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 <linux/security.h>
  25 #include "sysfs.h"
  26
  27 DEFINE_MUTEX(sysfs_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_active - get an active reference to sysfs_dirent
  88  *      @sd: sysfs_dirent to get an active reference to
  89  *
  90  *      Get an active reference of @sd.  This function is noop if @sd
  91  *      is NULL.
  92  *
  93  *      RETURNS:
  94  *      Pointer to @sd on success, NULL on failure.
  95  */
  96 struct sysfs_dirent *sysfs_get_active(struct sysfs_dirent *sd)
  97 {
  98         if (unlikely(!sd))
  99                 return NULL;
 100
 101         while (1) {
 102                 int v, t;
 103
 104                 v = atomic_read(&sd->s_active);
 105                 if (unlikely(v < 0))
 106                         return NULL;
 107
 108                 t = atomic_cmpxchg(&sd->s_active, v, v + 1);
 109                 if (likely(t == v)) {
 110                         rwsem_acquire_read(&sd->dep_map, 0, 1, _RET_IP_);
 111                         return sd;
 112                 }
 113                 if (t < 0)
 114                         return NULL;
 115
 116                 cpu_relax();
 117         }
 118 }
 119
 120 /**
 121  *      sysfs_put_active - put an active reference to sysfs_dirent
 122  *      @sd: sysfs_dirent to put an active reference to
 123  *
 124  *      Put an active reference to @sd.  This function is noop if @sd
 125  *      is NULL.
 126  */
 127 void sysfs_put_active(struct sysfs_dirent *sd)
 128 {
 129         struct completion *cmpl;
 130         int v;
 131
 132         if (unlikely(!sd))
 133                 return;
 134
 135         rwsem_release(&sd->dep_map, 1, _RET_IP_);
 136         v = atomic_dec_return(&sd->s_active);
 137         if (likely(v != SD_DEACTIVATED_BIAS))
 138                 return;
 139
 140         /* atomic_dec_return() is a mb(), we'll always see the updated
 141          * sd->s_sibling.
 142          */
 143         cmpl = (void *)sd->s_sibling;
 144         complete(cmpl);
 145 }
 146
 147 /**
 148  *      sysfs_deactivate - deactivate sysfs_dirent
 149  *      @sd: sysfs_dirent to deactivate
 150  *
 151  *      Deny new active references and drain existing ones.
 152  */
 153 static void sysfs_deactivate(struct sysfs_dirent *sd)
 154 {
 155         DECLARE_COMPLETION_ONSTACK(wait);
 156         int v;
 157
 158         BUG_ON(sd->s_sibling || !(sd->s_flags & SYSFS_FLAG_REMOVED));
 159         sd->s_sibling = (void *)&wait;
 160
 161         rwsem_acquire(&sd->dep_map, 0, 0, _RET_IP_);
 162         /* atomic_add_return() is a mb(), put_active() will always see
 163          * the updated sd->s_sibling.
 164          */
 165         v = atomic_add_return(SD_DEACTIVATED_BIAS, &sd->s_active);
 166
 167         if (v != SD_DEACTIVATED_BIAS) {
 168                 lock_contended(&sd->dep_map, _RET_IP_);
 169                 wait_for_completion(&wait);
 170         }
 171
 172         sd->s_sibling = NULL;
 173
 174         lock_acquired(&sd->dep_map, _RET_IP_);
 175         rwsem_release(&sd->dep_map, 1, _RET_IP_);
 176 }
 177
 178 static int sysfs_alloc_ino(ino_t *pino)
 179 {
 180         int ino, rc;
 181
 182  retry:
 183         spin_lock(&sysfs_ino_lock);
 184         rc = ida_get_new_above(&sysfs_ino_ida, 2, &ino);
 185         spin_unlock(&sysfs_ino_lock);
 186
 187         if (rc == -EAGAIN) {
 188                 if (ida_pre_get(&sysfs_ino_ida, GFP_KERNEL))
 189                         goto retry;
 190                 rc = -ENOMEM;
 191         }
 192
 193         *pino = ino;
 194         return rc;
 195 }
 196
 197 static void sysfs_free_ino(ino_t ino)
 198 {
 199         spin_lock(&sysfs_ino_lock);
 200         ida_remove(&sysfs_ino_ida, ino);
 201         spin_unlock(&sysfs_ino_lock);
 202 }
 203
 204 void release_sysfs_dirent(struct sysfs_dirent * sd)
 205 {
 206         struct sysfs_dirent *parent_sd;
 207
 208  repeat:
 209         /* Moving/renaming is always done while holding reference.
 210          * sd->s_parent won't change beneath us.
 211          */
 212         parent_sd = sd->s_parent;
 213
 214         if (sysfs_type(sd) == SYSFS_KOBJ_LINK)
 215                 sysfs_put(sd->s_symlink.target_sd);
 216         if (sysfs_type(sd) & SYSFS_COPY_NAME)
 217                 kfree(sd->s_name);
 218         if (sd->s_iattr && sd->s_iattr->ia_secdata)
 219                 security_release_secctx(sd->s_iattr->ia_secdata,
 220                                         sd->s_iattr->ia_secdata_len);
 221         kfree(sd->s_iattr);
 222         sysfs_free_ino(sd->s_ino);
 223         kmem_cache_free(sysfs_dir_cachep, sd);
 224
 225         sd = parent_sd;
 226         if (sd && atomic_dec_and_test(&sd->s_count))
 227                 goto repeat;
 228 }
 229
 230 static int sysfs_dentry_delete(struct dentry *dentry)
 231 {
 232         struct sysfs_dirent *sd = dentry->d_fsdata;
 233         return !!(sd->s_flags & SYSFS_FLAG_REMOVED);
 234 }
 235
 236 static int sysfs_dentry_revalidate(struct dentry *dentry, struct nameidata *nd)
 237 {
 238         struct sysfs_dirent *sd = dentry->d_fsdata;
 239         int is_dir;
 240
 241         mutex_lock(&sysfs_mutex);
 242
 243         /* The sysfs dirent has been deleted */
 244         if (sd->s_flags & SYSFS_FLAG_REMOVED)
 245                 goto out_bad;
 246
 247         /* The sysfs dirent has been moved? */
 248         if (dentry->d_parent->d_fsdata != sd->s_parent)
 249                 goto out_bad;
 250
 251         /* The sysfs dirent has been renamed */
 252         if (strcmp(dentry->d_name.name, sd->s_name) != 0)
 253                 goto out_bad;
 254
 255         mutex_unlock(&sysfs_mutex);
 256 out_valid:
 257         return 1;
 258 out_bad:
 259         /* Remove the dentry from the dcache hashes.
 260          * If this is a deleted dentry we use d_drop instead of d_delete
 261          * so sysfs doesn't need to cope with negative dentries.
 262          *
 263          * If this is a dentry that has simply been renamed we
 264          * use d_drop to remove it from the dcache lookup on its
 265          * old parent.  If this dentry persists later when a lookup
 266          * is performed at its new name the dentry will be readded
 267          * to the dcache hashes.
 268          */
 269         is_dir = (sysfs_type(sd) == SYSFS_DIR);
 270         mutex_unlock(&sysfs_mutex);
 271         if (is_dir) {
 272                 /* If we have submounts we must allow the vfs caches
 273                  * to lie about the state of the filesystem to prevent
 274                  * leaks and other nasty things.
 275                  */
 276                 if (have_submounts(dentry))
 277                         goto out_valid;
 278                 shrink_dcache_parent(dentry);
 279         }
 280         d_drop(dentry);
 281         return 0;
 282 }
 283
 284 static void sysfs_dentry_iput(struct dentry *dentry, struct inode *inode)
 285 {
 286         struct sysfs_dirent * sd = dentry->d_fsdata;
 287
 288         sysfs_put(sd);
 289         iput(inode);
 290 }
 291
 292 static const struct dentry_operations sysfs_dentry_ops = {
 293         .d_revalidate   = sysfs_dentry_revalidate,
 294         .d_delete       = sysfs_dentry_delete,
 295         .d_iput         = sysfs_dentry_iput,
 296 };
 297
 298 struct sysfs_dirent *sysfs_new_dirent(const char *name, umode_t mode, int type)
 299 {
 300         char *dup_name = NULL;
 301         struct sysfs_dirent *sd;
 302
 303         if (type & SYSFS_COPY_NAME) {
 304                 name = dup_name = kstrdup(name, GFP_KERNEL);
 305                 if (!name)
 306                         return NULL;
 307         }
 308
 309         sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL);
 310         if (!sd)
 311                 goto err_out1;
 312
 313         if (sysfs_alloc_ino(&sd->s_ino))
 314                 goto err_out2;
 315
 316         atomic_set(&sd->s_count, 1);
 317         atomic_set(&sd->s_active, 0);
 318         sysfs_dirent_init_lockdep(sd);
 319
 320         sd->s_name = name;
 321         sd->s_mode = mode;
 322         sd->s_flags = type;
 323
 324         return sd;
 325
 326  err_out2:
 327         kmem_cache_free(sysfs_dir_cachep, sd);
 328  err_out1:
 329         kfree(dup_name);
 330         return NULL;
 331 }
 332
 333 /**
 334  *      sysfs_addrm_start - prepare for sysfs_dirent add/remove
 335  *      @acxt: pointer to sysfs_addrm_cxt to be used
 336  *      @parent_sd: parent sysfs_dirent
 337  *
 338  *      This function is called when the caller is about to add or
 339  *      remove sysfs_dirent under @parent_sd.  This function acquires
 340  *      sysfs_mutex.  @acxt is used to keep and pass context to
 341  *      other addrm functions.
 342  *
 343  *      LOCKING:
 344  *      Kernel thread context (may sleep).  sysfs_mutex is locked on
 345  *      return.
 346  */
 347 void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt,
 348                        struct sysfs_dirent *parent_sd)
 349 {
 350         memset(acxt, 0, sizeof(*acxt));
 351         acxt->parent_sd = parent_sd;
 352
 353         mutex_lock(&sysfs_mutex);
 354 }
 355
 356 /**
 357  *      __sysfs_add_one - add sysfs_dirent to parent without warning
 358  *      @acxt: addrm context to use
 359  *      @sd: sysfs_dirent to be added
 360  *
 361  *      Get @acxt->parent_sd and set sd->s_parent to it and increment
 362  *      nlink of parent inode if @sd is a directory and link into the
 363  *      children list of the parent.
 364  *
 365  *      This function should be called between calls to
 366  *      sysfs_addrm_start() and sysfs_addrm_finish() and should be
 367  *      passed the same @acxt as passed to sysfs_addrm_start().
 368  *
 369  *      LOCKING:
 370  *      Determined by sysfs_addrm_start().
 371  *
 372  *      RETURNS:
 373  *      0 on success, -EEXIST if entry with the given name already
 374  *      exists.
 375  */
 376 int __sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
 377 {
 378         struct sysfs_inode_attrs *ps_iattr;
 379
 380         if (sysfs_find_dirent(acxt->parent_sd, sd->s_name))
 381                 return -EEXIST;
 382
 383         sd->s_parent = sysfs_get(acxt->parent_sd);
 384
 385         sysfs_link_sibling(sd);
 386
 387         /* Update timestamps on the parent */
 388         ps_iattr = acxt->parent_sd->s_iattr;
 389         if (ps_iattr) {
 390                 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
 391                 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
 392         }
 393
 394         return 0;
 395 }
 396
 397 /**
 398  *      sysfs_pathname - return full path to sysfs dirent
 399  *      @sd: sysfs_dirent whose path we want
 400  *      @path: caller allocated buffer
 401  *
 402  *      Gives the name "/" to the sysfs_root entry; any path returned
 403  *      is relative to wherever sysfs is mounted.
 404  *
 405  *      XXX: does no error checking on @path size
 406  */
 407 static char *sysfs_pathname(struct sysfs_dirent *sd, char *path)
 408 {
 409         if (sd->s_parent) {
 410                 sysfs_pathname(sd->s_parent, path);
 411                 strcat(path, "/");
 412         }
 413         strcat(path, sd->s_name);
 414         return path;
 415 }
 416
 417 /**
 418  *      sysfs_add_one - add sysfs_dirent to parent
 419  *      @acxt: addrm context to use
 420  *      @sd: sysfs_dirent to be added
 421  *
 422  *      Get @acxt->parent_sd and set sd->s_parent to it and increment
 423  *      nlink of parent inode if @sd is a directory and link into the
 424  *      children list of the parent.
 425  *
 426  *      This function should be called between calls to
 427  *      sysfs_addrm_start() and sysfs_addrm_finish() and should be
 428  *      passed the same @acxt as passed to sysfs_addrm_start().
 429  *
 430  *      LOCKING:
 431  *      Determined by sysfs_addrm_start().
 432  *
 433  *      RETURNS:
 434  *      0 on success, -EEXIST if entry with the given name already
 435  *      exists.
 436  */
 437 int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
 438 {
 439         int ret;
 440
 441         ret = __sysfs_add_one(acxt, sd);
 442         if (ret == -EEXIST) {
 443                 char *path = kzalloc(PATH_MAX, GFP_KERNEL);
 444                 WARN(1, KERN_WARNING
 445                      "sysfs: cannot create duplicate filename '%s'\n",
 446                      (path == NULL) ? sd->s_name :
 447                      strcat(strcat(sysfs_pathname(acxt->parent_sd, path), "/"),
 448                             sd->s_name));
 449                 kfree(path);
 450         }
 451
 452         return ret;
 453 }
 454
 455 /**
 456  *      sysfs_remove_one - remove sysfs_dirent from parent
 457  *      @acxt: addrm context to use
 458  *      @sd: sysfs_dirent to be removed
 459  *
 460  *      Mark @sd removed and drop nlink of parent inode if @sd is a
 461  *      directory.  @sd is unlinked from the children list.
 462  *
 463  *      This function should be called between calls to
 464  *      sysfs_addrm_start() and sysfs_addrm_finish() and should be
 465  *      passed the same @acxt as passed to sysfs_addrm_start().
 466  *
 467  *      LOCKING:
 468  *      Determined by sysfs_addrm_start().
 469  */
 470 void sysfs_remove_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
 471 {
 472         struct sysfs_inode_attrs *ps_iattr;
 473
 474         BUG_ON(sd->s_flags & SYSFS_FLAG_REMOVED);
 475
 476         sysfs_unlink_sibling(sd);
 477
 478         /* Update timestamps on the parent */
 479         ps_iattr = acxt->parent_sd->s_iattr;
 480         if (ps_iattr) {
 481                 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
 482                 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
 483         }
 484
 485         sd->s_flags |= SYSFS_FLAG_REMOVED;
 486         sd->s_sibling = acxt->removed;
 487         acxt->removed = sd;
 488 }
 489
 490 /**
 491  *      sysfs_addrm_finish - finish up sysfs_dirent add/remove
 492  *      @acxt: addrm context to finish up
 493  *
 494  *      Finish up sysfs_dirent add/remove.  Resources acquired by
 495  *      sysfs_addrm_start() are released and removed sysfs_dirents are
 496  *      cleaned up.
 497  *
 498  *      LOCKING:
 499  *      sysfs_mutex is released.
 500  */
 501 void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
 502 {
 503         /* release resources acquired by sysfs_addrm_start() */
 504         mutex_unlock(&sysfs_mutex);
 505
 506         /* kill removed sysfs_dirents */
 507         while (acxt->removed) {
 508                 struct sysfs_dirent *sd = acxt->removed;
 509
 510                 acxt->removed = sd->s_sibling;
 511                 sd->s_sibling = NULL;
 512
 513                 sysfs_deactivate(sd);
 514                 unmap_bin_file(sd);
 515                 sysfs_put(sd);
 516         }
 517 }
 518
 519 /**
 520  *      sysfs_find_dirent - find sysfs_dirent with the given name
 521  *      @parent_sd: sysfs_dirent to search under
 522  *      @name: name to look for
 523  *
 524  *      Look for sysfs_dirent with name @name under @parent_sd.
 525  *
 526  *      LOCKING:
 527  *      mutex_lock(sysfs_mutex)
 528  *
 529  *      RETURNS:
 530  *      Pointer to sysfs_dirent if found, NULL if not.
 531  */
 532 struct sysfs_dirent *sysfs_find_dirent(struct sysfs_dirent *parent_sd,
 533                                        const unsigned char *name)
 534 {
 535         struct sysfs_dirent *sd;
 536
 537         for (sd = parent_sd->s_dir.children; sd; sd = sd->s_sibling)
 538                 if (!strcmp(sd->s_name, name))
 539                         return sd;
 540         return NULL;
 541 }
 542
 543 /**
 544  *      sysfs_get_dirent - find and get sysfs_dirent with the given name
 545  *      @parent_sd: sysfs_dirent to search under
 546  *      @name: name to look for
 547  *
 548  *      Look for sysfs_dirent with name @name under @parent_sd and get
 549  *      it if found.
 550  *
 551  *      LOCKING:
 552  *      Kernel thread context (may sleep).  Grabs sysfs_mutex.
 553  *
 554  *      RETURNS:
 555  *      Pointer to sysfs_dirent if found, NULL if not.
 556  */
 557 struct sysfs_dirent *sysfs_get_dirent(struct sysfs_dirent *parent_sd,
 558                                       const unsigned char *name)
 559 {
 560         struct sysfs_dirent *sd;
 561
 562         mutex_lock(&sysfs_mutex);
 563         sd = sysfs_find_dirent(parent_sd, name);
 564         sysfs_get(sd);
 565         mutex_unlock(&sysfs_mutex);
 566
 567         return sd;
 568 }
 569 EXPORT_SYMBOL_GPL(sysfs_get_dirent);
 570
 571 static int create_dir(struct kobject *kobj, struct sysfs_dirent *parent_sd,
 572                       const char *name, struct sysfs_dirent **p_sd)
 573 {
 574         umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
 575         struct sysfs_addrm_cxt acxt;
 576         struct sysfs_dirent *sd;
 577         int rc;
 578
 579         /* allocate */
 580         sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
 581         if (!sd)
 582                 return -ENOMEM;
 583         sd->s_dir.kobj = kobj;
 584
 585         /* link in */
 586         sysfs_addrm_start(&acxt, parent_sd);
 587         rc = sysfs_add_one(&acxt, sd);
 588         sysfs_addrm_finish(&acxt);
 589
 590         if (rc == 0)
 591                 *p_sd = sd;
 592         else
 593                 sysfs_put(sd);
 594
 595         return rc;
 596 }
 597
 598 int sysfs_create_subdir(struct kobject *kobj, const char *name,
 599                         struct sysfs_dirent **p_sd)
 600 {
 601         return create_dir(kobj, kobj->sd, name, p_sd);
 602 }
 603
 604 /**
 605  *      sysfs_create_dir - create a directory for an object.
 606  *      @kobj:          object we're creating directory for.
 607  */
 608 int sysfs_create_dir(struct kobject * kobj)
 609 {
 610         struct sysfs_dirent *parent_sd, *sd;
 611         int error = 0;
 612
 613         BUG_ON(!kobj);
 614
 615         if (kobj->parent)
 616                 parent_sd = kobj->parent->sd;
 617         else
 618                 parent_sd = &sysfs_root;
 619
 620         error = create_dir(kobj, parent_sd, kobject_name(kobj), &sd);
 621         if (!error)
 622                 kobj->sd = sd;
 623         return error;
 624 }
 625
 626 static struct dentry * sysfs_lookup(struct inode *dir, struct dentry *dentry,
 627                                 struct nameidata *nd)
 628 {
 629         struct dentry *ret = NULL;
 630         struct sysfs_dirent *parent_sd = dentry->d_parent->d_fsdata;
 631         struct sysfs_dirent *sd;
 632         struct inode *inode;
 633
 634         mutex_lock(&sysfs_mutex);
 635
 636         sd = sysfs_find_dirent(parent_sd, dentry->d_name.name);
 637
 638         /* no such entry */
 639         if (!sd) {
 640                 ret = ERR_PTR(-ENOENT);
 641                 goto out_unlock;
 642         }
 643
 644         /* attach dentry and inode */
 645         inode = sysfs_get_inode(sd);
 646         if (!inode) {
 647                 ret = ERR_PTR(-ENOMEM);
 648                 goto out_unlock;
 649         }
 650
 651         /* instantiate and hash dentry */
 652         ret = d_find_alias(inode);
 653         if (!ret) {
 654                 dentry->d_op = &sysfs_dentry_ops;
 655                 dentry->d_fsdata = sysfs_get(sd);
 656                 d_add(dentry, inode);
 657         } else {
 658                 d_move(ret, dentry);
 659                 iput(inode);
 660         }
 661
 662  out_unlock:
 663         mutex_unlock(&sysfs_mutex);
 664         return ret;
 665 }
 666
 667 const struct inode_operations sysfs_dir_inode_operations = {
 668         .lookup         = sysfs_lookup,
 669         .permission     = sysfs_permission,
 670         .setattr        = sysfs_setattr,
 671         .getattr        = sysfs_getattr,
 672         .setxattr       = sysfs_setxattr,
 673 };
 674
 675 static void remove_dir(struct sysfs_dirent *sd)
 676 {
 677         struct sysfs_addrm_cxt acxt;
 678
 679         sysfs_addrm_start(&acxt, sd->s_parent);
 680         sysfs_remove_one(&acxt, sd);
 681         sysfs_addrm_finish(&acxt);
 682 }
 683
 684 void sysfs_remove_subdir(struct sysfs_dirent *sd)
 685 {
 686         remove_dir(sd);
 687 }
 688
 689
 690 static void __sysfs_remove_dir(struct sysfs_dirent *dir_sd)
 691 {
 692         struct sysfs_addrm_cxt acxt;
 693         struct sysfs_dirent **pos;
 694
 695         if (!dir_sd)
 696                 return;
 697
 698         pr_debug("sysfs %s: removing dir\n", dir_sd->s_name);
 699         sysfs_addrm_start(&acxt, dir_sd);
 700         pos = &dir_sd->s_dir.children;
 701         while (*pos) {
 702                 struct sysfs_dirent *sd = *pos;
 703
 704                 if (sysfs_type(sd) != SYSFS_DIR)
 705                         sysfs_remove_one(&acxt, sd);
 706                 else
 707                         pos = &(*pos)->s_sibling;
 708         }
 709         sysfs_addrm_finish(&acxt);
 710
 711         remove_dir(dir_sd);
 712 }
 713
 714 /**
 715  *      sysfs_remove_dir - remove an object's directory.
 716  *      @kobj:  object.
 717  *
 718  *      The only thing special about this is that we remove any files in
 719  *      the directory before we remove the directory, and we've inlined
 720  *      what used to be sysfs_rmdir() below, instead of calling separately.
 721  */
 722
 723 void sysfs_remove_dir(struct kobject * kobj)
 724 {
 725         struct sysfs_dirent *sd = kobj->sd;
 726
 727         spin_lock(&sysfs_assoc_lock);
 728         kobj->sd = NULL;
 729         spin_unlock(&sysfs_assoc_lock);
 730
 731         __sysfs_remove_dir(sd);
 732 }
 733
 734 int sysfs_rename(struct sysfs_dirent *sd,
 735         struct sysfs_dirent *new_parent_sd, const char *new_name)
 736 {
 737         const char *dup_name = NULL;
 738         int error;
 739
 740         mutex_lock(&sysfs_mutex);
 741
 742         error = 0;
 743         if ((sd->s_parent == new_parent_sd) &&
 744             (strcmp(sd->s_name, new_name) == 0))
 745                 goto out;       /* nothing to rename */
 746
 747         error = -EEXIST;
 748         if (sysfs_find_dirent(new_parent_sd, new_name))
 749                 goto out;
 750
 751         /* rename sysfs_dirent */
 752         if (strcmp(sd->s_name, new_name) != 0) {
 753                 error = -ENOMEM;
 754                 new_name = dup_name = kstrdup(new_name, GFP_KERNEL);
 755                 if (!new_name)
 756                         goto out;
 757
 758                 dup_name = sd->s_name;
 759                 sd->s_name = new_name;
 760         }
 761
 762         /* Remove from old parent's list and insert into new parent's list. */
 763         if (sd->s_parent != new_parent_sd) {
 764                 sysfs_unlink_sibling(sd);
 765                 sysfs_get(new_parent_sd);
 766                 sysfs_put(sd->s_parent);
 767                 sd->s_parent = new_parent_sd;
 768                 sysfs_link_sibling(sd);
 769         }
 770
 771         error = 0;
 772  out:
 773         mutex_unlock(&sysfs_mutex);
 774         kfree(dup_name);
 775         return error;
 776 }
 777
 778 int sysfs_rename_dir(struct kobject *kobj, const char *new_name)
 779 {
 780         return sysfs_rename(kobj->sd, kobj->sd->s_parent, new_name);
 781 }
 782
 783 int sysfs_move_dir(struct kobject *kobj, struct kobject *new_parent_kobj)
 784 {
 785         struct sysfs_dirent *sd = kobj->sd;
 786         struct sysfs_dirent *new_parent_sd;
 787
 788         BUG_ON(!sd->s_parent);
 789         new_parent_sd = new_parent_kobj && new_parent_kobj->sd ?
 790                 new_parent_kobj->sd : &sysfs_root;
 791
 792         return sysfs_rename(sd, new_parent_sd, sd->s_name);
 793 }
 794
 795 /* Relationship between s_mode and the DT_xxx types */
 796 static inline unsigned char dt_type(struct sysfs_dirent *sd)
 797 {
 798         return (sd->s_mode >> 12) & 15;
 799 }
 800
 801 static int sysfs_dir_release(struct inode *inode, struct file *filp)
 802 {
 803         sysfs_put(filp->private_data);
 804         return 0;
 805 }
 806
 807 static struct sysfs_dirent *sysfs_dir_pos(struct sysfs_dirent *parent_sd,
 808         ino_t ino, struct sysfs_dirent *pos)
 809 {
 810         if (pos) {
 811                 int valid = !(pos->s_flags & SYSFS_FLAG_REMOVED) &&
 812                         pos->s_parent == parent_sd &&
 813                         ino == pos->s_ino;
 814                 sysfs_put(pos);
 815                 if (valid)
 816                         return pos;
 817         }
 818         pos = NULL;
 819         if ((ino > 1) && (ino < INT_MAX)) {
 820                 pos = parent_sd->s_dir.children;
 821                 while (pos && (ino > pos->s_ino))
 822                         pos = pos->s_sibling;
 823         }
 824         return pos;
 825 }
 826
 827 static struct sysfs_dirent *sysfs_dir_next_pos(struct sysfs_dirent *parent_sd,
 828         ino_t ino, struct sysfs_dirent *pos)
 829 {
 830         pos = sysfs_dir_pos(parent_sd, ino, pos);
 831         if (pos)
 832                 pos = pos->s_sibling;
 833         return pos;
 834 }
 835
 836 static int sysfs_readdir(struct file * filp, void * dirent, filldir_t filldir)
 837 {
 838         struct dentry *dentry = filp->f_path.dentry;
 839         struct sysfs_dirent * parent_sd = dentry->d_fsdata;
 840         struct sysfs_dirent *pos = filp->private_data;
 841         ino_t ino;
 842
 843         if (filp->f_pos == 0) {
 844                 ino = parent_sd->s_ino;
 845                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) == 0)
 846                         filp->f_pos++;
 847         }
 848         if (filp->f_pos == 1) {
 849                 if (parent_sd->s_parent)
 850                         ino = parent_sd->s_parent->s_ino;
 851                 else
 852                         ino = parent_sd->s_ino;
 853                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) == 0)
 854                         filp->f_pos++;
 855         }
 856         mutex_lock(&sysfs_mutex);
 857         for (pos = sysfs_dir_pos(parent_sd, filp->f_pos, pos);
 858              pos;
 859              pos = sysfs_dir_next_pos(parent_sd, filp->f_pos, pos)) {
 860                 const char * name;
 861                 unsigned int type;
 862                 int len, ret;
 863
 864                 name = pos->s_name;
 865                 len = strlen(name);
 866                 ino = pos->s_ino;
 867                 type = dt_type(pos);
 868                 filp->f_pos = ino;
 869                 filp->private_data = sysfs_get(pos);
 870
 871                 mutex_unlock(&sysfs_mutex);
 872                 ret = filldir(dirent, name, len, filp->f_pos, ino, type);
 873                 mutex_lock(&sysfs_mutex);
 874                 if (ret < 0)
 875                         break;
 876         }
 877         mutex_unlock(&sysfs_mutex);
 878         if ((filp->f_pos > 1) && !pos) { /* EOF */
 879                 filp->f_pos = INT_MAX;
 880                 filp->private_data = NULL;
 881         }
 882         return 0;
 883 }
 884
 885
 886 const struct file_operations sysfs_dir_operations = {
 887         .read           = generic_read_dir,
 888         .readdir        = sysfs_readdir,
 889         .release        = sysfs_dir_release,
 890         .llseek         = generic_file_llseek,
 891 };