1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * dir.c - Operations for configfs directories.
   4 *
   5 * Based on sysfs:
   6 * 	sysfs is Copyright (C) 2001, 2002, 2003 Patrick Mochel
   7 *
   8 * configfs Copyright (C) 2005 Oracle.  All rights reserved.
   9 */
  10
  11#undef DEBUG
  12
  13#include <linux/fs.h>
  14#include <linux/fsnotify.h>
  15#include <linux/mount.h>
  16#include <linux/module.h>
  17#include <linux/slab.h>
  18#include <linux/err.h>
  19
  20#include <linux/configfs.h>
  21#include "configfs_internal.h"
  22
  23/*
  24 * Protects mutations of configfs_dirent linkage together with proper i_mutex
  25 * Also protects mutations of symlinks linkage to target configfs_dirent
  26 * Mutators of configfs_dirent linkage must *both* have the proper inode locked
  27 * and configfs_dirent_lock locked, in that order.
  28 * This allows one to safely traverse configfs_dirent trees and symlinks without
  29 * having to lock inodes.
  30 *
  31 * Protects setting of CONFIGFS_USET_DROPPING: checking the flag
  32 * unlocked is not reliable unless in detach_groups() called from
  33 * rmdir()/unregister() and from configfs_attach_group()
  34 */
  35DEFINE_SPINLOCK(configfs_dirent_lock);
  36
  37/*
  38 * All of link_obj/unlink_obj/link_group/unlink_group require that
  39 * subsys->su_mutex is held.
  40 * But parent configfs_subsystem is NULL when config_item is root.
  41 * Use this mutex when config_item is root.
  42 */
  43static DEFINE_MUTEX(configfs_subsystem_mutex);
  44
  45static void configfs_d_iput(struct dentry * dentry,
  46			    struct inode * inode)
  47{
  48	struct configfs_dirent *sd = dentry->d_fsdata;
  49
  50	if (sd) {
  51		/* Coordinate with configfs_readdir */
  52		spin_lock(&configfs_dirent_lock);
  53		/*
  54		 * Set sd->s_dentry to null only when this dentry is the one
  55		 * that is going to be killed.  Otherwise configfs_d_iput may
  56		 * run just after configfs_lookup and set sd->s_dentry to
  57		 * NULL even it's still in use.
  58		 */
  59		if (sd->s_dentry == dentry)
  60			sd->s_dentry = NULL;
  61
  62		spin_unlock(&configfs_dirent_lock);
  63		configfs_put(sd);
  64	}
  65	iput(inode);
  66}
  67
  68const struct dentry_operations configfs_dentry_ops = {
  69	.d_iput		= configfs_d_iput,
  70	.d_delete	= always_delete_dentry,
  71};
  72
  73#ifdef CONFIG_LOCKDEP
  74
  75/*
  76 * Helpers to make lockdep happy with our recursive locking of default groups'
  77 * inodes (see configfs_attach_group() and configfs_detach_group()).
  78 * We put default groups i_mutexes in separate classes according to their depth
  79 * from the youngest non-default group ancestor.
  80 *
  81 * For a non-default group A having default groups A/B, A/C, and A/C/D, default
  82 * groups A/B and A/C will have their inode's mutex in class
  83 * default_group_class[0], and default group A/C/D will be in
  84 * default_group_class[1].
  85 *
  86 * The lock classes are declared and assigned in inode.c, according to the
  87 * s_depth value.
  88 * The s_depth value is initialized to -1, adjusted to >= 0 when attaching
  89 * default groups, and reset to -1 when all default groups are attached. During
  90 * attachment, if configfs_create() sees s_depth > 0, the lock class of the new
  91 * inode's mutex is set to default_group_class[s_depth - 1].
  92 */
  93
  94static void configfs_init_dirent_depth(struct configfs_dirent *sd)
  95{
  96	sd->s_depth = -1;
  97}
  98
  99static void configfs_set_dir_dirent_depth(struct configfs_dirent *parent_sd,
 100					  struct configfs_dirent *sd)
 101{
 102	int parent_depth = parent_sd->s_depth;
 103
 104	if (parent_depth >= 0)
 105		sd->s_depth = parent_depth + 1;
 106}
 107
 108static void
 109configfs_adjust_dir_dirent_depth_before_populate(struct configfs_dirent *sd)
 110{
 111	/*
 112	 * item's i_mutex class is already setup, so s_depth is now only
 113	 * used to set new sub-directories s_depth, which is always done
 114	 * with item's i_mutex locked.
 115	 */
 116	/*
 117	 *  sd->s_depth == -1 iff we are a non default group.
 118	 *  else (we are a default group) sd->s_depth > 0 (see
 119	 *  create_dir()).
 120	 */
 121	if (sd->s_depth == -1)
 122		/*
 123		 * We are a non default group and we are going to create
 124		 * default groups.
 125		 */
 126		sd->s_depth = 0;
 127}
 128
 129static void
 130configfs_adjust_dir_dirent_depth_after_populate(struct configfs_dirent *sd)
 131{
 132	/* We will not create default groups anymore. */
 133	sd->s_depth = -1;
 134}
 135
 136#else /* CONFIG_LOCKDEP */
 137
 138static void configfs_init_dirent_depth(struct configfs_dirent *sd)
 139{
 140}
 141
 142static void configfs_set_dir_dirent_depth(struct configfs_dirent *parent_sd,
 143					  struct configfs_dirent *sd)
 144{
 145}
 146
 147static void
 148configfs_adjust_dir_dirent_depth_before_populate(struct configfs_dirent *sd)
 149{
 150}
 151
 152static void
 153configfs_adjust_dir_dirent_depth_after_populate(struct configfs_dirent *sd)
 154{
 155}
 156
 157#endif /* CONFIG_LOCKDEP */
 158
 159static struct configfs_fragment *new_fragment(void)
 160{
 161	struct configfs_fragment *p;
 162
 163	p = kmalloc(sizeof(struct configfs_fragment), GFP_KERNEL);
 164	if (p) {
 165		atomic_set(&p->frag_count, 1);
 166		init_rwsem(&p->frag_sem);
 167		p->frag_dead = false;
 168	}
 169	return p;
 170}
 171
 172void put_fragment(struct configfs_fragment *frag)
 173{
 174	if (frag && atomic_dec_and_test(&frag->frag_count))
 175		kfree(frag);
 176}
 177
 178struct configfs_fragment *get_fragment(struct configfs_fragment *frag)
 179{
 180	if (likely(frag))
 181		atomic_inc(&frag->frag_count);
 182	return frag;
 183}
 184
 185/*
 186 * Allocates a new configfs_dirent and links it to the parent configfs_dirent
 187 */
 188static struct configfs_dirent *configfs_new_dirent(struct configfs_dirent *parent_sd,
 189						   void *element, int type,
 190						   struct configfs_fragment *frag)
 191{
 192	struct configfs_dirent * sd;
 193
 194	sd = kmem_cache_zalloc(configfs_dir_cachep, GFP_KERNEL);
 195	if (!sd)
 196		return ERR_PTR(-ENOMEM);
 197
 198	atomic_set(&sd->s_count, 1);
 199	INIT_LIST_HEAD(&sd->s_children);
 200	sd->s_element = element;
 201	sd->s_type = type;
 202	configfs_init_dirent_depth(sd);
 203	spin_lock(&configfs_dirent_lock);
 204	if (parent_sd->s_type & CONFIGFS_USET_DROPPING) {
 205		spin_unlock(&configfs_dirent_lock);
 206		kmem_cache_free(configfs_dir_cachep, sd);
 207		return ERR_PTR(-ENOENT);
 208	}
 209	sd->s_frag = get_fragment(frag);
 210	list_add(&sd->s_sibling, &parent_sd->s_children);
 
 
 
 
 
 
 
 
 
 
 211	spin_unlock(&configfs_dirent_lock);
 212
 213	return sd;
 214}
 215
 216/*
 217 *
 218 * Return -EEXIST if there is already a configfs element with the same
 219 * name for the same parent.
 220 *
 221 * called with parent inode's i_mutex held
 222 */
 223static int configfs_dirent_exists(struct configfs_dirent *parent_sd,
 224				  const unsigned char *new)
 225{
 226	struct configfs_dirent * sd;
 
 
 227
 228	list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
 229		if (sd->s_element) {
 230			const unsigned char *existing = configfs_get_name(sd);
 231			if (strcmp(existing, new))
 232				continue;
 233			else
 234				return -EEXIST;
 235		}
 236	}
 237
 238	return 0;
 239}
 240
 241
 242int configfs_make_dirent(struct configfs_dirent * parent_sd,
 243			 struct dentry * dentry, void * element,
 244			 umode_t mode, int type, struct configfs_fragment *frag)
 245{
 246	struct configfs_dirent * sd;
 247
 248	sd = configfs_new_dirent(parent_sd, element, type, frag);
 249	if (IS_ERR(sd))
 250		return PTR_ERR(sd);
 251
 252	sd->s_mode = mode;
 253	sd->s_dentry = dentry;
 254	if (dentry)
 255		dentry->d_fsdata = configfs_get(sd);
 256
 257	return 0;
 258}
 259
 260static void configfs_remove_dirent(struct dentry *dentry)
 261{
 262	struct configfs_dirent *sd = dentry->d_fsdata;
 263
 264	if (!sd)
 265		return;
 266	spin_lock(&configfs_dirent_lock);
 267	list_del_init(&sd->s_sibling);
 268	spin_unlock(&configfs_dirent_lock);
 269	configfs_put(sd);
 270}
 271
 272/**
 273 *	configfs_create_dir - create a directory for an config_item.
 274 *	@item:		config_itemwe're creating directory for.
 275 *	@dentry:	config_item's dentry.
 276 *	@frag:		config_item's fragment.
 277 *
 278 *	Note: user-created entries won't be allowed under this new directory
 279 *	until it is validated by configfs_dir_set_ready()
 280 */
 281
 282static int configfs_create_dir(struct config_item *item, struct dentry *dentry,
 283				struct configfs_fragment *frag)
 284{
 285	int error;
 286	umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
 287	struct dentry *p = dentry->d_parent;
 288	struct inode *inode;
 289
 290	BUG_ON(!item);
 291
 292	error = configfs_dirent_exists(p->d_fsdata, dentry->d_name.name);
 293	if (unlikely(error))
 294		return error;
 295
 296	error = configfs_make_dirent(p->d_fsdata, dentry, item, mode,
 297				     CONFIGFS_DIR | CONFIGFS_USET_CREATING,
 298				     frag);
 299	if (unlikely(error))
 300		return error;
 301
 302	configfs_set_dir_dirent_depth(p->d_fsdata, dentry->d_fsdata);
 303	inode = configfs_create(dentry, mode);
 304	if (IS_ERR(inode))
 305		goto out_remove;
 306
 307	inode->i_op = &configfs_dir_inode_operations;
 308	inode->i_fop = &configfs_dir_operations;
 309	/* directory inodes start off with i_nlink == 2 (for "." entry) */
 310	inc_nlink(inode);
 311	d_instantiate(dentry, inode);
 312	/* already hashed */
 313	dget(dentry);  /* pin directory dentries in core */
 314	inc_nlink(d_inode(p));
 315	item->ci_dentry = dentry;
 316	return 0;
 317
 318out_remove:
 319	configfs_put(dentry->d_fsdata);
 320	configfs_remove_dirent(dentry);
 321	return PTR_ERR(inode);
 322}
 323
 324/*
 325 * Allow userspace to create new entries under a new directory created with
 326 * configfs_create_dir(), and under all of its chidlren directories recursively.
 327 * @sd		configfs_dirent of the new directory to validate
 328 *
 329 * Caller must hold configfs_dirent_lock.
 330 */
 331static void configfs_dir_set_ready(struct configfs_dirent *sd)
 332{
 333	struct configfs_dirent *child_sd;
 334
 335	sd->s_type &= ~CONFIGFS_USET_CREATING;
 336	list_for_each_entry(child_sd, &sd->s_children, s_sibling)
 337		if (child_sd->s_type & CONFIGFS_USET_CREATING)
 338			configfs_dir_set_ready(child_sd);
 339}
 340
 341/*
 342 * Check that a directory does not belong to a directory hierarchy being
 343 * attached and not validated yet.
 344 * @sd		configfs_dirent of the directory to check
 345 *
 346 * @return	non-zero iff the directory was validated
 347 *
 348 * Note: takes configfs_dirent_lock, so the result may change from false to true
 349 * in two consecutive calls, but never from true to false.
 350 */
 351int configfs_dirent_is_ready(struct configfs_dirent *sd)
 352{
 353	int ret;
 354
 355	spin_lock(&configfs_dirent_lock);
 356	ret = !(sd->s_type & CONFIGFS_USET_CREATING);
 357	spin_unlock(&configfs_dirent_lock);
 358
 359	return ret;
 360}
 361
 362int configfs_create_link(struct configfs_dirent *target, struct dentry *parent,
 363		struct dentry *dentry, char *body)
 364{
 365	int err = 0;
 366	umode_t mode = S_IFLNK | S_IRWXUGO;
 367	struct configfs_dirent *p = parent->d_fsdata;
 368	struct inode *inode;
 369
 370	err = configfs_make_dirent(p, dentry, target, mode, CONFIGFS_ITEM_LINK,
 371			p->s_frag);
 372	if (err)
 373		return err;
 374
 375	inode = configfs_create(dentry, mode);
 376	if (IS_ERR(inode))
 377		goto out_remove;
 378
 379	inode->i_link = body;
 380	inode->i_op = &configfs_symlink_inode_operations;
 381	d_instantiate(dentry, inode);
 382	dget(dentry);  /* pin link dentries in core */
 383	return 0;
 384
 385out_remove:
 386	configfs_put(dentry->d_fsdata);
 387	configfs_remove_dirent(dentry);
 388	return PTR_ERR(inode);
 389}
 390
 391static void remove_dir(struct dentry * d)
 392{
 393	struct dentry * parent = dget(d->d_parent);
 394
 395	configfs_remove_dirent(d);
 396
 397	if (d_really_is_positive(d))
 398		simple_rmdir(d_inode(parent),d);
 399
 400	pr_debug(" o %pd removing done (%d)\n", d, d_count(d));
 401
 402	dput(parent);
 403}
 404
 405/**
 406 * configfs_remove_dir - remove an config_item's directory.
 407 * @item:	config_item we're removing.
 408 *
 409 * The only thing special about this is that we remove any files in
 410 * the directory before we remove the directory, and we've inlined
 411 * what used to be configfs_rmdir() below, instead of calling separately.
 412 *
 413 * Caller holds the mutex of the item's inode
 414 */
 415
 416static void configfs_remove_dir(struct config_item * item)
 417{
 418	struct dentry * dentry = dget(item->ci_dentry);
 419
 420	if (!dentry)
 421		return;
 422
 423	remove_dir(dentry);
 424	/**
 425	 * Drop reference from dget() on entrance.
 426	 */
 427	dput(dentry);
 428}
 429
 430static struct dentry * configfs_lookup(struct inode *dir,
 431				       struct dentry *dentry,
 432				       unsigned int flags)
 433{
 434	struct configfs_dirent * parent_sd = dentry->d_parent->d_fsdata;
 435	struct configfs_dirent * sd;
 436	struct inode *inode = NULL;
 437
 438	if (dentry->d_name.len > NAME_MAX)
 439		return ERR_PTR(-ENAMETOOLONG);
 440
 441	/*
 442	 * Fake invisibility if dir belongs to a group/default groups hierarchy
 443	 * being attached
 444	 *
 445	 * This forbids userspace to read/write attributes of items which may
 446	 * not complete their initialization, since the dentries of the
 447	 * attributes won't be instantiated.
 448	 */
 449	if (!configfs_dirent_is_ready(parent_sd))
 450		return ERR_PTR(-ENOENT);
 451
 452	spin_lock(&configfs_dirent_lock);
 453	list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
 
 
 
 
 
 
 
 
 
 
 
 
 454		if ((sd->s_type & CONFIGFS_NOT_PINNED) &&
 455		    !strcmp(configfs_get_name(sd), dentry->d_name.name)) {
 456			struct configfs_attribute *attr = sd->s_element;
 457			umode_t mode = (attr->ca_mode & S_IALLUGO) | S_IFREG;
 458
 459			dentry->d_fsdata = configfs_get(sd);
 460			sd->s_dentry = dentry;
 461			spin_unlock(&configfs_dirent_lock);
 462
 463			inode = configfs_create(dentry, mode);
 464			if (IS_ERR(inode)) {
 465				configfs_put(sd);
 466				return ERR_CAST(inode);
 467			}
 468			if (sd->s_type & CONFIGFS_ITEM_BIN_ATTR) {
 469				inode->i_size = 0;
 470				inode->i_fop = &configfs_bin_file_operations;
 471			} else {
 472				inode->i_size = PAGE_SIZE;
 473				inode->i_fop = &configfs_file_operations;
 474			}
 475			goto done;
 476		}
 477	}
 478	spin_unlock(&configfs_dirent_lock);
 479done:
 480	d_add(dentry, inode);
 481	return NULL;
 482}
 483
 484/*
 485 * Only subdirectories count here.  Files (CONFIGFS_NOT_PINNED) are
 486 * attributes and are removed by rmdir().  We recurse, setting
 487 * CONFIGFS_USET_DROPPING on all children that are candidates for
 488 * default detach.
 489 * If there is an error, the caller will reset the flags via
 490 * configfs_detach_rollback().
 491 */
 492static int configfs_detach_prep(struct dentry *dentry, struct dentry **wait)
 493{
 494	struct configfs_dirent *parent_sd = dentry->d_fsdata;
 495	struct configfs_dirent *sd;
 496	int ret;
 497
 498	/* Mark that we're trying to drop the group */
 499	parent_sd->s_type |= CONFIGFS_USET_DROPPING;
 500
 501	ret = -EBUSY;
 502	if (parent_sd->s_links)
 503		goto out;
 504
 505	ret = 0;
 506	list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
 507		if (!sd->s_element ||
 508		    (sd->s_type & CONFIGFS_NOT_PINNED))
 509			continue;
 510		if (sd->s_type & CONFIGFS_USET_DEFAULT) {
 511			/* Abort if racing with mkdir() */
 512			if (sd->s_type & CONFIGFS_USET_IN_MKDIR) {
 513				if (wait)
 514					*wait= dget(sd->s_dentry);
 515				return -EAGAIN;
 516			}
 517
 518			/*
 519			 * Yup, recursive.  If there's a problem, blame
 520			 * deep nesting of default_groups
 521			 */
 522			ret = configfs_detach_prep(sd->s_dentry, wait);
 523			if (!ret)
 524				continue;
 525		} else
 526			ret = -ENOTEMPTY;
 527
 528		break;
 529	}
 530
 531out:
 532	return ret;
 533}
 534
 535/*
 536 * Walk the tree, resetting CONFIGFS_USET_DROPPING wherever it was
 537 * set.
 538 */
 539static void configfs_detach_rollback(struct dentry *dentry)
 540{
 541	struct configfs_dirent *parent_sd = dentry->d_fsdata;
 542	struct configfs_dirent *sd;
 543
 544	parent_sd->s_type &= ~CONFIGFS_USET_DROPPING;
 545
 546	list_for_each_entry(sd, &parent_sd->s_children, s_sibling)
 547		if (sd->s_type & CONFIGFS_USET_DEFAULT)
 548			configfs_detach_rollback(sd->s_dentry);
 549}
 550
 551static void detach_attrs(struct config_item * item)
 552{
 553	struct dentry * dentry = dget(item->ci_dentry);
 554	struct configfs_dirent * parent_sd;
 555	struct configfs_dirent * sd, * tmp;
 556
 557	if (!dentry)
 558		return;
 559
 560	pr_debug("configfs %s: dropping attrs for  dir\n",
 561		 dentry->d_name.name);
 562
 563	parent_sd = dentry->d_fsdata;
 564	list_for_each_entry_safe(sd, tmp, &parent_sd->s_children, s_sibling) {
 565		if (!sd->s_element || !(sd->s_type & CONFIGFS_NOT_PINNED))
 566			continue;
 567		spin_lock(&configfs_dirent_lock);
 568		list_del_init(&sd->s_sibling);
 569		spin_unlock(&configfs_dirent_lock);
 570		configfs_drop_dentry(sd, dentry);
 571		configfs_put(sd);
 572	}
 573
 574	/**
 575	 * Drop reference from dget() on entrance.
 576	 */
 577	dput(dentry);
 578}
 579
 580static int populate_attrs(struct config_item *item)
 581{
 582	const struct config_item_type *t = item->ci_type;
 
 583	struct configfs_attribute *attr;
 584	struct configfs_bin_attribute *bin_attr;
 585	int error = 0;
 586	int i;
 587
 588	if (!t)
 589		return -EINVAL;
 
 
 
 590	if (t->ct_attrs) {
 591		for (i = 0; (attr = t->ct_attrs[i]) != NULL; i++) {
 
 
 
 592			if ((error = configfs_create_file(item, attr)))
 593				break;
 594		}
 595	}
 596	if (t->ct_bin_attrs) {
 597		for (i = 0; (bin_attr = t->ct_bin_attrs[i]) != NULL; i++) {
 
 
 
 598			error = configfs_create_bin_file(item, bin_attr);
 599			if (error)
 600				break;
 601		}
 602	}
 603
 604	if (error)
 605		detach_attrs(item);
 606
 607	return error;
 608}
 609
 610static int configfs_attach_group(struct config_item *parent_item,
 611				 struct config_item *item,
 612				 struct dentry *dentry,
 613				 struct configfs_fragment *frag);
 614static void configfs_detach_group(struct config_item *item);
 615
 616static void detach_groups(struct config_group *group)
 617{
 618	struct dentry * dentry = dget(group->cg_item.ci_dentry);
 619	struct dentry *child;
 620	struct configfs_dirent *parent_sd;
 621	struct configfs_dirent *sd, *tmp;
 622
 623	if (!dentry)
 624		return;
 625
 626	parent_sd = dentry->d_fsdata;
 627	list_for_each_entry_safe(sd, tmp, &parent_sd->s_children, s_sibling) {
 628		if (!sd->s_element ||
 629		    !(sd->s_type & CONFIGFS_USET_DEFAULT))
 630			continue;
 631
 632		child = sd->s_dentry;
 633
 634		inode_lock(d_inode(child));
 635
 636		configfs_detach_group(sd->s_element);
 637		d_inode(child)->i_flags |= S_DEAD;
 638		dont_mount(child);
 639
 640		inode_unlock(d_inode(child));
 641
 642		d_delete(child);
 643		dput(child);
 644	}
 645
 646	/**
 647	 * Drop reference from dget() on entrance.
 648	 */
 649	dput(dentry);
 650}
 651
 652/*
 653 * This fakes mkdir(2) on a default_groups[] entry.  It
 654 * creates a dentry, attachs it, and then does fixup
 655 * on the sd->s_type.
 656 *
 657 * We could, perhaps, tweak our parent's ->mkdir for a minute and
 658 * try using vfs_mkdir.  Just a thought.
 659 */
 660static int create_default_group(struct config_group *parent_group,
 661				struct config_group *group,
 662				struct configfs_fragment *frag)
 663{
 664	int ret;
 665	struct configfs_dirent *sd;
 666	/* We trust the caller holds a reference to parent */
 667	struct dentry *child, *parent = parent_group->cg_item.ci_dentry;
 668
 669	if (!group->cg_item.ci_name)
 670		group->cg_item.ci_name = group->cg_item.ci_namebuf;
 671
 672	ret = -ENOMEM;
 673	child = d_alloc_name(parent, group->cg_item.ci_name);
 674	if (child) {
 675		d_add(child, NULL);
 676
 677		ret = configfs_attach_group(&parent_group->cg_item,
 678					    &group->cg_item, child, frag);
 679		if (!ret) {
 680			sd = child->d_fsdata;
 681			sd->s_type |= CONFIGFS_USET_DEFAULT;
 682		} else {
 683			BUG_ON(d_inode(child));
 684			d_drop(child);
 685			dput(child);
 686		}
 687	}
 688
 689	return ret;
 690}
 691
 692static int populate_groups(struct config_group *group,
 693			   struct configfs_fragment *frag)
 694{
 695	struct config_group *new_group;
 696	int ret = 0;
 697
 698	list_for_each_entry(new_group, &group->default_groups, group_entry) {
 699		ret = create_default_group(group, new_group, frag);
 700		if (ret) {
 701			detach_groups(group);
 702			break;
 703		}
 704	}
 705
 706	return ret;
 707}
 708
 709void configfs_remove_default_groups(struct config_group *group)
 710{
 711	struct config_group *g, *n;
 712
 713	list_for_each_entry_safe(g, n, &group->default_groups, group_entry) {
 714		list_del(&g->group_entry);
 715		config_item_put(&g->cg_item);
 716	}
 717}
 718EXPORT_SYMBOL(configfs_remove_default_groups);
 719
 720/*
 721 * All of link_obj/unlink_obj/link_group/unlink_group require that
 722 * subsys->su_mutex is held.
 723 */
 724
 725static void unlink_obj(struct config_item *item)
 726{
 727	struct config_group *group;
 728
 729	group = item->ci_group;
 730	if (group) {
 731		list_del_init(&item->ci_entry);
 732
 733		item->ci_group = NULL;
 734		item->ci_parent = NULL;
 735
 736		/* Drop the reference for ci_entry */
 737		config_item_put(item);
 738
 739		/* Drop the reference for ci_parent */
 740		config_group_put(group);
 741	}
 742}
 743
 744static void link_obj(struct config_item *parent_item, struct config_item *item)
 745{
 746	/*
 747	 * Parent seems redundant with group, but it makes certain
 748	 * traversals much nicer.
 749	 */
 750	item->ci_parent = parent_item;
 751
 752	/*
 753	 * We hold a reference on the parent for the child's ci_parent
 754	 * link.
 755	 */
 756	item->ci_group = config_group_get(to_config_group(parent_item));
 757	list_add_tail(&item->ci_entry, &item->ci_group->cg_children);
 758
 759	/*
 760	 * We hold a reference on the child for ci_entry on the parent's
 761	 * cg_children
 762	 */
 763	config_item_get(item);
 764}
 765
 766static void unlink_group(struct config_group *group)
 767{
 768	struct config_group *new_group;
 769
 770	list_for_each_entry(new_group, &group->default_groups, group_entry)
 771		unlink_group(new_group);
 772
 773	group->cg_subsys = NULL;
 774	unlink_obj(&group->cg_item);
 775}
 776
 777static void link_group(struct config_group *parent_group, struct config_group *group)
 778{
 779	struct config_group *new_group;
 780	struct configfs_subsystem *subsys = NULL; /* gcc is a turd */
 781
 782	link_obj(&parent_group->cg_item, &group->cg_item);
 783
 784	if (parent_group->cg_subsys)
 785		subsys = parent_group->cg_subsys;
 786	else if (configfs_is_root(&parent_group->cg_item))
 787		subsys = to_configfs_subsystem(group);
 788	else
 789		BUG();
 790	group->cg_subsys = subsys;
 791
 792	list_for_each_entry(new_group, &group->default_groups, group_entry)
 793		link_group(group, new_group);
 794}
 795
 796/*
 797 * The goal is that configfs_attach_item() (and
 798 * configfs_attach_group()) can be called from either the VFS or this
 799 * module.  That is, they assume that the items have been created,
 800 * the dentry allocated, and the dcache is all ready to go.
 801 *
 802 * If they fail, they must clean up after themselves as if they
 803 * had never been called.  The caller (VFS or local function) will
 804 * handle cleaning up the dcache bits.
 805 *
 806 * configfs_detach_group() and configfs_detach_item() behave similarly on
 807 * the way out.  They assume that the proper semaphores are held, they
 808 * clean up the configfs items, and they expect their callers will
 809 * handle the dcache bits.
 810 */
 811static int configfs_attach_item(struct config_item *parent_item,
 812				struct config_item *item,
 813				struct dentry *dentry,
 814				struct configfs_fragment *frag)
 815{
 816	int ret;
 817
 818	ret = configfs_create_dir(item, dentry, frag);
 819	if (!ret) {
 820		ret = populate_attrs(item);
 821		if (ret) {
 822			/*
 823			 * We are going to remove an inode and its dentry but
 824			 * the VFS may already have hit and used them. Thus,
 825			 * we must lock them as rmdir() would.
 826			 */
 827			inode_lock(d_inode(dentry));
 828			configfs_remove_dir(item);
 829			d_inode(dentry)->i_flags |= S_DEAD;
 830			dont_mount(dentry);
 831			inode_unlock(d_inode(dentry));
 832			d_delete(dentry);
 833		}
 834	}
 835
 836	return ret;
 837}
 838
 839/* Caller holds the mutex of the item's inode */
 840static void configfs_detach_item(struct config_item *item)
 841{
 842	detach_attrs(item);
 843	configfs_remove_dir(item);
 844}
 845
 846static int configfs_attach_group(struct config_item *parent_item,
 847				 struct config_item *item,
 848				 struct dentry *dentry,
 849				 struct configfs_fragment *frag)
 850{
 851	int ret;
 852	struct configfs_dirent *sd;
 853
 854	ret = configfs_attach_item(parent_item, item, dentry, frag);
 855	if (!ret) {
 856		sd = dentry->d_fsdata;
 857		sd->s_type |= CONFIGFS_USET_DIR;
 858
 859		/*
 860		 * FYI, we're faking mkdir in populate_groups()
 861		 * We must lock the group's inode to avoid races with the VFS
 862		 * which can already hit the inode and try to add/remove entries
 863		 * under it.
 864		 *
 865		 * We must also lock the inode to remove it safely in case of
 866		 * error, as rmdir() would.
 867		 */
 868		inode_lock_nested(d_inode(dentry), I_MUTEX_CHILD);
 869		configfs_adjust_dir_dirent_depth_before_populate(sd);
 870		ret = populate_groups(to_config_group(item), frag);
 871		if (ret) {
 872			configfs_detach_item(item);
 873			d_inode(dentry)->i_flags |= S_DEAD;
 874			dont_mount(dentry);
 875		}
 876		configfs_adjust_dir_dirent_depth_after_populate(sd);
 877		inode_unlock(d_inode(dentry));
 878		if (ret)
 879			d_delete(dentry);
 880	}
 881
 882	return ret;
 883}
 884
 885/* Caller holds the mutex of the group's inode */
 886static void configfs_detach_group(struct config_item *item)
 887{
 888	detach_groups(to_config_group(item));
 889	configfs_detach_item(item);
 890}
 891
 892/*
 893 * After the item has been detached from the filesystem view, we are
 894 * ready to tear it out of the hierarchy.  Notify the client before
 895 * we do that so they can perform any cleanup that requires
 896 * navigating the hierarchy.  A client does not need to provide this
 897 * callback.  The subsystem semaphore MUST be held by the caller, and
 898 * references must be valid for both items.  It also assumes the
 899 * caller has validated ci_type.
 900 */
 901static void client_disconnect_notify(struct config_item *parent_item,
 902				     struct config_item *item)
 903{
 904	const struct config_item_type *type;
 905
 906	type = parent_item->ci_type;
 907	BUG_ON(!type);
 908
 909	if (type->ct_group_ops && type->ct_group_ops->disconnect_notify)
 910		type->ct_group_ops->disconnect_notify(to_config_group(parent_item),
 911						      item);
 912}
 913
 914/*
 915 * Drop the initial reference from make_item()/make_group()
 916 * This function assumes that reference is held on item
 917 * and that item holds a valid reference to the parent.  Also, it
 918 * assumes the caller has validated ci_type.
 919 */
 920static void client_drop_item(struct config_item *parent_item,
 921			     struct config_item *item)
 922{
 923	const struct config_item_type *type;
 924
 925	type = parent_item->ci_type;
 926	BUG_ON(!type);
 927
 928	/*
 929	 * If ->drop_item() exists, it is responsible for the
 930	 * config_item_put().
 931	 */
 932	if (type->ct_group_ops && type->ct_group_ops->drop_item)
 933		type->ct_group_ops->drop_item(to_config_group(parent_item),
 934					      item);
 935	else
 936		config_item_put(item);
 937}
 938
 939#ifdef DEBUG
 940static void configfs_dump_one(struct configfs_dirent *sd, int level)
 941{
 942	pr_info("%*s\"%s\":\n", level, " ", configfs_get_name(sd));
 943
 944#define type_print(_type) if (sd->s_type & _type) pr_info("%*s %s\n", level, " ", #_type);
 945	type_print(CONFIGFS_ROOT);
 946	type_print(CONFIGFS_DIR);
 947	type_print(CONFIGFS_ITEM_ATTR);
 948	type_print(CONFIGFS_ITEM_LINK);
 949	type_print(CONFIGFS_USET_DIR);
 950	type_print(CONFIGFS_USET_DEFAULT);
 951	type_print(CONFIGFS_USET_DROPPING);
 952#undef type_print
 953}
 954
 955static int configfs_dump(struct configfs_dirent *sd, int level)
 956{
 957	struct configfs_dirent *child_sd;
 958	int ret = 0;
 959
 960	configfs_dump_one(sd, level);
 961
 962	if (!(sd->s_type & (CONFIGFS_DIR|CONFIGFS_ROOT)))
 963		return 0;
 964
 965	list_for_each_entry(child_sd, &sd->s_children, s_sibling) {
 966		ret = configfs_dump(child_sd, level + 2);
 967		if (ret)
 968			break;
 969	}
 970
 971	return ret;
 972}
 973#endif
 974
 975
 976/*
 977 * configfs_depend_item() and configfs_undepend_item()
 978 *
 979 * WARNING: Do not call these from a configfs callback!
 980 *
 981 * This describes these functions and their helpers.
 982 *
 983 * Allow another kernel system to depend on a config_item.  If this
 984 * happens, the item cannot go away until the dependent can live without
 985 * it.  The idea is to give client modules as simple an interface as
 986 * possible.  When a system asks them to depend on an item, they just
 987 * call configfs_depend_item().  If the item is live and the client
 988 * driver is in good shape, we'll happily do the work for them.
 989 *
 990 * Why is the locking complex?  Because configfs uses the VFS to handle
 991 * all locking, but this function is called outside the normal
 992 * VFS->configfs path.  So it must take VFS locks to prevent the
 993 * VFS->configfs stuff (configfs_mkdir(), configfs_rmdir(), etc).  This is
 994 * why you can't call these functions underneath configfs callbacks.
 995 *
 996 * Note, btw, that this can be called at *any* time, even when a configfs
 997 * subsystem isn't registered, or when configfs is loading or unloading.
 998 * Just like configfs_register_subsystem().  So we take the same
 999 * precautions.  We pin the filesystem.  We lock configfs_dirent_lock.
1000 * If we can find the target item in the
1001 * configfs tree, it must be part of the subsystem tree as well, so we
1002 * do not need the subsystem semaphore.  Holding configfs_dirent_lock helps
1003 * locking out mkdir() and rmdir(), who might be racing us.
1004 */
1005
1006/*
1007 * configfs_depend_prep()
1008 *
1009 * Only subdirectories count here.  Files (CONFIGFS_NOT_PINNED) are
1010 * attributes.  This is similar but not the same to configfs_detach_prep().
1011 * Note that configfs_detach_prep() expects the parent to be locked when it
1012 * is called, but we lock the parent *inside* configfs_depend_prep().  We
1013 * do that so we can unlock it if we find nothing.
1014 *
1015 * Here we do a depth-first search of the dentry hierarchy looking for
1016 * our object.
1017 * We deliberately ignore items tagged as dropping since they are virtually
1018 * dead, as well as items in the middle of attachment since they virtually
1019 * do not exist yet. This completes the locking out of racing mkdir() and
1020 * rmdir().
1021 * Note: subdirectories in the middle of attachment start with s_type =
1022 * CONFIGFS_DIR|CONFIGFS_USET_CREATING set by create_dir().  When
1023 * CONFIGFS_USET_CREATING is set, we ignore the item.  The actual set of
1024 * s_type is in configfs_new_dirent(), which has configfs_dirent_lock.
1025 *
1026 * If the target is not found, -ENOENT is bubbled up.
1027 *
1028 * This adds a requirement that all config_items be unique!
1029 *
1030 * This is recursive.  There isn't
1031 * much on the stack, though, so folks that need this function - be careful
1032 * about your stack!  Patches will be accepted to make it iterative.
1033 */
1034static int configfs_depend_prep(struct dentry *origin,
1035				struct config_item *target)
1036{
1037	struct configfs_dirent *child_sd, *sd;
1038	int ret = 0;
1039
1040	BUG_ON(!origin || !origin->d_fsdata);
1041	sd = origin->d_fsdata;
1042
1043	if (sd->s_element == target)  /* Boo-yah */
1044		goto out;
1045
1046	list_for_each_entry(child_sd, &sd->s_children, s_sibling) {
1047		if ((child_sd->s_type & CONFIGFS_DIR) &&
1048		    !(child_sd->s_type & CONFIGFS_USET_DROPPING) &&
1049		    !(child_sd->s_type & CONFIGFS_USET_CREATING)) {
1050			ret = configfs_depend_prep(child_sd->s_dentry,
1051						   target);
1052			if (!ret)
1053				goto out;  /* Child path boo-yah */
1054		}
1055	}
1056
1057	/* We looped all our children and didn't find target */
1058	ret = -ENOENT;
1059
1060out:
1061	return ret;
1062}
1063
1064static int configfs_do_depend_item(struct dentry *subsys_dentry,
1065				   struct config_item *target)
1066{
1067	struct configfs_dirent *p;
1068	int ret;
1069
1070	spin_lock(&configfs_dirent_lock);
1071	/* Scan the tree, return 0 if found */
1072	ret = configfs_depend_prep(subsys_dentry, target);
1073	if (ret)
1074		goto out_unlock_dirent_lock;
1075
1076	/*
1077	 * We are sure that the item is not about to be removed by rmdir(), and
1078	 * not in the middle of attachment by mkdir().
1079	 */
1080	p = target->ci_dentry->d_fsdata;
1081	p->s_dependent_count += 1;
1082
1083out_unlock_dirent_lock:
1084	spin_unlock(&configfs_dirent_lock);
1085
1086	return ret;
1087}
1088
1089static inline struct configfs_dirent *
1090configfs_find_subsys_dentry(struct configfs_dirent *root_sd,
1091			    struct config_item *subsys_item)
1092{
1093	struct configfs_dirent *p;
1094	struct configfs_dirent *ret = NULL;
1095
1096	list_for_each_entry(p, &root_sd->s_children, s_sibling) {
1097		if (p->s_type & CONFIGFS_DIR &&
1098		    p->s_element == subsys_item) {
1099			ret = p;
1100			break;
1101		}
1102	}
1103
1104	return ret;
1105}
1106
1107
1108int configfs_depend_item(struct configfs_subsystem *subsys,
1109			 struct config_item *target)
1110{
1111	int ret;
1112	struct configfs_dirent *subsys_sd;
1113	struct config_item *s_item = &subsys->su_group.cg_item;
1114	struct dentry *root;
1115
1116	/*
1117	 * Pin the configfs filesystem.  This means we can safely access
1118	 * the root of the configfs filesystem.
1119	 */
1120	root = configfs_pin_fs();
1121	if (IS_ERR(root))
1122		return PTR_ERR(root);
1123
1124	/*
1125	 * Next, lock the root directory.  We're going to check that the
1126	 * subsystem is really registered, and so we need to lock out
1127	 * configfs_[un]register_subsystem().
1128	 */
1129	inode_lock(d_inode(root));
1130
1131	subsys_sd = configfs_find_subsys_dentry(root->d_fsdata, s_item);
1132	if (!subsys_sd) {
1133		ret = -ENOENT;
1134		goto out_unlock_fs;
1135	}
1136
1137	/* Ok, now we can trust subsys/s_item */
1138	ret = configfs_do_depend_item(subsys_sd->s_dentry, target);
1139
1140out_unlock_fs:
1141	inode_unlock(d_inode(root));
1142
1143	/*
1144	 * If we succeeded, the fs is pinned via other methods.  If not,
1145	 * we're done with it anyway.  So release_fs() is always right.
1146	 */
1147	configfs_release_fs();
1148
1149	return ret;
1150}
1151EXPORT_SYMBOL(configfs_depend_item);
1152
1153/*
1154 * Release the dependent linkage.  This is much simpler than
1155 * configfs_depend_item() because we know that the client driver is
1156 * pinned, thus the subsystem is pinned, and therefore configfs is pinned.
1157 */
1158void configfs_undepend_item(struct config_item *target)
1159{
1160	struct configfs_dirent *sd;
1161
1162	/*
1163	 * Since we can trust everything is pinned, we just need
1164	 * configfs_dirent_lock.
1165	 */
1166	spin_lock(&configfs_dirent_lock);
1167
1168	sd = target->ci_dentry->d_fsdata;
1169	BUG_ON(sd->s_dependent_count < 1);
1170
1171	sd->s_dependent_count -= 1;
1172
1173	/*
1174	 * After this unlock, we cannot trust the item to stay alive!
1175	 * DO NOT REFERENCE item after this unlock.
1176	 */
1177	spin_unlock(&configfs_dirent_lock);
1178}
1179EXPORT_SYMBOL(configfs_undepend_item);
1180
1181/*
1182 * caller_subsys is a caller's subsystem not target's. This is used to
1183 * determine if we should lock root and check subsys or not. When we are
1184 * in the same subsystem as our target there is no need to do locking as
1185 * we know that subsys is valid and is not unregistered during this function
1186 * as we are called from callback of one of his children and VFS holds a lock
1187 * on some inode. Otherwise we have to lock our root to  ensure that target's
1188 * subsystem it is not unregistered during this function.
1189 */
1190int configfs_depend_item_unlocked(struct configfs_subsystem *caller_subsys,
1191				  struct config_item *target)
1192{
1193	struct configfs_subsystem *target_subsys;
1194	struct config_group *root, *parent;
1195	struct configfs_dirent *subsys_sd;
1196	int ret = -ENOENT;
1197
1198	/* Disallow this function for configfs root */
1199	if (configfs_is_root(target))
1200		return -EINVAL;
1201
1202	parent = target->ci_group;
1203	/*
1204	 * This may happen when someone is trying to depend root
1205	 * directory of some subsystem
1206	 */
1207	if (configfs_is_root(&parent->cg_item)) {
1208		target_subsys = to_configfs_subsystem(to_config_group(target));
1209		root = parent;
1210	} else {
1211		target_subsys = parent->cg_subsys;
1212		/* Find a cofnigfs root as we may need it for locking */
1213		for (root = parent; !configfs_is_root(&root->cg_item);
1214		     root = root->cg_item.ci_group)
1215			;
1216	}
1217
1218	if (target_subsys != caller_subsys) {
1219		/*
1220		 * We are in other configfs subsystem, so we have to do
1221		 * additional locking to prevent other subsystem from being
1222		 * unregistered
1223		 */
1224		inode_lock(d_inode(root->cg_item.ci_dentry));
1225
1226		/*
1227		 * As we are trying to depend item from other subsystem
1228		 * we have to check if this subsystem is still registered
1229		 */
1230		subsys_sd = configfs_find_subsys_dentry(
1231				root->cg_item.ci_dentry->d_fsdata,
1232				&target_subsys->su_group.cg_item);
1233		if (!subsys_sd)
1234			goto out_root_unlock;
1235	} else {
1236		subsys_sd = target_subsys->su_group.cg_item.ci_dentry->d_fsdata;
1237	}
1238
1239	/* Now we can execute core of depend item */
1240	ret = configfs_do_depend_item(subsys_sd->s_dentry, target);
1241
1242	if (target_subsys != caller_subsys)
1243out_root_unlock:
1244		/*
1245		 * We were called from subsystem other than our target so we
1246		 * took some locks so now it's time to release them
1247		 */
1248		inode_unlock(d_inode(root->cg_item.ci_dentry));
1249
1250	return ret;
1251}
1252EXPORT_SYMBOL(configfs_depend_item_unlocked);
1253
1254static int configfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1255			  struct dentry *dentry, umode_t mode)
1256{
1257	int ret = 0;
1258	int module_got = 0;
1259	struct config_group *group = NULL;
1260	struct config_item *item = NULL;
1261	struct config_item *parent_item;
1262	struct configfs_subsystem *subsys;
1263	struct configfs_dirent *sd;
1264	const struct config_item_type *type;
1265	struct module *subsys_owner = NULL, *new_item_owner = NULL;
1266	struct configfs_fragment *frag;
1267	char *name;
1268
1269	sd = dentry->d_parent->d_fsdata;
1270
1271	/*
1272	 * Fake invisibility if dir belongs to a group/default groups hierarchy
1273	 * being attached
1274	 */
1275	if (!configfs_dirent_is_ready(sd)) {
1276		ret = -ENOENT;
1277		goto out;
1278	}
1279
1280	if (!(sd->s_type & CONFIGFS_USET_DIR)) {
1281		ret = -EPERM;
1282		goto out;
1283	}
1284
1285	frag = new_fragment();
1286	if (!frag) {
1287		ret = -ENOMEM;
1288		goto out;
1289	}
1290
1291	/* Get a working ref for the duration of this function */
1292	parent_item = configfs_get_config_item(dentry->d_parent);
1293	type = parent_item->ci_type;
1294	subsys = to_config_group(parent_item)->cg_subsys;
1295	BUG_ON(!subsys);
1296
1297	if (!type || !type->ct_group_ops ||
1298	    (!type->ct_group_ops->make_group &&
1299	     !type->ct_group_ops->make_item)) {
1300		ret = -EPERM;  /* Lack-of-mkdir returns -EPERM */
1301		goto out_put;
1302	}
1303
1304	/*
1305	 * The subsystem may belong to a different module than the item
1306	 * being created.  We don't want to safely pin the new item but
1307	 * fail to pin the subsystem it sits under.
1308	 */
1309	if (!subsys->su_group.cg_item.ci_type) {
1310		ret = -EINVAL;
1311		goto out_put;
1312	}
1313	subsys_owner = subsys->su_group.cg_item.ci_type->ct_owner;
1314	if (!try_module_get(subsys_owner)) {
1315		ret = -EINVAL;
1316		goto out_put;
1317	}
1318
1319	name = kmalloc(dentry->d_name.len + 1, GFP_KERNEL);
1320	if (!name) {
1321		ret = -ENOMEM;
1322		goto out_subsys_put;
1323	}
1324
1325	snprintf(name, dentry->d_name.len + 1, "%s", dentry->d_name.name);
1326
1327	mutex_lock(&subsys->su_mutex);
1328	if (type->ct_group_ops->make_group) {
1329		group = type->ct_group_ops->make_group(to_config_group(parent_item), name);
1330		if (!group)
1331			group = ERR_PTR(-ENOMEM);
1332		if (!IS_ERR(group)) {
1333			link_group(to_config_group(parent_item), group);
1334			item = &group->cg_item;
1335		} else
1336			ret = PTR_ERR(group);
1337	} else {
1338		item = type->ct_group_ops->make_item(to_config_group(parent_item), name);
1339		if (!item)
1340			item = ERR_PTR(-ENOMEM);
1341		if (!IS_ERR(item))
1342			link_obj(parent_item, item);
1343		else
1344			ret = PTR_ERR(item);
1345	}
1346	mutex_unlock(&subsys->su_mutex);
1347
1348	kfree(name);
1349	if (ret) {
1350		/*
1351		 * If ret != 0, then link_obj() was never called.
1352		 * There are no extra references to clean up.
1353		 */
1354		goto out_subsys_put;
1355	}
1356
1357	/*
1358	 * link_obj() has been called (via link_group() for groups).
1359	 * From here on out, errors must clean that up.
1360	 */
1361
1362	type = item->ci_type;
1363	if (!type) {
1364		ret = -EINVAL;
1365		goto out_unlink;
1366	}
1367
1368	new_item_owner = type->ct_owner;
1369	if (!try_module_get(new_item_owner)) {
1370		ret = -EINVAL;
1371		goto out_unlink;
1372	}
1373
1374	/*
1375	 * I hate doing it this way, but if there is
1376	 * an error,  module_put() probably should
1377	 * happen after any cleanup.
1378	 */
1379	module_got = 1;
1380
1381	/*
1382	 * Make racing rmdir() fail if it did not tag parent with
1383	 * CONFIGFS_USET_DROPPING
1384	 * Note: if CONFIGFS_USET_DROPPING is already set, attach_group() will
1385	 * fail and let rmdir() terminate correctly
1386	 */
1387	spin_lock(&configfs_dirent_lock);
1388	/* This will make configfs_detach_prep() fail */
1389	sd->s_type |= CONFIGFS_USET_IN_MKDIR;
1390	spin_unlock(&configfs_dirent_lock);
1391
1392	if (group)
1393		ret = configfs_attach_group(parent_item, item, dentry, frag);
1394	else
1395		ret = configfs_attach_item(parent_item, item, dentry, frag);
1396
1397	spin_lock(&configfs_dirent_lock);
1398	sd->s_type &= ~CONFIGFS_USET_IN_MKDIR;
1399	if (!ret)
1400		configfs_dir_set_ready(dentry->d_fsdata);
1401	spin_unlock(&configfs_dirent_lock);
1402
1403out_unlink:
1404	if (ret) {
1405		/* Tear down everything we built up */
1406		mutex_lock(&subsys->su_mutex);
1407
1408		client_disconnect_notify(parent_item, item);
1409		if (group)
1410			unlink_group(group);
1411		else
1412			unlink_obj(item);
1413		client_drop_item(parent_item, item);
1414
1415		mutex_unlock(&subsys->su_mutex);
1416
1417		if (module_got)
1418			module_put(new_item_owner);
1419	}
1420
1421out_subsys_put:
1422	if (ret)
1423		module_put(subsys_owner);
1424
1425out_put:
1426	/*
1427	 * link_obj()/link_group() took a reference from child->parent,
1428	 * so the parent is safely pinned.  We can drop our working
1429	 * reference.
1430	 */
1431	config_item_put(parent_item);
1432	put_fragment(frag);
1433
1434out:
1435	return ret;
1436}
1437
1438static int configfs_rmdir(struct inode *dir, struct dentry *dentry)
1439{
1440	struct config_item *parent_item;
1441	struct config_item *item;
1442	struct configfs_subsystem *subsys;
1443	struct configfs_dirent *sd;
1444	struct configfs_fragment *frag;
1445	struct module *subsys_owner = NULL, *dead_item_owner = NULL;
1446	int ret;
1447
1448	sd = dentry->d_fsdata;
1449	if (sd->s_type & CONFIGFS_USET_DEFAULT)
1450		return -EPERM;
1451
1452	/* Get a working ref until we have the child */
1453	parent_item = configfs_get_config_item(dentry->d_parent);
1454	subsys = to_config_group(parent_item)->cg_subsys;
1455	BUG_ON(!subsys);
1456
1457	if (!parent_item->ci_type) {
1458		config_item_put(parent_item);
1459		return -EINVAL;
1460	}
1461
1462	/* configfs_mkdir() shouldn't have allowed this */
1463	BUG_ON(!subsys->su_group.cg_item.ci_type);
1464	subsys_owner = subsys->su_group.cg_item.ci_type->ct_owner;
1465
1466	/*
1467	 * Ensure that no racing symlink() will make detach_prep() fail while
1468	 * the new link is temporarily attached
1469	 */
1470	do {
1471		struct dentry *wait;
1472
1473		mutex_lock(&configfs_symlink_mutex);
1474		spin_lock(&configfs_dirent_lock);
1475		/*
1476		 * Here's where we check for dependents.  We're protected by
1477		 * configfs_dirent_lock.
1478		 * If no dependent, atomically tag the item as dropping.
1479		 */
1480		ret = sd->s_dependent_count ? -EBUSY : 0;
1481		if (!ret) {
1482			ret = configfs_detach_prep(dentry, &wait);
1483			if (ret)
1484				configfs_detach_rollback(dentry);
1485		}
1486		spin_unlock(&configfs_dirent_lock);
1487		mutex_unlock(&configfs_symlink_mutex);
1488
1489		if (ret) {
1490			if (ret != -EAGAIN) {
1491				config_item_put(parent_item);
1492				return ret;
1493			}
1494
1495			/* Wait until the racing operation terminates */
1496			inode_lock(d_inode(wait));
1497			inode_unlock(d_inode(wait));
1498			dput(wait);
1499		}
1500	} while (ret == -EAGAIN);
1501
1502	frag = sd->s_frag;
1503	if (down_write_killable(&frag->frag_sem)) {
1504		spin_lock(&configfs_dirent_lock);
1505		configfs_detach_rollback(dentry);
1506		spin_unlock(&configfs_dirent_lock);
1507		config_item_put(parent_item);
1508		return -EINTR;
1509	}
1510	frag->frag_dead = true;
1511	up_write(&frag->frag_sem);
1512
1513	/* Get a working ref for the duration of this function */
1514	item = configfs_get_config_item(dentry);
1515
1516	/* Drop reference from above, item already holds one. */
1517	config_item_put(parent_item);
1518
1519	if (item->ci_type)
1520		dead_item_owner = item->ci_type->ct_owner;
1521
1522	if (sd->s_type & CONFIGFS_USET_DIR) {
1523		configfs_detach_group(item);
1524
1525		mutex_lock(&subsys->su_mutex);
1526		client_disconnect_notify(parent_item, item);
1527		unlink_group(to_config_group(item));
1528	} else {
1529		configfs_detach_item(item);
1530
1531		mutex_lock(&subsys->su_mutex);
1532		client_disconnect_notify(parent_item, item);
1533		unlink_obj(item);
1534	}
1535
1536	client_drop_item(parent_item, item);
1537	mutex_unlock(&subsys->su_mutex);
1538
1539	/* Drop our reference from above */
1540	config_item_put(item);
1541
1542	module_put(dead_item_owner);
1543	module_put(subsys_owner);
1544
1545	return 0;
1546}
1547
1548const struct inode_operations configfs_dir_inode_operations = {
1549	.mkdir		= configfs_mkdir,
1550	.rmdir		= configfs_rmdir,
1551	.symlink	= configfs_symlink,
1552	.unlink		= configfs_unlink,
1553	.lookup		= configfs_lookup,
1554	.setattr	= configfs_setattr,
1555};
1556
1557const struct inode_operations configfs_root_inode_operations = {
1558	.lookup		= configfs_lookup,
1559	.setattr	= configfs_setattr,
1560};
1561
1562static int configfs_dir_open(struct inode *inode, struct file *file)
1563{
1564	struct dentry * dentry = file->f_path.dentry;
1565	struct configfs_dirent * parent_sd = dentry->d_fsdata;
1566	int err;
1567
1568	inode_lock(d_inode(dentry));
1569	/*
1570	 * Fake invisibility if dir belongs to a group/default groups hierarchy
1571	 * being attached
1572	 */
1573	err = -ENOENT;
1574	if (configfs_dirent_is_ready(parent_sd)) {
1575		file->private_data = configfs_new_dirent(parent_sd, NULL, 0, NULL);
1576		if (IS_ERR(file->private_data))
1577			err = PTR_ERR(file->private_data);
1578		else
1579			err = 0;
1580	}
1581	inode_unlock(d_inode(dentry));
1582
1583	return err;
1584}
1585
1586static int configfs_dir_close(struct inode *inode, struct file *file)
1587{
1588	struct dentry * dentry = file->f_path.dentry;
1589	struct configfs_dirent * cursor = file->private_data;
1590
1591	inode_lock(d_inode(dentry));
1592	spin_lock(&configfs_dirent_lock);
1593	list_del_init(&cursor->s_sibling);
1594	spin_unlock(&configfs_dirent_lock);
1595	inode_unlock(d_inode(dentry));
1596
1597	release_configfs_dirent(cursor);
1598
1599	return 0;
1600}
1601
1602/* Relationship between s_mode and the DT_xxx types */
1603static inline unsigned char dt_type(struct configfs_dirent *sd)
1604{
1605	return (sd->s_mode >> 12) & 15;
1606}
1607
1608static int configfs_readdir(struct file *file, struct dir_context *ctx)
1609{
1610	struct dentry *dentry = file->f_path.dentry;
1611	struct super_block *sb = dentry->d_sb;
1612	struct configfs_dirent * parent_sd = dentry->d_fsdata;
1613	struct configfs_dirent *cursor = file->private_data;
1614	struct list_head *p, *q = &cursor->s_sibling;
1615	ino_t ino = 0;
1616
1617	if (!dir_emit_dots(file, ctx))
1618		return 0;
1619	spin_lock(&configfs_dirent_lock);
1620	if (ctx->pos == 2)
1621		list_move(q, &parent_sd->s_children);
1622	for (p = q->next; p != &parent_sd->s_children; p = p->next) {
1623		struct configfs_dirent *next;
1624		const char *name;
1625		int len;
1626		struct inode *inode = NULL;
1627
1628		next = list_entry(p, struct configfs_dirent, s_sibling);
1629		if (!next->s_element)
1630			continue;
1631
1632		/*
1633		 * We'll have a dentry and an inode for
1634		 * PINNED items and for open attribute
1635		 * files.  We lock here to prevent a race
1636		 * with configfs_d_iput() clearing
1637		 * s_dentry before calling iput().
1638		 *
1639		 * Why do we go to the trouble?  If
1640		 * someone has an attribute file open,
1641		 * the inode number should match until
1642		 * they close it.  Beyond that, we don't
1643		 * care.
1644		 */
1645		dentry = next->s_dentry;
1646		if (dentry)
1647			inode = d_inode(dentry);
1648		if (inode)
1649			ino = inode->i_ino;
1650		spin_unlock(&configfs_dirent_lock);
1651		if (!inode)
1652			ino = iunique(sb, 2);
1653
1654		name = configfs_get_name(next);
1655		len = strlen(name);
1656
1657		if (!dir_emit(ctx, name, len, ino, dt_type(next)))
 
1658			return 0;
1659
1660		spin_lock(&configfs_dirent_lock);
1661		list_move(q, p);
1662		p = q;
1663		ctx->pos++;
1664	}
1665	spin_unlock(&configfs_dirent_lock);
1666	return 0;
1667}
1668
1669static loff_t configfs_dir_lseek(struct file *file, loff_t offset, int whence)
1670{
1671	struct dentry * dentry = file->f_path.dentry;
1672
1673	switch (whence) {
1674		case 1:
1675			offset += file->f_pos;
1676			fallthrough;
1677		case 0:
1678			if (offset >= 0)
1679				break;
1680			fallthrough;
1681		default:
1682			return -EINVAL;
1683	}
1684	if (offset != file->f_pos) {
1685		file->f_pos = offset;
1686		if (file->f_pos >= 2) {
1687			struct configfs_dirent *sd = dentry->d_fsdata;
1688			struct configfs_dirent *cursor = file->private_data;
1689			struct list_head *p;
1690			loff_t n = file->f_pos - 2;
1691
1692			spin_lock(&configfs_dirent_lock);
1693			list_del(&cursor->s_sibling);
1694			p = sd->s_children.next;
1695			while (n && p != &sd->s_children) {
1696				struct configfs_dirent *next;
1697				next = list_entry(p, struct configfs_dirent,
1698						   s_sibling);
1699				if (next->s_element)
1700					n--;
1701				p = p->next;
1702			}
1703			list_add_tail(&cursor->s_sibling, p);
1704			spin_unlock(&configfs_dirent_lock);
1705		}
1706	}
1707	return offset;
1708}
1709
1710const struct file_operations configfs_dir_operations = {
1711	.open		= configfs_dir_open,
1712	.release	= configfs_dir_close,
1713	.llseek		= configfs_dir_lseek,
1714	.read		= generic_read_dir,
1715	.iterate_shared	= configfs_readdir,
1716};
1717
1718/**
1719 * configfs_register_group - creates a parent-child relation between two groups
1720 * @parent_group:	parent group
1721 * @group:		child group
1722 *
1723 * link groups, creates dentry for the child and attaches it to the
1724 * parent dentry.
1725 *
1726 * Return: 0 on success, negative errno code on error
1727 */
1728int configfs_register_group(struct config_group *parent_group,
1729			    struct config_group *group)
1730{
1731	struct configfs_subsystem *subsys = parent_group->cg_subsys;
1732	struct dentry *parent;
1733	struct configfs_fragment *frag;
1734	int ret;
1735
1736	frag = new_fragment();
1737	if (!frag)
1738		return -ENOMEM;
1739
1740	mutex_lock(&subsys->su_mutex);
1741	link_group(parent_group, group);
1742	mutex_unlock(&subsys->su_mutex);
1743
1744	parent = parent_group->cg_item.ci_dentry;
1745
1746	inode_lock_nested(d_inode(parent), I_MUTEX_PARENT);
1747	ret = create_default_group(parent_group, group, frag);
1748	if (ret)
1749		goto err_out;
1750
1751	spin_lock(&configfs_dirent_lock);
1752	configfs_dir_set_ready(group->cg_item.ci_dentry->d_fsdata);
1753	spin_unlock(&configfs_dirent_lock);
1754	inode_unlock(d_inode(parent));
1755	put_fragment(frag);
1756	return 0;
1757err_out:
1758	inode_unlock(d_inode(parent));
1759	mutex_lock(&subsys->su_mutex);
1760	unlink_group(group);
1761	mutex_unlock(&subsys->su_mutex);
1762	put_fragment(frag);
1763	return ret;
1764}
1765EXPORT_SYMBOL(configfs_register_group);
1766
1767/**
1768 * configfs_unregister_group() - unregisters a child group from its parent
1769 * @group: parent group to be unregistered
1770 *
1771 * Undoes configfs_register_group()
1772 */
1773void configfs_unregister_group(struct config_group *group)
1774{
1775	struct configfs_subsystem *subsys = group->cg_subsys;
1776	struct dentry *dentry = group->cg_item.ci_dentry;
1777	struct dentry *parent = group->cg_item.ci_parent->ci_dentry;
1778	struct configfs_dirent *sd = dentry->d_fsdata;
1779	struct configfs_fragment *frag = sd->s_frag;
1780
1781	down_write(&frag->frag_sem);
1782	frag->frag_dead = true;
1783	up_write(&frag->frag_sem);
1784
1785	inode_lock_nested(d_inode(parent), I_MUTEX_PARENT);
1786	spin_lock(&configfs_dirent_lock);
1787	configfs_detach_prep(dentry, NULL);
1788	spin_unlock(&configfs_dirent_lock);
1789
1790	configfs_detach_group(&group->cg_item);
1791	d_inode(dentry)->i_flags |= S_DEAD;
1792	dont_mount(dentry);
1793	d_drop(dentry);
1794	fsnotify_rmdir(d_inode(parent), dentry);
1795	inode_unlock(d_inode(parent));
1796
1797	dput(dentry);
1798
1799	mutex_lock(&subsys->su_mutex);
1800	unlink_group(group);
1801	mutex_unlock(&subsys->su_mutex);
1802}
1803EXPORT_SYMBOL(configfs_unregister_group);
1804
1805/**
1806 * configfs_register_default_group() - allocates and registers a child group
1807 * @parent_group:	parent group
1808 * @name:		child group name
1809 * @item_type:		child item type description
1810 *
1811 * boilerplate to allocate and register a child group with its parent. We need
1812 * kzalloc'ed memory because child's default_group is initially empty.
1813 *
1814 * Return: allocated config group or ERR_PTR() on error
1815 */
1816struct config_group *
1817configfs_register_default_group(struct config_group *parent_group,
1818				const char *name,
1819				const struct config_item_type *item_type)
1820{
1821	int ret;
1822	struct config_group *group;
1823
1824	group = kzalloc(sizeof(*group), GFP_KERNEL);
1825	if (!group)
1826		return ERR_PTR(-ENOMEM);
1827	config_group_init_type_name(group, name, item_type);
1828
1829	ret = configfs_register_group(parent_group, group);
1830	if (ret) {
1831		kfree(group);
1832		return ERR_PTR(ret);
1833	}
1834	return group;
1835}
1836EXPORT_SYMBOL(configfs_register_default_group);
1837
1838/**
1839 * configfs_unregister_default_group() - unregisters and frees a child group
1840 * @group:	the group to act on
1841 */
1842void configfs_unregister_default_group(struct config_group *group)
1843{
1844	configfs_unregister_group(group);
1845	kfree(group);
1846}
1847EXPORT_SYMBOL(configfs_unregister_default_group);
1848
1849int configfs_register_subsystem(struct configfs_subsystem *subsys)
1850{
1851	int err;
1852	struct config_group *group = &subsys->su_group;
1853	struct dentry *dentry;
1854	struct dentry *root;
1855	struct configfs_dirent *sd;
1856	struct configfs_fragment *frag;
1857
1858	frag = new_fragment();
1859	if (!frag)
1860		return -ENOMEM;
1861
1862	root = configfs_pin_fs();
1863	if (IS_ERR(root)) {
1864		put_fragment(frag);
1865		return PTR_ERR(root);
1866	}
1867
1868	if (!group->cg_item.ci_name)
1869		group->cg_item.ci_name = group->cg_item.ci_namebuf;
1870
1871	sd = root->d_fsdata;
1872	mutex_lock(&configfs_subsystem_mutex);
1873	link_group(to_config_group(sd->s_element), group);
1874	mutex_unlock(&configfs_subsystem_mutex);
1875
1876	inode_lock_nested(d_inode(root), I_MUTEX_PARENT);
1877
1878	err = -ENOMEM;
1879	dentry = d_alloc_name(root, group->cg_item.ci_name);
1880	if (dentry) {
1881		d_add(dentry, NULL);
1882
1883		err = configfs_attach_group(sd->s_element, &group->cg_item,
1884					    dentry, frag);
 
 
 
1885		if (err) {
1886			BUG_ON(d_inode(dentry));
1887			d_drop(dentry);
1888			dput(dentry);
1889		} else {
1890			spin_lock(&configfs_dirent_lock);
1891			configfs_dir_set_ready(dentry->d_fsdata);
1892			spin_unlock(&configfs_dirent_lock);
1893		}
1894	}
1895
1896	inode_unlock(d_inode(root));
1897
1898	if (err) {
1899		mutex_lock(&configfs_subsystem_mutex);
1900		unlink_group(group);
1901		mutex_unlock(&configfs_subsystem_mutex);
1902		configfs_release_fs();
1903	}
1904	put_fragment(frag);
1905
1906	return err;
1907}
1908
1909void configfs_unregister_subsystem(struct configfs_subsystem *subsys)
1910{
1911	struct config_group *group = &subsys->su_group;
1912	struct dentry *dentry = group->cg_item.ci_dentry;
1913	struct dentry *root = dentry->d_sb->s_root;
1914	struct configfs_dirent *sd = dentry->d_fsdata;
1915	struct configfs_fragment *frag = sd->s_frag;
1916
1917	if (dentry->d_parent != root) {
1918		pr_err("Tried to unregister non-subsystem!\n");
1919		return;
1920	}
1921
1922	down_write(&frag->frag_sem);
1923	frag->frag_dead = true;
1924	up_write(&frag->frag_sem);
1925
1926	inode_lock_nested(d_inode(root),
1927			  I_MUTEX_PARENT);
1928	inode_lock_nested(d_inode(dentry), I_MUTEX_CHILD);
1929	mutex_lock(&configfs_symlink_mutex);
1930	spin_lock(&configfs_dirent_lock);
1931	if (configfs_detach_prep(dentry, NULL)) {
1932		pr_err("Tried to unregister non-empty subsystem!\n");
1933	}
1934	spin_unlock(&configfs_dirent_lock);
1935	mutex_unlock(&configfs_symlink_mutex);
1936	configfs_detach_group(&group->cg_item);
1937	d_inode(dentry)->i_flags |= S_DEAD;
1938	dont_mount(dentry);
1939	inode_unlock(d_inode(dentry));
1940
1941	d_drop(dentry);
1942	fsnotify_rmdir(d_inode(root), dentry);
1943
1944	inode_unlock(d_inode(root));
1945
1946	dput(dentry);
1947
1948	mutex_lock(&configfs_subsystem_mutex);
1949	unlink_group(group);
1950	mutex_unlock(&configfs_subsystem_mutex);
1951	configfs_release_fs();
1952}
1953
1954EXPORT_SYMBOL(configfs_register_subsystem);
1955EXPORT_SYMBOL(configfs_unregister_subsystem);