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() - 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(struct idr *idr, int *nextid)
 231 {
 232         struct radix_tree_iter iter;
 233         void __rcu **slot;
 234         unsigned long base = idr->idr_base;
 235         unsigned long id = *nextid;
 236
 237         id = (id < base) ? 0 : id - base;
 238         slot = radix_tree_iter_find(&idr->idr_rt, &iter, id);
 239         if (!slot)
 240                 return NULL;
 241         id = iter.index + base;
 242
 243         if (WARN_ON_ONCE(id > INT_MAX))
 244                 return NULL;
 245
 246         *nextid = id;
 247         return rcu_dereference_raw(*slot);
 248 }
 249 EXPORT_SYMBOL(idr_get_next);
 250
 251 /**
 252  * idr_get_next_ul() - Find next populated entry.
 253  * @idr: IDR handle.
 254  * @nextid: Pointer to an ID.
 255  *
 256  * Returns the next populated entry in the tree with an ID greater than
 257  * or equal to the value pointed to by @nextid.  On exit, @nextid is updated
 258  * to the ID of the found value.  To use in a loop, the value pointed to by
 259  * nextid must be incremented by the user.
 260  */
 261 void *idr_get_next_ul(struct idr *idr, unsigned long *nextid)
 262 {
 263         struct radix_tree_iter iter;
 264         void __rcu **slot;
 265         unsigned long base = idr->idr_base;
 266         unsigned long id = *nextid;
 267
 268         id = (id < base) ? 0 : id - base;
 269         slot = radix_tree_iter_find(&idr->idr_rt, &iter, id);
 270         if (!slot)
 271                 return NULL;
 272
 273         *nextid = iter.index + base;
 274         return rcu_dereference_raw(*slot);
 275 }
 276 EXPORT_SYMBOL(idr_get_next_ul);
 277
 278 /**
 279  * idr_replace() - replace pointer for given ID.
 280  * @idr: IDR handle.
 281  * @ptr: New pointer to associate with the ID.
 282  * @id: ID to change.
 283  *
 284  * Replace the pointer registered with an ID and return the old value.
 285  * This function can be called under the RCU read lock concurrently with
 286  * idr_alloc() and idr_remove() (as long as the ID being removed is not
 287  * the one being replaced!).
 288  *
 289  * Returns: the old value on success.  %-ENOENT indicates that @id was not
 290  * found.  %-EINVAL indicates that @ptr was not valid.
 291  */
 292 void *idr_replace(struct idr *idr, void *ptr, unsigned long id)
 293 {
 294         struct radix_tree_node *node;
 295         void __rcu **slot = NULL;
 296         void *entry;
 297
 298         if (WARN_ON_ONCE(radix_tree_is_internal_node(ptr)))
 299                 return ERR_PTR(-EINVAL);
 300         id -= idr->idr_base;
 301
 302         entry = __radix_tree_lookup(&idr->idr_rt, id, &node, &slot);
 303         if (!slot || radix_tree_tag_get(&idr->idr_rt, id, IDR_FREE))
 304                 return ERR_PTR(-ENOENT);
 305
 306         __radix_tree_replace(&idr->idr_rt, node, slot, ptr, NULL);
 307
 308         return entry;
 309 }
 310 EXPORT_SYMBOL(idr_replace);
 311
 312 /**
 313  * DOC: IDA description
 314  *
 315  * The IDA is an ID allocator which does not provide the ability to
 316  * associate an ID with a pointer.  As such, it only needs to store one
 317  * bit per ID, and so is more space efficient than an IDR.  To use an IDA,
 318  * define it using DEFINE_IDA() (or embed a &struct ida in a data structure,
 319  * then initialise it using ida_init()).  To allocate a new ID, call
 320  * ida_alloc(), ida_alloc_min(), ida_alloc_max() or ida_alloc_range().
 321  * To free an ID, call ida_free().
 322  *
 323  * ida_destroy() can be used to dispose of an IDA without needing to
 324  * free the individual IDs in it.  You can use ida_is_empty() to find
 325  * out whether the IDA has any IDs currently allocated.
 326  *
 327  * IDs are currently limited to the range [0-INT_MAX].  If this is an awkward
 328  * limitation, it should be quite straightforward to raise the maximum.
 329  */
 330
 331 /*
 332  * Developer's notes:
 333  *
 334  * The IDA uses the functionality provided by the IDR & radix tree to store
 335  * bitmaps in each entry.  The IDR_FREE tag means there is at least one bit
 336  * free, unlike the IDR where it means at least one entry is free.
 337  *
 338  * I considered telling the radix tree that each slot is an order-10 node
 339  * and storing the bit numbers in the radix tree, but the radix tree can't
 340  * allow a single multiorder entry at index 0, which would significantly
 341  * increase memory consumption for the IDA.  So instead we divide the index
 342  * by the number of bits in the leaf bitmap before doing a radix tree lookup.
 343  *
 344  * As an optimisation, if there are only a few low bits set in any given
 345  * leaf, instead of allocating a 128-byte bitmap, we use the 'exceptional
 346  * entry' functionality of the radix tree to store BITS_PER_LONG - 2 bits
 347  * directly in the entry.  By being really tricksy, we could store
 348  * BITS_PER_LONG - 1 bits, but there're diminishing returns after optimising
 349  * for 0-3 allocated IDs.
 350  *
 351  * We allow the radix tree 'exceptional' count to get out of date.  Nothing
 352  * in the IDA nor the radix tree code checks it.  If it becomes important
 353  * to maintain an accurate exceptional count, switch the rcu_assign_pointer()
 354  * calls to radix_tree_iter_replace() which will correct the exceptional
 355  * count.
 356  *
 357  * The IDA always requires a lock to alloc/free.  If we add a 'test_bit'
 358  * equivalent, it will still need locking.  Going to RCU lookup would require
 359  * using RCU to free bitmaps, and that's not trivial without embedding an
 360  * RCU head in the bitmap, which adds a 2-pointer overhead to each 128-byte
 361  * bitmap, which is excessive.
 362  */
 363
 364 #define IDA_MAX (0x80000000U / IDA_BITMAP_BITS - 1)
 365
 366 static int ida_get_new_above(struct ida *ida, int start)
 367 {
 368         struct radix_tree_root *root = &ida->ida_rt;
 369         void __rcu **slot;
 370         struct radix_tree_iter iter;
 371         struct ida_bitmap *bitmap;
 372         unsigned long index;
 373         unsigned bit, ebit;
 374         int new;
 375
 376         index = start / IDA_BITMAP_BITS;
 377         bit = start % IDA_BITMAP_BITS;
 378         ebit = bit + RADIX_TREE_EXCEPTIONAL_SHIFT;
 379
 380         slot = radix_tree_iter_init(&iter, index);
 381         for (;;) {
 382                 if (slot)
 383                         slot = radix_tree_next_slot(slot, &iter,
 384                                                 RADIX_TREE_ITER_TAGGED);
 385                 if (!slot) {
 386                         slot = idr_get_free(root, &iter, GFP_NOWAIT, IDA_MAX);
 387                         if (IS_ERR(slot)) {
 388                                 if (slot == ERR_PTR(-ENOMEM))
 389                                         return -EAGAIN;
 390                                 return PTR_ERR(slot);
 391                         }
 392                 }
 393                 if (iter.index > index) {
 394                         bit = 0;
 395                         ebit = RADIX_TREE_EXCEPTIONAL_SHIFT;
 396                 }
 397                 new = iter.index * IDA_BITMAP_BITS;
 398                 bitmap = rcu_dereference_raw(*slot);
 399                 if (radix_tree_exception(bitmap)) {
 400                         unsigned long tmp = (unsigned long)bitmap;
 401                         ebit = find_next_zero_bit(&tmp, BITS_PER_LONG, ebit);
 402                         if (ebit < BITS_PER_LONG) {
 403                                 tmp |= 1UL << ebit;
 404                                 rcu_assign_pointer(*slot, (void *)tmp);
 405                                 return new + ebit -
 406                                         RADIX_TREE_EXCEPTIONAL_SHIFT;
 407                         }
 408                         bitmap = this_cpu_xchg(ida_bitmap, NULL);
 409                         if (!bitmap)
 410                                 return -EAGAIN;
 411                         bitmap->bitmap[0] = tmp >> RADIX_TREE_EXCEPTIONAL_SHIFT;
 412                         rcu_assign_pointer(*slot, bitmap);
 413                 }
 414
 415                 if (bitmap) {
 416                         bit = find_next_zero_bit(bitmap->bitmap,
 417                                                         IDA_BITMAP_BITS, bit);
 418                         new += bit;
 419                         if (new < 0)
 420                                 return -ENOSPC;
 421                         if (bit == IDA_BITMAP_BITS)
 422                                 continue;
 423
 424                         __set_bit(bit, bitmap->bitmap);
 425                         if (bitmap_full(bitmap->bitmap, IDA_BITMAP_BITS))
 426                                 radix_tree_iter_tag_clear(root, &iter,
 427                                                                 IDR_FREE);
 428                 } else {
 429                         new += bit;
 430                         if (new < 0)
 431                                 return -ENOSPC;
 432                         if (ebit < BITS_PER_LONG) {
 433                                 bitmap = (void *)((1UL << ebit) |
 434                                                 RADIX_TREE_EXCEPTIONAL_ENTRY);
 435                                 radix_tree_iter_replace(root, &iter, slot,
 436                                                 bitmap);
 437                                 return new;
 438                         }
 439                         bitmap = this_cpu_xchg(ida_bitmap, NULL);
 440                         if (!bitmap)
 441                                 return -EAGAIN;
 442                         __set_bit(bit, bitmap->bitmap);
 443                         radix_tree_iter_replace(root, &iter, slot, bitmap);
 444                 }
 445
 446                 return new;
 447         }
 448 }
 449
 450 static void ida_remove(struct ida *ida, int id)
 451 {
 452         unsigned long index = id / IDA_BITMAP_BITS;
 453         unsigned offset = id % IDA_BITMAP_BITS;
 454         struct ida_bitmap *bitmap;
 455         unsigned long *btmp;
 456         struct radix_tree_iter iter;
 457         void __rcu **slot;
 458
 459         slot = radix_tree_iter_lookup(&ida->ida_rt, &iter, index);
 460         if (!slot)
 461                 goto err;
 462
 463         bitmap = rcu_dereference_raw(*slot);
 464         if (radix_tree_exception(bitmap)) {
 465                 btmp = (unsigned long *)slot;
 466                 offset += RADIX_TREE_EXCEPTIONAL_SHIFT;
 467                 if (offset >= BITS_PER_LONG)
 468                         goto err;
 469         } else {
 470                 btmp = bitmap->bitmap;
 471         }
 472         if (!test_bit(offset, btmp))
 473                 goto err;
 474
 475         __clear_bit(offset, btmp);
 476         radix_tree_iter_tag_set(&ida->ida_rt, &iter, IDR_FREE);
 477         if (radix_tree_exception(bitmap)) {
 478                 if (rcu_dereference_raw(*slot) ==
 479                                         (void *)RADIX_TREE_EXCEPTIONAL_ENTRY)
 480                         radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
 481         } else if (bitmap_empty(btmp, IDA_BITMAP_BITS)) {
 482                 kfree(bitmap);
 483                 radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
 484         }
 485         return;
 486  err:
 487         WARN(1, "ida_free called for id=%d which is not allocated.\n", id);
 488 }
 489
 490 /**
 491  * ida_destroy() - Free all IDs.
 492  * @ida: IDA handle.
 493  *
 494  * Calling this function frees all IDs and releases all resources used
 495  * by an IDA.  When this call returns, the IDA is empty and can be reused
 496  * or freed.  If the IDA is already empty, there is no need to call this
 497  * function.
 498  *
 499  * Context: Any context.
 500  */
 501 void ida_destroy(struct ida *ida)
 502 {
 503         unsigned long flags;
 504         struct radix_tree_iter iter;
 505         void __rcu **slot;
 506
 507         xa_lock_irqsave(&ida->ida_rt, flags);
 508         radix_tree_for_each_slot(slot, &ida->ida_rt, &iter, 0) {
 509                 struct ida_bitmap *bitmap = rcu_dereference_raw(*slot);
 510                 if (!radix_tree_exception(bitmap))
 511                         kfree(bitmap);
 512                 radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
 513         }
 514         xa_unlock_irqrestore(&ida->ida_rt, flags);
 515 }
 516 EXPORT_SYMBOL(ida_destroy);
 517
 518 /**
 519  * ida_alloc_range() - Allocate an unused ID.
 520  * @ida: IDA handle.
 521  * @min: Lowest ID to allocate.
 522  * @max: Highest ID to allocate.
 523  * @gfp: Memory allocation flags.
 524  *
 525  * Allocate an ID between @min and @max, inclusive.  The allocated ID will
 526  * not exceed %INT_MAX, even if @max is larger.
 527  *
 528  * Context: Any context.
 529  * Return: The allocated ID, or %-ENOMEM if memory could not be allocated,
 530  * or %-ENOSPC if there are no free IDs.
 531  */
 532 int ida_alloc_range(struct ida *ida, unsigned int min, unsigned int max,
 533                         gfp_t gfp)
 534 {
 535         int id = 0;
 536         unsigned long flags;
 537
 538         if ((int)min < 0)
 539                 return -ENOSPC;
 540
 541         if ((int)max < 0)
 542                 max = INT_MAX;
 543
 544 again:
 545         xa_lock_irqsave(&ida->ida_rt, flags);
 546         id = ida_get_new_above(ida, min);
 547         if (id > (int)max) {
 548                 ida_remove(ida, id);
 549                 id = -ENOSPC;
 550         }
 551         xa_unlock_irqrestore(&ida->ida_rt, flags);
 552
 553         if (unlikely(id == -EAGAIN)) {
 554                 if (!ida_pre_get(ida, gfp))
 555                         return -ENOMEM;
 556                 goto again;
 557         }
 558
 559         return id;
 560 }
 561 EXPORT_SYMBOL(ida_alloc_range);
 562
 563 /**
 564  * ida_free() - Release an allocated ID.
 565  * @ida: IDA handle.
 566  * @id: Previously allocated ID.
 567  *
 568  * Context: Any context.
 569  */
 570 void ida_free(struct ida *ida, unsigned int id)
 571 {
 572         unsigned long flags;
 573
 574         BUG_ON((int)id < 0);
 575         xa_lock_irqsave(&ida->ida_rt, flags);
 576         ida_remove(ida, id);
 577         xa_unlock_irqrestore(&ida->ida_rt, flags);
 578 }
 579 EXPORT_SYMBOL(ida_free);