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
 160         if (!(sysfs_type(sd) & SYSFS_ACTIVE_REF))
 161                 return;
 162
 163         sd->s_sibling = (void *)&wait;
 164
 165         rwsem_acquire(&sd->dep_map, 0, 0, _RET_IP_);
 166         /* atomic_add_return() is a mb(), put_active() will always see
 167          * the updated sd->s_sibling.
 168          */
 169         v = atomic_add_return(SD_DEACTIVATED_BIAS, &sd->s_active);
 170
 171         if (v != SD_DEACTIVATED_BIAS) {
 172                 lock_contended(&sd->dep_map, _RET_IP_);
 173                 wait_for_completion(&wait);
 174         }
 175
 176         sd->s_sibling = NULL;
 177
 178         lock_acquired(&sd->dep_map, _RET_IP_);
 179         rwsem_release(&sd->dep_map, 1, _RET_IP_);
 180 }
 181
 182 static int sysfs_alloc_ino(ino_t *pino)
 183 {
 184         int ino, rc;
 185
 186  retry:
 187         spin_lock(&sysfs_ino_lock);
 188         rc = ida_get_new_above(&sysfs_ino_ida, 2, &ino);
 189         spin_unlock(&sysfs_ino_lock);
 190
 191         if (rc == -EAGAIN) {
 192                 if (ida_pre_get(&sysfs_ino_ida, GFP_KERNEL))
 193                         goto retry;
 194                 rc = -ENOMEM;
 195         }
 196
 197         *pino = ino;
 198         return rc;
 199 }
 200
 201 static void sysfs_free_ino(ino_t ino)
 202 {
 203         spin_lock(&sysfs_ino_lock);
 204         ida_remove(&sysfs_ino_ida, ino);
 205         spin_unlock(&sysfs_ino_lock);
 206 }
 207
 208 void release_sysfs_dirent(struct sysfs_dirent * sd)
 209 {
 210         struct sysfs_dirent *parent_sd;
 211
 212  repeat:
 213         /* Moving/renaming is always done while holding reference.
 214          * sd->s_parent won't change beneath us.
 215          */
 216         parent_sd = sd->s_parent;
 217
 218         if (sysfs_type(sd) == SYSFS_KOBJ_LINK)
 219                 sysfs_put(sd->s_symlink.target_sd);
 220         if (sysfs_type(sd) & SYSFS_COPY_NAME)
 221                 kfree(sd->s_name);
 222         if (sd->s_iattr && sd->s_iattr->ia_secdata)
 223                 security_release_secctx(sd->s_iattr->ia_secdata,
 224                                         sd->s_iattr->ia_secdata_len);
 225         kfree(sd->s_iattr);
 226         sysfs_free_ino(sd->s_ino);
 227         kmem_cache_free(sysfs_dir_cachep, sd);
 228
 229         sd = parent_sd;
 230         if (sd && atomic_dec_and_test(&sd->s_count))
 231                 goto repeat;
 232 }
 233
 234 static int sysfs_dentry_delete(struct dentry *dentry)
 235 {
 236         struct sysfs_dirent *sd = dentry->d_fsdata;
 237         return !!(sd->s_flags & SYSFS_FLAG_REMOVED);
 238 }
 239
 240 static int sysfs_dentry_revalidate(struct dentry *dentry, struct nameidata *nd)
 241 {
 242         struct sysfs_dirent *sd = dentry->d_fsdata;
 243         int is_dir;
 244
 245         mutex_lock(&sysfs_mutex);
 246
 247         /* The sysfs dirent has been deleted */
 248         if (sd->s_flags & SYSFS_FLAG_REMOVED)
 249                 goto out_bad;
 250
 251         /* The sysfs dirent has been moved? */
 252         if (dentry->d_parent->d_fsdata != sd->s_parent)
 253                 goto out_bad;
 254
 255         /* The sysfs dirent has been renamed */
 256         if (strcmp(dentry->d_name.name, sd->s_name) != 0)
 257                 goto out_bad;
 258
 259         mutex_unlock(&sysfs_mutex);
 260 out_valid:
 261         return 1;
 262 out_bad:
 263         /* Remove the dentry from the dcache hashes.
 264          * If this is a deleted dentry we use d_drop instead of d_delete
 265          * so sysfs doesn't need to cope with negative dentries.
 266          *
 267          * If this is a dentry that has simply been renamed we
 268          * use d_drop to remove it from the dcache lookup on its
 269          * old parent.  If this dentry persists later when a lookup
 270          * is performed at its new name the dentry will be readded
 271          * to the dcache hashes.
 272          */
 273         is_dir = (sysfs_type(sd) == SYSFS_DIR);
 274         mutex_unlock(&sysfs_mutex);
 275         if (is_dir) {
 276                 /* If we have submounts we must allow the vfs caches
 277                  * to lie about the state of the filesystem to prevent
 278                  * leaks and other nasty things.
 279                  */
 280                 if (have_submounts(dentry))
 281                         goto out_valid;
 282                 shrink_dcache_parent(dentry);
 283         }
 284         d_drop(dentry);
 285         return 0;
 286 }
 287
 288 static void sysfs_dentry_iput(struct dentry *dentry, struct inode *inode)
 289 {
 290         struct sysfs_dirent * sd = dentry->d_fsdata;
 291
 292         sysfs_put(sd);
 293         iput(inode);
 294 }
 295
 296 static const struct dentry_operations sysfs_dentry_ops = {
 297         .d_revalidate   = sysfs_dentry_revalidate,
 298         .d_delete       = sysfs_dentry_delete,
 299         .d_iput         = sysfs_dentry_iput,
 300 };
 301
 302 struct sysfs_dirent *sysfs_new_dirent(const char *name, umode_t mode, int type)
 303 {
 304         char *dup_name = NULL;
 305         struct sysfs_dirent *sd;
 306
 307         if (type & SYSFS_COPY_NAME) {
 308                 name = dup_name = kstrdup(name, GFP_KERNEL);
 309                 if (!name)
 310                         return NULL;
 311         }
 312
 313         sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL);
 314         if (!sd)
 315                 goto err_out1;
 316
 317         if (sysfs_alloc_ino(&sd->s_ino))
 318                 goto err_out2;
 319
 320         atomic_set(&sd->s_count, 1);
 321         atomic_set(&sd->s_active, 0);
 322
 323         sd->s_name = name;
 324         sd->s_mode = mode;
 325         sd->s_flags = type;
 326
 327         return sd;
 328
 329  err_out2:
 330         kmem_cache_free(sysfs_dir_cachep, sd);
 331  err_out1:
 332         kfree(dup_name);
 333         return NULL;
 334 }
 335
 336 /**
 337  *      sysfs_addrm_start - prepare for sysfs_dirent add/remove
 338  *      @acxt: pointer to sysfs_addrm_cxt to be used
 339  *      @parent_sd: parent sysfs_dirent
 340  *
 341  *      This function is called when the caller is about to add or
 342  *      remove sysfs_dirent under @parent_sd.  This function acquires
 343  *      sysfs_mutex.  @acxt is used to keep and pass context to
 344  *      other addrm functions.
 345  *
 346  *      LOCKING:
 347  *      Kernel thread context (may sleep).  sysfs_mutex is locked on
 348  *      return.
 349  */
 350 void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt,
 351                        struct sysfs_dirent *parent_sd)
 352 {
 353         memset(acxt, 0, sizeof(*acxt));
 354         acxt->parent_sd = parent_sd;
 355
 356         mutex_lock(&sysfs_mutex);
 357 }
 358
 359 /**
 360  *      __sysfs_add_one - add sysfs_dirent to parent without warning
 361  *      @acxt: addrm context to use
 362  *      @sd: sysfs_dirent to be added
 363  *
 364  *      Get @acxt->parent_sd and set sd->s_parent to it and increment
 365  *      nlink of parent inode if @sd is a directory and link into the
 366  *      children list of the parent.
 367  *
 368  *      This function should be called between calls to
 369  *      sysfs_addrm_start() and sysfs_addrm_finish() and should be
 370  *      passed the same @acxt as passed to sysfs_addrm_start().
 371  *
 372  *      LOCKING:
 373  *      Determined by sysfs_addrm_start().
 374  *
 375  *      RETURNS:
 376  *      0 on success, -EEXIST if entry with the given name already
 377  *      exists.
 378  */
 379 int __sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
 380 {
 381         struct sysfs_inode_attrs *ps_iattr;
 382
 383         if (sysfs_find_dirent(acxt->parent_sd, sd->s_ns, sd->s_name))
 384                 return -EEXIST;
 385
 386         sd->s_parent = sysfs_get(acxt->parent_sd);
 387
 388         sysfs_link_sibling(sd);
 389
 390         /* Update timestamps on the parent */
 391         ps_iattr = acxt->parent_sd->s_iattr;
 392         if (ps_iattr) {
 393                 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
 394                 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
 395         }
 396
 397         return 0;
 398 }
 399
 400 /**
 401  *      sysfs_pathname - return full path to sysfs dirent
 402  *      @sd: sysfs_dirent whose path we want
 403  *      @path: caller allocated buffer
 404  *
 405  *      Gives the name "/" to the sysfs_root entry; any path returned
 406  *      is relative to wherever sysfs is mounted.
 407  *
 408  *      XXX: does no error checking on @path size
 409  */
 410 static char *sysfs_pathname(struct sysfs_dirent *sd, char *path)
 411 {
 412         if (sd->s_parent) {
 413                 sysfs_pathname(sd->s_parent, path);
 414                 strcat(path, "/");
 415         }
 416         strcat(path, sd->s_name);
 417         return path;
 418 }
 419
 420 /**
 421  *      sysfs_add_one - add sysfs_dirent to parent
 422  *      @acxt: addrm context to use
 423  *      @sd: sysfs_dirent to be added
 424  *
 425  *      Get @acxt->parent_sd and set sd->s_parent to it and increment
 426  *      nlink of parent inode if @sd is a directory and link into the
 427  *      children list of the parent.
 428  *
 429  *      This function should be called between calls to
 430  *      sysfs_addrm_start() and sysfs_addrm_finish() and should be
 431  *      passed the same @acxt as passed to sysfs_addrm_start().
 432  *
 433  *      LOCKING:
 434  *      Determined by sysfs_addrm_start().
 435  *
 436  *      RETURNS:
 437  *      0 on success, -EEXIST if entry with the given name already
 438  *      exists.
 439  */
 440 int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
 441 {
 442         int ret;
 443
 444         ret = __sysfs_add_one(acxt, sd);
 445         if (ret == -EEXIST) {
 446                 char *path = kzalloc(PATH_MAX, GFP_KERNEL);
 447                 WARN(1, KERN_WARNING
 448                      "sysfs: cannot create duplicate filename '%s'\n",
 449                      (path == NULL) ? sd->s_name :
 450                      strcat(strcat(sysfs_pathname(acxt->parent_sd, path), "/"),
 451                             sd->s_name));
 452                 kfree(path);
 453         }
 454
 455         return ret;
 456 }
 457
 458 /**
 459  *      sysfs_remove_one - remove sysfs_dirent from parent
 460  *      @acxt: addrm context to use
 461  *      @sd: sysfs_dirent to be removed
 462  *
 463  *      Mark @sd removed and drop nlink of parent inode if @sd is a
 464  *      directory.  @sd is unlinked from the children list.
 465  *
 466  *      This function should be called between calls to
 467  *      sysfs_addrm_start() and sysfs_addrm_finish() and should be
 468  *      passed the same @acxt as passed to sysfs_addrm_start().
 469  *
 470  *      LOCKING:
 471  *      Determined by sysfs_addrm_start().
 472  */
 473 void sysfs_remove_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
 474 {
 475         struct sysfs_inode_attrs *ps_iattr;
 476
 477         BUG_ON(sd->s_flags & SYSFS_FLAG_REMOVED);
 478
 479         sysfs_unlink_sibling(sd);
 480
 481         /* Update timestamps on the parent */
 482         ps_iattr = acxt->parent_sd->s_iattr;
 483         if (ps_iattr) {
 484                 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
 485                 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
 486         }
 487
 488         sd->s_flags |= SYSFS_FLAG_REMOVED;
 489         sd->s_sibling = acxt->removed;
 490         acxt->removed = sd;
 491 }
 492
 493 /**
 494  *      sysfs_addrm_finish - finish up sysfs_dirent add/remove
 495  *      @acxt: addrm context to finish up
 496  *
 497  *      Finish up sysfs_dirent add/remove.  Resources acquired by
 498  *      sysfs_addrm_start() are released and removed sysfs_dirents are
 499  *      cleaned up.
 500  *
 501  *      LOCKING:
 502  *      sysfs_mutex is released.
 503  */
 504 void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
 505 {
 506         /* release resources acquired by sysfs_addrm_start() */
 507         mutex_unlock(&sysfs_mutex);
 508
 509         /* kill removed sysfs_dirents */
 510         while (acxt->removed) {
 511                 struct sysfs_dirent *sd = acxt->removed;
 512
 513                 acxt->removed = sd->s_sibling;
 514                 sd->s_sibling = NULL;
 515
 516                 sysfs_deactivate(sd);
 517                 unmap_bin_file(sd);
 518                 sysfs_put(sd);
 519         }
 520 }
 521
 522 /**
 523  *      sysfs_find_dirent - find sysfs_dirent with the given name
 524  *      @parent_sd: sysfs_dirent to search under
 525  *      @name: name to look for
 526  *
 527  *      Look for sysfs_dirent with name @name under @parent_sd.
 528  *
 529  *      LOCKING:
 530  *      mutex_lock(sysfs_mutex)
 531  *
 532  *      RETURNS:
 533  *      Pointer to sysfs_dirent if found, NULL if not.
 534  */
 535 struct sysfs_dirent *sysfs_find_dirent(struct sysfs_dirent *parent_sd,
 536                                        const void *ns,
 537                                        const unsigned char *name)
 538 {
 539         struct sysfs_dirent *sd;
 540
 541         for (sd = parent_sd->s_dir.children; sd; sd = sd->s_sibling) {
 542                 if (ns && sd->s_ns && (sd->s_ns != ns))
 543                         continue;
 544                 if (!strcmp(sd->s_name, name))
 545                         return sd;
 546         }
 547         return NULL;
 548 }
 549
 550 /**
 551  *      sysfs_get_dirent - find and get sysfs_dirent with the given name
 552  *      @parent_sd: sysfs_dirent to search under
 553  *      @name: name to look for
 554  *
 555  *      Look for sysfs_dirent with name @name under @parent_sd and get
 556  *      it if found.
 557  *
 558  *      LOCKING:
 559  *      Kernel thread context (may sleep).  Grabs sysfs_mutex.
 560  *
 561  *      RETURNS:
 562  *      Pointer to sysfs_dirent if found, NULL if not.
 563  */
 564 struct sysfs_dirent *sysfs_get_dirent(struct sysfs_dirent *parent_sd,
 565                                       const void *ns,
 566                                       const unsigned char *name)
 567 {
 568         struct sysfs_dirent *sd;
 569
 570         mutex_lock(&sysfs_mutex);
 571         sd = sysfs_find_dirent(parent_sd, ns, name);
 572         sysfs_get(sd);
 573         mutex_unlock(&sysfs_mutex);
 574
 575         return sd;
 576 }
 577 EXPORT_SYMBOL_GPL(sysfs_get_dirent);
 578
 579 static int create_dir(struct kobject *kobj, struct sysfs_dirent *parent_sd,
 580         enum kobj_ns_type type, const void *ns, const char *name,
 581         struct sysfs_dirent **p_sd)
 582 {
 583         umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
 584         struct sysfs_addrm_cxt acxt;
 585         struct sysfs_dirent *sd;
 586         int rc;
 587
 588         /* allocate */
 589         sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
 590         if (!sd)
 591                 return -ENOMEM;
 592
 593         sd->s_flags |= (type << SYSFS_NS_TYPE_SHIFT);
 594         sd->s_ns = ns;
 595         sd->s_dir.kobj = kobj;
 596
 597         /* link in */
 598         sysfs_addrm_start(&acxt, parent_sd);
 599         rc = sysfs_add_one(&acxt, sd);
 600         sysfs_addrm_finish(&acxt);
 601
 602         if (rc == 0)
 603                 *p_sd = sd;
 604         else
 605                 sysfs_put(sd);
 606
 607         return rc;
 608 }
 609
 610 int sysfs_create_subdir(struct kobject *kobj, const char *name,
 611                         struct sysfs_dirent **p_sd)
 612 {
 613         return create_dir(kobj, kobj->sd,
 614                           KOBJ_NS_TYPE_NONE, NULL, name, p_sd);
 615 }
 616
 617 /**
 618  *      sysfs_read_ns_type: return associated ns_type
 619  *      @kobj: the kobject being queried
 620  *
 621  *      Each kobject can be tagged with exactly one namespace type
 622  *      (i.e. network or user).  Return the ns_type associated with
 623  *      this object if any
 624  */
 625 static enum kobj_ns_type sysfs_read_ns_type(struct kobject *kobj)
 626 {
 627         const struct kobj_ns_type_operations *ops;
 628         enum kobj_ns_type type;
 629
 630         ops = kobj_child_ns_ops(kobj);
 631         if (!ops)
 632                 return KOBJ_NS_TYPE_NONE;
 633
 634         type = ops->type;
 635         BUG_ON(type <= KOBJ_NS_TYPE_NONE);
 636         BUG_ON(type >= KOBJ_NS_TYPES);
 637         BUG_ON(!kobj_ns_type_registered(type));
 638
 639         return type;
 640 }
 641
 642 /**
 643  *      sysfs_create_dir - create a directory for an object.
 644  *      @kobj:          object we're creating directory for.
 645  */
 646 int sysfs_create_dir(struct kobject * kobj)
 647 {
 648         enum kobj_ns_type type;
 649         struct sysfs_dirent *parent_sd, *sd;
 650         const void *ns = NULL;
 651         int error = 0;
 652
 653         BUG_ON(!kobj);
 654
 655         if (kobj->parent)
 656                 parent_sd = kobj->parent->sd;
 657         else
 658                 parent_sd = &sysfs_root;
 659
 660         if (sysfs_ns_type(parent_sd))
 661                 ns = kobj->ktype->namespace(kobj);
 662         type = sysfs_read_ns_type(kobj);
 663
 664         error = create_dir(kobj, parent_sd, type, ns, kobject_name(kobj), &sd);
 665         if (!error)
 666                 kobj->sd = sd;
 667         return error;
 668 }
 669
 670 static struct dentry * sysfs_lookup(struct inode *dir, struct dentry *dentry,
 671                                 struct nameidata *nd)
 672 {
 673         struct dentry *ret = NULL;
 674         struct dentry *parent = dentry->d_parent;
 675         struct sysfs_dirent *parent_sd = parent->d_fsdata;
 676         struct sysfs_dirent *sd;
 677         struct inode *inode;
 678         enum kobj_ns_type type;
 679         const void *ns;
 680
 681         mutex_lock(&sysfs_mutex);
 682
 683         type = sysfs_ns_type(parent_sd);
 684         ns = sysfs_info(dir->i_sb)->ns[type];
 685
 686         sd = sysfs_find_dirent(parent_sd, ns, dentry->d_name.name);
 687
 688         /* no such entry */
 689         if (!sd) {
 690                 ret = ERR_PTR(-ENOENT);
 691                 goto out_unlock;
 692         }
 693
 694         /* attach dentry and inode */
 695         inode = sysfs_get_inode(dir->i_sb, sd);
 696         if (!inode) {
 697                 ret = ERR_PTR(-ENOMEM);
 698                 goto out_unlock;
 699         }
 700
 701         /* instantiate and hash dentry */
 702         ret = d_find_alias(inode);
 703         if (!ret) {
 704                 dentry->d_op = &sysfs_dentry_ops;
 705                 dentry->d_fsdata = sysfs_get(sd);
 706                 d_add(dentry, inode);
 707         } else {
 708                 d_move(ret, dentry);
 709                 iput(inode);
 710         }
 711
 712  out_unlock:
 713         mutex_unlock(&sysfs_mutex);
 714         return ret;
 715 }
 716
 717 const struct inode_operations sysfs_dir_inode_operations = {
 718         .lookup         = sysfs_lookup,
 719         .permission     = sysfs_permission,
 720         .setattr        = sysfs_setattr,
 721         .getattr        = sysfs_getattr,
 722         .setxattr       = sysfs_setxattr,
 723 };
 724
 725 static void remove_dir(struct sysfs_dirent *sd)
 726 {
 727         struct sysfs_addrm_cxt acxt;
 728
 729         sysfs_addrm_start(&acxt, sd->s_parent);
 730         sysfs_remove_one(&acxt, sd);
 731         sysfs_addrm_finish(&acxt);
 732 }
 733
 734 void sysfs_remove_subdir(struct sysfs_dirent *sd)
 735 {
 736         remove_dir(sd);
 737 }
 738
 739
 740 static void __sysfs_remove_dir(struct sysfs_dirent *dir_sd)
 741 {
 742         struct sysfs_addrm_cxt acxt;
 743         struct sysfs_dirent **pos;
 744
 745         if (!dir_sd)
 746                 return;
 747
 748         pr_debug("sysfs %s: removing dir\n", dir_sd->s_name);
 749         sysfs_addrm_start(&acxt, dir_sd);
 750         pos = &dir_sd->s_dir.children;
 751         while (*pos) {
 752                 struct sysfs_dirent *sd = *pos;
 753
 754                 if (sysfs_type(sd) != SYSFS_DIR)
 755                         sysfs_remove_one(&acxt, sd);
 756                 else
 757                         pos = &(*pos)->s_sibling;
 758         }
 759         sysfs_addrm_finish(&acxt);
 760
 761         remove_dir(dir_sd);
 762 }
 763
 764 /**
 765  *      sysfs_remove_dir - remove an object's directory.
 766  *      @kobj:  object.
 767  *
 768  *      The only thing special about this is that we remove any files in
 769  *      the directory before we remove the directory, and we've inlined
 770  *      what used to be sysfs_rmdir() below, instead of calling separately.
 771  */
 772
 773 void sysfs_remove_dir(struct kobject * kobj)
 774 {
 775         struct sysfs_dirent *sd = kobj->sd;
 776
 777         spin_lock(&sysfs_assoc_lock);
 778         kobj->sd = NULL;
 779         spin_unlock(&sysfs_assoc_lock);
 780
 781         __sysfs_remove_dir(sd);
 782 }
 783
 784 int sysfs_rename(struct sysfs_dirent *sd,
 785         struct sysfs_dirent *new_parent_sd, const void *new_ns,
 786         const char *new_name)
 787 {
 788         const char *dup_name = NULL;
 789         int error;
 790
 791         mutex_lock(&sysfs_mutex);
 792
 793         error = 0;
 794         if ((sd->s_parent == new_parent_sd) && (sd->s_ns == new_ns) &&
 795             (strcmp(sd->s_name, new_name) == 0))
 796                 goto out;       /* nothing to rename */
 797
 798         error = -EEXIST;
 799         if (sysfs_find_dirent(new_parent_sd, new_ns, new_name))
 800                 goto out;
 801
 802         /* rename sysfs_dirent */
 803         if (strcmp(sd->s_name, new_name) != 0) {
 804                 error = -ENOMEM;
 805                 new_name = dup_name = kstrdup(new_name, GFP_KERNEL);
 806                 if (!new_name)
 807                         goto out;
 808
 809                 dup_name = sd->s_name;
 810                 sd->s_name = new_name;
 811         }
 812
 813         /* Remove from old parent's list and insert into new parent's list. */
 814         if (sd->s_parent != new_parent_sd) {
 815                 sysfs_unlink_sibling(sd);
 816                 sysfs_get(new_parent_sd);
 817                 sysfs_put(sd->s_parent);
 818                 sd->s_parent = new_parent_sd;
 819                 sysfs_link_sibling(sd);
 820         }
 821         sd->s_ns = new_ns;
 822
 823         error = 0;
 824  out:
 825         mutex_unlock(&sysfs_mutex);
 826         kfree(dup_name);
 827         return error;
 828 }
 829
 830 int sysfs_rename_dir(struct kobject *kobj, const char *new_name)
 831 {
 832         struct sysfs_dirent *parent_sd = kobj->sd->s_parent;
 833         const void *new_ns = NULL;
 834
 835         if (sysfs_ns_type(parent_sd))
 836                 new_ns = kobj->ktype->namespace(kobj);
 837
 838         return sysfs_rename(kobj->sd, parent_sd, new_ns, new_name);
 839 }
 840
 841 int sysfs_move_dir(struct kobject *kobj, struct kobject *new_parent_kobj)
 842 {
 843         struct sysfs_dirent *sd = kobj->sd;
 844         struct sysfs_dirent *new_parent_sd;
 845         const void *new_ns = NULL;
 846
 847         BUG_ON(!sd->s_parent);
 848         if (sysfs_ns_type(sd->s_parent))
 849                 new_ns = kobj->ktype->namespace(kobj);
 850         new_parent_sd = new_parent_kobj && new_parent_kobj->sd ?
 851                 new_parent_kobj->sd : &sysfs_root;
 852
 853         return sysfs_rename(sd, new_parent_sd, new_ns, sd->s_name);
 854 }
 855
 856 /* Relationship between s_mode and the DT_xxx types */
 857 static inline unsigned char dt_type(struct sysfs_dirent *sd)
 858 {
 859         return (sd->s_mode >> 12) & 15;
 860 }
 861
 862 static int sysfs_dir_release(struct inode *inode, struct file *filp)
 863 {
 864         sysfs_put(filp->private_data);
 865         return 0;
 866 }
 867
 868 static struct sysfs_dirent *sysfs_dir_pos(const void *ns,
 869         struct sysfs_dirent *parent_sd, ino_t ino, struct sysfs_dirent *pos)
 870 {
 871         if (pos) {
 872                 int valid = !(pos->s_flags & SYSFS_FLAG_REMOVED) &&
 873                         pos->s_parent == parent_sd &&
 874                         ino == pos->s_ino;
 875                 sysfs_put(pos);
 876                 if (!valid)
 877                         pos = NULL;
 878         }
 879         if (!pos && (ino > 1) && (ino < INT_MAX)) {
 880                 pos = parent_sd->s_dir.children;
 881                 while (pos && (ino > pos->s_ino))
 882                         pos = pos->s_sibling;
 883         }
 884         while (pos && pos->s_ns && pos->s_ns != ns)
 885                 pos = pos->s_sibling;
 886         return pos;
 887 }
 888
 889 static struct sysfs_dirent *sysfs_dir_next_pos(const void *ns,
 890         struct sysfs_dirent *parent_sd, ino_t ino, struct sysfs_dirent *pos)
 891 {
 892         pos = sysfs_dir_pos(ns, parent_sd, ino, pos);
 893         if (pos)
 894                 pos = pos->s_sibling;
 895         while (pos && pos->s_ns && pos->s_ns != ns)
 896                 pos = pos->s_sibling;
 897         return pos;
 898 }
 899
 900 static int sysfs_readdir(struct file * filp, void * dirent, filldir_t filldir)
 901 {
 902         struct dentry *dentry = filp->f_path.dentry;
 903         struct sysfs_dirent * parent_sd = dentry->d_fsdata;
 904         struct sysfs_dirent *pos = filp->private_data;
 905         enum kobj_ns_type type;
 906         const void *ns;
 907         ino_t ino;
 908
 909         type = sysfs_ns_type(parent_sd);
 910         ns = sysfs_info(dentry->d_sb)->ns[type];
 911
 912         if (filp->f_pos == 0) {
 913                 ino = parent_sd->s_ino;
 914                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) == 0)
 915                         filp->f_pos++;
 916         }
 917         if (filp->f_pos == 1) {
 918                 if (parent_sd->s_parent)
 919                         ino = parent_sd->s_parent->s_ino;
 920                 else
 921                         ino = parent_sd->s_ino;
 922                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) == 0)
 923                         filp->f_pos++;
 924         }
 925         mutex_lock(&sysfs_mutex);
 926         for (pos = sysfs_dir_pos(ns, parent_sd, filp->f_pos, pos);
 927              pos;
 928              pos = sysfs_dir_next_pos(ns, parent_sd, filp->f_pos, pos)) {
 929                 const char * name;
 930                 unsigned int type;
 931                 int len, ret;
 932
 933                 name = pos->s_name;
 934                 len = strlen(name);
 935                 ino = pos->s_ino;
 936                 type = dt_type(pos);
 937                 filp->f_pos = ino;
 938                 filp->private_data = sysfs_get(pos);
 939
 940                 mutex_unlock(&sysfs_mutex);
 941                 ret = filldir(dirent, name, len, filp->f_pos, ino, type);
 942                 mutex_lock(&sysfs_mutex);
 943                 if (ret < 0)
 944                         break;
 945         }
 946         mutex_unlock(&sysfs_mutex);
 947         if ((filp->f_pos > 1) && !pos) { /* EOF */
 948                 filp->f_pos = INT_MAX;
 949                 filp->private_data = NULL;
 950         }
 951         return 0;
 952 }
 953
 954
 955 const struct file_operations sysfs_dir_operations = {
 956         .read           = generic_read_dir,
 957         .readdir        = sysfs_readdir,
 958         .release        = sysfs_dir_release,
 959         .llseek         = generic_file_llseek,
 960 };