fs/kernfs/dir.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
kernfs_active
kernfs_lockdep
kernfs_name_locked
kernfs_depth
kernfs_common_ancestor
kernfs_path_from_node_locked
kernfs_name
kernfs_path_from_node
pr_cont_kernfs_name
pr_cont_kernfs_path
kernfs_get_parent
kernfs_name_hash
kernfs_name_compare
kernfs_sd_compare
kernfs_link_sibling
kernfs_unlink_sibling
kernfs_get_active
kernfs_put_active
kernfs_drain
kernfs_get
kernfs_put
kernfs_dop_revalidate
kernfs_node_from_dentry
__kernfs_new_node
kernfs_new_node
kernfs_find_and_get_node_by_ino
kernfs_add_one
kernfs_find_ns
kernfs_walk_ns
kernfs_find_and_get_ns
kernfs_walk_and_get_ns
kernfs_create_root
kernfs_destroy_root
kernfs_create_dir_ns
kernfs_create_empty_dir
kernfs_iop_lookup
kernfs_iop_mkdir
kernfs_iop_rmdir
kernfs_iop_rename
kernfs_leftmost_descendant
kernfs_next_descendant_post
kernfs_activate
__kernfs_remove
kernfs_remove
kernfs_break_active_protection
kernfs_unbreak_active_protection
kernfs_remove_self
kernfs_remove_by_name_ns
kernfs_rename_ns
dt_type
kernfs_dir_fop_release
kernfs_dir_pos
kernfs_dir_next_pos
kernfs_fop_readdir
   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 
  20 DEFINE_MUTEX(kernfs_mutex);
  21 static DEFINE_SPINLOCK(kernfs_rename_lock);     /* kn->parent and ->name */
  22 static char kernfs_pr_cont_buf[PATH_MAX];       /* protected by rename_lock */
  23 static DEFINE_SPINLOCK(kernfs_idr_lock);        /* root->ino_idr */
  24 
  25 #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb)
  26 
  27 static bool kernfs_active(struct kernfs_node *kn)
  28 {
  29         lockdep_assert_held(&kernfs_mutex);
  30         return atomic_read(&kn->active) >= 0;
  31 }
  32 
  33 static 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 
  42 static 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 */
  51 static 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 
  62 static 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  */
 122 static 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  */
 183 int 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  */
 210 int 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 }
 221 EXPORT_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  */
 229 void 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  */
 247 void 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 
 268 out:
 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  */
 279 struct 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  */
 299 static 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 
 315 static 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 
 329 static 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  */
 348 static 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  */
 390 static 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  */
 413 struct 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  */
 433 void kernfs_put_active(struct kernfs_node *kn)
 434 {
 435         int v;
 436 
 437         if (unlikely(!kn))
 438                 return;
 439 
 440         if (kernfs_lockdep(kn))
 441                 rwsem_release(&kn->dep_map, 1, _RET_IP_);
 442         v = atomic_dec_return(&kn->active);
 443         if (likely(v != KN_DEACTIVATED_BIAS))
 444                 return;
 445 
 446         wake_up_all(&kernfs_root(kn)->deactivate_waitq);
 447 }
 448 
 449 /**
 450  * kernfs_drain - drain kernfs_node
 451  * @kn: kernfs_node to drain
 452  *
 453  * Drain existing usages and nuke all existing mmaps of @kn.  Mutiple
 454  * removers may invoke this function concurrently on @kn and all will
 455  * return after draining is complete.
 456  */
 457 static void kernfs_drain(struct kernfs_node *kn)
 458         __releases(&kernfs_mutex) __acquires(&kernfs_mutex)
 459 {
 460         struct kernfs_root *root = kernfs_root(kn);
 461 
 462         lockdep_assert_held(&kernfs_mutex);
 463         WARN_ON_ONCE(kernfs_active(kn));
 464 
 465         mutex_unlock(&kernfs_mutex);
 466 
 467         if (kernfs_lockdep(kn)) {
 468                 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
 469                 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
 470                         lock_contended(&kn->dep_map, _RET_IP_);
 471         }
 472 
 473         /* but everyone should wait for draining */
 474         wait_event(root->deactivate_waitq,
 475                    atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
 476 
 477         if (kernfs_lockdep(kn)) {
 478                 lock_acquired(&kn->dep_map, _RET_IP_);
 479                 rwsem_release(&kn->dep_map, 1, _RET_IP_);
 480         }
 481 
 482         kernfs_drain_open_files(kn);
 483 
 484         mutex_lock(&kernfs_mutex);
 485 }
 486 
 487 /**
 488  * kernfs_get - get a reference count on a kernfs_node
 489  * @kn: the target kernfs_node
 490  */
 491 void kernfs_get(struct kernfs_node *kn)
 492 {
 493         if (kn) {
 494                 WARN_ON(!atomic_read(&kn->count));
 495                 atomic_inc(&kn->count);
 496         }
 497 }
 498 EXPORT_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  */
 506 void kernfs_put(struct kernfs_node *kn)
 507 {
 508         struct kernfs_node *parent;
 509         struct kernfs_root *root;
 510 
 511         /*
 512          * kernfs_node is freed with ->count 0, kernfs_find_and_get_node_by_ino
 513          * depends on this to filter reused stale node
 514          */
 515         if (!kn || !atomic_dec_and_test(&kn->count))
 516                 return;
 517         root = kernfs_root(kn);
 518  repeat:
 519         /*
 520          * Moving/renaming is always done while holding reference.
 521          * kn->parent won't change beneath us.
 522          */
 523         parent = kn->parent;
 524 
 525         WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
 526                   "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
 527                   parent ? parent->name : "", kn->name, atomic_read(&kn->active));
 528 
 529         if (kernfs_type(kn) == KERNFS_LINK)
 530                 kernfs_put(kn->symlink.target_kn);
 531 
 532         kfree_const(kn->name);
 533 
 534         if (kn->iattr) {
 535                 simple_xattrs_free(&kn->iattr->xattrs);
 536                 kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
 537         }
 538         spin_lock(&kernfs_idr_lock);
 539         idr_remove(&root->ino_idr, kn->id.ino);
 540         spin_unlock(&kernfs_idr_lock);
 541         kmem_cache_free(kernfs_node_cache, kn);
 542 
 543         kn = parent;
 544         if (kn) {
 545                 if (atomic_dec_and_test(&kn->count))
 546                         goto repeat;
 547         } else {
 548                 /* just released the root kn, free @root too */
 549                 idr_destroy(&root->ino_idr);
 550                 kfree(root);
 551         }
 552 }
 553 EXPORT_SYMBOL_GPL(kernfs_put);
 554 
 555 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
 556 {
 557         struct kernfs_node *kn;
 558 
 559         if (flags & LOOKUP_RCU)
 560                 return -ECHILD;
 561 
 562         /* Always perform fresh lookup for negatives */
 563         if (d_really_is_negative(dentry))
 564                 goto out_bad_unlocked;
 565 
 566         kn = kernfs_dentry_node(dentry);
 567         mutex_lock(&kernfs_mutex);
 568 
 569         /* The kernfs node has been deactivated */
 570         if (!kernfs_active(kn))
 571                 goto out_bad;
 572 
 573         /* The kernfs node has been moved? */
 574         if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
 575                 goto out_bad;
 576 
 577         /* The kernfs node has been renamed */
 578         if (strcmp(dentry->d_name.name, kn->name) != 0)
 579                 goto out_bad;
 580 
 581         /* The kernfs node has been moved to a different namespace */
 582         if (kn->parent && kernfs_ns_enabled(kn->parent) &&
 583             kernfs_info(dentry->d_sb)->ns != kn->ns)
 584                 goto out_bad;
 585 
 586         mutex_unlock(&kernfs_mutex);
 587         return 1;
 588 out_bad:
 589         mutex_unlock(&kernfs_mutex);
 590 out_bad_unlocked:
 591         return 0;
 592 }
 593 
 594 const struct dentry_operations kernfs_dops = {
 595         .d_revalidate   = kernfs_dop_revalidate,
 596 };
 597 
 598 /**
 599  * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
 600  * @dentry: the dentry in question
 601  *
 602  * Return the kernfs_node associated with @dentry.  If @dentry is not a
 603  * kernfs one, %NULL is returned.
 604  *
 605  * While the returned kernfs_node will stay accessible as long as @dentry
 606  * is accessible, the returned node can be in any state and the caller is
 607  * fully responsible for determining what's accessible.
 608  */
 609 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
 610 {
 611         if (dentry->d_sb->s_op == &kernfs_sops &&
 612             !d_really_is_negative(dentry))
 613                 return kernfs_dentry_node(dentry);
 614         return NULL;
 615 }
 616 
 617 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
 618                                              struct kernfs_node *parent,
 619                                              const char *name, umode_t mode,
 620                                              kuid_t uid, kgid_t gid,
 621                                              unsigned flags)
 622 {
 623         struct kernfs_node *kn;
 624         u32 gen;
 625         int ret;
 626 
 627         name = kstrdup_const(name, GFP_KERNEL);
 628         if (!name)
 629                 return NULL;
 630 
 631         kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
 632         if (!kn)
 633                 goto err_out1;
 634 
 635         idr_preload(GFP_KERNEL);
 636         spin_lock(&kernfs_idr_lock);
 637         ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
 638         if (ret >= 0 && ret < root->last_ino)
 639                 root->next_generation++;
 640         gen = root->next_generation;
 641         root->last_ino = ret;
 642         spin_unlock(&kernfs_idr_lock);
 643         idr_preload_end();
 644         if (ret < 0)
 645                 goto err_out2;
 646         kn->id.ino = ret;
 647         kn->id.generation = gen;
 648 
 649         /*
 650          * set ino first. This RELEASE is paired with atomic_inc_not_zero in
 651          * kernfs_find_and_get_node_by_ino
 652          */
 653         atomic_set_release(&kn->count, 1);
 654         atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
 655         RB_CLEAR_NODE(&kn->rb);
 656 
 657         kn->name = name;
 658         kn->mode = mode;
 659         kn->flags = flags;
 660 
 661         if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
 662                 struct iattr iattr = {
 663                         .ia_valid = ATTR_UID | ATTR_GID,
 664                         .ia_uid = uid,
 665                         .ia_gid = gid,
 666                 };
 667 
 668                 ret = __kernfs_setattr(kn, &iattr);
 669                 if (ret < 0)
 670                         goto err_out3;
 671         }
 672 
 673         if (parent) {
 674                 ret = security_kernfs_init_security(parent, kn);
 675                 if (ret)
 676                         goto err_out3;
 677         }
 678 
 679         return kn;
 680 
 681  err_out3:
 682         idr_remove(&root->ino_idr, kn->id.ino);
 683  err_out2:
 684         kmem_cache_free(kernfs_node_cache, kn);
 685  err_out1:
 686         kfree_const(name);
 687         return NULL;
 688 }
 689 
 690 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
 691                                     const char *name, umode_t mode,
 692                                     kuid_t uid, kgid_t gid,
 693                                     unsigned flags)
 694 {
 695         struct kernfs_node *kn;
 696 
 697         kn = __kernfs_new_node(kernfs_root(parent), parent,
 698                                name, mode, uid, gid, flags);
 699         if (kn) {
 700                 kernfs_get(parent);
 701                 kn->parent = parent;
 702         }
 703         return kn;
 704 }
 705 
 706 /*
 707  * kernfs_find_and_get_node_by_ino - get kernfs_node from inode number
 708  * @root: the kernfs root
 709  * @ino: inode number
 710  *
 711  * RETURNS:
 712  * NULL on failure. Return a kernfs node with reference counter incremented
 713  */
 714 struct kernfs_node *kernfs_find_and_get_node_by_ino(struct kernfs_root *root,
 715                                                     unsigned int ino)
 716 {
 717         struct kernfs_node *kn;
 718 
 719         rcu_read_lock();
 720         kn = idr_find(&root->ino_idr, ino);
 721         if (!kn)
 722                 goto out;
 723 
 724         /*
 725          * Since kernfs_node is freed in RCU, it's possible an old node for ino
 726          * is freed, but reused before RCU grace period. But a freed node (see
 727          * kernfs_put) or an incompletedly initialized node (see
 728          * __kernfs_new_node) should have 'count' 0. We can use this fact to
 729          * filter out such node.
 730          */
 731         if (!atomic_inc_not_zero(&kn->count)) {
 732                 kn = NULL;
 733                 goto out;
 734         }
 735 
 736         /*
 737          * The node could be a new node or a reused node. If it's a new node,
 738          * we are ok. If it's reused because of RCU (because of
 739          * SLAB_TYPESAFE_BY_RCU), the __kernfs_new_node always sets its 'ino'
 740          * before 'count'. So if 'count' is uptodate, 'ino' should be uptodate,
 741          * hence we can use 'ino' to filter stale node.
 742          */
 743         if (kn->id.ino != ino)
 744                 goto out;
 745         rcu_read_unlock();
 746 
 747         return kn;
 748 out:
 749         rcu_read_unlock();
 750         kernfs_put(kn);
 751         return NULL;
 752 }
 753 
 754 /**
 755  *      kernfs_add_one - add kernfs_node to parent without warning
 756  *      @kn: kernfs_node to be added
 757  *
 758  *      The caller must already have initialized @kn->parent.  This
 759  *      function increments nlink of the parent's inode if @kn is a
 760  *      directory and link into the children list of the parent.
 761  *
 762  *      RETURNS:
 763  *      0 on success, -EEXIST if entry with the given name already
 764  *      exists.
 765  */
 766 int kernfs_add_one(struct kernfs_node *kn)
 767 {
 768         struct kernfs_node *parent = kn->parent;
 769         struct kernfs_iattrs *ps_iattr;
 770         bool has_ns;
 771         int ret;
 772 
 773         mutex_lock(&kernfs_mutex);
 774 
 775         ret = -EINVAL;
 776         has_ns = kernfs_ns_enabled(parent);
 777         if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
 778                  has_ns ? "required" : "invalid", parent->name, kn->name))
 779                 goto out_unlock;
 780 
 781         if (kernfs_type(parent) != KERNFS_DIR)
 782                 goto out_unlock;
 783 
 784         ret = -ENOENT;
 785         if (parent->flags & KERNFS_EMPTY_DIR)
 786                 goto out_unlock;
 787 
 788         if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
 789                 goto out_unlock;
 790 
 791         kn->hash = kernfs_name_hash(kn->name, kn->ns);
 792 
 793         ret = kernfs_link_sibling(kn);
 794         if (ret)
 795                 goto out_unlock;
 796 
 797         /* Update timestamps on the parent */
 798         ps_iattr = parent->iattr;
 799         if (ps_iattr) {
 800                 ktime_get_real_ts64(&ps_iattr->ia_ctime);
 801                 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
 802         }
 803 
 804         mutex_unlock(&kernfs_mutex);
 805 
 806         /*
 807          * Activate the new node unless CREATE_DEACTIVATED is requested.
 808          * If not activated here, the kernfs user is responsible for
 809          * activating the node with kernfs_activate().  A node which hasn't
 810          * been activated is not visible to userland and its removal won't
 811          * trigger deactivation.
 812          */
 813         if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
 814                 kernfs_activate(kn);
 815         return 0;
 816 
 817 out_unlock:
 818         mutex_unlock(&kernfs_mutex);
 819         return ret;
 820 }
 821 
 822 /**
 823  * kernfs_find_ns - find kernfs_node with the given name
 824  * @parent: kernfs_node to search under
 825  * @name: name to look for
 826  * @ns: the namespace tag to use
 827  *
 828  * Look for kernfs_node with name @name under @parent.  Returns pointer to
 829  * the found kernfs_node on success, %NULL on failure.
 830  */
 831 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
 832                                           const unsigned char *name,
 833                                           const void *ns)
 834 {
 835         struct rb_node *node = parent->dir.children.rb_node;
 836         bool has_ns = kernfs_ns_enabled(parent);
 837         unsigned int hash;
 838 
 839         lockdep_assert_held(&kernfs_mutex);
 840 
 841         if (has_ns != (bool)ns) {
 842                 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
 843                      has_ns ? "required" : "invalid", parent->name, name);
 844                 return NULL;
 845         }
 846 
 847         hash = kernfs_name_hash(name, ns);
 848         while (node) {
 849                 struct kernfs_node *kn;
 850                 int result;
 851 
 852                 kn = rb_to_kn(node);
 853                 result = kernfs_name_compare(hash, name, ns, kn);
 854                 if (result < 0)
 855                         node = node->rb_left;
 856                 else if (result > 0)
 857                         node = node->rb_right;
 858                 else
 859                         return kn;
 860         }
 861         return NULL;
 862 }
 863 
 864 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
 865                                           const unsigned char *path,
 866                                           const void *ns)
 867 {
 868         size_t len;
 869         char *p, *name;
 870 
 871         lockdep_assert_held(&kernfs_mutex);
 872 
 873         /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
 874         spin_lock_irq(&kernfs_rename_lock);
 875 
 876         len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
 877 
 878         if (len >= sizeof(kernfs_pr_cont_buf)) {
 879                 spin_unlock_irq(&kernfs_rename_lock);
 880                 return NULL;
 881         }
 882 
 883         p = kernfs_pr_cont_buf;
 884 
 885         while ((name = strsep(&p, "/")) && parent) {
 886                 if (*name == '\0')
 887                         continue;
 888                 parent = kernfs_find_ns(parent, name, ns);
 889         }
 890 
 891         spin_unlock_irq(&kernfs_rename_lock);
 892 
 893         return parent;
 894 }
 895 
 896 /**
 897  * kernfs_find_and_get_ns - find and get kernfs_node with the given name
 898  * @parent: kernfs_node to search under
 899  * @name: name to look for
 900  * @ns: the namespace tag to use
 901  *
 902  * Look for kernfs_node with name @name under @parent and get a reference
 903  * if found.  This function may sleep and returns pointer to the found
 904  * kernfs_node on success, %NULL on failure.
 905  */
 906 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
 907                                            const char *name, const void *ns)
 908 {
 909         struct kernfs_node *kn;
 910 
 911         mutex_lock(&kernfs_mutex);
 912         kn = kernfs_find_ns(parent, name, ns);
 913         kernfs_get(kn);
 914         mutex_unlock(&kernfs_mutex);
 915 
 916         return kn;
 917 }
 918 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
 919 
 920 /**
 921  * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
 922  * @parent: kernfs_node to search under
 923  * @path: path to look for
 924  * @ns: the namespace tag to use
 925  *
 926  * Look for kernfs_node with path @path under @parent and get a reference
 927  * if found.  This function may sleep and returns pointer to the found
 928  * kernfs_node on success, %NULL on failure.
 929  */
 930 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
 931                                            const char *path, const void *ns)
 932 {
 933         struct kernfs_node *kn;
 934 
 935         mutex_lock(&kernfs_mutex);
 936         kn = kernfs_walk_ns(parent, path, ns);
 937         kernfs_get(kn);
 938         mutex_unlock(&kernfs_mutex);
 939 
 940         return kn;
 941 }
 942 
 943 /**
 944  * kernfs_create_root - create a new kernfs hierarchy
 945  * @scops: optional syscall operations for the hierarchy
 946  * @flags: KERNFS_ROOT_* flags
 947  * @priv: opaque data associated with the new directory
 948  *
 949  * Returns the root of the new hierarchy on success, ERR_PTR() value on
 950  * failure.
 951  */
 952 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
 953                                        unsigned int flags, void *priv)
 954 {
 955         struct kernfs_root *root;
 956         struct kernfs_node *kn;
 957 
 958         root = kzalloc(sizeof(*root), GFP_KERNEL);
 959         if (!root)
 960                 return ERR_PTR(-ENOMEM);
 961 
 962         idr_init(&root->ino_idr);
 963         INIT_LIST_HEAD(&root->supers);
 964         root->next_generation = 1;
 965 
 966         kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
 967                                GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
 968                                KERNFS_DIR);
 969         if (!kn) {
 970                 idr_destroy(&root->ino_idr);
 971                 kfree(root);
 972                 return ERR_PTR(-ENOMEM);
 973         }
 974 
 975         kn->priv = priv;
 976         kn->dir.root = root;
 977 
 978         root->syscall_ops = scops;
 979         root->flags = flags;
 980         root->kn = kn;
 981         init_waitqueue_head(&root->deactivate_waitq);
 982 
 983         if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
 984                 kernfs_activate(kn);
 985 
 986         return root;
 987 }
 988 
 989 /**
 990  * kernfs_destroy_root - destroy a kernfs hierarchy
 991  * @root: root of the hierarchy to destroy
 992  *
 993  * Destroy the hierarchy anchored at @root by removing all existing
 994  * directories and destroying @root.
 995  */
 996 void kernfs_destroy_root(struct kernfs_root *root)
 997 {
 998         kernfs_remove(root->kn);        /* will also free @root */
 999 }
1000 
1001 /**
1002  * kernfs_create_dir_ns - create a directory
1003  * @parent: parent in which to create a new directory
1004  * @name: name of the new directory
1005  * @mode: mode of the new directory
1006  * @uid: uid of the new directory
1007  * @gid: gid of the new directory
1008  * @priv: opaque data associated with the new directory
1009  * @ns: optional namespace tag of the directory
1010  *
1011  * Returns the created node on success, ERR_PTR() value on failure.
1012  */
1013 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1014                                          const char *name, umode_t mode,
1015                                          kuid_t uid, kgid_t gid,
1016                                          void *priv, const void *ns)
1017 {
1018         struct kernfs_node *kn;
1019         int rc;
1020 
1021         /* allocate */
1022         kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1023                              uid, gid, KERNFS_DIR);
1024         if (!kn)
1025                 return ERR_PTR(-ENOMEM);
1026 
1027         kn->dir.root = parent->dir.root;
1028         kn->ns = ns;
1029         kn->priv = priv;
1030 
1031         /* link in */
1032         rc = kernfs_add_one(kn);
1033         if (!rc)
1034                 return kn;
1035 
1036         kernfs_put(kn);
1037         return ERR_PTR(rc);
1038 }
1039 
1040 /**
1041  * kernfs_create_empty_dir - create an always empty directory
1042  * @parent: parent in which to create a new directory
1043  * @name: name of the new directory
1044  *
1045  * Returns the created node on success, ERR_PTR() value on failure.
1046  */
1047 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1048                                             const char *name)
1049 {
1050         struct kernfs_node *kn;
1051         int rc;
1052 
1053         /* allocate */
1054         kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1055                              GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1056         if (!kn)
1057                 return ERR_PTR(-ENOMEM);
1058 
1059         kn->flags |= KERNFS_EMPTY_DIR;
1060         kn->dir.root = parent->dir.root;
1061         kn->ns = NULL;
1062         kn->priv = NULL;
1063 
1064         /* link in */
1065         rc = kernfs_add_one(kn);
1066         if (!rc)
1067                 return kn;
1068 
1069         kernfs_put(kn);
1070         return ERR_PTR(rc);
1071 }
1072 
1073 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1074                                         struct dentry *dentry,
1075                                         unsigned int flags)
1076 {
1077         struct dentry *ret;
1078         struct kernfs_node *parent = dir->i_private;
1079         struct kernfs_node *kn;
1080         struct inode *inode;
1081         const void *ns = NULL;
1082 
1083         mutex_lock(&kernfs_mutex);
1084 
1085         if (kernfs_ns_enabled(parent))
1086                 ns = kernfs_info(dir->i_sb)->ns;
1087 
1088         kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1089 
1090         /* no such entry */
1091         if (!kn || !kernfs_active(kn)) {
1092                 ret = NULL;
1093                 goto out_unlock;
1094         }
1095 
1096         /* attach dentry and inode */
1097         inode = kernfs_get_inode(dir->i_sb, kn);
1098         if (!inode) {
1099                 ret = ERR_PTR(-ENOMEM);
1100                 goto out_unlock;
1101         }
1102 
1103         /* instantiate and hash dentry */
1104         ret = d_splice_alias(inode, dentry);
1105  out_unlock:
1106         mutex_unlock(&kernfs_mutex);
1107         return ret;
1108 }
1109 
1110 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1111                             umode_t mode)
1112 {
1113         struct kernfs_node *parent = dir->i_private;
1114         struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1115         int ret;
1116 
1117         if (!scops || !scops->mkdir)
1118                 return -EPERM;
1119 
1120         if (!kernfs_get_active(parent))
1121                 return -ENODEV;
1122 
1123         ret = scops->mkdir(parent, dentry->d_name.name, mode);
1124 
1125         kernfs_put_active(parent);
1126         return ret;
1127 }
1128 
1129 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1130 {
1131         struct kernfs_node *kn  = kernfs_dentry_node(dentry);
1132         struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1133         int ret;
1134 
1135         if (!scops || !scops->rmdir)
1136                 return -EPERM;
1137 
1138         if (!kernfs_get_active(kn))
1139                 return -ENODEV;
1140 
1141         ret = scops->rmdir(kn);
1142 
1143         kernfs_put_active(kn);
1144         return ret;
1145 }
1146 
1147 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1148                              struct inode *new_dir, struct dentry *new_dentry,
1149                              unsigned int flags)
1150 {
1151         struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1152         struct kernfs_node *new_parent = new_dir->i_private;
1153         struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1154         int ret;
1155 
1156         if (flags)
1157                 return -EINVAL;
1158 
1159         if (!scops || !scops->rename)
1160                 return -EPERM;
1161 
1162         if (!kernfs_get_active(kn))
1163                 return -ENODEV;
1164 
1165         if (!kernfs_get_active(new_parent)) {
1166                 kernfs_put_active(kn);
1167                 return -ENODEV;
1168         }
1169 
1170         ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1171 
1172         kernfs_put_active(new_parent);
1173         kernfs_put_active(kn);
1174         return ret;
1175 }
1176 
1177 const struct inode_operations kernfs_dir_iops = {
1178         .lookup         = kernfs_iop_lookup,
1179         .permission     = kernfs_iop_permission,
1180         .setattr        = kernfs_iop_setattr,
1181         .getattr        = kernfs_iop_getattr,
1182         .listxattr      = kernfs_iop_listxattr,
1183 
1184         .mkdir          = kernfs_iop_mkdir,
1185         .rmdir          = kernfs_iop_rmdir,
1186         .rename         = kernfs_iop_rename,
1187 };
1188 
1189 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1190 {
1191         struct kernfs_node *last;
1192 
1193         while (true) {
1194                 struct rb_node *rbn;
1195 
1196                 last = pos;
1197 
1198                 if (kernfs_type(pos) != KERNFS_DIR)
1199                         break;
1200 
1201                 rbn = rb_first(&pos->dir.children);
1202                 if (!rbn)
1203                         break;
1204 
1205                 pos = rb_to_kn(rbn);
1206         }
1207 
1208         return last;
1209 }
1210 
1211 /**
1212  * kernfs_next_descendant_post - find the next descendant for post-order walk
1213  * @pos: the current position (%NULL to initiate traversal)
1214  * @root: kernfs_node whose descendants to walk
1215  *
1216  * Find the next descendant to visit for post-order traversal of @root's
1217  * descendants.  @root is included in the iteration and the last node to be
1218  * visited.
1219  */
1220 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1221                                                        struct kernfs_node *root)
1222 {
1223         struct rb_node *rbn;
1224 
1225         lockdep_assert_held(&kernfs_mutex);
1226 
1227         /* if first iteration, visit leftmost descendant which may be root */
1228         if (!pos)
1229                 return kernfs_leftmost_descendant(root);
1230 
1231         /* if we visited @root, we're done */
1232         if (pos == root)
1233                 return NULL;
1234 
1235         /* if there's an unvisited sibling, visit its leftmost descendant */
1236         rbn = rb_next(&pos->rb);
1237         if (rbn)
1238                 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1239 
1240         /* no sibling left, visit parent */
1241         return pos->parent;
1242 }
1243 
1244 /**
1245  * kernfs_activate - activate a node which started deactivated
1246  * @kn: kernfs_node whose subtree is to be activated
1247  *
1248  * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1249  * needs to be explicitly activated.  A node which hasn't been activated
1250  * isn't visible to userland and deactivation is skipped during its
1251  * removal.  This is useful to construct atomic init sequences where
1252  * creation of multiple nodes should either succeed or fail atomically.
1253  *
1254  * The caller is responsible for ensuring that this function is not called
1255  * after kernfs_remove*() is invoked on @kn.
1256  */
1257 void kernfs_activate(struct kernfs_node *kn)
1258 {
1259         struct kernfs_node *pos;
1260 
1261         mutex_lock(&kernfs_mutex);
1262 
1263         pos = NULL;
1264         while ((pos = kernfs_next_descendant_post(pos, kn))) {
1265                 if (!pos || (pos->flags & KERNFS_ACTIVATED))
1266                         continue;
1267 
1268                 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1269                 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1270 
1271                 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1272                 pos->flags |= KERNFS_ACTIVATED;
1273         }
1274 
1275         mutex_unlock(&kernfs_mutex);
1276 }
1277 
1278 static void __kernfs_remove(struct kernfs_node *kn)
1279 {
1280         struct kernfs_node *pos;
1281 
1282         lockdep_assert_held(&kernfs_mutex);
1283 
1284         /*
1285          * Short-circuit if non-root @kn has already finished removal.
1286          * This is for kernfs_remove_self() which plays with active ref
1287          * after removal.
1288          */
1289         if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1290                 return;
1291 
1292         pr_debug("kernfs %s: removing\n", kn->name);
1293 
1294         /* prevent any new usage under @kn by deactivating all nodes */
1295         pos = NULL;
1296         while ((pos = kernfs_next_descendant_post(pos, kn)))
1297                 if (kernfs_active(pos))
1298                         atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1299 
1300         /* deactivate and unlink the subtree node-by-node */
1301         do {
1302                 pos = kernfs_leftmost_descendant(kn);
1303 
1304                 /*
1305                  * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1306                  * base ref could have been put by someone else by the time
1307                  * the function returns.  Make sure it doesn't go away
1308                  * underneath us.
1309                  */
1310                 kernfs_get(pos);
1311 
1312                 /*
1313                  * Drain iff @kn was activated.  This avoids draining and
1314                  * its lockdep annotations for nodes which have never been
1315                  * activated and allows embedding kernfs_remove() in create
1316                  * error paths without worrying about draining.
1317                  */
1318                 if (kn->flags & KERNFS_ACTIVATED)
1319                         kernfs_drain(pos);
1320                 else
1321                         WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1322 
1323                 /*
1324                  * kernfs_unlink_sibling() succeeds once per node.  Use it
1325                  * to decide who's responsible for cleanups.
1326                  */
1327                 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1328                         struct kernfs_iattrs *ps_iattr =
1329                                 pos->parent ? pos->parent->iattr : NULL;
1330 
1331                         /* update timestamps on the parent */
1332                         if (ps_iattr) {
1333                                 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1334                                 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1335                         }
1336 
1337                         kernfs_put(pos);
1338                 }
1339 
1340                 kernfs_put(pos);
1341         } while (pos != kn);
1342 }
1343 
1344 /**
1345  * kernfs_remove - remove a kernfs_node recursively
1346  * @kn: the kernfs_node to remove
1347  *
1348  * Remove @kn along with all its subdirectories and files.
1349  */
1350 void kernfs_remove(struct kernfs_node *kn)
1351 {
1352         mutex_lock(&kernfs_mutex);
1353         __kernfs_remove(kn);
1354         mutex_unlock(&kernfs_mutex);
1355 }
1356 
1357 /**
1358  * kernfs_break_active_protection - break out of active protection
1359  * @kn: the self kernfs_node
1360  *
1361  * The caller must be running off of a kernfs operation which is invoked
1362  * with an active reference - e.g. one of kernfs_ops.  Each invocation of
1363  * this function must also be matched with an invocation of
1364  * kernfs_unbreak_active_protection().
1365  *
1366  * This function releases the active reference of @kn the caller is
1367  * holding.  Once this function is called, @kn may be removed at any point
1368  * and the caller is solely responsible for ensuring that the objects it
1369  * dereferences are accessible.
1370  */
1371 void kernfs_break_active_protection(struct kernfs_node *kn)
1372 {
1373         /*
1374          * Take out ourself out of the active ref dependency chain.  If
1375          * we're called without an active ref, lockdep will complain.
1376          */
1377         kernfs_put_active(kn);
1378 }
1379 
1380 /**
1381  * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1382  * @kn: the self kernfs_node
1383  *
1384  * If kernfs_break_active_protection() was called, this function must be
1385  * invoked before finishing the kernfs operation.  Note that while this
1386  * function restores the active reference, it doesn't and can't actually
1387  * restore the active protection - @kn may already or be in the process of
1388  * being removed.  Once kernfs_break_active_protection() is invoked, that
1389  * protection is irreversibly gone for the kernfs operation instance.
1390  *
1391  * While this function may be called at any point after
1392  * kernfs_break_active_protection() is invoked, its most useful location
1393  * would be right before the enclosing kernfs operation returns.
1394  */
1395 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1396 {
1397         /*
1398          * @kn->active could be in any state; however, the increment we do
1399          * here will be undone as soon as the enclosing kernfs operation
1400          * finishes and this temporary bump can't break anything.  If @kn
1401          * is alive, nothing changes.  If @kn is being deactivated, the
1402          * soon-to-follow put will either finish deactivation or restore
1403          * deactivated state.  If @kn is already removed, the temporary
1404          * bump is guaranteed to be gone before @kn is released.
1405          */
1406         atomic_inc(&kn->active);
1407         if (kernfs_lockdep(kn))
1408                 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1409 }
1410 
1411 /**
1412  * kernfs_remove_self - remove a kernfs_node from its own method
1413  * @kn: the self kernfs_node to remove
1414  *
1415  * The caller must be running off of a kernfs operation which is invoked
1416  * with an active reference - e.g. one of kernfs_ops.  This can be used to
1417  * implement a file operation which deletes itself.
1418  *
1419  * For example, the "delete" file for a sysfs device directory can be
1420  * implemented by invoking kernfs_remove_self() on the "delete" file
1421  * itself.  This function breaks the circular dependency of trying to
1422  * deactivate self while holding an active ref itself.  It isn't necessary
1423  * to modify the usual removal path to use kernfs_remove_self().  The
1424  * "delete" implementation can simply invoke kernfs_remove_self() on self
1425  * before proceeding with the usual removal path.  kernfs will ignore later
1426  * kernfs_remove() on self.
1427  *
1428  * kernfs_remove_self() can be called multiple times concurrently on the
1429  * same kernfs_node.  Only the first one actually performs removal and
1430  * returns %true.  All others will wait until the kernfs operation which
1431  * won self-removal finishes and return %false.  Note that the losers wait
1432  * for the completion of not only the winning kernfs_remove_self() but also
1433  * the whole kernfs_ops which won the arbitration.  This can be used to
1434  * guarantee, for example, all concurrent writes to a "delete" file to
1435  * finish only after the whole operation is complete.
1436  */
1437 bool kernfs_remove_self(struct kernfs_node *kn)
1438 {
1439         bool ret;
1440 
1441         mutex_lock(&kernfs_mutex);
1442         kernfs_break_active_protection(kn);
1443 
1444         /*
1445          * SUICIDAL is used to arbitrate among competing invocations.  Only
1446          * the first one will actually perform removal.  When the removal
1447          * is complete, SUICIDED is set and the active ref is restored
1448          * while holding kernfs_mutex.  The ones which lost arbitration
1449          * waits for SUICDED && drained which can happen only after the
1450          * enclosing kernfs operation which executed the winning instance
1451          * of kernfs_remove_self() finished.
1452          */
1453         if (!(kn->flags & KERNFS_SUICIDAL)) {
1454                 kn->flags |= KERNFS_SUICIDAL;
1455                 __kernfs_remove(kn);
1456                 kn->flags |= KERNFS_SUICIDED;
1457                 ret = true;
1458         } else {
1459                 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1460                 DEFINE_WAIT(wait);
1461 
1462                 while (true) {
1463                         prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1464 
1465                         if ((kn->flags & KERNFS_SUICIDED) &&
1466                             atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1467                                 break;
1468 
1469                         mutex_unlock(&kernfs_mutex);
1470                         schedule();
1471                         mutex_lock(&kernfs_mutex);
1472                 }
1473                 finish_wait(waitq, &wait);
1474                 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1475                 ret = false;
1476         }
1477 
1478         /*
1479          * This must be done while holding kernfs_mutex; otherwise, waiting
1480          * for SUICIDED && deactivated could finish prematurely.
1481          */
1482         kernfs_unbreak_active_protection(kn);
1483 
1484         mutex_unlock(&kernfs_mutex);
1485         return ret;
1486 }
1487 
1488 /**
1489  * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1490  * @parent: parent of the target
1491  * @name: name of the kernfs_node to remove
1492  * @ns: namespace tag of the kernfs_node to remove
1493  *
1494  * Look for the kernfs_node with @name and @ns under @parent and remove it.
1495  * Returns 0 on success, -ENOENT if such entry doesn't exist.
1496  */
1497 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1498                              const void *ns)
1499 {
1500         struct kernfs_node *kn;
1501 
1502         if (!parent) {
1503                 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1504                         name);
1505                 return -ENOENT;
1506         }
1507 
1508         mutex_lock(&kernfs_mutex);
1509 
1510         kn = kernfs_find_ns(parent, name, ns);
1511         if (kn)
1512                 __kernfs_remove(kn);
1513 
1514         mutex_unlock(&kernfs_mutex);
1515 
1516         if (kn)
1517                 return 0;
1518         else
1519                 return -ENOENT;
1520 }
1521 
1522 /**
1523  * kernfs_rename_ns - move and rename a kernfs_node
1524  * @kn: target node
1525  * @new_parent: new parent to put @sd under
1526  * @new_name: new name
1527  * @new_ns: new namespace tag
1528  */
1529 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1530                      const char *new_name, const void *new_ns)
1531 {
1532         struct kernfs_node *old_parent;
1533         const char *old_name = NULL;
1534         int error;
1535 
1536         /* can't move or rename root */
1537         if (!kn->parent)
1538                 return -EINVAL;
1539 
1540         mutex_lock(&kernfs_mutex);
1541 
1542         error = -ENOENT;
1543         if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1544             (new_parent->flags & KERNFS_EMPTY_DIR))
1545                 goto out;
1546 
1547         error = 0;
1548         if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1549             (strcmp(kn->name, new_name) == 0))
1550                 goto out;       /* nothing to rename */
1551 
1552         error = -EEXIST;
1553         if (kernfs_find_ns(new_parent, new_name, new_ns))
1554                 goto out;
1555 
1556         /* rename kernfs_node */
1557         if (strcmp(kn->name, new_name) != 0) {
1558                 error = -ENOMEM;
1559                 new_name = kstrdup_const(new_name, GFP_KERNEL);
1560                 if (!new_name)
1561                         goto out;
1562         } else {
1563                 new_name = NULL;
1564         }
1565 
1566         /*
1567          * Move to the appropriate place in the appropriate directories rbtree.
1568          */
1569         kernfs_unlink_sibling(kn);
1570         kernfs_get(new_parent);
1571 
1572         /* rename_lock protects ->parent and ->name accessors */
1573         spin_lock_irq(&kernfs_rename_lock);
1574 
1575         old_parent = kn->parent;
1576         kn->parent = new_parent;
1577 
1578         kn->ns = new_ns;
1579         if (new_name) {
1580                 old_name = kn->name;
1581                 kn->name = new_name;
1582         }
1583 
1584         spin_unlock_irq(&kernfs_rename_lock);
1585 
1586         kn->hash = kernfs_name_hash(kn->name, kn->ns);
1587         kernfs_link_sibling(kn);
1588 
1589         kernfs_put(old_parent);
1590         kfree_const(old_name);
1591 
1592         error = 0;
1593  out:
1594         mutex_unlock(&kernfs_mutex);
1595         return error;
1596 }
1597 
1598 /* Relationship between s_mode and the DT_xxx types */
1599 static inline unsigned char dt_type(struct kernfs_node *kn)
1600 {
1601         return (kn->mode >> 12) & 15;
1602 }
1603 
1604 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1605 {
1606         kernfs_put(filp->private_data);
1607         return 0;
1608 }
1609 
1610 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1611         struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1612 {
1613         if (pos) {
1614                 int valid = kernfs_active(pos) &&
1615                         pos->parent == parent && hash == pos->hash;
1616                 kernfs_put(pos);
1617                 if (!valid)
1618                         pos = NULL;
1619         }
1620         if (!pos && (hash > 1) && (hash < INT_MAX)) {
1621                 struct rb_node *node = parent->dir.children.rb_node;
1622                 while (node) {
1623                         pos = rb_to_kn(node);
1624 
1625                         if (hash < pos->hash)
1626                                 node = node->rb_left;
1627                         else if (hash > pos->hash)
1628                                 node = node->rb_right;
1629                         else
1630                                 break;
1631                 }
1632         }
1633         /* Skip over entries which are dying/dead or in the wrong namespace */
1634         while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1635                 struct rb_node *node = rb_next(&pos->rb);
1636                 if (!node)
1637                         pos = NULL;
1638                 else
1639                         pos = rb_to_kn(node);
1640         }
1641         return pos;
1642 }
1643 
1644 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1645         struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1646 {
1647         pos = kernfs_dir_pos(ns, parent, ino, pos);
1648         if (pos) {
1649                 do {
1650                         struct rb_node *node = rb_next(&pos->rb);
1651                         if (!node)
1652                                 pos = NULL;
1653                         else
1654                                 pos = rb_to_kn(node);
1655                 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1656         }
1657         return pos;
1658 }
1659 
1660 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1661 {
1662         struct dentry *dentry = file->f_path.dentry;
1663         struct kernfs_node *parent = kernfs_dentry_node(dentry);
1664         struct kernfs_node *pos = file->private_data;
1665         const void *ns = NULL;
1666 
1667         if (!dir_emit_dots(file, ctx))
1668                 return 0;
1669         mutex_lock(&kernfs_mutex);
1670 
1671         if (kernfs_ns_enabled(parent))
1672                 ns = kernfs_info(dentry->d_sb)->ns;
1673 
1674         for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1675              pos;
1676              pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1677                 const char *name = pos->name;
1678                 unsigned int type = dt_type(pos);
1679                 int len = strlen(name);
1680                 ino_t ino = pos->id.ino;
1681 
1682                 ctx->pos = pos->hash;
1683                 file->private_data = pos;
1684                 kernfs_get(pos);
1685 
1686                 mutex_unlock(&kernfs_mutex);
1687                 if (!dir_emit(ctx, name, len, ino, type))
1688                         return 0;
1689                 mutex_lock(&kernfs_mutex);
1690         }
1691         mutex_unlock(&kernfs_mutex);
1692         file->private_data = NULL;
1693         ctx->pos = INT_MAX;
1694         return 0;
1695 }
1696 
1697 const struct file_operations kernfs_dir_fops = {
1698         .read           = generic_read_dir,
1699         .iterate_shared = kernfs_fop_readdir,
1700         .release        = kernfs_dir_fop_release,
1701         .llseek         = generic_file_llseek,
1702 };
/* [<][>][^][v][top][bottom][index][help] */
root/fs/kernfs/dir.c

DEFINITIONS
