4 #include <linux/init.h>
5 #include <linux/sysctl.h>
6 #include <linux/poll.h>
7 #include <linux/proc_fs.h>
8 #include <linux/printk.h>
9 #include <linux/security.h>
10 #include <linux/sched.h>
11 #include <linux/cred.h>
12 #include <linux/namei.h>
14 #include <linux/module.h>
17 static const struct dentry_operations proc_sys_dentry_operations
;
18 static const struct file_operations proc_sys_file_operations
;
19 static const struct inode_operations proc_sys_inode_operations
;
20 static const struct file_operations proc_sys_dir_file_operations
;
21 static const struct inode_operations proc_sys_dir_operations
;
23 /* Support for permanently empty directories */
25 struct ctl_table sysctl_mount_point
[] = {
29 static bool is_empty_dir(struct ctl_table_header
*head
)
31 return head
->ctl_table
[0].child
== sysctl_mount_point
;
34 static void set_empty_dir(struct ctl_dir
*dir
)
36 dir
->header
.ctl_table
[0].child
= sysctl_mount_point
;
39 static void clear_empty_dir(struct ctl_dir
*dir
)
42 dir
->header
.ctl_table
[0].child
= NULL
;
45 void proc_sys_poll_notify(struct ctl_table_poll
*poll
)
50 atomic_inc(&poll
->event
);
51 wake_up_interruptible(&poll
->wait
);
54 static struct ctl_table root_table
[] = {
57 .mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
,
61 static struct ctl_table_root sysctl_table_root
= {
62 .default_set
.dir
.header
= {
65 .ctl_table
= root_table
}},
66 .ctl_table_arg
= root_table
,
67 .root
= &sysctl_table_root
,
68 .set
= &sysctl_table_root
.default_set
,
72 static DEFINE_SPINLOCK(sysctl_lock
);
74 static void drop_sysctl_table(struct ctl_table_header
*header
);
75 static int sysctl_follow_link(struct ctl_table_header
**phead
,
76 struct ctl_table
**pentry
);
77 static int insert_links(struct ctl_table_header
*head
);
78 static void put_links(struct ctl_table_header
*header
);
80 static void sysctl_print_dir(struct ctl_dir
*dir
)
82 if (dir
->header
.parent
)
83 sysctl_print_dir(dir
->header
.parent
);
84 pr_cont("%s/", dir
->header
.ctl_table
[0].procname
);
87 static int namecmp(const char *name1
, int len1
, const char *name2
, int len2
)
96 cmp
= memcmp(name1
, name2
, minlen
);
102 /* Called under sysctl_lock */
103 static struct ctl_table
*find_entry(struct ctl_table_header
**phead
,
104 struct ctl_dir
*dir
, const char *name
, int namelen
)
106 struct ctl_table_header
*head
;
107 struct ctl_table
*entry
;
108 struct rb_node
*node
= dir
->root
.rb_node
;
112 struct ctl_node
*ctl_node
;
113 const char *procname
;
116 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
117 head
= ctl_node
->header
;
118 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
119 procname
= entry
->procname
;
121 cmp
= namecmp(name
, namelen
, procname
, strlen(procname
));
123 node
= node
->rb_left
;
125 node
= node
->rb_right
;
134 static int insert_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
136 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
137 struct rb_node
**p
= &head
->parent
->root
.rb_node
;
138 struct rb_node
*parent
= NULL
;
139 const char *name
= entry
->procname
;
140 int namelen
= strlen(name
);
143 struct ctl_table_header
*parent_head
;
144 struct ctl_table
*parent_entry
;
145 struct ctl_node
*parent_node
;
146 const char *parent_name
;
150 parent_node
= rb_entry(parent
, struct ctl_node
, node
);
151 parent_head
= parent_node
->header
;
152 parent_entry
= &parent_head
->ctl_table
[parent_node
- parent_head
->node
];
153 parent_name
= parent_entry
->procname
;
155 cmp
= namecmp(name
, namelen
, parent_name
, strlen(parent_name
));
161 pr_err("sysctl duplicate entry: ");
162 sysctl_print_dir(head
->parent
);
163 pr_cont("/%s\n", entry
->procname
);
168 rb_link_node(node
, parent
, p
);
169 rb_insert_color(node
, &head
->parent
->root
);
173 static void erase_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
175 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
177 rb_erase(node
, &head
->parent
->root
);
180 static void init_header(struct ctl_table_header
*head
,
181 struct ctl_table_root
*root
, struct ctl_table_set
*set
,
182 struct ctl_node
*node
, struct ctl_table
*table
)
184 head
->ctl_table
= table
;
185 head
->ctl_table_arg
= table
;
189 head
->unregistering
= NULL
;
194 INIT_HLIST_HEAD(&head
->inodes
);
196 struct ctl_table
*entry
;
197 for (entry
= table
; entry
->procname
; entry
++, node
++)
202 static void erase_header(struct ctl_table_header
*head
)
204 struct ctl_table
*entry
;
205 for (entry
= head
->ctl_table
; entry
->procname
; entry
++)
206 erase_entry(head
, entry
);
209 static int insert_header(struct ctl_dir
*dir
, struct ctl_table_header
*header
)
211 struct ctl_table
*entry
;
214 /* Is this a permanently empty directory? */
215 if (is_empty_dir(&dir
->header
))
218 /* Am I creating a permanently empty directory? */
219 if (header
->ctl_table
== sysctl_mount_point
) {
220 if (!RB_EMPTY_ROOT(&dir
->root
))
226 header
->parent
= dir
;
227 err
= insert_links(header
);
230 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
231 err
= insert_entry(header
, entry
);
237 erase_header(header
);
240 if (header
->ctl_table
== sysctl_mount_point
)
241 clear_empty_dir(dir
);
242 header
->parent
= NULL
;
243 drop_sysctl_table(&dir
->header
);
247 /* called under sysctl_lock */
248 static int use_table(struct ctl_table_header
*p
)
250 if (unlikely(p
->unregistering
))
256 /* called under sysctl_lock */
257 static void unuse_table(struct ctl_table_header
*p
)
260 if (unlikely(p
->unregistering
))
261 complete(p
->unregistering
);
264 static void proc_sys_prune_dcache(struct ctl_table_header
*head
)
267 struct proc_inode
*ei
;
268 struct hlist_node
*node
;
269 struct super_block
*sb
;
273 node
= hlist_first_rcu(&head
->inodes
);
276 ei
= hlist_entry(node
, struct proc_inode
, sysctl_inodes
);
277 spin_lock(&sysctl_lock
);
278 hlist_del_init_rcu(&ei
->sysctl_inodes
);
279 spin_unlock(&sysctl_lock
);
281 inode
= &ei
->vfs_inode
;
283 if (!atomic_inc_not_zero(&sb
->s_active
))
285 inode
= igrab(inode
);
287 if (unlikely(!inode
)) {
288 deactivate_super(sb
);
293 d_prune_aliases(inode
);
295 deactivate_super(sb
);
302 /* called under sysctl_lock, will reacquire if has to wait */
303 static void start_unregistering(struct ctl_table_header
*p
)
306 * if p->used is 0, nobody will ever touch that entry again;
307 * we'll eliminate all paths to it before dropping sysctl_lock
309 if (unlikely(p
->used
)) {
310 struct completion wait
;
311 init_completion(&wait
);
312 p
->unregistering
= &wait
;
313 spin_unlock(&sysctl_lock
);
314 wait_for_completion(&wait
);
316 /* anything non-NULL; we'll never dereference it */
317 p
->unregistering
= ERR_PTR(-EINVAL
);
318 spin_unlock(&sysctl_lock
);
321 * Prune dentries for unregistered sysctls: namespaced sysctls
322 * can have duplicate names and contaminate dcache very badly.
324 proc_sys_prune_dcache(p
);
326 * do not remove from the list until nobody holds it; walking the
327 * list in do_sysctl() relies on that.
329 spin_lock(&sysctl_lock
);
333 static struct ctl_table_header
*sysctl_head_grab(struct ctl_table_header
*head
)
336 spin_lock(&sysctl_lock
);
337 if (!use_table(head
))
338 head
= ERR_PTR(-ENOENT
);
339 spin_unlock(&sysctl_lock
);
343 static void sysctl_head_finish(struct ctl_table_header
*head
)
347 spin_lock(&sysctl_lock
);
349 spin_unlock(&sysctl_lock
);
352 static struct ctl_table_set
*
353 lookup_header_set(struct ctl_table_root
*root
)
355 struct ctl_table_set
*set
= &root
->default_set
;
357 set
= root
->lookup(root
);
361 static struct ctl_table
*lookup_entry(struct ctl_table_header
**phead
,
363 const char *name
, int namelen
)
365 struct ctl_table_header
*head
;
366 struct ctl_table
*entry
;
368 spin_lock(&sysctl_lock
);
369 entry
= find_entry(&head
, dir
, name
, namelen
);
370 if (entry
&& use_table(head
))
374 spin_unlock(&sysctl_lock
);
378 static struct ctl_node
*first_usable_entry(struct rb_node
*node
)
380 struct ctl_node
*ctl_node
;
382 for (;node
; node
= rb_next(node
)) {
383 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
384 if (use_table(ctl_node
->header
))
390 static void first_entry(struct ctl_dir
*dir
,
391 struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
393 struct ctl_table_header
*head
= NULL
;
394 struct ctl_table
*entry
= NULL
;
395 struct ctl_node
*ctl_node
;
397 spin_lock(&sysctl_lock
);
398 ctl_node
= first_usable_entry(rb_first(&dir
->root
));
399 spin_unlock(&sysctl_lock
);
401 head
= ctl_node
->header
;
402 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
408 static void next_entry(struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
410 struct ctl_table_header
*head
= *phead
;
411 struct ctl_table
*entry
= *pentry
;
412 struct ctl_node
*ctl_node
= &head
->node
[entry
- head
->ctl_table
];
414 spin_lock(&sysctl_lock
);
417 ctl_node
= first_usable_entry(rb_next(&ctl_node
->node
));
418 spin_unlock(&sysctl_lock
);
421 head
= ctl_node
->header
;
422 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
429 * sysctl_perm does NOT grant the superuser all rights automatically, because
430 * some sysctl variables are readonly even to root.
433 static int test_perm(int mode
, int op
)
435 if (uid_eq(current_euid(), GLOBAL_ROOT_UID
))
437 else if (in_egroup_p(GLOBAL_ROOT_GID
))
439 if ((op
& ~mode
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
)) == 0)
444 static int sysctl_perm(struct ctl_table_header
*head
, struct ctl_table
*table
, int op
)
446 struct ctl_table_root
*root
= head
->root
;
449 if (root
->permissions
)
450 mode
= root
->permissions(head
, table
);
454 return test_perm(mode
, op
);
457 static struct inode
*proc_sys_make_inode(struct super_block
*sb
,
458 struct ctl_table_header
*head
, struct ctl_table
*table
)
460 struct ctl_table_root
*root
= head
->root
;
462 struct proc_inode
*ei
;
464 inode
= new_inode(sb
);
468 inode
->i_ino
= get_next_ino();
472 spin_lock(&sysctl_lock
);
473 if (unlikely(head
->unregistering
)) {
474 spin_unlock(&sysctl_lock
);
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
);
506 void proc_sys_evict_inode(struct inode
*inode
, struct ctl_table_header
*head
)
508 spin_lock(&sysctl_lock
);
509 hlist_del_init_rcu(&PROC_I(inode
)->sysctl_inodes
);
511 kfree_rcu(head
, rcu
);
512 spin_unlock(&sysctl_lock
);
515 static struct ctl_table_header
*grab_header(struct inode
*inode
)
517 struct ctl_table_header
*head
= PROC_I(inode
)->sysctl
;
519 head
= &sysctl_table_root
.default_set
.dir
.header
;
520 return sysctl_head_grab(head
);
523 static struct dentry
*proc_sys_lookup(struct inode
*dir
, struct dentry
*dentry
,
526 struct ctl_table_header
*head
= grab_header(dir
);
527 struct ctl_table_header
*h
= NULL
;
528 const struct qstr
*name
= &dentry
->d_name
;
531 struct dentry
*err
= ERR_PTR(-ENOENT
);
532 struct ctl_dir
*ctl_dir
;
536 return ERR_CAST(head
);
538 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
540 p
= lookup_entry(&h
, ctl_dir
, name
->name
, name
->len
);
544 if (S_ISLNK(p
->mode
)) {
545 ret
= sysctl_follow_link(&h
, &p
);
551 err
= ERR_PTR(-ENOMEM
);
552 inode
= proc_sys_make_inode(dir
->i_sb
, h
? h
: head
, p
);
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
)
709 head
= sysctl_head_grab(head
);
711 if (S_ISLNK(table
->mode
)) {
712 /* It is not an error if we can not follow the link ignore it */
713 int err
= sysctl_follow_link(&head
, &table
);
718 ret
= proc_sys_fill_cache(file
, ctx
, head
, table
);
720 sysctl_head_finish(head
);
724 static int scan(struct ctl_table_header
*head
, struct ctl_table
*table
,
725 unsigned long *pos
, struct file
*file
,
726 struct dir_context
*ctx
)
730 if ((*pos
)++ < ctx
->pos
)
733 if (unlikely(S_ISLNK(table
->mode
)))
734 res
= proc_sys_link_fill_cache(file
, ctx
, head
, table
);
736 res
= proc_sys_fill_cache(file
, ctx
, head
, table
);
744 static int proc_sys_readdir(struct file
*file
, struct dir_context
*ctx
)
746 struct ctl_table_header
*head
= grab_header(file_inode(file
));
747 struct ctl_table_header
*h
= NULL
;
748 struct ctl_table
*entry
;
749 struct ctl_dir
*ctl_dir
;
753 return PTR_ERR(head
);
755 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
757 if (!dir_emit_dots(file
, ctx
))
762 for (first_entry(ctl_dir
, &h
, &entry
); h
; next_entry(&h
, &entry
)) {
763 if (!scan(h
, entry
, &pos
, file
, ctx
)) {
764 sysctl_head_finish(h
);
769 sysctl_head_finish(head
);
773 static int proc_sys_permission(struct inode
*inode
, int mask
)
776 * sysctl entries that are not writeable,
777 * are _NOT_ writeable, capabilities or not.
779 struct ctl_table_header
*head
;
780 struct ctl_table
*table
;
783 /* Executable files are not allowed under /proc/sys/ */
784 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
))
787 head
= grab_header(inode
);
789 return PTR_ERR(head
);
791 table
= PROC_I(inode
)->sysctl_entry
;
792 if (!table
) /* global root - r-xr-xr-x */
793 error
= mask
& MAY_WRITE
? -EACCES
: 0;
794 else /* Use the permissions on the sysctl table entry */
795 error
= sysctl_perm(head
, table
, mask
& ~MAY_NOT_BLOCK
);
797 sysctl_head_finish(head
);
801 static int proc_sys_setattr(struct dentry
*dentry
, struct iattr
*attr
)
803 struct inode
*inode
= d_inode(dentry
);
806 if (attr
->ia_valid
& (ATTR_MODE
| ATTR_UID
| ATTR_GID
))
809 error
= setattr_prepare(dentry
, attr
);
813 setattr_copy(inode
, attr
);
814 mark_inode_dirty(inode
);
818 static int proc_sys_getattr(const struct path
*path
, struct kstat
*stat
,
819 u32 request_mask
, unsigned int query_flags
)
821 struct inode
*inode
= d_inode(path
->dentry
);
822 struct ctl_table_header
*head
= grab_header(inode
);
823 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
826 return PTR_ERR(head
);
828 generic_fillattr(inode
, stat
);
830 stat
->mode
= (stat
->mode
& S_IFMT
) | table
->mode
;
832 sysctl_head_finish(head
);
836 static const struct file_operations proc_sys_file_operations
= {
837 .open
= proc_sys_open
,
838 .poll
= proc_sys_poll
,
839 .read
= proc_sys_read
,
840 .write
= proc_sys_write
,
841 .llseek
= default_llseek
,
844 static const struct file_operations proc_sys_dir_file_operations
= {
845 .read
= generic_read_dir
,
846 .iterate_shared
= proc_sys_readdir
,
847 .llseek
= generic_file_llseek
,
850 static const struct inode_operations proc_sys_inode_operations
= {
851 .permission
= proc_sys_permission
,
852 .setattr
= proc_sys_setattr
,
853 .getattr
= proc_sys_getattr
,
856 static const struct inode_operations proc_sys_dir_operations
= {
857 .lookup
= proc_sys_lookup
,
858 .permission
= proc_sys_permission
,
859 .setattr
= proc_sys_setattr
,
860 .getattr
= proc_sys_getattr
,
863 static int proc_sys_revalidate(struct dentry
*dentry
, unsigned int flags
)
865 if (flags
& LOOKUP_RCU
)
867 return !PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
870 static int proc_sys_delete(const struct dentry
*dentry
)
872 return !!PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
875 static int sysctl_is_seen(struct ctl_table_header
*p
)
877 struct ctl_table_set
*set
= p
->set
;
879 spin_lock(&sysctl_lock
);
880 if (p
->unregistering
)
882 else if (!set
->is_seen
)
885 res
= set
->is_seen(set
);
886 spin_unlock(&sysctl_lock
);
890 static int proc_sys_compare(const struct dentry
*dentry
,
891 unsigned int len
, const char *str
, const struct qstr
*name
)
893 struct ctl_table_header
*head
;
896 /* Although proc doesn't have negative dentries, rcu-walk means
897 * that inode here can be NULL */
898 /* AV: can it, indeed? */
899 inode
= d_inode_rcu(dentry
);
902 if (name
->len
!= len
)
904 if (memcmp(name
->name
, str
, len
))
906 head
= rcu_dereference(PROC_I(inode
)->sysctl
);
907 return !head
|| !sysctl_is_seen(head
);
910 static const struct dentry_operations proc_sys_dentry_operations
= {
911 .d_revalidate
= proc_sys_revalidate
,
912 .d_delete
= proc_sys_delete
,
913 .d_compare
= proc_sys_compare
,
916 static struct ctl_dir
*find_subdir(struct ctl_dir
*dir
,
917 const char *name
, int namelen
)
919 struct ctl_table_header
*head
;
920 struct ctl_table
*entry
;
922 entry
= find_entry(&head
, dir
, name
, namelen
);
924 return ERR_PTR(-ENOENT
);
925 if (!S_ISDIR(entry
->mode
))
926 return ERR_PTR(-ENOTDIR
);
927 return container_of(head
, struct ctl_dir
, header
);
930 static struct ctl_dir
*new_dir(struct ctl_table_set
*set
,
931 const char *name
, int namelen
)
933 struct ctl_table
*table
;
935 struct ctl_node
*node
;
938 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node
) +
939 sizeof(struct ctl_table
)*2 + namelen
+ 1,
944 node
= (struct ctl_node
*)(new + 1);
945 table
= (struct ctl_table
*)(node
+ 1);
946 new_name
= (char *)(table
+ 2);
947 memcpy(new_name
, name
, namelen
);
948 new_name
[namelen
] = '\0';
949 table
[0].procname
= new_name
;
950 table
[0].mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
;
951 init_header(&new->header
, set
->dir
.header
.root
, set
, node
, table
);
957 * get_subdir - find or create a subdir with the specified name.
958 * @dir: Directory to create the subdirectory in
959 * @name: The name of the subdirectory to find or create
960 * @namelen: The length of name
962 * Takes a directory with an elevated reference count so we know that
963 * if we drop the lock the directory will not go away. Upon success
964 * the reference is moved from @dir to the returned subdirectory.
965 * Upon error an error code is returned and the reference on @dir is
968 static struct ctl_dir
*get_subdir(struct ctl_dir
*dir
,
969 const char *name
, int namelen
)
971 struct ctl_table_set
*set
= dir
->header
.set
;
972 struct ctl_dir
*subdir
, *new = NULL
;
975 spin_lock(&sysctl_lock
);
976 subdir
= find_subdir(dir
, name
, namelen
);
979 if (PTR_ERR(subdir
) != -ENOENT
)
982 spin_unlock(&sysctl_lock
);
983 new = new_dir(set
, name
, namelen
);
984 spin_lock(&sysctl_lock
);
985 subdir
= ERR_PTR(-ENOMEM
);
989 /* Was the subdir added while we dropped the lock? */
990 subdir
= find_subdir(dir
, name
, namelen
);
993 if (PTR_ERR(subdir
) != -ENOENT
)
996 /* Nope. Use the our freshly made directory entry. */
997 err
= insert_header(dir
, &new->header
);
998 subdir
= ERR_PTR(err
);
1003 subdir
->header
.nreg
++;
1005 if (IS_ERR(subdir
)) {
1006 pr_err("sysctl could not get directory: ");
1007 sysctl_print_dir(dir
);
1008 pr_cont("/%*.*s %ld\n",
1009 namelen
, namelen
, name
, PTR_ERR(subdir
));
1011 drop_sysctl_table(&dir
->header
);
1013 drop_sysctl_table(&new->header
);
1014 spin_unlock(&sysctl_lock
);
1018 static struct ctl_dir
*xlate_dir(struct ctl_table_set
*set
, struct ctl_dir
*dir
)
1020 struct ctl_dir
*parent
;
1021 const char *procname
;
1022 if (!dir
->header
.parent
)
1024 parent
= xlate_dir(set
, dir
->header
.parent
);
1027 procname
= dir
->header
.ctl_table
[0].procname
;
1028 return find_subdir(parent
, procname
, strlen(procname
));
1031 static int sysctl_follow_link(struct ctl_table_header
**phead
,
1032 struct ctl_table
**pentry
)
1034 struct ctl_table_header
*head
;
1035 struct ctl_table_root
*root
;
1036 struct ctl_table_set
*set
;
1037 struct ctl_table
*entry
;
1038 struct ctl_dir
*dir
;
1042 spin_lock(&sysctl_lock
);
1043 root
= (*pentry
)->data
;
1044 set
= lookup_header_set(root
);
1045 dir
= xlate_dir(set
, (*phead
)->parent
);
1049 const char *procname
= (*pentry
)->procname
;
1051 entry
= find_entry(&head
, dir
, procname
, strlen(procname
));
1053 if (entry
&& use_table(head
)) {
1054 unuse_table(*phead
);
1061 spin_unlock(&sysctl_lock
);
1065 static int sysctl_err(const char *path
, struct ctl_table
*table
, char *fmt
, ...)
1067 struct va_format vaf
;
1070 va_start(args
, fmt
);
1074 pr_err("sysctl table check failed: %s/%s %pV\n",
1075 path
, table
->procname
, &vaf
);
1081 static int sysctl_check_table_array(const char *path
, struct ctl_table
*table
)
1085 if ((table
->proc_handler
== proc_douintvec
) ||
1086 (table
->proc_handler
== proc_douintvec_minmax
)) {
1087 if (table
->maxlen
!= sizeof(unsigned int))
1088 err
|= sysctl_err(path
, table
, "array now allowed");
1094 static int sysctl_check_table(const char *path
, struct ctl_table
*table
)
1097 for (; table
->procname
; table
++) {
1099 err
|= sysctl_err(path
, table
, "Not a file");
1101 if ((table
->proc_handler
== proc_dostring
) ||
1102 (table
->proc_handler
== proc_dointvec
) ||
1103 (table
->proc_handler
== proc_douintvec
) ||
1104 (table
->proc_handler
== proc_douintvec_minmax
) ||
1105 (table
->proc_handler
== proc_dointvec_minmax
) ||
1106 (table
->proc_handler
== proc_dointvec_jiffies
) ||
1107 (table
->proc_handler
== proc_dointvec_userhz_jiffies
) ||
1108 (table
->proc_handler
== proc_dointvec_ms_jiffies
) ||
1109 (table
->proc_handler
== proc_doulongvec_minmax
) ||
1110 (table
->proc_handler
== proc_doulongvec_ms_jiffies_minmax
)) {
1112 err
|= sysctl_err(path
, table
, "No data");
1114 err
|= sysctl_err(path
, table
, "No maxlen");
1116 err
|= sysctl_check_table_array(path
, table
);
1118 if (!table
->proc_handler
)
1119 err
|= sysctl_err(path
, table
, "No proc_handler");
1121 if ((table
->mode
& (S_IRUGO
|S_IWUGO
)) != table
->mode
)
1122 err
|= sysctl_err(path
, table
, "bogus .mode 0%o",
1128 static struct ctl_table_header
*new_links(struct ctl_dir
*dir
, struct ctl_table
*table
,
1129 struct ctl_table_root
*link_root
)
1131 struct ctl_table
*link_table
, *entry
, *link
;
1132 struct ctl_table_header
*links
;
1133 struct ctl_node
*node
;
1135 int nr_entries
, name_bytes
;
1139 for (entry
= table
; entry
->procname
; entry
++) {
1141 name_bytes
+= strlen(entry
->procname
) + 1;
1144 links
= kzalloc(sizeof(struct ctl_table_header
) +
1145 sizeof(struct ctl_node
)*nr_entries
+
1146 sizeof(struct ctl_table
)*(nr_entries
+ 1) +
1153 node
= (struct ctl_node
*)(links
+ 1);
1154 link_table
= (struct ctl_table
*)(node
+ nr_entries
);
1155 link_name
= (char *)&link_table
[nr_entries
+ 1];
1157 for (link
= link_table
, entry
= table
; entry
->procname
; link
++, entry
++) {
1158 int len
= strlen(entry
->procname
) + 1;
1159 memcpy(link_name
, entry
->procname
, len
);
1160 link
->procname
= link_name
;
1161 link
->mode
= S_IFLNK
|S_IRWXUGO
;
1162 link
->data
= link_root
;
1165 init_header(links
, dir
->header
.root
, dir
->header
.set
, node
, link_table
);
1166 links
->nreg
= nr_entries
;
1171 static bool get_links(struct ctl_dir
*dir
,
1172 struct ctl_table
*table
, struct ctl_table_root
*link_root
)
1174 struct ctl_table_header
*head
;
1175 struct ctl_table
*entry
, *link
;
1177 /* Are there links available for every entry in table? */
1178 for (entry
= table
; entry
->procname
; entry
++) {
1179 const char *procname
= entry
->procname
;
1180 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1183 if (S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
))
1185 if (S_ISLNK(link
->mode
) && (link
->data
== link_root
))
1190 /* The checks passed. Increase the registration count on the links */
1191 for (entry
= table
; entry
->procname
; entry
++) {
1192 const char *procname
= entry
->procname
;
1193 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1199 static int insert_links(struct ctl_table_header
*head
)
1201 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1202 struct ctl_dir
*core_parent
= NULL
;
1203 struct ctl_table_header
*links
;
1206 if (head
->set
== root_set
)
1209 core_parent
= xlate_dir(root_set
, head
->parent
);
1210 if (IS_ERR(core_parent
))
1213 if (get_links(core_parent
, head
->ctl_table
, head
->root
))
1216 core_parent
->header
.nreg
++;
1217 spin_unlock(&sysctl_lock
);
1219 links
= new_links(core_parent
, head
->ctl_table
, head
->root
);
1221 spin_lock(&sysctl_lock
);
1227 if (get_links(core_parent
, head
->ctl_table
, head
->root
)) {
1232 err
= insert_header(core_parent
, links
);
1236 drop_sysctl_table(&core_parent
->header
);
1241 * __register_sysctl_table - register a leaf sysctl table
1242 * @set: Sysctl tree to register on
1243 * @path: The path to the directory the sysctl table is in.
1244 * @table: the top-level table structure
1246 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1247 * array. A completely 0 filled entry terminates the table.
1249 * The members of the &struct ctl_table structure are used as follows:
1251 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1252 * enter a sysctl file
1254 * data - a pointer to data for use by proc_handler
1256 * maxlen - the maximum size in bytes of the data
1258 * mode - the file permissions for the /proc/sys file
1260 * child - must be %NULL.
1262 * proc_handler - the text handler routine (described below)
1264 * extra1, extra2 - extra pointers usable by the proc handler routines
1266 * Leaf nodes in the sysctl tree will be represented by a single file
1267 * under /proc; non-leaf nodes will be represented by directories.
1269 * There must be a proc_handler routine for any terminal nodes.
1270 * Several default handlers are available to cover common cases -
1272 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1273 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1274 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1276 * It is the handler's job to read the input buffer from user memory
1277 * and process it. The handler should return 0 on success.
1279 * This routine returns %NULL on a failure to register, and a pointer
1280 * to the table header on success.
1282 struct ctl_table_header
*__register_sysctl_table(
1283 struct ctl_table_set
*set
,
1284 const char *path
, struct ctl_table
*table
)
1286 struct ctl_table_root
*root
= set
->dir
.header
.root
;
1287 struct ctl_table_header
*header
;
1288 const char *name
, *nextname
;
1289 struct ctl_dir
*dir
;
1290 struct ctl_table
*entry
;
1291 struct ctl_node
*node
;
1294 for (entry
= table
; entry
->procname
; entry
++)
1297 header
= kzalloc(sizeof(struct ctl_table_header
) +
1298 sizeof(struct ctl_node
)*nr_entries
, GFP_KERNEL
);
1302 node
= (struct ctl_node
*)(header
+ 1);
1303 init_header(header
, root
, set
, node
, table
);
1304 if (sysctl_check_table(path
, table
))
1307 spin_lock(&sysctl_lock
);
1309 /* Reference moved down the diretory tree get_subdir */
1311 spin_unlock(&sysctl_lock
);
1313 /* Find the directory for the ctl_table */
1314 for (name
= path
; name
; name
= nextname
) {
1316 nextname
= strchr(name
, '/');
1318 namelen
= nextname
- name
;
1321 namelen
= strlen(name
);
1326 dir
= get_subdir(dir
, name
, namelen
);
1331 spin_lock(&sysctl_lock
);
1332 if (insert_header(dir
, header
))
1333 goto fail_put_dir_locked
;
1335 drop_sysctl_table(&dir
->header
);
1336 spin_unlock(&sysctl_lock
);
1340 fail_put_dir_locked
:
1341 drop_sysctl_table(&dir
->header
);
1342 spin_unlock(&sysctl_lock
);
1350 * register_sysctl - register a sysctl table
1351 * @path: The path to the directory the sysctl table is in.
1352 * @table: the table structure
1354 * Register a sysctl table. @table should be a filled in ctl_table
1355 * array. A completely 0 filled entry terminates the table.
1357 * See __register_sysctl_table for more details.
1359 struct ctl_table_header
*register_sysctl(const char *path
, struct ctl_table
*table
)
1361 return __register_sysctl_table(&sysctl_table_root
.default_set
,
1364 EXPORT_SYMBOL(register_sysctl
);
1366 static char *append_path(const char *path
, char *pos
, const char *name
)
1369 namelen
= strlen(name
);
1370 if (((pos
- path
) + namelen
+ 2) >= PATH_MAX
)
1372 memcpy(pos
, name
, namelen
);
1374 pos
[namelen
+ 1] = '\0';
1379 static int count_subheaders(struct ctl_table
*table
)
1382 int nr_subheaders
= 0;
1383 struct ctl_table
*entry
;
1385 /* special case: no directory and empty directory */
1386 if (!table
|| !table
->procname
)
1389 for (entry
= table
; entry
->procname
; entry
++) {
1391 nr_subheaders
+= count_subheaders(entry
->child
);
1395 return nr_subheaders
+ has_files
;
1398 static int register_leaf_sysctl_tables(const char *path
, char *pos
,
1399 struct ctl_table_header
***subheader
, struct ctl_table_set
*set
,
1400 struct ctl_table
*table
)
1402 struct ctl_table
*ctl_table_arg
= NULL
;
1403 struct ctl_table
*entry
, *files
;
1408 for (entry
= table
; entry
->procname
; entry
++) {
1416 /* If there are mixed files and directories we need a new table */
1417 if (nr_dirs
&& nr_files
) {
1418 struct ctl_table
*new;
1419 files
= kzalloc(sizeof(struct ctl_table
) * (nr_files
+ 1),
1424 ctl_table_arg
= files
;
1425 for (new = files
, entry
= table
; entry
->procname
; entry
++) {
1433 /* Register everything except a directory full of subdirectories */
1434 if (nr_files
|| !nr_dirs
) {
1435 struct ctl_table_header
*header
;
1436 header
= __register_sysctl_table(set
, path
, files
);
1438 kfree(ctl_table_arg
);
1442 /* Remember if we need to free the file table */
1443 header
->ctl_table_arg
= ctl_table_arg
;
1444 **subheader
= header
;
1448 /* Recurse into the subdirectories. */
1449 for (entry
= table
; entry
->procname
; entry
++) {
1455 err
= -ENAMETOOLONG
;
1456 child_pos
= append_path(path
, pos
, entry
->procname
);
1460 err
= register_leaf_sysctl_tables(path
, child_pos
, subheader
,
1468 /* On failure our caller will unregister all registered subheaders */
1473 * __register_sysctl_paths - register a sysctl table hierarchy
1474 * @set: Sysctl tree to register on
1475 * @path: The path to the directory the sysctl table is in.
1476 * @table: the top-level table structure
1478 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1479 * array. A completely 0 filled entry terminates the table.
1481 * See __register_sysctl_table for more details.
1483 struct ctl_table_header
*__register_sysctl_paths(
1484 struct ctl_table_set
*set
,
1485 const struct ctl_path
*path
, struct ctl_table
*table
)
1487 struct ctl_table
*ctl_table_arg
= table
;
1488 int nr_subheaders
= count_subheaders(table
);
1489 struct ctl_table_header
*header
= NULL
, **subheaders
, **subheader
;
1490 const struct ctl_path
*component
;
1491 char *new_path
, *pos
;
1493 pos
= new_path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1498 for (component
= path
; component
->procname
; component
++) {
1499 pos
= append_path(new_path
, pos
, component
->procname
);
1503 while (table
->procname
&& table
->child
&& !table
[1].procname
) {
1504 pos
= append_path(new_path
, pos
, table
->procname
);
1507 table
= table
->child
;
1509 if (nr_subheaders
== 1) {
1510 header
= __register_sysctl_table(set
, new_path
, table
);
1512 header
->ctl_table_arg
= ctl_table_arg
;
1514 header
= kzalloc(sizeof(*header
) +
1515 sizeof(*subheaders
)*nr_subheaders
, GFP_KERNEL
);
1519 subheaders
= (struct ctl_table_header
**) (header
+ 1);
1520 subheader
= subheaders
;
1521 header
->ctl_table_arg
= ctl_table_arg
;
1523 if (register_leaf_sysctl_tables(new_path
, pos
, &subheader
,
1525 goto err_register_leaves
;
1532 err_register_leaves
:
1533 while (subheader
> subheaders
) {
1534 struct ctl_table_header
*subh
= *(--subheader
);
1535 struct ctl_table
*table
= subh
->ctl_table_arg
;
1536 unregister_sysctl_table(subh
);
1545 * register_sysctl_table_path - register a sysctl table hierarchy
1546 * @path: The path to the directory the sysctl table is in.
1547 * @table: the top-level table structure
1549 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1550 * array. A completely 0 filled entry terminates the table.
1552 * See __register_sysctl_paths for more details.
1554 struct ctl_table_header
*register_sysctl_paths(const struct ctl_path
*path
,
1555 struct ctl_table
*table
)
1557 return __register_sysctl_paths(&sysctl_table_root
.default_set
,
1560 EXPORT_SYMBOL(register_sysctl_paths
);
1563 * register_sysctl_table - register a sysctl table hierarchy
1564 * @table: the top-level table structure
1566 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1567 * array. A completely 0 filled entry terminates the table.
1569 * See register_sysctl_paths for more details.
1571 struct ctl_table_header
*register_sysctl_table(struct ctl_table
*table
)
1573 static const struct ctl_path null_path
[] = { {} };
1575 return register_sysctl_paths(null_path
, table
);
1577 EXPORT_SYMBOL(register_sysctl_table
);
1579 static void put_links(struct ctl_table_header
*header
)
1581 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1582 struct ctl_table_root
*root
= header
->root
;
1583 struct ctl_dir
*parent
= header
->parent
;
1584 struct ctl_dir
*core_parent
;
1585 struct ctl_table
*entry
;
1587 if (header
->set
== root_set
)
1590 core_parent
= xlate_dir(root_set
, parent
);
1591 if (IS_ERR(core_parent
))
1594 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
1595 struct ctl_table_header
*link_head
;
1596 struct ctl_table
*link
;
1597 const char *name
= entry
->procname
;
1599 link
= find_entry(&link_head
, core_parent
, name
, strlen(name
));
1601 ((S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
)) ||
1602 (S_ISLNK(link
->mode
) && (link
->data
== root
)))) {
1603 drop_sysctl_table(link_head
);
1606 pr_err("sysctl link missing during unregister: ");
1607 sysctl_print_dir(parent
);
1608 pr_cont("/%s\n", name
);
1613 static void drop_sysctl_table(struct ctl_table_header
*header
)
1615 struct ctl_dir
*parent
= header
->parent
;
1621 start_unregistering(header
);
1622 if (!--header
->count
)
1623 kfree_rcu(header
, rcu
);
1626 drop_sysctl_table(&parent
->header
);
1630 * unregister_sysctl_table - unregister a sysctl table hierarchy
1631 * @header: the header returned from register_sysctl_table
1633 * Unregisters the sysctl table and all children. proc entries may not
1634 * actually be removed until they are no longer used by anyone.
1636 void unregister_sysctl_table(struct ctl_table_header
* header
)
1644 nr_subheaders
= count_subheaders(header
->ctl_table_arg
);
1645 if (unlikely(nr_subheaders
> 1)) {
1646 struct ctl_table_header
**subheaders
;
1649 subheaders
= (struct ctl_table_header
**)(header
+ 1);
1650 for (i
= nr_subheaders
-1; i
>= 0; i
--) {
1651 struct ctl_table_header
*subh
= subheaders
[i
];
1652 struct ctl_table
*table
= subh
->ctl_table_arg
;
1653 unregister_sysctl_table(subh
);
1660 spin_lock(&sysctl_lock
);
1661 drop_sysctl_table(header
);
1662 spin_unlock(&sysctl_lock
);
1664 EXPORT_SYMBOL(unregister_sysctl_table
);
1666 void setup_sysctl_set(struct ctl_table_set
*set
,
1667 struct ctl_table_root
*root
,
1668 int (*is_seen
)(struct ctl_table_set
*))
1670 memset(set
, 0, sizeof(*set
));
1671 set
->is_seen
= is_seen
;
1672 init_header(&set
->dir
.header
, root
, set
, NULL
, root_table
);
1675 void retire_sysctl_set(struct ctl_table_set
*set
)
1677 WARN_ON(!RB_EMPTY_ROOT(&set
->dir
.root
));
1680 int __init
proc_sys_init(void)
1682 struct proc_dir_entry
*proc_sys_root
;
1684 proc_sys_root
= proc_mkdir("sys", NULL
);
1685 proc_sys_root
->proc_iops
= &proc_sys_dir_operations
;
1686 proc_sys_root
->proc_fops
= &proc_sys_dir_file_operations
;
1687 proc_sys_root
->nlink
= 0;
1689 return sysctl_init();