4 #include <linux/init.h>
5 #include <linux/sysctl.h>
6 #include <linux/poll.h>
7 #include <linux/proc_fs.h>
8 #include <linux/security.h>
9 #include <linux/sched.h>
10 #include <linux/namei.h>
12 #include <linux/module.h>
15 static const struct dentry_operations proc_sys_dentry_operations
;
16 static const struct file_operations proc_sys_file_operations
;
17 static const struct inode_operations proc_sys_inode_operations
;
18 static const struct file_operations proc_sys_dir_file_operations
;
19 static const struct inode_operations proc_sys_dir_operations
;
21 void proc_sys_poll_notify(struct ctl_table_poll
*poll
)
26 atomic_inc(&poll
->event
);
27 wake_up_interruptible(&poll
->wait
);
30 static struct ctl_table root_table
[] = {
33 .mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
,
37 static struct ctl_table_root sysctl_table_root
= {
38 .default_set
.dir
.header
= {
41 .ctl_table
= root_table
}},
42 .ctl_table_arg
= root_table
,
43 .root
= &sysctl_table_root
,
44 .set
= &sysctl_table_root
.default_set
,
48 static DEFINE_SPINLOCK(sysctl_lock
);
50 static void drop_sysctl_table(struct ctl_table_header
*header
);
51 static int sysctl_follow_link(struct ctl_table_header
**phead
,
52 struct ctl_table
**pentry
, struct nsproxy
*namespaces
);
53 static int insert_links(struct ctl_table_header
*head
);
54 static void put_links(struct ctl_table_header
*header
);
56 static void sysctl_print_dir(struct ctl_dir
*dir
)
58 if (dir
->header
.parent
)
59 sysctl_print_dir(dir
->header
.parent
);
60 printk(KERN_CONT
"%s/", dir
->header
.ctl_table
[0].procname
);
63 static int namecmp(const char *name1
, int len1
, const char *name2
, int len2
)
72 cmp
= memcmp(name1
, name2
, minlen
);
78 /* Called under sysctl_lock */
79 static struct ctl_table
*find_entry(struct ctl_table_header
**phead
,
80 struct ctl_dir
*dir
, const char *name
, int namelen
)
82 struct ctl_table_header
*head
;
83 struct ctl_table
*entry
;
84 struct rb_node
*node
= dir
->root
.rb_node
;
88 struct ctl_node
*ctl_node
;
92 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
93 head
= ctl_node
->header
;
94 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
95 procname
= entry
->procname
;
97 cmp
= namecmp(name
, namelen
, procname
, strlen(procname
));
101 node
= node
->rb_right
;
110 static int insert_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
112 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
113 struct rb_node
**p
= &head
->parent
->root
.rb_node
;
114 struct rb_node
*parent
= NULL
;
115 const char *name
= entry
->procname
;
116 int namelen
= strlen(name
);
119 struct ctl_table_header
*parent_head
;
120 struct ctl_table
*parent_entry
;
121 struct ctl_node
*parent_node
;
122 const char *parent_name
;
126 parent_node
= rb_entry(parent
, struct ctl_node
, node
);
127 parent_head
= parent_node
->header
;
128 parent_entry
= &parent_head
->ctl_table
[parent_node
- parent_head
->node
];
129 parent_name
= parent_entry
->procname
;
131 cmp
= namecmp(name
, namelen
, parent_name
, strlen(parent_name
));
137 printk(KERN_ERR
"sysctl duplicate entry: ");
138 sysctl_print_dir(head
->parent
);
139 printk(KERN_CONT
"/%s\n", entry
->procname
);
144 rb_link_node(node
, parent
, p
);
145 rb_insert_color(node
, &head
->parent
->root
);
149 static void erase_entry(struct ctl_table_header
*head
, struct ctl_table
*entry
)
151 struct rb_node
*node
= &head
->node
[entry
- head
->ctl_table
].node
;
153 rb_erase(node
, &head
->parent
->root
);
156 static void init_header(struct ctl_table_header
*head
,
157 struct ctl_table_root
*root
, struct ctl_table_set
*set
,
158 struct ctl_node
*node
, struct ctl_table
*table
)
160 head
->ctl_table
= table
;
161 head
->ctl_table_arg
= table
;
165 head
->unregistering
= NULL
;
171 struct ctl_table
*entry
;
172 for (entry
= table
; entry
->procname
; entry
++, node
++)
177 static void erase_header(struct ctl_table_header
*head
)
179 struct ctl_table
*entry
;
180 for (entry
= head
->ctl_table
; entry
->procname
; entry
++)
181 erase_entry(head
, entry
);
184 static int insert_header(struct ctl_dir
*dir
, struct ctl_table_header
*header
)
186 struct ctl_table
*entry
;
190 header
->parent
= dir
;
191 err
= insert_links(header
);
194 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
195 err
= insert_entry(header
, entry
);
201 erase_header(header
);
204 header
->parent
= NULL
;
205 drop_sysctl_table(&dir
->header
);
209 /* called under sysctl_lock */
210 static int use_table(struct ctl_table_header
*p
)
212 if (unlikely(p
->unregistering
))
218 /* called under sysctl_lock */
219 static void unuse_table(struct ctl_table_header
*p
)
222 if (unlikely(p
->unregistering
))
223 complete(p
->unregistering
);
226 /* called under sysctl_lock, will reacquire if has to wait */
227 static void start_unregistering(struct ctl_table_header
*p
)
230 * if p->used is 0, nobody will ever touch that entry again;
231 * we'll eliminate all paths to it before dropping sysctl_lock
233 if (unlikely(p
->used
)) {
234 struct completion wait
;
235 init_completion(&wait
);
236 p
->unregistering
= &wait
;
237 spin_unlock(&sysctl_lock
);
238 wait_for_completion(&wait
);
239 spin_lock(&sysctl_lock
);
241 /* anything non-NULL; we'll never dereference it */
242 p
->unregistering
= ERR_PTR(-EINVAL
);
245 * do not remove from the list until nobody holds it; walking the
246 * list in do_sysctl() relies on that.
251 static void sysctl_head_get(struct ctl_table_header
*head
)
253 spin_lock(&sysctl_lock
);
255 spin_unlock(&sysctl_lock
);
258 void sysctl_head_put(struct ctl_table_header
*head
)
260 spin_lock(&sysctl_lock
);
262 kfree_rcu(head
, rcu
);
263 spin_unlock(&sysctl_lock
);
266 static struct ctl_table_header
*sysctl_head_grab(struct ctl_table_header
*head
)
269 spin_lock(&sysctl_lock
);
270 if (!use_table(head
))
271 head
= ERR_PTR(-ENOENT
);
272 spin_unlock(&sysctl_lock
);
276 static void sysctl_head_finish(struct ctl_table_header
*head
)
280 spin_lock(&sysctl_lock
);
282 spin_unlock(&sysctl_lock
);
285 static struct ctl_table_set
*
286 lookup_header_set(struct ctl_table_root
*root
, struct nsproxy
*namespaces
)
288 struct ctl_table_set
*set
= &root
->default_set
;
290 set
= root
->lookup(root
, namespaces
);
294 static struct ctl_table
*lookup_entry(struct ctl_table_header
**phead
,
296 const char *name
, int namelen
)
298 struct ctl_table_header
*head
;
299 struct ctl_table
*entry
;
301 spin_lock(&sysctl_lock
);
302 entry
= find_entry(&head
, dir
, name
, namelen
);
303 if (entry
&& use_table(head
))
307 spin_unlock(&sysctl_lock
);
311 static struct ctl_node
*first_usable_entry(struct rb_node
*node
)
313 struct ctl_node
*ctl_node
;
315 for (;node
; node
= rb_next(node
)) {
316 ctl_node
= rb_entry(node
, struct ctl_node
, node
);
317 if (use_table(ctl_node
->header
))
323 static void first_entry(struct ctl_dir
*dir
,
324 struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
326 struct ctl_table_header
*head
= NULL
;
327 struct ctl_table
*entry
= NULL
;
328 struct ctl_node
*ctl_node
;
330 spin_lock(&sysctl_lock
);
331 ctl_node
= first_usable_entry(rb_first(&dir
->root
));
332 spin_unlock(&sysctl_lock
);
334 head
= ctl_node
->header
;
335 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
341 static void next_entry(struct ctl_table_header
**phead
, struct ctl_table
**pentry
)
343 struct ctl_table_header
*head
= *phead
;
344 struct ctl_table
*entry
= *pentry
;
345 struct ctl_node
*ctl_node
= &head
->node
[entry
- head
->ctl_table
];
347 spin_lock(&sysctl_lock
);
350 ctl_node
= first_usable_entry(rb_next(&ctl_node
->node
));
351 spin_unlock(&sysctl_lock
);
354 head
= ctl_node
->header
;
355 entry
= &head
->ctl_table
[ctl_node
- head
->node
];
361 void register_sysctl_root(struct ctl_table_root
*root
)
366 * sysctl_perm does NOT grant the superuser all rights automatically, because
367 * some sysctl variables are readonly even to root.
370 static int test_perm(int mode
, int op
)
372 if (uid_eq(current_euid(), GLOBAL_ROOT_UID
))
374 else if (in_egroup_p(GLOBAL_ROOT_GID
))
376 if ((op
& ~mode
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
)) == 0)
381 static int sysctl_perm(struct ctl_table_root
*root
, struct ctl_table
*table
, int op
)
385 if (root
->permissions
)
386 mode
= root
->permissions(root
, current
->nsproxy
, table
);
390 return test_perm(mode
, op
);
393 static struct inode
*proc_sys_make_inode(struct super_block
*sb
,
394 struct ctl_table_header
*head
, struct ctl_table
*table
)
397 struct proc_inode
*ei
;
399 inode
= new_inode(sb
);
403 inode
->i_ino
= get_next_ino();
405 sysctl_head_get(head
);
408 ei
->sysctl_entry
= table
;
410 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
411 inode
->i_mode
= table
->mode
;
412 if (!S_ISDIR(table
->mode
)) {
413 inode
->i_mode
|= S_IFREG
;
414 inode
->i_op
= &proc_sys_inode_operations
;
415 inode
->i_fop
= &proc_sys_file_operations
;
417 inode
->i_mode
|= S_IFDIR
;
418 inode
->i_op
= &proc_sys_dir_operations
;
419 inode
->i_fop
= &proc_sys_dir_file_operations
;
425 static struct ctl_table_header
*grab_header(struct inode
*inode
)
427 struct ctl_table_header
*head
= PROC_I(inode
)->sysctl
;
429 head
= &sysctl_table_root
.default_set
.dir
.header
;
430 return sysctl_head_grab(head
);
433 static struct dentry
*proc_sys_lookup(struct inode
*dir
, struct dentry
*dentry
,
436 struct ctl_table_header
*head
= grab_header(dir
);
437 struct ctl_table_header
*h
= NULL
;
438 struct qstr
*name
= &dentry
->d_name
;
441 struct dentry
*err
= ERR_PTR(-ENOENT
);
442 struct ctl_dir
*ctl_dir
;
446 return ERR_CAST(head
);
448 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
450 p
= lookup_entry(&h
, ctl_dir
, name
->name
, name
->len
);
454 if (S_ISLNK(p
->mode
)) {
455 ret
= sysctl_follow_link(&h
, &p
, current
->nsproxy
);
461 err
= ERR_PTR(-ENOMEM
);
462 inode
= proc_sys_make_inode(dir
->i_sb
, h
? h
: head
, p
);
467 d_set_d_op(dentry
, &proc_sys_dentry_operations
);
468 d_add(dentry
, inode
);
472 sysctl_head_finish(h
);
473 sysctl_head_finish(head
);
477 static ssize_t
proc_sys_call_handler(struct file
*filp
, void __user
*buf
,
478 size_t count
, loff_t
*ppos
, int write
)
480 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
481 struct ctl_table_header
*head
= grab_header(inode
);
482 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
487 return PTR_ERR(head
);
490 * At this point we know that the sysctl was not unregistered
491 * and won't be until we finish.
494 if (sysctl_perm(head
->root
, table
, write
? MAY_WRITE
: MAY_READ
))
497 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
499 if (!table
->proc_handler
)
502 /* careful: calling conventions are nasty here */
504 error
= table
->proc_handler(table
, write
, buf
, &res
, ppos
);
508 sysctl_head_finish(head
);
513 static ssize_t
proc_sys_read(struct file
*filp
, char __user
*buf
,
514 size_t count
, loff_t
*ppos
)
516 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 0);
519 static ssize_t
proc_sys_write(struct file
*filp
, const char __user
*buf
,
520 size_t count
, loff_t
*ppos
)
522 return proc_sys_call_handler(filp
, (void __user
*)buf
, count
, ppos
, 1);
525 static int proc_sys_open(struct inode
*inode
, struct file
*filp
)
527 struct ctl_table_header
*head
= grab_header(inode
);
528 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
530 /* sysctl was unregistered */
532 return PTR_ERR(head
);
535 filp
->private_data
= proc_sys_poll_event(table
->poll
);
537 sysctl_head_finish(head
);
542 static unsigned int proc_sys_poll(struct file
*filp
, poll_table
*wait
)
544 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
545 struct ctl_table_header
*head
= grab_header(inode
);
546 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
547 unsigned int ret
= DEFAULT_POLLMASK
;
550 /* sysctl was unregistered */
552 return POLLERR
| POLLHUP
;
554 if (!table
->proc_handler
)
560 event
= (unsigned long)filp
->private_data
;
561 poll_wait(filp
, &table
->poll
->wait
, wait
);
563 if (event
!= atomic_read(&table
->poll
->event
)) {
564 filp
->private_data
= proc_sys_poll_event(table
->poll
);
565 ret
= POLLIN
| POLLRDNORM
| POLLERR
| POLLPRI
;
569 sysctl_head_finish(head
);
574 static int proc_sys_fill_cache(struct file
*filp
, void *dirent
,
576 struct ctl_table_header
*head
,
577 struct ctl_table
*table
)
579 struct dentry
*child
, *dir
= filp
->f_path
.dentry
;
583 unsigned type
= DT_UNKNOWN
;
585 qname
.name
= table
->procname
;
586 qname
.len
= strlen(table
->procname
);
587 qname
.hash
= full_name_hash(qname
.name
, qname
.len
);
589 child
= d_lookup(dir
, &qname
);
591 child
= d_alloc(dir
, &qname
);
593 inode
= proc_sys_make_inode(dir
->d_sb
, head
, table
);
598 d_set_d_op(child
, &proc_sys_dentry_operations
);
605 inode
= child
->d_inode
;
607 type
= inode
->i_mode
>> 12;
609 return !!filldir(dirent
, qname
.name
, qname
.len
, filp
->f_pos
, ino
, type
);
612 static int proc_sys_link_fill_cache(struct file
*filp
, void *dirent
,
614 struct ctl_table_header
*head
,
615 struct ctl_table
*table
)
618 head
= sysctl_head_grab(head
);
620 if (S_ISLNK(table
->mode
)) {
621 /* It is not an error if we can not follow the link ignore it */
622 err
= sysctl_follow_link(&head
, &table
, current
->nsproxy
);
627 ret
= proc_sys_fill_cache(filp
, dirent
, filldir
, head
, table
);
629 sysctl_head_finish(head
);
633 static int scan(struct ctl_table_header
*head
, ctl_table
*table
,
634 unsigned long *pos
, struct file
*file
,
635 void *dirent
, filldir_t filldir
)
639 if ((*pos
)++ < file
->f_pos
)
642 if (unlikely(S_ISLNK(table
->mode
)))
643 res
= proc_sys_link_fill_cache(file
, dirent
, filldir
, head
, table
);
645 res
= proc_sys_fill_cache(file
, dirent
, filldir
, head
, table
);
653 static int proc_sys_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
655 struct dentry
*dentry
= filp
->f_path
.dentry
;
656 struct inode
*inode
= dentry
->d_inode
;
657 struct ctl_table_header
*head
= grab_header(inode
);
658 struct ctl_table_header
*h
= NULL
;
659 struct ctl_table
*entry
;
660 struct ctl_dir
*ctl_dir
;
665 return PTR_ERR(head
);
667 ctl_dir
= container_of(head
, struct ctl_dir
, header
);
670 /* Avoid a switch here: arm builds fail with missing __cmpdi2 */
671 if (filp
->f_pos
== 0) {
672 if (filldir(dirent
, ".", 1, filp
->f_pos
,
673 inode
->i_ino
, DT_DIR
) < 0)
677 if (filp
->f_pos
== 1) {
678 if (filldir(dirent
, "..", 2, filp
->f_pos
,
679 parent_ino(dentry
), DT_DIR
) < 0)
685 for (first_entry(ctl_dir
, &h
, &entry
); h
; next_entry(&h
, &entry
)) {
686 ret
= scan(h
, entry
, &pos
, filp
, dirent
, filldir
);
688 sysctl_head_finish(h
);
694 sysctl_head_finish(head
);
698 static int proc_sys_permission(struct inode
*inode
, int mask
)
701 * sysctl entries that are not writeable,
702 * are _NOT_ writeable, capabilities or not.
704 struct ctl_table_header
*head
;
705 struct ctl_table
*table
;
708 /* Executable files are not allowed under /proc/sys/ */
709 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
))
712 head
= grab_header(inode
);
714 return PTR_ERR(head
);
716 table
= PROC_I(inode
)->sysctl_entry
;
717 if (!table
) /* global root - r-xr-xr-x */
718 error
= mask
& MAY_WRITE
? -EACCES
: 0;
719 else /* Use the permissions on the sysctl table entry */
720 error
= sysctl_perm(head
->root
, table
, mask
& ~MAY_NOT_BLOCK
);
722 sysctl_head_finish(head
);
726 static int proc_sys_setattr(struct dentry
*dentry
, struct iattr
*attr
)
728 struct inode
*inode
= dentry
->d_inode
;
731 if (attr
->ia_valid
& (ATTR_MODE
| ATTR_UID
| ATTR_GID
))
734 error
= inode_change_ok(inode
, attr
);
738 if ((attr
->ia_valid
& ATTR_SIZE
) &&
739 attr
->ia_size
!= i_size_read(inode
)) {
740 error
= vmtruncate(inode
, attr
->ia_size
);
745 setattr_copy(inode
, attr
);
746 mark_inode_dirty(inode
);
750 static int proc_sys_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
, struct kstat
*stat
)
752 struct inode
*inode
= dentry
->d_inode
;
753 struct ctl_table_header
*head
= grab_header(inode
);
754 struct ctl_table
*table
= PROC_I(inode
)->sysctl_entry
;
757 return PTR_ERR(head
);
759 generic_fillattr(inode
, stat
);
761 stat
->mode
= (stat
->mode
& S_IFMT
) | table
->mode
;
763 sysctl_head_finish(head
);
767 static const struct file_operations proc_sys_file_operations
= {
768 .open
= proc_sys_open
,
769 .poll
= proc_sys_poll
,
770 .read
= proc_sys_read
,
771 .write
= proc_sys_write
,
772 .llseek
= default_llseek
,
775 static const struct file_operations proc_sys_dir_file_operations
= {
776 .read
= generic_read_dir
,
777 .readdir
= proc_sys_readdir
,
778 .llseek
= generic_file_llseek
,
781 static const struct inode_operations proc_sys_inode_operations
= {
782 .permission
= proc_sys_permission
,
783 .setattr
= proc_sys_setattr
,
784 .getattr
= proc_sys_getattr
,
787 static const struct inode_operations proc_sys_dir_operations
= {
788 .lookup
= proc_sys_lookup
,
789 .permission
= proc_sys_permission
,
790 .setattr
= proc_sys_setattr
,
791 .getattr
= proc_sys_getattr
,
794 static int proc_sys_revalidate(struct dentry
*dentry
, unsigned int flags
)
796 if (flags
& LOOKUP_RCU
)
798 return !PROC_I(dentry
->d_inode
)->sysctl
->unregistering
;
801 static int proc_sys_delete(const struct dentry
*dentry
)
803 return !!PROC_I(dentry
->d_inode
)->sysctl
->unregistering
;
806 static int sysctl_is_seen(struct ctl_table_header
*p
)
808 struct ctl_table_set
*set
= p
->set
;
810 spin_lock(&sysctl_lock
);
811 if (p
->unregistering
)
813 else if (!set
->is_seen
)
816 res
= set
->is_seen(set
);
817 spin_unlock(&sysctl_lock
);
821 static int proc_sys_compare(const struct dentry
*parent
,
822 const struct inode
*pinode
,
823 const struct dentry
*dentry
, const struct inode
*inode
,
824 unsigned int len
, const char *str
, const struct qstr
*name
)
826 struct ctl_table_header
*head
;
827 /* Although proc doesn't have negative dentries, rcu-walk means
828 * that inode here can be NULL */
829 /* AV: can it, indeed? */
832 if (name
->len
!= len
)
834 if (memcmp(name
->name
, str
, len
))
836 head
= rcu_dereference(PROC_I(inode
)->sysctl
);
837 return !head
|| !sysctl_is_seen(head
);
840 static const struct dentry_operations proc_sys_dentry_operations
= {
841 .d_revalidate
= proc_sys_revalidate
,
842 .d_delete
= proc_sys_delete
,
843 .d_compare
= proc_sys_compare
,
846 static struct ctl_dir
*find_subdir(struct ctl_dir
*dir
,
847 const char *name
, int namelen
)
849 struct ctl_table_header
*head
;
850 struct ctl_table
*entry
;
852 entry
= find_entry(&head
, dir
, name
, namelen
);
854 return ERR_PTR(-ENOENT
);
855 if (!S_ISDIR(entry
->mode
))
856 return ERR_PTR(-ENOTDIR
);
857 return container_of(head
, struct ctl_dir
, header
);
860 static struct ctl_dir
*new_dir(struct ctl_table_set
*set
,
861 const char *name
, int namelen
)
863 struct ctl_table
*table
;
865 struct ctl_node
*node
;
868 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node
) +
869 sizeof(struct ctl_table
)*2 + namelen
+ 1,
874 node
= (struct ctl_node
*)(new + 1);
875 table
= (struct ctl_table
*)(node
+ 1);
876 new_name
= (char *)(table
+ 2);
877 memcpy(new_name
, name
, namelen
);
878 new_name
[namelen
] = '\0';
879 table
[0].procname
= new_name
;
880 table
[0].mode
= S_IFDIR
|S_IRUGO
|S_IXUGO
;
881 init_header(&new->header
, set
->dir
.header
.root
, set
, node
, table
);
887 * get_subdir - find or create a subdir with the specified name.
888 * @dir: Directory to create the subdirectory in
889 * @name: The name of the subdirectory to find or create
890 * @namelen: The length of name
892 * Takes a directory with an elevated reference count so we know that
893 * if we drop the lock the directory will not go away. Upon success
894 * the reference is moved from @dir to the returned subdirectory.
895 * Upon error an error code is returned and the reference on @dir is
898 static struct ctl_dir
*get_subdir(struct ctl_dir
*dir
,
899 const char *name
, int namelen
)
901 struct ctl_table_set
*set
= dir
->header
.set
;
902 struct ctl_dir
*subdir
, *new = NULL
;
905 spin_lock(&sysctl_lock
);
906 subdir
= find_subdir(dir
, name
, namelen
);
909 if (PTR_ERR(subdir
) != -ENOENT
)
912 spin_unlock(&sysctl_lock
);
913 new = new_dir(set
, name
, namelen
);
914 spin_lock(&sysctl_lock
);
915 subdir
= ERR_PTR(-ENOMEM
);
919 /* Was the subdir added while we dropped the lock? */
920 subdir
= find_subdir(dir
, name
, namelen
);
923 if (PTR_ERR(subdir
) != -ENOENT
)
926 /* Nope. Use the our freshly made directory entry. */
927 err
= insert_header(dir
, &new->header
);
928 subdir
= ERR_PTR(err
);
933 subdir
->header
.nreg
++;
935 if (unlikely(IS_ERR(subdir
))) {
936 printk(KERN_ERR
"sysctl could not get directory: ");
937 sysctl_print_dir(dir
);
938 printk(KERN_CONT
"/%*.*s %ld\n",
939 namelen
, namelen
, name
, PTR_ERR(subdir
));
941 drop_sysctl_table(&dir
->header
);
943 drop_sysctl_table(&new->header
);
944 spin_unlock(&sysctl_lock
);
948 static struct ctl_dir
*xlate_dir(struct ctl_table_set
*set
, struct ctl_dir
*dir
)
950 struct ctl_dir
*parent
;
951 const char *procname
;
952 if (!dir
->header
.parent
)
954 parent
= xlate_dir(set
, dir
->header
.parent
);
957 procname
= dir
->header
.ctl_table
[0].procname
;
958 return find_subdir(parent
, procname
, strlen(procname
));
961 static int sysctl_follow_link(struct ctl_table_header
**phead
,
962 struct ctl_table
**pentry
, struct nsproxy
*namespaces
)
964 struct ctl_table_header
*head
;
965 struct ctl_table_root
*root
;
966 struct ctl_table_set
*set
;
967 struct ctl_table
*entry
;
972 spin_lock(&sysctl_lock
);
973 root
= (*pentry
)->data
;
974 set
= lookup_header_set(root
, namespaces
);
975 dir
= xlate_dir(set
, (*phead
)->parent
);
979 const char *procname
= (*pentry
)->procname
;
981 entry
= find_entry(&head
, dir
, procname
, strlen(procname
));
983 if (entry
&& use_table(head
)) {
991 spin_unlock(&sysctl_lock
);
995 static int sysctl_err(const char *path
, struct ctl_table
*table
, char *fmt
, ...)
997 struct va_format vaf
;
1000 va_start(args
, fmt
);
1004 printk(KERN_ERR
"sysctl table check failed: %s/%s %pV\n",
1005 path
, table
->procname
, &vaf
);
1011 static int sysctl_check_table(const char *path
, struct ctl_table
*table
)
1014 for (; table
->procname
; table
++) {
1016 err
= sysctl_err(path
, table
, "Not a file");
1018 if ((table
->proc_handler
== proc_dostring
) ||
1019 (table
->proc_handler
== proc_dointvec
) ||
1020 (table
->proc_handler
== proc_dointvec_minmax
) ||
1021 (table
->proc_handler
== proc_dointvec_jiffies
) ||
1022 (table
->proc_handler
== proc_dointvec_userhz_jiffies
) ||
1023 (table
->proc_handler
== proc_dointvec_ms_jiffies
) ||
1024 (table
->proc_handler
== proc_doulongvec_minmax
) ||
1025 (table
->proc_handler
== proc_doulongvec_ms_jiffies_minmax
)) {
1027 err
= sysctl_err(path
, table
, "No data");
1029 err
= sysctl_err(path
, table
, "No maxlen");
1031 if (!table
->proc_handler
)
1032 err
= sysctl_err(path
, table
, "No proc_handler");
1034 if ((table
->mode
& (S_IRUGO
|S_IWUGO
)) != table
->mode
)
1035 err
= sysctl_err(path
, table
, "bogus .mode 0%o",
1041 static struct ctl_table_header
*new_links(struct ctl_dir
*dir
, struct ctl_table
*table
,
1042 struct ctl_table_root
*link_root
)
1044 struct ctl_table
*link_table
, *entry
, *link
;
1045 struct ctl_table_header
*links
;
1046 struct ctl_node
*node
;
1048 int nr_entries
, name_bytes
;
1052 for (entry
= table
; entry
->procname
; entry
++) {
1054 name_bytes
+= strlen(entry
->procname
) + 1;
1057 links
= kzalloc(sizeof(struct ctl_table_header
) +
1058 sizeof(struct ctl_node
)*nr_entries
+
1059 sizeof(struct ctl_table
)*(nr_entries
+ 1) +
1066 node
= (struct ctl_node
*)(links
+ 1);
1067 link_table
= (struct ctl_table
*)(node
+ nr_entries
);
1068 link_name
= (char *)&link_table
[nr_entries
+ 1];
1070 for (link
= link_table
, entry
= table
; entry
->procname
; link
++, entry
++) {
1071 int len
= strlen(entry
->procname
) + 1;
1072 memcpy(link_name
, entry
->procname
, len
);
1073 link
->procname
= link_name
;
1074 link
->mode
= S_IFLNK
|S_IRWXUGO
;
1075 link
->data
= link_root
;
1078 init_header(links
, dir
->header
.root
, dir
->header
.set
, node
, link_table
);
1079 links
->nreg
= nr_entries
;
1084 static bool get_links(struct ctl_dir
*dir
,
1085 struct ctl_table
*table
, struct ctl_table_root
*link_root
)
1087 struct ctl_table_header
*head
;
1088 struct ctl_table
*entry
, *link
;
1090 /* Are there links available for every entry in table? */
1091 for (entry
= table
; entry
->procname
; entry
++) {
1092 const char *procname
= entry
->procname
;
1093 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1096 if (S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
))
1098 if (S_ISLNK(link
->mode
) && (link
->data
== link_root
))
1103 /* The checks passed. Increase the registration count on the links */
1104 for (entry
= table
; entry
->procname
; entry
++) {
1105 const char *procname
= entry
->procname
;
1106 link
= find_entry(&head
, dir
, procname
, strlen(procname
));
1112 static int insert_links(struct ctl_table_header
*head
)
1114 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1115 struct ctl_dir
*core_parent
= NULL
;
1116 struct ctl_table_header
*links
;
1119 if (head
->set
== root_set
)
1122 core_parent
= xlate_dir(root_set
, head
->parent
);
1123 if (IS_ERR(core_parent
))
1126 if (get_links(core_parent
, head
->ctl_table
, head
->root
))
1129 core_parent
->header
.nreg
++;
1130 spin_unlock(&sysctl_lock
);
1132 links
= new_links(core_parent
, head
->ctl_table
, head
->root
);
1134 spin_lock(&sysctl_lock
);
1140 if (get_links(core_parent
, head
->ctl_table
, head
->root
)) {
1145 err
= insert_header(core_parent
, links
);
1149 drop_sysctl_table(&core_parent
->header
);
1154 * __register_sysctl_table - register a leaf sysctl table
1155 * @set: Sysctl tree to register on
1156 * @path: The path to the directory the sysctl table is in.
1157 * @table: the top-level table structure
1159 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1160 * array. A completely 0 filled entry terminates the table.
1162 * The members of the &struct ctl_table structure are used as follows:
1164 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1165 * enter a sysctl file
1167 * data - a pointer to data for use by proc_handler
1169 * maxlen - the maximum size in bytes of the data
1171 * mode - the file permissions for the /proc/sys file
1173 * child - must be %NULL.
1175 * proc_handler - the text handler routine (described below)
1177 * extra1, extra2 - extra pointers usable by the proc handler routines
1179 * Leaf nodes in the sysctl tree will be represented by a single file
1180 * under /proc; non-leaf nodes will be represented by directories.
1182 * There must be a proc_handler routine for any terminal nodes.
1183 * Several default handlers are available to cover common cases -
1185 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1186 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1187 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1189 * It is the handler's job to read the input buffer from user memory
1190 * and process it. The handler should return 0 on success.
1192 * This routine returns %NULL on a failure to register, and a pointer
1193 * to the table header on success.
1195 struct ctl_table_header
*__register_sysctl_table(
1196 struct ctl_table_set
*set
,
1197 const char *path
, struct ctl_table
*table
)
1199 struct ctl_table_root
*root
= set
->dir
.header
.root
;
1200 struct ctl_table_header
*header
;
1201 const char *name
, *nextname
;
1202 struct ctl_dir
*dir
;
1203 struct ctl_table
*entry
;
1204 struct ctl_node
*node
;
1207 for (entry
= table
; entry
->procname
; entry
++)
1210 header
= kzalloc(sizeof(struct ctl_table_header
) +
1211 sizeof(struct ctl_node
)*nr_entries
, GFP_KERNEL
);
1215 node
= (struct ctl_node
*)(header
+ 1);
1216 init_header(header
, root
, set
, node
, table
);
1217 if (sysctl_check_table(path
, table
))
1220 spin_lock(&sysctl_lock
);
1222 /* Reference moved down the diretory tree get_subdir */
1224 spin_unlock(&sysctl_lock
);
1226 /* Find the directory for the ctl_table */
1227 for (name
= path
; name
; name
= nextname
) {
1229 nextname
= strchr(name
, '/');
1231 namelen
= nextname
- name
;
1234 namelen
= strlen(name
);
1239 dir
= get_subdir(dir
, name
, namelen
);
1244 spin_lock(&sysctl_lock
);
1245 if (insert_header(dir
, header
))
1246 goto fail_put_dir_locked
;
1248 drop_sysctl_table(&dir
->header
);
1249 spin_unlock(&sysctl_lock
);
1253 fail_put_dir_locked
:
1254 drop_sysctl_table(&dir
->header
);
1255 spin_unlock(&sysctl_lock
);
1263 * register_sysctl - register a sysctl table
1264 * @path: The path to the directory the sysctl table is in.
1265 * @table: the table structure
1267 * Register a sysctl table. @table should be a filled in ctl_table
1268 * array. A completely 0 filled entry terminates the table.
1270 * See __register_sysctl_table for more details.
1272 struct ctl_table_header
*register_sysctl(const char *path
, struct ctl_table
*table
)
1274 return __register_sysctl_table(&sysctl_table_root
.default_set
,
1277 EXPORT_SYMBOL(register_sysctl
);
1279 static char *append_path(const char *path
, char *pos
, const char *name
)
1282 namelen
= strlen(name
);
1283 if (((pos
- path
) + namelen
+ 2) >= PATH_MAX
)
1285 memcpy(pos
, name
, namelen
);
1287 pos
[namelen
+ 1] = '\0';
1292 static int count_subheaders(struct ctl_table
*table
)
1295 int nr_subheaders
= 0;
1296 struct ctl_table
*entry
;
1298 /* special case: no directory and empty directory */
1299 if (!table
|| !table
->procname
)
1302 for (entry
= table
; entry
->procname
; entry
++) {
1304 nr_subheaders
+= count_subheaders(entry
->child
);
1308 return nr_subheaders
+ has_files
;
1311 static int register_leaf_sysctl_tables(const char *path
, char *pos
,
1312 struct ctl_table_header
***subheader
, struct ctl_table_set
*set
,
1313 struct ctl_table
*table
)
1315 struct ctl_table
*ctl_table_arg
= NULL
;
1316 struct ctl_table
*entry
, *files
;
1321 for (entry
= table
; entry
->procname
; entry
++) {
1329 /* If there are mixed files and directories we need a new table */
1330 if (nr_dirs
&& nr_files
) {
1331 struct ctl_table
*new;
1332 files
= kzalloc(sizeof(struct ctl_table
) * (nr_files
+ 1),
1337 ctl_table_arg
= files
;
1338 for (new = files
, entry
= table
; entry
->procname
; entry
++) {
1346 /* Register everything except a directory full of subdirectories */
1347 if (nr_files
|| !nr_dirs
) {
1348 struct ctl_table_header
*header
;
1349 header
= __register_sysctl_table(set
, path
, files
);
1351 kfree(ctl_table_arg
);
1355 /* Remember if we need to free the file table */
1356 header
->ctl_table_arg
= ctl_table_arg
;
1357 **subheader
= header
;
1361 /* Recurse into the subdirectories. */
1362 for (entry
= table
; entry
->procname
; entry
++) {
1368 err
= -ENAMETOOLONG
;
1369 child_pos
= append_path(path
, pos
, entry
->procname
);
1373 err
= register_leaf_sysctl_tables(path
, child_pos
, subheader
,
1381 /* On failure our caller will unregister all registered subheaders */
1386 * __register_sysctl_paths - register a sysctl table hierarchy
1387 * @set: Sysctl tree to register on
1388 * @path: The path to the directory the sysctl table is in.
1389 * @table: the top-level table structure
1391 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1392 * array. A completely 0 filled entry terminates the table.
1394 * See __register_sysctl_table for more details.
1396 struct ctl_table_header
*__register_sysctl_paths(
1397 struct ctl_table_set
*set
,
1398 const struct ctl_path
*path
, struct ctl_table
*table
)
1400 struct ctl_table
*ctl_table_arg
= table
;
1401 int nr_subheaders
= count_subheaders(table
);
1402 struct ctl_table_header
*header
= NULL
, **subheaders
, **subheader
;
1403 const struct ctl_path
*component
;
1404 char *new_path
, *pos
;
1406 pos
= new_path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1411 for (component
= path
; component
->procname
; component
++) {
1412 pos
= append_path(new_path
, pos
, component
->procname
);
1416 while (table
->procname
&& table
->child
&& !table
[1].procname
) {
1417 pos
= append_path(new_path
, pos
, table
->procname
);
1420 table
= table
->child
;
1422 if (nr_subheaders
== 1) {
1423 header
= __register_sysctl_table(set
, new_path
, table
);
1425 header
->ctl_table_arg
= ctl_table_arg
;
1427 header
= kzalloc(sizeof(*header
) +
1428 sizeof(*subheaders
)*nr_subheaders
, GFP_KERNEL
);
1432 subheaders
= (struct ctl_table_header
**) (header
+ 1);
1433 subheader
= subheaders
;
1434 header
->ctl_table_arg
= ctl_table_arg
;
1436 if (register_leaf_sysctl_tables(new_path
, pos
, &subheader
,
1438 goto err_register_leaves
;
1445 err_register_leaves
:
1446 while (subheader
> subheaders
) {
1447 struct ctl_table_header
*subh
= *(--subheader
);
1448 struct ctl_table
*table
= subh
->ctl_table_arg
;
1449 unregister_sysctl_table(subh
);
1458 * register_sysctl_table_path - register a sysctl table hierarchy
1459 * @path: The path to the directory the sysctl table is in.
1460 * @table: the top-level table structure
1462 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1463 * array. A completely 0 filled entry terminates the table.
1465 * See __register_sysctl_paths for more details.
1467 struct ctl_table_header
*register_sysctl_paths(const struct ctl_path
*path
,
1468 struct ctl_table
*table
)
1470 return __register_sysctl_paths(&sysctl_table_root
.default_set
,
1473 EXPORT_SYMBOL(register_sysctl_paths
);
1476 * register_sysctl_table - register a sysctl table hierarchy
1477 * @table: the top-level table structure
1479 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1480 * array. A completely 0 filled entry terminates the table.
1482 * See register_sysctl_paths for more details.
1484 struct ctl_table_header
*register_sysctl_table(struct ctl_table
*table
)
1486 static const struct ctl_path null_path
[] = { {} };
1488 return register_sysctl_paths(null_path
, table
);
1490 EXPORT_SYMBOL(register_sysctl_table
);
1492 static void put_links(struct ctl_table_header
*header
)
1494 struct ctl_table_set
*root_set
= &sysctl_table_root
.default_set
;
1495 struct ctl_table_root
*root
= header
->root
;
1496 struct ctl_dir
*parent
= header
->parent
;
1497 struct ctl_dir
*core_parent
;
1498 struct ctl_table
*entry
;
1500 if (header
->set
== root_set
)
1503 core_parent
= xlate_dir(root_set
, parent
);
1504 if (IS_ERR(core_parent
))
1507 for (entry
= header
->ctl_table
; entry
->procname
; entry
++) {
1508 struct ctl_table_header
*link_head
;
1509 struct ctl_table
*link
;
1510 const char *name
= entry
->procname
;
1512 link
= find_entry(&link_head
, core_parent
, name
, strlen(name
));
1514 ((S_ISDIR(link
->mode
) && S_ISDIR(entry
->mode
)) ||
1515 (S_ISLNK(link
->mode
) && (link
->data
== root
)))) {
1516 drop_sysctl_table(link_head
);
1519 printk(KERN_ERR
"sysctl link missing during unregister: ");
1520 sysctl_print_dir(parent
);
1521 printk(KERN_CONT
"/%s\n", name
);
1526 static void drop_sysctl_table(struct ctl_table_header
*header
)
1528 struct ctl_dir
*parent
= header
->parent
;
1534 start_unregistering(header
);
1535 if (!--header
->count
)
1536 kfree_rcu(header
, rcu
);
1539 drop_sysctl_table(&parent
->header
);
1543 * unregister_sysctl_table - unregister a sysctl table hierarchy
1544 * @header: the header returned from register_sysctl_table
1546 * Unregisters the sysctl table and all children. proc entries may not
1547 * actually be removed until they are no longer used by anyone.
1549 void unregister_sysctl_table(struct ctl_table_header
* header
)
1557 nr_subheaders
= count_subheaders(header
->ctl_table_arg
);
1558 if (unlikely(nr_subheaders
> 1)) {
1559 struct ctl_table_header
**subheaders
;
1562 subheaders
= (struct ctl_table_header
**)(header
+ 1);
1563 for (i
= nr_subheaders
-1; i
>= 0; i
--) {
1564 struct ctl_table_header
*subh
= subheaders
[i
];
1565 struct ctl_table
*table
= subh
->ctl_table_arg
;
1566 unregister_sysctl_table(subh
);
1573 spin_lock(&sysctl_lock
);
1574 drop_sysctl_table(header
);
1575 spin_unlock(&sysctl_lock
);
1577 EXPORT_SYMBOL(unregister_sysctl_table
);
1579 void setup_sysctl_set(struct ctl_table_set
*set
,
1580 struct ctl_table_root
*root
,
1581 int (*is_seen
)(struct ctl_table_set
*))
1583 memset(set
, 0, sizeof(*set
));
1584 set
->is_seen
= is_seen
;
1585 init_header(&set
->dir
.header
, root
, set
, NULL
, root_table
);
1588 void retire_sysctl_set(struct ctl_table_set
*set
)
1590 WARN_ON(!RB_EMPTY_ROOT(&set
->dir
.root
));
1593 int __init
proc_sys_init(void)
1595 struct proc_dir_entry
*proc_sys_root
;
1597 proc_sys_root
= proc_mkdir("sys", NULL
);
1598 proc_sys_root
->proc_iops
= &proc_sys_dir_operations
;
1599 proc_sys_root
->proc_fops
= &proc_sys_dir_file_operations
;
1600 proc_sys_root
->nlink
= 0;
1602 return sysctl_init();