linux/fs/kernfs/dir.c
<<
>>
Prefs
   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
 551static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
 552{
 553        struct kernfs_node *kn;
 554
 555        if (flags & LOOKUP_RCU)
 556                return -ECHILD;
 557
 558        /* Always perform fresh lookup for negatives */
 559        if (d_really_is_negative(dentry))
 560                goto out_bad_unlocked;
 561
 562        kn = kernfs_dentry_node(dentry);
 563        mutex_lock(&kernfs_mutex);
 564
 565        /* The kernfs node has been deactivated */
 566        if (!kernfs_active(kn))
 567                goto out_bad;
 568
 569        /* The kernfs node has been moved? */
 570        if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
 571                goto out_bad;
 572
 573        /* The kernfs node has been renamed */
 574        if (strcmp(dentry->d_name.name, kn->name) != 0)
 575                goto out_bad;
 576
 577        /* The kernfs node has been moved to a different namespace */
 578        if (kn->parent && kernfs_ns_enabled(kn->parent) &&
 579            kernfs_info(dentry->d_sb)->ns != kn->ns)
 580                goto out_bad;
 581
 582        mutex_unlock(&kernfs_mutex);
 583        return 1;
 584out_bad:
 585        mutex_unlock(&kernfs_mutex);
 586out_bad_unlocked:
 587        return 0;
 588}
 589
 590const struct dentry_operations kernfs_dops = {
 591        .d_revalidate   = kernfs_dop_revalidate,
 592};
 593
 594/**
 595 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
 596 * @dentry: the dentry in question
 597 *
 598 * Return the kernfs_node associated with @dentry.  If @dentry is not a
 599 * kernfs one, %NULL is returned.
 600 *
 601 * While the returned kernfs_node will stay accessible as long as @dentry
 602 * is accessible, the returned node can be in any state and the caller is
 603 * fully responsible for determining what's accessible.
 604 */
 605struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
 606{
 607        if (dentry->d_sb->s_op == &kernfs_sops)
 608                return kernfs_dentry_node(dentry);
 609        return NULL;
 610}
 611
 612static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
 613                                             struct kernfs_node *parent,
 614                                             const char *name, umode_t mode,
 615                                             kuid_t uid, kgid_t gid,
 616                                             unsigned flags)
 617{
 618        struct kernfs_node *kn;
 619        u32 id_highbits;
 620        int ret;
 621
 622        name = kstrdup_const(name, GFP_KERNEL);
 623        if (!name)
 624                return NULL;
 625
 626        kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
 627        if (!kn)
 628                goto err_out1;
 629
 630        idr_preload(GFP_KERNEL);
 631        spin_lock(&kernfs_idr_lock);
 632        ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
 633        if (ret >= 0 && ret < root->last_id_lowbits)
 634                root->id_highbits++;
 635        id_highbits = root->id_highbits;
 636        root->last_id_lowbits = ret;
 637        spin_unlock(&kernfs_idr_lock);
 638        idr_preload_end();
 639        if (ret < 0)
 640                goto err_out2;
 641
 642        kn->id = (u64)id_highbits << 32 | ret;
 643
 644        atomic_set(&kn->count, 1);
 645        atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
 646        RB_CLEAR_NODE(&kn->rb);
 647
 648        kn->name = name;
 649        kn->mode = mode;
 650        kn->flags = flags;
 651
 652        if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
 653                struct iattr iattr = {
 654                        .ia_valid = ATTR_UID | ATTR_GID,
 655                        .ia_uid = uid,
 656                        .ia_gid = gid,
 657                };
 658
 659                ret = __kernfs_setattr(kn, &iattr);
 660                if (ret < 0)
 661                        goto err_out3;
 662        }
 663
 664        if (parent) {
 665                ret = security_kernfs_init_security(parent, kn);
 666                if (ret)
 667                        goto err_out3;
 668        }
 669
 670        return kn;
 671
 672 err_out3:
 673        idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
 674 err_out2:
 675        kmem_cache_free(kernfs_node_cache, kn);
 676 err_out1:
 677        kfree_const(name);
 678        return NULL;
 679}
 680
 681struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
 682                                    const char *name, umode_t mode,
 683                                    kuid_t uid, kgid_t gid,
 684                                    unsigned flags)
 685{
 686        struct kernfs_node *kn;
 687
 688        kn = __kernfs_new_node(kernfs_root(parent), parent,
 689                               name, mode, uid, gid, flags);
 690        if (kn) {
 691                kernfs_get(parent);
 692                kn->parent = parent;
 693        }
 694        return kn;
 695}
 696
 697/*
 698 * kernfs_find_and_get_node_by_id - get kernfs_node from node id
 699 * @root: the kernfs root
 700 * @id: the target node id
 701 *
 702 * @id's lower 32bits encode ino and upper gen.  If the gen portion is
 703 * zero, all generations are matched.
 704 *
 705 * RETURNS:
 706 * NULL on failure. Return a kernfs node with reference counter incremented
 707 */
 708struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
 709                                                   u64 id)
 710{
 711        struct kernfs_node *kn;
 712        ino_t ino = kernfs_id_ino(id);
 713        u32 gen = kernfs_id_gen(id);
 714
 715        spin_lock(&kernfs_idr_lock);
 716
 717        kn = idr_find(&root->ino_idr, (u32)ino);
 718        if (!kn)
 719                goto err_unlock;
 720
 721        if (sizeof(ino_t) >= sizeof(u64)) {
 722                /* we looked up with the low 32bits, compare the whole */
 723                if (kernfs_ino(kn) != ino)
 724                        goto err_unlock;
 725        } else {
 726                /* 0 matches all generations */
 727                if (unlikely(gen && kernfs_gen(kn) != gen))
 728                        goto err_unlock;
 729        }
 730
 731        /*
 732         * ACTIVATED is protected with kernfs_mutex but it was clear when
 733         * @kn was added to idr and we just wanna see it set.  No need to
 734         * grab kernfs_mutex.
 735         */
 736        if (unlikely(!(kn->flags & KERNFS_ACTIVATED) ||
 737                     !atomic_inc_not_zero(&kn->count)))
 738                goto err_unlock;
 739
 740        spin_unlock(&kernfs_idr_lock);
 741        return kn;
 742err_unlock:
 743        spin_unlock(&kernfs_idr_lock);
 744        return NULL;
 745}
 746
 747/**
 748 *      kernfs_add_one - add kernfs_node to parent without warning
 749 *      @kn: kernfs_node to be added
 750 *
 751 *      The caller must already have initialized @kn->parent.  This
 752 *      function increments nlink of the parent's inode if @kn is a
 753 *      directory and link into the children list of the parent.
 754 *
 755 *      RETURNS:
 756 *      0 on success, -EEXIST if entry with the given name already
 757 *      exists.
 758 */
 759int kernfs_add_one(struct kernfs_node *kn)
 760{
 761        struct kernfs_node *parent = kn->parent;
 762        struct kernfs_iattrs *ps_iattr;
 763        bool has_ns;
 764        int ret;
 765
 766        mutex_lock(&kernfs_mutex);
 767
 768        ret = -EINVAL;
 769        has_ns = kernfs_ns_enabled(parent);
 770        if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
 771                 has_ns ? "required" : "invalid", parent->name, kn->name))
 772                goto out_unlock;
 773
 774        if (kernfs_type(parent) != KERNFS_DIR)
 775                goto out_unlock;
 776
 777        ret = -ENOENT;
 778        if (parent->flags & KERNFS_EMPTY_DIR)
 779                goto out_unlock;
 780
 781        if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
 782                goto out_unlock;
 783
 784        kn->hash = kernfs_name_hash(kn->name, kn->ns);
 785
 786        ret = kernfs_link_sibling(kn);
 787        if (ret)
 788                goto out_unlock;
 789
 790        /* Update timestamps on the parent */
 791        ps_iattr = parent->iattr;
 792        if (ps_iattr) {
 793                ktime_get_real_ts64(&ps_iattr->ia_ctime);
 794                ps_iattr->ia_mtime = ps_iattr->ia_ctime;
 795        }
 796
 797        mutex_unlock(&kernfs_mutex);
 798
 799        /*
 800         * Activate the new node unless CREATE_DEACTIVATED is requested.
 801         * If not activated here, the kernfs user is responsible for
 802         * activating the node with kernfs_activate().  A node which hasn't
 803         * been activated is not visible to userland and its removal won't
 804         * trigger deactivation.
 805         */
 806        if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
 807                kernfs_activate(kn);
 808        return 0;
 809
 810out_unlock:
 811        mutex_unlock(&kernfs_mutex);
 812        return ret;
 813}
 814
 815/**
 816 * kernfs_find_ns - find kernfs_node with the given name
 817 * @parent: kernfs_node to search under
 818 * @name: name to look for
 819 * @ns: the namespace tag to use
 820 *
 821 * Look for kernfs_node with name @name under @parent.  Returns pointer to
 822 * the found kernfs_node on success, %NULL on failure.
 823 */
 824static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
 825                                          const unsigned char *name,
 826                                          const void *ns)
 827{
 828        struct rb_node *node = parent->dir.children.rb_node;
 829        bool has_ns = kernfs_ns_enabled(parent);
 830        unsigned int hash;
 831
 832        lockdep_assert_held(&kernfs_mutex);
 833
 834        if (has_ns != (bool)ns) {
 835                WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
 836                     has_ns ? "required" : "invalid", parent->name, name);
 837                return NULL;
 838        }
 839
 840        hash = kernfs_name_hash(name, ns);
 841        while (node) {
 842                struct kernfs_node *kn;
 843                int result;
 844
 845                kn = rb_to_kn(node);
 846                result = kernfs_name_compare(hash, name, ns, kn);
 847                if (result < 0)
 848                        node = node->rb_left;
 849                else if (result > 0)
 850                        node = node->rb_right;
 851                else
 852                        return kn;
 853        }
 854        return NULL;
 855}
 856
 857static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
 858                                          const unsigned char *path,
 859                                          const void *ns)
 860{
 861        size_t len;
 862        char *p, *name;
 863
 864        lockdep_assert_held(&kernfs_mutex);
 865
 866        /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
 867        spin_lock_irq(&kernfs_rename_lock);
 868
 869        len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
 870
 871        if (len >= sizeof(kernfs_pr_cont_buf)) {
 872                spin_unlock_irq(&kernfs_rename_lock);
 873                return NULL;
 874        }
 875
 876        p = kernfs_pr_cont_buf;
 877
 878        while ((name = strsep(&p, "/")) && parent) {
 879                if (*name == '\0')
 880                        continue;
 881                parent = kernfs_find_ns(parent, name, ns);
 882        }
 883
 884        spin_unlock_irq(&kernfs_rename_lock);
 885
 886        return parent;
 887}
 888
 889/**
 890 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
 891 * @parent: kernfs_node to search under
 892 * @name: name to look for
 893 * @ns: the namespace tag to use
 894 *
 895 * Look for kernfs_node with name @name under @parent and get a reference
 896 * if found.  This function may sleep and returns pointer to the found
 897 * kernfs_node on success, %NULL on failure.
 898 */
 899struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
 900                                           const char *name, const void *ns)
 901{
 902        struct kernfs_node *kn;
 903
 904        mutex_lock(&kernfs_mutex);
 905        kn = kernfs_find_ns(parent, name, ns);
 906        kernfs_get(kn);
 907        mutex_unlock(&kernfs_mutex);
 908
 909        return kn;
 910}
 911EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
 912
 913/**
 914 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
 915 * @parent: kernfs_node to search under
 916 * @path: path to look for
 917 * @ns: the namespace tag to use
 918 *
 919 * Look for kernfs_node with path @path under @parent and get a reference
 920 * if found.  This function may sleep and returns pointer to the found
 921 * kernfs_node on success, %NULL on failure.
 922 */
 923struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
 924                                           const char *path, const void *ns)
 925{
 926        struct kernfs_node *kn;
 927
 928        mutex_lock(&kernfs_mutex);
 929        kn = kernfs_walk_ns(parent, path, ns);
 930        kernfs_get(kn);
 931        mutex_unlock(&kernfs_mutex);
 932
 933        return kn;
 934}
 935
 936/**
 937 * kernfs_create_root - create a new kernfs hierarchy
 938 * @scops: optional syscall operations for the hierarchy
 939 * @flags: KERNFS_ROOT_* flags
 940 * @priv: opaque data associated with the new directory
 941 *
 942 * Returns the root of the new hierarchy on success, ERR_PTR() value on
 943 * failure.
 944 */
 945struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
 946                                       unsigned int flags, void *priv)
 947{
 948        struct kernfs_root *root;
 949        struct kernfs_node *kn;
 950
 951        root = kzalloc(sizeof(*root), GFP_KERNEL);
 952        if (!root)
 953                return ERR_PTR(-ENOMEM);
 954
 955        idr_init(&root->ino_idr);
 956        INIT_LIST_HEAD(&root->supers);
 957
 958        /*
 959         * On 64bit ino setups, id is ino.  On 32bit, low 32bits are ino.
 960         * High bits generation.  The starting value for both ino and
 961         * genenration is 1.  Initialize upper 32bit allocation
 962         * accordingly.
 963         */
 964        if (sizeof(ino_t) >= sizeof(u64))
 965                root->id_highbits = 0;
 966        else
 967                root->id_highbits = 1;
 968
 969        kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
 970                               GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
 971                               KERNFS_DIR);
 972        if (!kn) {
 973                idr_destroy(&root->ino_idr);
 974                kfree(root);
 975                return ERR_PTR(-ENOMEM);
 976        }
 977
 978        kn->priv = priv;
 979        kn->dir.root = root;
 980
 981        root->syscall_ops = scops;
 982        root->flags = flags;
 983        root->kn = kn;
 984        init_waitqueue_head(&root->deactivate_waitq);
 985
 986        if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
 987                kernfs_activate(kn);
 988
 989        return root;
 990}
 991
 992/**
 993 * kernfs_destroy_root - destroy a kernfs hierarchy
 994 * @root: root of the hierarchy to destroy
 995 *
 996 * Destroy the hierarchy anchored at @root by removing all existing
 997 * directories and destroying @root.
 998 */
 999void kernfs_destroy_root(struct kernfs_root *root)
1000{
1001        kernfs_remove(root->kn);        /* will also free @root */
1002}
1003
1004/**
1005 * kernfs_create_dir_ns - create a directory
1006 * @parent: parent in which to create a new directory
1007 * @name: name of the new directory
1008 * @mode: mode of the new directory
1009 * @uid: uid of the new directory
1010 * @gid: gid of the new directory
1011 * @priv: opaque data associated with the new directory
1012 * @ns: optional namespace tag of the directory
1013 *
1014 * Returns the created node on success, ERR_PTR() value on failure.
1015 */
1016struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1017                                         const char *name, umode_t mode,
1018                                         kuid_t uid, kgid_t gid,
1019                                         void *priv, const void *ns)
1020{
1021        struct kernfs_node *kn;
1022        int rc;
1023
1024        /* allocate */
1025        kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1026                             uid, gid, KERNFS_DIR);
1027        if (!kn)
1028                return ERR_PTR(-ENOMEM);
1029
1030        kn->dir.root = parent->dir.root;
1031        kn->ns = ns;
1032        kn->priv = priv;
1033
1034        /* link in */
1035        rc = kernfs_add_one(kn);
1036        if (!rc)
1037                return kn;
1038
1039        kernfs_put(kn);
1040        return ERR_PTR(rc);
1041}
1042
1043/**
1044 * kernfs_create_empty_dir - create an always empty directory
1045 * @parent: parent in which to create a new directory
1046 * @name: name of the new directory
1047 *
1048 * Returns the created node on success, ERR_PTR() value on failure.
1049 */
1050struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1051                                            const char *name)
1052{
1053        struct kernfs_node *kn;
1054        int rc;
1055
1056        /* allocate */
1057        kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1058                             GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1059        if (!kn)
1060                return ERR_PTR(-ENOMEM);
1061
1062        kn->flags |= KERNFS_EMPTY_DIR;
1063        kn->dir.root = parent->dir.root;
1064        kn->ns = NULL;
1065        kn->priv = NULL;
1066
1067        /* link in */
1068        rc = kernfs_add_one(kn);
1069        if (!rc)
1070                return kn;
1071
1072        kernfs_put(kn);
1073        return ERR_PTR(rc);
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};
1708