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