Linux Audio

Check our new training course

Loading...
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * /proc/sys support
   4 */
   5#include <linux/init.h>
   6#include <linux/sysctl.h>
   7#include <linux/poll.h>
   8#include <linux/proc_fs.h>
   9#include <linux/printk.h>
  10#include <linux/security.h>
  11#include <linux/sched.h>
  12#include <linux/cred.h>
  13#include <linux/namei.h>
  14#include <linux/mm.h>
  15#include <linux/uio.h>
  16#include <linux/module.h>
  17#include <linux/bpf-cgroup.h>
  18#include <linux/mount.h>
  19#include <linux/kmemleak.h>
  20#include <linux/lockdep.h>
  21#include "internal.h"
  22
  23#define list_for_each_table_entry(entry, header)	\
  24	entry = header->ctl_table;			\
  25	for (size_t i = 0 ; i < header->ctl_table_size; ++i, entry++)
  26
  27static const struct dentry_operations proc_sys_dentry_operations;
  28static const struct file_operations proc_sys_file_operations;
  29static const struct inode_operations proc_sys_inode_operations;
  30static const struct file_operations proc_sys_dir_file_operations;
  31static const struct inode_operations proc_sys_dir_operations;
  32
  33/*
  34 * Support for permanently empty directories.
  35 * Must be non-empty to avoid sharing an address with other tables.
  36 */
  37static const struct ctl_table sysctl_mount_point[] = {
  38	{ }
  39};
  40
  41/**
  42 * register_sysctl_mount_point() - registers a sysctl mount point
  43 * @path: path for the mount point
  44 *
  45 * Used to create a permanently empty directory to serve as mount point.
  46 * There are some subtle but important permission checks this allows in the
  47 * case of unprivileged mounts.
  48 */
  49struct ctl_table_header *register_sysctl_mount_point(const char *path)
  50{
  51	return register_sysctl_sz(path, sysctl_mount_point, 0);
  52}
  53EXPORT_SYMBOL(register_sysctl_mount_point);
  54
 
 
  55#define sysctl_is_perm_empty_ctl_header(hptr)		\
  56	(hptr->type == SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
  57#define sysctl_set_perm_empty_ctl_header(hptr)		\
  58	(hptr->type = SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
  59#define sysctl_clear_perm_empty_ctl_header(hptr)	\
  60	(hptr->type = SYSCTL_TABLE_TYPE_DEFAULT)
  61
  62void proc_sys_poll_notify(struct ctl_table_poll *poll)
  63{
  64	if (!poll)
  65		return;
  66
  67	atomic_inc(&poll->event);
  68	wake_up_interruptible(&poll->wait);
  69}
  70
  71static const struct ctl_table root_table[] = {
  72	{
  73		.procname = "",
  74		.mode = S_IFDIR|S_IRUGO|S_IXUGO,
  75	},
  76};
  77static struct ctl_table_root sysctl_table_root = {
  78	.default_set.dir.header = {
  79		{{.count = 1,
  80		  .nreg = 1,
  81		  .ctl_table = root_table }},
  82		.ctl_table_arg = root_table,
  83		.root = &sysctl_table_root,
  84		.set = &sysctl_table_root.default_set,
  85	},
  86};
  87
  88static DEFINE_SPINLOCK(sysctl_lock);
  89
  90static void drop_sysctl_table(struct ctl_table_header *header);
  91static int sysctl_follow_link(struct ctl_table_header **phead,
  92	const struct ctl_table **pentry);
  93static int insert_links(struct ctl_table_header *head);
  94static void put_links(struct ctl_table_header *header);
  95
  96static void sysctl_print_dir(struct ctl_dir *dir)
  97{
  98	if (dir->header.parent)
  99		sysctl_print_dir(dir->header.parent);
 100	pr_cont("%s/", dir->header.ctl_table[0].procname);
 101}
 102
 103static int namecmp(const char *name1, int len1, const char *name2, int len2)
 104{
 105	int cmp;
 106
 107	cmp = memcmp(name1, name2, min(len1, len2));
 108	if (cmp == 0)
 109		cmp = len1 - len2;
 110	return cmp;
 111}
 112
 113static const struct ctl_table *find_entry(struct ctl_table_header **phead,
 
 114	struct ctl_dir *dir, const char *name, int namelen)
 115{
 116	struct ctl_table_header *head;
 117	const struct ctl_table *entry;
 118	struct rb_node *node = dir->root.rb_node;
 119
 120	lockdep_assert_held(&sysctl_lock);
 121
 122	while (node)
 123	{
 124		struct ctl_node *ctl_node;
 125		const char *procname;
 126		int cmp;
 127
 128		ctl_node = rb_entry(node, struct ctl_node, node);
 129		head = ctl_node->header;
 130		entry = &head->ctl_table[ctl_node - head->node];
 131		procname = entry->procname;
 132
 133		cmp = namecmp(name, namelen, procname, strlen(procname));
 134		if (cmp < 0)
 135			node = node->rb_left;
 136		else if (cmp > 0)
 137			node = node->rb_right;
 138		else {
 139			*phead = head;
 140			return entry;
 141		}
 142	}
 143	return NULL;
 144}
 145
 146static int insert_entry(struct ctl_table_header *head, const struct ctl_table *entry)
 147{
 148	struct rb_node *node = &head->node[entry - head->ctl_table].node;
 149	struct rb_node **p = &head->parent->root.rb_node;
 150	struct rb_node *parent = NULL;
 151	const char *name = entry->procname;
 152	int namelen = strlen(name);
 153
 154	while (*p) {
 155		struct ctl_table_header *parent_head;
 156		const struct ctl_table *parent_entry;
 157		struct ctl_node *parent_node;
 158		const char *parent_name;
 159		int cmp;
 160
 161		parent = *p;
 162		parent_node = rb_entry(parent, struct ctl_node, node);
 163		parent_head = parent_node->header;
 164		parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
 165		parent_name = parent_entry->procname;
 166
 167		cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
 168		if (cmp < 0)
 169			p = &(*p)->rb_left;
 170		else if (cmp > 0)
 171			p = &(*p)->rb_right;
 172		else {
 173			pr_err("sysctl duplicate entry: ");
 174			sysctl_print_dir(head->parent);
 175			pr_cont("%s\n", entry->procname);
 176			return -EEXIST;
 177		}
 178	}
 179
 180	rb_link_node(node, parent, p);
 181	rb_insert_color(node, &head->parent->root);
 182	return 0;
 183}
 184
 185static void erase_entry(struct ctl_table_header *head, const struct ctl_table *entry)
 186{
 187	struct rb_node *node = &head->node[entry - head->ctl_table].node;
 188
 189	rb_erase(node, &head->parent->root);
 190}
 191
 192static void init_header(struct ctl_table_header *head,
 193	struct ctl_table_root *root, struct ctl_table_set *set,
 194	struct ctl_node *node, const struct ctl_table *table, size_t table_size)
 195{
 196	head->ctl_table = table;
 197	head->ctl_table_size = table_size;
 198	head->ctl_table_arg = table;
 199	head->used = 0;
 200	head->count = 1;
 201	head->nreg = 1;
 202	head->unregistering = NULL;
 203	head->root = root;
 204	head->set = set;
 205	head->parent = NULL;
 206	head->node = node;
 207	INIT_HLIST_HEAD(&head->inodes);
 208	if (node) {
 209		const struct ctl_table *entry;
 210
 211		list_for_each_table_entry(entry, head) {
 212			node->header = head;
 213			node++;
 214		}
 215	}
 216	if (table == sysctl_mount_point)
 217		sysctl_set_perm_empty_ctl_header(head);
 218}
 219
 220static void erase_header(struct ctl_table_header *head)
 221{
 222	const struct ctl_table *entry;
 223
 224	list_for_each_table_entry(entry, head)
 225		erase_entry(head, entry);
 226}
 227
 228static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
 229{
 230	const struct ctl_table *entry;
 231	struct ctl_table_header *dir_h = &dir->header;
 232	int err;
 233
 234
 235	/* Is this a permanently empty directory? */
 236	if (sysctl_is_perm_empty_ctl_header(dir_h))
 237		return -EROFS;
 238
 239	/* Am I creating a permanently empty directory? */
 240	if (sysctl_is_perm_empty_ctl_header(header)) {
 
 241		if (!RB_EMPTY_ROOT(&dir->root))
 242			return -EINVAL;
 243		sysctl_set_perm_empty_ctl_header(dir_h);
 244	}
 245
 246	dir_h->nreg++;
 247	header->parent = dir;
 248	err = insert_links(header);
 249	if (err)
 250		goto fail_links;
 251	list_for_each_table_entry(entry, header) {
 252		err = insert_entry(header, entry);
 253		if (err)
 254			goto fail;
 255	}
 256	return 0;
 257fail:
 258	erase_header(header);
 259	put_links(header);
 260fail_links:
 261	if (header->ctl_table == sysctl_mount_point)
 262		sysctl_clear_perm_empty_ctl_header(dir_h);
 263	header->parent = NULL;
 264	drop_sysctl_table(dir_h);
 265	return err;
 266}
 267
 
 268static int use_table(struct ctl_table_header *p)
 269{
 270	lockdep_assert_held(&sysctl_lock);
 271
 272	if (unlikely(p->unregistering))
 273		return 0;
 274	p->used++;
 275	return 1;
 276}
 277
 
 278static void unuse_table(struct ctl_table_header *p)
 279{
 280	lockdep_assert_held(&sysctl_lock);
 281
 282	if (!--p->used)
 283		if (unlikely(p->unregistering))
 284			complete(p->unregistering);
 285}
 286
 287static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
 288{
 289	proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
 290}
 291
 
 292static void start_unregistering(struct ctl_table_header *p)
 293{
 294	/* will reacquire if has to wait */
 295	lockdep_assert_held(&sysctl_lock);
 296
 297	/*
 298	 * if p->used is 0, nobody will ever touch that entry again;
 299	 * we'll eliminate all paths to it before dropping sysctl_lock
 300	 */
 301	if (unlikely(p->used)) {
 302		struct completion wait;
 303		init_completion(&wait);
 304		p->unregistering = &wait;
 305		spin_unlock(&sysctl_lock);
 306		wait_for_completion(&wait);
 307	} else {
 308		/* anything non-NULL; we'll never dereference it */
 309		p->unregistering = ERR_PTR(-EINVAL);
 310		spin_unlock(&sysctl_lock);
 311	}
 312	/*
 313	 * Invalidate dentries for unregistered sysctls: namespaced sysctls
 314	 * can have duplicate names and contaminate dcache very badly.
 315	 */
 316	proc_sys_invalidate_dcache(p);
 317	/*
 318	 * do not remove from the list until nobody holds it; walking the
 319	 * list in do_sysctl() relies on that.
 320	 */
 321	spin_lock(&sysctl_lock);
 322	erase_header(p);
 323}
 324
 325static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
 326{
 327	BUG_ON(!head);
 328	spin_lock(&sysctl_lock);
 329	if (!use_table(head))
 330		head = ERR_PTR(-ENOENT);
 331	spin_unlock(&sysctl_lock);
 332	return head;
 333}
 334
 335static void sysctl_head_finish(struct ctl_table_header *head)
 336{
 337	if (!head)
 338		return;
 339	spin_lock(&sysctl_lock);
 340	unuse_table(head);
 341	spin_unlock(&sysctl_lock);
 342}
 343
 344static struct ctl_table_set *
 345lookup_header_set(struct ctl_table_root *root)
 346{
 347	struct ctl_table_set *set = &root->default_set;
 348	if (root->lookup)
 349		set = root->lookup(root);
 350	return set;
 351}
 352
 353static const struct ctl_table *lookup_entry(struct ctl_table_header **phead,
 354					    struct ctl_dir *dir,
 355					    const char *name, int namelen)
 356{
 357	struct ctl_table_header *head;
 358	const struct ctl_table *entry;
 359
 360	spin_lock(&sysctl_lock);
 361	entry = find_entry(&head, dir, name, namelen);
 362	if (entry && use_table(head))
 363		*phead = head;
 364	else
 365		entry = NULL;
 366	spin_unlock(&sysctl_lock);
 367	return entry;
 368}
 369
 370static struct ctl_node *first_usable_entry(struct rb_node *node)
 371{
 372	struct ctl_node *ctl_node;
 373
 374	for (;node; node = rb_next(node)) {
 375		ctl_node = rb_entry(node, struct ctl_node, node);
 376		if (use_table(ctl_node->header))
 377			return ctl_node;
 378	}
 379	return NULL;
 380}
 381
 382static void first_entry(struct ctl_dir *dir,
 383	struct ctl_table_header **phead, const struct ctl_table **pentry)
 384{
 385	struct ctl_table_header *head = NULL;
 386	const struct ctl_table *entry = NULL;
 387	struct ctl_node *ctl_node;
 388
 389	spin_lock(&sysctl_lock);
 390	ctl_node = first_usable_entry(rb_first(&dir->root));
 391	spin_unlock(&sysctl_lock);
 392	if (ctl_node) {
 393		head = ctl_node->header;
 394		entry = &head->ctl_table[ctl_node - head->node];
 395	}
 396	*phead = head;
 397	*pentry = entry;
 398}
 399
 400static void next_entry(struct ctl_table_header **phead, const struct ctl_table **pentry)
 401{
 402	struct ctl_table_header *head = *phead;
 403	const struct ctl_table *entry = *pentry;
 404	struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
 405
 406	spin_lock(&sysctl_lock);
 407	unuse_table(head);
 408
 409	ctl_node = first_usable_entry(rb_next(&ctl_node->node));
 410	spin_unlock(&sysctl_lock);
 411	head = NULL;
 412	if (ctl_node) {
 413		head = ctl_node->header;
 414		entry = &head->ctl_table[ctl_node - head->node];
 415	}
 416	*phead = head;
 417	*pentry = entry;
 418}
 419
 420/*
 421 * sysctl_perm does NOT grant the superuser all rights automatically, because
 422 * some sysctl variables are readonly even to root.
 423 */
 424
 425static int test_perm(int mode, int op)
 426{
 427	if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
 428		mode >>= 6;
 429	else if (in_egroup_p(GLOBAL_ROOT_GID))
 430		mode >>= 3;
 431	if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
 432		return 0;
 433	return -EACCES;
 434}
 435
 436static int sysctl_perm(struct ctl_table_header *head, const struct ctl_table *table, int op)
 437{
 438	struct ctl_table_root *root = head->root;
 439	int mode;
 440
 441	if (root->permissions)
 442		mode = root->permissions(head, table);
 443	else
 444		mode = table->mode;
 445
 446	return test_perm(mode, op);
 447}
 448
 449static struct inode *proc_sys_make_inode(struct super_block *sb,
 450		struct ctl_table_header *head, const struct ctl_table *table)
 451{
 452	struct ctl_table_root *root = head->root;
 453	struct inode *inode;
 454	struct proc_inode *ei;
 455
 456	inode = new_inode(sb);
 457	if (!inode)
 458		return ERR_PTR(-ENOMEM);
 459
 460	inode->i_ino = get_next_ino();
 461
 462	ei = PROC_I(inode);
 463
 464	spin_lock(&sysctl_lock);
 465	if (unlikely(head->unregistering)) {
 466		spin_unlock(&sysctl_lock);
 467		iput(inode);
 468		return ERR_PTR(-ENOENT);
 469	}
 470	ei->sysctl = head;
 471	ei->sysctl_entry = table;
 472	hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
 473	head->count++;
 474	spin_unlock(&sysctl_lock);
 475
 476	simple_inode_init_ts(inode);
 477	inode->i_mode = table->mode;
 478	if (!S_ISDIR(table->mode)) {
 479		inode->i_mode |= S_IFREG;
 480		inode->i_op = &proc_sys_inode_operations;
 481		inode->i_fop = &proc_sys_file_operations;
 482	} else {
 483		inode->i_mode |= S_IFDIR;
 484		inode->i_op = &proc_sys_dir_operations;
 485		inode->i_fop = &proc_sys_dir_file_operations;
 486		if (sysctl_is_perm_empty_ctl_header(head))
 487			make_empty_dir_inode(inode);
 488	}
 489
 490	inode->i_uid = GLOBAL_ROOT_UID;
 491	inode->i_gid = GLOBAL_ROOT_GID;
 492	if (root->set_ownership)
 493		root->set_ownership(head, &inode->i_uid, &inode->i_gid);
 
 
 
 
 494
 495	return inode;
 496}
 497
 498void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
 499{
 500	spin_lock(&sysctl_lock);
 501	hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
 502	if (!--head->count)
 503		kfree_rcu(head, rcu);
 504	spin_unlock(&sysctl_lock);
 505}
 506
 507static struct ctl_table_header *grab_header(struct inode *inode)
 508{
 509	struct ctl_table_header *head = PROC_I(inode)->sysctl;
 510	if (!head)
 511		head = &sysctl_table_root.default_set.dir.header;
 512	return sysctl_head_grab(head);
 513}
 514
 515static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
 516					unsigned int flags)
 517{
 518	struct ctl_table_header *head = grab_header(dir);
 519	struct ctl_table_header *h = NULL;
 520	const struct qstr *name = &dentry->d_name;
 521	const struct ctl_table *p;
 522	struct inode *inode;
 523	struct dentry *err = ERR_PTR(-ENOENT);
 524	struct ctl_dir *ctl_dir;
 525	int ret;
 526
 527	if (IS_ERR(head))
 528		return ERR_CAST(head);
 529
 530	ctl_dir = container_of(head, struct ctl_dir, header);
 531
 532	p = lookup_entry(&h, ctl_dir, name->name, name->len);
 533	if (!p)
 534		goto out;
 535
 536	if (S_ISLNK(p->mode)) {
 537		ret = sysctl_follow_link(&h, &p);
 538		err = ERR_PTR(ret);
 539		if (ret)
 540			goto out;
 541	}
 542
 543	d_set_d_op(dentry, &proc_sys_dentry_operations);
 544	inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
 545	err = d_splice_alias(inode, dentry);
 546
 547out:
 548	if (h)
 549		sysctl_head_finish(h);
 550	sysctl_head_finish(head);
 551	return err;
 552}
 553
 554static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
 555		int write)
 556{
 557	struct inode *inode = file_inode(iocb->ki_filp);
 558	struct ctl_table_header *head = grab_header(inode);
 559	const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 560	size_t count = iov_iter_count(iter);
 561	char *kbuf;
 562	ssize_t error;
 563
 564	if (IS_ERR(head))
 565		return PTR_ERR(head);
 566
 567	/*
 568	 * At this point we know that the sysctl was not unregistered
 569	 * and won't be until we finish.
 570	 */
 571	error = -EPERM;
 572	if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
 573		goto out;
 574
 575	/* if that can happen at all, it should be -EINVAL, not -EISDIR */
 576	error = -EINVAL;
 577	if (!table->proc_handler)
 578		goto out;
 579
 580	/* don't even try if the size is too large */
 581	error = -ENOMEM;
 582	if (count >= KMALLOC_MAX_SIZE)
 583		goto out;
 584	kbuf = kvzalloc(count + 1, GFP_KERNEL);
 585	if (!kbuf)
 586		goto out;
 587
 588	if (write) {
 589		error = -EFAULT;
 590		if (!copy_from_iter_full(kbuf, count, iter))
 591			goto out_free_buf;
 592		kbuf[count] = '\0';
 593	}
 594
 595	error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
 596					   &iocb->ki_pos);
 597	if (error)
 598		goto out_free_buf;
 599
 600	/* careful: calling conventions are nasty here */
 601	error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
 602	if (error)
 603		goto out_free_buf;
 604
 605	if (!write) {
 606		error = -EFAULT;
 607		if (copy_to_iter(kbuf, count, iter) < count)
 608			goto out_free_buf;
 609	}
 610
 611	error = count;
 612out_free_buf:
 613	kvfree(kbuf);
 614out:
 615	sysctl_head_finish(head);
 616
 617	return error;
 618}
 619
 620static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
 621{
 622	return proc_sys_call_handler(iocb, iter, 0);
 623}
 624
 625static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
 626{
 627	return proc_sys_call_handler(iocb, iter, 1);
 628}
 629
 630static int proc_sys_open(struct inode *inode, struct file *filp)
 631{
 632	struct ctl_table_header *head = grab_header(inode);
 633	const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 634
 635	/* sysctl was unregistered */
 636	if (IS_ERR(head))
 637		return PTR_ERR(head);
 638
 639	if (table->poll)
 640		filp->private_data = proc_sys_poll_event(table->poll);
 641
 642	sysctl_head_finish(head);
 643
 644	return 0;
 645}
 646
 647static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
 648{
 649	struct inode *inode = file_inode(filp);
 650	struct ctl_table_header *head = grab_header(inode);
 651	const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 652	__poll_t ret = DEFAULT_POLLMASK;
 653	unsigned long event;
 654
 655	/* sysctl was unregistered */
 656	if (IS_ERR(head))
 657		return EPOLLERR | EPOLLHUP;
 658
 659	if (!table->proc_handler)
 660		goto out;
 661
 662	if (!table->poll)
 663		goto out;
 664
 665	event = (unsigned long)filp->private_data;
 666	poll_wait(filp, &table->poll->wait, wait);
 667
 668	if (event != atomic_read(&table->poll->event)) {
 669		filp->private_data = proc_sys_poll_event(table->poll);
 670		ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
 671	}
 672
 673out:
 674	sysctl_head_finish(head);
 675
 676	return ret;
 677}
 678
 679static bool proc_sys_fill_cache(struct file *file,
 680				struct dir_context *ctx,
 681				struct ctl_table_header *head,
 682				const struct ctl_table *table)
 683{
 684	struct dentry *child, *dir = file->f_path.dentry;
 685	struct inode *inode;
 686	struct qstr qname;
 687	ino_t ino = 0;
 688	unsigned type = DT_UNKNOWN;
 689
 690	qname.name = table->procname;
 691	qname.len  = strlen(table->procname);
 692	qname.hash = full_name_hash(dir, qname.name, qname.len);
 693
 694	child = d_lookup(dir, &qname);
 695	if (!child) {
 696		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
 697		child = d_alloc_parallel(dir, &qname, &wq);
 698		if (IS_ERR(child))
 699			return false;
 700		if (d_in_lookup(child)) {
 701			struct dentry *res;
 702			d_set_d_op(child, &proc_sys_dentry_operations);
 703			inode = proc_sys_make_inode(dir->d_sb, head, table);
 704			res = d_splice_alias(inode, child);
 705			d_lookup_done(child);
 706			if (unlikely(res)) {
 707				dput(child);
 708
 709				if (IS_ERR(res))
 710					return false;
 711
 
 712				child = res;
 713			}
 714		}
 715	}
 716	inode = d_inode(child);
 717	ino  = inode->i_ino;
 718	type = inode->i_mode >> 12;
 719	dput(child);
 720	return dir_emit(ctx, qname.name, qname.len, ino, type);
 721}
 722
 723static bool proc_sys_link_fill_cache(struct file *file,
 724				    struct dir_context *ctx,
 725				    struct ctl_table_header *head,
 726				    const struct ctl_table *table)
 727{
 728	bool ret = true;
 729
 730	head = sysctl_head_grab(head);
 731	if (IS_ERR(head))
 732		return false;
 733
 734	/* It is not an error if we can not follow the link ignore it */
 735	if (sysctl_follow_link(&head, &table))
 736		goto out;
 737
 738	ret = proc_sys_fill_cache(file, ctx, head, table);
 739out:
 740	sysctl_head_finish(head);
 741	return ret;
 742}
 743
 744static int scan(struct ctl_table_header *head, const struct ctl_table *table,
 745		unsigned long *pos, struct file *file,
 746		struct dir_context *ctx)
 747{
 748	bool res;
 749
 750	if ((*pos)++ < ctx->pos)
 751		return true;
 752
 753	if (unlikely(S_ISLNK(table->mode)))
 754		res = proc_sys_link_fill_cache(file, ctx, head, table);
 755	else
 756		res = proc_sys_fill_cache(file, ctx, head, table);
 757
 758	if (res)
 759		ctx->pos = *pos;
 760
 761	return res;
 762}
 763
 764static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
 765{
 766	struct ctl_table_header *head = grab_header(file_inode(file));
 767	struct ctl_table_header *h = NULL;
 768	const struct ctl_table *entry;
 769	struct ctl_dir *ctl_dir;
 770	unsigned long pos;
 771
 772	if (IS_ERR(head))
 773		return PTR_ERR(head);
 774
 775	ctl_dir = container_of(head, struct ctl_dir, header);
 776
 777	if (!dir_emit_dots(file, ctx))
 778		goto out;
 779
 780	pos = 2;
 781
 782	for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
 783		if (!scan(h, entry, &pos, file, ctx)) {
 784			sysctl_head_finish(h);
 785			break;
 786		}
 787	}
 788out:
 789	sysctl_head_finish(head);
 790	return 0;
 791}
 792
 793static int proc_sys_permission(struct mnt_idmap *idmap,
 794			       struct inode *inode, int mask)
 795{
 796	/*
 797	 * sysctl entries that are not writeable,
 798	 * are _NOT_ writeable, capabilities or not.
 799	 */
 800	struct ctl_table_header *head;
 801	const struct ctl_table *table;
 802	int error;
 803
 804	/* Executable files are not allowed under /proc/sys/ */
 805	if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
 806		return -EACCES;
 807
 808	head = grab_header(inode);
 809	if (IS_ERR(head))
 810		return PTR_ERR(head);
 811
 812	table = PROC_I(inode)->sysctl_entry;
 813	if (!table) /* global root - r-xr-xr-x */
 814		error = mask & MAY_WRITE ? -EACCES : 0;
 815	else /* Use the permissions on the sysctl table entry */
 816		error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
 817
 818	sysctl_head_finish(head);
 819	return error;
 820}
 821
 822static int proc_sys_setattr(struct mnt_idmap *idmap,
 823			    struct dentry *dentry, struct iattr *attr)
 824{
 825	struct inode *inode = d_inode(dentry);
 826	int error;
 827
 828	if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
 829		return -EPERM;
 830
 831	error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
 832	if (error)
 833		return error;
 834
 835	setattr_copy(&nop_mnt_idmap, inode, attr);
 836	return 0;
 837}
 838
 839static int proc_sys_getattr(struct mnt_idmap *idmap,
 840			    const struct path *path, struct kstat *stat,
 841			    u32 request_mask, unsigned int query_flags)
 842{
 843	struct inode *inode = d_inode(path->dentry);
 844	struct ctl_table_header *head = grab_header(inode);
 845	const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 846
 847	if (IS_ERR(head))
 848		return PTR_ERR(head);
 849
 850	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
 851	if (table)
 852		stat->mode = (stat->mode & S_IFMT) | table->mode;
 853
 854	sysctl_head_finish(head);
 855	return 0;
 856}
 857
 858static const struct file_operations proc_sys_file_operations = {
 859	.open		= proc_sys_open,
 860	.poll		= proc_sys_poll,
 861	.read_iter	= proc_sys_read,
 862	.write_iter	= proc_sys_write,
 863	.splice_read	= copy_splice_read,
 864	.splice_write	= iter_file_splice_write,
 865	.llseek		= default_llseek,
 866};
 867
 868static const struct file_operations proc_sys_dir_file_operations = {
 869	.read		= generic_read_dir,
 870	.iterate_shared	= proc_sys_readdir,
 871	.llseek		= generic_file_llseek,
 872};
 873
 874static const struct inode_operations proc_sys_inode_operations = {
 875	.permission	= proc_sys_permission,
 876	.setattr	= proc_sys_setattr,
 877	.getattr	= proc_sys_getattr,
 878};
 879
 880static const struct inode_operations proc_sys_dir_operations = {
 881	.lookup		= proc_sys_lookup,
 882	.permission	= proc_sys_permission,
 883	.setattr	= proc_sys_setattr,
 884	.getattr	= proc_sys_getattr,
 885};
 886
 887static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
 888{
 889	if (flags & LOOKUP_RCU)
 890		return -ECHILD;
 891	return !PROC_I(d_inode(dentry))->sysctl->unregistering;
 892}
 893
 894static int proc_sys_delete(const struct dentry *dentry)
 895{
 896	return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
 897}
 898
 899static int sysctl_is_seen(struct ctl_table_header *p)
 900{
 901	struct ctl_table_set *set = p->set;
 902	int res;
 903	spin_lock(&sysctl_lock);
 904	if (p->unregistering)
 905		res = 0;
 906	else if (!set->is_seen)
 907		res = 1;
 908	else
 909		res = set->is_seen(set);
 910	spin_unlock(&sysctl_lock);
 911	return res;
 912}
 913
 914static int proc_sys_compare(const struct dentry *dentry,
 915		unsigned int len, const char *str, const struct qstr *name)
 916{
 917	struct ctl_table_header *head;
 918	struct inode *inode;
 919
 920	/* Although proc doesn't have negative dentries, rcu-walk means
 921	 * that inode here can be NULL */
 922	/* AV: can it, indeed? */
 923	inode = d_inode_rcu(dentry);
 924	if (!inode)
 925		return 1;
 926	if (name->len != len)
 927		return 1;
 928	if (memcmp(name->name, str, len))
 929		return 1;
 930	head = rcu_dereference(PROC_I(inode)->sysctl);
 931	return !head || !sysctl_is_seen(head);
 932}
 933
 934static const struct dentry_operations proc_sys_dentry_operations = {
 935	.d_revalidate	= proc_sys_revalidate,
 936	.d_delete	= proc_sys_delete,
 937	.d_compare	= proc_sys_compare,
 938};
 939
 940static struct ctl_dir *find_subdir(struct ctl_dir *dir,
 941				   const char *name, int namelen)
 942{
 943	struct ctl_table_header *head;
 944	const struct ctl_table *entry;
 945
 946	entry = find_entry(&head, dir, name, namelen);
 947	if (!entry)
 948		return ERR_PTR(-ENOENT);
 949	if (!S_ISDIR(entry->mode))
 950		return ERR_PTR(-ENOTDIR);
 951	return container_of(head, struct ctl_dir, header);
 952}
 953
 954static struct ctl_dir *new_dir(struct ctl_table_set *set,
 955			       const char *name, int namelen)
 956{
 957	struct ctl_table *table;
 958	struct ctl_dir *new;
 959	struct ctl_node *node;
 960	char *new_name;
 961
 962	new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
 963		      sizeof(struct ctl_table) +  namelen + 1,
 964		      GFP_KERNEL);
 965	if (!new)
 966		return NULL;
 967
 968	node = (struct ctl_node *)(new + 1);
 969	table = (struct ctl_table *)(node + 1);
 970	new_name = (char *)(table + 1);
 971	memcpy(new_name, name, namelen);
 972	table[0].procname = new_name;
 973	table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
 974	init_header(&new->header, set->dir.header.root, set, node, table, 1);
 975
 976	return new;
 977}
 978
 979/**
 980 * get_subdir - find or create a subdir with the specified name.
 981 * @dir:  Directory to create the subdirectory in
 982 * @name: The name of the subdirectory to find or create
 983 * @namelen: The length of name
 984 *
 985 * Takes a directory with an elevated reference count so we know that
 986 * if we drop the lock the directory will not go away.  Upon success
 987 * the reference is moved from @dir to the returned subdirectory.
 988 * Upon error an error code is returned and the reference on @dir is
 989 * simply dropped.
 990 */
 991static struct ctl_dir *get_subdir(struct ctl_dir *dir,
 992				  const char *name, int namelen)
 993{
 994	struct ctl_table_set *set = dir->header.set;
 995	struct ctl_dir *subdir, *new = NULL;
 996	int err;
 997
 998	spin_lock(&sysctl_lock);
 999	subdir = find_subdir(dir, name, namelen);
1000	if (!IS_ERR(subdir))
1001		goto found;
1002	if (PTR_ERR(subdir) != -ENOENT)
1003		goto failed;
1004
1005	spin_unlock(&sysctl_lock);
1006	new = new_dir(set, name, namelen);
1007	spin_lock(&sysctl_lock);
1008	subdir = ERR_PTR(-ENOMEM);
1009	if (!new)
1010		goto failed;
1011
1012	/* Was the subdir added while we dropped the lock? */
1013	subdir = find_subdir(dir, name, namelen);
1014	if (!IS_ERR(subdir))
1015		goto found;
1016	if (PTR_ERR(subdir) != -ENOENT)
1017		goto failed;
1018
1019	/* Nope.  Use the our freshly made directory entry. */
1020	err = insert_header(dir, &new->header);
1021	subdir = ERR_PTR(err);
1022	if (err)
1023		goto failed;
1024	subdir = new;
1025found:
1026	subdir->header.nreg++;
1027failed:
1028	if (IS_ERR(subdir)) {
1029		pr_err("sysctl could not get directory: ");
1030		sysctl_print_dir(dir);
1031		pr_cont("%*.*s %ld\n", namelen, namelen, name,
1032			PTR_ERR(subdir));
1033	}
1034	drop_sysctl_table(&dir->header);
1035	if (new)
1036		drop_sysctl_table(&new->header);
1037	spin_unlock(&sysctl_lock);
1038	return subdir;
1039}
1040
1041static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1042{
1043	struct ctl_dir *parent;
1044	const char *procname;
1045	if (!dir->header.parent)
1046		return &set->dir;
1047	parent = xlate_dir(set, dir->header.parent);
1048	if (IS_ERR(parent))
1049		return parent;
1050	procname = dir->header.ctl_table[0].procname;
1051	return find_subdir(parent, procname, strlen(procname));
1052}
1053
1054static int sysctl_follow_link(struct ctl_table_header **phead,
1055	const struct ctl_table **pentry)
1056{
1057	struct ctl_table_header *head;
1058	const struct ctl_table *entry;
1059	struct ctl_table_root *root;
1060	struct ctl_table_set *set;
 
1061	struct ctl_dir *dir;
1062	int ret;
1063
1064	spin_lock(&sysctl_lock);
1065	root = (*pentry)->data;
1066	set = lookup_header_set(root);
1067	dir = xlate_dir(set, (*phead)->parent);
1068	if (IS_ERR(dir))
1069		ret = PTR_ERR(dir);
1070	else {
1071		const char *procname = (*pentry)->procname;
1072		head = NULL;
1073		entry = find_entry(&head, dir, procname, strlen(procname));
1074		ret = -ENOENT;
1075		if (entry && use_table(head)) {
1076			unuse_table(*phead);
1077			*phead = head;
1078			*pentry = entry;
1079			ret = 0;
1080		}
1081	}
1082
1083	spin_unlock(&sysctl_lock);
1084	return ret;
1085}
1086
1087static int sysctl_err(const char *path, const struct ctl_table *table, char *fmt, ...)
1088{
1089	struct va_format vaf;
1090	va_list args;
1091
1092	va_start(args, fmt);
1093	vaf.fmt = fmt;
1094	vaf.va = &args;
1095
1096	pr_err("sysctl table check failed: %s/%s %pV\n",
1097	       path, table->procname, &vaf);
1098
1099	va_end(args);
1100	return -EINVAL;
1101}
1102
1103static int sysctl_check_table_array(const char *path, const struct ctl_table *table)
1104{
1105	unsigned int extra;
1106	int err = 0;
1107
1108	if ((table->proc_handler == proc_douintvec) ||
1109	    (table->proc_handler == proc_douintvec_minmax)) {
1110		if (table->maxlen != sizeof(unsigned int))
1111			err |= sysctl_err(path, table, "array not allowed");
1112	}
1113
1114	if (table->proc_handler == proc_dou8vec_minmax) {
1115		if (table->maxlen != sizeof(u8))
1116			err |= sysctl_err(path, table, "array not allowed");
1117
1118		if (table->extra1) {
1119			extra = *(unsigned int *) table->extra1;
1120			if (extra > 255U)
1121				err |= sysctl_err(path, table,
1122						"range value too large for proc_dou8vec_minmax");
1123		}
1124		if (table->extra2) {
1125			extra = *(unsigned int *) table->extra2;
1126			if (extra > 255U)
1127				err |= sysctl_err(path, table,
1128						"range value too large for proc_dou8vec_minmax");
1129		}
1130	}
1131
1132	if (table->proc_handler == proc_dobool) {
1133		if (table->maxlen != sizeof(bool))
1134			err |= sysctl_err(path, table, "array not allowed");
1135	}
1136
1137	return err;
1138}
1139
1140static int sysctl_check_table(const char *path, struct ctl_table_header *header)
1141{
1142	const struct ctl_table *entry;
1143	int err = 0;
1144	list_for_each_table_entry(entry, header) {
1145		if (!entry->procname)
1146			err |= sysctl_err(path, entry, "procname is null");
1147		if ((entry->proc_handler == proc_dostring) ||
1148		    (entry->proc_handler == proc_dobool) ||
1149		    (entry->proc_handler == proc_dointvec) ||
1150		    (entry->proc_handler == proc_douintvec) ||
1151		    (entry->proc_handler == proc_douintvec_minmax) ||
1152		    (entry->proc_handler == proc_dointvec_minmax) ||
1153		    (entry->proc_handler == proc_dou8vec_minmax) ||
1154		    (entry->proc_handler == proc_dointvec_jiffies) ||
1155		    (entry->proc_handler == proc_dointvec_userhz_jiffies) ||
1156		    (entry->proc_handler == proc_dointvec_ms_jiffies) ||
1157		    (entry->proc_handler == proc_doulongvec_minmax) ||
1158		    (entry->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1159			if (!entry->data)
1160				err |= sysctl_err(path, entry, "No data");
1161			if (!entry->maxlen)
1162				err |= sysctl_err(path, entry, "No maxlen");
1163			else
1164				err |= sysctl_check_table_array(path, entry);
1165		}
1166		if (!entry->proc_handler)
1167			err |= sysctl_err(path, entry, "No proc_handler");
1168
1169		if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode)
1170			err |= sysctl_err(path, entry, "bogus .mode 0%o",
1171				entry->mode);
1172	}
1173	return err;
1174}
1175
1176static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table_header *head)
1177{
1178	struct ctl_table *link_table, *link;
1179	struct ctl_table_header *links;
1180	const struct ctl_table *entry;
1181	struct ctl_node *node;
1182	char *link_name;
1183	int name_bytes;
1184
1185	name_bytes = 0;
 
1186	list_for_each_table_entry(entry, head) {
 
1187		name_bytes += strlen(entry->procname) + 1;
1188	}
1189
1190	links = kzalloc(sizeof(struct ctl_table_header) +
1191			sizeof(struct ctl_node)*head->ctl_table_size +
1192			sizeof(struct ctl_table)*head->ctl_table_size +
1193			name_bytes,
1194			GFP_KERNEL);
1195
1196	if (!links)
1197		return NULL;
1198
1199	node = (struct ctl_node *)(links + 1);
1200	link_table = (struct ctl_table *)(node + head->ctl_table_size);
1201	link_name = (char *)(link_table + head->ctl_table_size);
1202	link = link_table;
1203
1204	list_for_each_table_entry(entry, head) {
1205		int len = strlen(entry->procname) + 1;
1206		memcpy(link_name, entry->procname, len);
1207		link->procname = link_name;
1208		link->mode = S_IFLNK|S_IRWXUGO;
1209		link->data = head->root;
1210		link_name += len;
1211		link++;
1212	}
1213	init_header(links, dir->header.root, dir->header.set, node, link_table,
1214		    head->ctl_table_size);
1215	links->nreg = head->ctl_table_size;
1216
1217	return links;
1218}
1219
1220static bool get_links(struct ctl_dir *dir,
1221		      struct ctl_table_header *header,
1222		      struct ctl_table_root *link_root)
1223{
1224	struct ctl_table_header *tmp_head;
1225	const struct ctl_table *entry, *link;
1226
1227	if (header->ctl_table_size == 0 ||
1228	    sysctl_is_perm_empty_ctl_header(header))
1229		return true;
1230
1231	/* Are there links available for every entry in table? */
1232	list_for_each_table_entry(entry, header) {
1233		const char *procname = entry->procname;
1234		link = find_entry(&tmp_head, dir, procname, strlen(procname));
1235		if (!link)
1236			return false;
1237		if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1238			continue;
1239		if (S_ISLNK(link->mode) && (link->data == link_root))
1240			continue;
1241		return false;
1242	}
1243
1244	/* The checks passed.  Increase the registration count on the links */
1245	list_for_each_table_entry(entry, header) {
1246		const char *procname = entry->procname;
1247		link = find_entry(&tmp_head, dir, procname, strlen(procname));
1248		tmp_head->nreg++;
1249	}
1250	return true;
1251}
1252
1253static int insert_links(struct ctl_table_header *head)
1254{
1255	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1256	struct ctl_dir *core_parent;
1257	struct ctl_table_header *links;
1258	int err;
1259
1260	if (head->set == root_set)
1261		return 0;
1262
1263	core_parent = xlate_dir(root_set, head->parent);
1264	if (IS_ERR(core_parent))
1265		return 0;
1266
1267	if (get_links(core_parent, head, head->root))
1268		return 0;
1269
1270	core_parent->header.nreg++;
1271	spin_unlock(&sysctl_lock);
1272
1273	links = new_links(core_parent, head);
1274
1275	spin_lock(&sysctl_lock);
1276	err = -ENOMEM;
1277	if (!links)
1278		goto out;
1279
1280	err = 0;
1281	if (get_links(core_parent, head, head->root)) {
1282		kfree(links);
1283		goto out;
1284	}
1285
1286	err = insert_header(core_parent, links);
1287	if (err)
1288		kfree(links);
1289out:
1290	drop_sysctl_table(&core_parent->header);
1291	return err;
1292}
1293
1294/* Find the directory for the ctl_table. If one is not found create it. */
1295static struct ctl_dir *sysctl_mkdir_p(struct ctl_dir *dir, const char *path)
1296{
1297	const char *name, *nextname;
1298
1299	for (name = path; name; name = nextname) {
1300		int namelen;
1301		nextname = strchr(name, '/');
1302		if (nextname) {
1303			namelen = nextname - name;
1304			nextname++;
1305		} else {
1306			namelen = strlen(name);
1307		}
1308		if (namelen == 0)
1309			continue;
1310
1311		/*
1312		 * namelen ensures if name is "foo/bar/yay" only foo is
1313		 * registered first. We traverse as if using mkdir -p and
1314		 * return a ctl_dir for the last directory entry.
1315		 */
1316		dir = get_subdir(dir, name, namelen);
1317		if (IS_ERR(dir))
1318			break;
1319	}
1320	return dir;
1321}
1322
1323/**
1324 * __register_sysctl_table - register a leaf sysctl table
1325 * @set: Sysctl tree to register on
1326 * @path: The path to the directory the sysctl table is in.
1327 *
1328 * @table: the top-level table structure. This table should not be free'd
1329 *         after registration. So it should not be used on stack. It can either
1330 *         be a global or dynamically allocated by the caller and free'd later
1331 *         after sysctl unregistration.
1332 * @table_size : The number of elements in table
1333 *
1334 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1335 * array.
1336 *
1337 * The members of the &struct ctl_table structure are used as follows:
 
1338 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1339 *            enter a sysctl file
1340 * data     - a pointer to data for use by proc_handler
1341 * maxlen   - the maximum size in bytes of the data
1342 * mode     - the file permissions for the /proc/sys file
1343 * type     - Defines the target type (described in struct definition)
 
 
 
 
 
1344 * proc_handler - the text handler routine (described below)
1345 *
1346 * extra1, extra2 - extra pointers usable by the proc handler routines
1347 * XXX: we should eventually modify these to use long min / max [0]
1348 * [0] https://lkml.kernel.org/87zgpte9o4.fsf@email.froward.int.ebiederm.org
1349 *
1350 * Leaf nodes in the sysctl tree will be represented by a single file
1351 * under /proc; non-leaf nodes are not allowed.
 
1352 *
1353 * There must be a proc_handler routine for any terminal nodes.
1354 * Several default handlers are available to cover common cases -
1355 *
1356 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1357 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1358 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1359 *
1360 * It is the handler's job to read the input buffer from user memory
1361 * and process it. The handler should return 0 on success.
1362 *
1363 * This routine returns %NULL on a failure to register, and a pointer
1364 * to the table header on success.
1365 */
1366struct ctl_table_header *__register_sysctl_table(
1367	struct ctl_table_set *set,
1368	const char *path, const struct ctl_table *table, size_t table_size)
1369{
1370	struct ctl_table_root *root = set->dir.header.root;
1371	struct ctl_table_header *header;
1372	struct ctl_dir *dir;
1373	struct ctl_node *node;
1374
1375	header = kzalloc(sizeof(struct ctl_table_header) +
1376			 sizeof(struct ctl_node)*table_size, GFP_KERNEL_ACCOUNT);
1377	if (!header)
1378		return NULL;
1379
1380	node = (struct ctl_node *)(header + 1);
1381	init_header(header, root, set, node, table, table_size);
1382	if (sysctl_check_table(path, header))
1383		goto fail;
1384
1385	spin_lock(&sysctl_lock);
1386	dir = &set->dir;
1387	/* Reference moved down the directory tree get_subdir */
1388	dir->header.nreg++;
1389	spin_unlock(&sysctl_lock);
1390
1391	dir = sysctl_mkdir_p(dir, path);
1392	if (IS_ERR(dir))
1393		goto fail;
1394	spin_lock(&sysctl_lock);
1395	if (insert_header(dir, header))
1396		goto fail_put_dir_locked;
1397
1398	drop_sysctl_table(&dir->header);
1399	spin_unlock(&sysctl_lock);
1400
1401	return header;
1402
1403fail_put_dir_locked:
1404	drop_sysctl_table(&dir->header);
1405	spin_unlock(&sysctl_lock);
1406fail:
1407	kfree(header);
1408	return NULL;
1409}
1410
1411/**
1412 * register_sysctl_sz - register a sysctl table
1413 * @path: The path to the directory the sysctl table is in. If the path
1414 * 	doesn't exist we will create it for you.
1415 * @table: the table structure. The calller must ensure the life of the @table
1416 * 	will be kept during the lifetime use of the syctl. It must not be freed
1417 * 	until unregister_sysctl_table() is called with the given returned table
1418 * 	with this registration. If your code is non modular then you don't need
1419 * 	to call unregister_sysctl_table() and can instead use something like
1420 * 	register_sysctl_init() which does not care for the result of the syctl
1421 * 	registration.
1422 * @table_size: The number of elements in table.
1423 *
1424 * Register a sysctl table. @table should be a filled in ctl_table
1425 * array. A completely 0 filled entry terminates the table.
1426 *
1427 * See __register_sysctl_table for more details.
1428 */
1429struct ctl_table_header *register_sysctl_sz(const char *path, const struct ctl_table *table,
1430					    size_t table_size)
1431{
1432	return __register_sysctl_table(&sysctl_table_root.default_set,
1433					path, table, table_size);
1434}
1435EXPORT_SYMBOL(register_sysctl_sz);
1436
1437/**
1438 * __register_sysctl_init() - register sysctl table to path
1439 * @path: path name for sysctl base. If that path doesn't exist we will create
1440 * 	it for you.
1441 * @table: This is the sysctl table that needs to be registered to the path.
1442 * 	The caller must ensure the life of the @table will be kept during the
1443 * 	lifetime use of the sysctl.
1444 * @table_name: The name of sysctl table, only used for log printing when
1445 *              registration fails
1446 * @table_size: The number of elements in table
1447 *
1448 * The sysctl interface is used by userspace to query or modify at runtime
1449 * a predefined value set on a variable. These variables however have default
1450 * values pre-set. Code which depends on these variables will always work even
1451 * if register_sysctl() fails. If register_sysctl() fails you'd just loose the
1452 * ability to query or modify the sysctls dynamically at run time. Chances of
1453 * register_sysctl() failing on init are extremely low, and so for both reasons
1454 * this function does not return any error as it is used by initialization code.
1455 *
1456 * Context: if your base directory does not exist it will be created for you.
1457 */
1458void __init __register_sysctl_init(const char *path, const struct ctl_table *table,
1459				 const char *table_name, size_t table_size)
1460{
1461	struct ctl_table_header *hdr = register_sysctl_sz(path, table, table_size);
1462
1463	if (unlikely(!hdr)) {
1464		pr_err("failed when register_sysctl_sz %s to %s\n", table_name, path);
1465		return;
1466	}
1467	kmemleak_not_leak(hdr);
1468}
1469
1470static void put_links(struct ctl_table_header *header)
1471{
1472	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1473	struct ctl_table_root *root = header->root;
1474	struct ctl_dir *parent = header->parent;
1475	struct ctl_dir *core_parent;
1476	const struct ctl_table *entry;
1477
1478	if (header->set == root_set)
1479		return;
1480
1481	core_parent = xlate_dir(root_set, parent);
1482	if (IS_ERR(core_parent))
1483		return;
1484
1485	list_for_each_table_entry(entry, header) {
1486		struct ctl_table_header *link_head;
1487		const struct ctl_table *link;
1488		const char *name = entry->procname;
1489
1490		link = find_entry(&link_head, core_parent, name, strlen(name));
1491		if (link &&
1492		    ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1493		     (S_ISLNK(link->mode) && (link->data == root)))) {
1494			drop_sysctl_table(link_head);
1495		}
1496		else {
1497			pr_err("sysctl link missing during unregister: ");
1498			sysctl_print_dir(parent);
1499			pr_cont("%s\n", name);
1500		}
1501	}
1502}
1503
1504static void drop_sysctl_table(struct ctl_table_header *header)
1505{
1506	struct ctl_dir *parent = header->parent;
1507
1508	if (--header->nreg)
1509		return;
1510
1511	if (parent) {
1512		put_links(header);
1513		start_unregistering(header);
1514	}
1515
1516	if (!--header->count)
1517		kfree_rcu(header, rcu);
1518
1519	if (parent)
1520		drop_sysctl_table(&parent->header);
1521}
1522
1523/**
1524 * unregister_sysctl_table - unregister a sysctl table hierarchy
1525 * @header: the header returned from register_sysctl or __register_sysctl_table
1526 *
1527 * Unregisters the sysctl table and all children. proc entries may not
1528 * actually be removed until they are no longer used by anyone.
1529 */
1530void unregister_sysctl_table(struct ctl_table_header * header)
1531{
1532	might_sleep();
1533
1534	if (header == NULL)
1535		return;
1536
1537	spin_lock(&sysctl_lock);
1538	drop_sysctl_table(header);
1539	spin_unlock(&sysctl_lock);
1540}
1541EXPORT_SYMBOL(unregister_sysctl_table);
1542
1543void setup_sysctl_set(struct ctl_table_set *set,
1544	struct ctl_table_root *root,
1545	int (*is_seen)(struct ctl_table_set *))
1546{
1547	memset(set, 0, sizeof(*set));
1548	set->is_seen = is_seen;
1549	init_header(&set->dir.header, root, set, NULL, root_table, 1);
1550}
1551
1552void retire_sysctl_set(struct ctl_table_set *set)
1553{
1554	WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1555}
1556
1557int __init proc_sys_init(void)
1558{
1559	struct proc_dir_entry *proc_sys_root;
1560
1561	proc_sys_root = proc_mkdir("sys", NULL);
1562	proc_sys_root->proc_iops = &proc_sys_dir_operations;
1563	proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1564	proc_sys_root->nlink = 0;
1565
1566	return sysctl_init_bases();
1567}
1568
1569struct sysctl_alias {
1570	const char *kernel_param;
1571	const char *sysctl_param;
1572};
1573
1574/*
1575 * Historically some settings had both sysctl and a command line parameter.
1576 * With the generic sysctl. parameter support, we can handle them at a single
1577 * place and only keep the historical name for compatibility. This is not meant
1578 * to add brand new aliases. When adding existing aliases, consider whether
1579 * the possibly different moment of changing the value (e.g. from early_param
1580 * to the moment do_sysctl_args() is called) is an issue for the specific
1581 * parameter.
1582 */
1583static const struct sysctl_alias sysctl_aliases[] = {
1584	{"hardlockup_all_cpu_backtrace",	"kernel.hardlockup_all_cpu_backtrace" },
1585	{"hung_task_panic",			"kernel.hung_task_panic" },
1586	{"numa_zonelist_order",			"vm.numa_zonelist_order" },
1587	{"softlockup_all_cpu_backtrace",	"kernel.softlockup_all_cpu_backtrace" },
1588	{ }
1589};
1590
1591static const char *sysctl_find_alias(char *param)
1592{
1593	const struct sysctl_alias *alias;
1594
1595	for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1596		if (strcmp(alias->kernel_param, param) == 0)
1597			return alias->sysctl_param;
1598	}
1599
1600	return NULL;
1601}
1602
1603bool sysctl_is_alias(char *param)
1604{
1605	const char *alias = sysctl_find_alias(param);
1606
1607	return alias != NULL;
1608}
1609
1610/* Set sysctl value passed on kernel command line. */
1611static int process_sysctl_arg(char *param, char *val,
1612			       const char *unused, void *arg)
1613{
1614	char *path;
1615	struct vfsmount **proc_mnt = arg;
1616	struct file_system_type *proc_fs_type;
1617	struct file *file;
1618	int len;
1619	int err;
1620	loff_t pos = 0;
1621	ssize_t wret;
1622
1623	if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1624		param += sizeof("sysctl") - 1;
1625
1626		if (param[0] != '/' && param[0] != '.')
1627			return 0;
1628
1629		param++;
1630	} else {
1631		param = (char *) sysctl_find_alias(param);
1632		if (!param)
1633			return 0;
1634	}
1635
1636	if (!val)
1637		return -EINVAL;
1638	len = strlen(val);
1639	if (len == 0)
1640		return -EINVAL;
1641
1642	/*
1643	 * To set sysctl options, we use a temporary mount of proc, look up the
1644	 * respective sys/ file and write to it. To avoid mounting it when no
1645	 * options were given, we mount it only when the first sysctl option is
1646	 * found. Why not a persistent mount? There are problems with a
1647	 * persistent mount of proc in that it forces userspace not to use any
1648	 * proc mount options.
1649	 */
1650	if (!*proc_mnt) {
1651		proc_fs_type = get_fs_type("proc");
1652		if (!proc_fs_type) {
1653			pr_err("Failed to find procfs to set sysctl from command line\n");
1654			return 0;
1655		}
1656		*proc_mnt = kern_mount(proc_fs_type);
1657		put_filesystem(proc_fs_type);
1658		if (IS_ERR(*proc_mnt)) {
1659			pr_err("Failed to mount procfs to set sysctl from command line\n");
1660			return 0;
1661		}
1662	}
1663
1664	path = kasprintf(GFP_KERNEL, "sys/%s", param);
1665	if (!path)
1666		panic("%s: Failed to allocate path for %s\n", __func__, param);
1667	strreplace(path, '.', '/');
1668
1669	file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1670	if (IS_ERR(file)) {
1671		err = PTR_ERR(file);
1672		if (err == -ENOENT)
1673			pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1674				param, val);
1675		else if (err == -EACCES)
1676			pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1677				param, val);
1678		else
1679			pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1680				file, param, val);
1681		goto out;
1682	}
1683	wret = kernel_write(file, val, len, &pos);
1684	if (wret < 0) {
1685		err = wret;
1686		if (err == -EINVAL)
1687			pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1688				param, val);
1689		else
1690			pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1691				ERR_PTR(err), param, val);
1692	} else if (wret != len) {
1693		pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1694			wret, len, path, param, val);
1695	}
1696
1697	err = filp_close(file, NULL);
1698	if (err)
1699		pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1700			ERR_PTR(err), param, val);
1701out:
1702	kfree(path);
1703	return 0;
1704}
1705
1706void do_sysctl_args(void)
1707{
1708	char *command_line;
1709	struct vfsmount *proc_mnt = NULL;
1710
1711	command_line = kstrdup(saved_command_line, GFP_KERNEL);
1712	if (!command_line)
1713		panic("%s: Failed to allocate copy of command line\n", __func__);
1714
1715	parse_args("Setting sysctl args", command_line,
1716		   NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1717
1718	if (proc_mnt)
1719		kern_unmount(proc_mnt);
1720
1721	kfree(command_line);
1722}
v6.8
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * /proc/sys support
   4 */
   5#include <linux/init.h>
   6#include <linux/sysctl.h>
   7#include <linux/poll.h>
   8#include <linux/proc_fs.h>
   9#include <linux/printk.h>
  10#include <linux/security.h>
  11#include <linux/sched.h>
  12#include <linux/cred.h>
  13#include <linux/namei.h>
  14#include <linux/mm.h>
  15#include <linux/uio.h>
  16#include <linux/module.h>
  17#include <linux/bpf-cgroup.h>
  18#include <linux/mount.h>
  19#include <linux/kmemleak.h>
 
  20#include "internal.h"
  21
  22#define list_for_each_table_entry(entry, header)	\
  23	entry = header->ctl_table;			\
  24	for (size_t i = 0 ; i < header->ctl_table_size && entry->procname; ++i, entry++)
  25
  26static const struct dentry_operations proc_sys_dentry_operations;
  27static const struct file_operations proc_sys_file_operations;
  28static const struct inode_operations proc_sys_inode_operations;
  29static const struct file_operations proc_sys_dir_file_operations;
  30static const struct inode_operations proc_sys_dir_operations;
  31
  32/* Support for permanently empty directories */
  33static struct ctl_table sysctl_mount_point[] = {
  34	{.type = SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY }
 
 
 
  35};
  36
  37/**
  38 * register_sysctl_mount_point() - registers a sysctl mount point
  39 * @path: path for the mount point
  40 *
  41 * Used to create a permanently empty directory to serve as mount point.
  42 * There are some subtle but important permission checks this allows in the
  43 * case of unprivileged mounts.
  44 */
  45struct ctl_table_header *register_sysctl_mount_point(const char *path)
  46{
  47	return register_sysctl(path, sysctl_mount_point);
  48}
  49EXPORT_SYMBOL(register_sysctl_mount_point);
  50
  51#define sysctl_is_perm_empty_ctl_table(tptr)		\
  52	(tptr[0].type == SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
  53#define sysctl_is_perm_empty_ctl_header(hptr)		\
  54	(sysctl_is_perm_empty_ctl_table(hptr->ctl_table))
  55#define sysctl_set_perm_empty_ctl_header(hptr)		\
  56	(hptr->ctl_table[0].type = SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
  57#define sysctl_clear_perm_empty_ctl_header(hptr)	\
  58	(hptr->ctl_table[0].type = SYSCTL_TABLE_TYPE_DEFAULT)
  59
  60void proc_sys_poll_notify(struct ctl_table_poll *poll)
  61{
  62	if (!poll)
  63		return;
  64
  65	atomic_inc(&poll->event);
  66	wake_up_interruptible(&poll->wait);
  67}
  68
  69static struct ctl_table root_table[] = {
  70	{
  71		.procname = "",
  72		.mode = S_IFDIR|S_IRUGO|S_IXUGO,
  73	},
  74};
  75static struct ctl_table_root sysctl_table_root = {
  76	.default_set.dir.header = {
  77		{{.count = 1,
  78		  .nreg = 1,
  79		  .ctl_table = root_table }},
  80		.ctl_table_arg = root_table,
  81		.root = &sysctl_table_root,
  82		.set = &sysctl_table_root.default_set,
  83	},
  84};
  85
  86static DEFINE_SPINLOCK(sysctl_lock);
  87
  88static void drop_sysctl_table(struct ctl_table_header *header);
  89static int sysctl_follow_link(struct ctl_table_header **phead,
  90	struct ctl_table **pentry);
  91static int insert_links(struct ctl_table_header *head);
  92static void put_links(struct ctl_table_header *header);
  93
  94static void sysctl_print_dir(struct ctl_dir *dir)
  95{
  96	if (dir->header.parent)
  97		sysctl_print_dir(dir->header.parent);
  98	pr_cont("%s/", dir->header.ctl_table[0].procname);
  99}
 100
 101static int namecmp(const char *name1, int len1, const char *name2, int len2)
 102{
 103	int cmp;
 104
 105	cmp = memcmp(name1, name2, min(len1, len2));
 106	if (cmp == 0)
 107		cmp = len1 - len2;
 108	return cmp;
 109}
 110
 111/* Called under sysctl_lock */
 112static struct ctl_table *find_entry(struct ctl_table_header **phead,
 113	struct ctl_dir *dir, const char *name, int namelen)
 114{
 115	struct ctl_table_header *head;
 116	struct ctl_table *entry;
 117	struct rb_node *node = dir->root.rb_node;
 118
 
 
 119	while (node)
 120	{
 121		struct ctl_node *ctl_node;
 122		const char *procname;
 123		int cmp;
 124
 125		ctl_node = rb_entry(node, struct ctl_node, node);
 126		head = ctl_node->header;
 127		entry = &head->ctl_table[ctl_node - head->node];
 128		procname = entry->procname;
 129
 130		cmp = namecmp(name, namelen, procname, strlen(procname));
 131		if (cmp < 0)
 132			node = node->rb_left;
 133		else if (cmp > 0)
 134			node = node->rb_right;
 135		else {
 136			*phead = head;
 137			return entry;
 138		}
 139	}
 140	return NULL;
 141}
 142
 143static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
 144{
 145	struct rb_node *node = &head->node[entry - head->ctl_table].node;
 146	struct rb_node **p = &head->parent->root.rb_node;
 147	struct rb_node *parent = NULL;
 148	const char *name = entry->procname;
 149	int namelen = strlen(name);
 150
 151	while (*p) {
 152		struct ctl_table_header *parent_head;
 153		struct ctl_table *parent_entry;
 154		struct ctl_node *parent_node;
 155		const char *parent_name;
 156		int cmp;
 157
 158		parent = *p;
 159		parent_node = rb_entry(parent, struct ctl_node, node);
 160		parent_head = parent_node->header;
 161		parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
 162		parent_name = parent_entry->procname;
 163
 164		cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
 165		if (cmp < 0)
 166			p = &(*p)->rb_left;
 167		else if (cmp > 0)
 168			p = &(*p)->rb_right;
 169		else {
 170			pr_err("sysctl duplicate entry: ");
 171			sysctl_print_dir(head->parent);
 172			pr_cont("%s\n", entry->procname);
 173			return -EEXIST;
 174		}
 175	}
 176
 177	rb_link_node(node, parent, p);
 178	rb_insert_color(node, &head->parent->root);
 179	return 0;
 180}
 181
 182static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
 183{
 184	struct rb_node *node = &head->node[entry - head->ctl_table].node;
 185
 186	rb_erase(node, &head->parent->root);
 187}
 188
 189static void init_header(struct ctl_table_header *head,
 190	struct ctl_table_root *root, struct ctl_table_set *set,
 191	struct ctl_node *node, struct ctl_table *table, size_t table_size)
 192{
 193	head->ctl_table = table;
 194	head->ctl_table_size = table_size;
 195	head->ctl_table_arg = table;
 196	head->used = 0;
 197	head->count = 1;
 198	head->nreg = 1;
 199	head->unregistering = NULL;
 200	head->root = root;
 201	head->set = set;
 202	head->parent = NULL;
 203	head->node = node;
 204	INIT_HLIST_HEAD(&head->inodes);
 205	if (node) {
 206		struct ctl_table *entry;
 207
 208		list_for_each_table_entry(entry, head) {
 209			node->header = head;
 210			node++;
 211		}
 212	}
 
 
 213}
 214
 215static void erase_header(struct ctl_table_header *head)
 216{
 217	struct ctl_table *entry;
 218
 219	list_for_each_table_entry(entry, head)
 220		erase_entry(head, entry);
 221}
 222
 223static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
 224{
 225	struct ctl_table *entry;
 226	struct ctl_table_header *dir_h = &dir->header;
 227	int err;
 228
 229
 230	/* Is this a permanently empty directory? */
 231	if (sysctl_is_perm_empty_ctl_header(dir_h))
 232		return -EROFS;
 233
 234	/* Am I creating a permanently empty directory? */
 235	if (header->ctl_table_size > 0 &&
 236	    sysctl_is_perm_empty_ctl_table(header->ctl_table)) {
 237		if (!RB_EMPTY_ROOT(&dir->root))
 238			return -EINVAL;
 239		sysctl_set_perm_empty_ctl_header(dir_h);
 240	}
 241
 242	dir_h->nreg++;
 243	header->parent = dir;
 244	err = insert_links(header);
 245	if (err)
 246		goto fail_links;
 247	list_for_each_table_entry(entry, header) {
 248		err = insert_entry(header, entry);
 249		if (err)
 250			goto fail;
 251	}
 252	return 0;
 253fail:
 254	erase_header(header);
 255	put_links(header);
 256fail_links:
 257	if (header->ctl_table == sysctl_mount_point)
 258		sysctl_clear_perm_empty_ctl_header(dir_h);
 259	header->parent = NULL;
 260	drop_sysctl_table(dir_h);
 261	return err;
 262}
 263
 264/* called under sysctl_lock */
 265static int use_table(struct ctl_table_header *p)
 266{
 
 
 267	if (unlikely(p->unregistering))
 268		return 0;
 269	p->used++;
 270	return 1;
 271}
 272
 273/* called under sysctl_lock */
 274static void unuse_table(struct ctl_table_header *p)
 275{
 
 
 276	if (!--p->used)
 277		if (unlikely(p->unregistering))
 278			complete(p->unregistering);
 279}
 280
 281static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
 282{
 283	proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
 284}
 285
 286/* called under sysctl_lock, will reacquire if has to wait */
 287static void start_unregistering(struct ctl_table_header *p)
 288{
 
 
 
 289	/*
 290	 * if p->used is 0, nobody will ever touch that entry again;
 291	 * we'll eliminate all paths to it before dropping sysctl_lock
 292	 */
 293	if (unlikely(p->used)) {
 294		struct completion wait;
 295		init_completion(&wait);
 296		p->unregistering = &wait;
 297		spin_unlock(&sysctl_lock);
 298		wait_for_completion(&wait);
 299	} else {
 300		/* anything non-NULL; we'll never dereference it */
 301		p->unregistering = ERR_PTR(-EINVAL);
 302		spin_unlock(&sysctl_lock);
 303	}
 304	/*
 305	 * Invalidate dentries for unregistered sysctls: namespaced sysctls
 306	 * can have duplicate names and contaminate dcache very badly.
 307	 */
 308	proc_sys_invalidate_dcache(p);
 309	/*
 310	 * do not remove from the list until nobody holds it; walking the
 311	 * list in do_sysctl() relies on that.
 312	 */
 313	spin_lock(&sysctl_lock);
 314	erase_header(p);
 315}
 316
 317static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
 318{
 319	BUG_ON(!head);
 320	spin_lock(&sysctl_lock);
 321	if (!use_table(head))
 322		head = ERR_PTR(-ENOENT);
 323	spin_unlock(&sysctl_lock);
 324	return head;
 325}
 326
 327static void sysctl_head_finish(struct ctl_table_header *head)
 328{
 329	if (!head)
 330		return;
 331	spin_lock(&sysctl_lock);
 332	unuse_table(head);
 333	spin_unlock(&sysctl_lock);
 334}
 335
 336static struct ctl_table_set *
 337lookup_header_set(struct ctl_table_root *root)
 338{
 339	struct ctl_table_set *set = &root->default_set;
 340	if (root->lookup)
 341		set = root->lookup(root);
 342	return set;
 343}
 344
 345static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
 346				      struct ctl_dir *dir,
 347				      const char *name, int namelen)
 348{
 349	struct ctl_table_header *head;
 350	struct ctl_table *entry;
 351
 352	spin_lock(&sysctl_lock);
 353	entry = find_entry(&head, dir, name, namelen);
 354	if (entry && use_table(head))
 355		*phead = head;
 356	else
 357		entry = NULL;
 358	spin_unlock(&sysctl_lock);
 359	return entry;
 360}
 361
 362static struct ctl_node *first_usable_entry(struct rb_node *node)
 363{
 364	struct ctl_node *ctl_node;
 365
 366	for (;node; node = rb_next(node)) {
 367		ctl_node = rb_entry(node, struct ctl_node, node);
 368		if (use_table(ctl_node->header))
 369			return ctl_node;
 370	}
 371	return NULL;
 372}
 373
 374static void first_entry(struct ctl_dir *dir,
 375	struct ctl_table_header **phead, struct ctl_table **pentry)
 376{
 377	struct ctl_table_header *head = NULL;
 378	struct ctl_table *entry = NULL;
 379	struct ctl_node *ctl_node;
 380
 381	spin_lock(&sysctl_lock);
 382	ctl_node = first_usable_entry(rb_first(&dir->root));
 383	spin_unlock(&sysctl_lock);
 384	if (ctl_node) {
 385		head = ctl_node->header;
 386		entry = &head->ctl_table[ctl_node - head->node];
 387	}
 388	*phead = head;
 389	*pentry = entry;
 390}
 391
 392static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
 393{
 394	struct ctl_table_header *head = *phead;
 395	struct ctl_table *entry = *pentry;
 396	struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
 397
 398	spin_lock(&sysctl_lock);
 399	unuse_table(head);
 400
 401	ctl_node = first_usable_entry(rb_next(&ctl_node->node));
 402	spin_unlock(&sysctl_lock);
 403	head = NULL;
 404	if (ctl_node) {
 405		head = ctl_node->header;
 406		entry = &head->ctl_table[ctl_node - head->node];
 407	}
 408	*phead = head;
 409	*pentry = entry;
 410}
 411
 412/*
 413 * sysctl_perm does NOT grant the superuser all rights automatically, because
 414 * some sysctl variables are readonly even to root.
 415 */
 416
 417static int test_perm(int mode, int op)
 418{
 419	if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
 420		mode >>= 6;
 421	else if (in_egroup_p(GLOBAL_ROOT_GID))
 422		mode >>= 3;
 423	if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
 424		return 0;
 425	return -EACCES;
 426}
 427
 428static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
 429{
 430	struct ctl_table_root *root = head->root;
 431	int mode;
 432
 433	if (root->permissions)
 434		mode = root->permissions(head, table);
 435	else
 436		mode = table->mode;
 437
 438	return test_perm(mode, op);
 439}
 440
 441static struct inode *proc_sys_make_inode(struct super_block *sb,
 442		struct ctl_table_header *head, struct ctl_table *table)
 443{
 444	struct ctl_table_root *root = head->root;
 445	struct inode *inode;
 446	struct proc_inode *ei;
 447
 448	inode = new_inode(sb);
 449	if (!inode)
 450		return ERR_PTR(-ENOMEM);
 451
 452	inode->i_ino = get_next_ino();
 453
 454	ei = PROC_I(inode);
 455
 456	spin_lock(&sysctl_lock);
 457	if (unlikely(head->unregistering)) {
 458		spin_unlock(&sysctl_lock);
 459		iput(inode);
 460		return ERR_PTR(-ENOENT);
 461	}
 462	ei->sysctl = head;
 463	ei->sysctl_entry = table;
 464	hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
 465	head->count++;
 466	spin_unlock(&sysctl_lock);
 467
 468	simple_inode_init_ts(inode);
 469	inode->i_mode = table->mode;
 470	if (!S_ISDIR(table->mode)) {
 471		inode->i_mode |= S_IFREG;
 472		inode->i_op = &proc_sys_inode_operations;
 473		inode->i_fop = &proc_sys_file_operations;
 474	} else {
 475		inode->i_mode |= S_IFDIR;
 476		inode->i_op = &proc_sys_dir_operations;
 477		inode->i_fop = &proc_sys_dir_file_operations;
 478		if (sysctl_is_perm_empty_ctl_header(head))
 479			make_empty_dir_inode(inode);
 480	}
 481
 
 
 482	if (root->set_ownership)
 483		root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
 484	else {
 485		inode->i_uid = GLOBAL_ROOT_UID;
 486		inode->i_gid = GLOBAL_ROOT_GID;
 487	}
 488
 489	return inode;
 490}
 491
 492void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
 493{
 494	spin_lock(&sysctl_lock);
 495	hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
 496	if (!--head->count)
 497		kfree_rcu(head, rcu);
 498	spin_unlock(&sysctl_lock);
 499}
 500
 501static struct ctl_table_header *grab_header(struct inode *inode)
 502{
 503	struct ctl_table_header *head = PROC_I(inode)->sysctl;
 504	if (!head)
 505		head = &sysctl_table_root.default_set.dir.header;
 506	return sysctl_head_grab(head);
 507}
 508
 509static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
 510					unsigned int flags)
 511{
 512	struct ctl_table_header *head = grab_header(dir);
 513	struct ctl_table_header *h = NULL;
 514	const struct qstr *name = &dentry->d_name;
 515	struct ctl_table *p;
 516	struct inode *inode;
 517	struct dentry *err = ERR_PTR(-ENOENT);
 518	struct ctl_dir *ctl_dir;
 519	int ret;
 520
 521	if (IS_ERR(head))
 522		return ERR_CAST(head);
 523
 524	ctl_dir = container_of(head, struct ctl_dir, header);
 525
 526	p = lookup_entry(&h, ctl_dir, name->name, name->len);
 527	if (!p)
 528		goto out;
 529
 530	if (S_ISLNK(p->mode)) {
 531		ret = sysctl_follow_link(&h, &p);
 532		err = ERR_PTR(ret);
 533		if (ret)
 534			goto out;
 535	}
 536
 537	d_set_d_op(dentry, &proc_sys_dentry_operations);
 538	inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
 539	err = d_splice_alias(inode, dentry);
 540
 541out:
 542	if (h)
 543		sysctl_head_finish(h);
 544	sysctl_head_finish(head);
 545	return err;
 546}
 547
 548static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
 549		int write)
 550{
 551	struct inode *inode = file_inode(iocb->ki_filp);
 552	struct ctl_table_header *head = grab_header(inode);
 553	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 554	size_t count = iov_iter_count(iter);
 555	char *kbuf;
 556	ssize_t error;
 557
 558	if (IS_ERR(head))
 559		return PTR_ERR(head);
 560
 561	/*
 562	 * At this point we know that the sysctl was not unregistered
 563	 * and won't be until we finish.
 564	 */
 565	error = -EPERM;
 566	if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
 567		goto out;
 568
 569	/* if that can happen at all, it should be -EINVAL, not -EISDIR */
 570	error = -EINVAL;
 571	if (!table->proc_handler)
 572		goto out;
 573
 574	/* don't even try if the size is too large */
 575	error = -ENOMEM;
 576	if (count >= KMALLOC_MAX_SIZE)
 577		goto out;
 578	kbuf = kvzalloc(count + 1, GFP_KERNEL);
 579	if (!kbuf)
 580		goto out;
 581
 582	if (write) {
 583		error = -EFAULT;
 584		if (!copy_from_iter_full(kbuf, count, iter))
 585			goto out_free_buf;
 586		kbuf[count] = '\0';
 587	}
 588
 589	error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
 590					   &iocb->ki_pos);
 591	if (error)
 592		goto out_free_buf;
 593
 594	/* careful: calling conventions are nasty here */
 595	error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
 596	if (error)
 597		goto out_free_buf;
 598
 599	if (!write) {
 600		error = -EFAULT;
 601		if (copy_to_iter(kbuf, count, iter) < count)
 602			goto out_free_buf;
 603	}
 604
 605	error = count;
 606out_free_buf:
 607	kvfree(kbuf);
 608out:
 609	sysctl_head_finish(head);
 610
 611	return error;
 612}
 613
 614static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
 615{
 616	return proc_sys_call_handler(iocb, iter, 0);
 617}
 618
 619static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
 620{
 621	return proc_sys_call_handler(iocb, iter, 1);
 622}
 623
 624static int proc_sys_open(struct inode *inode, struct file *filp)
 625{
 626	struct ctl_table_header *head = grab_header(inode);
 627	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 628
 629	/* sysctl was unregistered */
 630	if (IS_ERR(head))
 631		return PTR_ERR(head);
 632
 633	if (table->poll)
 634		filp->private_data = proc_sys_poll_event(table->poll);
 635
 636	sysctl_head_finish(head);
 637
 638	return 0;
 639}
 640
 641static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
 642{
 643	struct inode *inode = file_inode(filp);
 644	struct ctl_table_header *head = grab_header(inode);
 645	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 646	__poll_t ret = DEFAULT_POLLMASK;
 647	unsigned long event;
 648
 649	/* sysctl was unregistered */
 650	if (IS_ERR(head))
 651		return EPOLLERR | EPOLLHUP;
 652
 653	if (!table->proc_handler)
 654		goto out;
 655
 656	if (!table->poll)
 657		goto out;
 658
 659	event = (unsigned long)filp->private_data;
 660	poll_wait(filp, &table->poll->wait, wait);
 661
 662	if (event != atomic_read(&table->poll->event)) {
 663		filp->private_data = proc_sys_poll_event(table->poll);
 664		ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
 665	}
 666
 667out:
 668	sysctl_head_finish(head);
 669
 670	return ret;
 671}
 672
 673static bool proc_sys_fill_cache(struct file *file,
 674				struct dir_context *ctx,
 675				struct ctl_table_header *head,
 676				struct ctl_table *table)
 677{
 678	struct dentry *child, *dir = file->f_path.dentry;
 679	struct inode *inode;
 680	struct qstr qname;
 681	ino_t ino = 0;
 682	unsigned type = DT_UNKNOWN;
 683
 684	qname.name = table->procname;
 685	qname.len  = strlen(table->procname);
 686	qname.hash = full_name_hash(dir, qname.name, qname.len);
 687
 688	child = d_lookup(dir, &qname);
 689	if (!child) {
 690		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
 691		child = d_alloc_parallel(dir, &qname, &wq);
 692		if (IS_ERR(child))
 693			return false;
 694		if (d_in_lookup(child)) {
 695			struct dentry *res;
 696			d_set_d_op(child, &proc_sys_dentry_operations);
 697			inode = proc_sys_make_inode(dir->d_sb, head, table);
 698			res = d_splice_alias(inode, child);
 699			d_lookup_done(child);
 700			if (unlikely(res)) {
 701				if (IS_ERR(res)) {
 702					dput(child);
 
 703					return false;
 704				}
 705				dput(child);
 706				child = res;
 707			}
 708		}
 709	}
 710	inode = d_inode(child);
 711	ino  = inode->i_ino;
 712	type = inode->i_mode >> 12;
 713	dput(child);
 714	return dir_emit(ctx, qname.name, qname.len, ino, type);
 715}
 716
 717static bool proc_sys_link_fill_cache(struct file *file,
 718				    struct dir_context *ctx,
 719				    struct ctl_table_header *head,
 720				    struct ctl_table *table)
 721{
 722	bool ret = true;
 723
 724	head = sysctl_head_grab(head);
 725	if (IS_ERR(head))
 726		return false;
 727
 728	/* It is not an error if we can not follow the link ignore it */
 729	if (sysctl_follow_link(&head, &table))
 730		goto out;
 731
 732	ret = proc_sys_fill_cache(file, ctx, head, table);
 733out:
 734	sysctl_head_finish(head);
 735	return ret;
 736}
 737
 738static int scan(struct ctl_table_header *head, struct ctl_table *table,
 739		unsigned long *pos, struct file *file,
 740		struct dir_context *ctx)
 741{
 742	bool res;
 743
 744	if ((*pos)++ < ctx->pos)
 745		return true;
 746
 747	if (unlikely(S_ISLNK(table->mode)))
 748		res = proc_sys_link_fill_cache(file, ctx, head, table);
 749	else
 750		res = proc_sys_fill_cache(file, ctx, head, table);
 751
 752	if (res)
 753		ctx->pos = *pos;
 754
 755	return res;
 756}
 757
 758static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
 759{
 760	struct ctl_table_header *head = grab_header(file_inode(file));
 761	struct ctl_table_header *h = NULL;
 762	struct ctl_table *entry;
 763	struct ctl_dir *ctl_dir;
 764	unsigned long pos;
 765
 766	if (IS_ERR(head))
 767		return PTR_ERR(head);
 768
 769	ctl_dir = container_of(head, struct ctl_dir, header);
 770
 771	if (!dir_emit_dots(file, ctx))
 772		goto out;
 773
 774	pos = 2;
 775
 776	for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
 777		if (!scan(h, entry, &pos, file, ctx)) {
 778			sysctl_head_finish(h);
 779			break;
 780		}
 781	}
 782out:
 783	sysctl_head_finish(head);
 784	return 0;
 785}
 786
 787static int proc_sys_permission(struct mnt_idmap *idmap,
 788			       struct inode *inode, int mask)
 789{
 790	/*
 791	 * sysctl entries that are not writeable,
 792	 * are _NOT_ writeable, capabilities or not.
 793	 */
 794	struct ctl_table_header *head;
 795	struct ctl_table *table;
 796	int error;
 797
 798	/* Executable files are not allowed under /proc/sys/ */
 799	if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
 800		return -EACCES;
 801
 802	head = grab_header(inode);
 803	if (IS_ERR(head))
 804		return PTR_ERR(head);
 805
 806	table = PROC_I(inode)->sysctl_entry;
 807	if (!table) /* global root - r-xr-xr-x */
 808		error = mask & MAY_WRITE ? -EACCES : 0;
 809	else /* Use the permissions on the sysctl table entry */
 810		error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
 811
 812	sysctl_head_finish(head);
 813	return error;
 814}
 815
 816static int proc_sys_setattr(struct mnt_idmap *idmap,
 817			    struct dentry *dentry, struct iattr *attr)
 818{
 819	struct inode *inode = d_inode(dentry);
 820	int error;
 821
 822	if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
 823		return -EPERM;
 824
 825	error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
 826	if (error)
 827		return error;
 828
 829	setattr_copy(&nop_mnt_idmap, inode, attr);
 830	return 0;
 831}
 832
 833static int proc_sys_getattr(struct mnt_idmap *idmap,
 834			    const struct path *path, struct kstat *stat,
 835			    u32 request_mask, unsigned int query_flags)
 836{
 837	struct inode *inode = d_inode(path->dentry);
 838	struct ctl_table_header *head = grab_header(inode);
 839	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 840
 841	if (IS_ERR(head))
 842		return PTR_ERR(head);
 843
 844	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
 845	if (table)
 846		stat->mode = (stat->mode & S_IFMT) | table->mode;
 847
 848	sysctl_head_finish(head);
 849	return 0;
 850}
 851
 852static const struct file_operations proc_sys_file_operations = {
 853	.open		= proc_sys_open,
 854	.poll		= proc_sys_poll,
 855	.read_iter	= proc_sys_read,
 856	.write_iter	= proc_sys_write,
 857	.splice_read	= copy_splice_read,
 858	.splice_write	= iter_file_splice_write,
 859	.llseek		= default_llseek,
 860};
 861
 862static const struct file_operations proc_sys_dir_file_operations = {
 863	.read		= generic_read_dir,
 864	.iterate_shared	= proc_sys_readdir,
 865	.llseek		= generic_file_llseek,
 866};
 867
 868static const struct inode_operations proc_sys_inode_operations = {
 869	.permission	= proc_sys_permission,
 870	.setattr	= proc_sys_setattr,
 871	.getattr	= proc_sys_getattr,
 872};
 873
 874static const struct inode_operations proc_sys_dir_operations = {
 875	.lookup		= proc_sys_lookup,
 876	.permission	= proc_sys_permission,
 877	.setattr	= proc_sys_setattr,
 878	.getattr	= proc_sys_getattr,
 879};
 880
 881static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
 882{
 883	if (flags & LOOKUP_RCU)
 884		return -ECHILD;
 885	return !PROC_I(d_inode(dentry))->sysctl->unregistering;
 886}
 887
 888static int proc_sys_delete(const struct dentry *dentry)
 889{
 890	return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
 891}
 892
 893static int sysctl_is_seen(struct ctl_table_header *p)
 894{
 895	struct ctl_table_set *set = p->set;
 896	int res;
 897	spin_lock(&sysctl_lock);
 898	if (p->unregistering)
 899		res = 0;
 900	else if (!set->is_seen)
 901		res = 1;
 902	else
 903		res = set->is_seen(set);
 904	spin_unlock(&sysctl_lock);
 905	return res;
 906}
 907
 908static int proc_sys_compare(const struct dentry *dentry,
 909		unsigned int len, const char *str, const struct qstr *name)
 910{
 911	struct ctl_table_header *head;
 912	struct inode *inode;
 913
 914	/* Although proc doesn't have negative dentries, rcu-walk means
 915	 * that inode here can be NULL */
 916	/* AV: can it, indeed? */
 917	inode = d_inode_rcu(dentry);
 918	if (!inode)
 919		return 1;
 920	if (name->len != len)
 921		return 1;
 922	if (memcmp(name->name, str, len))
 923		return 1;
 924	head = rcu_dereference(PROC_I(inode)->sysctl);
 925	return !head || !sysctl_is_seen(head);
 926}
 927
 928static const struct dentry_operations proc_sys_dentry_operations = {
 929	.d_revalidate	= proc_sys_revalidate,
 930	.d_delete	= proc_sys_delete,
 931	.d_compare	= proc_sys_compare,
 932};
 933
 934static struct ctl_dir *find_subdir(struct ctl_dir *dir,
 935				   const char *name, int namelen)
 936{
 937	struct ctl_table_header *head;
 938	struct ctl_table *entry;
 939
 940	entry = find_entry(&head, dir, name, namelen);
 941	if (!entry)
 942		return ERR_PTR(-ENOENT);
 943	if (!S_ISDIR(entry->mode))
 944		return ERR_PTR(-ENOTDIR);
 945	return container_of(head, struct ctl_dir, header);
 946}
 947
 948static struct ctl_dir *new_dir(struct ctl_table_set *set,
 949			       const char *name, int namelen)
 950{
 951	struct ctl_table *table;
 952	struct ctl_dir *new;
 953	struct ctl_node *node;
 954	char *new_name;
 955
 956	new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
 957		      sizeof(struct ctl_table)*2 +  namelen + 1,
 958		      GFP_KERNEL);
 959	if (!new)
 960		return NULL;
 961
 962	node = (struct ctl_node *)(new + 1);
 963	table = (struct ctl_table *)(node + 1);
 964	new_name = (char *)(table + 2);
 965	memcpy(new_name, name, namelen);
 966	table[0].procname = new_name;
 967	table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
 968	init_header(&new->header, set->dir.header.root, set, node, table, 1);
 969
 970	return new;
 971}
 972
 973/**
 974 * get_subdir - find or create a subdir with the specified name.
 975 * @dir:  Directory to create the subdirectory in
 976 * @name: The name of the subdirectory to find or create
 977 * @namelen: The length of name
 978 *
 979 * Takes a directory with an elevated reference count so we know that
 980 * if we drop the lock the directory will not go away.  Upon success
 981 * the reference is moved from @dir to the returned subdirectory.
 982 * Upon error an error code is returned and the reference on @dir is
 983 * simply dropped.
 984 */
 985static struct ctl_dir *get_subdir(struct ctl_dir *dir,
 986				  const char *name, int namelen)
 987{
 988	struct ctl_table_set *set = dir->header.set;
 989	struct ctl_dir *subdir, *new = NULL;
 990	int err;
 991
 992	spin_lock(&sysctl_lock);
 993	subdir = find_subdir(dir, name, namelen);
 994	if (!IS_ERR(subdir))
 995		goto found;
 996	if (PTR_ERR(subdir) != -ENOENT)
 997		goto failed;
 998
 999	spin_unlock(&sysctl_lock);
1000	new = new_dir(set, name, namelen);
1001	spin_lock(&sysctl_lock);
1002	subdir = ERR_PTR(-ENOMEM);
1003	if (!new)
1004		goto failed;
1005
1006	/* Was the subdir added while we dropped the lock? */
1007	subdir = find_subdir(dir, name, namelen);
1008	if (!IS_ERR(subdir))
1009		goto found;
1010	if (PTR_ERR(subdir) != -ENOENT)
1011		goto failed;
1012
1013	/* Nope.  Use the our freshly made directory entry. */
1014	err = insert_header(dir, &new->header);
1015	subdir = ERR_PTR(err);
1016	if (err)
1017		goto failed;
1018	subdir = new;
1019found:
1020	subdir->header.nreg++;
1021failed:
1022	if (IS_ERR(subdir)) {
1023		pr_err("sysctl could not get directory: ");
1024		sysctl_print_dir(dir);
1025		pr_cont("%*.*s %ld\n", namelen, namelen, name,
1026			PTR_ERR(subdir));
1027	}
1028	drop_sysctl_table(&dir->header);
1029	if (new)
1030		drop_sysctl_table(&new->header);
1031	spin_unlock(&sysctl_lock);
1032	return subdir;
1033}
1034
1035static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1036{
1037	struct ctl_dir *parent;
1038	const char *procname;
1039	if (!dir->header.parent)
1040		return &set->dir;
1041	parent = xlate_dir(set, dir->header.parent);
1042	if (IS_ERR(parent))
1043		return parent;
1044	procname = dir->header.ctl_table[0].procname;
1045	return find_subdir(parent, procname, strlen(procname));
1046}
1047
1048static int sysctl_follow_link(struct ctl_table_header **phead,
1049	struct ctl_table **pentry)
1050{
1051	struct ctl_table_header *head;
 
1052	struct ctl_table_root *root;
1053	struct ctl_table_set *set;
1054	struct ctl_table *entry;
1055	struct ctl_dir *dir;
1056	int ret;
1057
1058	spin_lock(&sysctl_lock);
1059	root = (*pentry)->data;
1060	set = lookup_header_set(root);
1061	dir = xlate_dir(set, (*phead)->parent);
1062	if (IS_ERR(dir))
1063		ret = PTR_ERR(dir);
1064	else {
1065		const char *procname = (*pentry)->procname;
1066		head = NULL;
1067		entry = find_entry(&head, dir, procname, strlen(procname));
1068		ret = -ENOENT;
1069		if (entry && use_table(head)) {
1070			unuse_table(*phead);
1071			*phead = head;
1072			*pentry = entry;
1073			ret = 0;
1074		}
1075	}
1076
1077	spin_unlock(&sysctl_lock);
1078	return ret;
1079}
1080
1081static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1082{
1083	struct va_format vaf;
1084	va_list args;
1085
1086	va_start(args, fmt);
1087	vaf.fmt = fmt;
1088	vaf.va = &args;
1089
1090	pr_err("sysctl table check failed: %s/%s %pV\n",
1091	       path, table->procname, &vaf);
1092
1093	va_end(args);
1094	return -EINVAL;
1095}
1096
1097static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1098{
 
1099	int err = 0;
1100
1101	if ((table->proc_handler == proc_douintvec) ||
1102	    (table->proc_handler == proc_douintvec_minmax)) {
1103		if (table->maxlen != sizeof(unsigned int))
1104			err |= sysctl_err(path, table, "array not allowed");
1105	}
1106
1107	if (table->proc_handler == proc_dou8vec_minmax) {
1108		if (table->maxlen != sizeof(u8))
1109			err |= sysctl_err(path, table, "array not allowed");
 
 
 
 
 
 
 
 
 
 
 
 
 
1110	}
1111
1112	if (table->proc_handler == proc_dobool) {
1113		if (table->maxlen != sizeof(bool))
1114			err |= sysctl_err(path, table, "array not allowed");
1115	}
1116
1117	return err;
1118}
1119
1120static int sysctl_check_table(const char *path, struct ctl_table_header *header)
1121{
1122	struct ctl_table *entry;
1123	int err = 0;
1124	list_for_each_table_entry(entry, header) {
 
 
1125		if ((entry->proc_handler == proc_dostring) ||
1126		    (entry->proc_handler == proc_dobool) ||
1127		    (entry->proc_handler == proc_dointvec) ||
1128		    (entry->proc_handler == proc_douintvec) ||
1129		    (entry->proc_handler == proc_douintvec_minmax) ||
1130		    (entry->proc_handler == proc_dointvec_minmax) ||
1131		    (entry->proc_handler == proc_dou8vec_minmax) ||
1132		    (entry->proc_handler == proc_dointvec_jiffies) ||
1133		    (entry->proc_handler == proc_dointvec_userhz_jiffies) ||
1134		    (entry->proc_handler == proc_dointvec_ms_jiffies) ||
1135		    (entry->proc_handler == proc_doulongvec_minmax) ||
1136		    (entry->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1137			if (!entry->data)
1138				err |= sysctl_err(path, entry, "No data");
1139			if (!entry->maxlen)
1140				err |= sysctl_err(path, entry, "No maxlen");
1141			else
1142				err |= sysctl_check_table_array(path, entry);
1143		}
1144		if (!entry->proc_handler)
1145			err |= sysctl_err(path, entry, "No proc_handler");
1146
1147		if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode)
1148			err |= sysctl_err(path, entry, "bogus .mode 0%o",
1149				entry->mode);
1150	}
1151	return err;
1152}
1153
1154static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table_header *head)
1155{
1156	struct ctl_table *link_table, *entry, *link;
1157	struct ctl_table_header *links;
 
1158	struct ctl_node *node;
1159	char *link_name;
1160	int nr_entries, name_bytes;
1161
1162	name_bytes = 0;
1163	nr_entries = 0;
1164	list_for_each_table_entry(entry, head) {
1165		nr_entries++;
1166		name_bytes += strlen(entry->procname) + 1;
1167	}
1168
1169	links = kzalloc(sizeof(struct ctl_table_header) +
1170			sizeof(struct ctl_node)*nr_entries +
1171			sizeof(struct ctl_table)*(nr_entries + 1) +
1172			name_bytes,
1173			GFP_KERNEL);
1174
1175	if (!links)
1176		return NULL;
1177
1178	node = (struct ctl_node *)(links + 1);
1179	link_table = (struct ctl_table *)(node + nr_entries);
1180	link_name = (char *)&link_table[nr_entries + 1];
1181	link = link_table;
1182
1183	list_for_each_table_entry(entry, head) {
1184		int len = strlen(entry->procname) + 1;
1185		memcpy(link_name, entry->procname, len);
1186		link->procname = link_name;
1187		link->mode = S_IFLNK|S_IRWXUGO;
1188		link->data = head->root;
1189		link_name += len;
1190		link++;
1191	}
1192	init_header(links, dir->header.root, dir->header.set, node, link_table,
1193		    head->ctl_table_size);
1194	links->nreg = nr_entries;
1195
1196	return links;
1197}
1198
1199static bool get_links(struct ctl_dir *dir,
1200		      struct ctl_table_header *header,
1201		      struct ctl_table_root *link_root)
1202{
1203	struct ctl_table_header *tmp_head;
1204	struct ctl_table *entry, *link;
1205
1206	if (header->ctl_table_size == 0 ||
1207	    sysctl_is_perm_empty_ctl_table(header->ctl_table))
1208		return true;
1209
1210	/* Are there links available for every entry in table? */
1211	list_for_each_table_entry(entry, header) {
1212		const char *procname = entry->procname;
1213		link = find_entry(&tmp_head, dir, procname, strlen(procname));
1214		if (!link)
1215			return false;
1216		if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1217			continue;
1218		if (S_ISLNK(link->mode) && (link->data == link_root))
1219			continue;
1220		return false;
1221	}
1222
1223	/* The checks passed.  Increase the registration count on the links */
1224	list_for_each_table_entry(entry, header) {
1225		const char *procname = entry->procname;
1226		link = find_entry(&tmp_head, dir, procname, strlen(procname));
1227		tmp_head->nreg++;
1228	}
1229	return true;
1230}
1231
1232static int insert_links(struct ctl_table_header *head)
1233{
1234	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1235	struct ctl_dir *core_parent;
1236	struct ctl_table_header *links;
1237	int err;
1238
1239	if (head->set == root_set)
1240		return 0;
1241
1242	core_parent = xlate_dir(root_set, head->parent);
1243	if (IS_ERR(core_parent))
1244		return 0;
1245
1246	if (get_links(core_parent, head, head->root))
1247		return 0;
1248
1249	core_parent->header.nreg++;
1250	spin_unlock(&sysctl_lock);
1251
1252	links = new_links(core_parent, head);
1253
1254	spin_lock(&sysctl_lock);
1255	err = -ENOMEM;
1256	if (!links)
1257		goto out;
1258
1259	err = 0;
1260	if (get_links(core_parent, head, head->root)) {
1261		kfree(links);
1262		goto out;
1263	}
1264
1265	err = insert_header(core_parent, links);
1266	if (err)
1267		kfree(links);
1268out:
1269	drop_sysctl_table(&core_parent->header);
1270	return err;
1271}
1272
1273/* Find the directory for the ctl_table. If one is not found create it. */
1274static struct ctl_dir *sysctl_mkdir_p(struct ctl_dir *dir, const char *path)
1275{
1276	const char *name, *nextname;
1277
1278	for (name = path; name; name = nextname) {
1279		int namelen;
1280		nextname = strchr(name, '/');
1281		if (nextname) {
1282			namelen = nextname - name;
1283			nextname++;
1284		} else {
1285			namelen = strlen(name);
1286		}
1287		if (namelen == 0)
1288			continue;
1289
1290		/*
1291		 * namelen ensures if name is "foo/bar/yay" only foo is
1292		 * registered first. We traverse as if using mkdir -p and
1293		 * return a ctl_dir for the last directory entry.
1294		 */
1295		dir = get_subdir(dir, name, namelen);
1296		if (IS_ERR(dir))
1297			break;
1298	}
1299	return dir;
1300}
1301
1302/**
1303 * __register_sysctl_table - register a leaf sysctl table
1304 * @set: Sysctl tree to register on
1305 * @path: The path to the directory the sysctl table is in.
1306 * @table: the top-level table structure without any child. This table
1307 * 	 should not be free'd after registration. So it should not be
1308 * 	 used on stack. It can either be a global or dynamically allocated
1309 * 	 by the caller and free'd later after sysctl unregistration.
 
1310 * @table_size : The number of elements in table
1311 *
1312 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1313 * array. A completely 0 filled entry terminates the table.
1314 *
1315 * The members of the &struct ctl_table structure are used as follows:
1316 *
1317 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1318 *            enter a sysctl file
1319 *
1320 * data - a pointer to data for use by proc_handler
1321 *
1322 * maxlen - the maximum size in bytes of the data
1323 *
1324 * mode - the file permissions for the /proc/sys file
1325 *
1326 * child - must be %NULL.
1327 *
1328 * proc_handler - the text handler routine (described below)
1329 *
1330 * extra1, extra2 - extra pointers usable by the proc handler routines
1331 * XXX: we should eventually modify these to use long min / max [0]
1332 * [0] https://lkml.kernel.org/87zgpte9o4.fsf@email.froward.int.ebiederm.org
1333 *
1334 * Leaf nodes in the sysctl tree will be represented by a single file
1335 * under /proc; non-leaf nodes (where child is not NULL) are not allowed,
1336 * sysctl_check_table() verifies this.
1337 *
1338 * There must be a proc_handler routine for any terminal nodes.
1339 * Several default handlers are available to cover common cases -
1340 *
1341 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1342 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1343 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1344 *
1345 * It is the handler's job to read the input buffer from user memory
1346 * and process it. The handler should return 0 on success.
1347 *
1348 * This routine returns %NULL on a failure to register, and a pointer
1349 * to the table header on success.
1350 */
1351struct ctl_table_header *__register_sysctl_table(
1352	struct ctl_table_set *set,
1353	const char *path, struct ctl_table *table, size_t table_size)
1354{
1355	struct ctl_table_root *root = set->dir.header.root;
1356	struct ctl_table_header *header;
1357	struct ctl_dir *dir;
1358	struct ctl_node *node;
1359
1360	header = kzalloc(sizeof(struct ctl_table_header) +
1361			 sizeof(struct ctl_node)*table_size, GFP_KERNEL_ACCOUNT);
1362	if (!header)
1363		return NULL;
1364
1365	node = (struct ctl_node *)(header + 1);
1366	init_header(header, root, set, node, table, table_size);
1367	if (sysctl_check_table(path, header))
1368		goto fail;
1369
1370	spin_lock(&sysctl_lock);
1371	dir = &set->dir;
1372	/* Reference moved down the directory tree get_subdir */
1373	dir->header.nreg++;
1374	spin_unlock(&sysctl_lock);
1375
1376	dir = sysctl_mkdir_p(dir, path);
1377	if (IS_ERR(dir))
1378		goto fail;
1379	spin_lock(&sysctl_lock);
1380	if (insert_header(dir, header))
1381		goto fail_put_dir_locked;
1382
1383	drop_sysctl_table(&dir->header);
1384	spin_unlock(&sysctl_lock);
1385
1386	return header;
1387
1388fail_put_dir_locked:
1389	drop_sysctl_table(&dir->header);
1390	spin_unlock(&sysctl_lock);
1391fail:
1392	kfree(header);
1393	return NULL;
1394}
1395
1396/**
1397 * register_sysctl_sz - register a sysctl table
1398 * @path: The path to the directory the sysctl table is in. If the path
1399 * 	doesn't exist we will create it for you.
1400 * @table: the table structure. The calller must ensure the life of the @table
1401 * 	will be kept during the lifetime use of the syctl. It must not be freed
1402 * 	until unregister_sysctl_table() is called with the given returned table
1403 * 	with this registration. If your code is non modular then you don't need
1404 * 	to call unregister_sysctl_table() and can instead use something like
1405 * 	register_sysctl_init() which does not care for the result of the syctl
1406 * 	registration.
1407 * @table_size: The number of elements in table.
1408 *
1409 * Register a sysctl table. @table should be a filled in ctl_table
1410 * array. A completely 0 filled entry terminates the table.
1411 *
1412 * See __register_sysctl_table for more details.
1413 */
1414struct ctl_table_header *register_sysctl_sz(const char *path, struct ctl_table *table,
1415					    size_t table_size)
1416{
1417	return __register_sysctl_table(&sysctl_table_root.default_set,
1418					path, table, table_size);
1419}
1420EXPORT_SYMBOL(register_sysctl_sz);
1421
1422/**
1423 * __register_sysctl_init() - register sysctl table to path
1424 * @path: path name for sysctl base. If that path doesn't exist we will create
1425 * 	it for you.
1426 * @table: This is the sysctl table that needs to be registered to the path.
1427 * 	The caller must ensure the life of the @table will be kept during the
1428 * 	lifetime use of the sysctl.
1429 * @table_name: The name of sysctl table, only used for log printing when
1430 *              registration fails
1431 * @table_size: The number of elements in table
1432 *
1433 * The sysctl interface is used by userspace to query or modify at runtime
1434 * a predefined value set on a variable. These variables however have default
1435 * values pre-set. Code which depends on these variables will always work even
1436 * if register_sysctl() fails. If register_sysctl() fails you'd just loose the
1437 * ability to query or modify the sysctls dynamically at run time. Chances of
1438 * register_sysctl() failing on init are extremely low, and so for both reasons
1439 * this function does not return any error as it is used by initialization code.
1440 *
1441 * Context: if your base directory does not exist it will be created for you.
1442 */
1443void __init __register_sysctl_init(const char *path, struct ctl_table *table,
1444				 const char *table_name, size_t table_size)
1445{
1446	struct ctl_table_header *hdr = register_sysctl_sz(path, table, table_size);
1447
1448	if (unlikely(!hdr)) {
1449		pr_err("failed when register_sysctl_sz %s to %s\n", table_name, path);
1450		return;
1451	}
1452	kmemleak_not_leak(hdr);
1453}
1454
1455static void put_links(struct ctl_table_header *header)
1456{
1457	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1458	struct ctl_table_root *root = header->root;
1459	struct ctl_dir *parent = header->parent;
1460	struct ctl_dir *core_parent;
1461	struct ctl_table *entry;
1462
1463	if (header->set == root_set)
1464		return;
1465
1466	core_parent = xlate_dir(root_set, parent);
1467	if (IS_ERR(core_parent))
1468		return;
1469
1470	list_for_each_table_entry(entry, header) {
1471		struct ctl_table_header *link_head;
1472		struct ctl_table *link;
1473		const char *name = entry->procname;
1474
1475		link = find_entry(&link_head, core_parent, name, strlen(name));
1476		if (link &&
1477		    ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1478		     (S_ISLNK(link->mode) && (link->data == root)))) {
1479			drop_sysctl_table(link_head);
1480		}
1481		else {
1482			pr_err("sysctl link missing during unregister: ");
1483			sysctl_print_dir(parent);
1484			pr_cont("%s\n", name);
1485		}
1486	}
1487}
1488
1489static void drop_sysctl_table(struct ctl_table_header *header)
1490{
1491	struct ctl_dir *parent = header->parent;
1492
1493	if (--header->nreg)
1494		return;
1495
1496	if (parent) {
1497		put_links(header);
1498		start_unregistering(header);
1499	}
1500
1501	if (!--header->count)
1502		kfree_rcu(header, rcu);
1503
1504	if (parent)
1505		drop_sysctl_table(&parent->header);
1506}
1507
1508/**
1509 * unregister_sysctl_table - unregister a sysctl table hierarchy
1510 * @header: the header returned from register_sysctl or __register_sysctl_table
1511 *
1512 * Unregisters the sysctl table and all children. proc entries may not
1513 * actually be removed until they are no longer used by anyone.
1514 */
1515void unregister_sysctl_table(struct ctl_table_header * header)
1516{
1517	might_sleep();
1518
1519	if (header == NULL)
1520		return;
1521
1522	spin_lock(&sysctl_lock);
1523	drop_sysctl_table(header);
1524	spin_unlock(&sysctl_lock);
1525}
1526EXPORT_SYMBOL(unregister_sysctl_table);
1527
1528void setup_sysctl_set(struct ctl_table_set *set,
1529	struct ctl_table_root *root,
1530	int (*is_seen)(struct ctl_table_set *))
1531{
1532	memset(set, 0, sizeof(*set));
1533	set->is_seen = is_seen;
1534	init_header(&set->dir.header, root, set, NULL, root_table, 1);
1535}
1536
1537void retire_sysctl_set(struct ctl_table_set *set)
1538{
1539	WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1540}
1541
1542int __init proc_sys_init(void)
1543{
1544	struct proc_dir_entry *proc_sys_root;
1545
1546	proc_sys_root = proc_mkdir("sys", NULL);
1547	proc_sys_root->proc_iops = &proc_sys_dir_operations;
1548	proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1549	proc_sys_root->nlink = 0;
1550
1551	return sysctl_init_bases();
1552}
1553
1554struct sysctl_alias {
1555	const char *kernel_param;
1556	const char *sysctl_param;
1557};
1558
1559/*
1560 * Historically some settings had both sysctl and a command line parameter.
1561 * With the generic sysctl. parameter support, we can handle them at a single
1562 * place and only keep the historical name for compatibility. This is not meant
1563 * to add brand new aliases. When adding existing aliases, consider whether
1564 * the possibly different moment of changing the value (e.g. from early_param
1565 * to the moment do_sysctl_args() is called) is an issue for the specific
1566 * parameter.
1567 */
1568static const struct sysctl_alias sysctl_aliases[] = {
1569	{"hardlockup_all_cpu_backtrace",	"kernel.hardlockup_all_cpu_backtrace" },
1570	{"hung_task_panic",			"kernel.hung_task_panic" },
1571	{"numa_zonelist_order",			"vm.numa_zonelist_order" },
1572	{"softlockup_all_cpu_backtrace",	"kernel.softlockup_all_cpu_backtrace" },
1573	{ }
1574};
1575
1576static const char *sysctl_find_alias(char *param)
1577{
1578	const struct sysctl_alias *alias;
1579
1580	for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1581		if (strcmp(alias->kernel_param, param) == 0)
1582			return alias->sysctl_param;
1583	}
1584
1585	return NULL;
1586}
1587
1588bool sysctl_is_alias(char *param)
1589{
1590	const char *alias = sysctl_find_alias(param);
1591
1592	return alias != NULL;
1593}
1594
1595/* Set sysctl value passed on kernel command line. */
1596static int process_sysctl_arg(char *param, char *val,
1597			       const char *unused, void *arg)
1598{
1599	char *path;
1600	struct vfsmount **proc_mnt = arg;
1601	struct file_system_type *proc_fs_type;
1602	struct file *file;
1603	int len;
1604	int err;
1605	loff_t pos = 0;
1606	ssize_t wret;
1607
1608	if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1609		param += sizeof("sysctl") - 1;
1610
1611		if (param[0] != '/' && param[0] != '.')
1612			return 0;
1613
1614		param++;
1615	} else {
1616		param = (char *) sysctl_find_alias(param);
1617		if (!param)
1618			return 0;
1619	}
1620
1621	if (!val)
1622		return -EINVAL;
1623	len = strlen(val);
1624	if (len == 0)
1625		return -EINVAL;
1626
1627	/*
1628	 * To set sysctl options, we use a temporary mount of proc, look up the
1629	 * respective sys/ file and write to it. To avoid mounting it when no
1630	 * options were given, we mount it only when the first sysctl option is
1631	 * found. Why not a persistent mount? There are problems with a
1632	 * persistent mount of proc in that it forces userspace not to use any
1633	 * proc mount options.
1634	 */
1635	if (!*proc_mnt) {
1636		proc_fs_type = get_fs_type("proc");
1637		if (!proc_fs_type) {
1638			pr_err("Failed to find procfs to set sysctl from command line\n");
1639			return 0;
1640		}
1641		*proc_mnt = kern_mount(proc_fs_type);
1642		put_filesystem(proc_fs_type);
1643		if (IS_ERR(*proc_mnt)) {
1644			pr_err("Failed to mount procfs to set sysctl from command line\n");
1645			return 0;
1646		}
1647	}
1648
1649	path = kasprintf(GFP_KERNEL, "sys/%s", param);
1650	if (!path)
1651		panic("%s: Failed to allocate path for %s\n", __func__, param);
1652	strreplace(path, '.', '/');
1653
1654	file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1655	if (IS_ERR(file)) {
1656		err = PTR_ERR(file);
1657		if (err == -ENOENT)
1658			pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1659				param, val);
1660		else if (err == -EACCES)
1661			pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1662				param, val);
1663		else
1664			pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1665				file, param, val);
1666		goto out;
1667	}
1668	wret = kernel_write(file, val, len, &pos);
1669	if (wret < 0) {
1670		err = wret;
1671		if (err == -EINVAL)
1672			pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1673				param, val);
1674		else
1675			pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1676				ERR_PTR(err), param, val);
1677	} else if (wret != len) {
1678		pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1679			wret, len, path, param, val);
1680	}
1681
1682	err = filp_close(file, NULL);
1683	if (err)
1684		pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1685			ERR_PTR(err), param, val);
1686out:
1687	kfree(path);
1688	return 0;
1689}
1690
1691void do_sysctl_args(void)
1692{
1693	char *command_line;
1694	struct vfsmount *proc_mnt = NULL;
1695
1696	command_line = kstrdup(saved_command_line, GFP_KERNEL);
1697	if (!command_line)
1698		panic("%s: Failed to allocate copy of command line\n", __func__);
1699
1700	parse_args("Setting sysctl args", command_line,
1701		   NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1702
1703	if (proc_mnt)
1704		kern_unmount(proc_mnt);
1705
1706	kfree(command_line);
1707}