proc_sysctl.c

   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	inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
 544	err = d_splice_alias_ops(inode, dentry, &proc_sys_dentry_operations);
 545
 546out:
 547	if (h)
 548		sysctl_head_finish(h);
 549	sysctl_head_finish(head);
 550	return err;
 551}
 552
 553static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
 554		int write)
 555{
 556	struct inode *inode = file_inode(iocb->ki_filp);
 557	struct ctl_table_header *head = grab_header(inode);
 558	const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 559	size_t count = iov_iter_count(iter);
 560	char *kbuf;
 561	ssize_t error;
 562
 563	if (IS_ERR(head))
 564		return PTR_ERR(head);
 565
 566	/*
 567	 * At this point we know that the sysctl was not unregistered
 568	 * and won't be until we finish.
 569	 */
 570	error = -EPERM;
 571	if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
 572		goto out;
 573
 574	/* if that can happen at all, it should be -EINVAL, not -EISDIR */
 575	error = -EINVAL;
 576	if (!table->proc_handler)
 577		goto out;
 578
 579	/* don't even try if the size is too large */
 580	error = -ENOMEM;
 581	if (count >= KMALLOC_MAX_SIZE)
 582		goto out;
 583	kbuf = kvzalloc(count + 1, GFP_KERNEL);
 584	if (!kbuf)
 585		goto out;
 586
 587	if (write) {
 588		error = -EFAULT;
 589		if (!copy_from_iter_full(kbuf, count, iter))
 590			goto out_free_buf;
 591		kbuf[count] = '\0';
 592	}
 593
 594	error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
 595					   &iocb->ki_pos);
 596	if (error)
 597		goto out_free_buf;
 598
 599	/* careful: calling conventions are nasty here */
 600	error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
 601	if (error)
 602		goto out_free_buf;
 603
 604	if (!write) {
 605		error = -EFAULT;
 606		if (copy_to_iter(kbuf, count, iter) < count)
 607			goto out_free_buf;
 608	}
 609
 610	error = count;
 611out_free_buf:
 612	kvfree(kbuf);
 613out:
 614	sysctl_head_finish(head);
 615
 616	return error;
 617}
 618
 619static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
 620{
 621	return proc_sys_call_handler(iocb, iter, 0);
 622}
 623
 624static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
 625{
 626	return proc_sys_call_handler(iocb, iter, 1);
 627}
 628
 629static int proc_sys_open(struct inode *inode, struct file *filp)
 630{
 631	struct ctl_table_header *head = grab_header(inode);
 632	const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 633
 634	/* sysctl was unregistered */
 635	if (IS_ERR(head))
 636		return PTR_ERR(head);
 637
 638	if (table->poll)
 639		filp->private_data = proc_sys_poll_event(table->poll);
 640
 641	sysctl_head_finish(head);
 642
 643	return 0;
 644}
 645
 646static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
 647{
 648	struct inode *inode = file_inode(filp);
 649	struct ctl_table_header *head = grab_header(inode);
 650	const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 651	__poll_t ret = DEFAULT_POLLMASK;
 652	unsigned long event;
 653
 654	/* sysctl was unregistered */
 655	if (IS_ERR(head))
 656		return EPOLLERR | EPOLLHUP;
 657
 658	if (!table->proc_handler)
 659		goto out;
 660
 661	if (!table->poll)
 662		goto out;
 663
 664	event = (unsigned long)filp->private_data;
 665	poll_wait(filp, &table->poll->wait, wait);
 666
 667	if (event != atomic_read(&table->poll->event)) {
 668		filp->private_data = proc_sys_poll_event(table->poll);
 669		ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
 670	}
 671
 672out:
 673	sysctl_head_finish(head);
 674
 675	return ret;
 676}
 677
 678static bool proc_sys_fill_cache(struct file *file,
 679				struct dir_context *ctx,
 680				struct ctl_table_header *head,
 681				const struct ctl_table *table)
 682{
 683	struct dentry *child, *dir = file->f_path.dentry;
 684	struct inode *inode;
 685	struct qstr qname;
 686	ino_t ino = 0;
 687	unsigned type = DT_UNKNOWN;
 688
 689	qname.name = table->procname;
 690	qname.len  = strlen(table->procname);
 691	qname.hash = full_name_hash(dir, qname.name, qname.len);
 692
 693	child = d_lookup(dir, &qname);
 694	if (!child) {
 695		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
 696		child = d_alloc_parallel(dir, &qname, &wq);
 697		if (IS_ERR(child))
 698			return false;
 699		if (d_in_lookup(child)) {
 700			struct dentry *res;
 701			inode = proc_sys_make_inode(dir->d_sb, head, table);
 702			res = d_splice_alias_ops(inode, child,
 703						 &proc_sys_dentry_operations);
 704			d_lookup_done(child);
 705			if (unlikely(res)) {
 706				dput(child);
 707
 708				if (IS_ERR(res))
 709					return false;
 710
 711				child = res;
 712			}
 713		}
 714	}
 715	inode = d_inode(child);
 716	ino  = inode->i_ino;
 717	type = inode->i_mode >> 12;
 718	dput(child);
 719	return dir_emit(ctx, qname.name, qname.len, ino, type);
 720}
 721
 722static bool proc_sys_link_fill_cache(struct file *file,
 723				    struct dir_context *ctx,
 724				    struct ctl_table_header *head,
 725				    const struct ctl_table *table)
 726{
 727	bool ret = true;
 728
 729	head = sysctl_head_grab(head);
 730	if (IS_ERR(head))
 731		return false;
 732
 733	/* It is not an error if we can not follow the link ignore it */
 734	if (sysctl_follow_link(&head, &table))
 735		goto out;
 736
 737	ret = proc_sys_fill_cache(file, ctx, head, table);
 738out:
 739	sysctl_head_finish(head);
 740	return ret;
 741}
 742
 743static int scan(struct ctl_table_header *head, const struct ctl_table *table,
 744		unsigned long *pos, struct file *file,
 745		struct dir_context *ctx)
 746{
 747	bool res;
 748
 749	if ((*pos)++ < ctx->pos)
 750		return true;
 751
 752	if (unlikely(S_ISLNK(table->mode)))
 753		res = proc_sys_link_fill_cache(file, ctx, head, table);
 754	else
 755		res = proc_sys_fill_cache(file, ctx, head, table);
 756
 757	if (res)
 758		ctx->pos = *pos;
 759
 760	return res;
 761}
 762
 763static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
 764{
 765	struct ctl_table_header *head = grab_header(file_inode(file));
 766	struct ctl_table_header *h = NULL;
 767	const struct ctl_table *entry;
 768	struct ctl_dir *ctl_dir;
 769	unsigned long pos;
 770
 771	if (IS_ERR(head))
 772		return PTR_ERR(head);
 773
 774	ctl_dir = container_of(head, struct ctl_dir, header);
 775
 776	if (!dir_emit_dots(file, ctx))
 777		goto out;
 778
 779	pos = 2;
 780
 781	for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
 782		if (!scan(h, entry, &pos, file, ctx)) {
 783			sysctl_head_finish(h);
 784			break;
 785		}
 786	}
 787out:
 788	sysctl_head_finish(head);
 789	return 0;
 790}
 791
 792static int proc_sys_permission(struct mnt_idmap *idmap,
 793			       struct inode *inode, int mask)
 794{
 795	/*
 796	 * sysctl entries that are not writeable,
 797	 * are _NOT_ writeable, capabilities or not.
 798	 */
 799	struct ctl_table_header *head;
 800	const struct ctl_table *table;
 801	int error;
 802
 803	/* Executable files are not allowed under /proc/sys/ */
 804	if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
 805		return -EACCES;
 806
 807	head = grab_header(inode);
 808	if (IS_ERR(head))
 809		return PTR_ERR(head);
 810
 811	table = PROC_I(inode)->sysctl_entry;
 812	if (!table) /* global root - r-xr-xr-x */
 813		error = mask & MAY_WRITE ? -EACCES : 0;
 814	else /* Use the permissions on the sysctl table entry */
 815		error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
 816
 817	sysctl_head_finish(head);
 818	return error;
 819}
 820
 821static int proc_sys_setattr(struct mnt_idmap *idmap,
 822			    struct dentry *dentry, struct iattr *attr)
 823{
 824	struct inode *inode = d_inode(dentry);
 825	int error;
 826
 827	if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
 828		return -EPERM;
 829
 830	error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
 831	if (error)
 832		return error;
 833
 834	setattr_copy(&nop_mnt_idmap, inode, attr);
 835	return 0;
 836}
 837
 838static int proc_sys_getattr(struct mnt_idmap *idmap,
 839			    const struct path *path, struct kstat *stat,
 840			    u32 request_mask, unsigned int query_flags)
 841{
 842	struct inode *inode = d_inode(path->dentry);
 843	struct ctl_table_header *head = grab_header(inode);
 844	const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
 845
 846	if (IS_ERR(head))
 847		return PTR_ERR(head);
 848
 849	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
 850	if (table)
 851		stat->mode = (stat->mode & S_IFMT) | table->mode;
 852
 853	sysctl_head_finish(head);
 854	return 0;
 855}
 856
 857static const struct file_operations proc_sys_file_operations = {
 858	.open		= proc_sys_open,
 859	.poll		= proc_sys_poll,
 860	.read_iter	= proc_sys_read,
 861	.write_iter	= proc_sys_write,
 862	.splice_read	= copy_splice_read,
 863	.splice_write	= iter_file_splice_write,
 864	.llseek		= default_llseek,
 865};
 866
 867static const struct file_operations proc_sys_dir_file_operations = {
 868	.read		= generic_read_dir,
 869	.iterate_shared	= proc_sys_readdir,
 870	.llseek		= generic_file_llseek,
 871};
 872
 873static const struct inode_operations proc_sys_inode_operations = {
 874	.permission	= proc_sys_permission,
 875	.setattr	= proc_sys_setattr,
 876	.getattr	= proc_sys_getattr,
 877};
 878
 879static const struct inode_operations proc_sys_dir_operations = {
 880	.lookup		= proc_sys_lookup,
 881	.permission	= proc_sys_permission,
 882	.setattr	= proc_sys_setattr,
 883	.getattr	= proc_sys_getattr,
 884};
 885
 886static int proc_sys_revalidate(struct inode *dir, const struct qstr *name,
 887			       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	if (name->len != len)
 921		return 1;
 922	if (memcmp(name->name, str, len))
 923		return 1;
 924
 925	// false positive is fine here - we'll recheck anyway
 926	if (d_in_lookup(dentry))
 927		return 0;
 928
 929	inode = d_inode_rcu(dentry);
 930	// we just might have run into dentry in the middle of __dentry_kill()
 931	if (!inode)
 932		return 1;
 933
 934	head = READ_ONCE(PROC_I(inode)->sysctl);
 935	return !head || !sysctl_is_seen(head);
 936}
 937
 938static const struct dentry_operations proc_sys_dentry_operations = {
 939	.d_revalidate	= proc_sys_revalidate,
 940	.d_delete	= proc_sys_delete,
 941	.d_compare	= proc_sys_compare,
 942};
 943
 944static struct ctl_dir *find_subdir(struct ctl_dir *dir,
 945				   const char *name, int namelen)
 946{
 947	struct ctl_table_header *head;
 948	const struct ctl_table *entry;
 949
 950	entry = find_entry(&head, dir, name, namelen);
 951	if (!entry)
 952		return ERR_PTR(-ENOENT);
 953	if (!S_ISDIR(entry->mode))
 954		return ERR_PTR(-ENOTDIR);
 955	return container_of(head, struct ctl_dir, header);
 956}
 957
 958static struct ctl_dir *new_dir(struct ctl_table_set *set,
 959			       const char *name, int namelen)
 960{
 961	struct ctl_table *table;
 962	struct ctl_dir *new;
 963	struct ctl_node *node;
 964	char *new_name;
 965
 966	new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
 967		      sizeof(struct ctl_table) +  namelen + 1,
 968		      GFP_KERNEL);
 969	if (!new)
 970		return NULL;
 971
 972	node = (struct ctl_node *)(new + 1);
 973	table = (struct ctl_table *)(node + 1);
 974	new_name = (char *)(table + 1);
 975	memcpy(new_name, name, namelen);
 976	table[0].procname = new_name;
 977	table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
 978	init_header(&new->header, set->dir.header.root, set, node, table, 1);
 979
 980	return new;
 981}
 982
 983/**
 984 * get_subdir - find or create a subdir with the specified name.
 985 * @dir:  Directory to create the subdirectory in
 986 * @name: The name of the subdirectory to find or create
 987 * @namelen: The length of name
 988 *
 989 * Takes a directory with an elevated reference count so we know that
 990 * if we drop the lock the directory will not go away.  Upon success
 991 * the reference is moved from @dir to the returned subdirectory.
 992 * Upon error an error code is returned and the reference on @dir is
 993 * simply dropped.
 994 */
 995static struct ctl_dir *get_subdir(struct ctl_dir *dir,
 996				  const char *name, int namelen)
 997{
 998	struct ctl_table_set *set = dir->header.set;
 999	struct ctl_dir *subdir, *new = NULL;
1000	int err;
1001
1002	spin_lock(&sysctl_lock);
1003	subdir = find_subdir(dir, name, namelen);
1004	if (!IS_ERR(subdir))
1005		goto found;
1006	if (PTR_ERR(subdir) != -ENOENT)
1007		goto failed;
1008
1009	spin_unlock(&sysctl_lock);
1010	new = new_dir(set, name, namelen);
1011	spin_lock(&sysctl_lock);
1012	subdir = ERR_PTR(-ENOMEM);
1013	if (!new)
1014		goto failed;
1015
1016	/* Was the subdir added while we dropped the lock? */
1017	subdir = find_subdir(dir, name, namelen);
1018	if (!IS_ERR(subdir))
1019		goto found;
1020	if (PTR_ERR(subdir) != -ENOENT)
1021		goto failed;
1022
1023	/* Nope.  Use the our freshly made directory entry. */
1024	err = insert_header(dir, &new->header);
1025	subdir = ERR_PTR(err);
1026	if (err)
1027		goto failed;
1028	subdir = new;
1029found:
1030	subdir->header.nreg++;
1031failed:
1032	if (IS_ERR(subdir)) {
1033		pr_err("sysctl could not get directory: ");
1034		sysctl_print_dir(dir);
1035		pr_cont("%*.*s %ld\n", namelen, namelen, name,
1036			PTR_ERR(subdir));
1037	}
1038	drop_sysctl_table(&dir->header);
1039	if (new)
1040		drop_sysctl_table(&new->header);
1041	spin_unlock(&sysctl_lock);
1042	return subdir;
1043}
1044
1045static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1046{
1047	struct ctl_dir *parent;
1048	const char *procname;
1049	if (!dir->header.parent)
1050		return &set->dir;
1051	parent = xlate_dir(set, dir->header.parent);
1052	if (IS_ERR(parent))
1053		return parent;
1054	procname = dir->header.ctl_table[0].procname;
1055	return find_subdir(parent, procname, strlen(procname));
1056}
1057
1058static int sysctl_follow_link(struct ctl_table_header **phead,
1059	const struct ctl_table **pentry)
1060{
1061	struct ctl_table_header *head;
1062	const struct ctl_table *entry;
1063	struct ctl_table_root *root;
1064	struct ctl_table_set *set;
1065	struct ctl_dir *dir;
1066	int ret;
1067
1068	spin_lock(&sysctl_lock);
1069	root = (*pentry)->data;
1070	set = lookup_header_set(root);
1071	dir = xlate_dir(set, (*phead)->parent);
1072	if (IS_ERR(dir))
1073		ret = PTR_ERR(dir);
1074	else {
1075		const char *procname = (*pentry)->procname;
1076		head = NULL;
1077		entry = find_entry(&head, dir, procname, strlen(procname));
1078		ret = -ENOENT;
1079		if (entry && use_table(head)) {
1080			unuse_table(*phead);
1081			*phead = head;
1082			*pentry = entry;
1083			ret = 0;
1084		}
1085	}
1086
1087	spin_unlock(&sysctl_lock);
1088	return ret;
1089}
1090
1091static int sysctl_err(const char *path, const struct ctl_table *table, char *fmt, ...)
1092{
1093	struct va_format vaf;
1094	va_list args;
1095
1096	va_start(args, fmt);
1097	vaf.fmt = fmt;
1098	vaf.va = &args;
1099
1100	pr_err("sysctl table check failed: %s/%s %pV\n",
1101	       path, table->procname, &vaf);
1102
1103	va_end(args);
1104	return -EINVAL;
1105}
1106
1107static int sysctl_check_table_array(const char *path, const struct ctl_table *table)
1108{
1109	unsigned int extra;
1110	int err = 0;
1111
1112	if ((table->proc_handler == proc_douintvec) ||
1113	    (table->proc_handler == proc_douintvec_minmax)) {
1114		if (table->maxlen != sizeof(unsigned int))
1115			err |= sysctl_err(path, table, "array not allowed");
1116	}
1117
1118	if (table->proc_handler == proc_dou8vec_minmax) {
1119		if (table->maxlen != sizeof(u8))
1120			err |= sysctl_err(path, table, "array not allowed");
1121
1122		if (table->extra1) {
1123			extra = *(unsigned int *) table->extra1;
1124			if (extra > 255U)
1125				err |= sysctl_err(path, table,
1126						"range value too large for proc_dou8vec_minmax");
1127		}
1128		if (table->extra2) {
1129			extra = *(unsigned int *) table->extra2;
1130			if (extra > 255U)
1131				err |= sysctl_err(path, table,
1132						"range value too large for proc_dou8vec_minmax");
1133		}
1134	}
1135
1136	if (table->proc_handler == proc_dobool) {
1137		if (table->maxlen != sizeof(bool))
1138			err |= sysctl_err(path, table, "array not allowed");
1139	}
1140
1141	return err;
1142}
1143
1144static int sysctl_check_table(const char *path, struct ctl_table_header *header)
1145{
1146	const struct ctl_table *entry;
1147	int err = 0;
1148	list_for_each_table_entry(entry, header) {
1149		if (!entry->procname)
1150			err |= sysctl_err(path, entry, "procname is null");
1151		if ((entry->proc_handler == proc_dostring) ||
1152		    (entry->proc_handler == proc_dobool) ||
1153		    (entry->proc_handler == proc_dointvec) ||
1154		    (entry->proc_handler == proc_douintvec) ||
1155		    (entry->proc_handler == proc_douintvec_minmax) ||
1156		    (entry->proc_handler == proc_dointvec_minmax) ||
1157		    (entry->proc_handler == proc_dou8vec_minmax) ||
1158		    (entry->proc_handler == proc_dointvec_jiffies) ||
1159		    (entry->proc_handler == proc_dointvec_userhz_jiffies) ||
1160		    (entry->proc_handler == proc_dointvec_ms_jiffies) ||
1161		    (entry->proc_handler == proc_doulongvec_minmax) ||
1162		    (entry->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1163			if (!entry->data)
1164				err |= sysctl_err(path, entry, "No data");
1165			if (!entry->maxlen)
1166				err |= sysctl_err(path, entry, "No maxlen");
1167			else
1168				err |= sysctl_check_table_array(path, entry);
1169		}
1170		if (!entry->proc_handler)
1171			err |= sysctl_err(path, entry, "No proc_handler");
1172
1173		if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode)
1174			err |= sysctl_err(path, entry, "bogus .mode 0%o",
1175				entry->mode);
1176	}
1177	return err;
1178}
1179
1180static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table_header *head)
1181{
1182	struct ctl_table *link_table, *link;
1183	struct ctl_table_header *links;
1184	const struct ctl_table *entry;
1185	struct ctl_node *node;
1186	char *link_name;
1187	int name_bytes;
1188
1189	name_bytes = 0;
1190	list_for_each_table_entry(entry, head) {
1191		name_bytes += strlen(entry->procname) + 1;
1192	}
1193
1194	links = kzalloc(sizeof(struct ctl_table_header) +
1195			sizeof(struct ctl_node)*head->ctl_table_size +
1196			sizeof(struct ctl_table)*head->ctl_table_size +
1197			name_bytes,
1198			GFP_KERNEL);
1199
1200	if (!links)
1201		return NULL;
1202
1203	node = (struct ctl_node *)(links + 1);
1204	link_table = (struct ctl_table *)(node + head->ctl_table_size);
1205	link_name = (char *)(link_table + head->ctl_table_size);
1206	link = link_table;
1207
1208	list_for_each_table_entry(entry, head) {
1209		int len = strlen(entry->procname) + 1;
1210		memcpy(link_name, entry->procname, len);
1211		link->procname = link_name;
1212		link->mode = S_IFLNK|S_IRWXUGO;
1213		link->data = head->root;
1214		link_name += len;
1215		link++;
1216	}
1217	init_header(links, dir->header.root, dir->header.set, node, link_table,
1218		    head->ctl_table_size);
1219	links->nreg = head->ctl_table_size;
1220
1221	return links;
1222}
1223
1224static bool get_links(struct ctl_dir *dir,
1225		      struct ctl_table_header *header,
1226		      struct ctl_table_root *link_root)
1227{
1228	struct ctl_table_header *tmp_head;
1229	const struct ctl_table *entry, *link;
1230
1231	if (header->ctl_table_size == 0 ||
1232	    sysctl_is_perm_empty_ctl_header(header))
1233		return true;
1234
1235	/* Are there links available for every entry in table? */
1236	list_for_each_table_entry(entry, header) {
1237		const char *procname = entry->procname;
1238		link = find_entry(&tmp_head, dir, procname, strlen(procname));
1239		if (!link)
1240			return false;
1241		if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1242			continue;
1243		if (S_ISLNK(link->mode) && (link->data == link_root))
1244			continue;
1245		return false;
1246	}
1247
1248	/* The checks passed.  Increase the registration count on the links */
1249	list_for_each_table_entry(entry, header) {
1250		const char *procname = entry->procname;
1251		link = find_entry(&tmp_head, dir, procname, strlen(procname));
1252		tmp_head->nreg++;
1253	}
1254	return true;
1255}
1256
1257static int insert_links(struct ctl_table_header *head)
1258{
1259	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1260	struct ctl_dir *core_parent;
1261	struct ctl_table_header *links;
1262	int err;
1263
1264	if (head->set == root_set)
1265		return 0;
1266
1267	core_parent = xlate_dir(root_set, head->parent);
1268	if (IS_ERR(core_parent))
1269		return 0;
1270
1271	if (get_links(core_parent, head, head->root))
1272		return 0;
1273
1274	core_parent->header.nreg++;
1275	spin_unlock(&sysctl_lock);
1276
1277	links = new_links(core_parent, head);
1278
1279	spin_lock(&sysctl_lock);
1280	err = -ENOMEM;
1281	if (!links)
1282		goto out;
1283
1284	err = 0;
1285	if (get_links(core_parent, head, head->root)) {
1286		kfree(links);
1287		goto out;
1288	}
1289
1290	err = insert_header(core_parent, links);
1291	if (err)
1292		kfree(links);
1293out:
1294	drop_sysctl_table(&core_parent->header);
1295	return err;
1296}
1297
1298/* Find the directory for the ctl_table. If one is not found create it. */
1299static struct ctl_dir *sysctl_mkdir_p(struct ctl_dir *dir, const char *path)
1300{
1301	const char *name, *nextname;
1302
1303	for (name = path; name; name = nextname) {
1304		int namelen;
1305		nextname = strchr(name, '/');
1306		if (nextname) {
1307			namelen = nextname - name;
1308			nextname++;
1309		} else {
1310			namelen = strlen(name);
1311		}
1312		if (namelen == 0)
1313			continue;
1314
1315		/*
1316		 * namelen ensures if name is "foo/bar/yay" only foo is
1317		 * registered first. We traverse as if using mkdir -p and
1318		 * return a ctl_dir for the last directory entry.
1319		 */
1320		dir = get_subdir(dir, name, namelen);
1321		if (IS_ERR(dir))
1322			break;
1323	}
1324	return dir;
1325}
1326
1327/**
1328 * __register_sysctl_table - register a leaf sysctl table
1329 * @set: Sysctl tree to register on
1330 * @path: The path to the directory the sysctl table is in.
1331 *
1332 * @table: the top-level table structure. This table should not be free'd
1333 *         after registration. So it should not be used on stack. It can either
1334 *         be a global or dynamically allocated by the caller and free'd later
1335 *         after sysctl unregistration.
1336 * @table_size : The number of elements in table
1337 *
1338 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1339 * array.
1340 *
1341 * The members of the &struct ctl_table structure are used as follows:
1342 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1343 *            enter a sysctl file
1344 * data     - a pointer to data for use by proc_handler
1345 * maxlen   - the maximum size in bytes of the data
1346 * mode     - the file permissions for the /proc/sys file
1347 * type     - Defines the target type (described in struct definition)
1348 * proc_handler - the text handler routine (described below)
1349 *
1350 * extra1, extra2 - extra pointers usable by the proc handler routines
1351 * XXX: we should eventually modify these to use long min / max [0]
1352 * [0] https://lkml.kernel.org/87zgpte9o4.fsf@email.froward.int.ebiederm.org
1353 *
1354 * Leaf nodes in the sysctl tree will be represented by a single file
1355 * under /proc; non-leaf nodes are not allowed.
1356 *
1357 * There must be a proc_handler routine for any terminal nodes.
1358 * Several default handlers are available to cover common cases -
1359 *
1360 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1361 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1362 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1363 *
1364 * It is the handler's job to read the input buffer from user memory
1365 * and process it. The handler should return 0 on success.
1366 *
1367 * This routine returns %NULL on a failure to register, and a pointer
1368 * to the table header on success.
1369 */
1370struct ctl_table_header *__register_sysctl_table(
1371	struct ctl_table_set *set,
1372	const char *path, const struct ctl_table *table, size_t table_size)
1373{
1374	struct ctl_table_root *root = set->dir.header.root;
1375	struct ctl_table_header *header;
1376	struct ctl_dir *dir;
1377	struct ctl_node *node;
1378
1379	header = kzalloc(sizeof(struct ctl_table_header) +
1380			 sizeof(struct ctl_node)*table_size, GFP_KERNEL_ACCOUNT);
1381	if (!header)
1382		return NULL;
1383
1384	node = (struct ctl_node *)(header + 1);
1385	init_header(header, root, set, node, table, table_size);
1386	if (sysctl_check_table(path, header))
1387		goto fail;
1388
1389	spin_lock(&sysctl_lock);
1390	dir = &set->dir;
1391	/* Reference moved down the directory tree get_subdir */
1392	dir->header.nreg++;
1393	spin_unlock(&sysctl_lock);
1394
1395	dir = sysctl_mkdir_p(dir, path);
1396	if (IS_ERR(dir))
1397		goto fail;
1398	spin_lock(&sysctl_lock);
1399	if (insert_header(dir, header))
1400		goto fail_put_dir_locked;
1401
1402	drop_sysctl_table(&dir->header);
1403	spin_unlock(&sysctl_lock);
1404
1405	return header;
1406
1407fail_put_dir_locked:
1408	drop_sysctl_table(&dir->header);
1409	spin_unlock(&sysctl_lock);
1410fail:
1411	kfree(header);
1412	return NULL;
1413}
1414
1415/**
1416 * register_sysctl_sz - register a sysctl table
1417 * @path: The path to the directory the sysctl table is in. If the path
1418 * 	doesn't exist we will create it for you.
1419 * @table: the table structure. The calller must ensure the life of the @table
1420 * 	will be kept during the lifetime use of the syctl. It must not be freed
1421 * 	until unregister_sysctl_table() is called with the given returned table
1422 * 	with this registration. If your code is non modular then you don't need
1423 * 	to call unregister_sysctl_table() and can instead use something like
1424 * 	register_sysctl_init() which does not care for the result of the syctl
1425 * 	registration.
1426 * @table_size: The number of elements in table.
1427 *
1428 * Register a sysctl table. @table should be a filled in ctl_table
1429 * array. A completely 0 filled entry terminates the table.
1430 *
1431 * See __register_sysctl_table for more details.
1432 */
1433struct ctl_table_header *register_sysctl_sz(const char *path, const struct ctl_table *table,
1434					    size_t table_size)
1435{
1436	return __register_sysctl_table(&sysctl_table_root.default_set,
1437					path, table, table_size);
1438}
1439EXPORT_SYMBOL(register_sysctl_sz);
1440
1441/**
1442 * __register_sysctl_init() - register sysctl table to path
1443 * @path: path name for sysctl base. If that path doesn't exist we will create
1444 * 	it for you.
1445 * @table: This is the sysctl table that needs to be registered to the path.
1446 * 	The caller must ensure the life of the @table will be kept during the
1447 * 	lifetime use of the sysctl.
1448 * @table_name: The name of sysctl table, only used for log printing when
1449 *              registration fails
1450 * @table_size: The number of elements in table
1451 *
1452 * The sysctl interface is used by userspace to query or modify at runtime
1453 * a predefined value set on a variable. These variables however have default
1454 * values pre-set. Code which depends on these variables will always work even
1455 * if register_sysctl() fails. If register_sysctl() fails you'd just loose the
1456 * ability to query or modify the sysctls dynamically at run time. Chances of
1457 * register_sysctl() failing on init are extremely low, and so for both reasons
1458 * this function does not return any error as it is used by initialization code.
1459 *
1460 * Context: if your base directory does not exist it will be created for you.
1461 */
1462void __init __register_sysctl_init(const char *path, const struct ctl_table *table,
1463				 const char *table_name, size_t table_size)
1464{
1465	struct ctl_table_header *hdr = register_sysctl_sz(path, table, table_size);
1466
1467	if (unlikely(!hdr)) {
1468		pr_err("failed when register_sysctl_sz %s to %s\n", table_name, path);
1469		return;
1470	}
1471	kmemleak_not_leak(hdr);
1472}
1473
1474static void put_links(struct ctl_table_header *header)
1475{
1476	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1477	struct ctl_table_root *root = header->root;
1478	struct ctl_dir *parent = header->parent;
1479	struct ctl_dir *core_parent;
1480	const struct ctl_table *entry;
1481
1482	if (header->set == root_set)
1483		return;
1484
1485	core_parent = xlate_dir(root_set, parent);
1486	if (IS_ERR(core_parent))
1487		return;
1488
1489	list_for_each_table_entry(entry, header) {
1490		struct ctl_table_header *link_head;
1491		const struct ctl_table *link;
1492		const char *name = entry->procname;
1493
1494		link = find_entry(&link_head, core_parent, name, strlen(name));
1495		if (link &&
1496		    ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1497		     (S_ISLNK(link->mode) && (link->data == root)))) {
1498			drop_sysctl_table(link_head);
1499		}
1500		else {
1501			pr_err("sysctl link missing during unregister: ");
1502			sysctl_print_dir(parent);
1503			pr_cont("%s\n", name);
1504		}
1505	}
1506}
1507
1508static void drop_sysctl_table(struct ctl_table_header *header)
1509{
1510	struct ctl_dir *parent = header->parent;
1511
1512	if (--header->nreg)
1513		return;
1514
1515	if (parent) {
1516		put_links(header);
1517		start_unregistering(header);
1518	}
1519
1520	if (!--header->count)
1521		kfree_rcu(header, rcu);
1522
1523	if (parent)
1524		drop_sysctl_table(&parent->header);
1525}
1526
1527/**
1528 * unregister_sysctl_table - unregister a sysctl table hierarchy
1529 * @header: the header returned from register_sysctl or __register_sysctl_table
1530 *
1531 * Unregisters the sysctl table and all children. proc entries may not
1532 * actually be removed until they are no longer used by anyone.
1533 */
1534void unregister_sysctl_table(struct ctl_table_header * header)
1535{
1536	might_sleep();
1537
1538	if (header == NULL)
1539		return;
1540
1541	spin_lock(&sysctl_lock);
1542	drop_sysctl_table(header);
1543	spin_unlock(&sysctl_lock);
1544}
1545EXPORT_SYMBOL(unregister_sysctl_table);
1546
1547void setup_sysctl_set(struct ctl_table_set *set,
1548	struct ctl_table_root *root,
1549	int (*is_seen)(struct ctl_table_set *))
1550{
1551	memset(set, 0, sizeof(*set));
1552	set->is_seen = is_seen;
1553	init_header(&set->dir.header, root, set, NULL, root_table, 1);
1554}
1555
1556void retire_sysctl_set(struct ctl_table_set *set)
1557{
1558	WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1559}
1560
1561int __init proc_sys_init(void)
1562{
1563	struct proc_dir_entry *proc_sys_root;
1564
1565	proc_sys_root = proc_mkdir("sys", NULL);
1566	proc_sys_root->proc_iops = &proc_sys_dir_operations;
1567	proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1568	proc_sys_root->nlink = 0;
1569
1570	return sysctl_init_bases();
1571}
1572
1573struct sysctl_alias {
1574	const char *kernel_param;
1575	const char *sysctl_param;
1576};
1577
1578/*
1579 * Historically some settings had both sysctl and a command line parameter.
1580 * With the generic sysctl. parameter support, we can handle them at a single
1581 * place and only keep the historical name for compatibility. This is not meant
1582 * to add brand new aliases. When adding existing aliases, consider whether
1583 * the possibly different moment of changing the value (e.g. from early_param
1584 * to the moment do_sysctl_args() is called) is an issue for the specific
1585 * parameter.
1586 */
1587static const struct sysctl_alias sysctl_aliases[] = {
1588	{"hardlockup_all_cpu_backtrace",	"kernel.hardlockup_all_cpu_backtrace" },
1589	{"hung_task_panic",			"kernel.hung_task_panic" },
1590	{"numa_zonelist_order",			"vm.numa_zonelist_order" },
1591	{"softlockup_all_cpu_backtrace",	"kernel.softlockup_all_cpu_backtrace" },
1592	{ }
1593};
1594
1595static const char *sysctl_find_alias(char *param)
1596{
1597	const struct sysctl_alias *alias;
1598
1599	for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1600		if (strcmp(alias->kernel_param, param) == 0)
1601			return alias->sysctl_param;
1602	}
1603
1604	return NULL;
1605}
1606
1607bool sysctl_is_alias(char *param)
1608{
1609	const char *alias = sysctl_find_alias(param);
1610
1611	return alias != NULL;
1612}
1613
1614/* Set sysctl value passed on kernel command line. */
1615static int process_sysctl_arg(char *param, char *val,
1616			       const char *unused, void *arg)
1617{
1618	char *path;
1619	struct vfsmount **proc_mnt = arg;
1620	struct file_system_type *proc_fs_type;
1621	struct file *file;
1622	int len;
1623	int err;
1624	loff_t pos = 0;
1625	ssize_t wret;
1626
1627	if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1628		param += sizeof("sysctl") - 1;
1629
1630		if (param[0] != '/' && param[0] != '.')
1631			return 0;
1632
1633		param++;
1634	} else {
1635		param = (char *) sysctl_find_alias(param);
1636		if (!param)
1637			return 0;
1638	}
1639
1640	if (!val)
1641		return -EINVAL;
1642	len = strlen(val);
1643	if (len == 0)
1644		return -EINVAL;
1645
1646	/*
1647	 * To set sysctl options, we use a temporary mount of proc, look up the
1648	 * respective sys/ file and write to it. To avoid mounting it when no
1649	 * options were given, we mount it only when the first sysctl option is
1650	 * found. Why not a persistent mount? There are problems with a
1651	 * persistent mount of proc in that it forces userspace not to use any
1652	 * proc mount options.
1653	 */
1654	if (!*proc_mnt) {
1655		proc_fs_type = get_fs_type("proc");
1656		if (!proc_fs_type) {
1657			pr_err("Failed to find procfs to set sysctl from command line\n");
1658			return 0;
1659		}
1660		*proc_mnt = kern_mount(proc_fs_type);
1661		put_filesystem(proc_fs_type);
1662		if (IS_ERR(*proc_mnt)) {
1663			pr_err("Failed to mount procfs to set sysctl from command line\n");
1664			return 0;
1665		}
1666	}
1667
1668	path = kasprintf(GFP_KERNEL, "sys/%s", param);
1669	if (!path)
1670		panic("%s: Failed to allocate path for %s\n", __func__, param);
1671	strreplace(path, '.', '/');
1672
1673	file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1674	if (IS_ERR(file)) {
1675		err = PTR_ERR(file);
1676		if (err == -ENOENT)
1677			pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1678				param, val);
1679		else if (err == -EACCES)
1680			pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1681				param, val);
1682		else
1683			pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1684				file, param, val);
1685		goto out;
1686	}
1687	wret = kernel_write(file, val, len, &pos);
1688	if (wret < 0) {
1689		err = wret;
1690		if (err == -EINVAL)
1691			pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1692				param, val);
1693		else
1694			pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1695				ERR_PTR(err), param, val);
1696	} else if (wret != len) {
1697		pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1698			wret, len, path, param, val);
1699	}
1700
1701	err = filp_close(file, NULL);
1702	if (err)
1703		pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1704			ERR_PTR(err), param, val);
1705out:
1706	kfree(path);
1707	return 0;
1708}
1709
1710void do_sysctl_args(void)
1711{
1712	char *command_line;
1713	struct vfsmount *proc_mnt = NULL;
1714
1715	command_line = kstrdup(saved_command_line, GFP_KERNEL);
1716	if (!command_line)
1717		panic("%s: Failed to allocate copy of command line\n", __func__);
1718
1719	parse_args("Setting sysctl args", command_line,
1720		   NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1721
1722	if (proc_mnt)
1723		kern_unmount(proc_mnt);
1724
1725	kfree(command_line);
1726}