entry(struct ctl_table_header
**phead
,
332 const char *name
, int namelen
)
334 struct ctl_table_header
*head
;
335 struct ctl_table
*entry
;
337 spin_lock(&sysctl_lock
);
338 entry
= find_entry(&head
, dir
, name
, namelen
);
339 if (entry
&& use_table(head
))
343 spin_unlock(&sysctl_lock
);
347 static struct ctl_node
*first_usable_entry(struct rb_node
*node
)
349 struct ctl_node
*ctl_node
;
351 for (;node
; node
= rb_next(node
)) {
352 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
353 if (use_table(ctl_node
->header
))
359 static void first_entry(struct ctl_dir
*dir
,
360 struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
362 struct ctl_table_header
*head
= NULL
;
363 struct ctl_table
*entry
= NULL
;
364 struct ctl_node
*ctl_node
;
366 spin_lock(&sysctl_lock
);
367 ctl_node
= first_usable_entry(rb_first(&dir
->root
));
368 spin_unlock(&sysctl_lock
);
370 head
= ctl_node
->header
;
371 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
377 static void next_entry(struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
379 struct ctl_table_header
*head
= *phead
;
380 struct ctl_table
*entry
= *pentry
;
381 struct ctl_node
*ctl_node
= &head
->node
[entry
- head
->ctl_table
];
383 spin_lock(&sysctl_lock
);
386 ctl_node
= first_usable_entry(rb_next(&ctl_node
->node
));
387 spin_unlock(&sysctl_lock
);
390 head
= ctl_node
->header
;
391 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
397 void register_sysctl_root(struct ctl_table_root
*root
)
402 * sysctl_perm does NOT grant the superuser all rights automatically, because
403 * some sysctl variables are readonly even to root.
406 static int test_perm(int mode
, int op
)
408 if (uid_eq(current_euid(), GLOBAL_ROOT_UID
))
410 else if (in_egroup_p(GLOBAL_ROOT_GID
))
412 if ((op
& ~mode
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
)) == 0)
417 static int sysctl_perm(struct ctl_table_header
*head
, struct ctl_table
*table
, int op
)
419 struct ctl_table_root
*root
= head
->root
;
422 if (root
->permissions
)
423 mode
= root
->permissions(head
, table
);
427 return test_perm(mode
, op
);
430 static struct inode
*proc_sys_make_inode(struct super_block
*sb
,
431 struct ctl_table_header
*head
, struct ctl_table
*table
)
433 struct ctl_table_root
*root
= head
->root
;
435 struct proc_inode
*ei
;
437 inode
= new_inode(sb
);
441 inode
->i_ino
= get_next_ino();
443 sysctl_head_get(head
);
446 ei
->sysctl_entry
= table
;
448 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= current_time(inode
);
449 inode
->i_mode
= table
->mode
;
450 if (!S_ISDIR(table
->mode
)) {
451 inode
->i_mode
|= S_IFREG
;
452 inode
->i_op
= &proc_sys_inode_operations
;
453 inode
->i_fop
= &proc_sys_file_operations
;
455 inode
->i_mode
|= S_IFDIR
;
456 inode
->i_op
= &proc_sys_dir_operations
;
457 inode
->i_fop
= &proc_sys_dir_file_operations
;
458 if (is_empty_dir(head
))
459 make_empty_dir_inode(inode
);
462 if (root
->set_ownership
)
463 root
->set_ownership(head
, table
, &inode
->i_uid
, &inode
->i_gid
);
469 static struct ctl_table_header
*grab_header(struct inode
*inode
)
471 struct ctl_table_header
*head
= PROC_I(inode
)->sysctl
;
473 head
= &sysctl_table_root
.default_set
.dir
.header
;
474 return sysctl_head_grab(head
);
477 static struct dentry
*proc_sys_lookup(struct inode
*dir
, struct dentry
*dentry
,
480 struct ctl_table_header
*head
= grab_header(dir
);
481 struct ctl_table_header
*h
= NULL
;
482 const struct qstr
*name
= &dentry
->d_name
;
485 struct dentry
*err
= ERR_PTR(-ENOENT
);
486 struct ctl_dir
*ctl_dir
;
490 return ERR_CAST(head
);
492 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
494 p
= lookup_entry(&h
, ctl_dir
, name
->name
, name
->len
);
498 if (S_ISLNK(p
->mode
)) {
499 ret
= sysctl_follow_link(&h
, &p
);
505 err
= ERR_PTR(-ENOMEM
);
506 inode
= proc_sys_make_inode(dir
->i_sb
, h
? h
: head
, p
);
511 d_set_d_op(dentry
, &proc_sys_dentry_operations
);
512 d_add(dentry
, inode
);
516 sysctl_head_finish(h
);
517 sysctl_head_finish(head
);
521 static ssize_t
proc_sys_call_handler(struct file
*filp
, void __user
*buf
,
522 size_t count
, loff_t
*ppos
, int write
)
524 struct inode
*inode
= file_inode(filp
);
525 struct ctl_table_header
*head
= grab_header(inode
);
526 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
531 return PTR_ERR(head
);
534 * At this point we know that the sysctl was not unregistered
535 * and won't be until we finish.
538 if (sysctl_perm(head
, table
, write
? MAY_WRITE
: MAY_READ
))
541 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
543 if (!table
->proc_handler
)
546 /* careful: calling conventions are nasty here */
548 error
= table
->proc_handler(table
, write
, buf
, &res
, ppos
);
552 sysctl_head_finish(head
);
557 static ssize_t
proc_sys_read(struct file
*filp
, char __user
*buf
,
558 size_t count
, loff_t
*ppos
)
560 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 0);
563 static ssize_t
proc_sys_write(struct file
*filp
, const char __user
*buf
,
564 size_t count
, loff_t
*ppos
)
566 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 1);
569 static int proc_sys_open(struct inode
*inode
, struct file
*filp
)
571 struct ctl_table_header
*head
= grab_header(inode
);
572 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
574 /* sysctl was unregistered */
576 return PTR_ERR(head
);
579 filp
->private_data
= proc_sys_poll_event(table
->poll
);
581 sysctl_head_finish(head
);
586 static unsigned int proc_sys_poll(struct file
*filp
, poll_table
*wait
)
588 struct inode
*inode
= file_inode(filp
);
589 struct ctl_table_header
*head
= grab_header(inode
);
590 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
591 unsigned int ret
= DEFAULT_POLLMASK
;
594 /* sysctl was unregistered */
596 return POLLERR
| POLLHUP
;
598 if (!table
->proc_handler
)
604 event
= (unsigned long)filp
->private_data
;
605 poll_wait(filp
, &table
->poll
->wait
, wait
);
607 if (event
!= atomic_read(&table
->poll
->event
)) {
608 filp
->private_data
= proc_sys_poll_event(table
->poll
);
609 ret
= POLLIN
| POLLRDNORM
| POLLERR
| POLLPRI
;
613 sysctl_head_finish(head
);
618 static bool proc_sys_fill_cache(struct file
*file
,
619 struct dir_context
*ctx
,
620 struct ctl_table_header
*head
,
621 struct ctl_table
*table
)
623 struct dentry
*child
, *dir
= file
->f_path
.dentry
;
627 unsigned type
= DT_UNKNOWN
;
629 qname
.name
= table
->procname
;
630 qname
.len
= strlen(table
->procname
);
631 qname
.hash
= full_name_hash(dir
, qname
.name
, qname
.len
);
633 child
= d_lookup(dir
, &qname
);
635 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
636 child
= d_alloc_parallel(dir
, &qname
, &wq
);
639 if (d_in_lookup(child
)) {
640 inode
= proc_sys_make_inode(dir
->d_sb
, head
, table
);
642 d_lookup_done(child
);
646 d_set_d_op(child
, &proc_sys_dentry_operations
);
650 inode
= d_inode(child
);
652 type
= inode
->i_mode
>> 12;
654 return dir_emit(ctx
, qname
.name
, qname
.len
, ino
, type
);
657 static bool proc_sys_link_fill_cache(struct file
*file
,
658 struct dir_context
*ctx
,
659 struct ctl_table_header
*head
,
660 struct ctl_table
*table
)
663 head
= sysctl_head_grab(head
);
665 if (S_ISLNK(table
->mode
)) {
666 /* It is not an error if we can not follow the link ignore it */
667 int err
= sysctl_follow_link(&head
, &table
);
672 ret
= proc_sys_fill_cache(file
, ctx
, head
, table
);
674 sysctl_head_finish(head
);
678 static int scan(struct ctl_table_header
*head
, struct ctl_table
*table
,
679 unsigned long *pos
, struct file
*file
,
680 struct dir_context
*ctx
)
684 if ((*pos
)++ < ctx
->pos
)
687 if (unlikely(S_ISLNK(table
->mode
)))
688 res
= proc_sys_link_fill_cache(file
, ctx
, head
, table
);
690 res
= proc_sys_fill_cache(file
, ctx
, head
, table
);
698 static int proc_sys_readdir(struct file
*file
, struct dir_context
*ctx
)
700 struct ctl_table_header
*head
= grab_header(file_inode(file
));
701 struct ctl_table_header
*h
= NULL
;
702 struct ctl_table
*entry
;
703 struct ctl_dir
*ctl_dir
;
707 return PTR_ERR(head
);
709 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
711 if (!dir_emit_dots(file
, ctx
))
716 for (first_entry(ctl_dir
, &h
, &entry
); h
; next_entry(&h
, &entry
)) {
717 if (!scan(h
, entry
, &pos
, file
, ctx
)) {
718 sysctl_head_finish(h
);
723 sysctl_head_finish(head
);
727 static int proc_sys_permission(struct inode
*inode
, int mask
)
730 * sysctl entries that are not writeable,
731 * are _NOT_ writeable, capabilities or not.
733 struct ctl_table_header
*head
;
734 struct ctl_table
*table
;
737 /* Executable files are not allowed under /proc/sys/ */
738 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
))
741 head
= grab_header(inode
);
743 return PTR_ERR(head
);
745 table
= PROC_I(inode
)->sysctl_entry
;
746 if (!table
) /* global root - r-xr-xr-x */
747 error
= mask
& MAY_WRITE
? -EACCES
: 0;
748 else /* Use the permissions on the sysctl table entry */
749 error
= sysctl_perm(head
, table
, mask
& ~MAY_NOT_BLOCK
);
751 sysctl_head_finish(head
);
755 static int proc_sys_setattr(struct dentry
*dentry
, struct iattr
*attr
)
757 struct inode
*inode
= d_inode(dentry
);
758 struct user_namespace
*s_user_ns
;
761 if (attr
->ia_valid
& (ATTR_MODE
| ATTR_UID
| ATTR_GID
))
764 /* Don't let anyone mess with weird proc files */
765 s_user_ns
= inode
->i_sb
->s_user_ns
;
766 if (!kuid_has_mapping(s_user_ns
, inode
->i_uid
) ||
767 !kgid_has_mapping(s_user_ns
, inode
->i_gid
))
770 error
= setattr_prepare(dentry
, attr
);
774 setattr_copy(inode
, attr
);
775 mark_inode_dirty(inode
);
779 static int proc_sys_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
, struct kstat
*stat
)
781 struct inode
*inode
= d_inode(dentry
);
782 struct ctl_table_header
*head
= grab_header(inode
);
783 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
786 return PTR_ERR(head
);
788 generic_fillattr(inode
, stat
);
790 stat
->mode
= (stat
->mode
& S_IFMT
) | table
->mode
;
792 sysctl_head_finish(head
);
796 static const struct file_operations proc_sys_file_operations
= {
797 .open
= proc_sys_open
,
798 .poll
= proc_sys_poll
,
799 .read
= proc_sys_read
,
800 .write
= proc_sys_write
,
801 .llseek
= default_llseek
,
804 static const struct file_operations proc_sys_dir_file_operations
= {
805 .read
= generic_read_dir
,
806 .iterate_shared
= proc_sys_readdir
,
807 .llseek
= generic_file_llseek
,
810 static const struct inode_operations proc_sys_inode_operations
= {
811 .permission
= proc_sys_permission
,
812 .setattr
= proc_sys_setattr
,
813 .getattr
= proc_sys_getattr
,
816 static const struct inode_operations proc_sys_dir_operations
= {
817 .lookup
= proc_sys_lookup
,
818 .permission
= proc_sys_permission
,
819 .setattr
= proc_sys_setattr
,
820 .getattr
= proc_sys_getattr
,
823 static int proc_sys_revalidate(struct dentry
*dentry
, unsigned int flags
)
825 if (flags
& LOOKUP_RCU
)
827 return !PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
830 static int proc_sys_delete(const struct dentry
*dentry
)
832 return !!PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
835 static int sysctl_is_seen(struct ctl_table_header
*p
)
837 struct ctl_table_set
*set
= p
->set
;
839 spin_lock(&sysctl_lock
);
840 if (p
->unregistering
)
842 else if (!set
->is_seen
)
845 res
= set
->is_seen(set
);
846 spin_unlock(&sysctl_lock
);
850 static int proc_sys_compare(const struct dentry
*dentry
,
851 unsigned int len
, const char *str
, const struct qstr
*name
)
853 struct ctl_table_header
*head
;
856 /* Although proc doesn't have negative dentries, rcu-walk means
857 * that inode here can be NULL */
858 /* AV: can it, indeed? */
859 inode
= d_inode_rcu(dentry
);
862 if (name
->len
!= len
)
864 if (memcmp(name
->name
, str
, len
))
866 head
= rcu_dereference(PROC_I(inode
)->sysctl
);
867 return !head
|| !sysctl_is_seen(head
);
870 static const struct dentry_operations proc_sys_dentry_operations
= {
871 .d_revalidate
= proc_sys_revalidate
,
872 .d_delete
= proc_sys_delete
,
873 .d_compare
= proc_sys_compare
,
876 static struct ctl_dir
*find_subdir(struct ctl_dir
*dir
,
877 const char *name
, int namelen
)
879 struct ctl_table_header
*head
;
880 struct ctl_table
*entry
;
882 entry
= find_entry(&head
, dir
, name
, namelen
);
884 return ERR_PTR(-ENOENT
);
885 if (!S_ISDIR(entry
->mode
))
886 return ERR_PTR(-ENOTDIR
);
887 return container_of(head
, struct ctl_dir
, header
);
890 static struct ctl_dir
*new_dir(struct ctl_table_set
*set
,
891 const char *name
, int namelen
)
893 struct ctl_table
*table
;
895 struct ctl_node
*node
;
898 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node
) +
899 sizeof(struct ctl_table
)*2 + namelen
+ 1,
904 node
= (struct ctl_node
*)(new + 1);
905 table
= (struct ctl_table
*)(node
+ 1);
906 new_name
= (char *)(table
+ 2);
907 memcpy(new_name
, name
, namelen
);
908 new_name
[namelen
] = '\0';
909 table
[0].procname
= new_name
;
910 table
[0].mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
;
911 init_header(&new->header
, set
->dir
.header
.root
, set
, node
, table
);
917 * get_subdir - find or create a subdir with the specified name.
918 * @dir: Directory to create the subdirectory in
919 * @name: The name of the subdirectory to find or create
920 * @namelen: The length of name
922 * Takes a directory with an elevated reference count so we know that
923 * if we drop the lock the directory will not go away. Upon success
924 * the reference is moved from @dir to the returned subdirectory.
925 * Upon error an error code is returned and the reference on @dir is
928 static struct ctl_dir
*get_subdir(struct ctl_dir
*dir
,
929 const char *name
, int namelen
)
931 struct ctl_table_set
*set
= dir
->header
.set
;
932 struct ctl_dir
*subdir
, *new = NULL
;
935 spin_lock(&sysctl_lock
);
936 subdir
= find_subdir(dir
, name
, namelen
);
939 if (PTR_ERR(subdir
) != -ENOENT
)
942 spin_unlock(&sysctl_lock
);
943 new = new_dir(set
, name
, namelen
);
944 spin_lock(&sysctl_lock
);
945 subdir
= ERR_PTR(-ENOMEM
);
949 /* Was the subdir added while we dropped the lock? */
950 subdir
= find_subdir(dir
, name
, namelen
);
953 if (PTR_ERR(subdir
) != -ENOENT
)
956 /* Nope. Use the our freshly made directory entry. */
957 err
= insert_header(dir
, &new->header
);
958 subdir
= ERR_PTR(err
);
963 subdir
->header
.nreg
++;
965 if (IS_ERR(subdir
)) {
966 pr_err("sysctl could not get directory: ");
967 sysctl_print_dir(dir
);
968 pr_cont("/%*.*s %ld\n",
969 namelen
, namelen
, name
, PTR_ERR(subdir
));
971 drop_sysctl_table(&dir
->header
);
973 drop_sysctl_table(&new->header
);
974 spin_unlock(&sysctl_lock
);
978 static struct ctl_dir
*xlate_dir(struct ctl_table_set
*set
, struct ctl_dir
*dir
)
980 struct ctl_dir
*parent
;
981 const char *procname
;
982 if (!dir
->header
.parent
)
984 parent
= xlate_dir(set
, dir
->header
.parent
);
987 procname
= dir
->header
.ctl_table
[0].procname
;
988 return find_subdir(parent
, procname
, strlen(procname
));
991 static int sysctl_follow_link(struct ctl_table_header
**phead
,
992 struct ctl_table
**pentry
)
994 struct ctl_table_header
*head
;
995 struct ctl_table_root
*root
;
996 struct ctl_table_set
*set
;
997 struct ctl_table
*entry
;
1002 spin_lock(&sysctl_lock
);
1003 root
= (*pentry
)->data
;
1004 set
= lookup_header_set(root
);
1005 dir
= xlate_dir(set
, (*phead
)->parent
);
1009 const char *procname
= (*pentry
)->procname
;
1011 entry
= find_entry(&head
, dir
, procname
, strlen(procname
));
1013 if (entry
&& use_table(head
)) {
1014 unuse_table(*phead
);
1021 spin_unlock(&sysctl_lock
);
1025 static int sysctl_err(const char *path
, struct ctl_table
*table
, char *fmt
, ...)
1027 struct va_format vaf
;
1030 va_start(args
, fmt
);
1034 pr_err("sysctl table check failed: %s/%s %pV\n",
1035 path
, table
->procname
, &vaf
);
1041 static int sysctl_check_table(const char *path
, struct ctl_table
*table
)
1044 for (; table
->procname
; table
++) {
1046 err
= sysctl_err(path
, table
, "Not a file");
1048 if ((table
->proc_handler
== proc_dostring
) ||
1049 (table
->proc_handler
== proc_dointvec
) ||
1050 (table
->proc_handler
== proc_dointvec_minmax
) ||
1051 (table
->proc_handler
== proc_dointvec_jiffies
) ||
1052 (table
->proc_handler
== proc_dointvec_userhz_jiffies
) ||
1053 (table
->proc_handler
== proc_dointvec_ms_jiffies
) ||
1054 (table
->proc_handler
== proc_doulongvec_minmax
) ||
1055 (table
->proc_handler
== proc_doulongvec_ms_jiffies_minmax
)) {
1057 err
= sysctl_err(path
, table
, "No data");
1059 err
= sysctl_err(path
, table
, "No maxlen");
1061 if (!table
->proc_handler
)
1062 err
= sysctl_err(path
, table
, "No proc_handler");
1064 if ((table
->mode
& (S_IRUGO
|S_IWUGO
)) != table
->mode
)
1065 err
= sysctl_err(path
, table
, "bogus .mode 0%o",
1071 static struct ctl_table_header
*new_links(struct ctl_dir
*dir
, struct ctl_table
*table
,
1072 struct ctl_table_root
*link_root
)
1074 struct ctl_table
*link_table
, *entry
, *link
;
1075 struct ctl_table_header
*links
;
1076 struct ctl_node
*node
;
1078 int nr_entries
, name_bytes
;
1082 for (entry
= table
; entry
->procname
; entry
++) {
1084 name_bytes
+= strlen(entry
->procname
) + 1;
1087 links
= kzalloc(sizeof(struct ctl_table_header
) +
1088 sizeof(struct ctl_node
)*nr_entries
+
1089 sizeof(struct ctl_table
)*(nr_entries
+ 1) +
1096 node
= (struct ctl_node
*)(links
+ 1);
1097 link_table
= (struct ctl_table
*)(node
+ nr_entries
);
1098 link_name
= (char *)&link_table
[nr_entries
+ 1];
1100 for (link
= link_table
, entry
= table
; entry
->procname
; link
++, entry
++) {
1101 int len
= strlen(entry
->procname
) + 1;
1102 memcpy(link_name
, entry
->procname
, len
);
1103 link
->procname
= link_name
;
1104 link
->mode
= S_IFLNK
|S_IRWXUGO
;
1105 link
->data
= link_root
;
1108 init_header(links
, dir
->header
.root
, dir
->header
.set
, node
, link_table
);
1109 links
->nreg
= nr_entries
;
1114 static bool get_links(struct ctl_dir
*dir
,
1115 struct ctl_table
*table
, struct ctl_table_root
*link_root
)
1117 struct ctl_table_header
*head
;
1118 struct ctl_table
*entry
, *link
;
1120 /* Are there links available for every entry in table? */
1121 for (entry
= table
; entry
->procname
; entry
++) {
1122 const char *procname
= entry
->procname
;
1123 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1126 if (S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
))
1128 if (S_ISLNK(link
->mode
) && (link
->data
== link_root
))
1133 /* The checks passed. Increase the registration count on the links */
1134 for (entry
= table
; entry
->procname
; entry
++) {
1135 const char *procname
= entry
->procname
;
1136 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1142 static int insert_links(struct ctl_table_header
*head
)
1144 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1145 struct ctl_dir
*core_parent
= NULL
;
1146 struct ctl_table_header
*links
;
1149 if (head
->set
== root_set
)
1152 core_parent
= xlate_dir(root_set
, head
->parent
);
1153 if (IS_ERR(core_parent
))
1156 if (get_links(core_parent
, head
->ctl_table
, head
->root
))
1159 core_parent
->header
.nreg
++;
1160 spin_unlock(&sysctl_lock
);
1162 links
= new_links(core_parent
, head
->ctl_table
, head
->root
);
1164 spin_lock(&sysctl_lock
);
1170 if (get_links(core_parent
, head
->ctl_table
, head
->root
)) {
1175 err
= insert_header(core_parent
, links
);
1179 drop_sysctl_table(&core_parent
->header
);
1184 * __register_sysctl_table - register a leaf sysctl table
1185 * @set: Sysctl tree to register on
1186 * @path: The path to the directory the sysctl table is in.
1187 * @table: the top-level table structure
1189 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1190 * array. A completely 0 filled entry terminates the table.
1192 * The members of the &struct ctl_table structure are used as follows:
1194 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1195 * enter a sysctl file
1197 * data - a pointer to data for use by proc_handler
1199 * maxlen - the maximum size in bytes of the data
1201 * mode - the file permissions for the /proc/sys file
1203 * child - must be %NULL.
1205 * proc_handler - the text handler routine (described below)
1207 * extra1, extra2 - extra pointers usable by the proc handler routines
1209 * Leaf nodes in the sysctl tree will be represented by a single file
1210 * under /proc; non-leaf nodes will be represented by directories.
1212 * There must be a proc_handler routine for any terminal nodes.
1213 * Several default handlers are available to cover common cases -
1215 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1216 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1217 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1219 * It is the handler's job to read the input buffer from user memory
1220 * and process it. The handler should return 0 on success.
1222 * This routine returns %NULL on a failure to register, and a pointer
1223 * to the table header on success.
1225 struct ctl_table_header
*__register_sysctl_table(
1226 struct ctl_table_set
*set
,
1227 const char *path
, struct ctl_table
*table
)
1229 struct ctl_table_root
*root
= set
->dir
.header
.root
;
1230 struct ctl_table_header
*header
;
1231 const char *name
, *nextname
;
1232 struct ctl_dir
*dir
;
1233 struct ctl_table
*entry
;
1234 struct ctl_node
*node
;
1237 for (entry
= table
; entry
->procname
; entry
++)
1240 header
= kzalloc(sizeof(struct ctl_table_header
) +
1241 sizeof(struct ctl_node
)*nr_entries
, GFP_KERNEL
);
1245 node
= (struct ctl_node
*)(header
+ 1);
1246 init_header(header
, root
, set
, node
, table
);
1247 if (sysctl_check_table(path
, table
))
1250 spin_lock(&sysctl_lock
);
1252 /* Reference moved down the diretory tree get_subdir */
1254 spin_unlock(&sysctl_lock
);
1256 /* Find the directory for the ctl_table */
1257 for (name
= path
; name
; name
= nextname
) {
1259 nextname
= strchr(name
, '/');
1261 namelen
= nextname
- name
;
1264 namelen
= strlen(name
);
1269 dir
= get_subdir(dir
, name
, namelen
);
1274 spin_lock(&sysctl_lock
);
1275 if (insert_header(dir
, header
))
1276 goto fail_put_dir_locked
;
1278 drop_sysctl_table(&dir
->header
);
1279 spin_unlock(&sysctl_lock
);
1283 fail_put_dir_locked
:
1284 drop_sysctl_table(&dir
->header
);
1285 spin_unlock(&sysctl_lock
);
1293 * register_sysctl - register a sysctl table
1294 * @path: The path to the directory the sysctl table is in.
1295 * @table: the table structure
1297 * Register a sysctl table. @table should be a filled in ctl_table
1298 * array. A completely 0 filled entry terminates the table.
1300 * See __register_sysctl_table for more details.
1302 struct ctl_table_header
*register_sysctl(const char *path
, struct ctl_table
*table
)
1304 return __register_sysctl_table(&sysctl_table_root
.default_set
,
1307 EXPORT_SYMBOL(register_sysctl
);
1309 static char *append_path(const char *path
, char *pos
, const char *name
)
1312 namelen
= strlen(name
);
1313 if (((pos
- path
) + namelen
+ 2) >= PATH_MAX
)
1315 memcpy(pos
, name
, namelen
);
1317 pos
[namelen
+ 1] = '\0';
1322 static int count_subheaders(struct ctl_table
*table
)
1325 int nr_subheaders
= 0;
1326 struct ctl_table
*entry
;
1328 /* special case: no directory and empty directory */
1329 if (!table
|| !table
->procname
)
1332 for (entry
= table
; entry
->procname
; entry
++) {
1334 nr_subheaders
+= count_subheaders(entry
->child
);
1338 return nr_subheaders
+ has_files
;
1341 static int register_leaf_sysctl_tables(const char *path
, char *pos
,
1342 struct ctl_table_header
***subheader
, struct ctl_table_set
*set
,
1343 struct ctl_table
*table
)
1345 struct ctl_table
*ctl_table_arg
= NULL
;
1346 struct ctl_table
*entry
, *files
;
1351 for (entry
= table
; entry
->procname
; entry
++) {
1359 /* If there are mixed files and directories we need a new table */
1360 if (nr_dirs
&& nr_files
) {
1361 struct ctl_table
*new;
1362 files
= kzalloc(sizeof(struct ctl_table
) * (nr_files
+ 1),
1367 ctl_table_arg
= files
;
1368 for (new = files
, entry
= table
; entry
->procname
; entry
++) {
1376 /* Register everything except a directory full of subdirectories */
1377 if (nr_files
|| !nr_dirs
) {
1378 struct ctl_table_header
*header
;
1379 header
= __register_sysctl_table(set
, path
, files
);
1381 kfree(ctl_table_arg
);
1385 /* Remember if we need to free the file table */
1386 header
->ctl_table_arg
= ctl_table_arg
;
1387 **subheader
= header
;
1391 /* Recurse into the subdirectories. */
1392 for (entry
= table
; entry
->procname
; entry
++) {
1398 err
= -ENAMETOOLONG
;
1399 child_pos
= append_path(path
, pos
, entry
->procname
);
1403 err
= register_leaf_sysctl_tables(path
, child_pos
, subheader
,
1411 /* On failure our caller will unregister all registered subheaders */
1416 * __register_sysctl_paths - register a sysctl table hierarchy
1417 * @set: Sysctl tree to register on
1418 * @path: The path to the directory the sysctl table is in.
1419 * @table: the top-level table structure
1421 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1422 * array. A completely 0 filled entry terminates the table.
1424 * See __register_sysctl_table for more details.
1426 struct ctl_table_header
*__register_sysctl_paths(
1427 struct ctl_table_set
*set
,
1428 const struct ctl_path
*path
, struct ctl_table
*table
)
1430 struct ctl_table
*ctl_table_arg
= table
;
1431 int nr_subheaders
= count_subheaders(table
);
1432 struct ctl_table_header
*header
= NULL
, **subheaders
, **subheader
;
1433 const struct ctl_path
*component
;
1434 char *new_path
, *pos
;
1436 pos
= new_path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1441 for (component
= path
; component
->procname
; component
++) {
1442 pos
= append_path(new_path
, pos
, component
->procname
);
1446 while (table
->procname
&& table
->child
&& !table
[1].procname
) {
1447 pos
= append_path(new_path
, pos
, table
->procname
);
1450 table
= table
->child
;
1452 if (nr_subheaders
== 1) {
1453 header
= __register_sysctl_table(set
, new_path
, table
);
1455 header
->ctl_table_arg
= ctl_table_arg
;
1457 header
= kzalloc(sizeof(*header
) +
1458 sizeof(*subheaders
)*nr_subheaders
, GFP_KERNEL
);
1462 subheaders
= (struct ctl_table_header
**) (header
+ 1);
1463 subheader
= subheaders
;
1464 header
->ctl_table_arg
= ctl_table_arg
;
1466 if (register_leaf_sysctl_tables(new_path
, pos
, &subheader
,
1468 goto err_register_leaves
;
1475 err_register_leaves
:
1476 while (subheader
> subheaders
) {
1477 struct ctl_table_header
*subh
= *(--subheader
);
1478 struct ctl_table
*table
= subh
->ctl_table_arg
;
1479 unregister_sysctl_table(subh
);
1488 * register_sysctl_table_path - register a sysctl table hierarchy
1489 * @path: The path to the directory the sysctl table is in.
1490 * @table: the top-level table structure
1492 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1493 * array. A completely 0 filled entry terminates the table.
1495 * See __register_sysctl_paths for more details.
1497 struct ctl_table_header
*register_sysctl_paths(const struct ctl_path
*path
,
1498 struct ctl_table
*table
)
1500 return __register_sysctl_paths(&sysctl_table_root
.default_set
,
1503 EXPORT_SYMBOL(register_sysctl_paths
);
1506 * register_sysctl_table - register a sysctl table hierarchy
1507 * @table: the top-level table structure
1509 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1510 * array. A completely 0 filled entry terminates the table.
1512 * See register_sysctl_paths for more details.
1514 struct ctl_table_header
*register_sysctl_table(struct ctl_table
*table
)
1516 static const struct ctl_path null_path
[] = { {} };
1518 return register_sysctl_paths(null_path
, table
);
1520 EXPORT_SYMBOL(register_sysctl_table
);
1522 static void put_links(struct ctl_table_header
*header
)
1524 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1525 struct ctl_table_root
*root
= header
->root
;
1526 struct ctl_dir
*parent
= header
->parent
;
1527 struct ctl_dir
*core_parent
;
1528 struct ctl_table
*entry
;
1530 if (header
->set
== root_set
)
1533 core_parent
= xlate_dir(root_set
, parent
);
1534 if (IS_ERR(core_parent
))
1537 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
1538 struct ctl_table_header
*link_head
;
1539 struct ctl_table
*link
;
1540 const char *name
= entry
->procname
;
1542 link
= find_entry(&link_head
, core_parent
, name
, strlen(name
));
1544 ((S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
)) ||
1545 (S_ISLNK(link
->mode
) && (link
->data
== root
)))) {
1546 drop_sysctl_table(link_head
);
1549 pr_err("sysctl link missing during unregister: ");
1550 sysctl_print_dir(parent
);
1551 pr_cont("/%s\n", name
);
1556 static void drop_sysctl_table(struct ctl_table_header
*header
)
1558 struct ctl_dir
*parent
= header
->parent
;
1564 start_unregistering(header
);
1565 if (!--header
->count
)
1566 kfree_rcu(header
, rcu
);
1569 drop_sysctl_table(&parent
->header
);
1573 * unregister_sysctl_table - unregister a sysctl table hierarchy
1574 * @header: the header returned from register_sysctl_table
1576 * Unregisters the sysctl table and all children. proc entries may not
1577 * actually be removed until they are no longer used by anyone.
1579 void unregister_sysctl_table(struct ctl_table_header
* header
)
1587 nr_subheaders
= count_subheaders(header
->ctl_table_arg
);
1588 if (unlikely(nr_subheaders
> 1)) {
1589 struct ctl_table_header
**subheaders
;
1592 subheaders
= (struct ctl_table_header
**)(header
+ 1);
1593 for (i
= nr_subheaders
-1; i
>= 0; i
--) {
1594 struct ctl_table_header
*subh
= subheaders
[i
];
1595 struct ctl_table
*table
= subh
->ctl_table_arg
;
1596 unregister_sysctl_table(subh
);
1603 spin_lock(&sysctl_lock
);
1604 drop_sysctl_table(header
);
1605 spin_unlock(&sysctl_lock
);
1607 EXPORT_SYMBOL(unregister_sysctl_table
);
1609 void setup_sysctl_set(struct ctl_table_set
*set
,
1610 struct ctl_table_root
*root
,
1611 int (*is_seen
)(struct ctl_table_set
*))
1613 memset(set
, 0, sizeof(*set
));
1614 set
->is_seen
= is_seen
;
1615 init_header(&set
->dir
.header
, root
, set
, NULL
, root_table
);
1618 void retire_sysctl_set(struct ctl_table_set
*set
)
1620 WARN_ON(!RB_EMPTY_ROOT(&set
->dir
.root
));
1623 int __init
proc_sys_init(void)
1625 struct proc_dir_entry
*proc_sys_root
;
1627 proc_sys_root
= proc_mkdir("sys", NULL
);
1628 proc_sys_root
->proc_iops
= &proc_sys_dir_operations
;
1629 proc_sys_root
->proc_fops
= &proc_sys_dir_file_operations
;
1630 proc_sys_root
->nlink
= 0;
1632 return sysctl_init();