Linux 4.8.3
[linux/fpc-iii.git] / fs / proc / proc_sysctl.c
blob1b93650dda2fc5f51cb9c0d42c94efcc2f561a92
1 /*
2 * /proc/sys support
3 */
4 #include <linux/init.h>
5 #include <linux/sysctl.h>
6 #include <linux/poll.h>
7 #include <linux/proc_fs.h>
8 #include <linux/printk.h>
9 #include <linux/security.h>
10 #include <linux/sched.h>
11 #include <linux/namei.h>
12 #include <linux/mm.h>
13 #include <linux/module.h>
14 #include "internal.h"
16 static const struct dentry_operations proc_sys_dentry_operations;
17 static const struct file_operations proc_sys_file_operations;
18 static const struct inode_operations proc_sys_inode_operations;
19 static const struct file_operations proc_sys_dir_file_operations;
20 static const struct inode_operations proc_sys_dir_operations;
22 /* Support for permanently empty directories */
24 struct ctl_table sysctl_mount_point[] = {
25 { }
28 static bool is_empty_dir(struct ctl_table_header *head)
30 return head->ctl_table[0].child == sysctl_mount_point;
33 static void set_empty_dir(struct ctl_dir *dir)
35 dir->header.ctl_table[0].child = sysctl_mount_point;
38 static void clear_empty_dir(struct ctl_dir *dir)
41 dir->header.ctl_table[0].child = NULL;
44 void proc_sys_poll_notify(struct ctl_table_poll *poll)
46 if (!poll)
47 return;
49 atomic_inc(&poll->event);
50 wake_up_interruptible(&poll->wait);
53 static struct ctl_table root_table[] = {
55 .procname = "",
56 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
58 { }
60 static struct ctl_table_root sysctl_table_root = {
61 .default_set.dir.header = {
62 {{.count = 1,
63 .nreg = 1,
64 .ctl_table = root_table }},
65 .ctl_table_arg = root_table,
66 .root = &sysctl_table_root,
67 .set = &sysctl_table_root.default_set,
71 static DEFINE_SPINLOCK(sysctl_lock);
73 static void drop_sysctl_table(struct ctl_table_header *header);
74 static int sysctl_follow_link(struct ctl_table_header **phead,
75 struct ctl_table **pentry, struct nsproxy *namespaces);
76 static int insert_links(struct ctl_table_header *head);
77 static void put_links(struct ctl_table_header *header);
79 static void sysctl_print_dir(struct ctl_dir *dir)
81 if (dir->header.parent)
82 sysctl_print_dir(dir->header.parent);
83 pr_cont("%s/", dir->header.ctl_table[0].procname);
86 static int namecmp(const char *name1, int len1, const char *name2, int len2)
88 int minlen;
89 int cmp;
91 minlen = len1;
92 if (minlen > len2)
93 minlen = len2;
95 cmp = memcmp(name1, name2, minlen);
96 if (cmp == 0)
97 cmp = len1 - len2;
98 return cmp;
101 /* Called under sysctl_lock */
102 static struct ctl_table *find_entry(struct ctl_table_header **phead,
103 struct ctl_dir *dir, const char *name, int namelen)
105 struct ctl_table_header *head;
106 struct ctl_table *entry;
107 struct rb_node *node = dir->root.rb_node;
109 while (node)
111 struct ctl_node *ctl_node;
112 const char *procname;
113 int cmp;
115 ctl_node = rb_entry(node, struct ctl_node, node);
116 head = ctl_node->header;
117 entry = &head->ctl_table[ctl_node - head->node];
118 procname = entry->procname;
120 cmp = namecmp(name, namelen, procname, strlen(procname));
121 if (cmp < 0)
122 node = node->rb_left;
123 else if (cmp > 0)
124 node = node->rb_right;
125 else {
126 *phead = head;
127 return entry;
130 return NULL;
133 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
135 struct rb_node *node = &head->node[entry - head->ctl_table].node;
136 struct rb_node **p = &head->parent->root.rb_node;
137 struct rb_node *parent = NULL;
138 const char *name = entry->procname;
139 int namelen = strlen(name);
141 while (*p) {
142 struct ctl_table_header *parent_head;
143 struct ctl_table *parent_entry;
144 struct ctl_node *parent_node;
145 const char *parent_name;
146 int cmp;
148 parent = *p;
149 parent_node = rb_entry(parent, struct ctl_node, node);
150 parent_head = parent_node->header;
151 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
152 parent_name = parent_entry->procname;
154 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
155 if (cmp < 0)
156 p = &(*p)->rb_left;
157 else if (cmp > 0)
158 p = &(*p)->rb_right;
159 else {
160 pr_err("sysctl duplicate entry: ");
161 sysctl_print_dir(head->parent);
162 pr_cont("/%s\n", entry->procname);
163 return -EEXIST;
167 rb_link_node(node, parent, p);
168 rb_insert_color(node, &head->parent->root);
169 return 0;
172 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
174 struct rb_node *node = &head->node[entry - head->ctl_table].node;
176 rb_erase(node, &head->parent->root);
179 static void init_header(struct ctl_table_header *head,
180 struct ctl_table_root *root, struct ctl_table_set *set,
181 struct ctl_node *node, struct ctl_table *table)
183 head->ctl_table = table;
184 head->ctl_table_arg = table;
185 head->used = 0;
186 head->count = 1;
187 head->nreg = 1;
188 head->unregistering = NULL;
189 head->root = root;
190 head->set = set;
191 head->parent = NULL;
192 head->node = node;
193 if (node) {
194 struct ctl_table *entry;
195 for (entry = table; entry->procname; entry++, node++)
196 node->header = head;
200 static void erase_header(struct ctl_table_header *head)
202 struct ctl_table *entry;
203 for (entry = head->ctl_table; entry->procname; entry++)
204 erase_entry(head, entry);
207 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
209 struct ctl_table *entry;
210 int err;
212 /* Is this a permanently empty directory? */
213 if (is_empty_dir(&dir->header))
214 return -EROFS;
216 /* Am I creating a permanently empty directory? */
217 if (header->ctl_table == sysctl_mount_point) {
218 if (!RB_EMPTY_ROOT(&dir->root))
219 return -EINVAL;
220 set_empty_dir(dir);
223 dir->header.nreg++;
224 header->parent = dir;
225 err = insert_links(header);
226 if (err)
227 goto fail_links;
228 for (entry = header->ctl_table; entry->procname; entry++) {
229 err = insert_entry(header, entry);
230 if (err)
231 goto fail;
233 return 0;
234 fail:
235 erase_header(header);
236 put_links(header);
237 fail_links:
238 if (header->ctl_table == sysctl_mount_point)
239 clear_empty_dir(dir);
240 header->parent = NULL;
241 drop_sysctl_table(&dir->header);
242 return err;
245 /* called under sysctl_lock */
246 static int use_table(struct ctl_table_header *p)
248 if (unlikely(p->unregistering))
249 return 0;
250 p->used++;
251 return 1;
254 /* called under sysctl_lock */
255 static void unuse_table(struct ctl_table_header *p)
257 if (!--p->used)
258 if (unlikely(p->unregistering))
259 complete(p->unregistering);
262 /* called under sysctl_lock, will reacquire if has to wait */
263 static void start_unregistering(struct ctl_table_header *p)
266 * if p->used is 0, nobody will ever touch that entry again;
267 * we'll eliminate all paths to it before dropping sysctl_lock
269 if (unlikely(p->used)) {
270 struct completion wait;
271 init_completion(&wait);
272 p->unregistering = &wait;
273 spin_unlock(&sysctl_lock);
274 wait_for_completion(&wait);
275 spin_lock(&sysctl_lock);
276 } else {
277 /* anything non-NULL; we'll never dereference it */
278 p->unregistering = ERR_PTR(-EINVAL);
281 * do not remove from the list until nobody holds it; walking the
282 * list in do_sysctl() relies on that.
284 erase_header(p);
287 static void sysctl_head_get(struct ctl_table_header *head)
289 spin_lock(&sysctl_lock);
290 head->count++;
291 spin_unlock(&sysctl_lock);
294 void sysctl_head_put(struct ctl_table_header *head)
296 spin_lock(&sysctl_lock);
297 if (!--head->count)
298 kfree_rcu(head, rcu);
299 spin_unlock(&sysctl_lock);
302 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
304 BUG_ON(!head);
305 spin_lock(&sysctl_lock);
306 if (!use_table(head))
307 head = ERR_PTR(-ENOENT);
308 spin_unlock(&sysctl_lock);
309 return head;
312 static void sysctl_head_finish(struct ctl_table_header *head)
314 if (!head)
315 return;
316 spin_lock(&sysctl_lock);
317 unuse_table(head);
318 spin_unlock(&sysctl_lock);
321 static struct ctl_table_set *
322 lookup_header_set(struct ctl_table_root *root, struct nsproxy *namespaces)
324 struct ctl_table_set *set = &root->default_set;
325 if (root->lookup)
326 set = root->lookup(root, namespaces);
327 return set;
330 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
331 struct ctl_dir *dir,
332 const char *name, int namelen)
334 struct ctl_table_header *head;
335 struct ctl_table *entry;
337 spin_lock(&sysctl_lock);
338 entry = find_entry(&head, dir, name, namelen);
339 if (entry && use_table(head))
340 *phead = head;
341 else
342 entry = NULL;
343 spin_unlock(&sysctl_lock);
344 return entry;
347 static struct ctl_node *first_usable_entry(struct rb_node *node)
349 struct ctl_node *ctl_node;
351 for (;node; node = rb_next(node)) {
352 ctl_node = rb_entry(node, struct ctl_node, node);
353 if (use_table(ctl_node->header))
354 return ctl_node;
356 return NULL;
359 static void first_entry(struct ctl_dir *dir,
360 struct ctl_table_header **phead, struct ctl_table **pentry)
362 struct ctl_table_header *head = NULL;
363 struct ctl_table *entry = NULL;
364 struct ctl_node *ctl_node;
366 spin_lock(&sysctl_lock);
367 ctl_node = first_usable_entry(rb_first(&dir->root));
368 spin_unlock(&sysctl_lock);
369 if (ctl_node) {
370 head = ctl_node->header;
371 entry = &head->ctl_table[ctl_node - head->node];
373 *phead = head;
374 *pentry = entry;
377 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
379 struct ctl_table_header *head = *phead;
380 struct ctl_table *entry = *pentry;
381 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
383 spin_lock(&sysctl_lock);
384 unuse_table(head);
386 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
387 spin_unlock(&sysctl_lock);
388 head = NULL;
389 if (ctl_node) {
390 head = ctl_node->header;
391 entry = &head->ctl_table[ctl_node - head->node];
393 *phead = head;
394 *pentry = entry;
397 void register_sysctl_root(struct ctl_table_root *root)
402 * sysctl_perm does NOT grant the superuser all rights automatically, because
403 * some sysctl variables are readonly even to root.
406 static int test_perm(int mode, int op)
408 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
409 mode >>= 6;
410 else if (in_egroup_p(GLOBAL_ROOT_GID))
411 mode >>= 3;
412 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
413 return 0;
414 return -EACCES;
417 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
419 struct ctl_table_root *root = head->root;
420 int mode;
422 if (root->permissions)
423 mode = root->permissions(head, table);
424 else
425 mode = table->mode;
427 return test_perm(mode, op);
430 static struct inode *proc_sys_make_inode(struct super_block *sb,
431 struct ctl_table_header *head, struct ctl_table *table)
433 struct inode *inode;
434 struct proc_inode *ei;
436 inode = new_inode(sb);
437 if (!inode)
438 goto out;
440 inode->i_ino = get_next_ino();
442 sysctl_head_get(head);
443 ei = PROC_I(inode);
444 ei->sysctl = head;
445 ei->sysctl_entry = table;
447 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
448 inode->i_mode = table->mode;
449 if (!S_ISDIR(table->mode)) {
450 inode->i_mode |= S_IFREG;
451 inode->i_op = &proc_sys_inode_operations;
452 inode->i_fop = &proc_sys_file_operations;
453 } else {
454 inode->i_mode |= S_IFDIR;
455 inode->i_op = &proc_sys_dir_operations;
456 inode->i_fop = &proc_sys_dir_file_operations;
457 if (is_empty_dir(head))
458 make_empty_dir_inode(inode);
460 out:
461 return inode;
464 static struct ctl_table_header *grab_header(struct inode *inode)
466 struct ctl_table_header *head = PROC_I(inode)->sysctl;
467 if (!head)
468 head = &sysctl_table_root.default_set.dir.header;
469 return sysctl_head_grab(head);
472 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
473 unsigned int flags)
475 struct ctl_table_header *head = grab_header(dir);
476 struct ctl_table_header *h = NULL;
477 const struct qstr *name = &dentry->d_name;
478 struct ctl_table *p;
479 struct inode *inode;
480 struct dentry *err = ERR_PTR(-ENOENT);
481 struct ctl_dir *ctl_dir;
482 int ret;
484 if (IS_ERR(head))
485 return ERR_CAST(head);
487 ctl_dir = container_of(head, struct ctl_dir, header);
489 p = lookup_entry(&h, ctl_dir, name->name, name->len);
490 if (!p)
491 goto out;
493 if (S_ISLNK(p->mode)) {
494 ret = sysctl_follow_link(&h, &p, current->nsproxy);
495 err = ERR_PTR(ret);
496 if (ret)
497 goto out;
500 err = ERR_PTR(-ENOMEM);
501 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
502 if (!inode)
503 goto out;
505 err = NULL;
506 d_set_d_op(dentry, &proc_sys_dentry_operations);
507 d_add(dentry, inode);
509 out:
510 if (h)
511 sysctl_head_finish(h);
512 sysctl_head_finish(head);
513 return err;
516 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
517 size_t count, loff_t *ppos, int write)
519 struct inode *inode = file_inode(filp);
520 struct ctl_table_header *head = grab_header(inode);
521 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
522 ssize_t error;
523 size_t res;
525 if (IS_ERR(head))
526 return PTR_ERR(head);
529 * At this point we know that the sysctl was not unregistered
530 * and won't be until we finish.
532 error = -EPERM;
533 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
534 goto out;
536 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
537 error = -EINVAL;
538 if (!table->proc_handler)
539 goto out;
541 /* careful: calling conventions are nasty here */
542 res = count;
543 error = table->proc_handler(table, write, buf, &res, ppos);
544 if (!error)
545 error = res;
546 out:
547 sysctl_head_finish(head);
549 return error;
552 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
553 size_t count, loff_t *ppos)
555 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
558 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
559 size_t count, loff_t *ppos)
561 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
564 static int proc_sys_open(struct inode *inode, struct file *filp)
566 struct ctl_table_header *head = grab_header(inode);
567 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
569 /* sysctl was unregistered */
570 if (IS_ERR(head))
571 return PTR_ERR(head);
573 if (table->poll)
574 filp->private_data = proc_sys_poll_event(table->poll);
576 sysctl_head_finish(head);
578 return 0;
581 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
583 struct inode *inode = file_inode(filp);
584 struct ctl_table_header *head = grab_header(inode);
585 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
586 unsigned int ret = DEFAULT_POLLMASK;
587 unsigned long event;
589 /* sysctl was unregistered */
590 if (IS_ERR(head))
591 return POLLERR | POLLHUP;
593 if (!table->proc_handler)
594 goto out;
596 if (!table->poll)
597 goto out;
599 event = (unsigned long)filp->private_data;
600 poll_wait(filp, &table->poll->wait, wait);
602 if (event != atomic_read(&table->poll->event)) {
603 filp->private_data = proc_sys_poll_event(table->poll);
604 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
607 out:
608 sysctl_head_finish(head);
610 return ret;
613 static bool proc_sys_fill_cache(struct file *file,
614 struct dir_context *ctx,
615 struct ctl_table_header *head,
616 struct ctl_table *table)
618 struct dentry *child, *dir = file->f_path.dentry;
619 struct inode *inode;
620 struct qstr qname;
621 ino_t ino = 0;
622 unsigned type = DT_UNKNOWN;
624 qname.name = table->procname;
625 qname.len = strlen(table->procname);
626 qname.hash = full_name_hash(dir, qname.name, qname.len);
628 child = d_lookup(dir, &qname);
629 if (!child) {
630 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
631 child = d_alloc_parallel(dir, &qname, &wq);
632 if (IS_ERR(child))
633 return false;
634 if (d_in_lookup(child)) {
635 inode = proc_sys_make_inode(dir->d_sb, head, table);
636 if (!inode) {
637 d_lookup_done(child);
638 dput(child);
639 return false;
641 d_set_d_op(child, &proc_sys_dentry_operations);
642 d_add(child, inode);
645 inode = d_inode(child);
646 ino = inode->i_ino;
647 type = inode->i_mode >> 12;
648 dput(child);
649 return dir_emit(ctx, qname.name, qname.len, ino, type);
652 static bool proc_sys_link_fill_cache(struct file *file,
653 struct dir_context *ctx,
654 struct ctl_table_header *head,
655 struct ctl_table *table)
657 bool ret = true;
658 head = sysctl_head_grab(head);
660 if (S_ISLNK(table->mode)) {
661 /* It is not an error if we can not follow the link ignore it */
662 int err = sysctl_follow_link(&head, &table, current->nsproxy);
663 if (err)
664 goto out;
667 ret = proc_sys_fill_cache(file, ctx, head, table);
668 out:
669 sysctl_head_finish(head);
670 return ret;
673 static int scan(struct ctl_table_header *head, struct ctl_table *table,
674 unsigned long *pos, struct file *file,
675 struct dir_context *ctx)
677 bool res;
679 if ((*pos)++ < ctx->pos)
680 return true;
682 if (unlikely(S_ISLNK(table->mode)))
683 res = proc_sys_link_fill_cache(file, ctx, head, table);
684 else
685 res = proc_sys_fill_cache(file, ctx, head, table);
687 if (res)
688 ctx->pos = *pos;
690 return res;
693 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
695 struct ctl_table_header *head = grab_header(file_inode(file));
696 struct ctl_table_header *h = NULL;
697 struct ctl_table *entry;
698 struct ctl_dir *ctl_dir;
699 unsigned long pos;
701 if (IS_ERR(head))
702 return PTR_ERR(head);
704 ctl_dir = container_of(head, struct ctl_dir, header);
706 if (!dir_emit_dots(file, ctx))
707 return 0;
709 pos = 2;
711 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
712 if (!scan(h, entry, &pos, file, ctx)) {
713 sysctl_head_finish(h);
714 break;
717 sysctl_head_finish(head);
718 return 0;
721 static int proc_sys_permission(struct inode *inode, int mask)
724 * sysctl entries that are not writeable,
725 * are _NOT_ writeable, capabilities or not.
727 struct ctl_table_header *head;
728 struct ctl_table *table;
729 int error;
731 /* Executable files are not allowed under /proc/sys/ */
732 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
733 return -EACCES;
735 head = grab_header(inode);
736 if (IS_ERR(head))
737 return PTR_ERR(head);
739 table = PROC_I(inode)->sysctl_entry;
740 if (!table) /* global root - r-xr-xr-x */
741 error = mask & MAY_WRITE ? -EACCES : 0;
742 else /* Use the permissions on the sysctl table entry */
743 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
745 sysctl_head_finish(head);
746 return error;
749 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
751 struct inode *inode = d_inode(dentry);
752 int error;
754 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
755 return -EPERM;
757 error = inode_change_ok(inode, attr);
758 if (error)
759 return error;
761 setattr_copy(inode, attr);
762 mark_inode_dirty(inode);
763 return 0;
766 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
768 struct inode *inode = d_inode(dentry);
769 struct ctl_table_header *head = grab_header(inode);
770 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
772 if (IS_ERR(head))
773 return PTR_ERR(head);
775 generic_fillattr(inode, stat);
776 if (table)
777 stat->mode = (stat->mode & S_IFMT) | table->mode;
779 sysctl_head_finish(head);
780 return 0;
783 static const struct file_operations proc_sys_file_operations = {
784 .open = proc_sys_open,
785 .poll = proc_sys_poll,
786 .read = proc_sys_read,
787 .write = proc_sys_write,
788 .llseek = default_llseek,
791 static const struct file_operations proc_sys_dir_file_operations = {
792 .read = generic_read_dir,
793 .iterate_shared = proc_sys_readdir,
794 .llseek = generic_file_llseek,
797 static const struct inode_operations proc_sys_inode_operations = {
798 .permission = proc_sys_permission,
799 .setattr = proc_sys_setattr,
800 .getattr = proc_sys_getattr,
803 static const struct inode_operations proc_sys_dir_operations = {
804 .lookup = proc_sys_lookup,
805 .permission = proc_sys_permission,
806 .setattr = proc_sys_setattr,
807 .getattr = proc_sys_getattr,
810 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
812 if (flags & LOOKUP_RCU)
813 return -ECHILD;
814 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
817 static int proc_sys_delete(const struct dentry *dentry)
819 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
822 static int sysctl_is_seen(struct ctl_table_header *p)
824 struct ctl_table_set *set = p->set;
825 int res;
826 spin_lock(&sysctl_lock);
827 if (p->unregistering)
828 res = 0;
829 else if (!set->is_seen)
830 res = 1;
831 else
832 res = set->is_seen(set);
833 spin_unlock(&sysctl_lock);
834 return res;
837 static int proc_sys_compare(const struct dentry *dentry,
838 unsigned int len, const char *str, const struct qstr *name)
840 struct ctl_table_header *head;
841 struct inode *inode;
843 /* Although proc doesn't have negative dentries, rcu-walk means
844 * that inode here can be NULL */
845 /* AV: can it, indeed? */
846 inode = d_inode_rcu(dentry);
847 if (!inode)
848 return 1;
849 if (name->len != len)
850 return 1;
851 if (memcmp(name->name, str, len))
852 return 1;
853 head = rcu_dereference(PROC_I(inode)->sysctl);
854 return !head || !sysctl_is_seen(head);
857 static const struct dentry_operations proc_sys_dentry_operations = {
858 .d_revalidate = proc_sys_revalidate,
859 .d_delete = proc_sys_delete,
860 .d_compare = proc_sys_compare,
863 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
864 const char *name, int namelen)
866 struct ctl_table_header *head;
867 struct ctl_table *entry;
869 entry = find_entry(&head, dir, name, namelen);
870 if (!entry)
871 return ERR_PTR(-ENOENT);
872 if (!S_ISDIR(entry->mode))
873 return ERR_PTR(-ENOTDIR);
874 return container_of(head, struct ctl_dir, header);
877 static struct ctl_dir *new_dir(struct ctl_table_set *set,
878 const char *name, int namelen)
880 struct ctl_table *table;
881 struct ctl_dir *new;
882 struct ctl_node *node;
883 char *new_name;
885 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
886 sizeof(struct ctl_table)*2 + namelen + 1,
887 GFP_KERNEL);
888 if (!new)
889 return NULL;
891 node = (struct ctl_node *)(new + 1);
892 table = (struct ctl_table *)(node + 1);
893 new_name = (char *)(table + 2);
894 memcpy(new_name, name, namelen);
895 new_name[namelen] = '\0';
896 table[0].procname = new_name;
897 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
898 init_header(&new->header, set->dir.header.root, set, node, table);
900 return new;
904 * get_subdir - find or create a subdir with the specified name.
905 * @dir: Directory to create the subdirectory in
906 * @name: The name of the subdirectory to find or create
907 * @namelen: The length of name
909 * Takes a directory with an elevated reference count so we know that
910 * if we drop the lock the directory will not go away. Upon success
911 * the reference is moved from @dir to the returned subdirectory.
912 * Upon error an error code is returned and the reference on @dir is
913 * simply dropped.
915 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
916 const char *name, int namelen)
918 struct ctl_table_set *set = dir->header.set;
919 struct ctl_dir *subdir, *new = NULL;
920 int err;
922 spin_lock(&sysctl_lock);
923 subdir = find_subdir(dir, name, namelen);
924 if (!IS_ERR(subdir))
925 goto found;
926 if (PTR_ERR(subdir) != -ENOENT)
927 goto failed;
929 spin_unlock(&sysctl_lock);
930 new = new_dir(set, name, namelen);
931 spin_lock(&sysctl_lock);
932 subdir = ERR_PTR(-ENOMEM);
933 if (!new)
934 goto failed;
936 /* Was the subdir added while we dropped the lock? */
937 subdir = find_subdir(dir, name, namelen);
938 if (!IS_ERR(subdir))
939 goto found;
940 if (PTR_ERR(subdir) != -ENOENT)
941 goto failed;
943 /* Nope. Use the our freshly made directory entry. */
944 err = insert_header(dir, &new->header);
945 subdir = ERR_PTR(err);
946 if (err)
947 goto failed;
948 subdir = new;
949 found:
950 subdir->header.nreg++;
951 failed:
952 if (IS_ERR(subdir)) {
953 pr_err("sysctl could not get directory: ");
954 sysctl_print_dir(dir);
955 pr_cont("/%*.*s %ld\n",
956 namelen, namelen, name, PTR_ERR(subdir));
958 drop_sysctl_table(&dir->header);
959 if (new)
960 drop_sysctl_table(&new->header);
961 spin_unlock(&sysctl_lock);
962 return subdir;
965 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
967 struct ctl_dir *parent;
968 const char *procname;
969 if (!dir->header.parent)
970 return &set->dir;
971 parent = xlate_dir(set, dir->header.parent);
972 if (IS_ERR(parent))
973 return parent;
974 procname = dir->header.ctl_table[0].procname;
975 return find_subdir(parent, procname, strlen(procname));
978 static int sysctl_follow_link(struct ctl_table_header **phead,
979 struct ctl_table **pentry, struct nsproxy *namespaces)
981 struct ctl_table_header *head;
982 struct ctl_table_root *root;
983 struct ctl_table_set *set;
984 struct ctl_table *entry;
985 struct ctl_dir *dir;
986 int ret;
988 ret = 0;
989 spin_lock(&sysctl_lock);
990 root = (*pentry)->data;
991 set = lookup_header_set(root, namespaces);
992 dir = xlate_dir(set, (*phead)->parent);
993 if (IS_ERR(dir))
994 ret = PTR_ERR(dir);
995 else {
996 const char *procname = (*pentry)->procname;
997 head = NULL;
998 entry = find_entry(&head, dir, procname, strlen(procname));
999 ret = -ENOENT;
1000 if (entry && use_table(head)) {
1001 unuse_table(*phead);
1002 *phead = head;
1003 *pentry = entry;
1004 ret = 0;
1008 spin_unlock(&sysctl_lock);
1009 return ret;
1012 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1014 struct va_format vaf;
1015 va_list args;
1017 va_start(args, fmt);
1018 vaf.fmt = fmt;
1019 vaf.va = &args;
1021 pr_err("sysctl table check failed: %s/%s %pV\n",
1022 path, table->procname, &vaf);
1024 va_end(args);
1025 return -EINVAL;
1028 static int sysctl_check_table(const char *path, struct ctl_table *table)
1030 int err = 0;
1031 for (; table->procname; table++) {
1032 if (table->child)
1033 err = sysctl_err(path, table, "Not a file");
1035 if ((table->proc_handler == proc_dostring) ||
1036 (table->proc_handler == proc_dointvec) ||
1037 (table->proc_handler == proc_dointvec_minmax) ||
1038 (table->proc_handler == proc_dointvec_jiffies) ||
1039 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1040 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1041 (table->proc_handler == proc_doulongvec_minmax) ||
1042 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1043 if (!table->data)
1044 err = sysctl_err(path, table, "No data");
1045 if (!table->maxlen)
1046 err = sysctl_err(path, table, "No maxlen");
1048 if (!table->proc_handler)
1049 err = sysctl_err(path, table, "No proc_handler");
1051 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1052 err = sysctl_err(path, table, "bogus .mode 0%o",
1053 table->mode);
1055 return err;
1058 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1059 struct ctl_table_root *link_root)
1061 struct ctl_table *link_table, *entry, *link;
1062 struct ctl_table_header *links;
1063 struct ctl_node *node;
1064 char *link_name;
1065 int nr_entries, name_bytes;
1067 name_bytes = 0;
1068 nr_entries = 0;
1069 for (entry = table; entry->procname; entry++) {
1070 nr_entries++;
1071 name_bytes += strlen(entry->procname) + 1;
1074 links = kzalloc(sizeof(struct ctl_table_header) +
1075 sizeof(struct ctl_node)*nr_entries +
1076 sizeof(struct ctl_table)*(nr_entries + 1) +
1077 name_bytes,
1078 GFP_KERNEL);
1080 if (!links)
1081 return NULL;
1083 node = (struct ctl_node *)(links + 1);
1084 link_table = (struct ctl_table *)(node + nr_entries);
1085 link_name = (char *)&link_table[nr_entries + 1];
1087 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1088 int len = strlen(entry->procname) + 1;
1089 memcpy(link_name, entry->procname, len);
1090 link->procname = link_name;
1091 link->mode = S_IFLNK|S_IRWXUGO;
1092 link->data = link_root;
1093 link_name += len;
1095 init_header(links, dir->header.root, dir->header.set, node, link_table);
1096 links->nreg = nr_entries;
1098 return links;
1101 static bool get_links(struct ctl_dir *dir,
1102 struct ctl_table *table, struct ctl_table_root *link_root)
1104 struct ctl_table_header *head;
1105 struct ctl_table *entry, *link;
1107 /* Are there links available for every entry in table? */
1108 for (entry = table; entry->procname; entry++) {
1109 const char *procname = entry->procname;
1110 link = find_entry(&head, dir, procname, strlen(procname));
1111 if (!link)
1112 return false;
1113 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1114 continue;
1115 if (S_ISLNK(link->mode) && (link->data == link_root))
1116 continue;
1117 return false;
1120 /* The checks passed. Increase the registration count on the links */
1121 for (entry = table; entry->procname; entry++) {
1122 const char *procname = entry->procname;
1123 link = find_entry(&head, dir, procname, strlen(procname));
1124 head->nreg++;
1126 return true;
1129 static int insert_links(struct ctl_table_header *head)
1131 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1132 struct ctl_dir *core_parent = NULL;
1133 struct ctl_table_header *links;
1134 int err;
1136 if (head->set == root_set)
1137 return 0;
1139 core_parent = xlate_dir(root_set, head->parent);
1140 if (IS_ERR(core_parent))
1141 return 0;
1143 if (get_links(core_parent, head->ctl_table, head->root))
1144 return 0;
1146 core_parent->header.nreg++;
1147 spin_unlock(&sysctl_lock);
1149 links = new_links(core_parent, head->ctl_table, head->root);
1151 spin_lock(&sysctl_lock);
1152 err = -ENOMEM;
1153 if (!links)
1154 goto out;
1156 err = 0;
1157 if (get_links(core_parent, head->ctl_table, head->root)) {
1158 kfree(links);
1159 goto out;
1162 err = insert_header(core_parent, links);
1163 if (err)
1164 kfree(links);
1165 out:
1166 drop_sysctl_table(&core_parent->header);
1167 return err;
1171 * __register_sysctl_table - register a leaf sysctl table
1172 * @set: Sysctl tree to register on
1173 * @path: The path to the directory the sysctl table is in.
1174 * @table: the top-level table structure
1176 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1177 * array. A completely 0 filled entry terminates the table.
1179 * The members of the &struct ctl_table structure are used as follows:
1181 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1182 * enter a sysctl file
1184 * data - a pointer to data for use by proc_handler
1186 * maxlen - the maximum size in bytes of the data
1188 * mode - the file permissions for the /proc/sys file
1190 * child - must be %NULL.
1192 * proc_handler - the text handler routine (described below)
1194 * extra1, extra2 - extra pointers usable by the proc handler routines
1196 * Leaf nodes in the sysctl tree will be represented by a single file
1197 * under /proc; non-leaf nodes will be represented by directories.
1199 * There must be a proc_handler routine for any terminal nodes.
1200 * Several default handlers are available to cover common cases -
1202 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1203 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1204 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1206 * It is the handler's job to read the input buffer from user memory
1207 * and process it. The handler should return 0 on success.
1209 * This routine returns %NULL on a failure to register, and a pointer
1210 * to the table header on success.
1212 struct ctl_table_header *__register_sysctl_table(
1213 struct ctl_table_set *set,
1214 const char *path, struct ctl_table *table)
1216 struct ctl_table_root *root = set->dir.header.root;
1217 struct ctl_table_header *header;
1218 const char *name, *nextname;
1219 struct ctl_dir *dir;
1220 struct ctl_table *entry;
1221 struct ctl_node *node;
1222 int nr_entries = 0;
1224 for (entry = table; entry->procname; entry++)
1225 nr_entries++;
1227 header = kzalloc(sizeof(struct ctl_table_header) +
1228 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1229 if (!header)
1230 return NULL;
1232 node = (struct ctl_node *)(header + 1);
1233 init_header(header, root, set, node, table);
1234 if (sysctl_check_table(path, table))
1235 goto fail;
1237 spin_lock(&sysctl_lock);
1238 dir = &set->dir;
1239 /* Reference moved down the diretory tree get_subdir */
1240 dir->header.nreg++;
1241 spin_unlock(&sysctl_lock);
1243 /* Find the directory for the ctl_table */
1244 for (name = path; name; name = nextname) {
1245 int namelen;
1246 nextname = strchr(name, '/');
1247 if (nextname) {
1248 namelen = nextname - name;
1249 nextname++;
1250 } else {
1251 namelen = strlen(name);
1253 if (namelen == 0)
1254 continue;
1256 dir = get_subdir(dir, name, namelen);
1257 if (IS_ERR(dir))
1258 goto fail;
1261 spin_lock(&sysctl_lock);
1262 if (insert_header(dir, header))
1263 goto fail_put_dir_locked;
1265 drop_sysctl_table(&dir->header);
1266 spin_unlock(&sysctl_lock);
1268 return header;
1270 fail_put_dir_locked:
1271 drop_sysctl_table(&dir->header);
1272 spin_unlock(&sysctl_lock);
1273 fail:
1274 kfree(header);
1275 dump_stack();
1276 return NULL;
1280 * register_sysctl - register a sysctl table
1281 * @path: The path to the directory the sysctl table is in.
1282 * @table: the table structure
1284 * Register a sysctl table. @table should be a filled in ctl_table
1285 * array. A completely 0 filled entry terminates the table.
1287 * See __register_sysctl_table for more details.
1289 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1291 return __register_sysctl_table(&sysctl_table_root.default_set,
1292 path, table);
1294 EXPORT_SYMBOL(register_sysctl);
1296 static char *append_path(const char *path, char *pos, const char *name)
1298 int namelen;
1299 namelen = strlen(name);
1300 if (((pos - path) + namelen + 2) >= PATH_MAX)
1301 return NULL;
1302 memcpy(pos, name, namelen);
1303 pos[namelen] = '/';
1304 pos[namelen + 1] = '\0';
1305 pos += namelen + 1;
1306 return pos;
1309 static int count_subheaders(struct ctl_table *table)
1311 int has_files = 0;
1312 int nr_subheaders = 0;
1313 struct ctl_table *entry;
1315 /* special case: no directory and empty directory */
1316 if (!table || !table->procname)
1317 return 1;
1319 for (entry = table; entry->procname; entry++) {
1320 if (entry->child)
1321 nr_subheaders += count_subheaders(entry->child);
1322 else
1323 has_files = 1;
1325 return nr_subheaders + has_files;
1328 static int register_leaf_sysctl_tables(const char *path, char *pos,
1329 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1330 struct ctl_table *table)
1332 struct ctl_table *ctl_table_arg = NULL;
1333 struct ctl_table *entry, *files;
1334 int nr_files = 0;
1335 int nr_dirs = 0;
1336 int err = -ENOMEM;
1338 for (entry = table; entry->procname; entry++) {
1339 if (entry->child)
1340 nr_dirs++;
1341 else
1342 nr_files++;
1345 files = table;
1346 /* If there are mixed files and directories we need a new table */
1347 if (nr_dirs && nr_files) {
1348 struct ctl_table *new;
1349 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1350 GFP_KERNEL);
1351 if (!files)
1352 goto out;
1354 ctl_table_arg = files;
1355 for (new = files, entry = table; entry->procname; entry++) {
1356 if (entry->child)
1357 continue;
1358 *new = *entry;
1359 new++;
1363 /* Register everything except a directory full of subdirectories */
1364 if (nr_files || !nr_dirs) {
1365 struct ctl_table_header *header;
1366 header = __register_sysctl_table(set, path, files);
1367 if (!header) {
1368 kfree(ctl_table_arg);
1369 goto out;
1372 /* Remember if we need to free the file table */
1373 header->ctl_table_arg = ctl_table_arg;
1374 **subheader = header;
1375 (*subheader)++;
1378 /* Recurse into the subdirectories. */
1379 for (entry = table; entry->procname; entry++) {
1380 char *child_pos;
1382 if (!entry->child)
1383 continue;
1385 err = -ENAMETOOLONG;
1386 child_pos = append_path(path, pos, entry->procname);
1387 if (!child_pos)
1388 goto out;
1390 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1391 set, entry->child);
1392 pos[0] = '\0';
1393 if (err)
1394 goto out;
1396 err = 0;
1397 out:
1398 /* On failure our caller will unregister all registered subheaders */
1399 return err;
1403 * __register_sysctl_paths - register a sysctl table hierarchy
1404 * @set: Sysctl tree to register on
1405 * @path: The path to the directory the sysctl table is in.
1406 * @table: the top-level table structure
1408 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1409 * array. A completely 0 filled entry terminates the table.
1411 * See __register_sysctl_table for more details.
1413 struct ctl_table_header *__register_sysctl_paths(
1414 struct ctl_table_set *set,
1415 const struct ctl_path *path, struct ctl_table *table)
1417 struct ctl_table *ctl_table_arg = table;
1418 int nr_subheaders = count_subheaders(table);
1419 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1420 const struct ctl_path *component;
1421 char *new_path, *pos;
1423 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1424 if (!new_path)
1425 return NULL;
1427 pos[0] = '\0';
1428 for (component = path; component->procname; component++) {
1429 pos = append_path(new_path, pos, component->procname);
1430 if (!pos)
1431 goto out;
1433 while (table->procname && table->child && !table[1].procname) {
1434 pos = append_path(new_path, pos, table->procname);
1435 if (!pos)
1436 goto out;
1437 table = table->child;
1439 if (nr_subheaders == 1) {
1440 header = __register_sysctl_table(set, new_path, table);
1441 if (header)
1442 header->ctl_table_arg = ctl_table_arg;
1443 } else {
1444 header = kzalloc(sizeof(*header) +
1445 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1446 if (!header)
1447 goto out;
1449 subheaders = (struct ctl_table_header **) (header + 1);
1450 subheader = subheaders;
1451 header->ctl_table_arg = ctl_table_arg;
1453 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1454 set, table))
1455 goto err_register_leaves;
1458 out:
1459 kfree(new_path);
1460 return header;
1462 err_register_leaves:
1463 while (subheader > subheaders) {
1464 struct ctl_table_header *subh = *(--subheader);
1465 struct ctl_table *table = subh->ctl_table_arg;
1466 unregister_sysctl_table(subh);
1467 kfree(table);
1469 kfree(header);
1470 header = NULL;
1471 goto out;
1475 * register_sysctl_table_path - register a sysctl table hierarchy
1476 * @path: The path to the directory the sysctl table is in.
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_paths(const struct ctl_path *path,
1485 struct ctl_table *table)
1487 return __register_sysctl_paths(&sysctl_table_root.default_set,
1488 path, table);
1490 EXPORT_SYMBOL(register_sysctl_paths);
1493 * register_sysctl_table - register a sysctl table hierarchy
1494 * @table: the top-level table structure
1496 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1497 * array. A completely 0 filled entry terminates the table.
1499 * See register_sysctl_paths for more details.
1501 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1503 static const struct ctl_path null_path[] = { {} };
1505 return register_sysctl_paths(null_path, table);
1507 EXPORT_SYMBOL(register_sysctl_table);
1509 static void put_links(struct ctl_table_header *header)
1511 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1512 struct ctl_table_root *root = header->root;
1513 struct ctl_dir *parent = header->parent;
1514 struct ctl_dir *core_parent;
1515 struct ctl_table *entry;
1517 if (header->set == root_set)
1518 return;
1520 core_parent = xlate_dir(root_set, parent);
1521 if (IS_ERR(core_parent))
1522 return;
1524 for (entry = header->ctl_table; entry->procname; entry++) {
1525 struct ctl_table_header *link_head;
1526 struct ctl_table *link;
1527 const char *name = entry->procname;
1529 link = find_entry(&link_head, core_parent, name, strlen(name));
1530 if (link &&
1531 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1532 (S_ISLNK(link->mode) && (link->data == root)))) {
1533 drop_sysctl_table(link_head);
1535 else {
1536 pr_err("sysctl link missing during unregister: ");
1537 sysctl_print_dir(parent);
1538 pr_cont("/%s\n", name);
1543 static void drop_sysctl_table(struct ctl_table_header *header)
1545 struct ctl_dir *parent = header->parent;
1547 if (--header->nreg)
1548 return;
1550 put_links(header);
1551 start_unregistering(header);
1552 if (!--header->count)
1553 kfree_rcu(header, rcu);
1555 if (parent)
1556 drop_sysctl_table(&parent->header);
1560 * unregister_sysctl_table - unregister a sysctl table hierarchy
1561 * @header: the header returned from register_sysctl_table
1563 * Unregisters the sysctl table and all children. proc entries may not
1564 * actually be removed until they are no longer used by anyone.
1566 void unregister_sysctl_table(struct ctl_table_header * header)
1568 int nr_subheaders;
1569 might_sleep();
1571 if (header == NULL)
1572 return;
1574 nr_subheaders = count_subheaders(header->ctl_table_arg);
1575 if (unlikely(nr_subheaders > 1)) {
1576 struct ctl_table_header **subheaders;
1577 int i;
1579 subheaders = (struct ctl_table_header **)(header + 1);
1580 for (i = nr_subheaders -1; i >= 0; i--) {
1581 struct ctl_table_header *subh = subheaders[i];
1582 struct ctl_table *table = subh->ctl_table_arg;
1583 unregister_sysctl_table(subh);
1584 kfree(table);
1586 kfree(header);
1587 return;
1590 spin_lock(&sysctl_lock);
1591 drop_sysctl_table(header);
1592 spin_unlock(&sysctl_lock);
1594 EXPORT_SYMBOL(unregister_sysctl_table);
1596 void setup_sysctl_set(struct ctl_table_set *set,
1597 struct ctl_table_root *root,
1598 int (*is_seen)(struct ctl_table_set *))
1600 memset(set, 0, sizeof(*set));
1601 set->is_seen = is_seen;
1602 init_header(&set->dir.header, root, set, NULL, root_table);
1605 void retire_sysctl_set(struct ctl_table_set *set)
1607 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1610 int __init proc_sys_init(void)
1612 struct proc_dir_entry *proc_sys_root;
1614 proc_sys_root = proc_mkdir("sys", NULL);
1615 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1616 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1617 proc_sys_root->nlink = 0;
1619 return sysctl_init();