lib/idr.c

   1 #include <linux/bitmap.h>
   2 #include <linux/bug.h>
   3 #include <linux/export.h>
   4 #include <linux/idr.h>
   5 #include <linux/slab.h>
   6 #include <linux/spinlock.h>
   7 #include <linux/xarray.h>
   8
   9 DEFINE_PER_CPU(struct ida_bitmap *, ida_bitmap);
  10
  11 /**
  12  * idr_alloc_u32() - Allocate an ID.
  13  * @idr: IDR handle.
  14  * @ptr: Pointer to be associated with the new ID.
  15  * @nextid: Pointer to an ID.
  16  * @max: The maximum ID to allocate (inclusive).
  17  * @gfp: Memory allocation flags.
  18  *
  19  * Allocates an unused ID in the range specified by @nextid and @max.
  20  * Note that @max is inclusive whereas the @end parameter to idr_alloc()
  21  * is exclusive.  The new ID is assigned to @nextid before the pointer
  22  * is inserted into the IDR, so if @nextid points into the object pointed
  23  * to by @ptr, a concurrent lookup will not find an uninitialised ID.
  24  *
  25  * The caller should provide their own locking to ensure that two
  26  * concurrent modifications to the IDR are not possible.  Read-only
  27  * accesses to the IDR may be done under the RCU read lock or may
  28  * exclude simultaneous writers.
  29  *
  30  * Return: 0 if an ID was allocated, -ENOMEM if memory allocation failed,
  31  * or -ENOSPC if no free IDs could be found.  If an error occurred,
  32  * @nextid is unchanged.
  33  */
  34 int idr_alloc_u32(struct idr *idr, void *ptr, u32 *nextid,
  35                         unsigned long max, gfp_t gfp)
  36 {
  37         struct radix_tree_iter iter;
  38         void __rcu **slot;
  39         unsigned int base = idr->idr_base;
  40         unsigned int id = *nextid;
  41
  42         if (WARN_ON_ONCE(radix_tree_is_internal_node(ptr)))
  43                 return -EINVAL;
  44         if (WARN_ON_ONCE(!(idr->idr_rt.gfp_mask & ROOT_IS_IDR)))
  45                 idr->idr_rt.gfp_mask |= IDR_RT_MARKER;
  46
  47         id = (id < base) ? 0 : id - base;
  48         radix_tree_iter_init(&iter, id);
  49         slot = idr_get_free(&idr->idr_rt, &iter, gfp, max - base);
  50         if (IS_ERR(slot))
  51                 return PTR_ERR(slot);
  52
  53         *nextid = iter.index + base;
  54         /* there is a memory barrier inside radix_tree_iter_replace() */
  55         radix_tree_iter_replace(&idr->idr_rt, &iter, slot, ptr);
  56         radix_tree_iter_tag_clear(&idr->idr_rt, &iter, IDR_FREE);
  57
  58         return 0;
  59 }
  60 EXPORT_SYMBOL_GPL(idr_alloc_u32);
  61
  62 /**
  63  * idr_alloc() - Allocate an ID.
  64  * @idr: IDR handle.
  65  * @ptr: Pointer to be associated with the new ID.
  66  * @start: The minimum ID (inclusive).
  67  * @end: The maximum ID (exclusive).
  68  * @gfp: Memory allocation flags.
  69  *
  70  * Allocates an unused ID in the range specified by @start and @end.  If
  71  * @end is <= 0, it is treated as one larger than %INT_MAX.  This allows
  72  * callers to use @start + N as @end as long as N is within integer range.
  73  *
  74  * The caller should provide their own locking to ensure that two
  75  * concurrent modifications to the IDR are not possible.  Read-only
  76  * accesses to the IDR may be done under the RCU read lock or may
  77  * exclude simultaneous writers.
  78  *
  79  * Return: The newly allocated ID, -ENOMEM if memory allocation failed,
  80  * or -ENOSPC if no free IDs could be found.
  81  */
  82 int idr_alloc(struct idr *idr, void *ptr, int start, int end, gfp_t gfp)
  83 {
  84         u32 id = start;
  85         int ret;
  86
  87         if (WARN_ON_ONCE(start < 0))
  88                 return -EINVAL;
  89
  90         ret = idr_alloc_u32(idr, ptr, &id, end > 0 ? end - 1 : INT_MAX, gfp);
  91         if (ret)
  92                 return ret;
  93
  94         return id;
  95 }
  96 EXPORT_SYMBOL_GPL(idr_alloc);
  97
  98 /**
  99  * idr_alloc_cyclic() - Allocate an ID cyclically.
 100  * @idr: IDR handle.
 101  * @ptr: Pointer to be associated with the new ID.
 102  * @start: The minimum ID (inclusive).
 103  * @end: The maximum ID (exclusive).
 104  * @gfp: Memory allocation flags.
 105  *
 106  * Allocates an unused ID in the range specified by @nextid and @end.  If
 107  * @end is <= 0, it is treated as one larger than %INT_MAX.  This allows
 108  * callers to use @start + N as @end as long as N is within integer range.
 109  * The search for an unused ID will start at the last ID allocated and will
 110  * wrap around to @start if no free IDs are found before reaching @end.
 111  *
 112  * The caller should provide their own locking to ensure that two
 113  * concurrent modifications to the IDR are not possible.  Read-only
 114  * accesses to the IDR may be done under the RCU read lock or may
 115  * exclude simultaneous writers.
 116  *
 117  * Return: The newly allocated ID, -ENOMEM if memory allocation failed,
 118  * or -ENOSPC if no free IDs could be found.
 119  */
 120 int idr_alloc_cyclic(struct idr *idr, void *ptr, int start, int end, gfp_t gfp)
 121 {
 122         u32 id = idr->idr_next;
 123         int err, max = end > 0 ? end - 1 : INT_MAX;
 124
 125         if ((int)id < start)
 126                 id = start;
 127
 128         err = idr_alloc_u32(idr, ptr, &id, max, gfp);
 129         if ((err == -ENOSPC) && (id > start)) {
 130                 id = start;
 131                 err = idr_alloc_u32(idr, ptr, &id, max, gfp);
 132         }
 133         if (err)
 134                 return err;
 135
 136         idr->idr_next = id + 1;
 137         return id;
 138 }
 139 EXPORT_SYMBOL(idr_alloc_cyclic);
 140
 141 /**
 142  * idr_remove() - Remove an ID from the IDR.
 143  * @idr: IDR handle.
 144  * @id: Pointer ID.
 145  *
 146  * Removes this ID from the IDR.  If the ID was not previously in the IDR,
 147  * this function returns %NULL.
 148  *
 149  * Since this function modifies the IDR, the caller should provide their
 150  * own locking to ensure that concurrent modification of the same IDR is
 151  * not possible.
 152  *
 153  * Return: The pointer formerly associated with this ID.
 154  */
 155 void *idr_remove(struct idr *idr, unsigned long id)
 156 {
 157         return radix_tree_delete_item(&idr->idr_rt, id - idr->idr_base, NULL);
 158 }
 159 EXPORT_SYMBOL_GPL(idr_remove);
 160
 161 /**
 162  * idr_find() - Return pointer for given ID.
 163  * @idr: IDR handle.
 164  * @id: Pointer ID.
 165  *
 166  * Looks up the pointer associated with this ID.  A %NULL pointer may
 167  * indicate that @id is not allocated or that the %NULL pointer was
 168  * associated with this ID.
 169  *
 170  * This function can be called under rcu_read_lock(), given that the leaf
 171  * pointers lifetimes are correctly managed.
 172  *
 173  * Return: The pointer associated with this ID.
 174  */
 175 void *idr_find(const struct idr *idr, unsigned long id)
 176 {
 177         return radix_tree_lookup(&idr->idr_rt, id - idr->idr_base);
 178 }
 179 EXPORT_SYMBOL_GPL(idr_find);
 180
 181 /**
 182  * idr_for_each() - Iterate through all stored pointers.
 183  * @idr: IDR handle.
 184  * @fn: Function to be called for each pointer.
 185  * @data: Data passed to callback function.
 186  *
 187  * The callback function will be called for each entry in @idr, passing
 188  * the ID, the entry and @data.
 189  *
 190  * If @fn returns anything other than %0, the iteration stops and that
 191  * value is returned from this function.
 192  *
 193  * idr_for_each() can be called concurrently with idr_alloc() and
 194  * idr_remove() if protected by RCU.  Newly added entries may not be
 195  * seen and deleted entries may be seen, but adding and removing entries
 196  * will not cause other entries to be skipped, nor spurious ones to be seen.
 197  */
 198 int idr_for_each(const struct idr *idr,
 199                 int (*fn)(int id, void *p, void *data), void *data)
 200 {
 201         struct radix_tree_iter iter;
 202         void __rcu **slot;
 203         int base = idr->idr_base;
 204
 205         radix_tree_for_each_slot(slot, &idr->idr_rt, &iter, 0) {
 206                 int ret;
 207                 unsigned long id = iter.index + base;
 208
 209                 if (WARN_ON_ONCE(id > INT_MAX))
 210                         break;
 211                 ret = fn(id, rcu_dereference_raw(*slot), data);
 212                 if (ret)
 213                         return ret;
 214         }
 215
 216         return 0;
 217 }
 218 EXPORT_SYMBOL(idr_for_each);
 219
 220 /**
 221  * idr_get_next_ul() - Find next populated entry.
 222  * @idr: IDR handle.
 223  * @nextid: Pointer to an ID.
 224  *
 225  * Returns the next populated entry in the tree with an ID greater than
 226  * or equal to the value pointed to by @nextid.  On exit, @nextid is updated
 227  * to the ID of the found value.  To use in a loop, the value pointed to by
 228  * nextid must be incremented by the user.
 229  */
 230 void *idr_get_next_ul(struct idr *idr, unsigned long *nextid)
 231 {
 232         struct radix_tree_iter iter;
 233         void __rcu **slot;
 234         void *entry = NULL;
 235         unsigned long base = idr->idr_base;
 236         unsigned long id = *nextid;
 237
 238         id = (id < base) ? 0 : id - base;
 239         radix_tree_for_each_slot(slot, &idr->idr_rt, &iter, id) {
 240                 entry = rcu_dereference_raw(*slot);
 241                 if (!entry)
 242                         continue;
 243                 if (!radix_tree_deref_retry(entry))
 244                         break;
 245                 if (slot != (void *)&idr->idr_rt.rnode &&
 246                                 entry != (void *)RADIX_TREE_INTERNAL_NODE)
 247                         break;
 248                 slot = radix_tree_iter_retry(&iter);
 249         }
 250         if (!slot)
 251                 return NULL;
 252
 253         *nextid = iter.index + base;
 254         return entry;
 255 }
 256 EXPORT_SYMBOL(idr_get_next_ul);
 257
 258 /**
 259  * idr_get_next() - Find next populated entry.
 260  * @idr: IDR handle.
 261  * @nextid: Pointer to an ID.
 262  *
 263  * Returns the next populated entry in the tree with an ID greater than
 264  * or equal to the value pointed to by @nextid.  On exit, @nextid is updated
 265  * to the ID of the found value.  To use in a loop, the value pointed to by
 266  * nextid must be incremented by the user.
 267  */
 268 void *idr_get_next(struct idr *idr, int *nextid)
 269 {
 270         unsigned long id = *nextid;
 271         void *entry = idr_get_next_ul(idr, &id);
 272
 273         if (WARN_ON_ONCE(id > INT_MAX))
 274                 return NULL;
 275         *nextid = id;
 276         return entry;
 277 }
 278 EXPORT_SYMBOL(idr_get_next);
 279
 280 /**
 281  * idr_replace() - replace pointer for given ID.
 282  * @idr: IDR handle.
 283  * @ptr: New pointer to associate with the ID.
 284  * @id: ID to change.
 285  *
 286  * Replace the pointer registered with an ID and return the old value.
 287  * This function can be called under the RCU read lock concurrently with
 288  * idr_alloc() and idr_remove() (as long as the ID being removed is not
 289  * the one being replaced!).
 290  *
 291  * Returns: the old value on success.  %-ENOENT indicates that @id was not
 292  * found.  %-EINVAL indicates that @ptr was not valid.
 293  */
 294 void *idr_replace(struct idr *idr, void *ptr, unsigned long id)
 295 {
 296         struct radix_tree_node *node;
 297         void __rcu **slot = NULL;
 298         void *entry;
 299
 300         if (WARN_ON_ONCE(radix_tree_is_internal_node(ptr)))
 301                 return ERR_PTR(-EINVAL);
 302         id -= idr->idr_base;
 303
 304         entry = __radix_tree_lookup(&idr->idr_rt, id, &node, &slot);
 305         if (!slot || radix_tree_tag_get(&idr->idr_rt, id, IDR_FREE))
 306                 return ERR_PTR(-ENOENT);
 307
 308         __radix_tree_replace(&idr->idr_rt, node, slot, ptr, NULL);
 309
 310         return entry;
 311 }
 312 EXPORT_SYMBOL(idr_replace);
 313
 314 /**
 315  * DOC: IDA description
 316  *
 317  * The IDA is an ID allocator which does not provide the ability to
 318  * associate an ID with a pointer.  As such, it only needs to store one
 319  * bit per ID, and so is more space efficient than an IDR.  To use an IDA,
 320  * define it using DEFINE_IDA() (or embed a &struct ida in a data structure,
 321  * then initialise it using ida_init()).  To allocate a new ID, call
 322  * ida_alloc(), ida_alloc_min(), ida_alloc_max() or ida_alloc_range().
 323  * To free an ID, call ida_free().
 324  *
 325  * ida_destroy() can be used to dispose of an IDA without needing to
 326  * free the individual IDs in it.  You can use ida_is_empty() to find
 327  * out whether the IDA has any IDs currently allocated.
 328  *
 329  * IDs are currently limited to the range [0-INT_MAX].  If this is an awkward
 330  * limitation, it should be quite straightforward to raise the maximum.
 331  */
 332
 333 /*
 334  * Developer's notes:
 335  *
 336  * The IDA uses the functionality provided by the IDR & radix tree to store
 337  * bitmaps in each entry.  The IDR_FREE tag means there is at least one bit
 338  * free, unlike the IDR where it means at least one entry is free.
 339  *
 340  * I considered telling the radix tree that each slot is an order-10 node
 341  * and storing the bit numbers in the radix tree, but the radix tree can't
 342  * allow a single multiorder entry at index 0, which would significantly
 343  * increase memory consumption for the IDA.  So instead we divide the index
 344  * by the number of bits in the leaf bitmap before doing a radix tree lookup.
 345  *
 346  * As an optimisation, if there are only a few low bits set in any given
 347  * leaf, instead of allocating a 128-byte bitmap, we use the 'exceptional
 348  * entry' functionality of the radix tree to store BITS_PER_LONG - 2 bits
 349  * directly in the entry.  By being really tricksy, we could store
 350  * BITS_PER_LONG - 1 bits, but there're diminishing returns after optimising
 351  * for 0-3 allocated IDs.
 352  *
 353  * We allow the radix tree 'exceptional' count to get out of date.  Nothing
 354  * in the IDA nor the radix tree code checks it.  If it becomes important
 355  * to maintain an accurate exceptional count, switch the rcu_assign_pointer()
 356  * calls to radix_tree_iter_replace() which will correct the exceptional
 357  * count.
 358  *
 359  * The IDA always requires a lock to alloc/free.  If we add a 'test_bit'
 360  * equivalent, it will still need locking.  Going to RCU lookup would require
 361  * using RCU to free bitmaps, and that's not trivial without embedding an
 362  * RCU head in the bitmap, which adds a 2-pointer overhead to each 128-byte
 363  * bitmap, which is excessive.
 364  */
 365
 366 #define IDA_MAX (0x80000000U / IDA_BITMAP_BITS - 1)
 367
 368 static int ida_get_new_above(struct ida *ida, int start)
 369 {
 370         struct radix_tree_root *root = &ida->ida_rt;
 371         void __rcu **slot;
 372         struct radix_tree_iter iter;
 373         struct ida_bitmap *bitmap;
 374         unsigned long index;
 375         unsigned bit, ebit;
 376         int new;
 377
 378         index = start / IDA_BITMAP_BITS;
 379         bit = start % IDA_BITMAP_BITS;
 380         ebit = bit + RADIX_TREE_EXCEPTIONAL_SHIFT;
 381
 382         slot = radix_tree_iter_init(&iter, index);
 383         for (;;) {
 384                 if (slot)
 385                         slot = radix_tree_next_slot(slot, &iter,
 386                                                 RADIX_TREE_ITER_TAGGED);
 387                 if (!slot) {
 388                         slot = idr_get_free(root, &iter, GFP_NOWAIT, IDA_MAX);
 389                         if (IS_ERR(slot)) {
 390                                 if (slot == ERR_PTR(-ENOMEM))
 391                                         return -EAGAIN;
 392                                 return PTR_ERR(slot);
 393                         }
 394                 }
 395                 if (iter.index > index) {
 396                         bit = 0;
 397                         ebit = RADIX_TREE_EXCEPTIONAL_SHIFT;
 398                 }
 399                 new = iter.index * IDA_BITMAP_BITS;
 400                 bitmap = rcu_dereference_raw(*slot);
 401                 if (radix_tree_exception(bitmap)) {
 402                         unsigned long tmp = (unsigned long)bitmap;
 403                         ebit = find_next_zero_bit(&tmp, BITS_PER_LONG, ebit);
 404                         if (ebit < BITS_PER_LONG) {
 405                                 tmp |= 1UL << ebit;
 406                                 rcu_assign_pointer(*slot, (void *)tmp);
 407                                 return new + ebit -
 408                                         RADIX_TREE_EXCEPTIONAL_SHIFT;
 409                         }
 410                         bitmap = this_cpu_xchg(ida_bitmap, NULL);
 411                         if (!bitmap)
 412                                 return -EAGAIN;
 413                         bitmap->bitmap[0] = tmp >> RADIX_TREE_EXCEPTIONAL_SHIFT;
 414                         rcu_assign_pointer(*slot, bitmap);
 415                 }
 416
 417                 if (bitmap) {
 418                         bit = find_next_zero_bit(bitmap->bitmap,
 419                                                         IDA_BITMAP_BITS, bit);
 420                         new += bit;
 421                         if (new < 0)
 422                                 return -ENOSPC;
 423                         if (bit == IDA_BITMAP_BITS)
 424                                 continue;
 425
 426                         __set_bit(bit, bitmap->bitmap);
 427                         if (bitmap_full(bitmap->bitmap, IDA_BITMAP_BITS))
 428                                 radix_tree_iter_tag_clear(root, &iter,
 429                                                                 IDR_FREE);
 430                 } else {
 431                         new += bit;
 432                         if (new < 0)
 433                                 return -ENOSPC;
 434                         if (ebit < BITS_PER_LONG) {
 435                                 bitmap = (void *)((1UL << ebit) |
 436                                                 RADIX_TREE_EXCEPTIONAL_ENTRY);
 437                                 radix_tree_iter_replace(root, &iter, slot,
 438                                                 bitmap);
 439                                 return new;
 440                         }
 441                         bitmap = this_cpu_xchg(ida_bitmap, NULL);
 442                         if (!bitmap)
 443                                 return -EAGAIN;
 444                         __set_bit(bit, bitmap->bitmap);
 445                         radix_tree_iter_replace(root, &iter, slot, bitmap);
 446                 }
 447
 448                 return new;
 449         }
 450 }
 451
 452 static void ida_remove(struct ida *ida, int id)
 453 {
 454         unsigned long index = id / IDA_BITMAP_BITS;
 455         unsigned offset = id % IDA_BITMAP_BITS;
 456         struct ida_bitmap *bitmap;
 457         unsigned long *btmp;
 458         struct radix_tree_iter iter;
 459         void __rcu **slot;
 460
 461         slot = radix_tree_iter_lookup(&ida->ida_rt, &iter, index);
 462         if (!slot)
 463                 goto err;
 464
 465         bitmap = rcu_dereference_raw(*slot);
 466         if (radix_tree_exception(bitmap)) {
 467                 btmp = (unsigned long *)slot;
 468                 offset += RADIX_TREE_EXCEPTIONAL_SHIFT;
 469                 if (offset >= BITS_PER_LONG)
 470                         goto err;
 471         } else {
 472                 btmp = bitmap->bitmap;
 473         }
 474         if (!test_bit(offset, btmp))
 475                 goto err;
 476
 477         __clear_bit(offset, btmp);
 478         radix_tree_iter_tag_set(&ida->ida_rt, &iter, IDR_FREE);
 479         if (radix_tree_exception(bitmap)) {
 480                 if (rcu_dereference_raw(*slot) ==
 481                                         (void *)RADIX_TREE_EXCEPTIONAL_ENTRY)
 482                         radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
 483         } else if (bitmap_empty(btmp, IDA_BITMAP_BITS)) {
 484                 kfree(bitmap);
 485                 radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
 486         }
 487         return;
 488  err:
 489         WARN(1, "ida_free called for id=%d which is not allocated.\n", id);
 490 }
 491
 492 /**
 493  * ida_destroy() - Free all IDs.
 494  * @ida: IDA handle.
 495  *
 496  * Calling this function frees all IDs and releases all resources used
 497  * by an IDA.  When this call returns, the IDA is empty and can be reused
 498  * or freed.  If the IDA is already empty, there is no need to call this
 499  * function.
 500  *
 501  * Context: Any context.
 502  */
 503 void ida_destroy(struct ida *ida)
 504 {
 505         unsigned long flags;
 506         struct radix_tree_iter iter;
 507         void __rcu **slot;
 508
 509         xa_lock_irqsave(&ida->ida_rt, flags);
 510         radix_tree_for_each_slot(slot, &ida->ida_rt, &iter, 0) {
 511                 struct ida_bitmap *bitmap = rcu_dereference_raw(*slot);
 512                 if (!radix_tree_exception(bitmap))
 513                         kfree(bitmap);
 514                 radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
 515         }
 516         xa_unlock_irqrestore(&ida->ida_rt, flags);
 517 }
 518 EXPORT_SYMBOL(ida_destroy);
 519
 520 /**
 521  * ida_alloc_range() - Allocate an unused ID.
 522  * @ida: IDA handle.
 523  * @min: Lowest ID to allocate.
 524  * @max: Highest ID to allocate.
 525  * @gfp: Memory allocation flags.
 526  *
 527  * Allocate an ID between @min and @max, inclusive.  The allocated ID will
 528  * not exceed %INT_MAX, even if @max is larger.
 529  *
 530  * Context: Any context.
 531  * Return: The allocated ID, or %-ENOMEM if memory could not be allocated,
 532  * or %-ENOSPC if there are no free IDs.
 533  */
 534 int ida_alloc_range(struct ida *ida, unsigned int min, unsigned int max,
 535                         gfp_t gfp)
 536 {
 537         int id = 0;
 538         unsigned long flags;
 539
 540         if ((int)min < 0)
 541                 return -ENOSPC;
 542
 543         if ((int)max < 0)
 544                 max = INT_MAX;
 545
 546 again:
 547         xa_lock_irqsave(&ida->ida_rt, flags);
 548         id = ida_get_new_above(ida, min);
 549         if (id > (int)max) {
 550                 ida_remove(ida, id);
 551                 id = -ENOSPC;
 552         }
 553         xa_unlock_irqrestore(&ida->ida_rt, flags);
 554
 555         if (unlikely(id == -EAGAIN)) {
 556                 if (!ida_pre_get(ida, gfp))
 557                         return -ENOMEM;
 558                 goto again;
 559         }
 560
 561         return id;
 562 }
 563 EXPORT_SYMBOL(ida_alloc_range);
 564
 565 /**
 566  * ida_free() - Release an allocated ID.
 567  * @ida: IDA handle.
 568  * @id: Previously allocated ID.
 569  *
 570  * Context: Any context.
 571  */
 572 void ida_free(struct ida *ida, unsigned int id)
 573 {
 574         unsigned long flags;
 575
 576         BUG_ON((int)id < 0);
 577         xa_lock_irqsave(&ida->ida_rt, flags);
 578         ida_remove(ida, id);
 579         xa_unlock_irqrestore(&ida->ida_rt, flags);
 580 }
 581 EXPORT_SYMBOL(ida_free);