1 // SPDX-License-Identifier: GPL-2.0
5 #include <linux/init.h>
6 #include <linux/sysctl.h>
7 #include <linux/poll.h>
8 #include <linux/proc_fs.h>
9 #include <linux/printk.h>
10 #include <linux/security.h>
11 #include <linux/sched.h>
12 #include <linux/cred.h>
13 #include <linux/namei.h>
15 #include <linux/module.h>
16 #include <linux/bpf-cgroup.h>
19 static const struct dentry_operations proc_sys_dentry_operations
;
20 static const struct file_operations proc_sys_file_operations
;
21 static const struct inode_operations proc_sys_inode_operations
;
22 static const struct file_operations proc_sys_dir_file_operations
;
23 static const struct inode_operations proc_sys_dir_operations
;
25 /* shared constants to be used in various sysctls */
26 const int sysctl_vals
[] = { 0, 1, INT_MAX
};
27 EXPORT_SYMBOL(sysctl_vals
);
29 /* Support for permanently empty directories */
31 struct ctl_table sysctl_mount_point
[] = {
35 static bool is_empty_dir(struct ctl_table_header
*head
)
37 return head
->ctl_table
[0].child
== sysctl_mount_point
;
40 static void set_empty_dir(struct ctl_dir
*dir
)
42 dir
->header
.ctl_table
[0].child
= sysctl_mount_point
;
45 static void clear_empty_dir(struct ctl_dir
*dir
)
48 dir
->header
.ctl_table
[0].child
= NULL
;
51 void proc_sys_poll_notify(struct ctl_table_poll
*poll
)
56 atomic_inc(&poll
->event
);
57 wake_up_interruptible(&poll
->wait
);
60 static struct ctl_table root_table
[] = {
63 .mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
,
67 static struct ctl_table_root sysctl_table_root
= {
68 .default_set
.dir
.header
= {
71 .ctl_table
= root_table
}},
72 .ctl_table_arg
= root_table
,
73 .root
= &sysctl_table_root
,
74 .set
= &sysctl_table_root
.default_set
,
78 static DEFINE_SPINLOCK(sysctl_lock
);
80 static void drop_sysctl_table(struct ctl_table_header
*header
);
81 static int sysctl_follow_link(struct ctl_table_header
**phead
,
82 struct ctl_table
**pentry
);
83 static int insert_links(struct ctl_table_header
*head
);
84 static void put_links(struct ctl_table_header
*header
);
86 static void sysctl_print_dir(struct ctl_dir
*dir
)
88 if (dir
->header
.parent
)
89 sysctl_print_dir(dir
->header
.parent
);
90 pr_cont("%s/", dir
->header
.ctl_table
[0].procname
);
93 static int namecmp(const char *name1
, int len1
, const char *name2
, int len2
)
102 cmp
= memcmp(name1
, name2
, minlen
);
108 /* Called under sysctl_lock */
109 static struct ctl_table
*find_entry(struct ctl_table_header
**phead
,
110 struct ctl_dir
*dir
, const char *name
, int namelen
)
112 struct ctl_table_header
*head
;
113 struct ctl_table
*entry
;
114 struct rb_node
*node
= dir
->root
.rb_node
;
118 struct ctl_node
*ctl_node
;
119 const char *procname
;
122 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
123 head
= ctl_node
->header
;
124 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
125 procname
= entry
->procname
;
127 cmp
= namecmp(name
, namelen
, procname
, strlen(procname
));
129 node
= node
->rb_left
;
131 node
= node
->rb_right
;
140 static int insert_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
142 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
143 struct rb_node
**p
= &head
->parent
->root
.rb_node
;
144 struct rb_node
*parent
= NULL
;
145 const char *name
= entry
->procname
;
146 int namelen
= strlen(name
);
149 struct ctl_table_header
*parent_head
;
150 struct ctl_table
*parent_entry
;
151 struct ctl_node
*parent_node
;
152 const char *parent_name
;
156 parent_node
= rb_entry(parent
, struct ctl_node
, node
);
157 parent_head
= parent_node
->header
;
158 parent_entry
= &parent_head
->ctl_table
[parent_node
- parent_head
->node
];
159 parent_name
= parent_entry
->procname
;
161 cmp
= namecmp(name
, namelen
, parent_name
, strlen(parent_name
));
167 pr_err("sysctl duplicate entry: ");
168 sysctl_print_dir(head
->parent
);
169 pr_cont("/%s\n", entry
->procname
);
174 rb_link_node(node
, parent
, p
);
175 rb_insert_color(node
, &head
->parent
->root
);
179 static void erase_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
181 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
183 rb_erase(node
, &head
->parent
->root
);
186 static void init_header(struct ctl_table_header
*head
,
187 struct ctl_table_root
*root
, struct ctl_table_set
*set
,
188 struct ctl_node
*node
, struct ctl_table
*table
)
190 head
->ctl_table
= table
;
191 head
->ctl_table_arg
= table
;
195 head
->unregistering
= NULL
;
200 INIT_HLIST_HEAD(&head
->inodes
);
202 struct ctl_table
*entry
;
203 for (entry
= table
; entry
->procname
; entry
++, node
++)
208 static void erase_header(struct ctl_table_header
*head
)
210 struct ctl_table
*entry
;
211 for (entry
= head
->ctl_table
; entry
->procname
; entry
++)
212 erase_entry(head
, entry
);
215 static int insert_header(struct ctl_dir
*dir
, struct ctl_table_header
*header
)
217 struct ctl_table
*entry
;
220 /* Is this a permanently empty directory? */
221 if (is_empty_dir(&dir
->header
))
224 /* Am I creating a permanently empty directory? */
225 if (header
->ctl_table
== sysctl_mount_point
) {
226 if (!RB_EMPTY_ROOT(&dir
->root
))
232 header
->parent
= dir
;
233 err
= insert_links(header
);
236 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
237 err
= insert_entry(header
, entry
);
243 erase_header(header
);
246 if (header
->ctl_table
== sysctl_mount_point
)
247 clear_empty_dir(dir
);
248 header
->parent
= NULL
;
249 drop_sysctl_table(&dir
->header
);
253 /* called under sysctl_lock */
254 static int use_table(struct ctl_table_header
*p
)
256 if (unlikely(p
->unregistering
))
262 /* called under sysctl_lock */
263 static void unuse_table(struct ctl_table_header
*p
)
266 if (unlikely(p
->unregistering
))
267 complete(p
->unregistering
);
270 static void proc_sys_prune_dcache(struct ctl_table_header
*head
)
273 struct proc_inode
*ei
;
274 struct hlist_node
*node
;
275 struct super_block
*sb
;
279 node
= hlist_first_rcu(&head
->inodes
);
282 ei
= hlist_entry(node
, struct proc_inode
, sysctl_inodes
);
283 spin_lock(&sysctl_lock
);
284 hlist_del_init_rcu(&ei
->sysctl_inodes
);
285 spin_unlock(&sysctl_lock
);
287 inode
= &ei
->vfs_inode
;
289 if (!atomic_inc_not_zero(&sb
->s_active
))
291 inode
= igrab(inode
);
293 if (unlikely(!inode
)) {
294 deactivate_super(sb
);
299 d_prune_aliases(inode
);
301 deactivate_super(sb
);
308 /* called under sysctl_lock, will reacquire if has to wait */
309 static void start_unregistering(struct ctl_table_header
*p
)
312 * if p->used is 0, nobody will ever touch that entry again;
313 * we'll eliminate all paths to it before dropping sysctl_lock
315 if (unlikely(p
->used
)) {
316 struct completion wait
;
317 init_completion(&wait
);
318 p
->unregistering
= &wait
;
319 spin_unlock(&sysctl_lock
);
320 wait_for_completion(&wait
);
322 /* anything non-NULL; we'll never dereference it */
323 p
->unregistering
= ERR_PTR(-EINVAL
);
324 spin_unlock(&sysctl_lock
);
327 * Prune dentries for unregistered sysctls: namespaced sysctls
328 * can have duplicate names and contaminate dcache very badly.
330 proc_sys_prune_dcache(p
);
332 * do not remove from the list until nobody holds it; walking the
333 * list in do_sysctl() relies on that.
335 spin_lock(&sysctl_lock
);
339 static struct ctl_table_header
*sysctl_head_grab(struct ctl_table_header
*head
)
342 spin_lock(&sysctl_lock
);
343 if (!use_table(head
))
344 head
= ERR_PTR(-ENOENT
);
345 spin_unlock(&sysctl_lock
);
349 static void sysctl_head_finish(struct ctl_table_header
*head
)
353 spin_lock(&sysctl_lock
);
355 spin_unlock(&sysctl_lock
);
358 static struct ctl_table_set
*
359 lookup_header_set(struct ctl_table_root
*root
)
361 struct ctl_table_set
*set
= &root
->default_set
;
363 set
= root
->lookup(root
);
367 static struct ctl_table
*lookup_entry(struct ctl_table_header
**phead
,
369 const char *name
, int namelen
)
371 struct ctl_table_header
*head
;
372 struct ctl_table
*entry
;
374 spin_lock(&sysctl_lock
);
375 entry
= find_entry(&head
, dir
, name
, namelen
);
376 if (entry
&& use_table(head
))
380 spin_unlock(&sysctl_lock
);
384 static struct ctl_node
*first_usable_entry(struct rb_node
*node
)
386 struct ctl_node
*ctl_node
;
388 for (;node
; node
= rb_next(node
)) {
389 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
390 if (use_table(ctl_node
->header
))
396 static void first_entry(struct ctl_dir
*dir
,
397 struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
399 struct ctl_table_header
*head
= NULL
;
400 struct ctl_table
*entry
= NULL
;
401 struct ctl_node
*ctl_node
;
403 spin_lock(&sysctl_lock
);
404 ctl_node
= first_usable_entry(rb_first(&dir
->root
));
405 spin_unlock(&sysctl_lock
);
407 head
= ctl_node
->header
;
408 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
414 static void next_entry(struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
416 struct ctl_table_header
*head
= *phead
;
417 struct ctl_table
*entry
= *pentry
;
418 struct ctl_node
*ctl_node
= &head
->node
[entry
- head
->ctl_table
];
420 spin_lock(&sysctl_lock
);
423 ctl_node
= first_usable_entry(rb_next(&ctl_node
->node
));
424 spin_unlock(&sysctl_lock
);
427 head
= ctl_node
->header
;
428 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
435 * sysctl_perm does NOT grant the superuser all rights automatically, because
436 * some sysctl variables are readonly even to root.
439 static int test_perm(int mode
, int op
)
441 if (uid_eq(current_euid(), GLOBAL_ROOT_UID
))
443 else if (in_egroup_p(GLOBAL_ROOT_GID
))
445 if ((op
& ~mode
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
)) == 0)
450 static int sysctl_perm(struct ctl_table_header
*head
, struct ctl_table
*table
, int op
)
452 struct ctl_table_root
*root
= head
->root
;
455 if (root
->permissions
)
456 mode
= root
->permissions(head
, table
);
460 return test_perm(mode
, op
);
463 static struct inode
*proc_sys_make_inode(struct super_block
*sb
,
464 struct ctl_table_header
*head
, struct ctl_table
*table
)
466 struct ctl_table_root
*root
= head
->root
;
468 struct proc_inode
*ei
;
470 inode
= new_inode(sb
);
472 return ERR_PTR(-ENOMEM
);
474 inode
->i_ino
= get_next_ino();
478 spin_lock(&sysctl_lock
);
479 if (unlikely(head
->unregistering
)) {
480 spin_unlock(&sysctl_lock
);
482 return ERR_PTR(-ENOENT
);
485 ei
->sysctl_entry
= table
;
486 hlist_add_head_rcu(&ei
->sysctl_inodes
, &head
->inodes
);
488 spin_unlock(&sysctl_lock
);
490 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= current_time(inode
);
491 inode
->i_mode
= table
->mode
;
492 if (!S_ISDIR(table
->mode
)) {
493 inode
->i_mode
|= S_IFREG
;
494 inode
->i_op
= &proc_sys_inode_operations
;
495 inode
->i_fop
= &proc_sys_file_operations
;
497 inode
->i_mode
|= S_IFDIR
;
498 inode
->i_op
= &proc_sys_dir_operations
;
499 inode
->i_fop
= &proc_sys_dir_file_operations
;
500 if (is_empty_dir(head
))
501 make_empty_dir_inode(inode
);
504 if (root
->set_ownership
)
505 root
->set_ownership(head
, table
, &inode
->i_uid
, &inode
->i_gid
);
507 inode
->i_uid
= GLOBAL_ROOT_UID
;
508 inode
->i_gid
= GLOBAL_ROOT_GID
;
514 void proc_sys_evict_inode(struct inode
*inode
, struct ctl_table_header
*head
)
516 spin_lock(&sysctl_lock
);
517 hlist_del_init_rcu(&PROC_I(inode
)->sysctl_inodes
);
519 kfree_rcu(head
, rcu
);
520 spin_unlock(&sysctl_lock
);
523 static struct ctl_table_header
*grab_header(struct inode
*inode
)
525 struct ctl_table_header
*head
= PROC_I(inode
)->sysctl
;
527 head
= &sysctl_table_root
.default_set
.dir
.header
;
528 return sysctl_head_grab(head
);
531 static struct dentry
*proc_sys_lookup(struct inode
*dir
, struct dentry
*dentry
,
534 struct ctl_table_header
*head
= grab_header(dir
);
535 struct ctl_table_header
*h
= NULL
;
536 const struct qstr
*name
= &dentry
->d_name
;
539 struct dentry
*err
= ERR_PTR(-ENOENT
);
540 struct ctl_dir
*ctl_dir
;
544 return ERR_CAST(head
);
546 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
548 p
= lookup_entry(&h
, ctl_dir
, name
->name
, name
->len
);
552 if (S_ISLNK(p
->mode
)) {
553 ret
= sysctl_follow_link(&h
, &p
);
559 inode
= proc_sys_make_inode(dir
->i_sb
, h
? h
: head
, p
);
561 err
= ERR_CAST(inode
);
565 d_set_d_op(dentry
, &proc_sys_dentry_operations
);
566 err
= d_splice_alias(inode
, dentry
);
570 sysctl_head_finish(h
);
571 sysctl_head_finish(head
);
575 static ssize_t
proc_sys_call_handler(struct file
*filp
, void __user
*buf
,
576 size_t count
, loff_t
*ppos
, int write
)
578 struct inode
*inode
= file_inode(filp
);
579 struct ctl_table_header
*head
= grab_header(inode
);
580 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
581 void *new_buf
= NULL
;
585 return PTR_ERR(head
);
588 * At this point we know that the sysctl was not unregistered
589 * and won't be until we finish.
592 if (sysctl_perm(head
, table
, write
? MAY_WRITE
: MAY_READ
))
595 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
597 if (!table
->proc_handler
)
600 error
= BPF_CGROUP_RUN_PROG_SYSCTL(head
, table
, write
, buf
, &count
,
605 /* careful: calling conventions are nasty here */
611 error
= table
->proc_handler(table
, write
, (void __user
*)new_buf
,
616 error
= table
->proc_handler(table
, write
, buf
, &count
, ppos
);
622 sysctl_head_finish(head
);
627 static ssize_t
proc_sys_read(struct file
*filp
, char __user
*buf
,
628 size_t count
, loff_t
*ppos
)
630 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 0);
633 static ssize_t
proc_sys_write(struct file
*filp
, const char __user
*buf
,
634 size_t count
, loff_t
*ppos
)
636 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 1);
639 static int proc_sys_open(struct inode
*inode
, struct file
*filp
)
641 struct ctl_table_header
*head
= grab_header(inode
);
642 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
644 /* sysctl was unregistered */
646 return PTR_ERR(head
);
649 filp
->private_data
= proc_sys_poll_event(table
->poll
);
651 sysctl_head_finish(head
);
656 static __poll_t
proc_sys_poll(struct file
*filp
, poll_table
*wait
)
658 struct inode
*inode
= file_inode(filp
);
659 struct ctl_table_header
*head
= grab_header(inode
);
660 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
661 __poll_t ret
= DEFAULT_POLLMASK
;
664 /* sysctl was unregistered */
666 return EPOLLERR
| EPOLLHUP
;
668 if (!table
->proc_handler
)
674 event
= (unsigned long)filp
->private_data
;
675 poll_wait(filp
, &table
->poll
->wait
, wait
);
677 if (event
!= atomic_read(&table
->poll
->event
)) {
678 filp
->private_data
= proc_sys_poll_event(table
->poll
);
679 ret
= EPOLLIN
| EPOLLRDNORM
| EPOLLERR
| EPOLLPRI
;
683 sysctl_head_finish(head
);
688 static bool proc_sys_fill_cache(struct file
*file
,
689 struct dir_context
*ctx
,
690 struct ctl_table_header
*head
,
691 struct ctl_table
*table
)
693 struct dentry
*child
, *dir
= file
->f_path
.dentry
;
697 unsigned type
= DT_UNKNOWN
;
699 qname
.name
= table
->procname
;
700 qname
.len
= strlen(table
->procname
);
701 qname
.hash
= full_name_hash(dir
, qname
.name
, qname
.len
);
703 child
= d_lookup(dir
, &qname
);
705 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
706 child
= d_alloc_parallel(dir
, &qname
, &wq
);
709 if (d_in_lookup(child
)) {
711 inode
= proc_sys_make_inode(dir
->d_sb
, head
, table
);
713 d_lookup_done(child
);
717 d_set_d_op(child
, &proc_sys_dentry_operations
);
718 res
= d_splice_alias(inode
, child
);
719 d_lookup_done(child
);
730 inode
= d_inode(child
);
732 type
= inode
->i_mode
>> 12;
734 return dir_emit(ctx
, qname
.name
, qname
.len
, ino
, type
);
737 static bool proc_sys_link_fill_cache(struct file
*file
,
738 struct dir_context
*ctx
,
739 struct ctl_table_header
*head
,
740 struct ctl_table
*table
)
744 head
= sysctl_head_grab(head
);
748 /* It is not an error if we can not follow the link ignore it */
749 if (sysctl_follow_link(&head
, &table
))
752 ret
= proc_sys_fill_cache(file
, ctx
, head
, table
);
754 sysctl_head_finish(head
);
758 static int scan(struct ctl_table_header
*head
, struct ctl_table
*table
,
759 unsigned long *pos
, struct file
*file
,
760 struct dir_context
*ctx
)
764 if ((*pos
)++ < ctx
->pos
)
767 if (unlikely(S_ISLNK(table
->mode
)))
768 res
= proc_sys_link_fill_cache(file
, ctx
, head
, table
);
770 res
= proc_sys_fill_cache(file
, ctx
, head
, table
);
778 static int proc_sys_readdir(struct file
*file
, struct dir_context
*ctx
)
780 struct ctl_table_header
*head
= grab_header(file_inode(file
));
781 struct ctl_table_header
*h
= NULL
;
782 struct ctl_table
*entry
;
783 struct ctl_dir
*ctl_dir
;
787 return PTR_ERR(head
);
789 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
791 if (!dir_emit_dots(file
, ctx
))
796 for (first_entry(ctl_dir
, &h
, &entry
); h
; next_entry(&h
, &entry
)) {
797 if (!scan(h
, entry
, &pos
, file
, ctx
)) {
798 sysctl_head_finish(h
);
803 sysctl_head_finish(head
);
807 static int proc_sys_permission(struct inode
*inode
, int mask
)
810 * sysctl entries that are not writeable,
811 * are _NOT_ writeable, capabilities or not.
813 struct ctl_table_header
*head
;
814 struct ctl_table
*table
;
817 /* Executable files are not allowed under /proc/sys/ */
818 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
))
821 head
= grab_header(inode
);
823 return PTR_ERR(head
);
825 table
= PROC_I(inode
)->sysctl_entry
;
826 if (!table
) /* global root - r-xr-xr-x */
827 error
= mask
& MAY_WRITE
? -EACCES
: 0;
828 else /* Use the permissions on the sysctl table entry */
829 error
= sysctl_perm(head
, table
, mask
& ~MAY_NOT_BLOCK
);
831 sysctl_head_finish(head
);
835 static int proc_sys_setattr(struct dentry
*dentry
, struct iattr
*attr
)
837 struct inode
*inode
= d_inode(dentry
);
840 if (attr
->ia_valid
& (ATTR_MODE
| ATTR_UID
| ATTR_GID
))
843 error
= setattr_prepare(dentry
, attr
);
847 setattr_copy(inode
, attr
);
848 mark_inode_dirty(inode
);
852 static int proc_sys_getattr(const struct path
*path
, struct kstat
*stat
,
853 u32 request_mask
, unsigned int query_flags
)
855 struct inode
*inode
= d_inode(path
->dentry
);
856 struct ctl_table_header
*head
= grab_header(inode
);
857 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
860 return PTR_ERR(head
);
862 generic_fillattr(inode
, stat
);
864 stat
->mode
= (stat
->mode
& S_IFMT
) | table
->mode
;
866 sysctl_head_finish(head
);
870 static const struct file_operations proc_sys_file_operations
= {
871 .open
= proc_sys_open
,
872 .poll
= proc_sys_poll
,
873 .read
= proc_sys_read
,
874 .write
= proc_sys_write
,
875 .llseek
= default_llseek
,
878 static const struct file_operations proc_sys_dir_file_operations
= {
879 .read
= generic_read_dir
,
880 .iterate_shared
= proc_sys_readdir
,
881 .llseek
= generic_file_llseek
,
884 static const struct inode_operations proc_sys_inode_operations
= {
885 .permission
= proc_sys_permission
,
886 .setattr
= proc_sys_setattr
,
887 .getattr
= proc_sys_getattr
,
890 static const struct inode_operations proc_sys_dir_operations
= {
891 .lookup
= proc_sys_lookup
,
892 .permission
= proc_sys_permission
,
893 .setattr
= proc_sys_setattr
,
894 .getattr
= proc_sys_getattr
,
897 static int proc_sys_revalidate(struct dentry
*dentry
, unsigned int flags
)
899 if (flags
& LOOKUP_RCU
)
901 return !PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
904 static int proc_sys_delete(const struct dentry
*dentry
)
906 return !!PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
909 static int sysctl_is_seen(struct ctl_table_header
*p
)
911 struct ctl_table_set
*set
= p
->set
;
913 spin_lock(&sysctl_lock
);
914 if (p
->unregistering
)
916 else if (!set
->is_seen
)
919 res
= set
->is_seen(set
);
920 spin_unlock(&sysctl_lock
);
924 static int proc_sys_compare(const struct dentry
*dentry
,
925 unsigned int len
, const char *str
, const struct qstr
*name
)
927 struct ctl_table_header
*head
;
930 /* Although proc doesn't have negative dentries, rcu-walk means
931 * that inode here can be NULL */
932 /* AV: can it, indeed? */
933 inode
= d_inode_rcu(dentry
);
936 if (name
->len
!= len
)
938 if (memcmp(name
->name
, str
, len
))
940 head
= rcu_dereference(PROC_I(inode
)->sysctl
);
941 return !head
|| !sysctl_is_seen(head
);
944 static const struct dentry_operations proc_sys_dentry_operations
= {
945 .d_revalidate
= proc_sys_revalidate
,
946 .d_delete
= proc_sys_delete
,
947 .d_compare
= proc_sys_compare
,
950 static struct ctl_dir
*find_subdir(struct ctl_dir
*dir
,
951 const char *name
, int namelen
)
953 struct ctl_table_header
*head
;
954 struct ctl_table
*entry
;
956 entry
= find_entry(&head
, dir
, name
, namelen
);
958 return ERR_PTR(-ENOENT
);
959 if (!S_ISDIR(entry
->mode
))
960 return ERR_PTR(-ENOTDIR
);
961 return container_of(head
, struct ctl_dir
, header
);
964 static struct ctl_dir
*new_dir(struct ctl_table_set
*set
,
965 const char *name
, int namelen
)
967 struct ctl_table
*table
;
969 struct ctl_node
*node
;
972 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node
) +
973 sizeof(struct ctl_table
)*2 + namelen
+ 1,
978 node
= (struct ctl_node
*)(new + 1);
979 table
= (struct ctl_table
*)(node
+ 1);
980 new_name
= (char *)(table
+ 2);
981 memcpy(new_name
, name
, namelen
);
982 new_name
[namelen
] = '\0';
983 table
[0].procname
= new_name
;
984 table
[0].mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
;
985 init_header(&new->header
, set
->dir
.header
.root
, set
, node
, table
);
991 * get_subdir - find or create a subdir with the specified name.
992 * @dir: Directory to create the subdirectory in
993 * @name: The name of the subdirectory to find or create
994 * @namelen: The length of name
996 * Takes a directory with an elevated reference count so we know that
997 * if we drop the lock the directory will not go away. Upon success
998 * the reference is moved from @dir to the returned subdirectory.
999 * Upon error an error code is returned and the reference on @dir is
1002 static struct ctl_dir
*get_subdir(struct ctl_dir
*dir
,
1003 const char *name
, int namelen
)
1005 struct ctl_table_set
*set
= dir
->header
.set
;
1006 struct ctl_dir
*subdir
, *new = NULL
;
1009 spin_lock(&sysctl_lock
);
1010 subdir
= find_subdir(dir
, name
, namelen
);
1011 if (!IS_ERR(subdir
))
1013 if (PTR_ERR(subdir
) != -ENOENT
)
1016 spin_unlock(&sysctl_lock
);
1017 new = new_dir(set
, name
, namelen
);
1018 spin_lock(&sysctl_lock
);
1019 subdir
= ERR_PTR(-ENOMEM
);
1023 /* Was the subdir added while we dropped the lock? */
1024 subdir
= find_subdir(dir
, name
, namelen
);
1025 if (!IS_ERR(subdir
))
1027 if (PTR_ERR(subdir
) != -ENOENT
)
1030 /* Nope. Use the our freshly made directory entry. */
1031 err
= insert_header(dir
, &new->header
);
1032 subdir
= ERR_PTR(err
);
1037 subdir
->header
.nreg
++;
1039 if (IS_ERR(subdir
)) {
1040 pr_err("sysctl could not get directory: ");
1041 sysctl_print_dir(dir
);
1042 pr_cont("/%*.*s %ld\n",
1043 namelen
, namelen
, name
, PTR_ERR(subdir
));
1045 drop_sysctl_table(&dir
->header
);
1047 drop_sysctl_table(&new->header
);
1048 spin_unlock(&sysctl_lock
);
1052 static struct ctl_dir
*xlate_dir(struct ctl_table_set
*set
, struct ctl_dir
*dir
)
1054 struct ctl_dir
*parent
;
1055 const char *procname
;
1056 if (!dir
->header
.parent
)
1058 parent
= xlate_dir(set
, dir
->header
.parent
);
1061 procname
= dir
->header
.ctl_table
[0].procname
;
1062 return find_subdir(parent
, procname
, strlen(procname
));
1065 static int sysctl_follow_link(struct ctl_table_header
**phead
,
1066 struct ctl_table
**pentry
)
1068 struct ctl_table_header
*head
;
1069 struct ctl_table_root
*root
;
1070 struct ctl_table_set
*set
;
1071 struct ctl_table
*entry
;
1072 struct ctl_dir
*dir
;
1076 spin_lock(&sysctl_lock
);
1077 root
= (*pentry
)->data
;
1078 set
= lookup_header_set(root
);
1079 dir
= xlate_dir(set
, (*phead
)->parent
);
1083 const char *procname
= (*pentry
)->procname
;
1085 entry
= find_entry(&head
, dir
, procname
, strlen(procname
));
1087 if (entry
&& use_table(head
)) {
1088 unuse_table(*phead
);
1095 spin_unlock(&sysctl_lock
);
1099 static int sysctl_err(const char *path
, struct ctl_table
*table
, char *fmt
, ...)
1101 struct va_format vaf
;
1104 va_start(args
, fmt
);
1108 pr_err("sysctl table check failed: %s/%s %pV\n",
1109 path
, table
->procname
, &vaf
);
1115 static int sysctl_check_table_array(const char *path
, struct ctl_table
*table
)
1119 if ((table
->proc_handler
== proc_douintvec
) ||
1120 (table
->proc_handler
== proc_douintvec_minmax
)) {
1121 if (table
->maxlen
!= sizeof(unsigned int))
1122 err
|= sysctl_err(path
, table
, "array not allowed");
1128 static int sysctl_check_table(const char *path
, struct ctl_table
*table
)
1131 for (; table
->procname
; table
++) {
1133 err
|= sysctl_err(path
, table
, "Not a file");
1135 if ((table
->proc_handler
== proc_dostring
) ||
1136 (table
->proc_handler
== proc_dointvec
) ||
1137 (table
->proc_handler
== proc_douintvec
) ||
1138 (table
->proc_handler
== proc_douintvec_minmax
) ||
1139 (table
->proc_handler
== proc_dointvec_minmax
) ||
1140 (table
->proc_handler
== proc_dointvec_jiffies
) ||
1141 (table
->proc_handler
== proc_dointvec_userhz_jiffies
) ||
1142 (table
->proc_handler
== proc_dointvec_ms_jiffies
) ||
1143 (table
->proc_handler
== proc_doulongvec_minmax
) ||
1144 (table
->proc_handler
== proc_doulongvec_ms_jiffies_minmax
)) {
1146 err
|= sysctl_err(path
, table
, "No data");
1148 err
|= sysctl_err(path
, table
, "No maxlen");
1150 err
|= sysctl_check_table_array(path
, table
);
1152 if (!table
->proc_handler
)
1153 err
|= sysctl_err(path
, table
, "No proc_handler");
1155 if ((table
->mode
& (S_IRUGO
|S_IWUGO
)) != table
->mode
)
1156 err
|= sysctl_err(path
, table
, "bogus .mode 0%o",
1162 static struct ctl_table_header
*new_links(struct ctl_dir
*dir
, struct ctl_table
*table
,
1163 struct ctl_table_root
*link_root
)
1165 struct ctl_table
*link_table
, *entry
, *link
;
1166 struct ctl_table_header
*links
;
1167 struct ctl_node
*node
;
1169 int nr_entries
, name_bytes
;
1173 for (entry
= table
; entry
->procname
; entry
++) {
1175 name_bytes
+= strlen(entry
->procname
) + 1;
1178 links
= kzalloc(sizeof(struct ctl_table_header
) +
1179 sizeof(struct ctl_node
)*nr_entries
+
1180 sizeof(struct ctl_table
)*(nr_entries
+ 1) +
1187 node
= (struct ctl_node
*)(links
+ 1);
1188 link_table
= (struct ctl_table
*)(node
+ nr_entries
);
1189 link_name
= (char *)&link_table
[nr_entries
+ 1];
1191 for (link
= link_table
, entry
= table
; entry
->procname
; link
++, entry
++) {
1192 int len
= strlen(entry
->procname
) + 1;
1193 memcpy(link_name
, entry
->procname
, len
);
1194 link
->procname
= link_name
;
1195 link
->mode
= S_IFLNK
|S_IRWXUGO
;
1196 link
->data
= link_root
;
1199 init_header(links
, dir
->header
.root
, dir
->header
.set
, node
, link_table
);
1200 links
->nreg
= nr_entries
;
1205 static bool get_links(struct ctl_dir
*dir
,
1206 struct ctl_table
*table
, struct ctl_table_root
*link_root
)
1208 struct ctl_table_header
*head
;
1209 struct ctl_table
*entry
, *link
;
1211 /* Are there links available for every entry in table? */
1212 for (entry
= table
; entry
->procname
; entry
++) {
1213 const char *procname
= entry
->procname
;
1214 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1217 if (S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
))
1219 if (S_ISLNK(link
->mode
) && (link
->data
== link_root
))
1224 /* The checks passed. Increase the registration count on the links */
1225 for (entry
= table
; entry
->procname
; entry
++) {
1226 const char *procname
= entry
->procname
;
1227 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1233 static int insert_links(struct ctl_table_header
*head
)
1235 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1236 struct ctl_dir
*core_parent
= NULL
;
1237 struct ctl_table_header
*links
;
1240 if (head
->set
== root_set
)
1243 core_parent
= xlate_dir(root_set
, head
->parent
);
1244 if (IS_ERR(core_parent
))
1247 if (get_links(core_parent
, head
->ctl_table
, head
->root
))
1250 core_parent
->header
.nreg
++;
1251 spin_unlock(&sysctl_lock
);
1253 links
= new_links(core_parent
, head
->ctl_table
, head
->root
);
1255 spin_lock(&sysctl_lock
);
1261 if (get_links(core_parent
, head
->ctl_table
, head
->root
)) {
1266 err
= insert_header(core_parent
, links
);
1270 drop_sysctl_table(&core_parent
->header
);
1275 * __register_sysctl_table - register a leaf sysctl table
1276 * @set: Sysctl tree to register on
1277 * @path: The path to the directory the sysctl table is in.
1278 * @table: the top-level table structure
1280 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1281 * array. A completely 0 filled entry terminates the table.
1283 * The members of the &struct ctl_table structure are used as follows:
1285 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1286 * enter a sysctl file
1288 * data - a pointer to data for use by proc_handler
1290 * maxlen - the maximum size in bytes of the data
1292 * mode - the file permissions for the /proc/sys file
1294 * child - must be %NULL.
1296 * proc_handler - the text handler routine (described below)
1298 * extra1, extra2 - extra pointers usable by the proc handler routines
1300 * Leaf nodes in the sysctl tree will be represented by a single file
1301 * under /proc; non-leaf nodes will be represented by directories.
1303 * There must be a proc_handler routine for any terminal nodes.
1304 * Several default handlers are available to cover common cases -
1306 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1307 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1308 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1310 * It is the handler's job to read the input buffer from user memory
1311 * and process it. The handler should return 0 on success.
1313 * This routine returns %NULL on a failure to register, and a pointer
1314 * to the table header on success.
1316 struct ctl_table_header
*__register_sysctl_table(
1317 struct ctl_table_set
*set
,
1318 const char *path
, struct ctl_table
*table
)
1320 struct ctl_table_root
*root
= set
->dir
.header
.root
;
1321 struct ctl_table_header
*header
;
1322 const char *name
, *nextname
;
1323 struct ctl_dir
*dir
;
1324 struct ctl_table
*entry
;
1325 struct ctl_node
*node
;
1328 for (entry
= table
; entry
->procname
; entry
++)
1331 header
= kzalloc(sizeof(struct ctl_table_header
) +
1332 sizeof(struct ctl_node
)*nr_entries
, GFP_KERNEL
);
1336 node
= (struct ctl_node
*)(header
+ 1);
1337 init_header(header
, root
, set
, node
, table
);
1338 if (sysctl_check_table(path
, table
))
1341 spin_lock(&sysctl_lock
);
1343 /* Reference moved down the diretory tree get_subdir */
1345 spin_unlock(&sysctl_lock
);
1347 /* Find the directory for the ctl_table */
1348 for (name
= path
; name
; name
= nextname
) {
1350 nextname
= strchr(name
, '/');
1352 namelen
= nextname
- name
;
1355 namelen
= strlen(name
);
1360 dir
= get_subdir(dir
, name
, namelen
);
1365 spin_lock(&sysctl_lock
);
1366 if (insert_header(dir
, header
))
1367 goto fail_put_dir_locked
;
1369 drop_sysctl_table(&dir
->header
);
1370 spin_unlock(&sysctl_lock
);
1374 fail_put_dir_locked
:
1375 drop_sysctl_table(&dir
->header
);
1376 spin_unlock(&sysctl_lock
);
1384 * register_sysctl - register a sysctl table
1385 * @path: The path to the directory the sysctl table is in.
1386 * @table: the table structure
1388 * Register a sysctl table. @table should be a filled in ctl_table
1389 * array. A completely 0 filled entry terminates the table.
1391 * See __register_sysctl_table for more details.
1393 struct ctl_table_header
*register_sysctl(const char *path
, struct ctl_table
*table
)
1395 return __register_sysctl_table(&sysctl_table_root
.default_set
,
1398 EXPORT_SYMBOL(register_sysctl
);
1400 static char *append_path(const char *path
, char *pos
, const char *name
)
1403 namelen
= strlen(name
);
1404 if (((pos
- path
) + namelen
+ 2) >= PATH_MAX
)
1406 memcpy(pos
, name
, namelen
);
1408 pos
[namelen
+ 1] = '\0';
1413 static int count_subheaders(struct ctl_table
*table
)
1416 int nr_subheaders
= 0;
1417 struct ctl_table
*entry
;
1419 /* special case: no directory and empty directory */
1420 if (!table
|| !table
->procname
)
1423 for (entry
= table
; entry
->procname
; entry
++) {
1425 nr_subheaders
+= count_subheaders(entry
->child
);
1429 return nr_subheaders
+ has_files
;
1432 static int register_leaf_sysctl_tables(const char *path
, char *pos
,
1433 struct ctl_table_header
***subheader
, struct ctl_table_set
*set
,
1434 struct ctl_table
*table
)
1436 struct ctl_table
*ctl_table_arg
= NULL
;
1437 struct ctl_table
*entry
, *files
;
1442 for (entry
= table
; entry
->procname
; entry
++) {
1450 /* If there are mixed files and directories we need a new table */
1451 if (nr_dirs
&& nr_files
) {
1452 struct ctl_table
*new;
1453 files
= kcalloc(nr_files
+ 1, sizeof(struct ctl_table
),
1458 ctl_table_arg
= files
;
1459 for (new = files
, entry
= table
; entry
->procname
; entry
++) {
1467 /* Register everything except a directory full of subdirectories */
1468 if (nr_files
|| !nr_dirs
) {
1469 struct ctl_table_header
*header
;
1470 header
= __register_sysctl_table(set
, path
, files
);
1472 kfree(ctl_table_arg
);
1476 /* Remember if we need to free the file table */
1477 header
->ctl_table_arg
= ctl_table_arg
;
1478 **subheader
= header
;
1482 /* Recurse into the subdirectories. */
1483 for (entry
= table
; entry
->procname
; entry
++) {
1489 err
= -ENAMETOOLONG
;
1490 child_pos
= append_path(path
, pos
, entry
->procname
);
1494 err
= register_leaf_sysctl_tables(path
, child_pos
, subheader
,
1502 /* On failure our caller will unregister all registered subheaders */
1507 * __register_sysctl_paths - register a sysctl table hierarchy
1508 * @set: Sysctl tree to register on
1509 * @path: The path to the directory the sysctl table is in.
1510 * @table: the top-level table structure
1512 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1513 * array. A completely 0 filled entry terminates the table.
1515 * See __register_sysctl_table for more details.
1517 struct ctl_table_header
*__register_sysctl_paths(
1518 struct ctl_table_set
*set
,
1519 const struct ctl_path
*path
, struct ctl_table
*table
)
1521 struct ctl_table
*ctl_table_arg
= table
;
1522 int nr_subheaders
= count_subheaders(table
);
1523 struct ctl_table_header
*header
= NULL
, **subheaders
, **subheader
;
1524 const struct ctl_path
*component
;
1525 char *new_path
, *pos
;
1527 pos
= new_path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1532 for (component
= path
; component
->procname
; component
++) {
1533 pos
= append_path(new_path
, pos
, component
->procname
);
1537 while (table
->procname
&& table
->child
&& !table
[1].procname
) {
1538 pos
= append_path(new_path
, pos
, table
->procname
);
1541 table
= table
->child
;
1543 if (nr_subheaders
== 1) {
1544 header
= __register_sysctl_table(set
, new_path
, table
);
1546 header
->ctl_table_arg
= ctl_table_arg
;
1548 header
= kzalloc(sizeof(*header
) +
1549 sizeof(*subheaders
)*nr_subheaders
, GFP_KERNEL
);
1553 subheaders
= (struct ctl_table_header
**) (header
+ 1);
1554 subheader
= subheaders
;
1555 header
->ctl_table_arg
= ctl_table_arg
;
1557 if (register_leaf_sysctl_tables(new_path
, pos
, &subheader
,
1559 goto err_register_leaves
;
1566 err_register_leaves
:
1567 while (subheader
> subheaders
) {
1568 struct ctl_table_header
*subh
= *(--subheader
);
1569 struct ctl_table
*table
= subh
->ctl_table_arg
;
1570 unregister_sysctl_table(subh
);
1579 * register_sysctl_table_path - register a sysctl table hierarchy
1580 * @path: The path to the directory the sysctl table is in.
1581 * @table: the top-level table structure
1583 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1584 * array. A completely 0 filled entry terminates the table.
1586 * See __register_sysctl_paths for more details.
1588 struct ctl_table_header
*register_sysctl_paths(const struct ctl_path
*path
,
1589 struct ctl_table
*table
)
1591 return __register_sysctl_paths(&sysctl_table_root
.default_set
,
1594 EXPORT_SYMBOL(register_sysctl_paths
);
1597 * register_sysctl_table - register a sysctl table hierarchy
1598 * @table: the top-level table structure
1600 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1601 * array. A completely 0 filled entry terminates the table.
1603 * See register_sysctl_paths for more details.
1605 struct ctl_table_header
*register_sysctl_table(struct ctl_table
*table
)
1607 static const struct ctl_path null_path
[] = { {} };
1609 return register_sysctl_paths(null_path
, table
);
1611 EXPORT_SYMBOL(register_sysctl_table
);
1613 static void put_links(struct ctl_table_header
*header
)
1615 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1616 struct ctl_table_root
*root
= header
->root
;
1617 struct ctl_dir
*parent
= header
->parent
;
1618 struct ctl_dir
*core_parent
;
1619 struct ctl_table
*entry
;
1621 if (header
->set
== root_set
)
1624 core_parent
= xlate_dir(root_set
, parent
);
1625 if (IS_ERR(core_parent
))
1628 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
1629 struct ctl_table_header
*link_head
;
1630 struct ctl_table
*link
;
1631 const char *name
= entry
->procname
;
1633 link
= find_entry(&link_head
, core_parent
, name
, strlen(name
));
1635 ((S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
)) ||
1636 (S_ISLNK(link
->mode
) && (link
->data
== root
)))) {
1637 drop_sysctl_table(link_head
);
1640 pr_err("sysctl link missing during unregister: ");
1641 sysctl_print_dir(parent
);
1642 pr_cont("/%s\n", name
);
1647 static void drop_sysctl_table(struct ctl_table_header
*header
)
1649 struct ctl_dir
*parent
= header
->parent
;
1656 start_unregistering(header
);
1659 if (!--header
->count
)
1660 kfree_rcu(header
, rcu
);
1663 drop_sysctl_table(&parent
->header
);
1667 * unregister_sysctl_table - unregister a sysctl table hierarchy
1668 * @header: the header returned from register_sysctl_table
1670 * Unregisters the sysctl table and all children. proc entries may not
1671 * actually be removed until they are no longer used by anyone.
1673 void unregister_sysctl_table(struct ctl_table_header
* header
)
1681 nr_subheaders
= count_subheaders(header
->ctl_table_arg
);
1682 if (unlikely(nr_subheaders
> 1)) {
1683 struct ctl_table_header
**subheaders
;
1686 subheaders
= (struct ctl_table_header
**)(header
+ 1);
1687 for (i
= nr_subheaders
-1; i
>= 0; i
--) {
1688 struct ctl_table_header
*subh
= subheaders
[i
];
1689 struct ctl_table
*table
= subh
->ctl_table_arg
;
1690 unregister_sysctl_table(subh
);
1697 spin_lock(&sysctl_lock
);
1698 drop_sysctl_table(header
);
1699 spin_unlock(&sysctl_lock
);
1701 EXPORT_SYMBOL(unregister_sysctl_table
);
1703 void setup_sysctl_set(struct ctl_table_set
*set
,
1704 struct ctl_table_root
*root
,
1705 int (*is_seen
)(struct ctl_table_set
*))
1707 memset(set
, 0, sizeof(*set
));
1708 set
->is_seen
= is_seen
;
1709 init_header(&set
->dir
.header
, root
, set
, NULL
, root_table
);
1712 void retire_sysctl_set(struct ctl_table_set
*set
)
1714 WARN_ON(!RB_EMPTY_ROOT(&set
->dir
.root
));
1717 int __init
proc_sys_init(void)
1719 struct proc_dir_entry
*proc_sys_root
;
1721 proc_sys_root
= proc_mkdir("sys", NULL
);
1722 proc_sys_root
->proc_iops
= &proc_sys_dir_operations
;
1723 proc_sys_root
->proc_dir_ops
= &proc_sys_dir_file_operations
;
1724 proc_sys_root
->nlink
= 0;
1726 return sysctl_init();