inet: frag: enforce memory limits earlier
[linux/fpc-iii.git] / fs / proc / proc_sysctl.c
blob847f23420b407ea4e7939efd5caf324d987ef5c3
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);
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)
324 struct ctl_table_set *set = &root->default_set;
325 if (root->lookup)
326 set = root->lookup(root);
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 ctl_table_root *root = head->root;
434 struct inode *inode;
435 struct proc_inode *ei;
437 inode = new_inode(sb);
438 if (!inode)
439 goto out;
441 inode->i_ino = get_next_ino();
443 sysctl_head_get(head);
444 ei = PROC_I(inode);
445 ei->sysctl = head;
446 ei->sysctl_entry = table;
448 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
449 inode->i_mode = table->mode;
450 if (!S_ISDIR(table->mode)) {
451 inode->i_mode |= S_IFREG;
452 inode->i_op = &proc_sys_inode_operations;
453 inode->i_fop = &proc_sys_file_operations;
454 } else {
455 inode->i_mode |= S_IFDIR;
456 inode->i_op = &proc_sys_dir_operations;
457 inode->i_fop = &proc_sys_dir_file_operations;
458 if (is_empty_dir(head))
459 make_empty_dir_inode(inode);
462 if (root->set_ownership)
463 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
465 out:
466 return inode;
469 static struct ctl_table_header *grab_header(struct inode *inode)
471 struct ctl_table_header *head = PROC_I(inode)->sysctl;
472 if (!head)
473 head = &sysctl_table_root.default_set.dir.header;
474 return sysctl_head_grab(head);
477 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
478 unsigned int flags)
480 struct ctl_table_header *head = grab_header(dir);
481 struct ctl_table_header *h = NULL;
482 const struct qstr *name = &dentry->d_name;
483 struct ctl_table *p;
484 struct inode *inode;
485 struct dentry *err = ERR_PTR(-ENOENT);
486 struct ctl_dir *ctl_dir;
487 int ret;
489 if (IS_ERR(head))
490 return ERR_CAST(head);
492 ctl_dir = container_of(head, struct ctl_dir, header);
494 p = lookup_entry(&h, ctl_dir, name->name, name->len);
495 if (!p)
496 goto out;
498 if (S_ISLNK(p->mode)) {
499 ret = sysctl_follow_link(&h, &p);
500 err = ERR_PTR(ret);
501 if (ret)
502 goto out;
505 err = ERR_PTR(-ENOMEM);
506 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
507 if (!inode)
508 goto out;
510 err = NULL;
511 d_set_d_op(dentry, &proc_sys_dentry_operations);
512 d_add(dentry, inode);
514 out:
515 if (h)
516 sysctl_head_finish(h);
517 sysctl_head_finish(head);
518 return err;
521 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
522 size_t count, loff_t *ppos, int write)
524 struct inode *inode = file_inode(filp);
525 struct ctl_table_header *head = grab_header(inode);
526 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
527 ssize_t error;
528 size_t res;
530 if (IS_ERR(head))
531 return PTR_ERR(head);
534 * At this point we know that the sysctl was not unregistered
535 * and won't be until we finish.
537 error = -EPERM;
538 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
539 goto out;
541 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
542 error = -EINVAL;
543 if (!table->proc_handler)
544 goto out;
546 /* careful: calling conventions are nasty here */
547 res = count;
548 error = table->proc_handler(table, write, buf, &res, ppos);
549 if (!error)
550 error = res;
551 out:
552 sysctl_head_finish(head);
554 return error;
557 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
558 size_t count, loff_t *ppos)
560 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
563 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
564 size_t count, loff_t *ppos)
566 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
569 static int proc_sys_open(struct inode *inode, struct file *filp)
571 struct ctl_table_header *head = grab_header(inode);
572 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
574 /* sysctl was unregistered */
575 if (IS_ERR(head))
576 return PTR_ERR(head);
578 if (table->poll)
579 filp->private_data = proc_sys_poll_event(table->poll);
581 sysctl_head_finish(head);
583 return 0;
586 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
588 struct inode *inode = file_inode(filp);
589 struct ctl_table_header *head = grab_header(inode);
590 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
591 unsigned int ret = DEFAULT_POLLMASK;
592 unsigned long event;
594 /* sysctl was unregistered */
595 if (IS_ERR(head))
596 return POLLERR | POLLHUP;
598 if (!table->proc_handler)
599 goto out;
601 if (!table->poll)
602 goto out;
604 event = (unsigned long)filp->private_data;
605 poll_wait(filp, &table->poll->wait, wait);
607 if (event != atomic_read(&table->poll->event)) {
608 filp->private_data = proc_sys_poll_event(table->poll);
609 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
612 out:
613 sysctl_head_finish(head);
615 return ret;
618 static bool proc_sys_fill_cache(struct file *file,
619 struct dir_context *ctx,
620 struct ctl_table_header *head,
621 struct ctl_table *table)
623 struct dentry *child, *dir = file->f_path.dentry;
624 struct inode *inode;
625 struct qstr qname;
626 ino_t ino = 0;
627 unsigned type = DT_UNKNOWN;
629 qname.name = table->procname;
630 qname.len = strlen(table->procname);
631 qname.hash = full_name_hash(dir, qname.name, qname.len);
633 child = d_lookup(dir, &qname);
634 if (!child) {
635 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
636 child = d_alloc_parallel(dir, &qname, &wq);
637 if (IS_ERR(child))
638 return false;
639 if (d_in_lookup(child)) {
640 inode = proc_sys_make_inode(dir->d_sb, head, table);
641 if (!inode) {
642 d_lookup_done(child);
643 dput(child);
644 return false;
646 d_set_d_op(child, &proc_sys_dentry_operations);
647 d_add(child, inode);
650 inode = d_inode(child);
651 ino = inode->i_ino;
652 type = inode->i_mode >> 12;
653 dput(child);
654 return dir_emit(ctx, qname.name, qname.len, ino, type);
657 static bool proc_sys_link_fill_cache(struct file *file,
658 struct dir_context *ctx,
659 struct ctl_table_header *head,
660 struct ctl_table *table)
662 bool ret = true;
664 head = sysctl_head_grab(head);
665 if (IS_ERR(head))
666 return false;
668 if (S_ISLNK(table->mode)) {
669 /* It is not an error if we can not follow the link ignore it */
670 int err = sysctl_follow_link(&head, &table);
671 if (err)
672 goto out;
675 ret = proc_sys_fill_cache(file, ctx, head, table);
676 out:
677 sysctl_head_finish(head);
678 return ret;
681 static int scan(struct ctl_table_header *head, struct ctl_table *table,
682 unsigned long *pos, struct file *file,
683 struct dir_context *ctx)
685 bool res;
687 if ((*pos)++ < ctx->pos)
688 return true;
690 if (unlikely(S_ISLNK(table->mode)))
691 res = proc_sys_link_fill_cache(file, ctx, head, table);
692 else
693 res = proc_sys_fill_cache(file, ctx, head, table);
695 if (res)
696 ctx->pos = *pos;
698 return res;
701 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
703 struct ctl_table_header *head = grab_header(file_inode(file));
704 struct ctl_table_header *h = NULL;
705 struct ctl_table *entry;
706 struct ctl_dir *ctl_dir;
707 unsigned long pos;
709 if (IS_ERR(head))
710 return PTR_ERR(head);
712 ctl_dir = container_of(head, struct ctl_dir, header);
714 if (!dir_emit_dots(file, ctx))
715 goto out;
717 pos = 2;
719 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
720 if (!scan(h, entry, &pos, file, ctx)) {
721 sysctl_head_finish(h);
722 break;
725 out:
726 sysctl_head_finish(head);
727 return 0;
730 static int proc_sys_permission(struct inode *inode, int mask)
733 * sysctl entries that are not writeable,
734 * are _NOT_ writeable, capabilities or not.
736 struct ctl_table_header *head;
737 struct ctl_table *table;
738 int error;
740 /* Executable files are not allowed under /proc/sys/ */
741 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
742 return -EACCES;
744 head = grab_header(inode);
745 if (IS_ERR(head))
746 return PTR_ERR(head);
748 table = PROC_I(inode)->sysctl_entry;
749 if (!table) /* global root - r-xr-xr-x */
750 error = mask & MAY_WRITE ? -EACCES : 0;
751 else /* Use the permissions on the sysctl table entry */
752 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
754 sysctl_head_finish(head);
755 return error;
758 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
760 struct inode *inode = d_inode(dentry);
761 int error;
763 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
764 return -EPERM;
766 error = setattr_prepare(dentry, attr);
767 if (error)
768 return error;
770 setattr_copy(inode, attr);
771 mark_inode_dirty(inode);
772 return 0;
775 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
777 struct inode *inode = d_inode(dentry);
778 struct ctl_table_header *head = grab_header(inode);
779 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
781 if (IS_ERR(head))
782 return PTR_ERR(head);
784 generic_fillattr(inode, stat);
785 if (table)
786 stat->mode = (stat->mode & S_IFMT) | table->mode;
788 sysctl_head_finish(head);
789 return 0;
792 static const struct file_operations proc_sys_file_operations = {
793 .open = proc_sys_open,
794 .poll = proc_sys_poll,
795 .read = proc_sys_read,
796 .write = proc_sys_write,
797 .llseek = default_llseek,
800 static const struct file_operations proc_sys_dir_file_operations = {
801 .read = generic_read_dir,
802 .iterate_shared = proc_sys_readdir,
803 .llseek = generic_file_llseek,
806 static const struct inode_operations proc_sys_inode_operations = {
807 .permission = proc_sys_permission,
808 .setattr = proc_sys_setattr,
809 .getattr = proc_sys_getattr,
812 static const struct inode_operations proc_sys_dir_operations = {
813 .lookup = proc_sys_lookup,
814 .permission = proc_sys_permission,
815 .setattr = proc_sys_setattr,
816 .getattr = proc_sys_getattr,
819 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
821 if (flags & LOOKUP_RCU)
822 return -ECHILD;
823 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
826 static int proc_sys_delete(const struct dentry *dentry)
828 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
831 static int sysctl_is_seen(struct ctl_table_header *p)
833 struct ctl_table_set *set = p->set;
834 int res;
835 spin_lock(&sysctl_lock);
836 if (p->unregistering)
837 res = 0;
838 else if (!set->is_seen)
839 res = 1;
840 else
841 res = set->is_seen(set);
842 spin_unlock(&sysctl_lock);
843 return res;
846 static int proc_sys_compare(const struct dentry *dentry,
847 unsigned int len, const char *str, const struct qstr *name)
849 struct ctl_table_header *head;
850 struct inode *inode;
852 /* Although proc doesn't have negative dentries, rcu-walk means
853 * that inode here can be NULL */
854 /* AV: can it, indeed? */
855 inode = d_inode_rcu(dentry);
856 if (!inode)
857 return 1;
858 if (name->len != len)
859 return 1;
860 if (memcmp(name->name, str, len))
861 return 1;
862 head = rcu_dereference(PROC_I(inode)->sysctl);
863 return !head || !sysctl_is_seen(head);
866 static const struct dentry_operations proc_sys_dentry_operations = {
867 .d_revalidate = proc_sys_revalidate,
868 .d_delete = proc_sys_delete,
869 .d_compare = proc_sys_compare,
872 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
873 const char *name, int namelen)
875 struct ctl_table_header *head;
876 struct ctl_table *entry;
878 entry = find_entry(&head, dir, name, namelen);
879 if (!entry)
880 return ERR_PTR(-ENOENT);
881 if (!S_ISDIR(entry->mode))
882 return ERR_PTR(-ENOTDIR);
883 return container_of(head, struct ctl_dir, header);
886 static struct ctl_dir *new_dir(struct ctl_table_set *set,
887 const char *name, int namelen)
889 struct ctl_table *table;
890 struct ctl_dir *new;
891 struct ctl_node *node;
892 char *new_name;
894 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
895 sizeof(struct ctl_table)*2 + namelen + 1,
896 GFP_KERNEL);
897 if (!new)
898 return NULL;
900 node = (struct ctl_node *)(new + 1);
901 table = (struct ctl_table *)(node + 1);
902 new_name = (char *)(table + 2);
903 memcpy(new_name, name, namelen);
904 new_name[namelen] = '\0';
905 table[0].procname = new_name;
906 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
907 init_header(&new->header, set->dir.header.root, set, node, table);
909 return new;
913 * get_subdir - find or create a subdir with the specified name.
914 * @dir: Directory to create the subdirectory in
915 * @name: The name of the subdirectory to find or create
916 * @namelen: The length of name
918 * Takes a directory with an elevated reference count so we know that
919 * if we drop the lock the directory will not go away. Upon success
920 * the reference is moved from @dir to the returned subdirectory.
921 * Upon error an error code is returned and the reference on @dir is
922 * simply dropped.
924 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
925 const char *name, int namelen)
927 struct ctl_table_set *set = dir->header.set;
928 struct ctl_dir *subdir, *new = NULL;
929 int err;
931 spin_lock(&sysctl_lock);
932 subdir = find_subdir(dir, name, namelen);
933 if (!IS_ERR(subdir))
934 goto found;
935 if (PTR_ERR(subdir) != -ENOENT)
936 goto failed;
938 spin_unlock(&sysctl_lock);
939 new = new_dir(set, name, namelen);
940 spin_lock(&sysctl_lock);
941 subdir = ERR_PTR(-ENOMEM);
942 if (!new)
943 goto failed;
945 /* Was the subdir added while we dropped the lock? */
946 subdir = find_subdir(dir, name, namelen);
947 if (!IS_ERR(subdir))
948 goto found;
949 if (PTR_ERR(subdir) != -ENOENT)
950 goto failed;
952 /* Nope. Use the our freshly made directory entry. */
953 err = insert_header(dir, &new->header);
954 subdir = ERR_PTR(err);
955 if (err)
956 goto failed;
957 subdir = new;
958 found:
959 subdir->header.nreg++;
960 failed:
961 if (IS_ERR(subdir)) {
962 pr_err("sysctl could not get directory: ");
963 sysctl_print_dir(dir);
964 pr_cont("/%*.*s %ld\n",
965 namelen, namelen, name, PTR_ERR(subdir));
967 drop_sysctl_table(&dir->header);
968 if (new)
969 drop_sysctl_table(&new->header);
970 spin_unlock(&sysctl_lock);
971 return subdir;
974 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
976 struct ctl_dir *parent;
977 const char *procname;
978 if (!dir->header.parent)
979 return &set->dir;
980 parent = xlate_dir(set, dir->header.parent);
981 if (IS_ERR(parent))
982 return parent;
983 procname = dir->header.ctl_table[0].procname;
984 return find_subdir(parent, procname, strlen(procname));
987 static int sysctl_follow_link(struct ctl_table_header **phead,
988 struct ctl_table **pentry)
990 struct ctl_table_header *head;
991 struct ctl_table_root *root;
992 struct ctl_table_set *set;
993 struct ctl_table *entry;
994 struct ctl_dir *dir;
995 int ret;
997 ret = 0;
998 spin_lock(&sysctl_lock);
999 root = (*pentry)->data;
1000 set = lookup_header_set(root);
1001 dir = xlate_dir(set, (*phead)->parent);
1002 if (IS_ERR(dir))
1003 ret = PTR_ERR(dir);
1004 else {
1005 const char *procname = (*pentry)->procname;
1006 head = NULL;
1007 entry = find_entry(&head, dir, procname, strlen(procname));
1008 ret = -ENOENT;
1009 if (entry && use_table(head)) {
1010 unuse_table(*phead);
1011 *phead = head;
1012 *pentry = entry;
1013 ret = 0;
1017 spin_unlock(&sysctl_lock);
1018 return ret;
1021 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1023 struct va_format vaf;
1024 va_list args;
1026 va_start(args, fmt);
1027 vaf.fmt = fmt;
1028 vaf.va = &args;
1030 pr_err("sysctl table check failed: %s/%s %pV\n",
1031 path, table->procname, &vaf);
1033 va_end(args);
1034 return -EINVAL;
1037 static int sysctl_check_table(const char *path, struct ctl_table *table)
1039 int err = 0;
1040 for (; table->procname; table++) {
1041 if (table->child)
1042 err = sysctl_err(path, table, "Not a file");
1044 if ((table->proc_handler == proc_dostring) ||
1045 (table->proc_handler == proc_dointvec) ||
1046 (table->proc_handler == proc_dointvec_minmax) ||
1047 (table->proc_handler == proc_dointvec_jiffies) ||
1048 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1049 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1050 (table->proc_handler == proc_doulongvec_minmax) ||
1051 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1052 if (!table->data)
1053 err = sysctl_err(path, table, "No data");
1054 if (!table->maxlen)
1055 err = sysctl_err(path, table, "No maxlen");
1057 if (!table->proc_handler)
1058 err = sysctl_err(path, table, "No proc_handler");
1060 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1061 err = sysctl_err(path, table, "bogus .mode 0%o",
1062 table->mode);
1064 return err;
1067 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1068 struct ctl_table_root *link_root)
1070 struct ctl_table *link_table, *entry, *link;
1071 struct ctl_table_header *links;
1072 struct ctl_node *node;
1073 char *link_name;
1074 int nr_entries, name_bytes;
1076 name_bytes = 0;
1077 nr_entries = 0;
1078 for (entry = table; entry->procname; entry++) {
1079 nr_entries++;
1080 name_bytes += strlen(entry->procname) + 1;
1083 links = kzalloc(sizeof(struct ctl_table_header) +
1084 sizeof(struct ctl_node)*nr_entries +
1085 sizeof(struct ctl_table)*(nr_entries + 1) +
1086 name_bytes,
1087 GFP_KERNEL);
1089 if (!links)
1090 return NULL;
1092 node = (struct ctl_node *)(links + 1);
1093 link_table = (struct ctl_table *)(node + nr_entries);
1094 link_name = (char *)&link_table[nr_entries + 1];
1096 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1097 int len = strlen(entry->procname) + 1;
1098 memcpy(link_name, entry->procname, len);
1099 link->procname = link_name;
1100 link->mode = S_IFLNK|S_IRWXUGO;
1101 link->data = link_root;
1102 link_name += len;
1104 init_header(links, dir->header.root, dir->header.set, node, link_table);
1105 links->nreg = nr_entries;
1107 return links;
1110 static bool get_links(struct ctl_dir *dir,
1111 struct ctl_table *table, struct ctl_table_root *link_root)
1113 struct ctl_table_header *head;
1114 struct ctl_table *entry, *link;
1116 /* Are there links available for every entry in table? */
1117 for (entry = table; entry->procname; entry++) {
1118 const char *procname = entry->procname;
1119 link = find_entry(&head, dir, procname, strlen(procname));
1120 if (!link)
1121 return false;
1122 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1123 continue;
1124 if (S_ISLNK(link->mode) && (link->data == link_root))
1125 continue;
1126 return false;
1129 /* The checks passed. Increase the registration count on the links */
1130 for (entry = table; entry->procname; entry++) {
1131 const char *procname = entry->procname;
1132 link = find_entry(&head, dir, procname, strlen(procname));
1133 head->nreg++;
1135 return true;
1138 static int insert_links(struct ctl_table_header *head)
1140 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1141 struct ctl_dir *core_parent = NULL;
1142 struct ctl_table_header *links;
1143 int err;
1145 if (head->set == root_set)
1146 return 0;
1148 core_parent = xlate_dir(root_set, head->parent);
1149 if (IS_ERR(core_parent))
1150 return 0;
1152 if (get_links(core_parent, head->ctl_table, head->root))
1153 return 0;
1155 core_parent->header.nreg++;
1156 spin_unlock(&sysctl_lock);
1158 links = new_links(core_parent, head->ctl_table, head->root);
1160 spin_lock(&sysctl_lock);
1161 err = -ENOMEM;
1162 if (!links)
1163 goto out;
1165 err = 0;
1166 if (get_links(core_parent, head->ctl_table, head->root)) {
1167 kfree(links);
1168 goto out;
1171 err = insert_header(core_parent, links);
1172 if (err)
1173 kfree(links);
1174 out:
1175 drop_sysctl_table(&core_parent->header);
1176 return err;
1180 * __register_sysctl_table - register a leaf sysctl table
1181 * @set: Sysctl tree to register on
1182 * @path: The path to the directory the sysctl table is in.
1183 * @table: the top-level table structure
1185 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1186 * array. A completely 0 filled entry terminates the table.
1188 * The members of the &struct ctl_table structure are used as follows:
1190 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1191 * enter a sysctl file
1193 * data - a pointer to data for use by proc_handler
1195 * maxlen - the maximum size in bytes of the data
1197 * mode - the file permissions for the /proc/sys file
1199 * child - must be %NULL.
1201 * proc_handler - the text handler routine (described below)
1203 * extra1, extra2 - extra pointers usable by the proc handler routines
1205 * Leaf nodes in the sysctl tree will be represented by a single file
1206 * under /proc; non-leaf nodes will be represented by directories.
1208 * There must be a proc_handler routine for any terminal nodes.
1209 * Several default handlers are available to cover common cases -
1211 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1212 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1213 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1215 * It is the handler's job to read the input buffer from user memory
1216 * and process it. The handler should return 0 on success.
1218 * This routine returns %NULL on a failure to register, and a pointer
1219 * to the table header on success.
1221 struct ctl_table_header *__register_sysctl_table(
1222 struct ctl_table_set *set,
1223 const char *path, struct ctl_table *table)
1225 struct ctl_table_root *root = set->dir.header.root;
1226 struct ctl_table_header *header;
1227 const char *name, *nextname;
1228 struct ctl_dir *dir;
1229 struct ctl_table *entry;
1230 struct ctl_node *node;
1231 int nr_entries = 0;
1233 for (entry = table; entry->procname; entry++)
1234 nr_entries++;
1236 header = kzalloc(sizeof(struct ctl_table_header) +
1237 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1238 if (!header)
1239 return NULL;
1241 node = (struct ctl_node *)(header + 1);
1242 init_header(header, root, set, node, table);
1243 if (sysctl_check_table(path, table))
1244 goto fail;
1246 spin_lock(&sysctl_lock);
1247 dir = &set->dir;
1248 /* Reference moved down the diretory tree get_subdir */
1249 dir->header.nreg++;
1250 spin_unlock(&sysctl_lock);
1252 /* Find the directory for the ctl_table */
1253 for (name = path; name; name = nextname) {
1254 int namelen;
1255 nextname = strchr(name, '/');
1256 if (nextname) {
1257 namelen = nextname - name;
1258 nextname++;
1259 } else {
1260 namelen = strlen(name);
1262 if (namelen == 0)
1263 continue;
1265 dir = get_subdir(dir, name, namelen);
1266 if (IS_ERR(dir))
1267 goto fail;
1270 spin_lock(&sysctl_lock);
1271 if (insert_header(dir, header))
1272 goto fail_put_dir_locked;
1274 drop_sysctl_table(&dir->header);
1275 spin_unlock(&sysctl_lock);
1277 return header;
1279 fail_put_dir_locked:
1280 drop_sysctl_table(&dir->header);
1281 spin_unlock(&sysctl_lock);
1282 fail:
1283 kfree(header);
1284 dump_stack();
1285 return NULL;
1289 * register_sysctl - register a sysctl table
1290 * @path: The path to the directory the sysctl table is in.
1291 * @table: the table structure
1293 * Register a sysctl table. @table should be a filled in ctl_table
1294 * array. A completely 0 filled entry terminates the table.
1296 * See __register_sysctl_table for more details.
1298 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1300 return __register_sysctl_table(&sysctl_table_root.default_set,
1301 path, table);
1303 EXPORT_SYMBOL(register_sysctl);
1305 static char *append_path(const char *path, char *pos, const char *name)
1307 int namelen;
1308 namelen = strlen(name);
1309 if (((pos - path) + namelen + 2) >= PATH_MAX)
1310 return NULL;
1311 memcpy(pos, name, namelen);
1312 pos[namelen] = '/';
1313 pos[namelen + 1] = '\0';
1314 pos += namelen + 1;
1315 return pos;
1318 static int count_subheaders(struct ctl_table *table)
1320 int has_files = 0;
1321 int nr_subheaders = 0;
1322 struct ctl_table *entry;
1324 /* special case: no directory and empty directory */
1325 if (!table || !table->procname)
1326 return 1;
1328 for (entry = table; entry->procname; entry++) {
1329 if (entry->child)
1330 nr_subheaders += count_subheaders(entry->child);
1331 else
1332 has_files = 1;
1334 return nr_subheaders + has_files;
1337 static int register_leaf_sysctl_tables(const char *path, char *pos,
1338 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1339 struct ctl_table *table)
1341 struct ctl_table *ctl_table_arg = NULL;
1342 struct ctl_table *entry, *files;
1343 int nr_files = 0;
1344 int nr_dirs = 0;
1345 int err = -ENOMEM;
1347 for (entry = table; entry->procname; entry++) {
1348 if (entry->child)
1349 nr_dirs++;
1350 else
1351 nr_files++;
1354 files = table;
1355 /* If there are mixed files and directories we need a new table */
1356 if (nr_dirs && nr_files) {
1357 struct ctl_table *new;
1358 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1359 GFP_KERNEL);
1360 if (!files)
1361 goto out;
1363 ctl_table_arg = files;
1364 for (new = files, entry = table; entry->procname; entry++) {
1365 if (entry->child)
1366 continue;
1367 *new = *entry;
1368 new++;
1372 /* Register everything except a directory full of subdirectories */
1373 if (nr_files || !nr_dirs) {
1374 struct ctl_table_header *header;
1375 header = __register_sysctl_table(set, path, files);
1376 if (!header) {
1377 kfree(ctl_table_arg);
1378 goto out;
1381 /* Remember if we need to free the file table */
1382 header->ctl_table_arg = ctl_table_arg;
1383 **subheader = header;
1384 (*subheader)++;
1387 /* Recurse into the subdirectories. */
1388 for (entry = table; entry->procname; entry++) {
1389 char *child_pos;
1391 if (!entry->child)
1392 continue;
1394 err = -ENAMETOOLONG;
1395 child_pos = append_path(path, pos, entry->procname);
1396 if (!child_pos)
1397 goto out;
1399 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1400 set, entry->child);
1401 pos[0] = '\0';
1402 if (err)
1403 goto out;
1405 err = 0;
1406 out:
1407 /* On failure our caller will unregister all registered subheaders */
1408 return err;
1412 * __register_sysctl_paths - register a sysctl table hierarchy
1413 * @set: Sysctl tree to register on
1414 * @path: The path to the directory the sysctl table is in.
1415 * @table: the top-level table structure
1417 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1418 * array. A completely 0 filled entry terminates the table.
1420 * See __register_sysctl_table for more details.
1422 struct ctl_table_header *__register_sysctl_paths(
1423 struct ctl_table_set *set,
1424 const struct ctl_path *path, struct ctl_table *table)
1426 struct ctl_table *ctl_table_arg = table;
1427 int nr_subheaders = count_subheaders(table);
1428 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1429 const struct ctl_path *component;
1430 char *new_path, *pos;
1432 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1433 if (!new_path)
1434 return NULL;
1436 pos[0] = '\0';
1437 for (component = path; component->procname; component++) {
1438 pos = append_path(new_path, pos, component->procname);
1439 if (!pos)
1440 goto out;
1442 while (table->procname && table->child && !table[1].procname) {
1443 pos = append_path(new_path, pos, table->procname);
1444 if (!pos)
1445 goto out;
1446 table = table->child;
1448 if (nr_subheaders == 1) {
1449 header = __register_sysctl_table(set, new_path, table);
1450 if (header)
1451 header->ctl_table_arg = ctl_table_arg;
1452 } else {
1453 header = kzalloc(sizeof(*header) +
1454 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1455 if (!header)
1456 goto out;
1458 subheaders = (struct ctl_table_header **) (header + 1);
1459 subheader = subheaders;
1460 header->ctl_table_arg = ctl_table_arg;
1462 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1463 set, table))
1464 goto err_register_leaves;
1467 out:
1468 kfree(new_path);
1469 return header;
1471 err_register_leaves:
1472 while (subheader > subheaders) {
1473 struct ctl_table_header *subh = *(--subheader);
1474 struct ctl_table *table = subh->ctl_table_arg;
1475 unregister_sysctl_table(subh);
1476 kfree(table);
1478 kfree(header);
1479 header = NULL;
1480 goto out;
1484 * register_sysctl_table_path - register a sysctl table hierarchy
1485 * @path: The path to the directory the sysctl table is in.
1486 * @table: the top-level table structure
1488 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1489 * array. A completely 0 filled entry terminates the table.
1491 * See __register_sysctl_paths for more details.
1493 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1494 struct ctl_table *table)
1496 return __register_sysctl_paths(&sysctl_table_root.default_set,
1497 path, table);
1499 EXPORT_SYMBOL(register_sysctl_paths);
1502 * register_sysctl_table - register a sysctl table hierarchy
1503 * @table: the top-level table structure
1505 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1506 * array. A completely 0 filled entry terminates the table.
1508 * See register_sysctl_paths for more details.
1510 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1512 static const struct ctl_path null_path[] = { {} };
1514 return register_sysctl_paths(null_path, table);
1516 EXPORT_SYMBOL(register_sysctl_table);
1518 static void put_links(struct ctl_table_header *header)
1520 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1521 struct ctl_table_root *root = header->root;
1522 struct ctl_dir *parent = header->parent;
1523 struct ctl_dir *core_parent;
1524 struct ctl_table *entry;
1526 if (header->set == root_set)
1527 return;
1529 core_parent = xlate_dir(root_set, parent);
1530 if (IS_ERR(core_parent))
1531 return;
1533 for (entry = header->ctl_table; entry->procname; entry++) {
1534 struct ctl_table_header *link_head;
1535 struct ctl_table *link;
1536 const char *name = entry->procname;
1538 link = find_entry(&link_head, core_parent, name, strlen(name));
1539 if (link &&
1540 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1541 (S_ISLNK(link->mode) && (link->data == root)))) {
1542 drop_sysctl_table(link_head);
1544 else {
1545 pr_err("sysctl link missing during unregister: ");
1546 sysctl_print_dir(parent);
1547 pr_cont("/%s\n", name);
1552 static void drop_sysctl_table(struct ctl_table_header *header)
1554 struct ctl_dir *parent = header->parent;
1556 if (--header->nreg)
1557 return;
1559 put_links(header);
1560 start_unregistering(header);
1561 if (!--header->count)
1562 kfree_rcu(header, rcu);
1564 if (parent)
1565 drop_sysctl_table(&parent->header);
1569 * unregister_sysctl_table - unregister a sysctl table hierarchy
1570 * @header: the header returned from register_sysctl_table
1572 * Unregisters the sysctl table and all children. proc entries may not
1573 * actually be removed until they are no longer used by anyone.
1575 void unregister_sysctl_table(struct ctl_table_header * header)
1577 int nr_subheaders;
1578 might_sleep();
1580 if (header == NULL)
1581 return;
1583 nr_subheaders = count_subheaders(header->ctl_table_arg);
1584 if (unlikely(nr_subheaders > 1)) {
1585 struct ctl_table_header **subheaders;
1586 int i;
1588 subheaders = (struct ctl_table_header **)(header + 1);
1589 for (i = nr_subheaders -1; i >= 0; i--) {
1590 struct ctl_table_header *subh = subheaders[i];
1591 struct ctl_table *table = subh->ctl_table_arg;
1592 unregister_sysctl_table(subh);
1593 kfree(table);
1595 kfree(header);
1596 return;
1599 spin_lock(&sysctl_lock);
1600 drop_sysctl_table(header);
1601 spin_unlock(&sysctl_lock);
1603 EXPORT_SYMBOL(unregister_sysctl_table);
1605 void setup_sysctl_set(struct ctl_table_set *set,
1606 struct ctl_table_root *root,
1607 int (*is_seen)(struct ctl_table_set *))
1609 memset(set, 0, sizeof(*set));
1610 set->is_seen = is_seen;
1611 init_header(&set->dir.header, root, set, NULL, root_table);
1614 void retire_sysctl_set(struct ctl_table_set *set)
1616 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1619 int __init proc_sys_init(void)
1621 struct proc_dir_entry *proc_sys_root;
1623 proc_sys_root = proc_mkdir("sys", NULL);
1624 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1625 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1626 proc_sys_root->nlink = 0;
1628 return sysctl_init();