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 /* Support for permanently empty directories */
27 struct ctl_table sysctl_mount_point
[] = {
31 static bool is_empty_dir(struct ctl_table_header
*head
)
33 return head
->ctl_table
[0].child
== sysctl_mount_point
;
36 static void set_empty_dir(struct ctl_dir
*dir
)
38 dir
->header
.ctl_table
[0].child
= sysctl_mount_point
;
41 static void clear_empty_dir(struct ctl_dir
*dir
)
44 dir
->header
.ctl_table
[0].child
= NULL
;
47 void proc_sys_poll_notify(struct ctl_table_poll
*poll
)
52 atomic_inc(&poll
->event
);
53 wake_up_interruptible(&poll
->wait
);
56 static struct ctl_table root_table
[] = {
59 .mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
,
63 static struct ctl_table_root sysctl_table_root
= {
64 .default_set
.dir
.header
= {
67 .ctl_table
= root_table
}},
68 .ctl_table_arg
= root_table
,
69 .root
= &sysctl_table_root
,
70 .set
= &sysctl_table_root
.default_set
,
74 static DEFINE_SPINLOCK(sysctl_lock
);
76 static void drop_sysctl_table(struct ctl_table_header
*header
);
77 static int sysctl_follow_link(struct ctl_table_header
**phead
,
78 struct ctl_table
**pentry
);
79 static int insert_links(struct ctl_table_header
*head
);
80 static void put_links(struct ctl_table_header
*header
);
82 static void sysctl_print_dir(struct ctl_dir
*dir
)
84 if (dir
->header
.parent
)
85 sysctl_print_dir(dir
->header
.parent
);
86 pr_cont("%s/", dir
->header
.ctl_table
[0].procname
);
89 static int namecmp(const char *name1
, int len1
, const char *name2
, int len2
)
98 cmp
= memcmp(name1
, name2
, minlen
);
104 /* Called under sysctl_lock */
105 static struct ctl_table
*find_entry(struct ctl_table_header
**phead
,
106 struct ctl_dir
*dir
, const char *name
, int namelen
)
108 struct ctl_table_header
*head
;
109 struct ctl_table
*entry
;
110 struct rb_node
*node
= dir
->root
.rb_node
;
114 struct ctl_node
*ctl_node
;
115 const char *procname
;
118 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
119 head
= ctl_node
->header
;
120 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
121 procname
= entry
->procname
;
123 cmp
= namecmp(name
, namelen
, procname
, strlen(procname
));
125 node
= node
->rb_left
;
127 node
= node
->rb_right
;
136 static int insert_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
138 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
139 struct rb_node
**p
= &head
->parent
->root
.rb_node
;
140 struct rb_node
*parent
= NULL
;
141 const char *name
= entry
->procname
;
142 int namelen
= strlen(name
);
145 struct ctl_table_header
*parent_head
;
146 struct ctl_table
*parent_entry
;
147 struct ctl_node
*parent_node
;
148 const char *parent_name
;
152 parent_node
= rb_entry(parent
, struct ctl_node
, node
);
153 parent_head
= parent_node
->header
;
154 parent_entry
= &parent_head
->ctl_table
[parent_node
- parent_head
->node
];
155 parent_name
= parent_entry
->procname
;
157 cmp
= namecmp(name
, namelen
, parent_name
, strlen(parent_name
));
163 pr_err("sysctl duplicate entry: ");
164 sysctl_print_dir(head
->parent
);
165 pr_cont("/%s\n", entry
->procname
);
170 rb_link_node(node
, parent
, p
);
171 rb_insert_color(node
, &head
->parent
->root
);
175 static void erase_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
177 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
179 rb_erase(node
, &head
->parent
->root
);
182 static void init_header(struct ctl_table_header
*head
,
183 struct ctl_table_root
*root
, struct ctl_table_set
*set
,
184 struct ctl_node
*node
, struct ctl_table
*table
)
186 head
->ctl_table
= table
;
187 head
->ctl_table_arg
= table
;
191 head
->unregistering
= NULL
;
196 INIT_HLIST_HEAD(&head
->inodes
);
198 struct ctl_table
*entry
;
199 for (entry
= table
; entry
->procname
; entry
++, node
++)
204 static void erase_header(struct ctl_table_header
*head
)
206 struct ctl_table
*entry
;
207 for (entry
= head
->ctl_table
; entry
->procname
; entry
++)
208 erase_entry(head
, entry
);
211 static int insert_header(struct ctl_dir
*dir
, struct ctl_table_header
*header
)
213 struct ctl_table
*entry
;
216 /* Is this a permanently empty directory? */
217 if (is_empty_dir(&dir
->header
))
220 /* Am I creating a permanently empty directory? */
221 if (header
->ctl_table
== sysctl_mount_point
) {
222 if (!RB_EMPTY_ROOT(&dir
->root
))
228 header
->parent
= dir
;
229 err
= insert_links(header
);
232 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
233 err
= insert_entry(header
, entry
);
239 erase_header(header
);
242 if (header
->ctl_table
== sysctl_mount_point
)
243 clear_empty_dir(dir
);
244 header
->parent
= NULL
;
245 drop_sysctl_table(&dir
->header
);
249 /* called under sysctl_lock */
250 static int use_table(struct ctl_table_header
*p
)
252 if (unlikely(p
->unregistering
))
258 /* called under sysctl_lock */
259 static void unuse_table(struct ctl_table_header
*p
)
262 if (unlikely(p
->unregistering
))
263 complete(p
->unregistering
);
266 static void proc_sys_prune_dcache(struct ctl_table_header
*head
)
269 struct proc_inode
*ei
;
270 struct hlist_node
*node
;
271 struct super_block
*sb
;
275 node
= hlist_first_rcu(&head
->inodes
);
278 ei
= hlist_entry(node
, struct proc_inode
, sysctl_inodes
);
279 spin_lock(&sysctl_lock
);
280 hlist_del_init_rcu(&ei
->sysctl_inodes
);
281 spin_unlock(&sysctl_lock
);
283 inode
= &ei
->vfs_inode
;
285 if (!atomic_inc_not_zero(&sb
->s_active
))
287 inode
= igrab(inode
);
289 if (unlikely(!inode
)) {
290 deactivate_super(sb
);
295 d_prune_aliases(inode
);
297 deactivate_super(sb
);
304 /* called under sysctl_lock, will reacquire if has to wait */
305 static void start_unregistering(struct ctl_table_header
*p
)
308 * if p->used is 0, nobody will ever touch that entry again;
309 * we'll eliminate all paths to it before dropping sysctl_lock
311 if (unlikely(p
->used
)) {
312 struct completion wait
;
313 init_completion(&wait
);
314 p
->unregistering
= &wait
;
315 spin_unlock(&sysctl_lock
);
316 wait_for_completion(&wait
);
318 /* anything non-NULL; we'll never dereference it */
319 p
->unregistering
= ERR_PTR(-EINVAL
);
320 spin_unlock(&sysctl_lock
);
323 * Prune dentries for unregistered sysctls: namespaced sysctls
324 * can have duplicate names and contaminate dcache very badly.
326 proc_sys_prune_dcache(p
);
328 * do not remove from the list until nobody holds it; walking the
329 * list in do_sysctl() relies on that.
331 spin_lock(&sysctl_lock
);
335 static struct ctl_table_header
*sysctl_head_grab(struct ctl_table_header
*head
)
338 spin_lock(&sysctl_lock
);
339 if (!use_table(head
))
340 head
= ERR_PTR(-ENOENT
);
341 spin_unlock(&sysctl_lock
);
345 static void sysctl_head_finish(struct ctl_table_header
*head
)
349 spin_lock(&sysctl_lock
);
351 spin_unlock(&sysctl_lock
);
354 static struct ctl_table_set
*
355 lookup_header_set(struct ctl_table_root
*root
)
357 struct ctl_table_set
*set
= &root
->default_set
;
359 set
= root
->lookup(root
);
363 static struct ctl_table
*lookup_entry(struct ctl_table_header
**phead
,
365 const char *name
, int namelen
)
367 struct ctl_table_header
*head
;
368 struct ctl_table
*entry
;
370 spin_lock(&sysctl_lock
);
371 entry
= find_entry(&head
, dir
, name
, namelen
);
372 if (entry
&& use_table(head
))
376 spin_unlock(&sysctl_lock
);
380 static struct ctl_node
*first_usable_entry(struct rb_node
*node
)
382 struct ctl_node
*ctl_node
;
384 for (;node
; node
= rb_next(node
)) {
385 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
386 if (use_table(ctl_node
->header
))
392 static void first_entry(struct ctl_dir
*dir
,
393 struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
395 struct ctl_table_header
*head
= NULL
;
396 struct ctl_table
*entry
= NULL
;
397 struct ctl_node
*ctl_node
;
399 spin_lock(&sysctl_lock
);
400 ctl_node
= first_usable_entry(rb_first(&dir
->root
));
401 spin_unlock(&sysctl_lock
);
403 head
= ctl_node
->header
;
404 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
410 static void next_entry(struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
412 struct ctl_table_header
*head
= *phead
;
413 struct ctl_table
*entry
= *pentry
;
414 struct ctl_node
*ctl_node
= &head
->node
[entry
- head
->ctl_table
];
416 spin_lock(&sysctl_lock
);
419 ctl_node
= first_usable_entry(rb_next(&ctl_node
->node
));
420 spin_unlock(&sysctl_lock
);
423 head
= ctl_node
->header
;
424 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
431 * sysctl_perm does NOT grant the superuser all rights automatically, because
432 * some sysctl variables are readonly even to root.
435 static int test_perm(int mode
, int op
)
437 if (uid_eq(current_euid(), GLOBAL_ROOT_UID
))
439 else if (in_egroup_p(GLOBAL_ROOT_GID
))
441 if ((op
& ~mode
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
)) == 0)
446 static int sysctl_perm(struct ctl_table_header
*head
, struct ctl_table
*table
, int op
)
448 struct ctl_table_root
*root
= head
->root
;
451 if (root
->permissions
)
452 mode
= root
->permissions(head
, table
);
456 return test_perm(mode
, op
);
459 static struct inode
*proc_sys_make_inode(struct super_block
*sb
,
460 struct ctl_table_header
*head
, struct ctl_table
*table
)
462 struct ctl_table_root
*root
= head
->root
;
464 struct proc_inode
*ei
;
466 inode
= new_inode(sb
);
468 return ERR_PTR(-ENOMEM
);
470 inode
->i_ino
= get_next_ino();
474 spin_lock(&sysctl_lock
);
475 if (unlikely(head
->unregistering
)) {
476 spin_unlock(&sysctl_lock
);
478 return ERR_PTR(-ENOENT
);
481 ei
->sysctl_entry
= table
;
482 hlist_add_head_rcu(&ei
->sysctl_inodes
, &head
->inodes
);
484 spin_unlock(&sysctl_lock
);
486 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= current_time(inode
);
487 inode
->i_mode
= table
->mode
;
488 if (!S_ISDIR(table
->mode
)) {
489 inode
->i_mode
|= S_IFREG
;
490 inode
->i_op
= &proc_sys_inode_operations
;
491 inode
->i_fop
= &proc_sys_file_operations
;
493 inode
->i_mode
|= S_IFDIR
;
494 inode
->i_op
= &proc_sys_dir_operations
;
495 inode
->i_fop
= &proc_sys_dir_file_operations
;
496 if (is_empty_dir(head
))
497 make_empty_dir_inode(inode
);
500 if (root
->set_ownership
)
501 root
->set_ownership(head
, table
, &inode
->i_uid
, &inode
->i_gid
);
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 inode
= proc_sys_make_inode(dir
->i_sb
, h
? h
: head
, p
);
553 err
= ERR_CAST(inode
);
557 d_set_d_op(dentry
, &proc_sys_dentry_operations
);
558 err
= d_splice_alias(inode
, dentry
);
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
;
573 void *new_buf
= NULL
;
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 error
= BPF_CGROUP_RUN_PROG_SYSCTL(head
, table
, write
, buf
, &count
,
597 /* careful: calling conventions are nasty here */
603 error
= table
->proc_handler(table
, write
, (void __user
*)new_buf
,
608 error
= table
->proc_handler(table
, write
, buf
, &count
, ppos
);
614 sysctl_head_finish(head
);
619 static ssize_t
proc_sys_read(struct file
*filp
, char __user
*buf
,
620 size_t count
, loff_t
*ppos
)
622 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 0);
625 static ssize_t
proc_sys_write(struct file
*filp
, const char __user
*buf
,
626 size_t count
, loff_t
*ppos
)
628 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 1);
631 static int proc_sys_open(struct inode
*inode
, struct file
*filp
)
633 struct ctl_table_header
*head
= grab_header(inode
);
634 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
636 /* sysctl was unregistered */
638 return PTR_ERR(head
);
641 filp
->private_data
= proc_sys_poll_event(table
->poll
);
643 sysctl_head_finish(head
);
648 static __poll_t
proc_sys_poll(struct file
*filp
, poll_table
*wait
)
650 struct inode
*inode
= file_inode(filp
);
651 struct ctl_table_header
*head
= grab_header(inode
);
652 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
653 __poll_t ret
= DEFAULT_POLLMASK
;
656 /* sysctl was unregistered */
658 return EPOLLERR
| EPOLLHUP
;
660 if (!table
->proc_handler
)
666 event
= (unsigned long)filp
->private_data
;
667 poll_wait(filp
, &table
->poll
->wait
, wait
);
669 if (event
!= atomic_read(&table
->poll
->event
)) {
670 filp
->private_data
= proc_sys_poll_event(table
->poll
);
671 ret
= EPOLLIN
| EPOLLRDNORM
| EPOLLERR
| EPOLLPRI
;
675 sysctl_head_finish(head
);
680 static bool proc_sys_fill_cache(struct file
*file
,
681 struct dir_context
*ctx
,
682 struct ctl_table_header
*head
,
683 struct ctl_table
*table
)
685 struct dentry
*child
, *dir
= file
->f_path
.dentry
;
689 unsigned type
= DT_UNKNOWN
;
691 qname
.name
= table
->procname
;
692 qname
.len
= strlen(table
->procname
);
693 qname
.hash
= full_name_hash(dir
, qname
.name
, qname
.len
);
695 child
= d_lookup(dir
, &qname
);
697 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
698 child
= d_alloc_parallel(dir
, &qname
, &wq
);
701 if (d_in_lookup(child
)) {
703 inode
= proc_sys_make_inode(dir
->d_sb
, head
, table
);
705 d_lookup_done(child
);
709 d_set_d_op(child
, &proc_sys_dentry_operations
);
710 res
= d_splice_alias(inode
, child
);
711 d_lookup_done(child
);
722 inode
= d_inode(child
);
724 type
= inode
->i_mode
>> 12;
726 return dir_emit(ctx
, qname
.name
, qname
.len
, ino
, type
);
729 static bool proc_sys_link_fill_cache(struct file
*file
,
730 struct dir_context
*ctx
,
731 struct ctl_table_header
*head
,
732 struct ctl_table
*table
)
736 head
= sysctl_head_grab(head
);
740 /* It is not an error if we can not follow the link ignore it */
741 if (sysctl_follow_link(&head
, &table
))
744 ret
= proc_sys_fill_cache(file
, ctx
, head
, table
);
746 sysctl_head_finish(head
);
750 static int scan(struct ctl_table_header
*head
, struct ctl_table
*table
,
751 unsigned long *pos
, struct file
*file
,
752 struct dir_context
*ctx
)
756 if ((*pos
)++ < ctx
->pos
)
759 if (unlikely(S_ISLNK(table
->mode
)))
760 res
= proc_sys_link_fill_cache(file
, ctx
, head
, table
);
762 res
= proc_sys_fill_cache(file
, ctx
, head
, table
);
770 static int proc_sys_readdir(struct file
*file
, struct dir_context
*ctx
)
772 struct ctl_table_header
*head
= grab_header(file_inode(file
));
773 struct ctl_table_header
*h
= NULL
;
774 struct ctl_table
*entry
;
775 struct ctl_dir
*ctl_dir
;
779 return PTR_ERR(head
);
781 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
783 if (!dir_emit_dots(file
, ctx
))
788 for (first_entry(ctl_dir
, &h
, &entry
); h
; next_entry(&h
, &entry
)) {
789 if (!scan(h
, entry
, &pos
, file
, ctx
)) {
790 sysctl_head_finish(h
);
795 sysctl_head_finish(head
);
799 static int proc_sys_permission(struct inode
*inode
, int mask
)
802 * sysctl entries that are not writeable,
803 * are _NOT_ writeable, capabilities or not.
805 struct ctl_table_header
*head
;
806 struct ctl_table
*table
;
809 /* Executable files are not allowed under /proc/sys/ */
810 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
))
813 head
= grab_header(inode
);
815 return PTR_ERR(head
);
817 table
= PROC_I(inode
)->sysctl_entry
;
818 if (!table
) /* global root - r-xr-xr-x */
819 error
= mask
& MAY_WRITE
? -EACCES
: 0;
820 else /* Use the permissions on the sysctl table entry */
821 error
= sysctl_perm(head
, table
, mask
& ~MAY_NOT_BLOCK
);
823 sysctl_head_finish(head
);
827 static int proc_sys_setattr(struct dentry
*dentry
, struct iattr
*attr
)
829 struct inode
*inode
= d_inode(dentry
);
832 if (attr
->ia_valid
& (ATTR_MODE
| ATTR_UID
| ATTR_GID
))
835 error
= setattr_prepare(dentry
, attr
);
839 setattr_copy(inode
, attr
);
840 mark_inode_dirty(inode
);
844 static int proc_sys_getattr(const struct path
*path
, struct kstat
*stat
,
845 u32 request_mask
, unsigned int query_flags
)
847 struct inode
*inode
= d_inode(path
->dentry
);
848 struct ctl_table_header
*head
= grab_header(inode
);
849 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
852 return PTR_ERR(head
);
854 generic_fillattr(inode
, stat
);
856 stat
->mode
= (stat
->mode
& S_IFMT
) | table
->mode
;
858 sysctl_head_finish(head
);
862 static const struct file_operations proc_sys_file_operations
= {
863 .open
= proc_sys_open
,
864 .poll
= proc_sys_poll
,
865 .read
= proc_sys_read
,
866 .write
= proc_sys_write
,
867 .llseek
= default_llseek
,
870 static const struct file_operations proc_sys_dir_file_operations
= {
871 .read
= generic_read_dir
,
872 .iterate_shared
= proc_sys_readdir
,
873 .llseek
= generic_file_llseek
,
876 static const struct inode_operations proc_sys_inode_operations
= {
877 .permission
= proc_sys_permission
,
878 .setattr
= proc_sys_setattr
,
879 .getattr
= proc_sys_getattr
,
882 static const struct inode_operations proc_sys_dir_operations
= {
883 .lookup
= proc_sys_lookup
,
884 .permission
= proc_sys_permission
,
885 .setattr
= proc_sys_setattr
,
886 .getattr
= proc_sys_getattr
,
889 static int proc_sys_revalidate(struct dentry
*dentry
, unsigned int flags
)
891 if (flags
& LOOKUP_RCU
)
893 return !PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
896 static int proc_sys_delete(const struct dentry
*dentry
)
898 return !!PROC_I(d_inode(dentry
))->sysctl
->unregistering
;
901 static int sysctl_is_seen(struct ctl_table_header
*p
)
903 struct ctl_table_set
*set
= p
->set
;
905 spin_lock(&sysctl_lock
);
906 if (p
->unregistering
)
908 else if (!set
->is_seen
)
911 res
= set
->is_seen(set
);
912 spin_unlock(&sysctl_lock
);
916 static int proc_sys_compare(const struct dentry
*dentry
,
917 unsigned int len
, const char *str
, const struct qstr
*name
)
919 struct ctl_table_header
*head
;
922 /* Although proc doesn't have negative dentries, rcu-walk means
923 * that inode here can be NULL */
924 /* AV: can it, indeed? */
925 inode
= d_inode_rcu(dentry
);
928 if (name
->len
!= len
)
930 if (memcmp(name
->name
, str
, len
))
932 head
= rcu_dereference(PROC_I(inode
)->sysctl
);
933 return !head
|| !sysctl_is_seen(head
);
936 static const struct dentry_operations proc_sys_dentry_operations
= {
937 .d_revalidate
= proc_sys_revalidate
,
938 .d_delete
= proc_sys_delete
,
939 .d_compare
= proc_sys_compare
,
942 static struct ctl_dir
*find_subdir(struct ctl_dir
*dir
,
943 const char *name
, int namelen
)
945 struct ctl_table_header
*head
;
946 struct ctl_table
*entry
;
948 entry
= find_entry(&head
, dir
, name
, namelen
);
950 return ERR_PTR(-ENOENT
);
951 if (!S_ISDIR(entry
->mode
))
952 return ERR_PTR(-ENOTDIR
);
953 return container_of(head
, struct ctl_dir
, header
);
956 static struct ctl_dir
*new_dir(struct ctl_table_set
*set
,
957 const char *name
, int namelen
)
959 struct ctl_table
*table
;
961 struct ctl_node
*node
;
964 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node
) +
965 sizeof(struct ctl_table
)*2 + namelen
+ 1,
970 node
= (struct ctl_node
*)(new + 1);
971 table
= (struct ctl_table
*)(node
+ 1);
972 new_name
= (char *)(table
+ 2);
973 memcpy(new_name
, name
, namelen
);
974 new_name
[namelen
] = '\0';
975 table
[0].procname
= new_name
;
976 table
[0].mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
;
977 init_header(&new->header
, set
->dir
.header
.root
, set
, node
, table
);
983 * get_subdir - find or create a subdir with the specified name.
984 * @dir: Directory to create the subdirectory in
985 * @name: The name of the subdirectory to find or create
986 * @namelen: The length of name
988 * Takes a directory with an elevated reference count so we know that
989 * if we drop the lock the directory will not go away. Upon success
990 * the reference is moved from @dir to the returned subdirectory.
991 * Upon error an error code is returned and the reference on @dir is
994 static struct ctl_dir
*get_subdir(struct ctl_dir
*dir
,
995 const char *name
, int namelen
)
997 struct ctl_table_set
*set
= dir
->header
.set
;
998 struct ctl_dir
*subdir
, *new = NULL
;
1001 spin_lock(&sysctl_lock
);
1002 subdir
= find_subdir(dir
, name
, namelen
);
1003 if (!IS_ERR(subdir
))
1005 if (PTR_ERR(subdir
) != -ENOENT
)
1008 spin_unlock(&sysctl_lock
);
1009 new = new_dir(set
, name
, namelen
);
1010 spin_lock(&sysctl_lock
);
1011 subdir
= ERR_PTR(-ENOMEM
);
1015 /* Was the subdir added while we dropped the lock? */
1016 subdir
= find_subdir(dir
, name
, namelen
);
1017 if (!IS_ERR(subdir
))
1019 if (PTR_ERR(subdir
) != -ENOENT
)
1022 /* Nope. Use the our freshly made directory entry. */
1023 err
= insert_header(dir
, &new->header
);
1024 subdir
= ERR_PTR(err
);
1029 subdir
->header
.nreg
++;
1031 if (IS_ERR(subdir
)) {
1032 pr_err("sysctl could not get directory: ");
1033 sysctl_print_dir(dir
);
1034 pr_cont("/%*.*s %ld\n",
1035 namelen
, namelen
, name
, PTR_ERR(subdir
));
1037 drop_sysctl_table(&dir
->header
);
1039 drop_sysctl_table(&new->header
);
1040 spin_unlock(&sysctl_lock
);
1044 static struct ctl_dir
*xlate_dir(struct ctl_table_set
*set
, struct ctl_dir
*dir
)
1046 struct ctl_dir
*parent
;
1047 const char *procname
;
1048 if (!dir
->header
.parent
)
1050 parent
= xlate_dir(set
, dir
->header
.parent
);
1053 procname
= dir
->header
.ctl_table
[0].procname
;
1054 return find_subdir(parent
, procname
, strlen(procname
));
1057 static int sysctl_follow_link(struct ctl_table_header
**phead
,
1058 struct ctl_table
**pentry
)
1060 struct ctl_table_header
*head
;
1061 struct ctl_table_root
*root
;
1062 struct ctl_table_set
*set
;
1063 struct ctl_table
*entry
;
1064 struct ctl_dir
*dir
;
1068 spin_lock(&sysctl_lock
);
1069 root
= (*pentry
)->data
;
1070 set
= lookup_header_set(root
);
1071 dir
= xlate_dir(set
, (*phead
)->parent
);
1075 const char *procname
= (*pentry
)->procname
;
1077 entry
= find_entry(&head
, dir
, procname
, strlen(procname
));
1079 if (entry
&& use_table(head
)) {
1080 unuse_table(*phead
);
1087 spin_unlock(&sysctl_lock
);
1091 static int sysctl_err(const char *path
, struct ctl_table
*table
, char *fmt
, ...)
1093 struct va_format vaf
;
1096 va_start(args
, fmt
);
1100 pr_err("sysctl table check failed: %s/%s %pV\n",
1101 path
, table
->procname
, &vaf
);
1107 static int sysctl_check_table_array(const char *path
, struct ctl_table
*table
)
1111 if ((table
->proc_handler
== proc_douintvec
) ||
1112 (table
->proc_handler
== proc_douintvec_minmax
)) {
1113 if (table
->maxlen
!= sizeof(unsigned int))
1114 err
|= sysctl_err(path
, table
, "array not allowed");
1120 static int sysctl_check_table(const char *path
, struct ctl_table
*table
)
1123 for (; table
->procname
; table
++) {
1125 err
|= sysctl_err(path
, table
, "Not a file");
1127 if ((table
->proc_handler
== proc_dostring
) ||
1128 (table
->proc_handler
== proc_dointvec
) ||
1129 (table
->proc_handler
== proc_douintvec
) ||
1130 (table
->proc_handler
== proc_douintvec_minmax
) ||
1131 (table
->proc_handler
== proc_dointvec_minmax
) ||
1132 (table
->proc_handler
== proc_dointvec_jiffies
) ||
1133 (table
->proc_handler
== proc_dointvec_userhz_jiffies
) ||
1134 (table
->proc_handler
== proc_dointvec_ms_jiffies
) ||
1135 (table
->proc_handler
== proc_doulongvec_minmax
) ||
1136 (table
->proc_handler
== proc_doulongvec_ms_jiffies_minmax
)) {
1138 err
|= sysctl_err(path
, table
, "No data");
1140 err
|= sysctl_err(path
, table
, "No maxlen");
1142 err
|= sysctl_check_table_array(path
, table
);
1144 if (!table
->proc_handler
)
1145 err
|= sysctl_err(path
, table
, "No proc_handler");
1147 if ((table
->mode
& (S_IRUGO
|S_IWUGO
)) != table
->mode
)
1148 err
|= sysctl_err(path
, table
, "bogus .mode 0%o",
1154 static struct ctl_table_header
*new_links(struct ctl_dir
*dir
, struct ctl_table
*table
,
1155 struct ctl_table_root
*link_root
)
1157 struct ctl_table
*link_table
, *entry
, *link
;
1158 struct ctl_table_header
*links
;
1159 struct ctl_node
*node
;
1161 int nr_entries
, name_bytes
;
1165 for (entry
= table
; entry
->procname
; entry
++) {
1167 name_bytes
+= strlen(entry
->procname
) + 1;
1170 links
= kzalloc(sizeof(struct ctl_table_header
) +
1171 sizeof(struct ctl_node
)*nr_entries
+
1172 sizeof(struct ctl_table
)*(nr_entries
+ 1) +
1179 node
= (struct ctl_node
*)(links
+ 1);
1180 link_table
= (struct ctl_table
*)(node
+ nr_entries
);
1181 link_name
= (char *)&link_table
[nr_entries
+ 1];
1183 for (link
= link_table
, entry
= table
; entry
->procname
; link
++, entry
++) {
1184 int len
= strlen(entry
->procname
) + 1;
1185 memcpy(link_name
, entry
->procname
, len
);
1186 link
->procname
= link_name
;
1187 link
->mode
= S_IFLNK
|S_IRWXUGO
;
1188 link
->data
= link_root
;
1191 init_header(links
, dir
->header
.root
, dir
->header
.set
, node
, link_table
);
1192 links
->nreg
= nr_entries
;
1197 static bool get_links(struct ctl_dir
*dir
,
1198 struct ctl_table
*table
, struct ctl_table_root
*link_root
)
1200 struct ctl_table_header
*head
;
1201 struct ctl_table
*entry
, *link
;
1203 /* Are there links available for every entry in table? */
1204 for (entry
= table
; entry
->procname
; entry
++) {
1205 const char *procname
= entry
->procname
;
1206 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1209 if (S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
))
1211 if (S_ISLNK(link
->mode
) && (link
->data
== link_root
))
1216 /* The checks passed. Increase the registration count on the links */
1217 for (entry
= table
; entry
->procname
; entry
++) {
1218 const char *procname
= entry
->procname
;
1219 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1225 static int insert_links(struct ctl_table_header
*head
)
1227 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1228 struct ctl_dir
*core_parent
= NULL
;
1229 struct ctl_table_header
*links
;
1232 if (head
->set
== root_set
)
1235 core_parent
= xlate_dir(root_set
, head
->parent
);
1236 if (IS_ERR(core_parent
))
1239 if (get_links(core_parent
, head
->ctl_table
, head
->root
))
1242 core_parent
->header
.nreg
++;
1243 spin_unlock(&sysctl_lock
);
1245 links
= new_links(core_parent
, head
->ctl_table
, head
->root
);
1247 spin_lock(&sysctl_lock
);
1253 if (get_links(core_parent
, head
->ctl_table
, head
->root
)) {
1258 err
= insert_header(core_parent
, links
);
1262 drop_sysctl_table(&core_parent
->header
);
1267 * __register_sysctl_table - register a leaf sysctl table
1268 * @set: Sysctl tree to register on
1269 * @path: The path to the directory the sysctl table is in.
1270 * @table: the top-level table structure
1272 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1273 * array. A completely 0 filled entry terminates the table.
1275 * The members of the &struct ctl_table structure are used as follows:
1277 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1278 * enter a sysctl file
1280 * data - a pointer to data for use by proc_handler
1282 * maxlen - the maximum size in bytes of the data
1284 * mode - the file permissions for the /proc/sys file
1286 * child - must be %NULL.
1288 * proc_handler - the text handler routine (described below)
1290 * extra1, extra2 - extra pointers usable by the proc handler routines
1292 * Leaf nodes in the sysctl tree will be represented by a single file
1293 * under /proc; non-leaf nodes will be represented by directories.
1295 * There must be a proc_handler routine for any terminal nodes.
1296 * Several default handlers are available to cover common cases -
1298 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1299 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1300 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1302 * It is the handler's job to read the input buffer from user memory
1303 * and process it. The handler should return 0 on success.
1305 * This routine returns %NULL on a failure to register, and a pointer
1306 * to the table header on success.
1308 struct ctl_table_header
*__register_sysctl_table(
1309 struct ctl_table_set
*set
,
1310 const char *path
, struct ctl_table
*table
)
1312 struct ctl_table_root
*root
= set
->dir
.header
.root
;
1313 struct ctl_table_header
*header
;
1314 const char *name
, *nextname
;
1315 struct ctl_dir
*dir
;
1316 struct ctl_table
*entry
;
1317 struct ctl_node
*node
;
1320 for (entry
= table
; entry
->procname
; entry
++)
1323 header
= kzalloc(sizeof(struct ctl_table_header
) +
1324 sizeof(struct ctl_node
)*nr_entries
, GFP_KERNEL
);
1328 node
= (struct ctl_node
*)(header
+ 1);
1329 init_header(header
, root
, set
, node
, table
);
1330 if (sysctl_check_table(path
, table
))
1333 spin_lock(&sysctl_lock
);
1335 /* Reference moved down the diretory tree get_subdir */
1337 spin_unlock(&sysctl_lock
);
1339 /* Find the directory for the ctl_table */
1340 for (name
= path
; name
; name
= nextname
) {
1342 nextname
= strchr(name
, '/');
1344 namelen
= nextname
- name
;
1347 namelen
= strlen(name
);
1352 dir
= get_subdir(dir
, name
, namelen
);
1357 spin_lock(&sysctl_lock
);
1358 if (insert_header(dir
, header
))
1359 goto fail_put_dir_locked
;
1361 drop_sysctl_table(&dir
->header
);
1362 spin_unlock(&sysctl_lock
);
1366 fail_put_dir_locked
:
1367 drop_sysctl_table(&dir
->header
);
1368 spin_unlock(&sysctl_lock
);
1376 * register_sysctl - register a sysctl table
1377 * @path: The path to the directory the sysctl table is in.
1378 * @table: the table structure
1380 * Register a sysctl table. @table should be a filled in ctl_table
1381 * array. A completely 0 filled entry terminates the table.
1383 * See __register_sysctl_table for more details.
1385 struct ctl_table_header
*register_sysctl(const char *path
, struct ctl_table
*table
)
1387 return __register_sysctl_table(&sysctl_table_root
.default_set
,
1390 EXPORT_SYMBOL(register_sysctl
);
1392 static char *append_path(const char *path
, char *pos
, const char *name
)
1395 namelen
= strlen(name
);
1396 if (((pos
- path
) + namelen
+ 2) >= PATH_MAX
)
1398 memcpy(pos
, name
, namelen
);
1400 pos
[namelen
+ 1] = '\0';
1405 static int count_subheaders(struct ctl_table
*table
)
1408 int nr_subheaders
= 0;
1409 struct ctl_table
*entry
;
1411 /* special case: no directory and empty directory */
1412 if (!table
|| !table
->procname
)
1415 for (entry
= table
; entry
->procname
; entry
++) {
1417 nr_subheaders
+= count_subheaders(entry
->child
);
1421 return nr_subheaders
+ has_files
;
1424 static int register_leaf_sysctl_tables(const char *path
, char *pos
,
1425 struct ctl_table_header
***subheader
, struct ctl_table_set
*set
,
1426 struct ctl_table
*table
)
1428 struct ctl_table
*ctl_table_arg
= NULL
;
1429 struct ctl_table
*entry
, *files
;
1434 for (entry
= table
; entry
->procname
; entry
++) {
1442 /* If there are mixed files and directories we need a new table */
1443 if (nr_dirs
&& nr_files
) {
1444 struct ctl_table
*new;
1445 files
= kcalloc(nr_files
+ 1, sizeof(struct ctl_table
),
1450 ctl_table_arg
= files
;
1451 for (new = files
, entry
= table
; entry
->procname
; entry
++) {
1459 /* Register everything except a directory full of subdirectories */
1460 if (nr_files
|| !nr_dirs
) {
1461 struct ctl_table_header
*header
;
1462 header
= __register_sysctl_table(set
, path
, files
);
1464 kfree(ctl_table_arg
);
1468 /* Remember if we need to free the file table */
1469 header
->ctl_table_arg
= ctl_table_arg
;
1470 **subheader
= header
;
1474 /* Recurse into the subdirectories. */
1475 for (entry
= table
; entry
->procname
; entry
++) {
1481 err
= -ENAMETOOLONG
;
1482 child_pos
= append_path(path
, pos
, entry
->procname
);
1486 err
= register_leaf_sysctl_tables(path
, child_pos
, subheader
,
1494 /* On failure our caller will unregister all registered subheaders */
1499 * __register_sysctl_paths - register a sysctl table hierarchy
1500 * @set: Sysctl tree to register on
1501 * @path: The path to the directory the sysctl table is in.
1502 * @table: the top-level table structure
1504 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1505 * array. A completely 0 filled entry terminates the table.
1507 * See __register_sysctl_table for more details.
1509 struct ctl_table_header
*__register_sysctl_paths(
1510 struct ctl_table_set
*set
,
1511 const struct ctl_path
*path
, struct ctl_table
*table
)
1513 struct ctl_table
*ctl_table_arg
= table
;
1514 int nr_subheaders
= count_subheaders(table
);
1515 struct ctl_table_header
*header
= NULL
, **subheaders
, **subheader
;
1516 const struct ctl_path
*component
;
1517 char *new_path
, *pos
;
1519 pos
= new_path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1524 for (component
= path
; component
->procname
; component
++) {
1525 pos
= append_path(new_path
, pos
, component
->procname
);
1529 while (table
->procname
&& table
->child
&& !table
[1].procname
) {
1530 pos
= append_path(new_path
, pos
, table
->procname
);
1533 table
= table
->child
;
1535 if (nr_subheaders
== 1) {
1536 header
= __register_sysctl_table(set
, new_path
, table
);
1538 header
->ctl_table_arg
= ctl_table_arg
;
1540 header
= kzalloc(sizeof(*header
) +
1541 sizeof(*subheaders
)*nr_subheaders
, GFP_KERNEL
);
1545 subheaders
= (struct ctl_table_header
**) (header
+ 1);
1546 subheader
= subheaders
;
1547 header
->ctl_table_arg
= ctl_table_arg
;
1549 if (register_leaf_sysctl_tables(new_path
, pos
, &subheader
,
1551 goto err_register_leaves
;
1558 err_register_leaves
:
1559 while (subheader
> subheaders
) {
1560 struct ctl_table_header
*subh
= *(--subheader
);
1561 struct ctl_table
*table
= subh
->ctl_table_arg
;
1562 unregister_sysctl_table(subh
);
1571 * register_sysctl_table_path - register a sysctl table hierarchy
1572 * @path: The path to the directory the sysctl table is in.
1573 * @table: the top-level table structure
1575 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1576 * array. A completely 0 filled entry terminates the table.
1578 * See __register_sysctl_paths for more details.
1580 struct ctl_table_header
*register_sysctl_paths(const struct ctl_path
*path
,
1581 struct ctl_table
*table
)
1583 return __register_sysctl_paths(&sysctl_table_root
.default_set
,
1586 EXPORT_SYMBOL(register_sysctl_paths
);
1589 * register_sysctl_table - register a sysctl table hierarchy
1590 * @table: the top-level table structure
1592 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1593 * array. A completely 0 filled entry terminates the table.
1595 * See register_sysctl_paths for more details.
1597 struct ctl_table_header
*register_sysctl_table(struct ctl_table
*table
)
1599 static const struct ctl_path null_path
[] = { {} };
1601 return register_sysctl_paths(null_path
, table
);
1603 EXPORT_SYMBOL(register_sysctl_table
);
1605 static void put_links(struct ctl_table_header
*header
)
1607 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1608 struct ctl_table_root
*root
= header
->root
;
1609 struct ctl_dir
*parent
= header
->parent
;
1610 struct ctl_dir
*core_parent
;
1611 struct ctl_table
*entry
;
1613 if (header
->set
== root_set
)
1616 core_parent
= xlate_dir(root_set
, parent
);
1617 if (IS_ERR(core_parent
))
1620 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
1621 struct ctl_table_header
*link_head
;
1622 struct ctl_table
*link
;
1623 const char *name
= entry
->procname
;
1625 link
= find_entry(&link_head
, core_parent
, name
, strlen(name
));
1627 ((S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
)) ||
1628 (S_ISLNK(link
->mode
) && (link
->data
== root
)))) {
1629 drop_sysctl_table(link_head
);
1632 pr_err("sysctl link missing during unregister: ");
1633 sysctl_print_dir(parent
);
1634 pr_cont("/%s\n", name
);
1639 static void drop_sysctl_table(struct ctl_table_header
*header
)
1641 struct ctl_dir
*parent
= header
->parent
;
1648 start_unregistering(header
);
1649 if (!--header
->count
)
1650 kfree_rcu(header
, rcu
);
1653 drop_sysctl_table(&parent
->header
);
1657 * unregister_sysctl_table - unregister a sysctl table hierarchy
1658 * @header: the header returned from register_sysctl_table
1660 * Unregisters the sysctl table and all children. proc entries may not
1661 * actually be removed until they are no longer used by anyone.
1663 void unregister_sysctl_table(struct ctl_table_header
* header
)
1671 nr_subheaders
= count_subheaders(header
->ctl_table_arg
);
1672 if (unlikely(nr_subheaders
> 1)) {
1673 struct ctl_table_header
**subheaders
;
1676 subheaders
= (struct ctl_table_header
**)(header
+ 1);
1677 for (i
= nr_subheaders
-1; i
>= 0; i
--) {
1678 struct ctl_table_header
*subh
= subheaders
[i
];
1679 struct ctl_table
*table
= subh
->ctl_table_arg
;
1680 unregister_sysctl_table(subh
);
1687 spin_lock(&sysctl_lock
);
1688 drop_sysctl_table(header
);
1689 spin_unlock(&sysctl_lock
);
1691 EXPORT_SYMBOL(unregister_sysctl_table
);
1693 void setup_sysctl_set(struct ctl_table_set
*set
,
1694 struct ctl_table_root
*root
,
1695 int (*is_seen
)(struct ctl_table_set
*))
1697 memset(set
, 0, sizeof(*set
));
1698 set
->is_seen
= is_seen
;
1699 init_header(&set
->dir
.header
, root
, set
, NULL
, root_table
);
1702 void retire_sysctl_set(struct ctl_table_set
*set
)
1704 WARN_ON(!RB_EMPTY_ROOT(&set
->dir
.root
));
1707 int __init
proc_sys_init(void)
1709 struct proc_dir_entry
*proc_sys_root
;
1711 proc_sys_root
= proc_mkdir("sys", NULL
);
1712 proc_sys_root
->proc_iops
= &proc_sys_dir_operations
;
1713 proc_sys_root
->proc_fops
= &proc_sys_dir_file_operations
;
1714 proc_sys_root
->nlink
= 0;
1716 return sysctl_init();