hfs: get rid of hfs_sync_super
[linux/fpc-iii.git] / fs / proc / proc_sysctl.c
blobdfafeb2b05a0e2ddc5481b2b2a4f445ec8086a0e
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/security.h>
9 #include <linux/sched.h>
10 #include <linux/namei.h>
11 #include <linux/mm.h>
12 #include <linux/module.h>
13 #include "internal.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)
23 if (!poll)
24 return;
26 atomic_inc(&poll->event);
27 wake_up_interruptible(&poll->wait);
30 static struct ctl_table root_table[] = {
32 .procname = "",
33 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
35 { }
37 static struct ctl_table_root sysctl_table_root = {
38 .default_set.dir.header = {
39 {{.count = 1,
40 .nreg = 1,
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)
65 int minlen;
66 int cmp;
68 minlen = len1;
69 if (minlen > len2)
70 minlen = len2;
72 cmp = memcmp(name1, name2, minlen);
73 if (cmp == 0)
74 cmp = len1 - len2;
75 return cmp;
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;
86 while (node)
88 struct ctl_node *ctl_node;
89 const char *procname;
90 int cmp;
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));
98 if (cmp < 0)
99 node = node->rb_left;
100 else if (cmp > 0)
101 node = node->rb_right;
102 else {
103 *phead = head;
104 return entry;
107 return NULL;
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);
118 while (*p) {
119 struct ctl_table_header *parent_head;
120 struct ctl_table *parent_entry;
121 struct ctl_node *parent_node;
122 const char *parent_name;
123 int cmp;
125 parent = *p;
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));
132 if (cmp < 0)
133 p = &(*p)->rb_left;
134 else if (cmp > 0)
135 p = &(*p)->rb_right;
136 else {
137 printk(KERN_ERR "sysctl duplicate entry: ");
138 sysctl_print_dir(head->parent);
139 printk(KERN_CONT "/%s\n", entry->procname);
140 return -EEXIST;
144 rb_link_node(node, parent, p);
145 return 0;
148 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
150 struct rb_node *node = &head->node[entry - head->ctl_table].node;
152 rb_erase(node, &head->parent->root);
155 static void init_header(struct ctl_table_header *head,
156 struct ctl_table_root *root, struct ctl_table_set *set,
157 struct ctl_node *node, struct ctl_table *table)
159 head->ctl_table = table;
160 head->ctl_table_arg = table;
161 head->used = 0;
162 head->count = 1;
163 head->nreg = 1;
164 head->unregistering = NULL;
165 head->root = root;
166 head->set = set;
167 head->parent = NULL;
168 head->node = node;
169 if (node) {
170 struct ctl_table *entry;
171 for (entry = table; entry->procname; entry++, node++) {
172 rb_init_node(&node->node);
173 node->header = head;
178 static void erase_header(struct ctl_table_header *head)
180 struct ctl_table *entry;
181 for (entry = head->ctl_table; entry->procname; entry++)
182 erase_entry(head, entry);
185 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
187 struct ctl_table *entry;
188 int err;
190 dir->header.nreg++;
191 header->parent = dir;
192 err = insert_links(header);
193 if (err)
194 goto fail_links;
195 for (entry = header->ctl_table; entry->procname; entry++) {
196 err = insert_entry(header, entry);
197 if (err)
198 goto fail;
200 return 0;
201 fail:
202 erase_header(header);
203 put_links(header);
204 fail_links:
205 header->parent = NULL;
206 drop_sysctl_table(&dir->header);
207 return err;
210 /* called under sysctl_lock */
211 static int use_table(struct ctl_table_header *p)
213 if (unlikely(p->unregistering))
214 return 0;
215 p->used++;
216 return 1;
219 /* called under sysctl_lock */
220 static void unuse_table(struct ctl_table_header *p)
222 if (!--p->used)
223 if (unlikely(p->unregistering))
224 complete(p->unregistering);
227 /* called under sysctl_lock, will reacquire if has to wait */
228 static void start_unregistering(struct ctl_table_header *p)
231 * if p->used is 0, nobody will ever touch that entry again;
232 * we'll eliminate all paths to it before dropping sysctl_lock
234 if (unlikely(p->used)) {
235 struct completion wait;
236 init_completion(&wait);
237 p->unregistering = &wait;
238 spin_unlock(&sysctl_lock);
239 wait_for_completion(&wait);
240 spin_lock(&sysctl_lock);
241 } else {
242 /* anything non-NULL; we'll never dereference it */
243 p->unregistering = ERR_PTR(-EINVAL);
246 * do not remove from the list until nobody holds it; walking the
247 * list in do_sysctl() relies on that.
249 erase_header(p);
252 static void sysctl_head_get(struct ctl_table_header *head)
254 spin_lock(&sysctl_lock);
255 head->count++;
256 spin_unlock(&sysctl_lock);
259 void sysctl_head_put(struct ctl_table_header *head)
261 spin_lock(&sysctl_lock);
262 if (!--head->count)
263 kfree_rcu(head, rcu);
264 spin_unlock(&sysctl_lock);
267 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
269 if (!head)
270 BUG();
271 spin_lock(&sysctl_lock);
272 if (!use_table(head))
273 head = ERR_PTR(-ENOENT);
274 spin_unlock(&sysctl_lock);
275 return head;
278 static void sysctl_head_finish(struct ctl_table_header *head)
280 if (!head)
281 return;
282 spin_lock(&sysctl_lock);
283 unuse_table(head);
284 spin_unlock(&sysctl_lock);
287 static struct ctl_table_set *
288 lookup_header_set(struct ctl_table_root *root, struct nsproxy *namespaces)
290 struct ctl_table_set *set = &root->default_set;
291 if (root->lookup)
292 set = root->lookup(root, namespaces);
293 return set;
296 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
297 struct ctl_dir *dir,
298 const char *name, int namelen)
300 struct ctl_table_header *head;
301 struct ctl_table *entry;
303 spin_lock(&sysctl_lock);
304 entry = find_entry(&head, dir, name, namelen);
305 if (entry && use_table(head))
306 *phead = head;
307 else
308 entry = NULL;
309 spin_unlock(&sysctl_lock);
310 return entry;
313 static struct ctl_node *first_usable_entry(struct rb_node *node)
315 struct ctl_node *ctl_node;
317 for (;node; node = rb_next(node)) {
318 ctl_node = rb_entry(node, struct ctl_node, node);
319 if (use_table(ctl_node->header))
320 return ctl_node;
322 return NULL;
325 static void first_entry(struct ctl_dir *dir,
326 struct ctl_table_header **phead, struct ctl_table **pentry)
328 struct ctl_table_header *head = NULL;
329 struct ctl_table *entry = NULL;
330 struct ctl_node *ctl_node;
332 spin_lock(&sysctl_lock);
333 ctl_node = first_usable_entry(rb_first(&dir->root));
334 spin_unlock(&sysctl_lock);
335 if (ctl_node) {
336 head = ctl_node->header;
337 entry = &head->ctl_table[ctl_node - head->node];
339 *phead = head;
340 *pentry = entry;
343 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
345 struct ctl_table_header *head = *phead;
346 struct ctl_table *entry = *pentry;
347 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
349 spin_lock(&sysctl_lock);
350 unuse_table(head);
352 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
353 spin_unlock(&sysctl_lock);
354 head = NULL;
355 if (ctl_node) {
356 head = ctl_node->header;
357 entry = &head->ctl_table[ctl_node - head->node];
359 *phead = head;
360 *pentry = entry;
363 void register_sysctl_root(struct ctl_table_root *root)
368 * sysctl_perm does NOT grant the superuser all rights automatically, because
369 * some sysctl variables are readonly even to root.
372 static int test_perm(int mode, int op)
374 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
375 mode >>= 6;
376 else if (in_egroup_p(GLOBAL_ROOT_GID))
377 mode >>= 3;
378 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
379 return 0;
380 return -EACCES;
383 static int sysctl_perm(struct ctl_table_root *root, struct ctl_table *table, int op)
385 int mode;
387 if (root->permissions)
388 mode = root->permissions(root, current->nsproxy, table);
389 else
390 mode = table->mode;
392 return test_perm(mode, op);
395 static struct inode *proc_sys_make_inode(struct super_block *sb,
396 struct ctl_table_header *head, struct ctl_table *table)
398 struct inode *inode;
399 struct proc_inode *ei;
401 inode = new_inode(sb);
402 if (!inode)
403 goto out;
405 inode->i_ino = get_next_ino();
407 sysctl_head_get(head);
408 ei = PROC_I(inode);
409 ei->sysctl = head;
410 ei->sysctl_entry = table;
412 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
413 inode->i_mode = table->mode;
414 if (!S_ISDIR(table->mode)) {
415 inode->i_mode |= S_IFREG;
416 inode->i_op = &proc_sys_inode_operations;
417 inode->i_fop = &proc_sys_file_operations;
418 } else {
419 inode->i_mode |= S_IFDIR;
420 inode->i_op = &proc_sys_dir_operations;
421 inode->i_fop = &proc_sys_dir_file_operations;
423 out:
424 return inode;
427 static struct ctl_table_header *grab_header(struct inode *inode)
429 struct ctl_table_header *head = PROC_I(inode)->sysctl;
430 if (!head)
431 head = &sysctl_table_root.default_set.dir.header;
432 return sysctl_head_grab(head);
435 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
436 unsigned int flags)
438 struct ctl_table_header *head = grab_header(dir);
439 struct ctl_table_header *h = NULL;
440 struct qstr *name = &dentry->d_name;
441 struct ctl_table *p;
442 struct inode *inode;
443 struct dentry *err = ERR_PTR(-ENOENT);
444 struct ctl_dir *ctl_dir;
445 int ret;
447 if (IS_ERR(head))
448 return ERR_CAST(head);
450 ctl_dir = container_of(head, struct ctl_dir, header);
452 p = lookup_entry(&h, ctl_dir, name->name, name->len);
453 if (!p)
454 goto out;
456 if (S_ISLNK(p->mode)) {
457 ret = sysctl_follow_link(&h, &p, current->nsproxy);
458 err = ERR_PTR(ret);
459 if (ret)
460 goto out;
463 err = ERR_PTR(-ENOMEM);
464 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
465 if (h)
466 sysctl_head_finish(h);
468 if (!inode)
469 goto out;
471 err = NULL;
472 d_set_d_op(dentry, &proc_sys_dentry_operations);
473 d_add(dentry, inode);
475 out:
476 sysctl_head_finish(head);
477 return err;
480 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
481 size_t count, loff_t *ppos, int write)
483 struct inode *inode = filp->f_path.dentry->d_inode;
484 struct ctl_table_header *head = grab_header(inode);
485 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
486 ssize_t error;
487 size_t res;
489 if (IS_ERR(head))
490 return PTR_ERR(head);
493 * At this point we know that the sysctl was not unregistered
494 * and won't be until we finish.
496 error = -EPERM;
497 if (sysctl_perm(head->root, table, write ? MAY_WRITE : MAY_READ))
498 goto out;
500 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
501 error = -EINVAL;
502 if (!table->proc_handler)
503 goto out;
505 /* careful: calling conventions are nasty here */
506 res = count;
507 error = table->proc_handler(table, write, buf, &res, ppos);
508 if (!error)
509 error = res;
510 out:
511 sysctl_head_finish(head);
513 return error;
516 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
517 size_t count, loff_t *ppos)
519 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
522 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
523 size_t count, loff_t *ppos)
525 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
528 static int proc_sys_open(struct inode *inode, struct file *filp)
530 struct ctl_table_header *head = grab_header(inode);
531 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
533 /* sysctl was unregistered */
534 if (IS_ERR(head))
535 return PTR_ERR(head);
537 if (table->poll)
538 filp->private_data = proc_sys_poll_event(table->poll);
540 sysctl_head_finish(head);
542 return 0;
545 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
547 struct inode *inode = filp->f_path.dentry->d_inode;
548 struct ctl_table_header *head = grab_header(inode);
549 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
550 unsigned int ret = DEFAULT_POLLMASK;
551 unsigned long event;
553 /* sysctl was unregistered */
554 if (IS_ERR(head))
555 return POLLERR | POLLHUP;
557 if (!table->proc_handler)
558 goto out;
560 if (!table->poll)
561 goto out;
563 event = (unsigned long)filp->private_data;
564 poll_wait(filp, &table->poll->wait, wait);
566 if (event != atomic_read(&table->poll->event)) {
567 filp->private_data = proc_sys_poll_event(table->poll);
568 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
571 out:
572 sysctl_head_finish(head);
574 return ret;
577 static int proc_sys_fill_cache(struct file *filp, void *dirent,
578 filldir_t filldir,
579 struct ctl_table_header *head,
580 struct ctl_table *table)
582 struct dentry *child, *dir = filp->f_path.dentry;
583 struct inode *inode;
584 struct qstr qname;
585 ino_t ino = 0;
586 unsigned type = DT_UNKNOWN;
588 qname.name = table->procname;
589 qname.len = strlen(table->procname);
590 qname.hash = full_name_hash(qname.name, qname.len);
592 child = d_lookup(dir, &qname);
593 if (!child) {
594 child = d_alloc(dir, &qname);
595 if (child) {
596 inode = proc_sys_make_inode(dir->d_sb, head, table);
597 if (!inode) {
598 dput(child);
599 return -ENOMEM;
600 } else {
601 d_set_d_op(child, &proc_sys_dentry_operations);
602 d_add(child, inode);
604 } else {
605 return -ENOMEM;
608 inode = child->d_inode;
609 ino = inode->i_ino;
610 type = inode->i_mode >> 12;
611 dput(child);
612 return !!filldir(dirent, qname.name, qname.len, filp->f_pos, ino, type);
615 static int proc_sys_link_fill_cache(struct file *filp, void *dirent,
616 filldir_t filldir,
617 struct ctl_table_header *head,
618 struct ctl_table *table)
620 int err, ret = 0;
621 head = sysctl_head_grab(head);
623 if (S_ISLNK(table->mode)) {
624 /* It is not an error if we can not follow the link ignore it */
625 err = sysctl_follow_link(&head, &table, current->nsproxy);
626 if (err)
627 goto out;
630 ret = proc_sys_fill_cache(filp, dirent, filldir, head, table);
631 out:
632 sysctl_head_finish(head);
633 return ret;
636 static int scan(struct ctl_table_header *head, ctl_table *table,
637 unsigned long *pos, struct file *file,
638 void *dirent, filldir_t filldir)
640 int res;
642 if ((*pos)++ < file->f_pos)
643 return 0;
645 if (unlikely(S_ISLNK(table->mode)))
646 res = proc_sys_link_fill_cache(file, dirent, filldir, head, table);
647 else
648 res = proc_sys_fill_cache(file, dirent, filldir, head, table);
650 if (res == 0)
651 file->f_pos = *pos;
653 return res;
656 static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir)
658 struct dentry *dentry = filp->f_path.dentry;
659 struct inode *inode = dentry->d_inode;
660 struct ctl_table_header *head = grab_header(inode);
661 struct ctl_table_header *h = NULL;
662 struct ctl_table *entry;
663 struct ctl_dir *ctl_dir;
664 unsigned long pos;
665 int ret = -EINVAL;
667 if (IS_ERR(head))
668 return PTR_ERR(head);
670 ctl_dir = container_of(head, struct ctl_dir, header);
672 ret = 0;
673 /* Avoid a switch here: arm builds fail with missing __cmpdi2 */
674 if (filp->f_pos == 0) {
675 if (filldir(dirent, ".", 1, filp->f_pos,
676 inode->i_ino, DT_DIR) < 0)
677 goto out;
678 filp->f_pos++;
680 if (filp->f_pos == 1) {
681 if (filldir(dirent, "..", 2, filp->f_pos,
682 parent_ino(dentry), DT_DIR) < 0)
683 goto out;
684 filp->f_pos++;
686 pos = 2;
688 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
689 ret = scan(h, entry, &pos, filp, dirent, filldir);
690 if (ret) {
691 sysctl_head_finish(h);
692 break;
695 ret = 1;
696 out:
697 sysctl_head_finish(head);
698 return ret;
701 static int proc_sys_permission(struct inode *inode, int mask)
704 * sysctl entries that are not writeable,
705 * are _NOT_ writeable, capabilities or not.
707 struct ctl_table_header *head;
708 struct ctl_table *table;
709 int error;
711 /* Executable files are not allowed under /proc/sys/ */
712 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
713 return -EACCES;
715 head = grab_header(inode);
716 if (IS_ERR(head))
717 return PTR_ERR(head);
719 table = PROC_I(inode)->sysctl_entry;
720 if (!table) /* global root - r-xr-xr-x */
721 error = mask & MAY_WRITE ? -EACCES : 0;
722 else /* Use the permissions on the sysctl table entry */
723 error = sysctl_perm(head->root, table, mask & ~MAY_NOT_BLOCK);
725 sysctl_head_finish(head);
726 return error;
729 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
731 struct inode *inode = dentry->d_inode;
732 int error;
734 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
735 return -EPERM;
737 error = inode_change_ok(inode, attr);
738 if (error)
739 return error;
741 if ((attr->ia_valid & ATTR_SIZE) &&
742 attr->ia_size != i_size_read(inode)) {
743 error = vmtruncate(inode, attr->ia_size);
744 if (error)
745 return error;
748 setattr_copy(inode, attr);
749 mark_inode_dirty(inode);
750 return 0;
753 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
755 struct inode *inode = dentry->d_inode;
756 struct ctl_table_header *head = grab_header(inode);
757 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
759 if (IS_ERR(head))
760 return PTR_ERR(head);
762 generic_fillattr(inode, stat);
763 if (table)
764 stat->mode = (stat->mode & S_IFMT) | table->mode;
766 sysctl_head_finish(head);
767 return 0;
770 static const struct file_operations proc_sys_file_operations = {
771 .open = proc_sys_open,
772 .poll = proc_sys_poll,
773 .read = proc_sys_read,
774 .write = proc_sys_write,
775 .llseek = default_llseek,
778 static const struct file_operations proc_sys_dir_file_operations = {
779 .read = generic_read_dir,
780 .readdir = proc_sys_readdir,
781 .llseek = generic_file_llseek,
784 static const struct inode_operations proc_sys_inode_operations = {
785 .permission = proc_sys_permission,
786 .setattr = proc_sys_setattr,
787 .getattr = proc_sys_getattr,
790 static const struct inode_operations proc_sys_dir_operations = {
791 .lookup = proc_sys_lookup,
792 .permission = proc_sys_permission,
793 .setattr = proc_sys_setattr,
794 .getattr = proc_sys_getattr,
797 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
799 if (flags & LOOKUP_RCU)
800 return -ECHILD;
801 return !PROC_I(dentry->d_inode)->sysctl->unregistering;
804 static int proc_sys_delete(const struct dentry *dentry)
806 return !!PROC_I(dentry->d_inode)->sysctl->unregistering;
809 static int sysctl_is_seen(struct ctl_table_header *p)
811 struct ctl_table_set *set = p->set;
812 int res;
813 spin_lock(&sysctl_lock);
814 if (p->unregistering)
815 res = 0;
816 else if (!set->is_seen)
817 res = 1;
818 else
819 res = set->is_seen(set);
820 spin_unlock(&sysctl_lock);
821 return res;
824 static int proc_sys_compare(const struct dentry *parent,
825 const struct inode *pinode,
826 const struct dentry *dentry, const struct inode *inode,
827 unsigned int len, const char *str, const struct qstr *name)
829 struct ctl_table_header *head;
830 /* Although proc doesn't have negative dentries, rcu-walk means
831 * that inode here can be NULL */
832 /* AV: can it, indeed? */
833 if (!inode)
834 return 1;
835 if (name->len != len)
836 return 1;
837 if (memcmp(name->name, str, len))
838 return 1;
839 head = rcu_dereference(PROC_I(inode)->sysctl);
840 return !head || !sysctl_is_seen(head);
843 static const struct dentry_operations proc_sys_dentry_operations = {
844 .d_revalidate = proc_sys_revalidate,
845 .d_delete = proc_sys_delete,
846 .d_compare = proc_sys_compare,
849 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
850 const char *name, int namelen)
852 struct ctl_table_header *head;
853 struct ctl_table *entry;
855 entry = find_entry(&head, dir, name, namelen);
856 if (!entry)
857 return ERR_PTR(-ENOENT);
858 if (!S_ISDIR(entry->mode))
859 return ERR_PTR(-ENOTDIR);
860 return container_of(head, struct ctl_dir, header);
863 static struct ctl_dir *new_dir(struct ctl_table_set *set,
864 const char *name, int namelen)
866 struct ctl_table *table;
867 struct ctl_dir *new;
868 struct ctl_node *node;
869 char *new_name;
871 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
872 sizeof(struct ctl_table)*2 + namelen + 1,
873 GFP_KERNEL);
874 if (!new)
875 return NULL;
877 node = (struct ctl_node *)(new + 1);
878 table = (struct ctl_table *)(node + 1);
879 new_name = (char *)(table + 2);
880 memcpy(new_name, name, namelen);
881 new_name[namelen] = '\0';
882 table[0].procname = new_name;
883 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
884 init_header(&new->header, set->dir.header.root, set, node, table);
886 return new;
890 * get_subdir - find or create a subdir with the specified name.
891 * @dir: Directory to create the subdirectory in
892 * @name: The name of the subdirectory to find or create
893 * @namelen: The length of name
895 * Takes a directory with an elevated reference count so we know that
896 * if we drop the lock the directory will not go away. Upon success
897 * the reference is moved from @dir to the returned subdirectory.
898 * Upon error an error code is returned and the reference on @dir is
899 * simply dropped.
901 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
902 const char *name, int namelen)
904 struct ctl_table_set *set = dir->header.set;
905 struct ctl_dir *subdir, *new = NULL;
906 int err;
908 spin_lock(&sysctl_lock);
909 subdir = find_subdir(dir, name, namelen);
910 if (!IS_ERR(subdir))
911 goto found;
912 if (PTR_ERR(subdir) != -ENOENT)
913 goto failed;
915 spin_unlock(&sysctl_lock);
916 new = new_dir(set, name, namelen);
917 spin_lock(&sysctl_lock);
918 subdir = ERR_PTR(-ENOMEM);
919 if (!new)
920 goto failed;
922 /* Was the subdir added while we dropped the 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 /* Nope. Use the our freshly made directory entry. */
930 err = insert_header(dir, &new->header);
931 subdir = ERR_PTR(err);
932 if (err)
933 goto failed;
934 subdir = new;
935 found:
936 subdir->header.nreg++;
937 failed:
938 if (unlikely(IS_ERR(subdir))) {
939 printk(KERN_ERR "sysctl could not get directory: ");
940 sysctl_print_dir(dir);
941 printk(KERN_CONT "/%*.*s %ld\n",
942 namelen, namelen, name, PTR_ERR(subdir));
944 drop_sysctl_table(&dir->header);
945 if (new)
946 drop_sysctl_table(&new->header);
947 spin_unlock(&sysctl_lock);
948 return subdir;
951 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
953 struct ctl_dir *parent;
954 const char *procname;
955 if (!dir->header.parent)
956 return &set->dir;
957 parent = xlate_dir(set, dir->header.parent);
958 if (IS_ERR(parent))
959 return parent;
960 procname = dir->header.ctl_table[0].procname;
961 return find_subdir(parent, procname, strlen(procname));
964 static int sysctl_follow_link(struct ctl_table_header **phead,
965 struct ctl_table **pentry, struct nsproxy *namespaces)
967 struct ctl_table_header *head;
968 struct ctl_table_root *root;
969 struct ctl_table_set *set;
970 struct ctl_table *entry;
971 struct ctl_dir *dir;
972 int ret;
974 ret = 0;
975 spin_lock(&sysctl_lock);
976 root = (*pentry)->data;
977 set = lookup_header_set(root, namespaces);
978 dir = xlate_dir(set, (*phead)->parent);
979 if (IS_ERR(dir))
980 ret = PTR_ERR(dir);
981 else {
982 const char *procname = (*pentry)->procname;
983 head = NULL;
984 entry = find_entry(&head, dir, procname, strlen(procname));
985 ret = -ENOENT;
986 if (entry && use_table(head)) {
987 unuse_table(*phead);
988 *phead = head;
989 *pentry = entry;
990 ret = 0;
994 spin_unlock(&sysctl_lock);
995 return ret;
998 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1000 struct va_format vaf;
1001 va_list args;
1003 va_start(args, fmt);
1004 vaf.fmt = fmt;
1005 vaf.va = &args;
1007 printk(KERN_ERR "sysctl table check failed: %s/%s %pV\n",
1008 path, table->procname, &vaf);
1010 va_end(args);
1011 return -EINVAL;
1014 static int sysctl_check_table(const char *path, struct ctl_table *table)
1016 int err = 0;
1017 for (; table->procname; table++) {
1018 if (table->child)
1019 err = sysctl_err(path, table, "Not a file");
1021 if ((table->proc_handler == proc_dostring) ||
1022 (table->proc_handler == proc_dointvec) ||
1023 (table->proc_handler == proc_dointvec_minmax) ||
1024 (table->proc_handler == proc_dointvec_jiffies) ||
1025 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1026 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1027 (table->proc_handler == proc_doulongvec_minmax) ||
1028 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1029 if (!table->data)
1030 err = sysctl_err(path, table, "No data");
1031 if (!table->maxlen)
1032 err = sysctl_err(path, table, "No maxlen");
1034 if (!table->proc_handler)
1035 err = sysctl_err(path, table, "No proc_handler");
1037 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1038 err = sysctl_err(path, table, "bogus .mode 0%o",
1039 table->mode);
1041 return err;
1044 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1045 struct ctl_table_root *link_root)
1047 struct ctl_table *link_table, *entry, *link;
1048 struct ctl_table_header *links;
1049 struct ctl_node *node;
1050 char *link_name;
1051 int nr_entries, name_bytes;
1053 name_bytes = 0;
1054 nr_entries = 0;
1055 for (entry = table; entry->procname; entry++) {
1056 nr_entries++;
1057 name_bytes += strlen(entry->procname) + 1;
1060 links = kzalloc(sizeof(struct ctl_table_header) +
1061 sizeof(struct ctl_node)*nr_entries +
1062 sizeof(struct ctl_table)*(nr_entries + 1) +
1063 name_bytes,
1064 GFP_KERNEL);
1066 if (!links)
1067 return NULL;
1069 node = (struct ctl_node *)(links + 1);
1070 link_table = (struct ctl_table *)(node + nr_entries);
1071 link_name = (char *)&link_table[nr_entries + 1];
1073 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1074 int len = strlen(entry->procname) + 1;
1075 memcpy(link_name, entry->procname, len);
1076 link->procname = link_name;
1077 link->mode = S_IFLNK|S_IRWXUGO;
1078 link->data = link_root;
1079 link_name += len;
1081 init_header(links, dir->header.root, dir->header.set, node, link_table);
1082 links->nreg = nr_entries;
1084 return links;
1087 static bool get_links(struct ctl_dir *dir,
1088 struct ctl_table *table, struct ctl_table_root *link_root)
1090 struct ctl_table_header *head;
1091 struct ctl_table *entry, *link;
1093 /* Are there links available for every entry in table? */
1094 for (entry = table; entry->procname; entry++) {
1095 const char *procname = entry->procname;
1096 link = find_entry(&head, dir, procname, strlen(procname));
1097 if (!link)
1098 return false;
1099 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1100 continue;
1101 if (S_ISLNK(link->mode) && (link->data == link_root))
1102 continue;
1103 return false;
1106 /* The checks passed. Increase the registration count on the links */
1107 for (entry = table; entry->procname; entry++) {
1108 const char *procname = entry->procname;
1109 link = find_entry(&head, dir, procname, strlen(procname));
1110 head->nreg++;
1112 return true;
1115 static int insert_links(struct ctl_table_header *head)
1117 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1118 struct ctl_dir *core_parent = NULL;
1119 struct ctl_table_header *links;
1120 int err;
1122 if (head->set == root_set)
1123 return 0;
1125 core_parent = xlate_dir(root_set, head->parent);
1126 if (IS_ERR(core_parent))
1127 return 0;
1129 if (get_links(core_parent, head->ctl_table, head->root))
1130 return 0;
1132 core_parent->header.nreg++;
1133 spin_unlock(&sysctl_lock);
1135 links = new_links(core_parent, head->ctl_table, head->root);
1137 spin_lock(&sysctl_lock);
1138 err = -ENOMEM;
1139 if (!links)
1140 goto out;
1142 err = 0;
1143 if (get_links(core_parent, head->ctl_table, head->root)) {
1144 kfree(links);
1145 goto out;
1148 err = insert_header(core_parent, links);
1149 if (err)
1150 kfree(links);
1151 out:
1152 drop_sysctl_table(&core_parent->header);
1153 return err;
1157 * __register_sysctl_table - register a leaf sysctl table
1158 * @set: Sysctl tree to register on
1159 * @path: The path to the directory the sysctl table is in.
1160 * @table: the top-level table structure
1162 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1163 * array. A completely 0 filled entry terminates the table.
1165 * The members of the &struct ctl_table structure are used as follows:
1167 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1168 * enter a sysctl file
1170 * data - a pointer to data for use by proc_handler
1172 * maxlen - the maximum size in bytes of the data
1174 * mode - the file permissions for the /proc/sys file
1176 * child - must be %NULL.
1178 * proc_handler - the text handler routine (described below)
1180 * extra1, extra2 - extra pointers usable by the proc handler routines
1182 * Leaf nodes in the sysctl tree will be represented by a single file
1183 * under /proc; non-leaf nodes will be represented by directories.
1185 * There must be a proc_handler routine for any terminal nodes.
1186 * Several default handlers are available to cover common cases -
1188 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1189 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1190 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1192 * It is the handler's job to read the input buffer from user memory
1193 * and process it. The handler should return 0 on success.
1195 * This routine returns %NULL on a failure to register, and a pointer
1196 * to the table header on success.
1198 struct ctl_table_header *__register_sysctl_table(
1199 struct ctl_table_set *set,
1200 const char *path, struct ctl_table *table)
1202 struct ctl_table_root *root = set->dir.header.root;
1203 struct ctl_table_header *header;
1204 const char *name, *nextname;
1205 struct ctl_dir *dir;
1206 struct ctl_table *entry;
1207 struct ctl_node *node;
1208 int nr_entries = 0;
1210 for (entry = table; entry->procname; entry++)
1211 nr_entries++;
1213 header = kzalloc(sizeof(struct ctl_table_header) +
1214 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1215 if (!header)
1216 return NULL;
1218 node = (struct ctl_node *)(header + 1);
1219 init_header(header, root, set, node, table);
1220 if (sysctl_check_table(path, table))
1221 goto fail;
1223 spin_lock(&sysctl_lock);
1224 dir = &set->dir;
1225 /* Reference moved down the diretory tree get_subdir */
1226 dir->header.nreg++;
1227 spin_unlock(&sysctl_lock);
1229 /* Find the directory for the ctl_table */
1230 for (name = path; name; name = nextname) {
1231 int namelen;
1232 nextname = strchr(name, '/');
1233 if (nextname) {
1234 namelen = nextname - name;
1235 nextname++;
1236 } else {
1237 namelen = strlen(name);
1239 if (namelen == 0)
1240 continue;
1242 dir = get_subdir(dir, name, namelen);
1243 if (IS_ERR(dir))
1244 goto fail;
1247 spin_lock(&sysctl_lock);
1248 if (insert_header(dir, header))
1249 goto fail_put_dir_locked;
1251 drop_sysctl_table(&dir->header);
1252 spin_unlock(&sysctl_lock);
1254 return header;
1256 fail_put_dir_locked:
1257 drop_sysctl_table(&dir->header);
1258 spin_unlock(&sysctl_lock);
1259 fail:
1260 kfree(header);
1261 dump_stack();
1262 return NULL;
1266 * register_sysctl - register a sysctl table
1267 * @path: The path to the directory the sysctl table is in.
1268 * @table: the table structure
1270 * Register a sysctl table. @table should be a filled in ctl_table
1271 * array. A completely 0 filled entry terminates the table.
1273 * See __register_sysctl_table for more details.
1275 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1277 return __register_sysctl_table(&sysctl_table_root.default_set,
1278 path, table);
1280 EXPORT_SYMBOL(register_sysctl);
1282 static char *append_path(const char *path, char *pos, const char *name)
1284 int namelen;
1285 namelen = strlen(name);
1286 if (((pos - path) + namelen + 2) >= PATH_MAX)
1287 return NULL;
1288 memcpy(pos, name, namelen);
1289 pos[namelen] = '/';
1290 pos[namelen + 1] = '\0';
1291 pos += namelen + 1;
1292 return pos;
1295 static int count_subheaders(struct ctl_table *table)
1297 int has_files = 0;
1298 int nr_subheaders = 0;
1299 struct ctl_table *entry;
1301 /* special case: no directory and empty directory */
1302 if (!table || !table->procname)
1303 return 1;
1305 for (entry = table; entry->procname; entry++) {
1306 if (entry->child)
1307 nr_subheaders += count_subheaders(entry->child);
1308 else
1309 has_files = 1;
1311 return nr_subheaders + has_files;
1314 static int register_leaf_sysctl_tables(const char *path, char *pos,
1315 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1316 struct ctl_table *table)
1318 struct ctl_table *ctl_table_arg = NULL;
1319 struct ctl_table *entry, *files;
1320 int nr_files = 0;
1321 int nr_dirs = 0;
1322 int err = -ENOMEM;
1324 for (entry = table; entry->procname; entry++) {
1325 if (entry->child)
1326 nr_dirs++;
1327 else
1328 nr_files++;
1331 files = table;
1332 /* If there are mixed files and directories we need a new table */
1333 if (nr_dirs && nr_files) {
1334 struct ctl_table *new;
1335 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1336 GFP_KERNEL);
1337 if (!files)
1338 goto out;
1340 ctl_table_arg = files;
1341 for (new = files, entry = table; entry->procname; entry++) {
1342 if (entry->child)
1343 continue;
1344 *new = *entry;
1345 new++;
1349 /* Register everything except a directory full of subdirectories */
1350 if (nr_files || !nr_dirs) {
1351 struct ctl_table_header *header;
1352 header = __register_sysctl_table(set, path, files);
1353 if (!header) {
1354 kfree(ctl_table_arg);
1355 goto out;
1358 /* Remember if we need to free the file table */
1359 header->ctl_table_arg = ctl_table_arg;
1360 **subheader = header;
1361 (*subheader)++;
1364 /* Recurse into the subdirectories. */
1365 for (entry = table; entry->procname; entry++) {
1366 char *child_pos;
1368 if (!entry->child)
1369 continue;
1371 err = -ENAMETOOLONG;
1372 child_pos = append_path(path, pos, entry->procname);
1373 if (!child_pos)
1374 goto out;
1376 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1377 set, entry->child);
1378 pos[0] = '\0';
1379 if (err)
1380 goto out;
1382 err = 0;
1383 out:
1384 /* On failure our caller will unregister all registered subheaders */
1385 return err;
1389 * __register_sysctl_paths - register a sysctl table hierarchy
1390 * @set: Sysctl tree to register on
1391 * @path: The path to the directory the sysctl table is in.
1392 * @table: the top-level table structure
1394 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1395 * array. A completely 0 filled entry terminates the table.
1397 * See __register_sysctl_table for more details.
1399 struct ctl_table_header *__register_sysctl_paths(
1400 struct ctl_table_set *set,
1401 const struct ctl_path *path, struct ctl_table *table)
1403 struct ctl_table *ctl_table_arg = table;
1404 int nr_subheaders = count_subheaders(table);
1405 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1406 const struct ctl_path *component;
1407 char *new_path, *pos;
1409 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1410 if (!new_path)
1411 return NULL;
1413 pos[0] = '\0';
1414 for (component = path; component->procname; component++) {
1415 pos = append_path(new_path, pos, component->procname);
1416 if (!pos)
1417 goto out;
1419 while (table->procname && table->child && !table[1].procname) {
1420 pos = append_path(new_path, pos, table->procname);
1421 if (!pos)
1422 goto out;
1423 table = table->child;
1425 if (nr_subheaders == 1) {
1426 header = __register_sysctl_table(set, new_path, table);
1427 if (header)
1428 header->ctl_table_arg = ctl_table_arg;
1429 } else {
1430 header = kzalloc(sizeof(*header) +
1431 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1432 if (!header)
1433 goto out;
1435 subheaders = (struct ctl_table_header **) (header + 1);
1436 subheader = subheaders;
1437 header->ctl_table_arg = ctl_table_arg;
1439 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1440 set, table))
1441 goto err_register_leaves;
1444 out:
1445 kfree(new_path);
1446 return header;
1448 err_register_leaves:
1449 while (subheader > subheaders) {
1450 struct ctl_table_header *subh = *(--subheader);
1451 struct ctl_table *table = subh->ctl_table_arg;
1452 unregister_sysctl_table(subh);
1453 kfree(table);
1455 kfree(header);
1456 header = NULL;
1457 goto out;
1461 * register_sysctl_table_path - register a sysctl table hierarchy
1462 * @path: The path to the directory the sysctl table is in.
1463 * @table: the top-level table structure
1465 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1466 * array. A completely 0 filled entry terminates the table.
1468 * See __register_sysctl_paths for more details.
1470 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1471 struct ctl_table *table)
1473 return __register_sysctl_paths(&sysctl_table_root.default_set,
1474 path, table);
1476 EXPORT_SYMBOL(register_sysctl_paths);
1479 * register_sysctl_table - register a sysctl table hierarchy
1480 * @table: the top-level table structure
1482 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1483 * array. A completely 0 filled entry terminates the table.
1485 * See register_sysctl_paths for more details.
1487 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1489 static const struct ctl_path null_path[] = { {} };
1491 return register_sysctl_paths(null_path, table);
1493 EXPORT_SYMBOL(register_sysctl_table);
1495 static void put_links(struct ctl_table_header *header)
1497 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1498 struct ctl_table_root *root = header->root;
1499 struct ctl_dir *parent = header->parent;
1500 struct ctl_dir *core_parent;
1501 struct ctl_table *entry;
1503 if (header->set == root_set)
1504 return;
1506 core_parent = xlate_dir(root_set, parent);
1507 if (IS_ERR(core_parent))
1508 return;
1510 for (entry = header->ctl_table; entry->procname; entry++) {
1511 struct ctl_table_header *link_head;
1512 struct ctl_table *link;
1513 const char *name = entry->procname;
1515 link = find_entry(&link_head, core_parent, name, strlen(name));
1516 if (link &&
1517 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1518 (S_ISLNK(link->mode) && (link->data == root)))) {
1519 drop_sysctl_table(link_head);
1521 else {
1522 printk(KERN_ERR "sysctl link missing during unregister: ");
1523 sysctl_print_dir(parent);
1524 printk(KERN_CONT "/%s\n", name);
1529 static void drop_sysctl_table(struct ctl_table_header *header)
1531 struct ctl_dir *parent = header->parent;
1533 if (--header->nreg)
1534 return;
1536 put_links(header);
1537 start_unregistering(header);
1538 if (!--header->count)
1539 kfree_rcu(header, rcu);
1541 if (parent)
1542 drop_sysctl_table(&parent->header);
1546 * unregister_sysctl_table - unregister a sysctl table hierarchy
1547 * @header: the header returned from register_sysctl_table
1549 * Unregisters the sysctl table and all children. proc entries may not
1550 * actually be removed until they are no longer used by anyone.
1552 void unregister_sysctl_table(struct ctl_table_header * header)
1554 int nr_subheaders;
1555 might_sleep();
1557 if (header == NULL)
1558 return;
1560 nr_subheaders = count_subheaders(header->ctl_table_arg);
1561 if (unlikely(nr_subheaders > 1)) {
1562 struct ctl_table_header **subheaders;
1563 int i;
1565 subheaders = (struct ctl_table_header **)(header + 1);
1566 for (i = nr_subheaders -1; i >= 0; i--) {
1567 struct ctl_table_header *subh = subheaders[i];
1568 struct ctl_table *table = subh->ctl_table_arg;
1569 unregister_sysctl_table(subh);
1570 kfree(table);
1572 kfree(header);
1573 return;
1576 spin_lock(&sysctl_lock);
1577 drop_sysctl_table(header);
1578 spin_unlock(&sysctl_lock);
1580 EXPORT_SYMBOL(unregister_sysctl_table);
1582 void setup_sysctl_set(struct ctl_table_set *set,
1583 struct ctl_table_root *root,
1584 int (*is_seen)(struct ctl_table_set *))
1586 memset(set, 0, sizeof(*set));
1587 set->is_seen = is_seen;
1588 init_header(&set->dir.header, root, set, NULL, root_table);
1591 void retire_sysctl_set(struct ctl_table_set *set)
1593 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1596 int __init proc_sys_init(void)
1598 struct proc_dir_entry *proc_sys_root;
1600 proc_sys_root = proc_mkdir("sys", NULL);
1601 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1602 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1603 proc_sys_root->nlink = 0;
1605 return sysctl_init();