fs/sysfs/dir.c

   1 /*
   2  * dir.c - Operations for sysfs directories.
   3  */
   4
   5 #undef DEBUG
   6
   7 #include <linux/fs.h>
   8 #include <linux/mount.h>
   9 #include <linux/module.h>
  10 #include <linux/kobject.h>
  11 #include <linux/namei.h>
  12 #include <linux/idr.h>
  13 #include <linux/completion.h>
  14 #include <asm/semaphore.h>
  15 #include "sysfs.h"
  16
  17 DECLARE_RWSEM(sysfs_rename_sem);
  18 spinlock_t sysfs_lock = SPIN_LOCK_UNLOCKED;
  19 spinlock_t kobj_sysfs_assoc_lock = SPIN_LOCK_UNLOCKED;
  20
  21 static spinlock_t sysfs_ino_lock = SPIN_LOCK_UNLOCKED;
  22 static DEFINE_IDA(sysfs_ino_ida);
  23
  24 /**
  25  *      sysfs_link_sibling - link sysfs_dirent into sibling list
  26  *      @sd: sysfs_dirent of interest
  27  *
  28  *      Link @sd into its sibling list which starts from
  29  *      sd->s_parent->s_children.
  30  *
  31  *      Locking:
  32  *      mutex_lock(sd->s_parent->dentry->d_inode->i_mutex)
  33  */
  34 static void sysfs_link_sibling(struct sysfs_dirent *sd)
  35 {
  36         struct sysfs_dirent *parent_sd = sd->s_parent;
  37
  38         BUG_ON(sd->s_sibling);
  39         sd->s_sibling = parent_sd->s_children;
  40         parent_sd->s_children = sd;
  41 }
  42
  43 /**
  44  *      sysfs_unlink_sibling - unlink sysfs_dirent from sibling list
  45  *      @sd: sysfs_dirent of interest
  46  *
  47  *      Unlink @sd from its sibling list which starts from
  48  *      sd->s_parent->s_children.
  49  *
  50  *      Locking:
  51  *      mutex_lock(sd->s_parent->dentry->d_inode->i_mutex)
  52  */
  53 static void sysfs_unlink_sibling(struct sysfs_dirent *sd)
  54 {
  55         struct sysfs_dirent **pos;
  56
  57         for (pos = &sd->s_parent->s_children; *pos; pos = &(*pos)->s_sibling) {
  58                 if (*pos == sd) {
  59                         *pos = sd->s_sibling;
  60                         sd->s_sibling = NULL;
  61                         break;
  62                 }
  63         }
  64 }
  65
  66 /**
  67  *      sysfs_get_active - get an active reference to sysfs_dirent
  68  *      @sd: sysfs_dirent to get an active reference to
  69  *
  70  *      Get an active reference of @sd.  This function is noop if @sd
  71  *      is NULL.
  72  *
  73  *      RETURNS:
  74  *      Pointer to @sd on success, NULL on failure.
  75  */
  76 struct sysfs_dirent *sysfs_get_active(struct sysfs_dirent *sd)
  77 {
  78         if (unlikely(!sd))
  79                 return NULL;
  80
  81         while (1) {
  82                 int v, t;
  83
  84                 v = atomic_read(&sd->s_active);
  85                 if (unlikely(v < 0))
  86                         return NULL;
  87
  88                 t = atomic_cmpxchg(&sd->s_active, v, v + 1);
  89                 if (likely(t == v))
  90                         return sd;
  91                 if (t < 0)
  92                         return NULL;
  93
  94                 cpu_relax();
  95         }
  96 }
  97
  98 /**
  99  *      sysfs_put_active - put an active reference to sysfs_dirent
 100  *      @sd: sysfs_dirent to put an active reference to
 101  *
 102  *      Put an active reference to @sd.  This function is noop if @sd
 103  *      is NULL.
 104  */
 105 void sysfs_put_active(struct sysfs_dirent *sd)
 106 {
 107         struct completion *cmpl;
 108         int v;
 109
 110         if (unlikely(!sd))
 111                 return;
 112
 113         v = atomic_dec_return(&sd->s_active);
 114         if (likely(v != SD_DEACTIVATED_BIAS))
 115                 return;
 116
 117         /* atomic_dec_return() is a mb(), we'll always see the updated
 118          * sd->s_sibling.
 119          */
 120         cmpl = (void *)sd->s_sibling;
 121         complete(cmpl);
 122 }
 123
 124 /**
 125  *      sysfs_get_active_two - get active references to sysfs_dirent and parent
 126  *      @sd: sysfs_dirent of interest
 127  *
 128  *      Get active reference to @sd and its parent.  Parent's active
 129  *      reference is grabbed first.  This function is noop if @sd is
 130  *      NULL.
 131  *
 132  *      RETURNS:
 133  *      Pointer to @sd on success, NULL on failure.
 134  */
 135 struct sysfs_dirent *sysfs_get_active_two(struct sysfs_dirent *sd)
 136 {
 137         if (sd) {
 138                 if (sd->s_parent && unlikely(!sysfs_get_active(sd->s_parent)))
 139                         return NULL;
 140                 if (unlikely(!sysfs_get_active(sd))) {
 141                         sysfs_put_active(sd->s_parent);
 142                         return NULL;
 143                 }
 144         }
 145         return sd;
 146 }
 147
 148 /**
 149  *      sysfs_put_active_two - put active references to sysfs_dirent and parent
 150  *      @sd: sysfs_dirent of interest
 151  *
 152  *      Put active references to @sd and its parent.  This function is
 153  *      noop if @sd is NULL.
 154  */
 155 void sysfs_put_active_two(struct sysfs_dirent *sd)
 156 {
 157         if (sd) {
 158                 sysfs_put_active(sd);
 159                 sysfs_put_active(sd->s_parent);
 160         }
 161 }
 162
 163 /**
 164  *      sysfs_deactivate - deactivate sysfs_dirent
 165  *      @sd: sysfs_dirent to deactivate
 166  *
 167  *      Deny new active references and drain existing ones.
 168  */
 169 void sysfs_deactivate(struct sysfs_dirent *sd)
 170 {
 171         DECLARE_COMPLETION_ONSTACK(wait);
 172         int v;
 173
 174         BUG_ON(sd->s_sibling);
 175         sd->s_sibling = (void *)&wait;
 176
 177         /* atomic_add_return() is a mb(), put_active() will always see
 178          * the updated sd->s_sibling.
 179          */
 180         v = atomic_add_return(SD_DEACTIVATED_BIAS, &sd->s_active);
 181
 182         if (v != SD_DEACTIVATED_BIAS)
 183                 wait_for_completion(&wait);
 184
 185         sd->s_sibling = NULL;
 186 }
 187
 188 static int sysfs_alloc_ino(ino_t *pino)
 189 {
 190         int ino, rc;
 191
 192  retry:
 193         spin_lock(&sysfs_ino_lock);
 194         rc = ida_get_new_above(&sysfs_ino_ida, 2, &ino);
 195         spin_unlock(&sysfs_ino_lock);
 196
 197         if (rc == -EAGAIN) {
 198                 if (ida_pre_get(&sysfs_ino_ida, GFP_KERNEL))
 199                         goto retry;
 200                 rc = -ENOMEM;
 201         }
 202
 203         *pino = ino;
 204         return rc;
 205 }
 206
 207 static void sysfs_free_ino(ino_t ino)
 208 {
 209         spin_lock(&sysfs_ino_lock);
 210         ida_remove(&sysfs_ino_ida, ino);
 211         spin_unlock(&sysfs_ino_lock);
 212 }
 213
 214 void release_sysfs_dirent(struct sysfs_dirent * sd)
 215 {
 216         struct sysfs_dirent *parent_sd;
 217
 218  repeat:
 219         parent_sd = sd->s_parent;
 220
 221         if (sysfs_type(sd) == SYSFS_KOBJ_LINK)
 222                 sysfs_put(sd->s_elem.symlink.target_sd);
 223         if (sysfs_type(sd) & SYSFS_COPY_NAME)
 224                 kfree(sd->s_name);
 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 void sysfs_d_iput(struct dentry * dentry, struct inode * inode)
 235 {
 236         struct sysfs_dirent * sd = dentry->d_fsdata;
 237
 238         if (sd) {
 239                 /* sd->s_dentry is protected with sysfs_lock.  This
 240                  * allows sysfs_drop_dentry() to dereference it.
 241                  */
 242                 spin_lock(&sysfs_lock);
 243
 244                 /* The dentry might have been deleted or another
 245                  * lookup could have happened updating sd->s_dentry to
 246                  * point the new dentry.  Ignore if it isn't pointing
 247                  * to this dentry.
 248                  */
 249                 if (sd->s_dentry == dentry)
 250                         sd->s_dentry = NULL;
 251                 spin_unlock(&sysfs_lock);
 252                 sysfs_put(sd);
 253         }
 254         iput(inode);
 255 }
 256
 257 static struct dentry_operations sysfs_dentry_ops = {
 258         .d_iput         = sysfs_d_iput,
 259 };
 260
 261 struct sysfs_dirent *sysfs_new_dirent(const char *name, umode_t mode, int type)
 262 {
 263         char *dup_name = NULL;
 264         struct sysfs_dirent *sd = NULL;
 265
 266         if (type & SYSFS_COPY_NAME) {
 267                 name = dup_name = kstrdup(name, GFP_KERNEL);
 268                 if (!name)
 269                         goto err_out;
 270         }
 271
 272         sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL);
 273         if (!sd)
 274                 goto err_out;
 275
 276         if (sysfs_alloc_ino(&sd->s_ino))
 277                 goto err_out;
 278
 279         atomic_set(&sd->s_count, 1);
 280         atomic_set(&sd->s_active, 0);
 281         atomic_set(&sd->s_event, 1);
 282
 283         sd->s_name = name;
 284         sd->s_mode = mode;
 285         sd->s_flags = type;
 286
 287         return sd;
 288
 289  err_out:
 290         kfree(dup_name);
 291         kmem_cache_free(sysfs_dir_cachep, sd);
 292         return NULL;
 293 }
 294
 295 static void sysfs_attach_dentry(struct sysfs_dirent *sd, struct dentry *dentry)
 296 {
 297         dentry->d_op = &sysfs_dentry_ops;
 298         dentry->d_fsdata = sysfs_get(sd);
 299
 300         /* protect sd->s_dentry against sysfs_d_iput */
 301         spin_lock(&sysfs_lock);
 302         sd->s_dentry = dentry;
 303         spin_unlock(&sysfs_lock);
 304
 305         d_rehash(dentry);
 306 }
 307
 308 void sysfs_attach_dirent(struct sysfs_dirent *sd,
 309                          struct sysfs_dirent *parent_sd, struct dentry *dentry)
 310 {
 311         if (dentry)
 312                 sysfs_attach_dentry(sd, dentry);
 313
 314         if (parent_sd) {
 315                 sd->s_parent = sysfs_get(parent_sd);
 316                 sysfs_link_sibling(sd);
 317         }
 318 }
 319
 320 /*
 321  *
 322  * Return -EEXIST if there is already a sysfs element with the same name for
 323  * the same parent.
 324  *
 325  * called with parent inode's i_mutex held
 326  */
 327 int sysfs_dirent_exist(struct sysfs_dirent *parent_sd,
 328                           const unsigned char *new)
 329 {
 330         struct sysfs_dirent * sd;
 331
 332         for (sd = parent_sd->s_children; sd; sd = sd->s_sibling) {
 333                 if (sysfs_type(sd)) {
 334                         if (strcmp(sd->s_name, new))
 335                                 continue;
 336                         else
 337                                 return -EEXIST;
 338                 }
 339         }
 340
 341         return 0;
 342 }
 343
 344 static int create_dir(struct kobject *kobj, struct dentry *parent,
 345                       const char *name, struct dentry **p_dentry)
 346 {
 347         int error;
 348         umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
 349         struct dentry *dentry;
 350         struct inode *inode;
 351         struct sysfs_dirent *sd;
 352
 353         mutex_lock(&parent->d_inode->i_mutex);
 354
 355         /* allocate */
 356         dentry = lookup_one_len(name, parent, strlen(name));
 357         if (IS_ERR(dentry)) {
 358                 error = PTR_ERR(dentry);
 359                 goto out_unlock;
 360         }
 361
 362         error = -EEXIST;
 363         if (dentry->d_inode)
 364                 goto out_dput;
 365
 366         error = -ENOMEM;
 367         sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
 368         if (!sd)
 369                 goto out_drop;
 370         sd->s_elem.dir.kobj = kobj;
 371
 372         inode = sysfs_get_inode(sd);
 373         if (!inode)
 374                 goto out_sput;
 375
 376         if (inode->i_state & I_NEW) {
 377                 inode->i_op = &sysfs_dir_inode_operations;
 378                 inode->i_fop = &sysfs_dir_operations;
 379                 /* directory inodes start off with i_nlink == 2 (for ".") */
 380                 inc_nlink(inode);
 381         }
 382
 383         /* link in */
 384         error = -EEXIST;
 385         if (sysfs_dirent_exist(parent->d_fsdata, name))
 386                 goto out_iput;
 387
 388         sysfs_instantiate(dentry, inode);
 389         inc_nlink(parent->d_inode);
 390         sysfs_attach_dirent(sd, parent->d_fsdata, dentry);
 391
 392         *p_dentry = dentry;
 393         error = 0;
 394         goto out_unlock;        /* pin directory dentry in core */
 395
 396  out_iput:
 397         iput(inode);
 398  out_sput:
 399         sysfs_put(sd);
 400  out_drop:
 401         d_drop(dentry);
 402  out_dput:
 403         dput(dentry);
 404  out_unlock:
 405         mutex_unlock(&parent->d_inode->i_mutex);
 406         return error;
 407 }
 408
 409
 410 int sysfs_create_subdir(struct kobject * k, const char * n, struct dentry ** d)
 411 {
 412         return create_dir(k,k->dentry,n,d);
 413 }
 414
 415 /**
 416  *      sysfs_create_dir - create a directory for an object.
 417  *      @kobj:          object we're creating directory for.
 418  *      @shadow_parent: parent parent object.
 419  */
 420
 421 int sysfs_create_dir(struct kobject * kobj, struct dentry *shadow_parent)
 422 {
 423         struct dentry * dentry = NULL;
 424         struct dentry * parent;
 425         int error = 0;
 426
 427         BUG_ON(!kobj);
 428
 429         if (shadow_parent)
 430                 parent = shadow_parent;
 431         else if (kobj->parent)
 432                 parent = kobj->parent->dentry;
 433         else if (sysfs_mount && sysfs_mount->mnt_sb)
 434                 parent = sysfs_mount->mnt_sb->s_root;
 435         else
 436                 return -EFAULT;
 437
 438         error = create_dir(kobj,parent,kobject_name(kobj),&dentry);
 439         if (!error)
 440                 kobj->dentry = dentry;
 441         return error;
 442 }
 443
 444 static struct dentry * sysfs_lookup(struct inode *dir, struct dentry *dentry,
 445                                 struct nameidata *nd)
 446 {
 447         struct sysfs_dirent * parent_sd = dentry->d_parent->d_fsdata;
 448         struct sysfs_dirent * sd;
 449         struct bin_attribute *bin_attr;
 450         struct inode *inode;
 451         int found = 0;
 452
 453         for (sd = parent_sd->s_children; sd; sd = sd->s_sibling) {
 454                 if ((sysfs_type(sd) & SYSFS_NOT_PINNED) &&
 455                     !strcmp(sd->s_name, dentry->d_name.name)) {
 456                         found = 1;
 457                         break;
 458                 }
 459         }
 460
 461         /* no such entry */
 462         if (!found)
 463                 return NULL;
 464
 465         /* attach dentry and inode */
 466         inode = sysfs_get_inode(sd);
 467         if (!inode)
 468                 return ERR_PTR(-ENOMEM);
 469
 470         if (inode->i_state & I_NEW) {
 471                 /* initialize inode according to type */
 472                 switch (sysfs_type(sd)) {
 473                 case SYSFS_KOBJ_ATTR:
 474                         inode->i_size = PAGE_SIZE;
 475                         inode->i_fop = &sysfs_file_operations;
 476                         break;
 477                 case SYSFS_KOBJ_BIN_ATTR:
 478                         bin_attr = sd->s_elem.bin_attr.bin_attr;
 479                         inode->i_size = bin_attr->size;
 480                         inode->i_fop = &bin_fops;
 481                         break;
 482                 case SYSFS_KOBJ_LINK:
 483                         inode->i_op = &sysfs_symlink_inode_operations;
 484                         break;
 485                 default:
 486                         BUG();
 487                 }
 488         }
 489
 490         sysfs_instantiate(dentry, inode);
 491         sysfs_attach_dentry(sd, dentry);
 492
 493         return NULL;
 494 }
 495
 496 const struct inode_operations sysfs_dir_inode_operations = {
 497         .lookup         = sysfs_lookup,
 498         .setattr        = sysfs_setattr,
 499 };
 500
 501 static void remove_dir(struct dentry * d)
 502 {
 503         struct dentry *parent = d->d_parent;
 504         struct sysfs_dirent *sd = d->d_fsdata;
 505
 506         mutex_lock(&parent->d_inode->i_mutex);
 507
 508         sysfs_unlink_sibling(sd);
 509
 510         pr_debug(" o %s removing done (%d)\n",d->d_name.name,
 511                  atomic_read(&d->d_count));
 512
 513         mutex_unlock(&parent->d_inode->i_mutex);
 514
 515         sysfs_drop_dentry(sd);
 516         sysfs_deactivate(sd);
 517         sysfs_put(sd);
 518 }
 519
 520 void sysfs_remove_subdir(struct dentry * d)
 521 {
 522         remove_dir(d);
 523 }
 524
 525
 526 static void __sysfs_remove_dir(struct dentry *dentry)
 527 {
 528         struct sysfs_dirent *removed = NULL;
 529         struct sysfs_dirent *parent_sd;
 530         struct sysfs_dirent **pos;
 531
 532         if (!dentry)
 533                 return;
 534
 535         pr_debug("sysfs %s: removing dir\n",dentry->d_name.name);
 536         mutex_lock(&dentry->d_inode->i_mutex);
 537         parent_sd = dentry->d_fsdata;
 538         pos = &parent_sd->s_children;
 539         while (*pos) {
 540                 struct sysfs_dirent *sd = *pos;
 541
 542                 if (sysfs_type(sd) && (sysfs_type(sd) & SYSFS_NOT_PINNED)) {
 543                         *pos = sd->s_sibling;
 544                         sd->s_sibling = removed;
 545                         removed = sd;
 546                 } else
 547                         pos = &(*pos)->s_sibling;
 548         }
 549         mutex_unlock(&dentry->d_inode->i_mutex);
 550
 551         while (removed) {
 552                 struct sysfs_dirent *sd = removed;
 553
 554                 removed = sd->s_sibling;
 555                 sd->s_sibling = NULL;
 556
 557                 sysfs_drop_dentry(sd);
 558                 sysfs_deactivate(sd);
 559                 sysfs_put(sd);
 560         }
 561
 562         remove_dir(dentry);
 563 }
 564
 565 /**
 566  *      sysfs_remove_dir - remove an object's directory.
 567  *      @kobj:  object.
 568  *
 569  *      The only thing special about this is that we remove any files in
 570  *      the directory before we remove the directory, and we've inlined
 571  *      what used to be sysfs_rmdir() below, instead of calling separately.
 572  */
 573
 574 void sysfs_remove_dir(struct kobject * kobj)
 575 {
 576         struct dentry *d = kobj->dentry;
 577
 578         spin_lock(&kobj_sysfs_assoc_lock);
 579         kobj->dentry = NULL;
 580         spin_unlock(&kobj_sysfs_assoc_lock);
 581
 582         __sysfs_remove_dir(d);
 583 }
 584
 585 int sysfs_rename_dir(struct kobject * kobj, struct dentry *new_parent,
 586                      const char *new_name)
 587 {
 588         struct sysfs_dirent *sd = kobj->dentry->d_fsdata;
 589         struct sysfs_dirent *parent_sd = new_parent->d_fsdata;
 590         struct dentry *new_dentry;
 591         char *dup_name;
 592         int error;
 593
 594         if (!new_parent)
 595                 return -EFAULT;
 596
 597         down_write(&sysfs_rename_sem);
 598         mutex_lock(&new_parent->d_inode->i_mutex);
 599
 600         new_dentry = lookup_one_len(new_name, new_parent, strlen(new_name));
 601         if (IS_ERR(new_dentry)) {
 602                 error = PTR_ERR(new_dentry);
 603                 goto out_unlock;
 604         }
 605
 606         /* By allowing two different directories with the same
 607          * d_parent we allow this routine to move between different
 608          * shadows of the same directory
 609          */
 610         error = -EINVAL;
 611         if (kobj->dentry->d_parent->d_inode != new_parent->d_inode ||
 612             new_dentry->d_parent->d_inode != new_parent->d_inode ||
 613             new_dentry == kobj->dentry)
 614                 goto out_dput;
 615
 616         error = -EEXIST;
 617         if (new_dentry->d_inode)
 618                 goto out_dput;
 619
 620         /* rename kobject and sysfs_dirent */
 621         error = -ENOMEM;
 622         new_name = dup_name = kstrdup(new_name, GFP_KERNEL);
 623         if (!new_name)
 624                 goto out_drop;
 625
 626         error = kobject_set_name(kobj, "%s", new_name);
 627         if (error)
 628                 goto out_free;
 629
 630         kfree(sd->s_name);
 631         sd->s_name = new_name;
 632
 633         /* move under the new parent */
 634         d_add(new_dentry, NULL);
 635         d_move(kobj->dentry, new_dentry);
 636
 637         sysfs_unlink_sibling(sd);
 638         sysfs_get(parent_sd);
 639         sysfs_put(sd->s_parent);
 640         sd->s_parent = parent_sd;
 641         sysfs_link_sibling(sd);
 642
 643         error = 0;
 644         goto out_unlock;
 645
 646  out_free:
 647         kfree(dup_name);
 648  out_drop:
 649         d_drop(new_dentry);
 650  out_dput:
 651         dput(new_dentry);
 652  out_unlock:
 653         mutex_unlock(&new_parent->d_inode->i_mutex);
 654         up_write(&sysfs_rename_sem);
 655         return error;
 656 }
 657
 658 int sysfs_move_dir(struct kobject *kobj, struct kobject *new_parent)
 659 {
 660         struct dentry *old_parent_dentry, *new_parent_dentry, *new_dentry;
 661         struct sysfs_dirent *new_parent_sd, *sd;
 662         int error;
 663
 664         old_parent_dentry = kobj->parent ?
 665                 kobj->parent->dentry : sysfs_mount->mnt_sb->s_root;
 666         new_parent_dentry = new_parent ?
 667                 new_parent->dentry : sysfs_mount->mnt_sb->s_root;
 668
 669         if (old_parent_dentry->d_inode == new_parent_dentry->d_inode)
 670                 return 0;       /* nothing to move */
 671 again:
 672         mutex_lock(&old_parent_dentry->d_inode->i_mutex);
 673         if (!mutex_trylock(&new_parent_dentry->d_inode->i_mutex)) {
 674                 mutex_unlock(&old_parent_dentry->d_inode->i_mutex);
 675                 goto again;
 676         }
 677
 678         new_parent_sd = new_parent_dentry->d_fsdata;
 679         sd = kobj->dentry->d_fsdata;
 680
 681         new_dentry = lookup_one_len(kobj->name, new_parent_dentry,
 682                                     strlen(kobj->name));
 683         if (IS_ERR(new_dentry)) {
 684                 error = PTR_ERR(new_dentry);
 685                 goto out;
 686         } else
 687                 error = 0;
 688         d_add(new_dentry, NULL);
 689         d_move(kobj->dentry, new_dentry);
 690         dput(new_dentry);
 691
 692         /* Remove from old parent's list and insert into new parent's list. */
 693         sysfs_unlink_sibling(sd);
 694         sysfs_get(new_parent_sd);
 695         sysfs_put(sd->s_parent);
 696         sd->s_parent = new_parent_sd;
 697         sysfs_link_sibling(sd);
 698
 699 out:
 700         mutex_unlock(&new_parent_dentry->d_inode->i_mutex);
 701         mutex_unlock(&old_parent_dentry->d_inode->i_mutex);
 702
 703         return error;
 704 }
 705
 706 static int sysfs_dir_open(struct inode *inode, struct file *file)
 707 {
 708         struct dentry * dentry = file->f_path.dentry;
 709         struct sysfs_dirent * parent_sd = dentry->d_fsdata;
 710         struct sysfs_dirent * sd;
 711
 712         mutex_lock(&dentry->d_inode->i_mutex);
 713         sd = sysfs_new_dirent("_DIR_", 0, 0);
 714         if (sd)
 715                 sysfs_attach_dirent(sd, parent_sd, NULL);
 716         mutex_unlock(&dentry->d_inode->i_mutex);
 717
 718         file->private_data = sd;
 719         return sd ? 0 : -ENOMEM;
 720 }
 721
 722 static int sysfs_dir_close(struct inode *inode, struct file *file)
 723 {
 724         struct dentry * dentry = file->f_path.dentry;
 725         struct sysfs_dirent * cursor = file->private_data;
 726
 727         mutex_lock(&dentry->d_inode->i_mutex);
 728         sysfs_unlink_sibling(cursor);
 729         mutex_unlock(&dentry->d_inode->i_mutex);
 730
 731         release_sysfs_dirent(cursor);
 732
 733         return 0;
 734 }
 735
 736 /* Relationship between s_mode and the DT_xxx types */
 737 static inline unsigned char dt_type(struct sysfs_dirent *sd)
 738 {
 739         return (sd->s_mode >> 12) & 15;
 740 }
 741
 742 static int sysfs_readdir(struct file * filp, void * dirent, filldir_t filldir)
 743 {
 744         struct dentry *dentry = filp->f_path.dentry;
 745         struct sysfs_dirent * parent_sd = dentry->d_fsdata;
 746         struct sysfs_dirent *cursor = filp->private_data;
 747         struct sysfs_dirent **pos;
 748         ino_t ino;
 749         int i = filp->f_pos;
 750
 751         switch (i) {
 752                 case 0:
 753                         ino = parent_sd->s_ino;
 754                         if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
 755                                 break;
 756                         filp->f_pos++;
 757                         i++;
 758                         /* fallthrough */
 759                 case 1:
 760                         if (parent_sd->s_parent)
 761                                 ino = parent_sd->s_parent->s_ino;
 762                         else
 763                                 ino = parent_sd->s_ino;
 764                         if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
 765                                 break;
 766                         filp->f_pos++;
 767                         i++;
 768                         /* fallthrough */
 769                 default:
 770                         pos = &parent_sd->s_children;
 771                         while (*pos != cursor)
 772                                 pos = &(*pos)->s_sibling;
 773
 774                         /* unlink cursor */
 775                         *pos = cursor->s_sibling;
 776
 777                         if (filp->f_pos == 2)
 778                                 pos = &parent_sd->s_children;
 779
 780                         for ( ; *pos; pos = &(*pos)->s_sibling) {
 781                                 struct sysfs_dirent *next = *pos;
 782                                 const char * name;
 783                                 int len;
 784
 785                                 if (!sysfs_type(next))
 786                                         continue;
 787
 788                                 name = next->s_name;
 789                                 len = strlen(name);
 790                                 ino = next->s_ino;
 791
 792                                 if (filldir(dirent, name, len, filp->f_pos, ino,
 793                                                  dt_type(next)) < 0)
 794                                         break;
 795
 796                                 filp->f_pos++;
 797                         }
 798
 799                         /* put cursor back in */
 800                         cursor->s_sibling = *pos;
 801                         *pos = cursor;
 802         }
 803         return 0;
 804 }
 805
 806 static loff_t sysfs_dir_lseek(struct file * file, loff_t offset, int origin)
 807 {
 808         struct dentry * dentry = file->f_path.dentry;
 809
 810         mutex_lock(&dentry->d_inode->i_mutex);
 811         switch (origin) {
 812                 case 1:
 813                         offset += file->f_pos;
 814                 case 0:
 815                         if (offset >= 0)
 816                                 break;
 817                 default:
 818                         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 819                         return -EINVAL;
 820         }
 821         if (offset != file->f_pos) {
 822                 file->f_pos = offset;
 823                 if (file->f_pos >= 2) {
 824                         struct sysfs_dirent *sd = dentry->d_fsdata;
 825                         struct sysfs_dirent *cursor = file->private_data;
 826                         struct sysfs_dirent **pos;
 827                         loff_t n = file->f_pos - 2;
 828
 829                         sysfs_unlink_sibling(cursor);
 830
 831                         pos = &sd->s_children;
 832                         while (n && *pos) {
 833                                 struct sysfs_dirent *next = *pos;
 834                                 if (sysfs_type(next))
 835                                         n--;
 836                                 pos = &(*pos)->s_sibling;
 837                         }
 838
 839                         cursor->s_sibling = *pos;
 840                         *pos = cursor;
 841                 }
 842         }
 843         mutex_unlock(&dentry->d_inode->i_mutex);
 844         return offset;
 845 }
 846
 847
 848 /**
 849  *      sysfs_make_shadowed_dir - Setup so a directory can be shadowed
 850  *      @kobj:  object we're creating shadow of.
 851  */
 852
 853 int sysfs_make_shadowed_dir(struct kobject *kobj,
 854         void * (*follow_link)(struct dentry *, struct nameidata *))
 855 {
 856         struct inode *inode;
 857         struct inode_operations *i_op;
 858
 859         inode = kobj->dentry->d_inode;
 860         if (inode->i_op != &sysfs_dir_inode_operations)
 861                 return -EINVAL;
 862
 863         i_op = kmalloc(sizeof(*i_op), GFP_KERNEL);
 864         if (!i_op)
 865                 return -ENOMEM;
 866
 867         memcpy(i_op, &sysfs_dir_inode_operations, sizeof(*i_op));
 868         i_op->follow_link = follow_link;
 869
 870         /* Locking of inode->i_op?
 871          * Since setting i_op is a single word write and they
 872          * are atomic we should be ok here.
 873          */
 874         inode->i_op = i_op;
 875         return 0;
 876 }
 877
 878 /**
 879  *      sysfs_create_shadow_dir - create a shadow directory for an object.
 880  *      @kobj:  object we're creating directory for.
 881  *
 882  *      sysfs_make_shadowed_dir must already have been called on this
 883  *      directory.
 884  */
 885
 886 struct dentry *sysfs_create_shadow_dir(struct kobject *kobj)
 887 {
 888         struct dentry *dir = kobj->dentry;
 889         struct inode *inode = dir->d_inode;
 890         struct dentry *parent = dir->d_parent;
 891         struct sysfs_dirent *parent_sd = parent->d_fsdata;
 892         struct dentry *shadow;
 893         struct sysfs_dirent *sd;
 894
 895         shadow = ERR_PTR(-EINVAL);
 896         if (!sysfs_is_shadowed_inode(inode))
 897                 goto out;
 898
 899         shadow = d_alloc(parent, &dir->d_name);
 900         if (!shadow)
 901                 goto nomem;
 902
 903         sd = sysfs_new_dirent("_SHADOW_", inode->i_mode, SYSFS_DIR);
 904         if (!sd)
 905                 goto nomem;
 906         sd->s_elem.dir.kobj = kobj;
 907         /* point to parent_sd but don't attach to it */
 908         sd->s_parent = sysfs_get(parent_sd);
 909         sysfs_attach_dirent(sd, NULL, shadow);
 910
 911         d_instantiate(shadow, igrab(inode));
 912         inc_nlink(inode);
 913         inc_nlink(parent->d_inode);
 914
 915         dget(shadow);           /* Extra count - pin the dentry in core */
 916
 917 out:
 918         return shadow;
 919 nomem:
 920         dput(shadow);
 921         shadow = ERR_PTR(-ENOMEM);
 922         goto out;
 923 }
 924
 925 /**
 926  *      sysfs_remove_shadow_dir - remove an object's directory.
 927  *      @shadow: dentry of shadow directory
 928  *
 929  *      The only thing special about this is that we remove any files in
 930  *      the directory before we remove the directory, and we've inlined
 931  *      what used to be sysfs_rmdir() below, instead of calling separately.
 932  */
 933
 934 void sysfs_remove_shadow_dir(struct dentry *shadow)
 935 {
 936         __sysfs_remove_dir(shadow);
 937 }
 938
 939 const struct file_operations sysfs_dir_operations = {
 940         .open           = sysfs_dir_open,
 941         .release        = sysfs_dir_close,
 942         .llseek         = sysfs_dir_lseek,
 943         .read           = generic_read_dir,
 944         .readdir        = sysfs_readdir,
 945 };