Linux Audio

Check our new training course

Loading...
v4.10.11
 
   1/*
   2 * fs/kernfs/dir.c - kernfs directory implementation
   3 *
   4 * Copyright (c) 2001-3 Patrick Mochel
   5 * Copyright (c) 2007 SUSE Linux Products GmbH
   6 * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org>
   7 *
   8 * This file is released under the GPLv2.
   9 */
  10
  11#include <linux/sched.h>
  12#include <linux/fs.h>
  13#include <linux/namei.h>
  14#include <linux/idr.h>
  15#include <linux/slab.h>
  16#include <linux/security.h>
  17#include <linux/hash.h>
  18
  19#include "kernfs-internal.h"
  20
  21DEFINE_MUTEX(kernfs_mutex);
  22static DEFINE_SPINLOCK(kernfs_rename_lock);	/* kn->parent and ->name */
  23static char kernfs_pr_cont_buf[PATH_MAX];	/* protected by rename_lock */
 
  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	return strlcpy(buf, kn->parent ? kn->name : "/", buflen);
  45}
  46
  47/* kernfs_node_depth - compute depth from @from to @to */
  48static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
  49{
  50	size_t depth = 0;
  51
  52	while (to->parent && to != from) {
  53		depth++;
  54		to = to->parent;
  55	}
  56	return depth;
  57}
  58
  59static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
  60						  struct kernfs_node *b)
  61{
  62	size_t da, db;
  63	struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b);
  64
  65	if (ra != rb)
  66		return NULL;
  67
  68	da = kernfs_depth(ra->kn, a);
  69	db = kernfs_depth(rb->kn, b);
  70
  71	while (da > db) {
  72		a = a->parent;
  73		da--;
  74	}
  75	while (db > da) {
  76		b = b->parent;
  77		db--;
  78	}
  79
  80	/* worst case b and a will be the same at root */
  81	while (b != a) {
  82		b = b->parent;
  83		a = a->parent;
  84	}
  85
  86	return a;
  87}
  88
  89/**
  90 * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to,
  91 * where kn_from is treated as root of the path.
  92 * @kn_from: kernfs node which should be treated as root for the path
  93 * @kn_to: kernfs node to which path is needed
  94 * @buf: buffer to copy the path into
  95 * @buflen: size of @buf
  96 *
  97 * We need to handle couple of scenarios here:
  98 * [1] when @kn_from is an ancestor of @kn_to at some level
  99 * kn_from: /n1/n2/n3
 100 * kn_to:   /n1/n2/n3/n4/n5
 101 * result:  /n4/n5
 102 *
 103 * [2] when @kn_from is on a different hierarchy and we need to find common
 104 * ancestor between @kn_from and @kn_to.
 105 * kn_from: /n1/n2/n3/n4
 106 * kn_to:   /n1/n2/n5
 107 * result:  /../../n5
 108 * OR
 109 * kn_from: /n1/n2/n3/n4/n5   [depth=5]
 110 * kn_to:   /n1/n2/n3         [depth=3]
 111 * result:  /../..
 112 *
 
 
 113 * Returns the length of the full path.  If the full length is equal to or
 114 * greater than @buflen, @buf contains the truncated path with the trailing
 115 * '\0'.  On error, -errno is returned.
 116 */
 117static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
 118					struct kernfs_node *kn_from,
 119					char *buf, size_t buflen)
 120{
 121	struct kernfs_node *kn, *common;
 122	const char parent_str[] = "/..";
 123	size_t depth_from, depth_to, len = 0;
 124	int i, j;
 125
 
 
 
 126	if (!kn_from)
 127		kn_from = kernfs_root(kn_to)->kn;
 128
 129	if (kn_from == kn_to)
 130		return strlcpy(buf, "/", buflen);
 131
 
 
 
 132	common = kernfs_common_ancestor(kn_from, kn_to);
 133	if (WARN_ON(!common))
 134		return -EINVAL;
 135
 136	depth_to = kernfs_depth(common, kn_to);
 137	depth_from = kernfs_depth(common, kn_from);
 138
 139	if (buf)
 140		buf[0] = '\0';
 141
 142	for (i = 0; i < depth_from; i++)
 143		len += strlcpy(buf + len, parent_str,
 144			       len < buflen ? buflen - len : 0);
 145
 146	/* Calculate how many bytes we need for the rest */
 147	for (i = depth_to - 1; i >= 0; i--) {
 148		for (kn = kn_to, j = 0; j < i; j++)
 149			kn = kn->parent;
 150		len += strlcpy(buf + len, "/",
 151			       len < buflen ? buflen - len : 0);
 152		len += strlcpy(buf + len, kn->name,
 153			       len < buflen ? buflen - len : 0);
 154	}
 155
 156	return len;
 157}
 158
 159/**
 160 * kernfs_name - obtain the name of a given node
 161 * @kn: kernfs_node of interest
 162 * @buf: buffer to copy @kn's name into
 163 * @buflen: size of @buf
 164 *
 165 * Copies the name of @kn into @buf of @buflen bytes.  The behavior is
 166 * similar to strlcpy().  It returns the length of @kn's name and if @buf
 167 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
 168 *
 
 
 169 * This function can be called from any context.
 170 */
 171int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
 172{
 173	unsigned long flags;
 174	int ret;
 175
 176	spin_lock_irqsave(&kernfs_rename_lock, flags);
 177	ret = kernfs_name_locked(kn, buf, buflen);
 178	spin_unlock_irqrestore(&kernfs_rename_lock, flags);
 179	return ret;
 180}
 181
 182/**
 183 * kernfs_path_from_node - build path of node @to relative to @from.
 184 * @from: parent kernfs_node relative to which we need to build the path
 185 * @to: kernfs_node of interest
 186 * @buf: buffer to copy @to's path into
 187 * @buflen: size of @buf
 188 *
 189 * Builds @to's path relative to @from in @buf. @from and @to must
 190 * be on the same kernfs-root. If @from is not parent of @to, then a relative
 191 * path (which includes '..'s) as needed to reach from @from to @to is
 192 * returned.
 193 *
 194 * Returns the length of the full path.  If the full length is equal to or
 195 * greater than @buflen, @buf contains the truncated path with the trailing
 196 * '\0'.  On error, -errno is returned.
 197 */
 198int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
 199			  char *buf, size_t buflen)
 200{
 201	unsigned long flags;
 202	int ret;
 203
 204	spin_lock_irqsave(&kernfs_rename_lock, flags);
 205	ret = kernfs_path_from_node_locked(to, from, buf, buflen);
 206	spin_unlock_irqrestore(&kernfs_rename_lock, flags);
 207	return ret;
 208}
 209EXPORT_SYMBOL_GPL(kernfs_path_from_node);
 210
 211/**
 212 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
 213 * @kn: kernfs_node of interest
 214 *
 215 * This function can be called from any context.
 216 */
 217void pr_cont_kernfs_name(struct kernfs_node *kn)
 218{
 219	unsigned long flags;
 220
 221	spin_lock_irqsave(&kernfs_rename_lock, flags);
 222
 223	kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
 224	pr_cont("%s", kernfs_pr_cont_buf);
 225
 226	spin_unlock_irqrestore(&kernfs_rename_lock, flags);
 227}
 228
 229/**
 230 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
 231 * @kn: kernfs_node of interest
 232 *
 233 * This function can be called from any context.
 234 */
 235void pr_cont_kernfs_path(struct kernfs_node *kn)
 236{
 237	unsigned long flags;
 238	int sz;
 239
 240	spin_lock_irqsave(&kernfs_rename_lock, flags);
 241
 242	sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf,
 243					  sizeof(kernfs_pr_cont_buf));
 244	if (sz < 0) {
 245		pr_cont("(error)");
 246		goto out;
 247	}
 248
 249	if (sz >= sizeof(kernfs_pr_cont_buf)) {
 250		pr_cont("(name too long)");
 251		goto out;
 252	}
 253
 254	pr_cont("%s", kernfs_pr_cont_buf);
 255
 256out:
 257	spin_unlock_irqrestore(&kernfs_rename_lock, flags);
 258}
 259
 260/**
 261 * kernfs_get_parent - determine the parent node and pin it
 262 * @kn: kernfs_node of interest
 263 *
 264 * Determines @kn's parent, pins and returns it.  This function can be
 265 * called from any context.
 266 */
 267struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
 268{
 269	struct kernfs_node *parent;
 270	unsigned long flags;
 271
 272	spin_lock_irqsave(&kernfs_rename_lock, flags);
 273	parent = kn->parent;
 274	kernfs_get(parent);
 275	spin_unlock_irqrestore(&kernfs_rename_lock, flags);
 276
 277	return parent;
 278}
 279
 280/**
 281 *	kernfs_name_hash
 282 *	@name: Null terminated string to hash
 283 *	@ns:   Namespace tag to hash
 284 *
 285 *	Returns 31 bit hash of ns + name (so it fits in an off_t )
 286 */
 287static unsigned int kernfs_name_hash(const char *name, const void *ns)
 288{
 289	unsigned long hash = init_name_hash(ns);
 290	unsigned int len = strlen(name);
 291	while (len--)
 292		hash = partial_name_hash(*name++, hash);
 293	hash = end_name_hash(hash);
 294	hash &= 0x7fffffffU;
 295	/* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
 296	if (hash < 2)
 297		hash += 2;
 298	if (hash >= INT_MAX)
 299		hash = INT_MAX - 1;
 300	return hash;
 301}
 302
 303static int kernfs_name_compare(unsigned int hash, const char *name,
 304			       const void *ns, const struct kernfs_node *kn)
 305{
 306	if (hash < kn->hash)
 307		return -1;
 308	if (hash > kn->hash)
 309		return 1;
 310	if (ns < kn->ns)
 311		return -1;
 312	if (ns > kn->ns)
 313		return 1;
 314	return strcmp(name, kn->name);
 315}
 316
 317static int kernfs_sd_compare(const struct kernfs_node *left,
 318			     const struct kernfs_node *right)
 319{
 320	return kernfs_name_compare(left->hash, left->name, left->ns, right);
 321}
 322
 323/**
 324 *	kernfs_link_sibling - link kernfs_node into sibling rbtree
 325 *	@kn: kernfs_node of interest
 326 *
 327 *	Link @kn into its sibling rbtree which starts from
 328 *	@kn->parent->dir.children.
 329 *
 330 *	Locking:
 331 *	mutex_lock(kernfs_mutex)
 332 *
 333 *	RETURNS:
 334 *	0 on susccess -EEXIST on failure.
 335 */
 336static int kernfs_link_sibling(struct kernfs_node *kn)
 337{
 338	struct rb_node **node = &kn->parent->dir.children.rb_node;
 339	struct rb_node *parent = NULL;
 340
 341	while (*node) {
 342		struct kernfs_node *pos;
 343		int result;
 344
 345		pos = rb_to_kn(*node);
 346		parent = *node;
 347		result = kernfs_sd_compare(kn, pos);
 348		if (result < 0)
 349			node = &pos->rb.rb_left;
 350		else if (result > 0)
 351			node = &pos->rb.rb_right;
 352		else
 353			return -EEXIST;
 354	}
 355
 356	/* add new node and rebalance the tree */
 357	rb_link_node(&kn->rb, parent, node);
 358	rb_insert_color(&kn->rb, &kn->parent->dir.children);
 359
 360	/* successfully added, account subdir number */
 361	if (kernfs_type(kn) == KERNFS_DIR)
 362		kn->parent->dir.subdirs++;
 363
 364	return 0;
 365}
 366
 367/**
 368 *	kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
 369 *	@kn: kernfs_node of interest
 370 *
 371 *	Try to unlink @kn from its sibling rbtree which starts from
 372 *	kn->parent->dir.children.  Returns %true if @kn was actually
 373 *	removed, %false if @kn wasn't on the rbtree.
 374 *
 375 *	Locking:
 376 *	mutex_lock(kernfs_mutex)
 377 */
 378static bool kernfs_unlink_sibling(struct kernfs_node *kn)
 379{
 380	if (RB_EMPTY_NODE(&kn->rb))
 381		return false;
 382
 383	if (kernfs_type(kn) == KERNFS_DIR)
 384		kn->parent->dir.subdirs--;
 385
 386	rb_erase(&kn->rb, &kn->parent->dir.children);
 387	RB_CLEAR_NODE(&kn->rb);
 388	return true;
 389}
 390
 391/**
 392 *	kernfs_get_active - get an active reference to kernfs_node
 393 *	@kn: kernfs_node to get an active reference to
 394 *
 395 *	Get an active reference of @kn.  This function is noop if @kn
 396 *	is NULL.
 397 *
 398 *	RETURNS:
 399 *	Pointer to @kn on success, NULL on failure.
 400 */
 401struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
 402{
 403	if (unlikely(!kn))
 404		return NULL;
 405
 406	if (!atomic_inc_unless_negative(&kn->active))
 407		return NULL;
 408
 409	if (kernfs_lockdep(kn))
 410		rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
 411	return kn;
 412}
 413
 414/**
 415 *	kernfs_put_active - put an active reference to kernfs_node
 416 *	@kn: kernfs_node to put an active reference to
 417 *
 418 *	Put an active reference to @kn.  This function is noop if @kn
 419 *	is NULL.
 420 */
 421void kernfs_put_active(struct kernfs_node *kn)
 422{
 423	struct kernfs_root *root = kernfs_root(kn);
 424	int v;
 425
 426	if (unlikely(!kn))
 427		return;
 428
 429	if (kernfs_lockdep(kn))
 430		rwsem_release(&kn->dep_map, 1, _RET_IP_);
 431	v = atomic_dec_return(&kn->active);
 432	if (likely(v != KN_DEACTIVATED_BIAS))
 433		return;
 434
 435	wake_up_all(&root->deactivate_waitq);
 436}
 437
 438/**
 439 * kernfs_drain - drain kernfs_node
 440 * @kn: kernfs_node to drain
 441 *
 442 * Drain existing usages and nuke all existing mmaps of @kn.  Mutiple
 443 * removers may invoke this function concurrently on @kn and all will
 444 * return after draining is complete.
 445 */
 446static void kernfs_drain(struct kernfs_node *kn)
 447	__releases(&kernfs_mutex) __acquires(&kernfs_mutex)
 448{
 449	struct kernfs_root *root = kernfs_root(kn);
 450
 451	lockdep_assert_held(&kernfs_mutex);
 452	WARN_ON_ONCE(kernfs_active(kn));
 453
 454	mutex_unlock(&kernfs_mutex);
 455
 456	if (kernfs_lockdep(kn)) {
 457		rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
 458		if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
 459			lock_contended(&kn->dep_map, _RET_IP_);
 460	}
 461
 462	/* but everyone should wait for draining */
 463	wait_event(root->deactivate_waitq,
 464		   atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
 465
 466	if (kernfs_lockdep(kn)) {
 467		lock_acquired(&kn->dep_map, _RET_IP_);
 468		rwsem_release(&kn->dep_map, 1, _RET_IP_);
 469	}
 470
 471	kernfs_unmap_bin_file(kn);
 472
 473	mutex_lock(&kernfs_mutex);
 474}
 475
 476/**
 477 * kernfs_get - get a reference count on a kernfs_node
 478 * @kn: the target kernfs_node
 479 */
 480void kernfs_get(struct kernfs_node *kn)
 481{
 482	if (kn) {
 483		WARN_ON(!atomic_read(&kn->count));
 484		atomic_inc(&kn->count);
 485	}
 486}
 487EXPORT_SYMBOL_GPL(kernfs_get);
 488
 489/**
 490 * kernfs_put - put a reference count on a kernfs_node
 491 * @kn: the target kernfs_node
 492 *
 493 * Put a reference count of @kn and destroy it if it reached zero.
 494 */
 495void kernfs_put(struct kernfs_node *kn)
 496{
 497	struct kernfs_node *parent;
 498	struct kernfs_root *root;
 499
 500	if (!kn || !atomic_dec_and_test(&kn->count))
 501		return;
 502	root = kernfs_root(kn);
 503 repeat:
 504	/*
 505	 * Moving/renaming is always done while holding reference.
 506	 * kn->parent won't change beneath us.
 507	 */
 508	parent = kn->parent;
 509
 510	WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
 511		  "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
 512		  parent ? parent->name : "", kn->name, atomic_read(&kn->active));
 513
 514	if (kernfs_type(kn) == KERNFS_LINK)
 515		kernfs_put(kn->symlink.target_kn);
 516
 517	kfree_const(kn->name);
 518
 519	if (kn->iattr) {
 520		if (kn->iattr->ia_secdata)
 521			security_release_secctx(kn->iattr->ia_secdata,
 522						kn->iattr->ia_secdata_len);
 523		simple_xattrs_free(&kn->iattr->xattrs);
 
 524	}
 525	kfree(kn->iattr);
 526	ida_simple_remove(&root->ino_ida, kn->ino);
 
 527	kmem_cache_free(kernfs_node_cache, kn);
 528
 529	kn = parent;
 530	if (kn) {
 531		if (atomic_dec_and_test(&kn->count))
 532			goto repeat;
 533	} else {
 534		/* just released the root kn, free @root too */
 535		ida_destroy(&root->ino_ida);
 536		kfree(root);
 537	}
 538}
 539EXPORT_SYMBOL_GPL(kernfs_put);
 540
 541static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
 542{
 543	struct kernfs_node *kn;
 544
 545	if (flags & LOOKUP_RCU)
 546		return -ECHILD;
 547
 548	/* Always perform fresh lookup for negatives */
 549	if (d_really_is_negative(dentry))
 550		goto out_bad_unlocked;
 551
 552	kn = dentry->d_fsdata;
 553	mutex_lock(&kernfs_mutex);
 554
 555	/* The kernfs node has been deactivated */
 556	if (!kernfs_active(kn))
 557		goto out_bad;
 558
 559	/* The kernfs node has been moved? */
 560	if (dentry->d_parent->d_fsdata != kn->parent)
 561		goto out_bad;
 562
 563	/* The kernfs node has been renamed */
 564	if (strcmp(dentry->d_name.name, kn->name) != 0)
 565		goto out_bad;
 566
 567	/* The kernfs node has been moved to a different namespace */
 568	if (kn->parent && kernfs_ns_enabled(kn->parent) &&
 569	    kernfs_info(dentry->d_sb)->ns != kn->ns)
 570		goto out_bad;
 571
 572	mutex_unlock(&kernfs_mutex);
 573	return 1;
 574out_bad:
 575	mutex_unlock(&kernfs_mutex);
 576out_bad_unlocked:
 577	return 0;
 578}
 579
 580static void kernfs_dop_release(struct dentry *dentry)
 581{
 582	kernfs_put(dentry->d_fsdata);
 583}
 584
 585const struct dentry_operations kernfs_dops = {
 586	.d_revalidate	= kernfs_dop_revalidate,
 587	.d_release	= kernfs_dop_release,
 588};
 589
 590/**
 591 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
 592 * @dentry: the dentry in question
 593 *
 594 * Return the kernfs_node associated with @dentry.  If @dentry is not a
 595 * kernfs one, %NULL is returned.
 596 *
 597 * While the returned kernfs_node will stay accessible as long as @dentry
 598 * is accessible, the returned node can be in any state and the caller is
 599 * fully responsible for determining what's accessible.
 600 */
 601struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
 602{
 603	if (dentry->d_sb->s_op == &kernfs_sops)
 604		return dentry->d_fsdata;
 605	return NULL;
 606}
 607
 608static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
 
 609					     const char *name, umode_t mode,
 
 610					     unsigned flags)
 611{
 612	struct kernfs_node *kn;
 
 613	int ret;
 614
 615	name = kstrdup_const(name, GFP_KERNEL);
 616	if (!name)
 617		return NULL;
 618
 619	kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
 620	if (!kn)
 621		goto err_out1;
 622
 623	ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL);
 
 
 
 
 
 
 
 
 624	if (ret < 0)
 625		goto err_out2;
 626	kn->ino = ret;
 
 627
 628	atomic_set(&kn->count, 1);
 629	atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
 630	RB_CLEAR_NODE(&kn->rb);
 631
 632	kn->name = name;
 633	kn->mode = mode;
 634	kn->flags = flags;
 635
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 636	return kn;
 637
 
 
 638 err_out2:
 639	kmem_cache_free(kernfs_node_cache, kn);
 640 err_out1:
 641	kfree_const(name);
 642	return NULL;
 643}
 644
 645struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
 646				    const char *name, umode_t mode,
 
 647				    unsigned flags)
 648{
 649	struct kernfs_node *kn;
 650
 651	kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags);
 
 652	if (kn) {
 653		kernfs_get(parent);
 654		kn->parent = parent;
 655	}
 656	return kn;
 657}
 658
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 659/**
 660 *	kernfs_add_one - add kernfs_node to parent without warning
 661 *	@kn: kernfs_node to be added
 662 *
 663 *	The caller must already have initialized @kn->parent.  This
 664 *	function increments nlink of the parent's inode if @kn is a
 665 *	directory and link into the children list of the parent.
 666 *
 667 *	RETURNS:
 668 *	0 on success, -EEXIST if entry with the given name already
 669 *	exists.
 670 */
 671int kernfs_add_one(struct kernfs_node *kn)
 672{
 673	struct kernfs_node *parent = kn->parent;
 674	struct kernfs_iattrs *ps_iattr;
 675	bool has_ns;
 676	int ret;
 677
 678	mutex_lock(&kernfs_mutex);
 679
 680	ret = -EINVAL;
 681	has_ns = kernfs_ns_enabled(parent);
 682	if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
 683		 has_ns ? "required" : "invalid", parent->name, kn->name))
 684		goto out_unlock;
 685
 686	if (kernfs_type(parent) != KERNFS_DIR)
 687		goto out_unlock;
 688
 689	ret = -ENOENT;
 690	if (parent->flags & KERNFS_EMPTY_DIR)
 691		goto out_unlock;
 692
 693	if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
 694		goto out_unlock;
 695
 696	kn->hash = kernfs_name_hash(kn->name, kn->ns);
 697
 698	ret = kernfs_link_sibling(kn);
 699	if (ret)
 700		goto out_unlock;
 701
 702	/* Update timestamps on the parent */
 703	ps_iattr = parent->iattr;
 704	if (ps_iattr) {
 705		struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
 706		ktime_get_real_ts(&ps_iattrs->ia_ctime);
 707		ps_iattrs->ia_mtime = ps_iattrs->ia_ctime;
 708	}
 709
 710	mutex_unlock(&kernfs_mutex);
 711
 712	/*
 713	 * Activate the new node unless CREATE_DEACTIVATED is requested.
 714	 * If not activated here, the kernfs user is responsible for
 715	 * activating the node with kernfs_activate().  A node which hasn't
 716	 * been activated is not visible to userland and its removal won't
 717	 * trigger deactivation.
 718	 */
 719	if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
 720		kernfs_activate(kn);
 721	return 0;
 722
 723out_unlock:
 724	mutex_unlock(&kernfs_mutex);
 725	return ret;
 726}
 727
 728/**
 729 * kernfs_find_ns - find kernfs_node with the given name
 730 * @parent: kernfs_node to search under
 731 * @name: name to look for
 732 * @ns: the namespace tag to use
 733 *
 734 * Look for kernfs_node with name @name under @parent.  Returns pointer to
 735 * the found kernfs_node on success, %NULL on failure.
 736 */
 737static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
 738					  const unsigned char *name,
 739					  const void *ns)
 740{
 741	struct rb_node *node = parent->dir.children.rb_node;
 742	bool has_ns = kernfs_ns_enabled(parent);
 743	unsigned int hash;
 744
 745	lockdep_assert_held(&kernfs_mutex);
 746
 747	if (has_ns != (bool)ns) {
 748		WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
 749		     has_ns ? "required" : "invalid", parent->name, name);
 750		return NULL;
 751	}
 752
 753	hash = kernfs_name_hash(name, ns);
 754	while (node) {
 755		struct kernfs_node *kn;
 756		int result;
 757
 758		kn = rb_to_kn(node);
 759		result = kernfs_name_compare(hash, name, ns, kn);
 760		if (result < 0)
 761			node = node->rb_left;
 762		else if (result > 0)
 763			node = node->rb_right;
 764		else
 765			return kn;
 766	}
 767	return NULL;
 768}
 769
 770static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
 771					  const unsigned char *path,
 772					  const void *ns)
 773{
 774	size_t len;
 775	char *p, *name;
 776
 777	lockdep_assert_held(&kernfs_mutex);
 778
 779	/* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
 780	spin_lock_irq(&kernfs_rename_lock);
 781
 782	len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
 783
 784	if (len >= sizeof(kernfs_pr_cont_buf)) {
 785		spin_unlock_irq(&kernfs_rename_lock);
 786		return NULL;
 787	}
 788
 789	p = kernfs_pr_cont_buf;
 790
 791	while ((name = strsep(&p, "/")) && parent) {
 792		if (*name == '\0')
 793			continue;
 794		parent = kernfs_find_ns(parent, name, ns);
 795	}
 796
 797	spin_unlock_irq(&kernfs_rename_lock);
 798
 799	return parent;
 800}
 801
 802/**
 803 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
 804 * @parent: kernfs_node to search under
 805 * @name: name to look for
 806 * @ns: the namespace tag to use
 807 *
 808 * Look for kernfs_node with name @name under @parent and get a reference
 809 * if found.  This function may sleep and returns pointer to the found
 810 * kernfs_node on success, %NULL on failure.
 811 */
 812struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
 813					   const char *name, const void *ns)
 814{
 815	struct kernfs_node *kn;
 816
 817	mutex_lock(&kernfs_mutex);
 818	kn = kernfs_find_ns(parent, name, ns);
 819	kernfs_get(kn);
 820	mutex_unlock(&kernfs_mutex);
 821
 822	return kn;
 823}
 824EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
 825
 826/**
 827 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
 828 * @parent: kernfs_node to search under
 829 * @path: path to look for
 830 * @ns: the namespace tag to use
 831 *
 832 * Look for kernfs_node with path @path under @parent and get a reference
 833 * if found.  This function may sleep and returns pointer to the found
 834 * kernfs_node on success, %NULL on failure.
 835 */
 836struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
 837					   const char *path, const void *ns)
 838{
 839	struct kernfs_node *kn;
 840
 841	mutex_lock(&kernfs_mutex);
 842	kn = kernfs_walk_ns(parent, path, ns);
 843	kernfs_get(kn);
 844	mutex_unlock(&kernfs_mutex);
 845
 846	return kn;
 847}
 848
 849/**
 850 * kernfs_create_root - create a new kernfs hierarchy
 851 * @scops: optional syscall operations for the hierarchy
 852 * @flags: KERNFS_ROOT_* flags
 853 * @priv: opaque data associated with the new directory
 854 *
 855 * Returns the root of the new hierarchy on success, ERR_PTR() value on
 856 * failure.
 857 */
 858struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
 859				       unsigned int flags, void *priv)
 860{
 861	struct kernfs_root *root;
 862	struct kernfs_node *kn;
 863
 864	root = kzalloc(sizeof(*root), GFP_KERNEL);
 865	if (!root)
 866		return ERR_PTR(-ENOMEM);
 867
 868	ida_init(&root->ino_ida);
 869	INIT_LIST_HEAD(&root->supers);
 870
 871	kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO,
 
 
 
 
 
 
 
 
 
 
 
 
 872			       KERNFS_DIR);
 873	if (!kn) {
 874		ida_destroy(&root->ino_ida);
 875		kfree(root);
 876		return ERR_PTR(-ENOMEM);
 877	}
 878
 879	kn->priv = priv;
 880	kn->dir.root = root;
 881
 882	root->syscall_ops = scops;
 883	root->flags = flags;
 884	root->kn = kn;
 885	init_waitqueue_head(&root->deactivate_waitq);
 886
 887	if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
 888		kernfs_activate(kn);
 889
 890	return root;
 891}
 892
 893/**
 894 * kernfs_destroy_root - destroy a kernfs hierarchy
 895 * @root: root of the hierarchy to destroy
 896 *
 897 * Destroy the hierarchy anchored at @root by removing all existing
 898 * directories and destroying @root.
 899 */
 900void kernfs_destroy_root(struct kernfs_root *root)
 901{
 902	kernfs_remove(root->kn);	/* will also free @root */
 903}
 904
 905/**
 906 * kernfs_create_dir_ns - create a directory
 907 * @parent: parent in which to create a new directory
 908 * @name: name of the new directory
 909 * @mode: mode of the new directory
 
 
 910 * @priv: opaque data associated with the new directory
 911 * @ns: optional namespace tag of the directory
 912 *
 913 * Returns the created node on success, ERR_PTR() value on failure.
 914 */
 915struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
 916					 const char *name, umode_t mode,
 
 917					 void *priv, const void *ns)
 918{
 919	struct kernfs_node *kn;
 920	int rc;
 921
 922	/* allocate */
 923	kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR);
 
 924	if (!kn)
 925		return ERR_PTR(-ENOMEM);
 926
 927	kn->dir.root = parent->dir.root;
 928	kn->ns = ns;
 929	kn->priv = priv;
 930
 931	/* link in */
 932	rc = kernfs_add_one(kn);
 933	if (!rc)
 934		return kn;
 935
 936	kernfs_put(kn);
 937	return ERR_PTR(rc);
 938}
 939
 940/**
 941 * kernfs_create_empty_dir - create an always empty directory
 942 * @parent: parent in which to create a new directory
 943 * @name: name of the new directory
 944 *
 945 * Returns the created node on success, ERR_PTR() value on failure.
 946 */
 947struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
 948					    const char *name)
 949{
 950	struct kernfs_node *kn;
 951	int rc;
 952
 953	/* allocate */
 954	kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, KERNFS_DIR);
 
 955	if (!kn)
 956		return ERR_PTR(-ENOMEM);
 957
 958	kn->flags |= KERNFS_EMPTY_DIR;
 959	kn->dir.root = parent->dir.root;
 960	kn->ns = NULL;
 961	kn->priv = NULL;
 962
 963	/* link in */
 964	rc = kernfs_add_one(kn);
 965	if (!rc)
 966		return kn;
 967
 968	kernfs_put(kn);
 969	return ERR_PTR(rc);
 970}
 971
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 972static struct dentry *kernfs_iop_lookup(struct inode *dir,
 973					struct dentry *dentry,
 974					unsigned int flags)
 975{
 976	struct dentry *ret;
 977	struct kernfs_node *parent = dentry->d_parent->d_fsdata;
 978	struct kernfs_node *kn;
 979	struct inode *inode;
 980	const void *ns = NULL;
 981
 982	mutex_lock(&kernfs_mutex);
 983
 984	if (kernfs_ns_enabled(parent))
 985		ns = kernfs_info(dir->i_sb)->ns;
 986
 987	kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
 988
 989	/* no such entry */
 990	if (!kn || !kernfs_active(kn)) {
 991		ret = NULL;
 992		goto out_unlock;
 993	}
 994	kernfs_get(kn);
 995	dentry->d_fsdata = kn;
 996
 997	/* attach dentry and inode */
 998	inode = kernfs_get_inode(dir->i_sb, kn);
 999	if (!inode) {
1000		ret = ERR_PTR(-ENOMEM);
1001		goto out_unlock;
1002	}
1003
1004	/* instantiate and hash dentry */
1005	ret = d_splice_alias(inode, dentry);
1006 out_unlock:
1007	mutex_unlock(&kernfs_mutex);
1008	return ret;
1009}
1010
1011static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
 
1012			    umode_t mode)
1013{
1014	struct kernfs_node *parent = dir->i_private;
1015	struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1016	int ret;
1017
1018	if (!scops || !scops->mkdir)
1019		return -EPERM;
1020
1021	if (!kernfs_get_active(parent))
1022		return -ENODEV;
1023
1024	ret = scops->mkdir(parent, dentry->d_name.name, mode);
1025
1026	kernfs_put_active(parent);
1027	return ret;
1028}
1029
1030static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1031{
1032	struct kernfs_node *kn  = dentry->d_fsdata;
1033	struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1034	int ret;
1035
1036	if (!scops || !scops->rmdir)
1037		return -EPERM;
1038
1039	if (!kernfs_get_active(kn))
1040		return -ENODEV;
1041
1042	ret = scops->rmdir(kn);
1043
1044	kernfs_put_active(kn);
1045	return ret;
1046}
1047
1048static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
 
1049			     struct inode *new_dir, struct dentry *new_dentry,
1050			     unsigned int flags)
1051{
1052	struct kernfs_node *kn  = old_dentry->d_fsdata;
1053	struct kernfs_node *new_parent = new_dir->i_private;
1054	struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1055	int ret;
1056
1057	if (flags)
1058		return -EINVAL;
1059
1060	if (!scops || !scops->rename)
1061		return -EPERM;
1062
1063	if (!kernfs_get_active(kn))
1064		return -ENODEV;
1065
1066	if (!kernfs_get_active(new_parent)) {
1067		kernfs_put_active(kn);
1068		return -ENODEV;
1069	}
1070
1071	ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1072
1073	kernfs_put_active(new_parent);
1074	kernfs_put_active(kn);
1075	return ret;
1076}
1077
1078const struct inode_operations kernfs_dir_iops = {
1079	.lookup		= kernfs_iop_lookup,
1080	.permission	= kernfs_iop_permission,
1081	.setattr	= kernfs_iop_setattr,
1082	.getattr	= kernfs_iop_getattr,
1083	.listxattr	= kernfs_iop_listxattr,
1084
1085	.mkdir		= kernfs_iop_mkdir,
1086	.rmdir		= kernfs_iop_rmdir,
1087	.rename		= kernfs_iop_rename,
1088};
1089
1090static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1091{
1092	struct kernfs_node *last;
1093
1094	while (true) {
1095		struct rb_node *rbn;
1096
1097		last = pos;
1098
1099		if (kernfs_type(pos) != KERNFS_DIR)
1100			break;
1101
1102		rbn = rb_first(&pos->dir.children);
1103		if (!rbn)
1104			break;
1105
1106		pos = rb_to_kn(rbn);
1107	}
1108
1109	return last;
1110}
1111
1112/**
1113 * kernfs_next_descendant_post - find the next descendant for post-order walk
1114 * @pos: the current position (%NULL to initiate traversal)
1115 * @root: kernfs_node whose descendants to walk
1116 *
1117 * Find the next descendant to visit for post-order traversal of @root's
1118 * descendants.  @root is included in the iteration and the last node to be
1119 * visited.
1120 */
1121static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1122						       struct kernfs_node *root)
1123{
1124	struct rb_node *rbn;
1125
1126	lockdep_assert_held(&kernfs_mutex);
1127
1128	/* if first iteration, visit leftmost descendant which may be root */
1129	if (!pos)
1130		return kernfs_leftmost_descendant(root);
1131
1132	/* if we visited @root, we're done */
1133	if (pos == root)
1134		return NULL;
1135
1136	/* if there's an unvisited sibling, visit its leftmost descendant */
1137	rbn = rb_next(&pos->rb);
1138	if (rbn)
1139		return kernfs_leftmost_descendant(rb_to_kn(rbn));
1140
1141	/* no sibling left, visit parent */
1142	return pos->parent;
1143}
1144
1145/**
1146 * kernfs_activate - activate a node which started deactivated
1147 * @kn: kernfs_node whose subtree is to be activated
1148 *
1149 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1150 * needs to be explicitly activated.  A node which hasn't been activated
1151 * isn't visible to userland and deactivation is skipped during its
1152 * removal.  This is useful to construct atomic init sequences where
1153 * creation of multiple nodes should either succeed or fail atomically.
1154 *
1155 * The caller is responsible for ensuring that this function is not called
1156 * after kernfs_remove*() is invoked on @kn.
1157 */
1158void kernfs_activate(struct kernfs_node *kn)
1159{
1160	struct kernfs_node *pos;
1161
1162	mutex_lock(&kernfs_mutex);
1163
1164	pos = NULL;
1165	while ((pos = kernfs_next_descendant_post(pos, kn))) {
1166		if (!pos || (pos->flags & KERNFS_ACTIVATED))
1167			continue;
1168
1169		WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1170		WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1171
1172		atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1173		pos->flags |= KERNFS_ACTIVATED;
1174	}
1175
1176	mutex_unlock(&kernfs_mutex);
1177}
1178
1179static void __kernfs_remove(struct kernfs_node *kn)
1180{
1181	struct kernfs_node *pos;
1182
1183	lockdep_assert_held(&kernfs_mutex);
1184
1185	/*
1186	 * Short-circuit if non-root @kn has already finished removal.
1187	 * This is for kernfs_remove_self() which plays with active ref
1188	 * after removal.
1189	 */
1190	if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1191		return;
1192
1193	pr_debug("kernfs %s: removing\n", kn->name);
1194
1195	/* prevent any new usage under @kn by deactivating all nodes */
1196	pos = NULL;
1197	while ((pos = kernfs_next_descendant_post(pos, kn)))
1198		if (kernfs_active(pos))
1199			atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1200
1201	/* deactivate and unlink the subtree node-by-node */
1202	do {
1203		pos = kernfs_leftmost_descendant(kn);
1204
1205		/*
1206		 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1207		 * base ref could have been put by someone else by the time
1208		 * the function returns.  Make sure it doesn't go away
1209		 * underneath us.
1210		 */
1211		kernfs_get(pos);
1212
1213		/*
1214		 * Drain iff @kn was activated.  This avoids draining and
1215		 * its lockdep annotations for nodes which have never been
1216		 * activated and allows embedding kernfs_remove() in create
1217		 * error paths without worrying about draining.
1218		 */
1219		if (kn->flags & KERNFS_ACTIVATED)
1220			kernfs_drain(pos);
1221		else
1222			WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1223
1224		/*
1225		 * kernfs_unlink_sibling() succeeds once per node.  Use it
1226		 * to decide who's responsible for cleanups.
1227		 */
1228		if (!pos->parent || kernfs_unlink_sibling(pos)) {
1229			struct kernfs_iattrs *ps_iattr =
1230				pos->parent ? pos->parent->iattr : NULL;
1231
1232			/* update timestamps on the parent */
1233			if (ps_iattr) {
1234				ktime_get_real_ts(&ps_iattr->ia_iattr.ia_ctime);
1235				ps_iattr->ia_iattr.ia_mtime =
1236					ps_iattr->ia_iattr.ia_ctime;
1237			}
1238
1239			kernfs_put(pos);
1240		}
1241
1242		kernfs_put(pos);
1243	} while (pos != kn);
1244}
1245
1246/**
1247 * kernfs_remove - remove a kernfs_node recursively
1248 * @kn: the kernfs_node to remove
1249 *
1250 * Remove @kn along with all its subdirectories and files.
1251 */
1252void kernfs_remove(struct kernfs_node *kn)
1253{
1254	mutex_lock(&kernfs_mutex);
1255	__kernfs_remove(kn);
1256	mutex_unlock(&kernfs_mutex);
1257}
1258
1259/**
1260 * kernfs_break_active_protection - break out of active protection
1261 * @kn: the self kernfs_node
1262 *
1263 * The caller must be running off of a kernfs operation which is invoked
1264 * with an active reference - e.g. one of kernfs_ops.  Each invocation of
1265 * this function must also be matched with an invocation of
1266 * kernfs_unbreak_active_protection().
1267 *
1268 * This function releases the active reference of @kn the caller is
1269 * holding.  Once this function is called, @kn may be removed at any point
1270 * and the caller is solely responsible for ensuring that the objects it
1271 * dereferences are accessible.
1272 */
1273void kernfs_break_active_protection(struct kernfs_node *kn)
1274{
1275	/*
1276	 * Take out ourself out of the active ref dependency chain.  If
1277	 * we're called without an active ref, lockdep will complain.
1278	 */
1279	kernfs_put_active(kn);
1280}
1281
1282/**
1283 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1284 * @kn: the self kernfs_node
1285 *
1286 * If kernfs_break_active_protection() was called, this function must be
1287 * invoked before finishing the kernfs operation.  Note that while this
1288 * function restores the active reference, it doesn't and can't actually
1289 * restore the active protection - @kn may already or be in the process of
1290 * being removed.  Once kernfs_break_active_protection() is invoked, that
1291 * protection is irreversibly gone for the kernfs operation instance.
1292 *
1293 * While this function may be called at any point after
1294 * kernfs_break_active_protection() is invoked, its most useful location
1295 * would be right before the enclosing kernfs operation returns.
1296 */
1297void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1298{
1299	/*
1300	 * @kn->active could be in any state; however, the increment we do
1301	 * here will be undone as soon as the enclosing kernfs operation
1302	 * finishes and this temporary bump can't break anything.  If @kn
1303	 * is alive, nothing changes.  If @kn is being deactivated, the
1304	 * soon-to-follow put will either finish deactivation or restore
1305	 * deactivated state.  If @kn is already removed, the temporary
1306	 * bump is guaranteed to be gone before @kn is released.
1307	 */
1308	atomic_inc(&kn->active);
1309	if (kernfs_lockdep(kn))
1310		rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1311}
1312
1313/**
1314 * kernfs_remove_self - remove a kernfs_node from its own method
1315 * @kn: the self kernfs_node to remove
1316 *
1317 * The caller must be running off of a kernfs operation which is invoked
1318 * with an active reference - e.g. one of kernfs_ops.  This can be used to
1319 * implement a file operation which deletes itself.
1320 *
1321 * For example, the "delete" file for a sysfs device directory can be
1322 * implemented by invoking kernfs_remove_self() on the "delete" file
1323 * itself.  This function breaks the circular dependency of trying to
1324 * deactivate self while holding an active ref itself.  It isn't necessary
1325 * to modify the usual removal path to use kernfs_remove_self().  The
1326 * "delete" implementation can simply invoke kernfs_remove_self() on self
1327 * before proceeding with the usual removal path.  kernfs will ignore later
1328 * kernfs_remove() on self.
1329 *
1330 * kernfs_remove_self() can be called multiple times concurrently on the
1331 * same kernfs_node.  Only the first one actually performs removal and
1332 * returns %true.  All others will wait until the kernfs operation which
1333 * won self-removal finishes and return %false.  Note that the losers wait
1334 * for the completion of not only the winning kernfs_remove_self() but also
1335 * the whole kernfs_ops which won the arbitration.  This can be used to
1336 * guarantee, for example, all concurrent writes to a "delete" file to
1337 * finish only after the whole operation is complete.
1338 */
1339bool kernfs_remove_self(struct kernfs_node *kn)
1340{
1341	bool ret;
1342
1343	mutex_lock(&kernfs_mutex);
1344	kernfs_break_active_protection(kn);
1345
1346	/*
1347	 * SUICIDAL is used to arbitrate among competing invocations.  Only
1348	 * the first one will actually perform removal.  When the removal
1349	 * is complete, SUICIDED is set and the active ref is restored
1350	 * while holding kernfs_mutex.  The ones which lost arbitration
1351	 * waits for SUICDED && drained which can happen only after the
1352	 * enclosing kernfs operation which executed the winning instance
1353	 * of kernfs_remove_self() finished.
1354	 */
1355	if (!(kn->flags & KERNFS_SUICIDAL)) {
1356		kn->flags |= KERNFS_SUICIDAL;
1357		__kernfs_remove(kn);
1358		kn->flags |= KERNFS_SUICIDED;
1359		ret = true;
1360	} else {
1361		wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1362		DEFINE_WAIT(wait);
1363
1364		while (true) {
1365			prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1366
1367			if ((kn->flags & KERNFS_SUICIDED) &&
1368			    atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1369				break;
1370
1371			mutex_unlock(&kernfs_mutex);
1372			schedule();
1373			mutex_lock(&kernfs_mutex);
1374		}
1375		finish_wait(waitq, &wait);
1376		WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1377		ret = false;
1378	}
1379
1380	/*
1381	 * This must be done while holding kernfs_mutex; otherwise, waiting
1382	 * for SUICIDED && deactivated could finish prematurely.
1383	 */
1384	kernfs_unbreak_active_protection(kn);
1385
1386	mutex_unlock(&kernfs_mutex);
1387	return ret;
1388}
1389
1390/**
1391 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1392 * @parent: parent of the target
1393 * @name: name of the kernfs_node to remove
1394 * @ns: namespace tag of the kernfs_node to remove
1395 *
1396 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1397 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1398 */
1399int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1400			     const void *ns)
1401{
1402	struct kernfs_node *kn;
1403
1404	if (!parent) {
1405		WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1406			name);
1407		return -ENOENT;
1408	}
1409
1410	mutex_lock(&kernfs_mutex);
1411
1412	kn = kernfs_find_ns(parent, name, ns);
1413	if (kn)
1414		__kernfs_remove(kn);
1415
1416	mutex_unlock(&kernfs_mutex);
1417
1418	if (kn)
1419		return 0;
1420	else
1421		return -ENOENT;
1422}
1423
1424/**
1425 * kernfs_rename_ns - move and rename a kernfs_node
1426 * @kn: target node
1427 * @new_parent: new parent to put @sd under
1428 * @new_name: new name
1429 * @new_ns: new namespace tag
1430 */
1431int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1432		     const char *new_name, const void *new_ns)
1433{
1434	struct kernfs_node *old_parent;
1435	const char *old_name = NULL;
1436	int error;
1437
1438	/* can't move or rename root */
1439	if (!kn->parent)
1440		return -EINVAL;
1441
1442	mutex_lock(&kernfs_mutex);
1443
1444	error = -ENOENT;
1445	if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1446	    (new_parent->flags & KERNFS_EMPTY_DIR))
1447		goto out;
1448
1449	error = 0;
1450	if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1451	    (strcmp(kn->name, new_name) == 0))
1452		goto out;	/* nothing to rename */
1453
1454	error = -EEXIST;
1455	if (kernfs_find_ns(new_parent, new_name, new_ns))
1456		goto out;
1457
1458	/* rename kernfs_node */
1459	if (strcmp(kn->name, new_name) != 0) {
1460		error = -ENOMEM;
1461		new_name = kstrdup_const(new_name, GFP_KERNEL);
1462		if (!new_name)
1463			goto out;
1464	} else {
1465		new_name = NULL;
1466	}
1467
1468	/*
1469	 * Move to the appropriate place in the appropriate directories rbtree.
1470	 */
1471	kernfs_unlink_sibling(kn);
1472	kernfs_get(new_parent);
1473
1474	/* rename_lock protects ->parent and ->name accessors */
1475	spin_lock_irq(&kernfs_rename_lock);
1476
1477	old_parent = kn->parent;
1478	kn->parent = new_parent;
1479
1480	kn->ns = new_ns;
1481	if (new_name) {
1482		old_name = kn->name;
1483		kn->name = new_name;
1484	}
1485
1486	spin_unlock_irq(&kernfs_rename_lock);
1487
1488	kn->hash = kernfs_name_hash(kn->name, kn->ns);
1489	kernfs_link_sibling(kn);
1490
1491	kernfs_put(old_parent);
1492	kfree_const(old_name);
1493
1494	error = 0;
1495 out:
1496	mutex_unlock(&kernfs_mutex);
1497	return error;
1498}
1499
1500/* Relationship between s_mode and the DT_xxx types */
1501static inline unsigned char dt_type(struct kernfs_node *kn)
1502{
1503	return (kn->mode >> 12) & 15;
1504}
1505
1506static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1507{
1508	kernfs_put(filp->private_data);
1509	return 0;
1510}
1511
1512static struct kernfs_node *kernfs_dir_pos(const void *ns,
1513	struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1514{
1515	if (pos) {
1516		int valid = kernfs_active(pos) &&
1517			pos->parent == parent && hash == pos->hash;
1518		kernfs_put(pos);
1519		if (!valid)
1520			pos = NULL;
1521	}
1522	if (!pos && (hash > 1) && (hash < INT_MAX)) {
1523		struct rb_node *node = parent->dir.children.rb_node;
1524		while (node) {
1525			pos = rb_to_kn(node);
1526
1527			if (hash < pos->hash)
1528				node = node->rb_left;
1529			else if (hash > pos->hash)
1530				node = node->rb_right;
1531			else
1532				break;
1533		}
1534	}
1535	/* Skip over entries which are dying/dead or in the wrong namespace */
1536	while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1537		struct rb_node *node = rb_next(&pos->rb);
1538		if (!node)
1539			pos = NULL;
1540		else
1541			pos = rb_to_kn(node);
1542	}
1543	return pos;
1544}
1545
1546static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1547	struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1548{
1549	pos = kernfs_dir_pos(ns, parent, ino, pos);
1550	if (pos) {
1551		do {
1552			struct rb_node *node = rb_next(&pos->rb);
1553			if (!node)
1554				pos = NULL;
1555			else
1556				pos = rb_to_kn(node);
1557		} while (pos && (!kernfs_active(pos) || pos->ns != ns));
1558	}
1559	return pos;
1560}
1561
1562static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1563{
1564	struct dentry *dentry = file->f_path.dentry;
1565	struct kernfs_node *parent = dentry->d_fsdata;
1566	struct kernfs_node *pos = file->private_data;
1567	const void *ns = NULL;
1568
1569	if (!dir_emit_dots(file, ctx))
1570		return 0;
1571	mutex_lock(&kernfs_mutex);
1572
1573	if (kernfs_ns_enabled(parent))
1574		ns = kernfs_info(dentry->d_sb)->ns;
1575
1576	for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1577	     pos;
1578	     pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1579		const char *name = pos->name;
1580		unsigned int type = dt_type(pos);
1581		int len = strlen(name);
1582		ino_t ino = pos->ino;
1583
1584		ctx->pos = pos->hash;
1585		file->private_data = pos;
1586		kernfs_get(pos);
1587
1588		mutex_unlock(&kernfs_mutex);
1589		if (!dir_emit(ctx, name, len, ino, type))
1590			return 0;
1591		mutex_lock(&kernfs_mutex);
1592	}
1593	mutex_unlock(&kernfs_mutex);
1594	file->private_data = NULL;
1595	ctx->pos = INT_MAX;
1596	return 0;
1597}
1598
1599const struct file_operations kernfs_dir_fops = {
1600	.read		= generic_read_dir,
1601	.iterate_shared	= kernfs_fop_readdir,
1602	.release	= kernfs_dir_fop_release,
1603	.llseek		= generic_file_llseek,
1604};
v5.14.15
   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, _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, _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	if (!kn || !atomic_dec_and_test(&kn->count))
 512		return;
 513	root = kernfs_root(kn);
 514 repeat:
 515	/*
 516	 * Moving/renaming is always done while holding reference.
 517	 * kn->parent won't change beneath us.
 518	 */
 519	parent = kn->parent;
 520
 521	WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
 522		  "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
 523		  parent ? parent->name : "", kn->name, atomic_read(&kn->active));
 524
 525	if (kernfs_type(kn) == KERNFS_LINK)
 526		kernfs_put(kn->symlink.target_kn);
 527
 528	kfree_const(kn->name);
 529
 530	if (kn->iattr) {
 
 
 
 531		simple_xattrs_free(&kn->iattr->xattrs);
 532		kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
 533	}
 534	spin_lock(&kernfs_idr_lock);
 535	idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
 536	spin_unlock(&kernfs_idr_lock);
 537	kmem_cache_free(kernfs_node_cache, kn);
 538
 539	kn = parent;
 540	if (kn) {
 541		if (atomic_dec_and_test(&kn->count))
 542			goto repeat;
 543	} else {
 544		/* just released the root kn, free @root too */
 545		idr_destroy(&root->ino_idr);
 546		kfree(root);
 547	}
 548}
 549EXPORT_SYMBOL_GPL(kernfs_put);
 550
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 551/**
 552 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
 553 * @dentry: the dentry in question
 554 *
 555 * Return the kernfs_node associated with @dentry.  If @dentry is not a
 556 * kernfs one, %NULL is returned.
 557 *
 558 * While the returned kernfs_node will stay accessible as long as @dentry
 559 * is accessible, the returned node can be in any state and the caller is
 560 * fully responsible for determining what's accessible.
 561 */
 562struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
 563{
 564	if (dentry->d_sb->s_op == &kernfs_sops)
 565		return kernfs_dentry_node(dentry);
 566	return NULL;
 567}
 568
 569static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
 570					     struct kernfs_node *parent,
 571					     const char *name, umode_t mode,
 572					     kuid_t uid, kgid_t gid,
 573					     unsigned flags)
 574{
 575	struct kernfs_node *kn;
 576	u32 id_highbits;
 577	int ret;
 578
 579	name = kstrdup_const(name, GFP_KERNEL);
 580	if (!name)
 581		return NULL;
 582
 583	kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
 584	if (!kn)
 585		goto err_out1;
 586
 587	idr_preload(GFP_KERNEL);
 588	spin_lock(&kernfs_idr_lock);
 589	ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
 590	if (ret >= 0 && ret < root->last_id_lowbits)
 591		root->id_highbits++;
 592	id_highbits = root->id_highbits;
 593	root->last_id_lowbits = ret;
 594	spin_unlock(&kernfs_idr_lock);
 595	idr_preload_end();
 596	if (ret < 0)
 597		goto err_out2;
 598
 599	kn->id = (u64)id_highbits << 32 | ret;
 600
 601	atomic_set(&kn->count, 1);
 602	atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
 603	RB_CLEAR_NODE(&kn->rb);
 604
 605	kn->name = name;
 606	kn->mode = mode;
 607	kn->flags = flags;
 608
 609	if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
 610		struct iattr iattr = {
 611			.ia_valid = ATTR_UID | ATTR_GID,
 612			.ia_uid = uid,
 613			.ia_gid = gid,
 614		};
 615
 616		ret = __kernfs_setattr(kn, &iattr);
 617		if (ret < 0)
 618			goto err_out3;
 619	}
 620
 621	if (parent) {
 622		ret = security_kernfs_init_security(parent, kn);
 623		if (ret)
 624			goto err_out3;
 625	}
 626
 627	return kn;
 628
 629 err_out3:
 630	idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
 631 err_out2:
 632	kmem_cache_free(kernfs_node_cache, kn);
 633 err_out1:
 634	kfree_const(name);
 635	return NULL;
 636}
 637
 638struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
 639				    const char *name, umode_t mode,
 640				    kuid_t uid, kgid_t gid,
 641				    unsigned flags)
 642{
 643	struct kernfs_node *kn;
 644
 645	kn = __kernfs_new_node(kernfs_root(parent), parent,
 646			       name, mode, uid, gid, flags);
 647	if (kn) {
 648		kernfs_get(parent);
 649		kn->parent = parent;
 650	}
 651	return kn;
 652}
 653
 654/*
 655 * kernfs_find_and_get_node_by_id - get kernfs_node from node id
 656 * @root: the kernfs root
 657 * @id: the target node id
 658 *
 659 * @id's lower 32bits encode ino and upper gen.  If the gen portion is
 660 * zero, all generations are matched.
 661 *
 662 * RETURNS:
 663 * NULL on failure. Return a kernfs node with reference counter incremented
 664 */
 665struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
 666						   u64 id)
 667{
 668	struct kernfs_node *kn;
 669	ino_t ino = kernfs_id_ino(id);
 670	u32 gen = kernfs_id_gen(id);
 671
 672	spin_lock(&kernfs_idr_lock);
 673
 674	kn = idr_find(&root->ino_idr, (u32)ino);
 675	if (!kn)
 676		goto err_unlock;
 677
 678	if (sizeof(ino_t) >= sizeof(u64)) {
 679		/* we looked up with the low 32bits, compare the whole */
 680		if (kernfs_ino(kn) != ino)
 681			goto err_unlock;
 682	} else {
 683		/* 0 matches all generations */
 684		if (unlikely(gen && kernfs_gen(kn) != gen))
 685			goto err_unlock;
 686	}
 687
 688	/*
 689	 * ACTIVATED is protected with kernfs_mutex but it was clear when
 690	 * @kn was added to idr and we just wanna see it set.  No need to
 691	 * grab kernfs_mutex.
 692	 */
 693	if (unlikely(!(kn->flags & KERNFS_ACTIVATED) ||
 694		     !atomic_inc_not_zero(&kn->count)))
 695		goto err_unlock;
 696
 697	spin_unlock(&kernfs_idr_lock);
 698	return kn;
 699err_unlock:
 700	spin_unlock(&kernfs_idr_lock);
 701	return NULL;
 702}
 703
 704/**
 705 *	kernfs_add_one - add kernfs_node to parent without warning
 706 *	@kn: kernfs_node to be added
 707 *
 708 *	The caller must already have initialized @kn->parent.  This
 709 *	function increments nlink of the parent's inode if @kn is a
 710 *	directory and link into the children list of the parent.
 711 *
 712 *	RETURNS:
 713 *	0 on success, -EEXIST if entry with the given name already
 714 *	exists.
 715 */
 716int kernfs_add_one(struct kernfs_node *kn)
 717{
 718	struct kernfs_node *parent = kn->parent;
 719	struct kernfs_iattrs *ps_iattr;
 720	bool has_ns;
 721	int ret;
 722
 723	mutex_lock(&kernfs_mutex);
 724
 725	ret = -EINVAL;
 726	has_ns = kernfs_ns_enabled(parent);
 727	if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
 728		 has_ns ? "required" : "invalid", parent->name, kn->name))
 729		goto out_unlock;
 730
 731	if (kernfs_type(parent) != KERNFS_DIR)
 732		goto out_unlock;
 733
 734	ret = -ENOENT;
 735	if (parent->flags & KERNFS_EMPTY_DIR)
 736		goto out_unlock;
 737
 738	if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
 739		goto out_unlock;
 740
 741	kn->hash = kernfs_name_hash(kn->name, kn->ns);
 742
 743	ret = kernfs_link_sibling(kn);
 744	if (ret)
 745		goto out_unlock;
 746
 747	/* Update timestamps on the parent */
 748	ps_iattr = parent->iattr;
 749	if (ps_iattr) {
 750		ktime_get_real_ts64(&ps_iattr->ia_ctime);
 751		ps_iattr->ia_mtime = ps_iattr->ia_ctime;
 
 752	}
 753
 754	mutex_unlock(&kernfs_mutex);
 755
 756	/*
 757	 * Activate the new node unless CREATE_DEACTIVATED is requested.
 758	 * If not activated here, the kernfs user is responsible for
 759	 * activating the node with kernfs_activate().  A node which hasn't
 760	 * been activated is not visible to userland and its removal won't
 761	 * trigger deactivation.
 762	 */
 763	if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
 764		kernfs_activate(kn);
 765	return 0;
 766
 767out_unlock:
 768	mutex_unlock(&kernfs_mutex);
 769	return ret;
 770}
 771
 772/**
 773 * kernfs_find_ns - find kernfs_node with the given name
 774 * @parent: kernfs_node to search under
 775 * @name: name to look for
 776 * @ns: the namespace tag to use
 777 *
 778 * Look for kernfs_node with name @name under @parent.  Returns pointer to
 779 * the found kernfs_node on success, %NULL on failure.
 780 */
 781static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
 782					  const unsigned char *name,
 783					  const void *ns)
 784{
 785	struct rb_node *node = parent->dir.children.rb_node;
 786	bool has_ns = kernfs_ns_enabled(parent);
 787	unsigned int hash;
 788
 789	lockdep_assert_held(&kernfs_mutex);
 790
 791	if (has_ns != (bool)ns) {
 792		WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
 793		     has_ns ? "required" : "invalid", parent->name, name);
 794		return NULL;
 795	}
 796
 797	hash = kernfs_name_hash(name, ns);
 798	while (node) {
 799		struct kernfs_node *kn;
 800		int result;
 801
 802		kn = rb_to_kn(node);
 803		result = kernfs_name_compare(hash, name, ns, kn);
 804		if (result < 0)
 805			node = node->rb_left;
 806		else if (result > 0)
 807			node = node->rb_right;
 808		else
 809			return kn;
 810	}
 811	return NULL;
 812}
 813
 814static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
 815					  const unsigned char *path,
 816					  const void *ns)
 817{
 818	size_t len;
 819	char *p, *name;
 820
 821	lockdep_assert_held(&kernfs_mutex);
 822
 823	/* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
 824	spin_lock_irq(&kernfs_rename_lock);
 825
 826	len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
 827
 828	if (len >= sizeof(kernfs_pr_cont_buf)) {
 829		spin_unlock_irq(&kernfs_rename_lock);
 830		return NULL;
 831	}
 832
 833	p = kernfs_pr_cont_buf;
 834
 835	while ((name = strsep(&p, "/")) && parent) {
 836		if (*name == '\0')
 837			continue;
 838		parent = kernfs_find_ns(parent, name, ns);
 839	}
 840
 841	spin_unlock_irq(&kernfs_rename_lock);
 842
 843	return parent;
 844}
 845
 846/**
 847 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
 848 * @parent: kernfs_node to search under
 849 * @name: name to look for
 850 * @ns: the namespace tag to use
 851 *
 852 * Look for kernfs_node with name @name under @parent and get a reference
 853 * if found.  This function may sleep and returns pointer to the found
 854 * kernfs_node on success, %NULL on failure.
 855 */
 856struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
 857					   const char *name, const void *ns)
 858{
 859	struct kernfs_node *kn;
 860
 861	mutex_lock(&kernfs_mutex);
 862	kn = kernfs_find_ns(parent, name, ns);
 863	kernfs_get(kn);
 864	mutex_unlock(&kernfs_mutex);
 865
 866	return kn;
 867}
 868EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
 869
 870/**
 871 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
 872 * @parent: kernfs_node to search under
 873 * @path: path to look for
 874 * @ns: the namespace tag to use
 875 *
 876 * Look for kernfs_node with path @path under @parent and get a reference
 877 * if found.  This function may sleep and returns pointer to the found
 878 * kernfs_node on success, %NULL on failure.
 879 */
 880struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
 881					   const char *path, const void *ns)
 882{
 883	struct kernfs_node *kn;
 884
 885	mutex_lock(&kernfs_mutex);
 886	kn = kernfs_walk_ns(parent, path, ns);
 887	kernfs_get(kn);
 888	mutex_unlock(&kernfs_mutex);
 889
 890	return kn;
 891}
 892
 893/**
 894 * kernfs_create_root - create a new kernfs hierarchy
 895 * @scops: optional syscall operations for the hierarchy
 896 * @flags: KERNFS_ROOT_* flags
 897 * @priv: opaque data associated with the new directory
 898 *
 899 * Returns the root of the new hierarchy on success, ERR_PTR() value on
 900 * failure.
 901 */
 902struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
 903				       unsigned int flags, void *priv)
 904{
 905	struct kernfs_root *root;
 906	struct kernfs_node *kn;
 907
 908	root = kzalloc(sizeof(*root), GFP_KERNEL);
 909	if (!root)
 910		return ERR_PTR(-ENOMEM);
 911
 912	idr_init(&root->ino_idr);
 913	INIT_LIST_HEAD(&root->supers);
 914
 915	/*
 916	 * On 64bit ino setups, id is ino.  On 32bit, low 32bits are ino.
 917	 * High bits generation.  The starting value for both ino and
 918	 * genenration is 1.  Initialize upper 32bit allocation
 919	 * accordingly.
 920	 */
 921	if (sizeof(ino_t) >= sizeof(u64))
 922		root->id_highbits = 0;
 923	else
 924		root->id_highbits = 1;
 925
 926	kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
 927			       GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
 928			       KERNFS_DIR);
 929	if (!kn) {
 930		idr_destroy(&root->ino_idr);
 931		kfree(root);
 932		return ERR_PTR(-ENOMEM);
 933	}
 934
 935	kn->priv = priv;
 936	kn->dir.root = root;
 937
 938	root->syscall_ops = scops;
 939	root->flags = flags;
 940	root->kn = kn;
 941	init_waitqueue_head(&root->deactivate_waitq);
 942
 943	if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
 944		kernfs_activate(kn);
 945
 946	return root;
 947}
 948
 949/**
 950 * kernfs_destroy_root - destroy a kernfs hierarchy
 951 * @root: root of the hierarchy to destroy
 952 *
 953 * Destroy the hierarchy anchored at @root by removing all existing
 954 * directories and destroying @root.
 955 */
 956void kernfs_destroy_root(struct kernfs_root *root)
 957{
 958	kernfs_remove(root->kn);	/* will also free @root */
 959}
 960
 961/**
 962 * kernfs_create_dir_ns - create a directory
 963 * @parent: parent in which to create a new directory
 964 * @name: name of the new directory
 965 * @mode: mode of the new directory
 966 * @uid: uid of the new directory
 967 * @gid: gid of the new directory
 968 * @priv: opaque data associated with the new directory
 969 * @ns: optional namespace tag of the directory
 970 *
 971 * Returns the created node on success, ERR_PTR() value on failure.
 972 */
 973struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
 974					 const char *name, umode_t mode,
 975					 kuid_t uid, kgid_t gid,
 976					 void *priv, const void *ns)
 977{
 978	struct kernfs_node *kn;
 979	int rc;
 980
 981	/* allocate */
 982	kn = kernfs_new_node(parent, name, mode | S_IFDIR,
 983			     uid, gid, KERNFS_DIR);
 984	if (!kn)
 985		return ERR_PTR(-ENOMEM);
 986
 987	kn->dir.root = parent->dir.root;
 988	kn->ns = ns;
 989	kn->priv = priv;
 990
 991	/* link in */
 992	rc = kernfs_add_one(kn);
 993	if (!rc)
 994		return kn;
 995
 996	kernfs_put(kn);
 997	return ERR_PTR(rc);
 998}
 999
1000/**
1001 * kernfs_create_empty_dir - create an always empty directory
1002 * @parent: parent in which to create a new directory
1003 * @name: name of the new directory
1004 *
1005 * Returns the created node on success, ERR_PTR() value on failure.
1006 */
1007struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1008					    const char *name)
1009{
1010	struct kernfs_node *kn;
1011	int rc;
1012
1013	/* allocate */
1014	kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1015			     GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1016	if (!kn)
1017		return ERR_PTR(-ENOMEM);
1018
1019	kn->flags |= KERNFS_EMPTY_DIR;
1020	kn->dir.root = parent->dir.root;
1021	kn->ns = NULL;
1022	kn->priv = NULL;
1023
1024	/* link in */
1025	rc = kernfs_add_one(kn);
1026	if (!rc)
1027		return kn;
1028
1029	kernfs_put(kn);
1030	return ERR_PTR(rc);
1031}
1032
1033static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
1034{
1035	struct kernfs_node *kn;
1036
1037	if (flags & LOOKUP_RCU)
1038		return -ECHILD;
1039
1040	/* Always perform fresh lookup for negatives */
1041	if (d_really_is_negative(dentry))
1042		goto out_bad_unlocked;
1043
1044	kn = kernfs_dentry_node(dentry);
1045	mutex_lock(&kernfs_mutex);
1046
1047	/* The kernfs node has been deactivated */
1048	if (!kernfs_active(kn))
1049		goto out_bad;
1050
1051	/* The kernfs node has been moved? */
1052	if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
1053		goto out_bad;
1054
1055	/* The kernfs node has been renamed */
1056	if (strcmp(dentry->d_name.name, kn->name) != 0)
1057		goto out_bad;
1058
1059	/* The kernfs node has been moved to a different namespace */
1060	if (kn->parent && kernfs_ns_enabled(kn->parent) &&
1061	    kernfs_info(dentry->d_sb)->ns != kn->ns)
1062		goto out_bad;
1063
1064	mutex_unlock(&kernfs_mutex);
1065	return 1;
1066out_bad:
1067	mutex_unlock(&kernfs_mutex);
1068out_bad_unlocked:
1069	return 0;
1070}
1071
1072const struct dentry_operations kernfs_dops = {
1073	.d_revalidate	= kernfs_dop_revalidate,
1074};
1075
1076static struct dentry *kernfs_iop_lookup(struct inode *dir,
1077					struct dentry *dentry,
1078					unsigned int flags)
1079{
1080	struct dentry *ret;
1081	struct kernfs_node *parent = dir->i_private;
1082	struct kernfs_node *kn;
1083	struct inode *inode;
1084	const void *ns = NULL;
1085
1086	mutex_lock(&kernfs_mutex);
1087
1088	if (kernfs_ns_enabled(parent))
1089		ns = kernfs_info(dir->i_sb)->ns;
1090
1091	kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1092
1093	/* no such entry */
1094	if (!kn || !kernfs_active(kn)) {
1095		ret = NULL;
1096		goto out_unlock;
1097	}
 
 
1098
1099	/* attach dentry and inode */
1100	inode = kernfs_get_inode(dir->i_sb, kn);
1101	if (!inode) {
1102		ret = ERR_PTR(-ENOMEM);
1103		goto out_unlock;
1104	}
1105
1106	/* instantiate and hash dentry */
1107	ret = d_splice_alias(inode, dentry);
1108 out_unlock:
1109	mutex_unlock(&kernfs_mutex);
1110	return ret;
1111}
1112
1113static int kernfs_iop_mkdir(struct user_namespace *mnt_userns,
1114			    struct inode *dir, struct dentry *dentry,
1115			    umode_t mode)
1116{
1117	struct kernfs_node *parent = dir->i_private;
1118	struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1119	int ret;
1120
1121	if (!scops || !scops->mkdir)
1122		return -EPERM;
1123
1124	if (!kernfs_get_active(parent))
1125		return -ENODEV;
1126
1127	ret = scops->mkdir(parent, dentry->d_name.name, mode);
1128
1129	kernfs_put_active(parent);
1130	return ret;
1131}
1132
1133static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1134{
1135	struct kernfs_node *kn  = kernfs_dentry_node(dentry);
1136	struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1137	int ret;
1138
1139	if (!scops || !scops->rmdir)
1140		return -EPERM;
1141
1142	if (!kernfs_get_active(kn))
1143		return -ENODEV;
1144
1145	ret = scops->rmdir(kn);
1146
1147	kernfs_put_active(kn);
1148	return ret;
1149}
1150
1151static int kernfs_iop_rename(struct user_namespace *mnt_userns,
1152			     struct inode *old_dir, struct dentry *old_dentry,
1153			     struct inode *new_dir, struct dentry *new_dentry,
1154			     unsigned int flags)
1155{
1156	struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1157	struct kernfs_node *new_parent = new_dir->i_private;
1158	struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1159	int ret;
1160
1161	if (flags)
1162		return -EINVAL;
1163
1164	if (!scops || !scops->rename)
1165		return -EPERM;
1166
1167	if (!kernfs_get_active(kn))
1168		return -ENODEV;
1169
1170	if (!kernfs_get_active(new_parent)) {
1171		kernfs_put_active(kn);
1172		return -ENODEV;
1173	}
1174
1175	ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1176
1177	kernfs_put_active(new_parent);
1178	kernfs_put_active(kn);
1179	return ret;
1180}
1181
1182const struct inode_operations kernfs_dir_iops = {
1183	.lookup		= kernfs_iop_lookup,
1184	.permission	= kernfs_iop_permission,
1185	.setattr	= kernfs_iop_setattr,
1186	.getattr	= kernfs_iop_getattr,
1187	.listxattr	= kernfs_iop_listxattr,
1188
1189	.mkdir		= kernfs_iop_mkdir,
1190	.rmdir		= kernfs_iop_rmdir,
1191	.rename		= kernfs_iop_rename,
1192};
1193
1194static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1195{
1196	struct kernfs_node *last;
1197
1198	while (true) {
1199		struct rb_node *rbn;
1200
1201		last = pos;
1202
1203		if (kernfs_type(pos) != KERNFS_DIR)
1204			break;
1205
1206		rbn = rb_first(&pos->dir.children);
1207		if (!rbn)
1208			break;
1209
1210		pos = rb_to_kn(rbn);
1211	}
1212
1213	return last;
1214}
1215
1216/**
1217 * kernfs_next_descendant_post - find the next descendant for post-order walk
1218 * @pos: the current position (%NULL to initiate traversal)
1219 * @root: kernfs_node whose descendants to walk
1220 *
1221 * Find the next descendant to visit for post-order traversal of @root's
1222 * descendants.  @root is included in the iteration and the last node to be
1223 * visited.
1224 */
1225static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1226						       struct kernfs_node *root)
1227{
1228	struct rb_node *rbn;
1229
1230	lockdep_assert_held(&kernfs_mutex);
1231
1232	/* if first iteration, visit leftmost descendant which may be root */
1233	if (!pos)
1234		return kernfs_leftmost_descendant(root);
1235
1236	/* if we visited @root, we're done */
1237	if (pos == root)
1238		return NULL;
1239
1240	/* if there's an unvisited sibling, visit its leftmost descendant */
1241	rbn = rb_next(&pos->rb);
1242	if (rbn)
1243		return kernfs_leftmost_descendant(rb_to_kn(rbn));
1244
1245	/* no sibling left, visit parent */
1246	return pos->parent;
1247}
1248
1249/**
1250 * kernfs_activate - activate a node which started deactivated
1251 * @kn: kernfs_node whose subtree is to be activated
1252 *
1253 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1254 * needs to be explicitly activated.  A node which hasn't been activated
1255 * isn't visible to userland and deactivation is skipped during its
1256 * removal.  This is useful to construct atomic init sequences where
1257 * creation of multiple nodes should either succeed or fail atomically.
1258 *
1259 * The caller is responsible for ensuring that this function is not called
1260 * after kernfs_remove*() is invoked on @kn.
1261 */
1262void kernfs_activate(struct kernfs_node *kn)
1263{
1264	struct kernfs_node *pos;
1265
1266	mutex_lock(&kernfs_mutex);
1267
1268	pos = NULL;
1269	while ((pos = kernfs_next_descendant_post(pos, kn))) {
1270		if (pos->flags & KERNFS_ACTIVATED)
1271			continue;
1272
1273		WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1274		WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1275
1276		atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1277		pos->flags |= KERNFS_ACTIVATED;
1278	}
1279
1280	mutex_unlock(&kernfs_mutex);
1281}
1282
1283static void __kernfs_remove(struct kernfs_node *kn)
1284{
1285	struct kernfs_node *pos;
1286
1287	lockdep_assert_held(&kernfs_mutex);
1288
1289	/*
1290	 * Short-circuit if non-root @kn has already finished removal.
1291	 * This is for kernfs_remove_self() which plays with active ref
1292	 * after removal.
1293	 */
1294	if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1295		return;
1296
1297	pr_debug("kernfs %s: removing\n", kn->name);
1298
1299	/* prevent any new usage under @kn by deactivating all nodes */
1300	pos = NULL;
1301	while ((pos = kernfs_next_descendant_post(pos, kn)))
1302		if (kernfs_active(pos))
1303			atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1304
1305	/* deactivate and unlink the subtree node-by-node */
1306	do {
1307		pos = kernfs_leftmost_descendant(kn);
1308
1309		/*
1310		 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1311		 * base ref could have been put by someone else by the time
1312		 * the function returns.  Make sure it doesn't go away
1313		 * underneath us.
1314		 */
1315		kernfs_get(pos);
1316
1317		/*
1318		 * Drain iff @kn was activated.  This avoids draining and
1319		 * its lockdep annotations for nodes which have never been
1320		 * activated and allows embedding kernfs_remove() in create
1321		 * error paths without worrying about draining.
1322		 */
1323		if (kn->flags & KERNFS_ACTIVATED)
1324			kernfs_drain(pos);
1325		else
1326			WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1327
1328		/*
1329		 * kernfs_unlink_sibling() succeeds once per node.  Use it
1330		 * to decide who's responsible for cleanups.
1331		 */
1332		if (!pos->parent || kernfs_unlink_sibling(pos)) {
1333			struct kernfs_iattrs *ps_iattr =
1334				pos->parent ? pos->parent->iattr : NULL;
1335
1336			/* update timestamps on the parent */
1337			if (ps_iattr) {
1338				ktime_get_real_ts64(&ps_iattr->ia_ctime);
1339				ps_iattr->ia_mtime = ps_iattr->ia_ctime;
 
1340			}
1341
1342			kernfs_put(pos);
1343		}
1344
1345		kernfs_put(pos);
1346	} while (pos != kn);
1347}
1348
1349/**
1350 * kernfs_remove - remove a kernfs_node recursively
1351 * @kn: the kernfs_node to remove
1352 *
1353 * Remove @kn along with all its subdirectories and files.
1354 */
1355void kernfs_remove(struct kernfs_node *kn)
1356{
1357	mutex_lock(&kernfs_mutex);
1358	__kernfs_remove(kn);
1359	mutex_unlock(&kernfs_mutex);
1360}
1361
1362/**
1363 * kernfs_break_active_protection - break out of active protection
1364 * @kn: the self kernfs_node
1365 *
1366 * The caller must be running off of a kernfs operation which is invoked
1367 * with an active reference - e.g. one of kernfs_ops.  Each invocation of
1368 * this function must also be matched with an invocation of
1369 * kernfs_unbreak_active_protection().
1370 *
1371 * This function releases the active reference of @kn the caller is
1372 * holding.  Once this function is called, @kn may be removed at any point
1373 * and the caller is solely responsible for ensuring that the objects it
1374 * dereferences are accessible.
1375 */
1376void kernfs_break_active_protection(struct kernfs_node *kn)
1377{
1378	/*
1379	 * Take out ourself out of the active ref dependency chain.  If
1380	 * we're called without an active ref, lockdep will complain.
1381	 */
1382	kernfs_put_active(kn);
1383}
1384
1385/**
1386 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1387 * @kn: the self kernfs_node
1388 *
1389 * If kernfs_break_active_protection() was called, this function must be
1390 * invoked before finishing the kernfs operation.  Note that while this
1391 * function restores the active reference, it doesn't and can't actually
1392 * restore the active protection - @kn may already or be in the process of
1393 * being removed.  Once kernfs_break_active_protection() is invoked, that
1394 * protection is irreversibly gone for the kernfs operation instance.
1395 *
1396 * While this function may be called at any point after
1397 * kernfs_break_active_protection() is invoked, its most useful location
1398 * would be right before the enclosing kernfs operation returns.
1399 */
1400void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1401{
1402	/*
1403	 * @kn->active could be in any state; however, the increment we do
1404	 * here will be undone as soon as the enclosing kernfs operation
1405	 * finishes and this temporary bump can't break anything.  If @kn
1406	 * is alive, nothing changes.  If @kn is being deactivated, the
1407	 * soon-to-follow put will either finish deactivation or restore
1408	 * deactivated state.  If @kn is already removed, the temporary
1409	 * bump is guaranteed to be gone before @kn is released.
1410	 */
1411	atomic_inc(&kn->active);
1412	if (kernfs_lockdep(kn))
1413		rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1414}
1415
1416/**
1417 * kernfs_remove_self - remove a kernfs_node from its own method
1418 * @kn: the self kernfs_node to remove
1419 *
1420 * The caller must be running off of a kernfs operation which is invoked
1421 * with an active reference - e.g. one of kernfs_ops.  This can be used to
1422 * implement a file operation which deletes itself.
1423 *
1424 * For example, the "delete" file for a sysfs device directory can be
1425 * implemented by invoking kernfs_remove_self() on the "delete" file
1426 * itself.  This function breaks the circular dependency of trying to
1427 * deactivate self while holding an active ref itself.  It isn't necessary
1428 * to modify the usual removal path to use kernfs_remove_self().  The
1429 * "delete" implementation can simply invoke kernfs_remove_self() on self
1430 * before proceeding with the usual removal path.  kernfs will ignore later
1431 * kernfs_remove() on self.
1432 *
1433 * kernfs_remove_self() can be called multiple times concurrently on the
1434 * same kernfs_node.  Only the first one actually performs removal and
1435 * returns %true.  All others will wait until the kernfs operation which
1436 * won self-removal finishes and return %false.  Note that the losers wait
1437 * for the completion of not only the winning kernfs_remove_self() but also
1438 * the whole kernfs_ops which won the arbitration.  This can be used to
1439 * guarantee, for example, all concurrent writes to a "delete" file to
1440 * finish only after the whole operation is complete.
1441 */
1442bool kernfs_remove_self(struct kernfs_node *kn)
1443{
1444	bool ret;
1445
1446	mutex_lock(&kernfs_mutex);
1447	kernfs_break_active_protection(kn);
1448
1449	/*
1450	 * SUICIDAL is used to arbitrate among competing invocations.  Only
1451	 * the first one will actually perform removal.  When the removal
1452	 * is complete, SUICIDED is set and the active ref is restored
1453	 * while holding kernfs_mutex.  The ones which lost arbitration
1454	 * waits for SUICDED && drained which can happen only after the
1455	 * enclosing kernfs operation which executed the winning instance
1456	 * of kernfs_remove_self() finished.
1457	 */
1458	if (!(kn->flags & KERNFS_SUICIDAL)) {
1459		kn->flags |= KERNFS_SUICIDAL;
1460		__kernfs_remove(kn);
1461		kn->flags |= KERNFS_SUICIDED;
1462		ret = true;
1463	} else {
1464		wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1465		DEFINE_WAIT(wait);
1466
1467		while (true) {
1468			prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1469
1470			if ((kn->flags & KERNFS_SUICIDED) &&
1471			    atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1472				break;
1473
1474			mutex_unlock(&kernfs_mutex);
1475			schedule();
1476			mutex_lock(&kernfs_mutex);
1477		}
1478		finish_wait(waitq, &wait);
1479		WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1480		ret = false;
1481	}
1482
1483	/*
1484	 * This must be done while holding kernfs_mutex; otherwise, waiting
1485	 * for SUICIDED && deactivated could finish prematurely.
1486	 */
1487	kernfs_unbreak_active_protection(kn);
1488
1489	mutex_unlock(&kernfs_mutex);
1490	return ret;
1491}
1492
1493/**
1494 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1495 * @parent: parent of the target
1496 * @name: name of the kernfs_node to remove
1497 * @ns: namespace tag of the kernfs_node to remove
1498 *
1499 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1500 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1501 */
1502int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1503			     const void *ns)
1504{
1505	struct kernfs_node *kn;
1506
1507	if (!parent) {
1508		WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1509			name);
1510		return -ENOENT;
1511	}
1512
1513	mutex_lock(&kernfs_mutex);
1514
1515	kn = kernfs_find_ns(parent, name, ns);
1516	if (kn)
1517		__kernfs_remove(kn);
1518
1519	mutex_unlock(&kernfs_mutex);
1520
1521	if (kn)
1522		return 0;
1523	else
1524		return -ENOENT;
1525}
1526
1527/**
1528 * kernfs_rename_ns - move and rename a kernfs_node
1529 * @kn: target node
1530 * @new_parent: new parent to put @sd under
1531 * @new_name: new name
1532 * @new_ns: new namespace tag
1533 */
1534int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1535		     const char *new_name, const void *new_ns)
1536{
1537	struct kernfs_node *old_parent;
1538	const char *old_name = NULL;
1539	int error;
1540
1541	/* can't move or rename root */
1542	if (!kn->parent)
1543		return -EINVAL;
1544
1545	mutex_lock(&kernfs_mutex);
1546
1547	error = -ENOENT;
1548	if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1549	    (new_parent->flags & KERNFS_EMPTY_DIR))
1550		goto out;
1551
1552	error = 0;
1553	if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1554	    (strcmp(kn->name, new_name) == 0))
1555		goto out;	/* nothing to rename */
1556
1557	error = -EEXIST;
1558	if (kernfs_find_ns(new_parent, new_name, new_ns))
1559		goto out;
1560
1561	/* rename kernfs_node */
1562	if (strcmp(kn->name, new_name) != 0) {
1563		error = -ENOMEM;
1564		new_name = kstrdup_const(new_name, GFP_KERNEL);
1565		if (!new_name)
1566			goto out;
1567	} else {
1568		new_name = NULL;
1569	}
1570
1571	/*
1572	 * Move to the appropriate place in the appropriate directories rbtree.
1573	 */
1574	kernfs_unlink_sibling(kn);
1575	kernfs_get(new_parent);
1576
1577	/* rename_lock protects ->parent and ->name accessors */
1578	spin_lock_irq(&kernfs_rename_lock);
1579
1580	old_parent = kn->parent;
1581	kn->parent = new_parent;
1582
1583	kn->ns = new_ns;
1584	if (new_name) {
1585		old_name = kn->name;
1586		kn->name = new_name;
1587	}
1588
1589	spin_unlock_irq(&kernfs_rename_lock);
1590
1591	kn->hash = kernfs_name_hash(kn->name, kn->ns);
1592	kernfs_link_sibling(kn);
1593
1594	kernfs_put(old_parent);
1595	kfree_const(old_name);
1596
1597	error = 0;
1598 out:
1599	mutex_unlock(&kernfs_mutex);
1600	return error;
1601}
1602
1603/* Relationship between mode and the DT_xxx types */
1604static inline unsigned char dt_type(struct kernfs_node *kn)
1605{
1606	return (kn->mode >> 12) & 15;
1607}
1608
1609static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1610{
1611	kernfs_put(filp->private_data);
1612	return 0;
1613}
1614
1615static struct kernfs_node *kernfs_dir_pos(const void *ns,
1616	struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1617{
1618	if (pos) {
1619		int valid = kernfs_active(pos) &&
1620			pos->parent == parent && hash == pos->hash;
1621		kernfs_put(pos);
1622		if (!valid)
1623			pos = NULL;
1624	}
1625	if (!pos && (hash > 1) && (hash < INT_MAX)) {
1626		struct rb_node *node = parent->dir.children.rb_node;
1627		while (node) {
1628			pos = rb_to_kn(node);
1629
1630			if (hash < pos->hash)
1631				node = node->rb_left;
1632			else if (hash > pos->hash)
1633				node = node->rb_right;
1634			else
1635				break;
1636		}
1637	}
1638	/* Skip over entries which are dying/dead or in the wrong namespace */
1639	while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1640		struct rb_node *node = rb_next(&pos->rb);
1641		if (!node)
1642			pos = NULL;
1643		else
1644			pos = rb_to_kn(node);
1645	}
1646	return pos;
1647}
1648
1649static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1650	struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1651{
1652	pos = kernfs_dir_pos(ns, parent, ino, pos);
1653	if (pos) {
1654		do {
1655			struct rb_node *node = rb_next(&pos->rb);
1656			if (!node)
1657				pos = NULL;
1658			else
1659				pos = rb_to_kn(node);
1660		} while (pos && (!kernfs_active(pos) || pos->ns != ns));
1661	}
1662	return pos;
1663}
1664
1665static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1666{
1667	struct dentry *dentry = file->f_path.dentry;
1668	struct kernfs_node *parent = kernfs_dentry_node(dentry);
1669	struct kernfs_node *pos = file->private_data;
1670	const void *ns = NULL;
1671
1672	if (!dir_emit_dots(file, ctx))
1673		return 0;
1674	mutex_lock(&kernfs_mutex);
1675
1676	if (kernfs_ns_enabled(parent))
1677		ns = kernfs_info(dentry->d_sb)->ns;
1678
1679	for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1680	     pos;
1681	     pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1682		const char *name = pos->name;
1683		unsigned int type = dt_type(pos);
1684		int len = strlen(name);
1685		ino_t ino = kernfs_ino(pos);
1686
1687		ctx->pos = pos->hash;
1688		file->private_data = pos;
1689		kernfs_get(pos);
1690
1691		mutex_unlock(&kernfs_mutex);
1692		if (!dir_emit(ctx, name, len, ino, type))
1693			return 0;
1694		mutex_lock(&kernfs_mutex);
1695	}
1696	mutex_unlock(&kernfs_mutex);
1697	file->private_data = NULL;
1698	ctx->pos = INT_MAX;
1699	return 0;
1700}
1701
1702const struct file_operations kernfs_dir_fops = {
1703	.read		= generic_read_dir,
1704	.iterate_shared	= kernfs_fop_readdir,
1705	.release	= kernfs_dir_fop_release,
1706	.llseek		= generic_file_llseek,
1707};