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