1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * fs/kernfs/dir.c - kernfs directory implementation
   4 *
   5 * Copyright (c) 2001-3 Patrick Mochel
   6 * Copyright (c) 2007 SUSE Linux Products GmbH
   7 * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org>
   8 */
   9
  10#include <linux/sched.h>
  11#include <linux/fs.h>
  12#include <linux/namei.h>
  13#include <linux/idr.h>
  14#include <linux/slab.h>
  15#include <linux/security.h>
  16#include <linux/hash.h>
  17
  18#include "kernfs-internal.h"
  19
  20static DEFINE_RWLOCK(kernfs_rename_lock);	/* kn->parent and ->name */
  21/*
  22 * Don't use rename_lock to piggy back on pr_cont_buf. We don't want to
  23 * call pr_cont() while holding rename_lock. Because sometimes pr_cont()
  24 * will perform wakeups when releasing console_sem. Holding rename_lock
  25 * will introduce deadlock if the scheduler reads the kernfs_name in the
  26 * wakeup path.
  27 */
  28static DEFINE_SPINLOCK(kernfs_pr_cont_lock);
  29static char kernfs_pr_cont_buf[PATH_MAX];	/* protected by pr_cont_lock */
  30static DEFINE_SPINLOCK(kernfs_idr_lock);	/* root->ino_idr */
  31
  32#define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb)
  33
  34static bool __kernfs_active(struct kernfs_node *kn)
  35{
  36	return atomic_read(&kn->active) >= 0;
  37}
  38
  39static bool kernfs_active(struct kernfs_node *kn)
  40{
  41	lockdep_assert_held(&kernfs_root(kn)->kernfs_rwsem);
  42	return __kernfs_active(kn);
  43}
  44
  45static bool kernfs_lockdep(struct kernfs_node *kn)
  46{
  47#ifdef CONFIG_DEBUG_LOCK_ALLOC
  48	return kn->flags & KERNFS_LOCKDEP;
  49#else
  50	return false;
  51#endif
  52}
  53
  54static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen)
  55{
  56	if (!kn)
  57		return strscpy(buf, "(null)", buflen);
  58
  59	return strscpy(buf, kn->parent ? kn->name : "/", buflen);
  60}
  61
  62/* kernfs_node_depth - compute depth from @from to @to */
  63static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
  64{
  65	size_t depth = 0;
  66
  67	while (to->parent && to != from) {
  68		depth++;
  69		to = to->parent;
  70	}
  71	return depth;
  72}
  73
  74static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
  75						  struct kernfs_node *b)
  76{
  77	size_t da, db;
  78	struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b);
  79
  80	if (ra != rb)
  81		return NULL;
  82
  83	da = kernfs_depth(ra->kn, a);
  84	db = kernfs_depth(rb->kn, b);
  85
  86	while (da > db) {
  87		a = a->parent;
  88		da--;
  89	}
  90	while (db > da) {
  91		b = b->parent;
  92		db--;
  93	}
  94
  95	/* worst case b and a will be the same at root */
  96	while (b != a) {
  97		b = b->parent;
  98		a = a->parent;
  99	}
 100
 101	return a;
 102}
 103
 104/**
 105 * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to,
 106 * where kn_from is treated as root of the path.
 107 * @kn_from: kernfs node which should be treated as root for the path
 108 * @kn_to: kernfs node to which path is needed
 109 * @buf: buffer to copy the path into
 110 * @buflen: size of @buf
 111 *
 112 * We need to handle couple of scenarios here:
 113 * [1] when @kn_from is an ancestor of @kn_to at some level
 114 * kn_from: /n1/n2/n3
 115 * kn_to:   /n1/n2/n3/n4/n5
 116 * result:  /n4/n5
 117 *
 118 * [2] when @kn_from is on a different hierarchy and we need to find common
 119 * ancestor between @kn_from and @kn_to.
 120 * kn_from: /n1/n2/n3/n4
 121 * kn_to:   /n1/n2/n5
 122 * result:  /../../n5
 123 * OR
 124 * kn_from: /n1/n2/n3/n4/n5   [depth=5]
 125 * kn_to:   /n1/n2/n3         [depth=3]
 126 * result:  /../..
 127 *
 128 * [3] when @kn_to is %NULL result will be "(null)"
 129 *
 130 * Return: the length of the constructed path.  If the path would have been
 131 * greater than @buflen, @buf contains the truncated path with the trailing
 132 * '\0'.  On error, -errno is returned.
 133 */
 134static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
 135					struct kernfs_node *kn_from,
 136					char *buf, size_t buflen)
 137{
 138	struct kernfs_node *kn, *common;
 139	const char parent_str[] = "/..";
 140	size_t depth_from, depth_to, len = 0;
 141	ssize_t copied;
 142	int i, j;
 143
 144	if (!kn_to)
 145		return strscpy(buf, "(null)", buflen);
 146
 147	if (!kn_from)
 148		kn_from = kernfs_root(kn_to)->kn;
 149
 150	if (kn_from == kn_to)
 151		return strscpy(buf, "/", buflen);
 152
 153	common = kernfs_common_ancestor(kn_from, kn_to);
 154	if (WARN_ON(!common))
 155		return -EINVAL;
 156
 157	depth_to = kernfs_depth(common, kn_to);
 158	depth_from = kernfs_depth(common, kn_from);
 159
 160	buf[0] = '\0';
 161
 162	for (i = 0; i < depth_from; i++) {
 163		copied = strscpy(buf + len, parent_str, buflen - len);
 164		if (copied < 0)
 165			return copied;
 166		len += copied;
 167	}
 168
 169	/* Calculate how many bytes we need for the rest */
 170	for (i = depth_to - 1; i >= 0; i--) {
 171		for (kn = kn_to, j = 0; j < i; j++)
 172			kn = kn->parent;
 173
 174		len += scnprintf(buf + len, buflen - len, "/%s", kn->name);
 175	}
 176
 177	return len;
 178}
 179
 180/**
 181 * kernfs_name - obtain the name of a given node
 182 * @kn: kernfs_node of interest
 183 * @buf: buffer to copy @kn's name into
 184 * @buflen: size of @buf
 185 *
 186 * Copies the name of @kn into @buf of @buflen bytes.  The behavior is
 187 * similar to strscpy().
 188 *
 189 * Fills buffer with "(null)" if @kn is %NULL.
 190 *
 191 * Return: the resulting length of @buf. If @buf isn't long enough,
 192 * it's filled up to @buflen-1 and nul terminated, and returns -E2BIG.
 193 *
 194 * This function can be called from any context.
 195 */
 196int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
 197{
 198	unsigned long flags;
 199	int ret;
 200
 201	read_lock_irqsave(&kernfs_rename_lock, flags);
 202	ret = kernfs_name_locked(kn, buf, buflen);
 203	read_unlock_irqrestore(&kernfs_rename_lock, flags);
 204	return ret;
 205}
 206
 207/**
 208 * kernfs_path_from_node - build path of node @to relative to @from.
 209 * @from: parent kernfs_node relative to which we need to build the path
 210 * @to: kernfs_node of interest
 211 * @buf: buffer to copy @to's path into
 212 * @buflen: size of @buf
 213 *
 214 * Builds @to's path relative to @from in @buf. @from and @to must
 215 * be on the same kernfs-root. If @from is not parent of @to, then a relative
 216 * path (which includes '..'s) as needed to reach from @from to @to is
 217 * returned.
 218 *
 219 * Return: the length of the constructed path.  If the path would have been
 220 * greater than @buflen, @buf contains the truncated path with the trailing
 221 * '\0'.  On error, -errno is returned.
 222 */
 223int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
 224			  char *buf, size_t buflen)
 225{
 226	unsigned long flags;
 227	int ret;
 228
 229	read_lock_irqsave(&kernfs_rename_lock, flags);
 230	ret = kernfs_path_from_node_locked(to, from, buf, buflen);
 231	read_unlock_irqrestore(&kernfs_rename_lock, flags);
 232	return ret;
 233}
 234EXPORT_SYMBOL_GPL(kernfs_path_from_node);
 235
 236/**
 237 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
 238 * @kn: kernfs_node of interest
 239 *
 240 * This function can be called from any context.
 241 */
 242void pr_cont_kernfs_name(struct kernfs_node *kn)
 243{
 244	unsigned long flags;
 245
 246	spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
 247
 248	kernfs_name(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
 249	pr_cont("%s", kernfs_pr_cont_buf);
 250
 251	spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
 252}
 253
 254/**
 255 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
 256 * @kn: kernfs_node of interest
 257 *
 258 * This function can be called from any context.
 259 */
 260void pr_cont_kernfs_path(struct kernfs_node *kn)
 261{
 262	unsigned long flags;
 263	int sz;
 264
 265	spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
 266
 267	sz = kernfs_path_from_node(kn, NULL, kernfs_pr_cont_buf,
 268				   sizeof(kernfs_pr_cont_buf));
 269	if (sz < 0) {
 270		if (sz == -E2BIG)
 271			pr_cont("(name too long)");
 272		else
 273			pr_cont("(error)");
 274		goto out;
 275	}
 276
 277	pr_cont("%s", kernfs_pr_cont_buf);
 278
 279out:
 280	spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
 281}
 282
 283/**
 284 * kernfs_get_parent - determine the parent node and pin it
 285 * @kn: kernfs_node of interest
 286 *
 287 * Determines @kn's parent, pins and returns it.  This function can be
 288 * called from any context.
 289 *
 290 * Return: parent node of @kn
 291 */
 292struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
 293{
 294	struct kernfs_node *parent;
 295	unsigned long flags;
 296
 297	read_lock_irqsave(&kernfs_rename_lock, flags);
 298	parent = kn->parent;
 299	kernfs_get(parent);
 300	read_unlock_irqrestore(&kernfs_rename_lock, flags);
 301
 302	return parent;
 303}
 304
 305/**
 306 *	kernfs_name_hash - calculate hash of @ns + @name
 307 *	@name: Null terminated string to hash
 308 *	@ns:   Namespace tag to hash
 309 *
 310 *	Return: 31-bit hash of ns + name (so it fits in an off_t)
 311 */
 312static unsigned int kernfs_name_hash(const char *name, const void *ns)
 313{
 314	unsigned long hash = init_name_hash(ns);
 315	unsigned int len = strlen(name);
 316	while (len--)
 317		hash = partial_name_hash(*name++, hash);
 318	hash = end_name_hash(hash);
 319	hash &= 0x7fffffffU;
 320	/* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
 321	if (hash < 2)
 322		hash += 2;
 323	if (hash >= INT_MAX)
 324		hash = INT_MAX - 1;
 325	return hash;
 326}
 327
 328static int kernfs_name_compare(unsigned int hash, const char *name,
 329			       const void *ns, const struct kernfs_node *kn)
 330{
 331	if (hash < kn->hash)
 332		return -1;
 333	if (hash > kn->hash)
 334		return 1;
 335	if (ns < kn->ns)
 336		return -1;
 337	if (ns > kn->ns)
 338		return 1;
 339	return strcmp(name, kn->name);
 340}
 341
 342static int kernfs_sd_compare(const struct kernfs_node *left,
 343			     const struct kernfs_node *right)
 344{
 345	return kernfs_name_compare(left->hash, left->name, left->ns, right);
 346}
 347
 348/**
 349 *	kernfs_link_sibling - link kernfs_node into sibling rbtree
 350 *	@kn: kernfs_node of interest
 351 *
 352 *	Link @kn into its sibling rbtree which starts from
 353 *	@kn->parent->dir.children.
 354 *
 355 *	Locking:
 356 *	kernfs_rwsem held exclusive
 357 *
 358 *	Return:
 359 *	%0 on success, -EEXIST on failure.
 360 */
 361static int kernfs_link_sibling(struct kernfs_node *kn)
 362{
 363	struct rb_node **node = &kn->parent->dir.children.rb_node;
 364	struct rb_node *parent = NULL;
 365
 366	while (*node) {
 367		struct kernfs_node *pos;
 368		int result;
 369
 370		pos = rb_to_kn(*node);
 371		parent = *node;
 372		result = kernfs_sd_compare(kn, pos);
 373		if (result < 0)
 374			node = &pos->rb.rb_left;
 375		else if (result > 0)
 376			node = &pos->rb.rb_right;
 377		else
 378			return -EEXIST;
 379	}
 380
 381	/* add new node and rebalance the tree */
 382	rb_link_node(&kn->rb, parent, node);
 383	rb_insert_color(&kn->rb, &kn->parent->dir.children);
 384
 385	/* successfully added, account subdir number */
 386	down_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
 387	if (kernfs_type(kn) == KERNFS_DIR)
 388		kn->parent->dir.subdirs++;
 389	kernfs_inc_rev(kn->parent);
 390	up_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
 391
 392	return 0;
 393}
 394
 395/**
 396 *	kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
 397 *	@kn: kernfs_node of interest
 398 *
 399 *	Try to unlink @kn from its sibling rbtree which starts from
 400 *	kn->parent->dir.children.
 401 *
 402 *	Return: %true if @kn was actually removed,
 403 *	%false if @kn wasn't on the rbtree.
 404 *
 405 *	Locking:
 406 *	kernfs_rwsem held exclusive
 407 */
 408static bool kernfs_unlink_sibling(struct kernfs_node *kn)
 409{
 410	if (RB_EMPTY_NODE(&kn->rb))
 411		return false;
 412
 413	down_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
 414	if (kernfs_type(kn) == KERNFS_DIR)
 415		kn->parent->dir.subdirs--;
 416	kernfs_inc_rev(kn->parent);
 417	up_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
 418
 419	rb_erase(&kn->rb, &kn->parent->dir.children);
 420	RB_CLEAR_NODE(&kn->rb);
 421	return true;
 422}
 423
 424/**
 425 *	kernfs_get_active - get an active reference to kernfs_node
 426 *	@kn: kernfs_node to get an active reference to
 427 *
 428 *	Get an active reference of @kn.  This function is noop if @kn
 429 *	is %NULL.
 430 *
 431 *	Return:
 432 *	Pointer to @kn on success, %NULL on failure.
 433 */
 434struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
 435{
 436	if (unlikely(!kn))
 437		return NULL;
 438
 439	if (!atomic_inc_unless_negative(&kn->active))
 440		return NULL;
 441
 442	if (kernfs_lockdep(kn))
 443		rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
 444	return kn;
 445}
 446
 447/**
 448 *	kernfs_put_active - put an active reference to kernfs_node
 449 *	@kn: kernfs_node to put an active reference to
 450 *
 451 *	Put an active reference to @kn.  This function is noop if @kn
 452 *	is %NULL.
 453 */
 454void kernfs_put_active(struct kernfs_node *kn)
 455{
 456	int v;
 457
 458	if (unlikely(!kn))
 459		return;
 460
 461	if (kernfs_lockdep(kn))
 462		rwsem_release(&kn->dep_map, _RET_IP_);
 463	v = atomic_dec_return(&kn->active);
 464	if (likely(v != KN_DEACTIVATED_BIAS))
 465		return;
 466
 467	wake_up_all(&kernfs_root(kn)->deactivate_waitq);
 468}
 469
 470/**
 471 * kernfs_drain - drain kernfs_node
 472 * @kn: kernfs_node to drain
 473 *
 474 * Drain existing usages and nuke all existing mmaps of @kn.  Multiple
 475 * removers may invoke this function concurrently on @kn and all will
 476 * return after draining is complete.
 477 */
 478static void kernfs_drain(struct kernfs_node *kn)
 479	__releases(&kernfs_root(kn)->kernfs_rwsem)
 480	__acquires(&kernfs_root(kn)->kernfs_rwsem)
 481{
 482	struct kernfs_root *root = kernfs_root(kn);
 483
 484	lockdep_assert_held_write(&root->kernfs_rwsem);
 485	WARN_ON_ONCE(kernfs_active(kn));
 486
 487	/*
 488	 * Skip draining if already fully drained. This avoids draining and its
 489	 * lockdep annotations for nodes which have never been activated
 490	 * allowing embedding kernfs_remove() in create error paths without
 491	 * worrying about draining.
 492	 */
 493	if (atomic_read(&kn->active) == KN_DEACTIVATED_BIAS &&
 494	    !kernfs_should_drain_open_files(kn))
 495		return;
 496
 497	up_write(&root->kernfs_rwsem);
 498
 499	if (kernfs_lockdep(kn)) {
 500		rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
 501		if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
 502			lock_contended(&kn->dep_map, _RET_IP_);
 503	}
 504
 505	wait_event(root->deactivate_waitq,
 506		   atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
 507
 508	if (kernfs_lockdep(kn)) {
 509		lock_acquired(&kn->dep_map, _RET_IP_);
 510		rwsem_release(&kn->dep_map, _RET_IP_);
 511	}
 512
 513	if (kernfs_should_drain_open_files(kn))
 514		kernfs_drain_open_files(kn);
 515
 516	down_write(&root->kernfs_rwsem);
 517}
 518
 519/**
 520 * kernfs_get - get a reference count on a kernfs_node
 521 * @kn: the target kernfs_node
 522 */
 523void kernfs_get(struct kernfs_node *kn)
 524{
 525	if (kn) {
 526		WARN_ON(!atomic_read(&kn->count));
 527		atomic_inc(&kn->count);
 528	}
 529}
 530EXPORT_SYMBOL_GPL(kernfs_get);
 531
 532static void kernfs_free_rcu(struct rcu_head *rcu)
 533{
 534	struct kernfs_node *kn = container_of(rcu, struct kernfs_node, rcu);
 535
 536	kfree_const(kn->name);
 537
 538	if (kn->iattr) {
 539		simple_xattrs_free(&kn->iattr->xattrs, NULL);
 540		kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
 541	}
 542
 543	kmem_cache_free(kernfs_node_cache, kn);
 544}
 545
 546/**
 547 * kernfs_put - put a reference count on a kernfs_node
 548 * @kn: the target kernfs_node
 549 *
 550 * Put a reference count of @kn and destroy it if it reached zero.
 551 */
 552void kernfs_put(struct kernfs_node *kn)
 553{
 554	struct kernfs_node *parent;
 555	struct kernfs_root *root;
 556
 557	if (!kn || !atomic_dec_and_test(&kn->count))
 558		return;
 559	root = kernfs_root(kn);
 560 repeat:
 561	/*
 562	 * Moving/renaming is always done while holding reference.
 563	 * kn->parent won't change beneath us.
 564	 */
 565	parent = kn->parent;
 566
 567	WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
 568		  "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
 569		  parent ? parent->name : "", kn->name, atomic_read(&kn->active));
 570
 571	if (kernfs_type(kn) == KERNFS_LINK)
 572		kernfs_put(kn->symlink.target_kn);
 573
 
 
 
 
 
 
 574	spin_lock(&kernfs_idr_lock);
 575	idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
 576	spin_unlock(&kernfs_idr_lock);
 577
 578	call_rcu(&kn->rcu, kernfs_free_rcu);
 579
 580	kn = parent;
 581	if (kn) {
 582		if (atomic_dec_and_test(&kn->count))
 583			goto repeat;
 584	} else {
 585		/* just released the root kn, free @root too */
 586		idr_destroy(&root->ino_idr);
 587		kfree_rcu(root, rcu);
 588	}
 589}
 590EXPORT_SYMBOL_GPL(kernfs_put);
 591
 592/**
 593 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
 594 * @dentry: the dentry in question
 595 *
 596 * Return: the kernfs_node associated with @dentry.  If @dentry is not a
 597 * kernfs one, %NULL is returned.
 598 *
 599 * While the returned kernfs_node will stay accessible as long as @dentry
 600 * is accessible, the returned node can be in any state and the caller is
 601 * fully responsible for determining what's accessible.
 602 */
 603struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
 604{
 605	if (dentry->d_sb->s_op == &kernfs_sops)
 606		return kernfs_dentry_node(dentry);
 607	return NULL;
 608}
 609
 610static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
 611					     struct kernfs_node *parent,
 612					     const char *name, umode_t mode,
 613					     kuid_t uid, kgid_t gid,
 614					     unsigned flags)
 615{
 616	struct kernfs_node *kn;
 617	u32 id_highbits;
 618	int ret;
 619
 620	name = kstrdup_const(name, GFP_KERNEL);
 621	if (!name)
 622		return NULL;
 623
 624	kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
 625	if (!kn)
 626		goto err_out1;
 627
 628	idr_preload(GFP_KERNEL);
 629	spin_lock(&kernfs_idr_lock);
 630	ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
 631	if (ret >= 0 && ret < root->last_id_lowbits)
 632		root->id_highbits++;
 633	id_highbits = root->id_highbits;
 634	root->last_id_lowbits = ret;
 635	spin_unlock(&kernfs_idr_lock);
 636	idr_preload_end();
 637	if (ret < 0)
 638		goto err_out2;
 639
 640	kn->id = (u64)id_highbits << 32 | ret;
 641
 642	atomic_set(&kn->count, 1);
 643	atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
 644	RB_CLEAR_NODE(&kn->rb);
 645
 646	kn->name = name;
 647	kn->mode = mode;
 648	kn->flags = flags;
 649
 650	if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
 651		struct iattr iattr = {
 652			.ia_valid = ATTR_UID | ATTR_GID,
 653			.ia_uid = uid,
 654			.ia_gid = gid,
 655		};
 656
 657		ret = __kernfs_setattr(kn, &iattr);
 658		if (ret < 0)
 659			goto err_out3;
 660	}
 661
 662	if (parent) {
 663		ret = security_kernfs_init_security(parent, kn);
 664		if (ret)
 665			goto err_out3;
 666	}
 667
 668	return kn;
 669
 670 err_out3:
 671	spin_lock(&kernfs_idr_lock);
 672	idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
 673	spin_unlock(&kernfs_idr_lock);
 674 err_out2:
 675	kmem_cache_free(kernfs_node_cache, kn);
 676 err_out1:
 677	kfree_const(name);
 678	return NULL;
 679}
 680
 681struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
 682				    const char *name, umode_t mode,
 683				    kuid_t uid, kgid_t gid,
 684				    unsigned flags)
 685{
 686	struct kernfs_node *kn;
 687
 688	if (parent->mode & S_ISGID) {
 689		/* this code block imitates inode_init_owner() for
 690		 * kernfs
 691		 */
 692
 693		if (parent->iattr)
 694			gid = parent->iattr->ia_gid;
 695
 696		if (flags & KERNFS_DIR)
 697			mode |= S_ISGID;
 698	}
 699
 700	kn = __kernfs_new_node(kernfs_root(parent), parent,
 701			       name, mode, uid, gid, flags);
 702	if (kn) {
 703		kernfs_get(parent);
 704		kn->parent = parent;
 705	}
 706	return kn;
 707}
 708
 709/*
 710 * kernfs_find_and_get_node_by_id - get kernfs_node from node id
 711 * @root: the kernfs root
 712 * @id: the target node id
 713 *
 714 * @id's lower 32bits encode ino and upper gen.  If the gen portion is
 715 * zero, all generations are matched.
 716 *
 717 * Return: %NULL on failure,
 718 * otherwise a kernfs node with reference counter incremented.
 719 */
 720struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
 721						   u64 id)
 722{
 723	struct kernfs_node *kn;
 724	ino_t ino = kernfs_id_ino(id);
 725	u32 gen = kernfs_id_gen(id);
 726
 727	rcu_read_lock();
 728
 729	kn = idr_find(&root->ino_idr, (u32)ino);
 730	if (!kn)
 731		goto err_unlock;
 732
 733	if (sizeof(ino_t) >= sizeof(u64)) {
 734		/* we looked up with the low 32bits, compare the whole */
 735		if (kernfs_ino(kn) != ino)
 736			goto err_unlock;
 737	} else {
 738		/* 0 matches all generations */
 739		if (unlikely(gen && kernfs_gen(kn) != gen))
 740			goto err_unlock;
 741	}
 742
 743	/*
 744	 * We should fail if @kn has never been activated and guarantee success
 745	 * if the caller knows that @kn is active. Both can be achieved by
 746	 * __kernfs_active() which tests @kn->active without kernfs_rwsem.
 747	 */
 748	if (unlikely(!__kernfs_active(kn) || !atomic_inc_not_zero(&kn->count)))
 749		goto err_unlock;
 750
 751	rcu_read_unlock();
 752	return kn;
 753err_unlock:
 754	rcu_read_unlock();
 755	return NULL;
 756}
 757
 758/**
 759 *	kernfs_add_one - add kernfs_node to parent without warning
 760 *	@kn: kernfs_node to be added
 761 *
 762 *	The caller must already have initialized @kn->parent.  This
 763 *	function increments nlink of the parent's inode if @kn is a
 764 *	directory and link into the children list of the parent.
 765 *
 766 *	Return:
 767 *	%0 on success, -EEXIST if entry with the given name already
 768 *	exists.
 769 */
 770int kernfs_add_one(struct kernfs_node *kn)
 771{
 772	struct kernfs_node *parent = kn->parent;
 773	struct kernfs_root *root = kernfs_root(parent);
 774	struct kernfs_iattrs *ps_iattr;
 775	bool has_ns;
 776	int ret;
 777
 778	down_write(&root->kernfs_rwsem);
 779
 780	ret = -EINVAL;
 781	has_ns = kernfs_ns_enabled(parent);
 782	if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
 783		 has_ns ? "required" : "invalid", parent->name, kn->name))
 784		goto out_unlock;
 785
 786	if (kernfs_type(parent) != KERNFS_DIR)
 787		goto out_unlock;
 788
 789	ret = -ENOENT;
 790	if (parent->flags & (KERNFS_REMOVING | KERNFS_EMPTY_DIR))
 791		goto out_unlock;
 792
 793	kn->hash = kernfs_name_hash(kn->name, kn->ns);
 794
 795	ret = kernfs_link_sibling(kn);
 796	if (ret)
 797		goto out_unlock;
 798
 799	/* Update timestamps on the parent */
 800	down_write(&root->kernfs_iattr_rwsem);
 801
 802	ps_iattr = parent->iattr;
 803	if (ps_iattr) {
 804		ktime_get_real_ts64(&ps_iattr->ia_ctime);
 805		ps_iattr->ia_mtime = ps_iattr->ia_ctime;
 806	}
 807
 808	up_write(&root->kernfs_iattr_rwsem);
 809	up_write(&root->kernfs_rwsem);
 810
 811	/*
 812	 * Activate the new node unless CREATE_DEACTIVATED is requested.
 813	 * If not activated here, the kernfs user is responsible for
 814	 * activating the node with kernfs_activate().  A node which hasn't
 815	 * been activated is not visible to userland and its removal won't
 816	 * trigger deactivation.
 817	 */
 818	if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
 819		kernfs_activate(kn);
 820	return 0;
 821
 822out_unlock:
 823	up_write(&root->kernfs_rwsem);
 824	return ret;
 825}
 826
 827/**
 828 * kernfs_find_ns - find kernfs_node with the given name
 829 * @parent: kernfs_node to search under
 830 * @name: name to look for
 831 * @ns: the namespace tag to use
 832 *
 833 * Look for kernfs_node with name @name under @parent.
 834 *
 835 * Return: pointer to the found kernfs_node on success, %NULL on failure.
 836 */
 837static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
 838					  const unsigned char *name,
 839					  const void *ns)
 840{
 841	struct rb_node *node = parent->dir.children.rb_node;
 842	bool has_ns = kernfs_ns_enabled(parent);
 843	unsigned int hash;
 844
 845	lockdep_assert_held(&kernfs_root(parent)->kernfs_rwsem);
 846
 847	if (has_ns != (bool)ns) {
 848		WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
 849		     has_ns ? "required" : "invalid", parent->name, name);
 850		return NULL;
 851	}
 852
 853	hash = kernfs_name_hash(name, ns);
 854	while (node) {
 855		struct kernfs_node *kn;
 856		int result;
 857
 858		kn = rb_to_kn(node);
 859		result = kernfs_name_compare(hash, name, ns, kn);
 860		if (result < 0)
 861			node = node->rb_left;
 862		else if (result > 0)
 863			node = node->rb_right;
 864		else
 865			return kn;
 866	}
 867	return NULL;
 868}
 869
 870static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
 871					  const unsigned char *path,
 872					  const void *ns)
 873{
 874	ssize_t len;
 875	char *p, *name;
 876
 877	lockdep_assert_held_read(&kernfs_root(parent)->kernfs_rwsem);
 878
 879	spin_lock_irq(&kernfs_pr_cont_lock);
 880
 881	len = strscpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
 882
 883	if (len < 0) {
 884		spin_unlock_irq(&kernfs_pr_cont_lock);
 885		return NULL;
 886	}
 887
 888	p = kernfs_pr_cont_buf;
 889
 890	while ((name = strsep(&p, "/")) && parent) {
 891		if (*name == '\0')
 892			continue;
 893		parent = kernfs_find_ns(parent, name, ns);
 894	}
 895
 896	spin_unlock_irq(&kernfs_pr_cont_lock);
 897
 898	return parent;
 899}
 900
 901/**
 902 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
 903 * @parent: kernfs_node to search under
 904 * @name: name to look for
 905 * @ns: the namespace tag to use
 906 *
 907 * Look for kernfs_node with name @name under @parent and get a reference
 908 * if found.  This function may sleep.
 909 *
 910 * Return: pointer to the found kernfs_node on success, %NULL on failure.
 911 */
 912struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
 913					   const char *name, const void *ns)
 914{
 915	struct kernfs_node *kn;
 916	struct kernfs_root *root = kernfs_root(parent);
 917
 918	down_read(&root->kernfs_rwsem);
 919	kn = kernfs_find_ns(parent, name, ns);
 920	kernfs_get(kn);
 921	up_read(&root->kernfs_rwsem);
 922
 923	return kn;
 924}
 925EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
 926
 927/**
 928 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
 929 * @parent: kernfs_node to search under
 930 * @path: path to look for
 931 * @ns: the namespace tag to use
 932 *
 933 * Look for kernfs_node with path @path under @parent and get a reference
 934 * if found.  This function may sleep.
 935 *
 936 * Return: pointer to the found kernfs_node on success, %NULL on failure.
 937 */
 938struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
 939					   const char *path, const void *ns)
 940{
 941	struct kernfs_node *kn;
 942	struct kernfs_root *root = kernfs_root(parent);
 943
 944	down_read(&root->kernfs_rwsem);
 945	kn = kernfs_walk_ns(parent, path, ns);
 946	kernfs_get(kn);
 947	up_read(&root->kernfs_rwsem);
 948
 949	return kn;
 950}
 951
 952/**
 953 * kernfs_create_root - create a new kernfs hierarchy
 954 * @scops: optional syscall operations for the hierarchy
 955 * @flags: KERNFS_ROOT_* flags
 956 * @priv: opaque data associated with the new directory
 957 *
 958 * Return: the root of the new hierarchy on success, ERR_PTR() value on
 959 * failure.
 960 */
 961struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
 962				       unsigned int flags, void *priv)
 963{
 964	struct kernfs_root *root;
 965	struct kernfs_node *kn;
 966
 967	root = kzalloc(sizeof(*root), GFP_KERNEL);
 968	if (!root)
 969		return ERR_PTR(-ENOMEM);
 970
 971	idr_init(&root->ino_idr);
 972	init_rwsem(&root->kernfs_rwsem);
 973	init_rwsem(&root->kernfs_iattr_rwsem);
 974	init_rwsem(&root->kernfs_supers_rwsem);
 975	INIT_LIST_HEAD(&root->supers);
 976
 977	/*
 978	 * On 64bit ino setups, id is ino.  On 32bit, low 32bits are ino.
 979	 * High bits generation.  The starting value for both ino and
 980	 * genenration is 1.  Initialize upper 32bit allocation
 981	 * accordingly.
 982	 */
 983	if (sizeof(ino_t) >= sizeof(u64))
 984		root->id_highbits = 0;
 985	else
 986		root->id_highbits = 1;
 987
 988	kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
 989			       GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
 990			       KERNFS_DIR);
 991	if (!kn) {
 992		idr_destroy(&root->ino_idr);
 993		kfree(root);
 994		return ERR_PTR(-ENOMEM);
 995	}
 996
 997	kn->priv = priv;
 998	kn->dir.root = root;
 999
1000	root->syscall_ops = scops;
1001	root->flags = flags;
1002	root->kn = kn;
1003	init_waitqueue_head(&root->deactivate_waitq);
1004
1005	if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
1006		kernfs_activate(kn);
1007
1008	return root;
1009}
1010
1011/**
1012 * kernfs_destroy_root - destroy a kernfs hierarchy
1013 * @root: root of the hierarchy to destroy
1014 *
1015 * Destroy the hierarchy anchored at @root by removing all existing
1016 * directories and destroying @root.
1017 */
1018void kernfs_destroy_root(struct kernfs_root *root)
1019{
1020	/*
1021	 *  kernfs_remove holds kernfs_rwsem from the root so the root
1022	 *  shouldn't be freed during the operation.
1023	 */
1024	kernfs_get(root->kn);
1025	kernfs_remove(root->kn);
1026	kernfs_put(root->kn); /* will also free @root */
1027}
1028
1029/**
1030 * kernfs_root_to_node - return the kernfs_node associated with a kernfs_root
1031 * @root: root to use to lookup
1032 *
1033 * Return: @root's kernfs_node
1034 */
1035struct kernfs_node *kernfs_root_to_node(struct kernfs_root *root)
1036{
1037	return root->kn;
1038}
1039
1040/**
1041 * kernfs_create_dir_ns - create a directory
1042 * @parent: parent in which to create a new directory
1043 * @name: name of the new directory
1044 * @mode: mode of the new directory
1045 * @uid: uid of the new directory
1046 * @gid: gid of the new directory
1047 * @priv: opaque data associated with the new directory
1048 * @ns: optional namespace tag of the directory
1049 *
1050 * Return: the created node on success, ERR_PTR() value on failure.
1051 */
1052struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1053					 const char *name, umode_t mode,
1054					 kuid_t uid, kgid_t gid,
1055					 void *priv, const void *ns)
1056{
1057	struct kernfs_node *kn;
1058	int rc;
1059
1060	/* allocate */
1061	kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1062			     uid, gid, KERNFS_DIR);
1063	if (!kn)
1064		return ERR_PTR(-ENOMEM);
1065
1066	kn->dir.root = parent->dir.root;
1067	kn->ns = ns;
1068	kn->priv = priv;
1069
1070	/* link in */
1071	rc = kernfs_add_one(kn);
1072	if (!rc)
1073		return kn;
1074
1075	kernfs_put(kn);
1076	return ERR_PTR(rc);
1077}
1078
1079/**
1080 * kernfs_create_empty_dir - create an always empty directory
1081 * @parent: parent in which to create a new directory
1082 * @name: name of the new directory
1083 *
1084 * Return: the created node on success, ERR_PTR() value on failure.
1085 */
1086struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1087					    const char *name)
1088{
1089	struct kernfs_node *kn;
1090	int rc;
1091
1092	/* allocate */
1093	kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1094			     GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1095	if (!kn)
1096		return ERR_PTR(-ENOMEM);
1097
1098	kn->flags |= KERNFS_EMPTY_DIR;
1099	kn->dir.root = parent->dir.root;
1100	kn->ns = NULL;
1101	kn->priv = NULL;
1102
1103	/* link in */
1104	rc = kernfs_add_one(kn);
1105	if (!rc)
1106		return kn;
1107
1108	kernfs_put(kn);
1109	return ERR_PTR(rc);
1110}
1111
1112static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
1113{
1114	struct kernfs_node *kn;
1115	struct kernfs_root *root;
1116
1117	if (flags & LOOKUP_RCU)
1118		return -ECHILD;
1119
1120	/* Negative hashed dentry? */
1121	if (d_really_is_negative(dentry)) {
1122		struct kernfs_node *parent;
1123
1124		/* If the kernfs parent node has changed discard and
1125		 * proceed to ->lookup.
1126		 *
1127		 * There's nothing special needed here when getting the
1128		 * dentry parent, even if a concurrent rename is in
1129		 * progress. That's because the dentry is negative so
1130		 * it can only be the target of the rename and it will
1131		 * be doing a d_move() not a replace. Consequently the
1132		 * dentry d_parent won't change over the d_move().
1133		 *
1134		 * Also kernfs negative dentries transitioning from
1135		 * negative to positive during revalidate won't happen
1136		 * because they are invalidated on containing directory
1137		 * changes and the lookup re-done so that a new positive
1138		 * dentry can be properly created.
1139		 */
1140		root = kernfs_root_from_sb(dentry->d_sb);
1141		down_read(&root->kernfs_rwsem);
1142		parent = kernfs_dentry_node(dentry->d_parent);
1143		if (parent) {
1144			if (kernfs_dir_changed(parent, dentry)) {
1145				up_read(&root->kernfs_rwsem);
1146				return 0;
1147			}
1148		}
1149		up_read(&root->kernfs_rwsem);
1150
1151		/* The kernfs parent node hasn't changed, leave the
1152		 * dentry negative and return success.
1153		 */
1154		return 1;
1155	}
1156
1157	kn = kernfs_dentry_node(dentry);
1158	root = kernfs_root(kn);
1159	down_read(&root->kernfs_rwsem);
1160
1161	/* The kernfs node has been deactivated */
1162	if (!kernfs_active(kn))
1163		goto out_bad;
1164
1165	/* The kernfs node has been moved? */
1166	if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
1167		goto out_bad;
1168
1169	/* The kernfs node has been renamed */
1170	if (strcmp(dentry->d_name.name, kn->name) != 0)
1171		goto out_bad;
1172
1173	/* The kernfs node has been moved to a different namespace */
1174	if (kn->parent && kernfs_ns_enabled(kn->parent) &&
1175	    kernfs_info(dentry->d_sb)->ns != kn->ns)
1176		goto out_bad;
1177
1178	up_read(&root->kernfs_rwsem);
1179	return 1;
1180out_bad:
1181	up_read(&root->kernfs_rwsem);
1182	return 0;
1183}
1184
1185const struct dentry_operations kernfs_dops = {
1186	.d_revalidate	= kernfs_dop_revalidate,
1187};
1188
1189static struct dentry *kernfs_iop_lookup(struct inode *dir,
1190					struct dentry *dentry,
1191					unsigned int flags)
1192{
1193	struct kernfs_node *parent = dir->i_private;
1194	struct kernfs_node *kn;
1195	struct kernfs_root *root;
1196	struct inode *inode = NULL;
1197	const void *ns = NULL;
1198
1199	root = kernfs_root(parent);
1200	down_read(&root->kernfs_rwsem);
1201	if (kernfs_ns_enabled(parent))
1202		ns = kernfs_info(dir->i_sb)->ns;
1203
1204	kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1205	/* attach dentry and inode */
1206	if (kn) {
1207		/* Inactive nodes are invisible to the VFS so don't
1208		 * create a negative.
1209		 */
1210		if (!kernfs_active(kn)) {
1211			up_read(&root->kernfs_rwsem);
1212			return NULL;
1213		}
1214		inode = kernfs_get_inode(dir->i_sb, kn);
1215		if (!inode)
1216			inode = ERR_PTR(-ENOMEM);
1217	}
1218	/*
1219	 * Needed for negative dentry validation.
1220	 * The negative dentry can be created in kernfs_iop_lookup()
1221	 * or transforms from positive dentry in dentry_unlink_inode()
1222	 * called from vfs_rmdir().
1223	 */
1224	if (!IS_ERR(inode))
1225		kernfs_set_rev(parent, dentry);
1226	up_read(&root->kernfs_rwsem);
1227
1228	/* instantiate and hash (possibly negative) dentry */
1229	return d_splice_alias(inode, dentry);
1230}
1231
1232static int kernfs_iop_mkdir(struct mnt_idmap *idmap,
1233			    struct inode *dir, struct dentry *dentry,
1234			    umode_t mode)
1235{
1236	struct kernfs_node *parent = dir->i_private;
1237	struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1238	int ret;
1239
1240	if (!scops || !scops->mkdir)
1241		return -EPERM;
1242
1243	if (!kernfs_get_active(parent))
1244		return -ENODEV;
1245
1246	ret = scops->mkdir(parent, dentry->d_name.name, mode);
1247
1248	kernfs_put_active(parent);
1249	return ret;
1250}
1251
1252static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1253{
1254	struct kernfs_node *kn  = kernfs_dentry_node(dentry);
1255	struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1256	int ret;
1257
1258	if (!scops || !scops->rmdir)
1259		return -EPERM;
1260
1261	if (!kernfs_get_active(kn))
1262		return -ENODEV;
1263
1264	ret = scops->rmdir(kn);
1265
1266	kernfs_put_active(kn);
1267	return ret;
1268}
1269
1270static int kernfs_iop_rename(struct mnt_idmap *idmap,
1271			     struct inode *old_dir, struct dentry *old_dentry,
1272			     struct inode *new_dir, struct dentry *new_dentry,
1273			     unsigned int flags)
1274{
1275	struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1276	struct kernfs_node *new_parent = new_dir->i_private;
1277	struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1278	int ret;
1279
1280	if (flags)
1281		return -EINVAL;
1282
1283	if (!scops || !scops->rename)
1284		return -EPERM;
1285
1286	if (!kernfs_get_active(kn))
1287		return -ENODEV;
1288
1289	if (!kernfs_get_active(new_parent)) {
1290		kernfs_put_active(kn);
1291		return -ENODEV;
1292	}
1293
1294	ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1295
1296	kernfs_put_active(new_parent);
1297	kernfs_put_active(kn);
1298	return ret;
1299}
1300
1301const struct inode_operations kernfs_dir_iops = {
1302	.lookup		= kernfs_iop_lookup,
1303	.permission	= kernfs_iop_permission,
1304	.setattr	= kernfs_iop_setattr,
1305	.getattr	= kernfs_iop_getattr,
1306	.listxattr	= kernfs_iop_listxattr,
1307
1308	.mkdir		= kernfs_iop_mkdir,
1309	.rmdir		= kernfs_iop_rmdir,
1310	.rename		= kernfs_iop_rename,
1311};
1312
1313static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1314{
1315	struct kernfs_node *last;
1316
1317	while (true) {
1318		struct rb_node *rbn;
1319
1320		last = pos;
1321
1322		if (kernfs_type(pos) != KERNFS_DIR)
1323			break;
1324
1325		rbn = rb_first(&pos->dir.children);
1326		if (!rbn)
1327			break;
1328
1329		pos = rb_to_kn(rbn);
1330	}
1331
1332	return last;
1333}
1334
1335/**
1336 * kernfs_next_descendant_post - find the next descendant for post-order walk
1337 * @pos: the current position (%NULL to initiate traversal)
1338 * @root: kernfs_node whose descendants to walk
1339 *
1340 * Find the next descendant to visit for post-order traversal of @root's
1341 * descendants.  @root is included in the iteration and the last node to be
1342 * visited.
1343 *
1344 * Return: the next descendant to visit or %NULL when done.
1345 */
1346static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1347						       struct kernfs_node *root)
1348{
1349	struct rb_node *rbn;
1350
1351	lockdep_assert_held_write(&kernfs_root(root)->kernfs_rwsem);
1352
1353	/* if first iteration, visit leftmost descendant which may be root */
1354	if (!pos)
1355		return kernfs_leftmost_descendant(root);
1356
1357	/* if we visited @root, we're done */
1358	if (pos == root)
1359		return NULL;
1360
1361	/* if there's an unvisited sibling, visit its leftmost descendant */
1362	rbn = rb_next(&pos->rb);
1363	if (rbn)
1364		return kernfs_leftmost_descendant(rb_to_kn(rbn));
1365
1366	/* no sibling left, visit parent */
1367	return pos->parent;
1368}
1369
1370static void kernfs_activate_one(struct kernfs_node *kn)
1371{
1372	lockdep_assert_held_write(&kernfs_root(kn)->kernfs_rwsem);
1373
1374	kn->flags |= KERNFS_ACTIVATED;
1375
1376	if (kernfs_active(kn) || (kn->flags & (KERNFS_HIDDEN | KERNFS_REMOVING)))
1377		return;
1378
1379	WARN_ON_ONCE(kn->parent && RB_EMPTY_NODE(&kn->rb));
1380	WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1381
1382	atomic_sub(KN_DEACTIVATED_BIAS, &kn->active);
1383}
1384
1385/**
1386 * kernfs_activate - activate a node which started deactivated
1387 * @kn: kernfs_node whose subtree is to be activated
1388 *
1389 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1390 * needs to be explicitly activated.  A node which hasn't been activated
1391 * isn't visible to userland and deactivation is skipped during its
1392 * removal.  This is useful to construct atomic init sequences where
1393 * creation of multiple nodes should either succeed or fail atomically.
1394 *
1395 * The caller is responsible for ensuring that this function is not called
1396 * after kernfs_remove*() is invoked on @kn.
1397 */
1398void kernfs_activate(struct kernfs_node *kn)
1399{
1400	struct kernfs_node *pos;
1401	struct kernfs_root *root = kernfs_root(kn);
1402
1403	down_write(&root->kernfs_rwsem);
1404
1405	pos = NULL;
1406	while ((pos = kernfs_next_descendant_post(pos, kn)))
1407		kernfs_activate_one(pos);
1408
1409	up_write(&root->kernfs_rwsem);
1410}
1411
1412/**
1413 * kernfs_show - show or hide a node
1414 * @kn: kernfs_node to show or hide
1415 * @show: whether to show or hide
1416 *
1417 * If @show is %false, @kn is marked hidden and deactivated. A hidden node is
1418 * ignored in future activaitons. If %true, the mark is removed and activation
1419 * state is restored. This function won't implicitly activate a new node in a
1420 * %KERNFS_ROOT_CREATE_DEACTIVATED root which hasn't been activated yet.
1421 *
1422 * To avoid recursion complexities, directories aren't supported for now.
1423 */
1424void kernfs_show(struct kernfs_node *kn, bool show)
1425{
1426	struct kernfs_root *root = kernfs_root(kn);
1427
1428	if (WARN_ON_ONCE(kernfs_type(kn) == KERNFS_DIR))
1429		return;
1430
1431	down_write(&root->kernfs_rwsem);
1432
1433	if (show) {
1434		kn->flags &= ~KERNFS_HIDDEN;
1435		if (kn->flags & KERNFS_ACTIVATED)
1436			kernfs_activate_one(kn);
1437	} else {
1438		kn->flags |= KERNFS_HIDDEN;
1439		if (kernfs_active(kn))
1440			atomic_add(KN_DEACTIVATED_BIAS, &kn->active);
1441		kernfs_drain(kn);
1442	}
1443
1444	up_write(&root->kernfs_rwsem);
1445}
1446
1447static void __kernfs_remove(struct kernfs_node *kn)
1448{
1449	struct kernfs_node *pos;
1450
1451	/* Short-circuit if non-root @kn has already finished removal. */
1452	if (!kn)
1453		return;
1454
1455	lockdep_assert_held_write(&kernfs_root(kn)->kernfs_rwsem);
1456
1457	/*
1458	 * This is for kernfs_remove_self() which plays with active ref
1459	 * after removal.
1460	 */
1461	if (kn->parent && RB_EMPTY_NODE(&kn->rb))
1462		return;
1463
1464	pr_debug("kernfs %s: removing\n", kn->name);
1465
1466	/* prevent new usage by marking all nodes removing and deactivating */
1467	pos = NULL;
1468	while ((pos = kernfs_next_descendant_post(pos, kn))) {
1469		pos->flags |= KERNFS_REMOVING;
1470		if (kernfs_active(pos))
1471			atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1472	}
1473
1474	/* deactivate and unlink the subtree node-by-node */
1475	do {
1476		pos = kernfs_leftmost_descendant(kn);
1477
1478		/*
1479		 * kernfs_drain() may drop kernfs_rwsem temporarily and @pos's
1480		 * base ref could have been put by someone else by the time
1481		 * the function returns.  Make sure it doesn't go away
1482		 * underneath us.
1483		 */
1484		kernfs_get(pos);
1485
1486		kernfs_drain(pos);
1487
1488		/*
1489		 * kernfs_unlink_sibling() succeeds once per node.  Use it
1490		 * to decide who's responsible for cleanups.
1491		 */
1492		if (!pos->parent || kernfs_unlink_sibling(pos)) {
1493			struct kernfs_iattrs *ps_iattr =
1494				pos->parent ? pos->parent->iattr : NULL;
1495
1496			/* update timestamps on the parent */
1497			down_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
1498
1499			if (ps_iattr) {
1500				ktime_get_real_ts64(&ps_iattr->ia_ctime);
1501				ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1502			}
1503
1504			up_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
1505			kernfs_put(pos);
1506		}
1507
1508		kernfs_put(pos);
1509	} while (pos != kn);
1510}
1511
1512/**
1513 * kernfs_remove - remove a kernfs_node recursively
1514 * @kn: the kernfs_node to remove
1515 *
1516 * Remove @kn along with all its subdirectories and files.
1517 */
1518void kernfs_remove(struct kernfs_node *kn)
1519{
1520	struct kernfs_root *root;
1521
1522	if (!kn)
1523		return;
1524
1525	root = kernfs_root(kn);
1526
1527	down_write(&root->kernfs_rwsem);
1528	__kernfs_remove(kn);
1529	up_write(&root->kernfs_rwsem);
1530}
1531
1532/**
1533 * kernfs_break_active_protection - break out of active protection
1534 * @kn: the self kernfs_node
1535 *
1536 * The caller must be running off of a kernfs operation which is invoked
1537 * with an active reference - e.g. one of kernfs_ops.  Each invocation of
1538 * this function must also be matched with an invocation of
1539 * kernfs_unbreak_active_protection().
1540 *
1541 * This function releases the active reference of @kn the caller is
1542 * holding.  Once this function is called, @kn may be removed at any point
1543 * and the caller is solely responsible for ensuring that the objects it
1544 * dereferences are accessible.
1545 */
1546void kernfs_break_active_protection(struct kernfs_node *kn)
1547{
1548	/*
1549	 * Take out ourself out of the active ref dependency chain.  If
1550	 * we're called without an active ref, lockdep will complain.
1551	 */
1552	kernfs_put_active(kn);
1553}
1554
1555/**
1556 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1557 * @kn: the self kernfs_node
1558 *
1559 * If kernfs_break_active_protection() was called, this function must be
1560 * invoked before finishing the kernfs operation.  Note that while this
1561 * function restores the active reference, it doesn't and can't actually
1562 * restore the active protection - @kn may already or be in the process of
1563 * being removed.  Once kernfs_break_active_protection() is invoked, that
1564 * protection is irreversibly gone for the kernfs operation instance.
1565 *
1566 * While this function may be called at any point after
1567 * kernfs_break_active_protection() is invoked, its most useful location
1568 * would be right before the enclosing kernfs operation returns.
1569 */
1570void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1571{
1572	/*
1573	 * @kn->active could be in any state; however, the increment we do
1574	 * here will be undone as soon as the enclosing kernfs operation
1575	 * finishes and this temporary bump can't break anything.  If @kn
1576	 * is alive, nothing changes.  If @kn is being deactivated, the
1577	 * soon-to-follow put will either finish deactivation or restore
1578	 * deactivated state.  If @kn is already removed, the temporary
1579	 * bump is guaranteed to be gone before @kn is released.
1580	 */
1581	atomic_inc(&kn->active);
1582	if (kernfs_lockdep(kn))
1583		rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1584}
1585
1586/**
1587 * kernfs_remove_self - remove a kernfs_node from its own method
1588 * @kn: the self kernfs_node to remove
1589 *
1590 * The caller must be running off of a kernfs operation which is invoked
1591 * with an active reference - e.g. one of kernfs_ops.  This can be used to
1592 * implement a file operation which deletes itself.
1593 *
1594 * For example, the "delete" file for a sysfs device directory can be
1595 * implemented by invoking kernfs_remove_self() on the "delete" file
1596 * itself.  This function breaks the circular dependency of trying to
1597 * deactivate self while holding an active ref itself.  It isn't necessary
1598 * to modify the usual removal path to use kernfs_remove_self().  The
1599 * "delete" implementation can simply invoke kernfs_remove_self() on self
1600 * before proceeding with the usual removal path.  kernfs will ignore later
1601 * kernfs_remove() on self.
1602 *
1603 * kernfs_remove_self() can be called multiple times concurrently on the
1604 * same kernfs_node.  Only the first one actually performs removal and
1605 * returns %true.  All others will wait until the kernfs operation which
1606 * won self-removal finishes and return %false.  Note that the losers wait
1607 * for the completion of not only the winning kernfs_remove_self() but also
1608 * the whole kernfs_ops which won the arbitration.  This can be used to
1609 * guarantee, for example, all concurrent writes to a "delete" file to
1610 * finish only after the whole operation is complete.
1611 *
1612 * Return: %true if @kn is removed by this call, otherwise %false.
1613 */
1614bool kernfs_remove_self(struct kernfs_node *kn)
1615{
1616	bool ret;
1617	struct kernfs_root *root = kernfs_root(kn);
1618
1619	down_write(&root->kernfs_rwsem);
1620	kernfs_break_active_protection(kn);
1621
1622	/*
1623	 * SUICIDAL is used to arbitrate among competing invocations.  Only
1624	 * the first one will actually perform removal.  When the removal
1625	 * is complete, SUICIDED is set and the active ref is restored
1626	 * while kernfs_rwsem for held exclusive.  The ones which lost
1627	 * arbitration waits for SUICIDED && drained which can happen only
1628	 * after the enclosing kernfs operation which executed the winning
1629	 * instance of kernfs_remove_self() finished.
1630	 */
1631	if (!(kn->flags & KERNFS_SUICIDAL)) {
1632		kn->flags |= KERNFS_SUICIDAL;
1633		__kernfs_remove(kn);
1634		kn->flags |= KERNFS_SUICIDED;
1635		ret = true;
1636	} else {
1637		wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1638		DEFINE_WAIT(wait);
1639
1640		while (true) {
1641			prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1642
1643			if ((kn->flags & KERNFS_SUICIDED) &&
1644			    atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1645				break;
1646
1647			up_write(&root->kernfs_rwsem);
1648			schedule();
1649			down_write(&root->kernfs_rwsem);
1650		}
1651		finish_wait(waitq, &wait);
1652		WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1653		ret = false;
1654	}
1655
1656	/*
1657	 * This must be done while kernfs_rwsem held exclusive; otherwise,
1658	 * waiting for SUICIDED && deactivated could finish prematurely.
1659	 */
1660	kernfs_unbreak_active_protection(kn);
1661
1662	up_write(&root->kernfs_rwsem);
1663	return ret;
1664}
1665
1666/**
1667 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1668 * @parent: parent of the target
1669 * @name: name of the kernfs_node to remove
1670 * @ns: namespace tag of the kernfs_node to remove
1671 *
1672 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1673 *
1674 * Return: %0 on success, -ENOENT if such entry doesn't exist.
1675 */
1676int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1677			     const void *ns)
1678{
1679	struct kernfs_node *kn;
1680	struct kernfs_root *root;
1681
1682	if (!parent) {
1683		WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1684			name);
1685		return -ENOENT;
1686	}
1687
1688	root = kernfs_root(parent);
1689	down_write(&root->kernfs_rwsem);
1690
1691	kn = kernfs_find_ns(parent, name, ns);
1692	if (kn) {
1693		kernfs_get(kn);
1694		__kernfs_remove(kn);
1695		kernfs_put(kn);
1696	}
1697
1698	up_write(&root->kernfs_rwsem);
1699
1700	if (kn)
1701		return 0;
1702	else
1703		return -ENOENT;
1704}
1705
1706/**
1707 * kernfs_rename_ns - move and rename a kernfs_node
1708 * @kn: target node
1709 * @new_parent: new parent to put @sd under
1710 * @new_name: new name
1711 * @new_ns: new namespace tag
1712 *
1713 * Return: %0 on success, -errno on failure.
1714 */
1715int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1716		     const char *new_name, const void *new_ns)
1717{
1718	struct kernfs_node *old_parent;
1719	struct kernfs_root *root;
1720	const char *old_name = NULL;
1721	int error;
1722
1723	/* can't move or rename root */
1724	if (!kn->parent)
1725		return -EINVAL;
1726
1727	root = kernfs_root(kn);
1728	down_write(&root->kernfs_rwsem);
1729
1730	error = -ENOENT;
1731	if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1732	    (new_parent->flags & KERNFS_EMPTY_DIR))
1733		goto out;
1734
1735	error = 0;
1736	if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1737	    (strcmp(kn->name, new_name) == 0))
1738		goto out;	/* nothing to rename */
1739
1740	error = -EEXIST;
1741	if (kernfs_find_ns(new_parent, new_name, new_ns))
1742		goto out;
1743
1744	/* rename kernfs_node */
1745	if (strcmp(kn->name, new_name) != 0) {
1746		error = -ENOMEM;
1747		new_name = kstrdup_const(new_name, GFP_KERNEL);
1748		if (!new_name)
1749			goto out;
1750	} else {
1751		new_name = NULL;
1752	}
1753
1754	/*
1755	 * Move to the appropriate place in the appropriate directories rbtree.
1756	 */
1757	kernfs_unlink_sibling(kn);
1758	kernfs_get(new_parent);
1759
1760	/* rename_lock protects ->parent and ->name accessors */
1761	write_lock_irq(&kernfs_rename_lock);
1762
1763	old_parent = kn->parent;
1764	kn->parent = new_parent;
1765
1766	kn->ns = new_ns;
1767	if (new_name) {
1768		old_name = kn->name;
1769		kn->name = new_name;
1770	}
1771
1772	write_unlock_irq(&kernfs_rename_lock);
1773
1774	kn->hash = kernfs_name_hash(kn->name, kn->ns);
1775	kernfs_link_sibling(kn);
1776
1777	kernfs_put(old_parent);
1778	kfree_const(old_name);
1779
1780	error = 0;
1781 out:
1782	up_write(&root->kernfs_rwsem);
1783	return error;
1784}
1785
1786static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1787{
1788	kernfs_put(filp->private_data);
1789	return 0;
1790}
1791
1792static struct kernfs_node *kernfs_dir_pos(const void *ns,
1793	struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1794{
1795	if (pos) {
1796		int valid = kernfs_active(pos) &&
1797			pos->parent == parent && hash == pos->hash;
1798		kernfs_put(pos);
1799		if (!valid)
1800			pos = NULL;
1801	}
1802	if (!pos && (hash > 1) && (hash < INT_MAX)) {
1803		struct rb_node *node = parent->dir.children.rb_node;
1804		while (node) {
1805			pos = rb_to_kn(node);
1806
1807			if (hash < pos->hash)
1808				node = node->rb_left;
1809			else if (hash > pos->hash)
1810				node = node->rb_right;
1811			else
1812				break;
1813		}
1814	}
1815	/* Skip over entries which are dying/dead or in the wrong namespace */
1816	while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1817		struct rb_node *node = rb_next(&pos->rb);
1818		if (!node)
1819			pos = NULL;
1820		else
1821			pos = rb_to_kn(node);
1822	}
1823	return pos;
1824}
1825
1826static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1827	struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1828{
1829	pos = kernfs_dir_pos(ns, parent, ino, pos);
1830	if (pos) {
1831		do {
1832			struct rb_node *node = rb_next(&pos->rb);
1833			if (!node)
1834				pos = NULL;
1835			else
1836				pos = rb_to_kn(node);
1837		} while (pos && (!kernfs_active(pos) || pos->ns != ns));
1838	}
1839	return pos;
1840}
1841
1842static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1843{
1844	struct dentry *dentry = file->f_path.dentry;
1845	struct kernfs_node *parent = kernfs_dentry_node(dentry);
1846	struct kernfs_node *pos = file->private_data;
1847	struct kernfs_root *root;
1848	const void *ns = NULL;
1849
1850	if (!dir_emit_dots(file, ctx))
1851		return 0;
1852
1853	root = kernfs_root(parent);
1854	down_read(&root->kernfs_rwsem);
1855
1856	if (kernfs_ns_enabled(parent))
1857		ns = kernfs_info(dentry->d_sb)->ns;
1858
1859	for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1860	     pos;
1861	     pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1862		const char *name = pos->name;
1863		unsigned int type = fs_umode_to_dtype(pos->mode);
1864		int len = strlen(name);
1865		ino_t ino = kernfs_ino(pos);
1866
1867		ctx->pos = pos->hash;
1868		file->private_data = pos;
1869		kernfs_get(pos);
1870
1871		up_read(&root->kernfs_rwsem);
1872		if (!dir_emit(ctx, name, len, ino, type))
1873			return 0;
1874		down_read(&root->kernfs_rwsem);
1875	}
1876	up_read(&root->kernfs_rwsem);
1877	file->private_data = NULL;
1878	ctx->pos = INT_MAX;
1879	return 0;
1880}
1881
1882const struct file_operations kernfs_dir_fops = {
1883	.read		= generic_read_dir,
1884	.iterate_shared	= kernfs_fop_readdir,
1885	.release	= kernfs_dir_fop_release,
1886	.llseek		= generic_file_llseek,
1887};