lib/idr.c

   1 #include <linux/bitmap.h>
   2 #include <linux/export.h>
   3 #include <linux/idr.h>
   4 #include <linux/slab.h>
   5 #include <linux/spinlock.h>
   6
   7 DEFINE_PER_CPU(struct ida_bitmap *, ida_bitmap);
   8 static DEFINE_SPINLOCK(simple_ida_lock);
   9
  10 int idr_alloc_cmn(struct idr *idr, void *ptr, unsigned long *index,
  11                   unsigned long start, unsigned long end, gfp_t gfp,
  12                   bool ext)
  13 {
  14         struct radix_tree_iter iter;
  15         void __rcu **slot;
  16
  17         if (WARN_ON_ONCE(radix_tree_is_internal_node(ptr)))
  18                 return -EINVAL;
  19
  20         radix_tree_iter_init(&iter, start);
  21         if (ext)
  22                 slot = idr_get_free_ext(&idr->idr_rt, &iter, gfp, end);
  23         else
  24                 slot = idr_get_free(&idr->idr_rt, &iter, gfp, end);
  25         if (IS_ERR(slot))
  26                 return PTR_ERR(slot);
  27
  28         radix_tree_iter_replace(&idr->idr_rt, &iter, slot, ptr);
  29         radix_tree_iter_tag_clear(&idr->idr_rt, &iter, IDR_FREE);
  30
  31         if (index)
  32                 *index = iter.index;
  33         return 0;
  34 }
  35 EXPORT_SYMBOL_GPL(idr_alloc_cmn);
  36
  37 /**
  38  * idr_alloc_cyclic - allocate new idr entry in a cyclical fashion
  39  * @idr: idr handle
  40  * @ptr: pointer to be associated with the new id
  41  * @start: the minimum id (inclusive)
  42  * @end: the maximum id (exclusive)
  43  * @gfp: memory allocation flags
  44  *
  45  * Allocates an ID larger than the last ID allocated if one is available.
  46  * If not, it will attempt to allocate the smallest ID that is larger or
  47  * equal to @start.
  48  */
  49 int idr_alloc_cyclic(struct idr *idr, void *ptr, int start, int end, gfp_t gfp)
  50 {
  51         int id, curr = idr->idr_next;
  52
  53         if (curr < start)
  54                 curr = start;
  55
  56         id = idr_alloc(idr, ptr, curr, end, gfp);
  57         if ((id == -ENOSPC) && (curr > start))
  58                 id = idr_alloc(idr, ptr, start, curr, gfp);
  59
  60         if (id >= 0)
  61                 idr->idr_next = id + 1U;
  62
  63         return id;
  64 }
  65 EXPORT_SYMBOL(idr_alloc_cyclic);
  66
  67 /**
  68  * idr_for_each - iterate through all stored pointers
  69  * @idr: idr handle
  70  * @fn: function to be called for each pointer
  71  * @data: data passed to callback function
  72  *
  73  * The callback function will be called for each entry in @idr, passing
  74  * the id, the pointer and the data pointer passed to this function.
  75  *
  76  * If @fn returns anything other than %0, the iteration stops and that
  77  * value is returned from this function.
  78  *
  79  * idr_for_each() can be called concurrently with idr_alloc() and
  80  * idr_remove() if protected by RCU.  Newly added entries may not be
  81  * seen and deleted entries may be seen, but adding and removing entries
  82  * will not cause other entries to be skipped, nor spurious ones to be seen.
  83  */
  84 int idr_for_each(const struct idr *idr,
  85                 int (*fn)(int id, void *p, void *data), void *data)
  86 {
  87         struct radix_tree_iter iter;
  88         void __rcu **slot;
  89
  90         radix_tree_for_each_slot(slot, &idr->idr_rt, &iter, 0) {
  91                 int ret = fn(iter.index, rcu_dereference_raw(*slot), data);
  92                 if (ret)
  93                         return ret;
  94         }
  95
  96         return 0;
  97 }
  98 EXPORT_SYMBOL(idr_for_each);
  99
 100 /**
 101  * idr_get_next - Find next populated entry
 102  * @idr: idr handle
 103  * @nextid: Pointer to lowest possible ID to return
 104  *
 105  * Returns the next populated entry in the tree with an ID greater than
 106  * or equal to the value pointed to by @nextid.  On exit, @nextid is updated
 107  * to the ID of the found value.  To use in a loop, the value pointed to by
 108  * nextid must be incremented by the user.
 109  */
 110 void *idr_get_next(struct idr *idr, int *nextid)
 111 {
 112         struct radix_tree_iter iter;
 113         void __rcu **slot;
 114         void *entry = NULL;
 115
 116         radix_tree_for_each_slot(slot, &idr->idr_rt, &iter, *nextid) {
 117                 entry = rcu_dereference_raw(*slot);
 118                 if (!entry)
 119                         continue;
 120                 if (!radix_tree_deref_retry(entry))
 121                         break;
 122                 if (slot != (void *)&idr->idr_rt.rnode &&
 123                                 entry != (void *)RADIX_TREE_INTERNAL_NODE)
 124                         break;
 125                 slot = radix_tree_iter_retry(&iter);
 126         }
 127         if (!slot)
 128                 return NULL;
 129
 130         if (WARN_ON_ONCE(iter.index > INT_MAX))
 131                 return NULL;
 132
 133         *nextid = iter.index;
 134         return entry;
 135 }
 136 EXPORT_SYMBOL(idr_get_next);
 137
 138 void *idr_get_next_ext(struct idr *idr, unsigned long *nextid)
 139 {
 140         struct radix_tree_iter iter;
 141         void __rcu **slot;
 142
 143         slot = radix_tree_iter_find(&idr->idr_rt, &iter, *nextid);
 144         if (!slot)
 145                 return NULL;
 146
 147         *nextid = iter.index;
 148         return rcu_dereference_raw(*slot);
 149 }
 150 EXPORT_SYMBOL(idr_get_next_ext);
 151
 152 /**
 153  * idr_replace - replace pointer for given id
 154  * @idr: idr handle
 155  * @ptr: New pointer to associate with the ID
 156  * @id: Lookup key
 157  *
 158  * Replace the pointer registered with an ID and return the old value.
 159  * This function can be called under the RCU read lock concurrently with
 160  * idr_alloc() and idr_remove() (as long as the ID being removed is not
 161  * the one being replaced!).
 162  *
 163  * Returns: the old value on success.  %-ENOENT indicates that @id was not
 164  * found.  %-EINVAL indicates that @id or @ptr were not valid.
 165  */
 166 void *idr_replace(struct idr *idr, void *ptr, int id)
 167 {
 168         if (id < 0)
 169                 return ERR_PTR(-EINVAL);
 170
 171         return idr_replace_ext(idr, ptr, id);
 172 }
 173 EXPORT_SYMBOL(idr_replace);
 174
 175 void *idr_replace_ext(struct idr *idr, void *ptr, unsigned long id)
 176 {
 177         struct radix_tree_node *node;
 178         void __rcu **slot = NULL;
 179         void *entry;
 180
 181         if (WARN_ON_ONCE(radix_tree_is_internal_node(ptr)))
 182                 return ERR_PTR(-EINVAL);
 183
 184         entry = __radix_tree_lookup(&idr->idr_rt, id, &node, &slot);
 185         if (!slot || radix_tree_tag_get(&idr->idr_rt, id, IDR_FREE))
 186                 return ERR_PTR(-ENOENT);
 187
 188         __radix_tree_replace(&idr->idr_rt, node, slot, ptr, NULL);
 189
 190         return entry;
 191 }
 192 EXPORT_SYMBOL(idr_replace_ext);
 193
 194 /**
 195  * DOC: IDA description
 196  *
 197  * The IDA is an ID allocator which does not provide the ability to
 198  * associate an ID with a pointer.  As such, it only needs to store one
 199  * bit per ID, and so is more space efficient than an IDR.  To use an IDA,
 200  * define it using DEFINE_IDA() (or embed a &struct ida in a data structure,
 201  * then initialise it using ida_init()).  To allocate a new ID, call
 202  * ida_simple_get().  To free an ID, call ida_simple_remove().
 203  *
 204  * If you have more complex locking requirements, use a loop around
 205  * ida_pre_get() and ida_get_new() to allocate a new ID.  Then use
 206  * ida_remove() to free an ID.  You must make sure that ida_get_new() and
 207  * ida_remove() cannot be called at the same time as each other for the
 208  * same IDA.
 209  *
 210  * You can also use ida_get_new_above() if you need an ID to be allocated
 211  * above a particular number.  ida_destroy() can be used to dispose of an
 212  * IDA without needing to free the individual IDs in it.  You can use
 213  * ida_is_empty() to find out whether the IDA has any IDs currently allocated.
 214  *
 215  * IDs are currently limited to the range [0-INT_MAX].  If this is an awkward
 216  * limitation, it should be quite straightforward to raise the maximum.
 217  */
 218
 219 /*
 220  * Developer's notes:
 221  *
 222  * The IDA uses the functionality provided by the IDR & radix tree to store
 223  * bitmaps in each entry.  The IDR_FREE tag means there is at least one bit
 224  * free, unlike the IDR where it means at least one entry is free.
 225  *
 226  * I considered telling the radix tree that each slot is an order-10 node
 227  * and storing the bit numbers in the radix tree, but the radix tree can't
 228  * allow a single multiorder entry at index 0, which would significantly
 229  * increase memory consumption for the IDA.  So instead we divide the index
 230  * by the number of bits in the leaf bitmap before doing a radix tree lookup.
 231  *
 232  * As an optimisation, if there are only a few low bits set in any given
 233  * leaf, instead of allocating a 128-byte bitmap, we use the 'exceptional
 234  * entry' functionality of the radix tree to store BITS_PER_LONG - 2 bits
 235  * directly in the entry.  By being really tricksy, we could store
 236  * BITS_PER_LONG - 1 bits, but there're diminishing returns after optimising
 237  * for 0-3 allocated IDs.
 238  *
 239  * We allow the radix tree 'exceptional' count to get out of date.  Nothing
 240  * in the IDA nor the radix tree code checks it.  If it becomes important
 241  * to maintain an accurate exceptional count, switch the rcu_assign_pointer()
 242  * calls to radix_tree_iter_replace() which will correct the exceptional
 243  * count.
 244  *
 245  * The IDA always requires a lock to alloc/free.  If we add a 'test_bit'
 246  * equivalent, it will still need locking.  Going to RCU lookup would require
 247  * using RCU to free bitmaps, and that's not trivial without embedding an
 248  * RCU head in the bitmap, which adds a 2-pointer overhead to each 128-byte
 249  * bitmap, which is excessive.
 250  */
 251
 252 #define IDA_MAX (0x80000000U / IDA_BITMAP_BITS)
 253
 254 /**
 255  * ida_get_new_above - allocate new ID above or equal to a start id
 256  * @ida: ida handle
 257  * @start: id to start search at
 258  * @id: pointer to the allocated handle
 259  *
 260  * Allocate new ID above or equal to @start.  It should be called
 261  * with any required locks to ensure that concurrent calls to
 262  * ida_get_new_above() / ida_get_new() / ida_remove() are not allowed.
 263  * Consider using ida_simple_get() if you do not have complex locking
 264  * requirements.
 265  *
 266  * If memory is required, it will return %-EAGAIN, you should unlock
 267  * and go back to the ida_pre_get() call.  If the ida is full, it will
 268  * return %-ENOSPC.  On success, it will return 0.
 269  *
 270  * @id returns a value in the range @start ... %0x7fffffff.
 271  */
 272 int ida_get_new_above(struct ida *ida, int start, int *id)
 273 {
 274         struct radix_tree_root *root = &ida->ida_rt;
 275         void __rcu **slot;
 276         struct radix_tree_iter iter;
 277         struct ida_bitmap *bitmap;
 278         unsigned long index;
 279         unsigned bit, ebit;
 280         int new;
 281
 282         index = start / IDA_BITMAP_BITS;
 283         bit = start % IDA_BITMAP_BITS;
 284         ebit = bit + RADIX_TREE_EXCEPTIONAL_SHIFT;
 285
 286         slot = radix_tree_iter_init(&iter, index);
 287         for (;;) {
 288                 if (slot)
 289                         slot = radix_tree_next_slot(slot, &iter,
 290                                                 RADIX_TREE_ITER_TAGGED);
 291                 if (!slot) {
 292                         slot = idr_get_free(root, &iter, GFP_NOWAIT, IDA_MAX);
 293                         if (IS_ERR(slot)) {
 294                                 if (slot == ERR_PTR(-ENOMEM))
 295                                         return -EAGAIN;
 296                                 return PTR_ERR(slot);
 297                         }
 298                 }
 299                 if (iter.index > index) {
 300                         bit = 0;
 301                         ebit = RADIX_TREE_EXCEPTIONAL_SHIFT;
 302                 }
 303                 new = iter.index * IDA_BITMAP_BITS;
 304                 bitmap = rcu_dereference_raw(*slot);
 305                 if (radix_tree_exception(bitmap)) {
 306                         unsigned long tmp = (unsigned long)bitmap;
 307                         ebit = find_next_zero_bit(&tmp, BITS_PER_LONG, ebit);
 308                         if (ebit < BITS_PER_LONG) {
 309                                 tmp |= 1UL << ebit;
 310                                 rcu_assign_pointer(*slot, (void *)tmp);
 311                                 *id = new + ebit - RADIX_TREE_EXCEPTIONAL_SHIFT;
 312                                 return 0;
 313                         }
 314                         bitmap = this_cpu_xchg(ida_bitmap, NULL);
 315                         if (!bitmap)
 316                                 return -EAGAIN;
 317                         memset(bitmap, 0, sizeof(*bitmap));
 318                         bitmap->bitmap[0] = tmp >> RADIX_TREE_EXCEPTIONAL_SHIFT;
 319                         rcu_assign_pointer(*slot, bitmap);
 320                 }
 321
 322                 if (bitmap) {
 323                         bit = find_next_zero_bit(bitmap->bitmap,
 324                                                         IDA_BITMAP_BITS, bit);
 325                         new += bit;
 326                         if (new < 0)
 327                                 return -ENOSPC;
 328                         if (bit == IDA_BITMAP_BITS)
 329                                 continue;
 330
 331                         __set_bit(bit, bitmap->bitmap);
 332                         if (bitmap_full(bitmap->bitmap, IDA_BITMAP_BITS))
 333                                 radix_tree_iter_tag_clear(root, &iter,
 334                                                                 IDR_FREE);
 335                 } else {
 336                         new += bit;
 337                         if (new < 0)
 338                                 return -ENOSPC;
 339                         if (ebit < BITS_PER_LONG) {
 340                                 bitmap = (void *)((1UL << ebit) |
 341                                                 RADIX_TREE_EXCEPTIONAL_ENTRY);
 342                                 radix_tree_iter_replace(root, &iter, slot,
 343                                                 bitmap);
 344                                 *id = new;
 345                                 return 0;
 346                         }
 347                         bitmap = this_cpu_xchg(ida_bitmap, NULL);
 348                         if (!bitmap)
 349                                 return -EAGAIN;
 350                         memset(bitmap, 0, sizeof(*bitmap));
 351                         __set_bit(bit, bitmap->bitmap);
 352                         radix_tree_iter_replace(root, &iter, slot, bitmap);
 353                 }
 354
 355                 *id = new;
 356                 return 0;
 357         }
 358 }
 359 EXPORT_SYMBOL(ida_get_new_above);
 360
 361 /**
 362  * ida_remove - Free the given ID
 363  * @ida: ida handle
 364  * @id: ID to free
 365  *
 366  * This function should not be called at the same time as ida_get_new_above().
 367  */
 368 void ida_remove(struct ida *ida, int id)
 369 {
 370         unsigned long index = id / IDA_BITMAP_BITS;
 371         unsigned offset = id % IDA_BITMAP_BITS;
 372         struct ida_bitmap *bitmap;
 373         unsigned long *btmp;
 374         struct radix_tree_iter iter;
 375         void __rcu **slot;
 376
 377         slot = radix_tree_iter_lookup(&ida->ida_rt, &iter, index);
 378         if (!slot)
 379                 goto err;
 380
 381         bitmap = rcu_dereference_raw(*slot);
 382         if (radix_tree_exception(bitmap)) {
 383                 btmp = (unsigned long *)slot;
 384                 offset += RADIX_TREE_EXCEPTIONAL_SHIFT;
 385                 if (offset >= BITS_PER_LONG)
 386                         goto err;
 387         } else {
 388                 btmp = bitmap->bitmap;
 389         }
 390         if (!test_bit(offset, btmp))
 391                 goto err;
 392
 393         __clear_bit(offset, btmp);
 394         radix_tree_iter_tag_set(&ida->ida_rt, &iter, IDR_FREE);
 395         if (radix_tree_exception(bitmap)) {
 396                 if (rcu_dereference_raw(*slot) ==
 397                                         (void *)RADIX_TREE_EXCEPTIONAL_ENTRY)
 398                         radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
 399         } else if (bitmap_empty(btmp, IDA_BITMAP_BITS)) {
 400                 kfree(bitmap);
 401                 radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
 402         }
 403         return;
 404  err:
 405         WARN(1, "ida_remove called for id=%d which is not allocated.\n", id);
 406 }
 407 EXPORT_SYMBOL(ida_remove);
 408
 409 /**
 410  * ida_destroy - Free the contents of an ida
 411  * @ida: ida handle
 412  *
 413  * Calling this function releases all resources associated with an IDA.  When
 414  * this call returns, the IDA is empty and can be reused or freed.  The caller
 415  * should not allow ida_remove() or ida_get_new_above() to be called at the
 416  * same time.
 417  */
 418 void ida_destroy(struct ida *ida)
 419 {
 420         struct radix_tree_iter iter;
 421         void __rcu **slot;
 422
 423         radix_tree_for_each_slot(slot, &ida->ida_rt, &iter, 0) {
 424                 struct ida_bitmap *bitmap = rcu_dereference_raw(*slot);
 425                 if (!radix_tree_exception(bitmap))
 426                         kfree(bitmap);
 427                 radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
 428         }
 429 }
 430 EXPORT_SYMBOL(ida_destroy);
 431
 432 /**
 433  * ida_simple_get - get a new id.
 434  * @ida: the (initialized) ida.
 435  * @start: the minimum id (inclusive, < 0x8000000)
 436  * @end: the maximum id (exclusive, < 0x8000000 or 0)
 437  * @gfp_mask: memory allocation flags
 438  *
 439  * Allocates an id in the range start <= id < end, or returns -ENOSPC.
 440  * On memory allocation failure, returns -ENOMEM.
 441  *
 442  * Compared to ida_get_new_above() this function does its own locking, and
 443  * should be used unless there are special requirements.
 444  *
 445  * Use ida_simple_remove() to get rid of an id.
 446  */
 447 int ida_simple_get(struct ida *ida, unsigned int start, unsigned int end,
 448                    gfp_t gfp_mask)
 449 {
 450         int ret, id;
 451         unsigned int max;
 452         unsigned long flags;
 453
 454         BUG_ON((int)start < 0);
 455         BUG_ON((int)end < 0);
 456
 457         if (end == 0)
 458                 max = 0x80000000;
 459         else {
 460                 BUG_ON(end < start);
 461                 max = end - 1;
 462         }
 463
 464 again:
 465         if (!ida_pre_get(ida, gfp_mask))
 466                 return -ENOMEM;
 467
 468         spin_lock_irqsave(&simple_ida_lock, flags);
 469         ret = ida_get_new_above(ida, start, &id);
 470         if (!ret) {
 471                 if (id > max) {
 472                         ida_remove(ida, id);
 473                         ret = -ENOSPC;
 474                 } else {
 475                         ret = id;
 476                 }
 477         }
 478         spin_unlock_irqrestore(&simple_ida_lock, flags);
 479
 480         if (unlikely(ret == -EAGAIN))
 481                 goto again;
 482
 483         return ret;
 484 }
 485 EXPORT_SYMBOL(ida_simple_get);
 486
 487 /**
 488  * ida_simple_remove - remove an allocated id.
 489  * @ida: the (initialized) ida.
 490  * @id: the id returned by ida_simple_get.
 491  *
 492  * Use to release an id allocated with ida_simple_get().
 493  *
 494  * Compared to ida_remove() this function does its own locking, and should be
 495  * used unless there are special requirements.
 496  */
 497 void ida_simple_remove(struct ida *ida, unsigned int id)
 498 {
 499         unsigned long flags;
 500
 501         BUG_ON((int)id < 0);
 502         spin_lock_irqsave(&simple_ida_lock, flags);
 503         ida_remove(ida, id);
 504         spin_unlock_irqrestore(&simple_ida_lock, flags);
 505 }
 506 EXPORT_SYMBOL(ida_simple_remove);