fs/sysfs/dir.c

   1 /*
   2  * fs/sysfs/dir.c - sysfs core and dir operation implementation
   3  *
   4  * Copyright (c) 2001-3 Patrick Mochel
   5  * Copyright (c) 2007 SUSE Linux Products GmbH
   6  * Copyright (c) 2007 Tejun Heo <teheo@suse.de>
   7  *
   8  * This file is released under the GPLv2.
   9  *
  10  * Please see Documentation/filesystems/sysfs.txt for more information.
  11  */
  12
  13 #undef DEBUG
  14
  15 #include <linux/fs.h>
  16 #include <linux/mount.h>
  17 #include <linux/module.h>
  18 #include <linux/kobject.h>
  19 #include <linux/namei.h>
  20 #include <linux/idr.h>
  21 #include <linux/completion.h>
  22 #include <linux/mutex.h>
  23 #include <linux/slab.h>
  24 #include "sysfs.h"
  25
  26 DEFINE_MUTEX(sysfs_mutex);
  27 DEFINE_MUTEX(sysfs_rename_mutex);
  28 DEFINE_SPINLOCK(sysfs_assoc_lock);
  29
  30 static DEFINE_SPINLOCK(sysfs_ino_lock);
  31 static DEFINE_IDA(sysfs_ino_ida);
  32
  33 /**
  34  *      sysfs_link_sibling - link sysfs_dirent into sibling list
  35  *      @sd: sysfs_dirent of interest
  36  *
  37  *      Link @sd into its sibling list which starts from
  38  *      sd->s_parent->s_dir.children.
  39  *
  40  *      Locking:
  41  *      mutex_lock(sysfs_mutex)
  42  */
  43 static void sysfs_link_sibling(struct sysfs_dirent *sd)
  44 {
  45         struct sysfs_dirent *parent_sd = sd->s_parent;
  46         struct sysfs_dirent **pos;
  47
  48         BUG_ON(sd->s_sibling);
  49
  50         /* Store directory entries in order by ino.  This allows
  51          * readdir to properly restart without having to add a
  52          * cursor into the s_dir.children list.
  53          */
  54         for (pos = &parent_sd->s_dir.children; *pos; pos = &(*pos)->s_sibling) {
  55                 if (sd->s_ino < (*pos)->s_ino)
  56                         break;
  57         }
  58         sd->s_sibling = *pos;
  59         *pos = sd;
  60 }
  61
  62 /**
  63  *      sysfs_unlink_sibling - unlink sysfs_dirent from sibling list
  64  *      @sd: sysfs_dirent of interest
  65  *
  66  *      Unlink @sd from its sibling list which starts from
  67  *      sd->s_parent->s_dir.children.
  68  *
  69  *      Locking:
  70  *      mutex_lock(sysfs_mutex)
  71  */
  72 static void sysfs_unlink_sibling(struct sysfs_dirent *sd)
  73 {
  74         struct sysfs_dirent **pos;
  75
  76         for (pos = &sd->s_parent->s_dir.children; *pos;
  77              pos = &(*pos)->s_sibling) {
  78                 if (*pos == sd) {
  79                         *pos = sd->s_sibling;
  80                         sd->s_sibling = NULL;
  81                         break;
  82                 }
  83         }
  84 }
  85
  86 /**
  87  *      sysfs_get_dentry - get dentry for the given sysfs_dirent
  88  *      @sd: sysfs_dirent of interest
  89  *
  90  *      Get dentry for @sd.  Dentry is looked up if currently not
  91  *      present.  This function descends from the root looking up
  92  *      dentry for each step.
  93  *
  94  *      LOCKING:
  95  *      mutex_lock(sysfs_rename_mutex)
  96  *
  97  *      RETURNS:
  98  *      Pointer to found dentry on success, ERR_PTR() value on error.
  99  */
 100 struct dentry *sysfs_get_dentry(struct sysfs_dirent *sd)
 101 {
 102         struct dentry *dentry = dget(sysfs_sb->s_root);
 103
 104         while (dentry->d_fsdata != sd) {
 105                 struct sysfs_dirent *cur;
 106                 struct dentry *parent;
 107
 108                 /* find the first ancestor which hasn't been looked up */
 109                 cur = sd;
 110                 while (cur->s_parent != dentry->d_fsdata)
 111                         cur = cur->s_parent;
 112
 113                 /* look it up */
 114                 parent = dentry;
 115                 mutex_lock(&parent->d_inode->i_mutex);
 116                 dentry = lookup_one_noperm(cur->s_name, parent);
 117                 mutex_unlock(&parent->d_inode->i_mutex);
 118                 dput(parent);
 119
 120                 if (IS_ERR(dentry))
 121                         break;
 122         }
 123         return dentry;
 124 }
 125
 126 /**
 127  *      sysfs_get_active - get an active reference to sysfs_dirent
 128  *      @sd: sysfs_dirent to get an active reference to
 129  *
 130  *      Get an active reference of @sd.  This function is noop if @sd
 131  *      is NULL.
 132  *
 133  *      RETURNS:
 134  *      Pointer to @sd on success, NULL on failure.
 135  */
 136 static struct sysfs_dirent *sysfs_get_active(struct sysfs_dirent *sd)
 137 {
 138         if (unlikely(!sd))
 139                 return NULL;
 140
 141         while (1) {
 142                 int v, t;
 143
 144                 v = atomic_read(&sd->s_active);
 145                 if (unlikely(v < 0))
 146                         return NULL;
 147
 148                 t = atomic_cmpxchg(&sd->s_active, v, v + 1);
 149                 if (likely(t == v))
 150                         return sd;
 151                 if (t < 0)
 152                         return NULL;
 153
 154                 cpu_relax();
 155         }
 156 }
 157
 158 /**
 159  *      sysfs_put_active - put an active reference to sysfs_dirent
 160  *      @sd: sysfs_dirent to put an active reference to
 161  *
 162  *      Put an active reference to @sd.  This function is noop if @sd
 163  *      is NULL.
 164  */
 165 static void sysfs_put_active(struct sysfs_dirent *sd)
 166 {
 167         struct completion *cmpl;
 168         int v;
 169
 170         if (unlikely(!sd))
 171                 return;
 172
 173         v = atomic_dec_return(&sd->s_active);
 174         if (likely(v != SD_DEACTIVATED_BIAS))
 175                 return;
 176
 177         /* atomic_dec_return() is a mb(), we'll always see the updated
 178          * sd->s_sibling.
 179          */
 180         cmpl = (void *)sd->s_sibling;
 181         complete(cmpl);
 182 }
 183
 184 /**
 185  *      sysfs_get_active_two - get active references to sysfs_dirent and parent
 186  *      @sd: sysfs_dirent of interest
 187  *
 188  *      Get active reference to @sd and its parent.  Parent's active
 189  *      reference is grabbed first.  This function is noop if @sd is
 190  *      NULL.
 191  *
 192  *      RETURNS:
 193  *      Pointer to @sd on success, NULL on failure.
 194  */
 195 struct sysfs_dirent *sysfs_get_active_two(struct sysfs_dirent *sd)
 196 {
 197         if (sd) {
 198                 if (sd->s_parent && unlikely(!sysfs_get_active(sd->s_parent)))
 199                         return NULL;
 200                 if (unlikely(!sysfs_get_active(sd))) {
 201                         sysfs_put_active(sd->s_parent);
 202                         return NULL;
 203                 }
 204         }
 205         return sd;
 206 }
 207
 208 /**
 209  *      sysfs_put_active_two - put active references to sysfs_dirent and parent
 210  *      @sd: sysfs_dirent of interest
 211  *
 212  *      Put active references to @sd and its parent.  This function is
 213  *      noop if @sd is NULL.
 214  */
 215 void sysfs_put_active_two(struct sysfs_dirent *sd)
 216 {
 217         if (sd) {
 218                 sysfs_put_active(sd);
 219                 sysfs_put_active(sd->s_parent);
 220         }
 221 }
 222
 223 /**
 224  *      sysfs_deactivate - deactivate sysfs_dirent
 225  *      @sd: sysfs_dirent to deactivate
 226  *
 227  *      Deny new active references and drain existing ones.
 228  */
 229 static void sysfs_deactivate(struct sysfs_dirent *sd)
 230 {
 231         DECLARE_COMPLETION_ONSTACK(wait);
 232         int v;
 233
 234         BUG_ON(sd->s_sibling || !(sd->s_flags & SYSFS_FLAG_REMOVED));
 235         sd->s_sibling = (void *)&wait;
 236
 237         /* atomic_add_return() is a mb(), put_active() will always see
 238          * the updated sd->s_sibling.
 239          */
 240         v = atomic_add_return(SD_DEACTIVATED_BIAS, &sd->s_active);
 241
 242         if (v != SD_DEACTIVATED_BIAS)
 243                 wait_for_completion(&wait);
 244
 245         sd->s_sibling = NULL;
 246 }
 247
 248 static int sysfs_alloc_ino(ino_t *pino)
 249 {
 250         int ino, rc;
 251
 252  retry:
 253         spin_lock(&sysfs_ino_lock);
 254         rc = ida_get_new_above(&sysfs_ino_ida, 2, &ino);
 255         spin_unlock(&sysfs_ino_lock);
 256
 257         if (rc == -EAGAIN) {
 258                 if (ida_pre_get(&sysfs_ino_ida, GFP_KERNEL))
 259                         goto retry;
 260                 rc = -ENOMEM;
 261         }
 262
 263         *pino = ino;
 264         return rc;
 265 }
 266
 267 static void sysfs_free_ino(ino_t ino)
 268 {
 269         spin_lock(&sysfs_ino_lock);
 270         ida_remove(&sysfs_ino_ida, ino);
 271         spin_unlock(&sysfs_ino_lock);
 272 }
 273
 274 void release_sysfs_dirent(struct sysfs_dirent * sd)
 275 {
 276         struct sysfs_dirent *parent_sd;
 277
 278  repeat:
 279         /* Moving/renaming is always done while holding reference.
 280          * sd->s_parent won't change beneath us.
 281          */
 282         parent_sd = sd->s_parent;
 283
 284         if (sysfs_type(sd) == SYSFS_KOBJ_LINK)
 285                 sysfs_put(sd->s_symlink.target_sd);
 286         if (sysfs_type(sd) & SYSFS_COPY_NAME)
 287                 kfree(sd->s_name);
 288         kfree(sd->s_iattr);
 289         sysfs_free_ino(sd->s_ino);
 290         kmem_cache_free(sysfs_dir_cachep, sd);
 291
 292         sd = parent_sd;
 293         if (sd && atomic_dec_and_test(&sd->s_count))
 294                 goto repeat;
 295 }
 296
 297 static void sysfs_d_iput(struct dentry * dentry, struct inode * inode)
 298 {
 299         struct sysfs_dirent * sd = dentry->d_fsdata;
 300
 301         sysfs_put(sd);
 302         iput(inode);
 303 }
 304
 305 static struct dentry_operations sysfs_dentry_ops = {
 306         .d_iput         = sysfs_d_iput,
 307 };
 308
 309 struct sysfs_dirent *sysfs_new_dirent(const char *name, umode_t mode, int type)
 310 {
 311         char *dup_name = NULL;
 312         struct sysfs_dirent *sd;
 313
 314         if (type & SYSFS_COPY_NAME) {
 315                 name = dup_name = kstrdup(name, GFP_KERNEL);
 316                 if (!name)
 317                         return NULL;
 318         }
 319
 320         sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL);
 321         if (!sd)
 322                 goto err_out1;
 323
 324         if (sysfs_alloc_ino(&sd->s_ino))
 325                 goto err_out2;
 326
 327         atomic_set(&sd->s_count, 1);
 328         atomic_set(&sd->s_active, 0);
 329
 330         sd->s_name = name;
 331         sd->s_mode = mode;
 332         sd->s_flags = type;
 333
 334         return sd;
 335
 336  err_out2:
 337         kmem_cache_free(sysfs_dir_cachep, sd);
 338  err_out1:
 339         kfree(dup_name);
 340         return NULL;
 341 }
 342
 343 static int sysfs_ilookup_test(struct inode *inode, void *arg)
 344 {
 345         struct sysfs_dirent *sd = arg;
 346         return inode->i_ino == sd->s_ino;
 347 }
 348
 349 /**
 350  *      sysfs_addrm_start - prepare for sysfs_dirent add/remove
 351  *      @acxt: pointer to sysfs_addrm_cxt to be used
 352  *      @parent_sd: parent sysfs_dirent
 353  *
 354  *      This function is called when the caller is about to add or
 355  *      remove sysfs_dirent under @parent_sd.  This function acquires
 356  *      sysfs_mutex, grabs inode for @parent_sd if available and lock
 357  *      i_mutex of it.  @acxt is used to keep and pass context to
 358  *      other addrm functions.
 359  *
 360  *      LOCKING:
 361  *      Kernel thread context (may sleep).  sysfs_mutex is locked on
 362  *      return.  i_mutex of parent inode is locked on return if
 363  *      available.
 364  */
 365 void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt,
 366                        struct sysfs_dirent *parent_sd)
 367 {
 368         struct inode *inode;
 369
 370         memset(acxt, 0, sizeof(*acxt));
 371         acxt->parent_sd = parent_sd;
 372
 373         /* Lookup parent inode.  inode initialization is protected by
 374          * sysfs_mutex, so inode existence can be determined by
 375          * looking up inode while holding sysfs_mutex.
 376          */
 377         mutex_lock(&sysfs_mutex);
 378
 379         inode = ilookup5(sysfs_sb, parent_sd->s_ino, sysfs_ilookup_test,
 380                          parent_sd);
 381         if (inode) {
 382                 WARN_ON(inode->i_state & I_NEW);
 383
 384                 /* parent inode available */
 385                 acxt->parent_inode = inode;
 386
 387                 /* sysfs_mutex is below i_mutex in lock hierarchy.
 388                  * First, trylock i_mutex.  If fails, unlock
 389                  * sysfs_mutex and lock them in order.
 390                  */
 391                 if (!mutex_trylock(&inode->i_mutex)) {
 392                         mutex_unlock(&sysfs_mutex);
 393                         mutex_lock(&inode->i_mutex);
 394                         mutex_lock(&sysfs_mutex);
 395                 }
 396         }
 397 }
 398
 399 /**
 400  *      __sysfs_add_one - add sysfs_dirent to parent without warning
 401  *      @acxt: addrm context to use
 402  *      @sd: sysfs_dirent to be added
 403  *
 404  *      Get @acxt->parent_sd and set sd->s_parent to it and increment
 405  *      nlink of parent inode if @sd is a directory and link into the
 406  *      children list of the parent.
 407  *
 408  *      This function should be called between calls to
 409  *      sysfs_addrm_start() and sysfs_addrm_finish() and should be
 410  *      passed the same @acxt as passed to sysfs_addrm_start().
 411  *
 412  *      LOCKING:
 413  *      Determined by sysfs_addrm_start().
 414  *
 415  *      RETURNS:
 416  *      0 on success, -EEXIST if entry with the given name already
 417  *      exists.
 418  */
 419 int __sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
 420 {
 421         if (sysfs_find_dirent(acxt->parent_sd, sd->s_name))
 422                 return -EEXIST;
 423
 424         sd->s_parent = sysfs_get(acxt->parent_sd);
 425
 426         if (sysfs_type(sd) == SYSFS_DIR && acxt->parent_inode)
 427                 inc_nlink(acxt->parent_inode);
 428
 429         acxt->cnt++;
 430
 431         sysfs_link_sibling(sd);
 432
 433         return 0;
 434 }
 435
 436 /**
 437  *      sysfs_add_one - add sysfs_dirent to parent
 438  *      @acxt: addrm context to use
 439  *      @sd: sysfs_dirent to be added
 440  *
 441  *      Get @acxt->parent_sd and set sd->s_parent to it and increment
 442  *      nlink of parent inode if @sd is a directory and link into the
 443  *      children list of the parent.
 444  *
 445  *      This function should be called between calls to
 446  *      sysfs_addrm_start() and sysfs_addrm_finish() and should be
 447  *      passed the same @acxt as passed to sysfs_addrm_start().
 448  *
 449  *      LOCKING:
 450  *      Determined by sysfs_addrm_start().
 451  *
 452  *      RETURNS:
 453  *      0 on success, -EEXIST if entry with the given name already
 454  *      exists.
 455  */
 456 int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
 457 {
 458         int ret;
 459
 460         ret = __sysfs_add_one(acxt, sd);
 461         WARN(ret == -EEXIST, KERN_WARNING "sysfs: duplicate filename '%s' "
 462                        "can not be created\n", sd->s_name);
 463         return ret;
 464 }
 465
 466 /**
 467  *      sysfs_remove_one - remove sysfs_dirent from parent
 468  *      @acxt: addrm context to use
 469  *      @sd: sysfs_dirent to be removed
 470  *
 471  *      Mark @sd removed and drop nlink of parent inode if @sd is a
 472  *      directory.  @sd is unlinked from the children list.
 473  *
 474  *      This function should be called between calls to
 475  *      sysfs_addrm_start() and sysfs_addrm_finish() and should be
 476  *      passed the same @acxt as passed to sysfs_addrm_start().
 477  *
 478  *      LOCKING:
 479  *      Determined by sysfs_addrm_start().
 480  */
 481 void sysfs_remove_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
 482 {
 483         BUG_ON(sd->s_flags & SYSFS_FLAG_REMOVED);
 484
 485         sysfs_unlink_sibling(sd);
 486
 487         sd->s_flags |= SYSFS_FLAG_REMOVED;
 488         sd->s_sibling = acxt->removed;
 489         acxt->removed = sd;
 490
 491         if (sysfs_type(sd) == SYSFS_DIR && acxt->parent_inode)
 492                 drop_nlink(acxt->parent_inode);
 493
 494         acxt->cnt++;
 495 }
 496
 497 /**
 498  *      sysfs_drop_dentry - drop dentry for the specified sysfs_dirent
 499  *      @sd: target sysfs_dirent
 500  *
 501  *      Drop dentry for @sd.  @sd must have been unlinked from its
 502  *      parent on entry to this function such that it can't be looked
 503  *      up anymore.
 504  */
 505 static void sysfs_drop_dentry(struct sysfs_dirent *sd)
 506 {
 507         struct inode *inode;
 508         struct dentry *dentry;
 509
 510         inode = ilookup(sysfs_sb, sd->s_ino);
 511         if (!inode)
 512                 return;
 513
 514         /* Drop any existing dentries associated with sd.
 515          *
 516          * For the dentry to be properly freed we need to grab a
 517          * reference to the dentry under the dcache lock,  unhash it,
 518          * and then put it.  The playing with the dentry count allows
 519          * dput to immediately free the dentry  if it is not in use.
 520          */
 521 repeat:
 522         spin_lock(&dcache_lock);
 523         list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
 524                 if (d_unhashed(dentry))
 525                         continue;
 526                 dget_locked(dentry);
 527                 spin_lock(&dentry->d_lock);
 528                 __d_drop(dentry);
 529                 spin_unlock(&dentry->d_lock);
 530                 spin_unlock(&dcache_lock);
 531                 dput(dentry);
 532                 goto repeat;
 533         }
 534         spin_unlock(&dcache_lock);
 535
 536         /* adjust nlink and update timestamp */
 537         mutex_lock(&inode->i_mutex);
 538
 539         inode->i_ctime = CURRENT_TIME;
 540         drop_nlink(inode);
 541         if (sysfs_type(sd) == SYSFS_DIR)
 542                 drop_nlink(inode);
 543
 544         mutex_unlock(&inode->i_mutex);
 545
 546         iput(inode);
 547 }
 548
 549 /**
 550  *      sysfs_addrm_finish - finish up sysfs_dirent add/remove
 551  *      @acxt: addrm context to finish up
 552  *
 553  *      Finish up sysfs_dirent add/remove.  Resources acquired by
 554  *      sysfs_addrm_start() are released and removed sysfs_dirents are
 555  *      cleaned up.  Timestamps on the parent inode are updated.
 556  *
 557  *      LOCKING:
 558  *      All mutexes acquired by sysfs_addrm_start() are released.
 559  */
 560 void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
 561 {
 562         /* release resources acquired by sysfs_addrm_start() */
 563         mutex_unlock(&sysfs_mutex);
 564         if (acxt->parent_inode) {
 565                 struct inode *inode = acxt->parent_inode;
 566
 567                 /* if added/removed, update timestamps on the parent */
 568                 if (acxt->cnt)
 569                         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 570
 571                 mutex_unlock(&inode->i_mutex);
 572                 iput(inode);
 573         }
 574
 575         /* kill removed sysfs_dirents */
 576         while (acxt->removed) {
 577                 struct sysfs_dirent *sd = acxt->removed;
 578
 579                 acxt->removed = sd->s_sibling;
 580                 sd->s_sibling = NULL;
 581
 582                 sysfs_drop_dentry(sd);
 583                 sysfs_deactivate(sd);
 584                 sysfs_put(sd);
 585         }
 586 }
 587
 588 /**
 589  *      sysfs_find_dirent - find sysfs_dirent with the given name
 590  *      @parent_sd: sysfs_dirent to search under
 591  *      @name: name to look for
 592  *
 593  *      Look for sysfs_dirent with name @name under @parent_sd.
 594  *
 595  *      LOCKING:
 596  *      mutex_lock(sysfs_mutex)
 597  *
 598  *      RETURNS:
 599  *      Pointer to sysfs_dirent if found, NULL if not.
 600  */
 601 struct sysfs_dirent *sysfs_find_dirent(struct sysfs_dirent *parent_sd,
 602                                        const unsigned char *name)
 603 {
 604         struct sysfs_dirent *sd;
 605
 606         for (sd = parent_sd->s_dir.children; sd; sd = sd->s_sibling)
 607                 if (!strcmp(sd->s_name, name))
 608                         return sd;
 609         return NULL;
 610 }
 611
 612 /**
 613  *      sysfs_get_dirent - find and get sysfs_dirent with the given name
 614  *      @parent_sd: sysfs_dirent to search under
 615  *      @name: name to look for
 616  *
 617  *      Look for sysfs_dirent with name @name under @parent_sd and get
 618  *      it if found.
 619  *
 620  *      LOCKING:
 621  *      Kernel thread context (may sleep).  Grabs sysfs_mutex.
 622  *
 623  *      RETURNS:
 624  *      Pointer to sysfs_dirent if found, NULL if not.
 625  */
 626 struct sysfs_dirent *sysfs_get_dirent(struct sysfs_dirent *parent_sd,
 627                                       const unsigned char *name)
 628 {
 629         struct sysfs_dirent *sd;
 630
 631         mutex_lock(&sysfs_mutex);
 632         sd = sysfs_find_dirent(parent_sd, name);
 633         sysfs_get(sd);
 634         mutex_unlock(&sysfs_mutex);
 635
 636         return sd;
 637 }
 638 EXPORT_SYMBOL_GPL(sysfs_get_dirent);
 639
 640 static int create_dir(struct kobject *kobj, struct sysfs_dirent *parent_sd,
 641                       const char *name, struct sysfs_dirent **p_sd)
 642 {
 643         umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
 644         struct sysfs_addrm_cxt acxt;
 645         struct sysfs_dirent *sd;
 646         int rc;
 647
 648         /* allocate */
 649         sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
 650         if (!sd)
 651                 return -ENOMEM;
 652         sd->s_dir.kobj = kobj;
 653
 654         /* link in */
 655         sysfs_addrm_start(&acxt, parent_sd);
 656         rc = sysfs_add_one(&acxt, sd);
 657         sysfs_addrm_finish(&acxt);
 658
 659         if (rc == 0)
 660                 *p_sd = sd;
 661         else
 662                 sysfs_put(sd);
 663
 664         return rc;
 665 }
 666
 667 int sysfs_create_subdir(struct kobject *kobj, const char *name,
 668                         struct sysfs_dirent **p_sd)
 669 {
 670         return create_dir(kobj, kobj->sd, name, p_sd);
 671 }
 672
 673 /**
 674  *      sysfs_create_dir - create a directory for an object.
 675  *      @kobj:          object we're creating directory for.
 676  */
 677 int sysfs_create_dir(struct kobject * kobj)
 678 {
 679         struct sysfs_dirent *parent_sd, *sd;
 680         int error = 0;
 681
 682         BUG_ON(!kobj);
 683
 684         if (kobj->parent)
 685                 parent_sd = kobj->parent->sd;
 686         else
 687                 parent_sd = &sysfs_root;
 688
 689         error = create_dir(kobj, parent_sd, kobject_name(kobj), &sd);
 690         if (!error)
 691                 kobj->sd = sd;
 692         return error;
 693 }
 694
 695 static struct dentry * sysfs_lookup(struct inode *dir, struct dentry *dentry,
 696                                 struct nameidata *nd)
 697 {
 698         struct dentry *ret = NULL;
 699         struct sysfs_dirent *parent_sd = dentry->d_parent->d_fsdata;
 700         struct sysfs_dirent *sd;
 701         struct inode *inode;
 702
 703         mutex_lock(&sysfs_mutex);
 704
 705         sd = sysfs_find_dirent(parent_sd, dentry->d_name.name);
 706
 707         /* no such entry */
 708         if (!sd) {
 709                 ret = ERR_PTR(-ENOENT);
 710                 goto out_unlock;
 711         }
 712
 713         /* attach dentry and inode */
 714         inode = sysfs_get_inode(sd);
 715         if (!inode) {
 716                 ret = ERR_PTR(-ENOMEM);
 717                 goto out_unlock;
 718         }
 719
 720         /* instantiate and hash dentry */
 721         dentry->d_op = &sysfs_dentry_ops;
 722         dentry->d_fsdata = sysfs_get(sd);
 723         d_instantiate(dentry, inode);
 724         d_rehash(dentry);
 725
 726  out_unlock:
 727         mutex_unlock(&sysfs_mutex);
 728         return ret;
 729 }
 730
 731 const struct inode_operations sysfs_dir_inode_operations = {
 732         .lookup         = sysfs_lookup,
 733         .setattr        = sysfs_setattr,
 734 };
 735
 736 static void remove_dir(struct sysfs_dirent *sd)
 737 {
 738         struct sysfs_addrm_cxt acxt;
 739
 740         sysfs_addrm_start(&acxt, sd->s_parent);
 741         sysfs_remove_one(&acxt, sd);
 742         sysfs_addrm_finish(&acxt);
 743 }
 744
 745 void sysfs_remove_subdir(struct sysfs_dirent *sd)
 746 {
 747         remove_dir(sd);
 748 }
 749
 750
 751 static void __sysfs_remove_dir(struct sysfs_dirent *dir_sd)
 752 {
 753         struct sysfs_addrm_cxt acxt;
 754         struct sysfs_dirent **pos;
 755
 756         if (!dir_sd)
 757                 return;
 758
 759         pr_debug("sysfs %s: removing dir\n", dir_sd->s_name);
 760         sysfs_addrm_start(&acxt, dir_sd);
 761         pos = &dir_sd->s_dir.children;
 762         while (*pos) {
 763                 struct sysfs_dirent *sd = *pos;
 764
 765                 if (sysfs_type(sd) != SYSFS_DIR)
 766                         sysfs_remove_one(&acxt, sd);
 767                 else
 768                         pos = &(*pos)->s_sibling;
 769         }
 770         sysfs_addrm_finish(&acxt);
 771
 772         remove_dir(dir_sd);
 773 }
 774
 775 /**
 776  *      sysfs_remove_dir - remove an object's directory.
 777  *      @kobj:  object.
 778  *
 779  *      The only thing special about this is that we remove any files in
 780  *      the directory before we remove the directory, and we've inlined
 781  *      what used to be sysfs_rmdir() below, instead of calling separately.
 782  */
 783
 784 void sysfs_remove_dir(struct kobject * kobj)
 785 {
 786         struct sysfs_dirent *sd = kobj->sd;
 787
 788         spin_lock(&sysfs_assoc_lock);
 789         kobj->sd = NULL;
 790         spin_unlock(&sysfs_assoc_lock);
 791
 792         __sysfs_remove_dir(sd);
 793 }
 794
 795 int sysfs_rename_dir(struct kobject * kobj, const char *new_name)
 796 {
 797         struct sysfs_dirent *sd = kobj->sd;
 798         struct dentry *parent = NULL;
 799         struct dentry *old_dentry = NULL, *new_dentry = NULL;
 800         const char *dup_name = NULL;
 801         int error;
 802
 803         mutex_lock(&sysfs_rename_mutex);
 804
 805         error = 0;
 806         if (strcmp(sd->s_name, new_name) == 0)
 807                 goto out;       /* nothing to rename */
 808
 809         /* get the original dentry */
 810         old_dentry = sysfs_get_dentry(sd);
 811         if (IS_ERR(old_dentry)) {
 812                 error = PTR_ERR(old_dentry);
 813                 old_dentry = NULL;
 814                 goto out;
 815         }
 816
 817         parent = old_dentry->d_parent;
 818
 819         /* lock parent and get dentry for new name */
 820         mutex_lock(&parent->d_inode->i_mutex);
 821         mutex_lock(&sysfs_mutex);
 822
 823         error = -EEXIST;
 824         if (sysfs_find_dirent(sd->s_parent, new_name))
 825                 goto out_unlock;
 826
 827         error = -ENOMEM;
 828         new_dentry = d_alloc_name(parent, new_name);
 829         if (!new_dentry)
 830                 goto out_unlock;
 831
 832         /* rename sysfs_dirent */
 833         error = -ENOMEM;
 834         new_name = dup_name = kstrdup(new_name, GFP_KERNEL);
 835         if (!new_name)
 836                 goto out_unlock;
 837
 838         dup_name = sd->s_name;
 839         sd->s_name = new_name;
 840
 841         /* rename */
 842         d_add(new_dentry, NULL);
 843         d_move(old_dentry, new_dentry);
 844
 845         error = 0;
 846  out_unlock:
 847         mutex_unlock(&sysfs_mutex);
 848         mutex_unlock(&parent->d_inode->i_mutex);
 849         kfree(dup_name);
 850         dput(old_dentry);
 851         dput(new_dentry);
 852  out:
 853         mutex_unlock(&sysfs_rename_mutex);
 854         return error;
 855 }
 856
 857 int sysfs_move_dir(struct kobject *kobj, struct kobject *new_parent_kobj)
 858 {
 859         struct sysfs_dirent *sd = kobj->sd;
 860         struct sysfs_dirent *new_parent_sd;
 861         struct dentry *old_parent, *new_parent = NULL;
 862         struct dentry *old_dentry = NULL, *new_dentry = NULL;
 863         int error;
 864
 865         mutex_lock(&sysfs_rename_mutex);
 866         BUG_ON(!sd->s_parent);
 867         new_parent_sd = new_parent_kobj->sd ? new_parent_kobj->sd : &sysfs_root;
 868
 869         error = 0;
 870         if (sd->s_parent == new_parent_sd)
 871                 goto out;       /* nothing to move */
 872
 873         /* get dentries */
 874         old_dentry = sysfs_get_dentry(sd);
 875         if (IS_ERR(old_dentry)) {
 876                 error = PTR_ERR(old_dentry);
 877                 old_dentry = NULL;
 878                 goto out;
 879         }
 880         old_parent = old_dentry->d_parent;
 881
 882         new_parent = sysfs_get_dentry(new_parent_sd);
 883         if (IS_ERR(new_parent)) {
 884                 error = PTR_ERR(new_parent);
 885                 new_parent = NULL;
 886                 goto out;
 887         }
 888
 889 again:
 890         mutex_lock(&old_parent->d_inode->i_mutex);
 891         if (!mutex_trylock(&new_parent->d_inode->i_mutex)) {
 892                 mutex_unlock(&old_parent->d_inode->i_mutex);
 893                 goto again;
 894         }
 895         mutex_lock(&sysfs_mutex);
 896
 897         error = -EEXIST;
 898         if (sysfs_find_dirent(new_parent_sd, sd->s_name))
 899                 goto out_unlock;
 900
 901         error = -ENOMEM;
 902         new_dentry = d_alloc_name(new_parent, sd->s_name);
 903         if (!new_dentry)
 904                 goto out_unlock;
 905
 906         error = 0;
 907         d_add(new_dentry, NULL);
 908         d_move(old_dentry, new_dentry);
 909
 910         /* Remove from old parent's list and insert into new parent's list. */
 911         sysfs_unlink_sibling(sd);
 912         sysfs_get(new_parent_sd);
 913         sysfs_put(sd->s_parent);
 914         sd->s_parent = new_parent_sd;
 915         sysfs_link_sibling(sd);
 916
 917  out_unlock:
 918         mutex_unlock(&sysfs_mutex);
 919         mutex_unlock(&new_parent->d_inode->i_mutex);
 920         mutex_unlock(&old_parent->d_inode->i_mutex);
 921  out:
 922         dput(new_parent);
 923         dput(old_dentry);
 924         dput(new_dentry);
 925         mutex_unlock(&sysfs_rename_mutex);
 926         return error;
 927 }
 928
 929 /* Relationship between s_mode and the DT_xxx types */
 930 static inline unsigned char dt_type(struct sysfs_dirent *sd)
 931 {
 932         return (sd->s_mode >> 12) & 15;
 933 }
 934
 935 static int sysfs_readdir(struct file * filp, void * dirent, filldir_t filldir)
 936 {
 937         struct dentry *dentry = filp->f_path.dentry;
 938         struct sysfs_dirent * parent_sd = dentry->d_fsdata;
 939         struct sysfs_dirent *pos;
 940         ino_t ino;
 941
 942         if (filp->f_pos == 0) {
 943                 ino = parent_sd->s_ino;
 944                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) == 0)
 945                         filp->f_pos++;
 946         }
 947         if (filp->f_pos == 1) {
 948                 if (parent_sd->s_parent)
 949                         ino = parent_sd->s_parent->s_ino;
 950                 else
 951                         ino = parent_sd->s_ino;
 952                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) == 0)
 953                         filp->f_pos++;
 954         }
 955         if ((filp->f_pos > 1) && (filp->f_pos < INT_MAX)) {
 956                 mutex_lock(&sysfs_mutex);
 957
 958                 /* Skip the dentries we have already reported */
 959                 pos = parent_sd->s_dir.children;
 960                 while (pos && (filp->f_pos > pos->s_ino))
 961                         pos = pos->s_sibling;
 962
 963                 for ( ; pos; pos = pos->s_sibling) {
 964                         const char * name;
 965                         int len;
 966
 967                         name = pos->s_name;
 968                         len = strlen(name);
 969                         filp->f_pos = ino = pos->s_ino;
 970
 971                         if (filldir(dirent, name, len, filp->f_pos, ino,
 972                                          dt_type(pos)) < 0)
 973                                 break;
 974                 }
 975                 if (!pos)
 976                         filp->f_pos = INT_MAX;
 977                 mutex_unlock(&sysfs_mutex);
 978         }
 979         return 0;
 980 }
 981
 982
 983 const struct file_operations sysfs_dir_operations = {
 984         .read           = generic_read_dir,
 985         .readdir        = sysfs_readdir,
 986         .llseek         = generic_file_llseek,
 987 };