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 and I_NEW
 374          * clearing are protected by sysfs_mutex.  By grabbing it and
 375          * looking up with _nowait variant, inode state can be
 376          * determined reliably.
 377          */
 378         mutex_lock(&sysfs_mutex);
 379
 380         inode = ilookup5_nowait(sysfs_sb, parent_sd->s_ino, sysfs_ilookup_test,
 381                                 parent_sd);
 382
 383         if (inode && !(inode->i_state & I_NEW)) {
 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         } else
 397                 iput(inode);
 398 }
 399
 400 /**
 401  *      sysfs_add_one - add sysfs_dirent to parent
 402  *      @acxt: addrm context to use
 403  *      @sd: sysfs_dirent to be added
 404  *
 405  *      Get @acxt->parent_sd and set sd->s_parent to it and increment
 406  *      nlink of parent inode if @sd is a directory and link into the
 407  *      children list of the parent.
 408  *
 409  *      This function should be called between calls to
 410  *      sysfs_addrm_start() and sysfs_addrm_finish() and should be
 411  *      passed the same @acxt as passed to sysfs_addrm_start().
 412  *
 413  *      LOCKING:
 414  *      Determined by sysfs_addrm_start().
 415  *
 416  *      RETURNS:
 417  *      0 on success, -EEXIST if entry with the given name already
 418  *      exists.
 419  */
 420 int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
 421 {
 422         if (sysfs_find_dirent(acxt->parent_sd, sd->s_name))
 423                 return -EEXIST;
 424
 425         sd->s_parent = sysfs_get(acxt->parent_sd);
 426
 427         if (sysfs_type(sd) == SYSFS_DIR && acxt->parent_inode)
 428                 inc_nlink(acxt->parent_inode);
 429
 430         acxt->cnt++;
 431
 432         sysfs_link_sibling(sd);
 433
 434         return 0;
 435 }
 436
 437 /**
 438  *      sysfs_remove_one - remove sysfs_dirent from parent
 439  *      @acxt: addrm context to use
 440  *      @sd: sysfs_dirent to be removed
 441  *
 442  *      Mark @sd removed and drop nlink of parent inode if @sd is a
 443  *      directory.  @sd is unlinked from the children list.
 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 void sysfs_remove_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
 453 {
 454         BUG_ON(sd->s_flags & SYSFS_FLAG_REMOVED);
 455
 456         sysfs_unlink_sibling(sd);
 457
 458         sd->s_flags |= SYSFS_FLAG_REMOVED;
 459         sd->s_sibling = acxt->removed;
 460         acxt->removed = sd;
 461
 462         if (sysfs_type(sd) == SYSFS_DIR && acxt->parent_inode)
 463                 drop_nlink(acxt->parent_inode);
 464
 465         acxt->cnt++;
 466 }
 467
 468 /**
 469  *      sysfs_drop_dentry - drop dentry for the specified sysfs_dirent
 470  *      @sd: target sysfs_dirent
 471  *
 472  *      Drop dentry for @sd.  @sd must have been unlinked from its
 473  *      parent on entry to this function such that it can't be looked
 474  *      up anymore.
 475  */
 476 static void sysfs_drop_dentry(struct sysfs_dirent *sd)
 477 {
 478         struct inode *inode;
 479         struct dentry *dentry;
 480
 481         inode = ilookup(sysfs_sb, sd->s_ino);
 482         if (!inode)
 483                 return;
 484
 485         /* Drop any existing dentries associated with sd.
 486          *
 487          * For the dentry to be properly freed we need to grab a
 488          * reference to the dentry under the dcache lock,  unhash it,
 489          * and then put it.  The playing with the dentry count allows
 490          * dput to immediately free the dentry  if it is not in use.
 491          */
 492 repeat:
 493         spin_lock(&dcache_lock);
 494         list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
 495                 if (d_unhashed(dentry))
 496                         continue;
 497                 dget_locked(dentry);
 498                 spin_lock(&dentry->d_lock);
 499                 __d_drop(dentry);
 500                 spin_unlock(&dentry->d_lock);
 501                 spin_unlock(&dcache_lock);
 502                 dput(dentry);
 503                 goto repeat;
 504         }
 505         spin_unlock(&dcache_lock);
 506
 507         /* adjust nlink and update timestamp */
 508         mutex_lock(&inode->i_mutex);
 509
 510         inode->i_ctime = CURRENT_TIME;
 511         drop_nlink(inode);
 512         if (sysfs_type(sd) == SYSFS_DIR)
 513                 drop_nlink(inode);
 514
 515         mutex_unlock(&inode->i_mutex);
 516
 517         iput(inode);
 518 }
 519
 520 /**
 521  *      sysfs_addrm_finish - finish up sysfs_dirent add/remove
 522  *      @acxt: addrm context to finish up
 523  *
 524  *      Finish up sysfs_dirent add/remove.  Resources acquired by
 525  *      sysfs_addrm_start() are released and removed sysfs_dirents are
 526  *      cleaned up.  Timestamps on the parent inode are updated.
 527  *
 528  *      LOCKING:
 529  *      All mutexes acquired by sysfs_addrm_start() are released.
 530  */
 531 void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
 532 {
 533         /* release resources acquired by sysfs_addrm_start() */
 534         mutex_unlock(&sysfs_mutex);
 535         if (acxt->parent_inode) {
 536                 struct inode *inode = acxt->parent_inode;
 537
 538                 /* if added/removed, update timestamps on the parent */
 539                 if (acxt->cnt)
 540                         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 541
 542                 mutex_unlock(&inode->i_mutex);
 543                 iput(inode);
 544         }
 545
 546         /* kill removed sysfs_dirents */
 547         while (acxt->removed) {
 548                 struct sysfs_dirent *sd = acxt->removed;
 549
 550                 acxt->removed = sd->s_sibling;
 551                 sd->s_sibling = NULL;
 552
 553                 sysfs_drop_dentry(sd);
 554                 sysfs_deactivate(sd);
 555                 sysfs_put(sd);
 556         }
 557 }
 558
 559 /**
 560  *      sysfs_find_dirent - find sysfs_dirent with the given name
 561  *      @parent_sd: sysfs_dirent to search under
 562  *      @name: name to look for
 563  *
 564  *      Look for sysfs_dirent with name @name under @parent_sd.
 565  *
 566  *      LOCKING:
 567  *      mutex_lock(sysfs_mutex)
 568  *
 569  *      RETURNS:
 570  *      Pointer to sysfs_dirent if found, NULL if not.
 571  */
 572 struct sysfs_dirent *sysfs_find_dirent(struct sysfs_dirent *parent_sd,
 573                                        const unsigned char *name)
 574 {
 575         struct sysfs_dirent *sd;
 576
 577         for (sd = parent_sd->s_dir.children; sd; sd = sd->s_sibling)
 578                 if (!strcmp(sd->s_name, name))
 579                         return sd;
 580         return NULL;
 581 }
 582
 583 /**
 584  *      sysfs_get_dirent - find and get sysfs_dirent with the given name
 585  *      @parent_sd: sysfs_dirent to search under
 586  *      @name: name to look for
 587  *
 588  *      Look for sysfs_dirent with name @name under @parent_sd and get
 589  *      it if found.
 590  *
 591  *      LOCKING:
 592  *      Kernel thread context (may sleep).  Grabs sysfs_mutex.
 593  *
 594  *      RETURNS:
 595  *      Pointer to sysfs_dirent if found, NULL if not.
 596  */
 597 struct sysfs_dirent *sysfs_get_dirent(struct sysfs_dirent *parent_sd,
 598                                       const unsigned char *name)
 599 {
 600         struct sysfs_dirent *sd;
 601
 602         mutex_lock(&sysfs_mutex);
 603         sd = sysfs_find_dirent(parent_sd, name);
 604         sysfs_get(sd);
 605         mutex_unlock(&sysfs_mutex);
 606
 607         return sd;
 608 }
 609
 610 static int create_dir(struct kobject *kobj, struct sysfs_dirent *parent_sd,
 611                       const char *name, struct sysfs_dirent **p_sd)
 612 {
 613         umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
 614         struct sysfs_addrm_cxt acxt;
 615         struct sysfs_dirent *sd;
 616         int rc;
 617
 618         /* allocate */
 619         sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
 620         if (!sd)
 621                 return -ENOMEM;
 622         sd->s_dir.kobj = kobj;
 623
 624         /* link in */
 625         sysfs_addrm_start(&acxt, parent_sd);
 626         rc = sysfs_add_one(&acxt, sd);
 627         sysfs_addrm_finish(&acxt);
 628
 629         if (rc == 0)
 630                 *p_sd = sd;
 631         else
 632                 sysfs_put(sd);
 633
 634         return rc;
 635 }
 636
 637 int sysfs_create_subdir(struct kobject *kobj, const char *name,
 638                         struct sysfs_dirent **p_sd)
 639 {
 640         return create_dir(kobj, kobj->sd, name, p_sd);
 641 }
 642
 643 /**
 644  *      sysfs_create_dir - create a directory for an object.
 645  *      @kobj:          object we're creating directory for.
 646  */
 647 int sysfs_create_dir(struct kobject * kobj)
 648 {
 649         struct sysfs_dirent *parent_sd, *sd;
 650         int error = 0;
 651
 652         BUG_ON(!kobj);
 653
 654         if (kobj->parent)
 655                 parent_sd = kobj->parent->sd;
 656         else
 657                 parent_sd = &sysfs_root;
 658
 659         error = create_dir(kobj, parent_sd, kobject_name(kobj), &sd);
 660         if (!error)
 661                 kobj->sd = sd;
 662         return error;
 663 }
 664
 665 static struct dentry * sysfs_lookup(struct inode *dir, struct dentry *dentry,
 666                                 struct nameidata *nd)
 667 {
 668         struct dentry *ret = NULL;
 669         struct sysfs_dirent *parent_sd = dentry->d_parent->d_fsdata;
 670         struct sysfs_dirent *sd;
 671         struct inode *inode;
 672
 673         mutex_lock(&sysfs_mutex);
 674
 675         sd = sysfs_find_dirent(parent_sd, dentry->d_name.name);
 676
 677         /* no such entry */
 678         if (!sd) {
 679                 ret = ERR_PTR(-ENOENT);
 680                 goto out_unlock;
 681         }
 682
 683         /* attach dentry and inode */
 684         inode = sysfs_get_inode(sd);
 685         if (!inode) {
 686                 ret = ERR_PTR(-ENOMEM);
 687                 goto out_unlock;
 688         }
 689
 690         /* instantiate and hash dentry */
 691         dentry->d_op = &sysfs_dentry_ops;
 692         dentry->d_fsdata = sysfs_get(sd);
 693         d_instantiate(dentry, inode);
 694         d_rehash(dentry);
 695
 696  out_unlock:
 697         mutex_unlock(&sysfs_mutex);
 698         return ret;
 699 }
 700
 701 const struct inode_operations sysfs_dir_inode_operations = {
 702         .lookup         = sysfs_lookup,
 703         .setattr        = sysfs_setattr,
 704 };
 705
 706 static void remove_dir(struct sysfs_dirent *sd)
 707 {
 708         struct sysfs_addrm_cxt acxt;
 709
 710         sysfs_addrm_start(&acxt, sd->s_parent);
 711         sysfs_remove_one(&acxt, sd);
 712         sysfs_addrm_finish(&acxt);
 713 }
 714
 715 void sysfs_remove_subdir(struct sysfs_dirent *sd)
 716 {
 717         remove_dir(sd);
 718 }
 719
 720
 721 static void __sysfs_remove_dir(struct sysfs_dirent *dir_sd)
 722 {
 723         struct sysfs_addrm_cxt acxt;
 724         struct sysfs_dirent **pos;
 725
 726         if (!dir_sd)
 727                 return;
 728
 729         pr_debug("sysfs %s: removing dir\n", dir_sd->s_name);
 730         sysfs_addrm_start(&acxt, dir_sd);
 731         pos = &dir_sd->s_dir.children;
 732         while (*pos) {
 733                 struct sysfs_dirent *sd = *pos;
 734
 735                 if (sysfs_type(sd) != SYSFS_DIR)
 736                         sysfs_remove_one(&acxt, sd);
 737                 else
 738                         pos = &(*pos)->s_sibling;
 739         }
 740         sysfs_addrm_finish(&acxt);
 741
 742         remove_dir(dir_sd);
 743 }
 744
 745 /**
 746  *      sysfs_remove_dir - remove an object's directory.
 747  *      @kobj:  object.
 748  *
 749  *      The only thing special about this is that we remove any files in
 750  *      the directory before we remove the directory, and we've inlined
 751  *      what used to be sysfs_rmdir() below, instead of calling separately.
 752  */
 753
 754 void sysfs_remove_dir(struct kobject * kobj)
 755 {
 756         struct sysfs_dirent *sd = kobj->sd;
 757
 758         spin_lock(&sysfs_assoc_lock);
 759         kobj->sd = NULL;
 760         spin_unlock(&sysfs_assoc_lock);
 761
 762         __sysfs_remove_dir(sd);
 763 }
 764
 765 int sysfs_rename_dir(struct kobject * kobj, const char *new_name)
 766 {
 767         struct sysfs_dirent *sd = kobj->sd;
 768         struct dentry *parent = NULL;
 769         struct dentry *old_dentry = NULL, *new_dentry = NULL;
 770         const char *dup_name = NULL;
 771         int error;
 772
 773         mutex_lock(&sysfs_rename_mutex);
 774
 775         error = 0;
 776         if (strcmp(sd->s_name, new_name) == 0)
 777                 goto out;       /* nothing to rename */
 778
 779         /* get the original dentry */
 780         old_dentry = sysfs_get_dentry(sd);
 781         if (IS_ERR(old_dentry)) {
 782                 error = PTR_ERR(old_dentry);
 783                 old_dentry = NULL;
 784                 goto out;
 785         }
 786
 787         parent = old_dentry->d_parent;
 788
 789         /* lock parent and get dentry for new name */
 790         mutex_lock(&parent->d_inode->i_mutex);
 791         mutex_lock(&sysfs_mutex);
 792
 793         error = -EEXIST;
 794         if (sysfs_find_dirent(sd->s_parent, new_name))
 795                 goto out_unlock;
 796
 797         error = -ENOMEM;
 798         new_dentry = d_alloc_name(parent, new_name);
 799         if (!new_dentry)
 800                 goto out_unlock;
 801
 802         /* rename kobject and sysfs_dirent */
 803         error = -ENOMEM;
 804         new_name = dup_name = kstrdup(new_name, GFP_KERNEL);
 805         if (!new_name)
 806                 goto out_unlock;
 807
 808         error = kobject_set_name(kobj, "%s", new_name);
 809         if (error)
 810                 goto out_unlock;
 811
 812         dup_name = sd->s_name;
 813         sd->s_name = new_name;
 814
 815         /* rename */
 816         d_add(new_dentry, NULL);
 817         d_move(old_dentry, new_dentry);
 818
 819         error = 0;
 820  out_unlock:
 821         mutex_unlock(&sysfs_mutex);
 822         mutex_unlock(&parent->d_inode->i_mutex);
 823         kfree(dup_name);
 824         dput(old_dentry);
 825         dput(new_dentry);
 826  out:
 827         mutex_unlock(&sysfs_rename_mutex);
 828         return error;
 829 }
 830
 831 int sysfs_move_dir(struct kobject *kobj, struct kobject *new_parent_kobj)
 832 {
 833         struct sysfs_dirent *sd = kobj->sd;
 834         struct sysfs_dirent *new_parent_sd;
 835         struct dentry *old_parent, *new_parent = NULL;
 836         struct dentry *old_dentry = NULL, *new_dentry = NULL;
 837         int error;
 838
 839         mutex_lock(&sysfs_rename_mutex);
 840         BUG_ON(!sd->s_parent);
 841         new_parent_sd = new_parent_kobj->sd ? new_parent_kobj->sd : &sysfs_root;
 842
 843         error = 0;
 844         if (sd->s_parent == new_parent_sd)
 845                 goto out;       /* nothing to move */
 846
 847         /* get dentries */
 848         old_dentry = sysfs_get_dentry(sd);
 849         if (IS_ERR(old_dentry)) {
 850                 error = PTR_ERR(old_dentry);
 851                 old_dentry = NULL;
 852                 goto out;
 853         }
 854         old_parent = old_dentry->d_parent;
 855
 856         new_parent = sysfs_get_dentry(new_parent_sd);
 857         if (IS_ERR(new_parent)) {
 858                 error = PTR_ERR(new_parent);
 859                 new_parent = NULL;
 860                 goto out;
 861         }
 862
 863 again:
 864         mutex_lock(&old_parent->d_inode->i_mutex);
 865         if (!mutex_trylock(&new_parent->d_inode->i_mutex)) {
 866                 mutex_unlock(&old_parent->d_inode->i_mutex);
 867                 goto again;
 868         }
 869         mutex_lock(&sysfs_mutex);
 870
 871         error = -EEXIST;
 872         if (sysfs_find_dirent(new_parent_sd, sd->s_name))
 873                 goto out_unlock;
 874
 875         error = -ENOMEM;
 876         new_dentry = d_alloc_name(new_parent, sd->s_name);
 877         if (!new_dentry)
 878                 goto out_unlock;
 879
 880         error = 0;
 881         d_add(new_dentry, NULL);
 882         d_move(old_dentry, new_dentry);
 883
 884         /* Remove from old parent's list and insert into new parent's list. */
 885         sysfs_unlink_sibling(sd);
 886         sysfs_get(new_parent_sd);
 887         sysfs_put(sd->s_parent);
 888         sd->s_parent = new_parent_sd;
 889         sysfs_link_sibling(sd);
 890
 891  out_unlock:
 892         mutex_unlock(&sysfs_mutex);
 893         mutex_unlock(&new_parent->d_inode->i_mutex);
 894         mutex_unlock(&old_parent->d_inode->i_mutex);
 895  out:
 896         dput(new_parent);
 897         dput(old_dentry);
 898         dput(new_dentry);
 899         mutex_unlock(&sysfs_rename_mutex);
 900         return error;
 901 }
 902
 903 /* Relationship between s_mode and the DT_xxx types */
 904 static inline unsigned char dt_type(struct sysfs_dirent *sd)
 905 {
 906         return (sd->s_mode >> 12) & 15;
 907 }
 908
 909 static int sysfs_readdir(struct file * filp, void * dirent, filldir_t filldir)
 910 {
 911         struct dentry *dentry = filp->f_path.dentry;
 912         struct sysfs_dirent * parent_sd = dentry->d_fsdata;
 913         struct sysfs_dirent *pos;
 914         ino_t ino;
 915
 916         if (filp->f_pos == 0) {
 917                 ino = parent_sd->s_ino;
 918                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) == 0)
 919                         filp->f_pos++;
 920         }
 921         if (filp->f_pos == 1) {
 922                 if (parent_sd->s_parent)
 923                         ino = parent_sd->s_parent->s_ino;
 924                 else
 925                         ino = parent_sd->s_ino;
 926                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) == 0)
 927                         filp->f_pos++;
 928         }
 929         if ((filp->f_pos > 1) && (filp->f_pos < INT_MAX)) {
 930                 mutex_lock(&sysfs_mutex);
 931
 932                 /* Skip the dentries we have already reported */
 933                 pos = parent_sd->s_dir.children;
 934                 while (pos && (filp->f_pos > pos->s_ino))
 935                         pos = pos->s_sibling;
 936
 937                 for ( ; pos; pos = pos->s_sibling) {
 938                         const char * name;
 939                         int len;
 940
 941                         name = pos->s_name;
 942                         len = strlen(name);
 943                         filp->f_pos = ino = pos->s_ino;
 944
 945                         if (filldir(dirent, name, len, filp->f_pos, ino,
 946                                          dt_type(pos)) < 0)
 947                                 break;
 948                 }
 949                 if (!pos)
 950                         filp->f_pos = INT_MAX;
 951                 mutex_unlock(&sysfs_mutex);
 952         }
 953         return 0;
 954 }
 955
 956
 957 const struct file_operations sysfs_dir_operations = {
 958         .read           = generic_read_dir,
 959         .readdir        = sysfs_readdir,
 960 };