s390: Fix runtime warning about negative pgtables_bytes
[linux-2.6/btrfs-unstable.git] / lib / idr.c
blobc98d77fcf3934d9e47b33ba4d02b509c25273d31
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>
8 DEFINE_PER_CPU(struct ida_bitmap *, ida_bitmap);
9 static DEFINE_SPINLOCK(simple_ida_lock);
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.
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.
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.
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.
34 int idr_alloc_u32(struct idr *idr, void *ptr, u32 *nextid,
35 unsigned long max, gfp_t gfp)
37 struct radix_tree_iter iter;
38 void __rcu **slot;
39 int base = idr->idr_base;
40 int id = *nextid;
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;
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);
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);
58 return 0;
60 EXPORT_SYMBOL_GPL(idr_alloc_u32);
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.
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.
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.
79 * Return: The newly allocated ID, -ENOMEM if memory allocation failed,
80 * or -ENOSPC if no free IDs could be found.
82 int idr_alloc(struct idr *idr, void *ptr, int start, int end, gfp_t gfp)
84 u32 id = start;
85 int ret;
87 if (WARN_ON_ONCE(start < 0))
88 return -EINVAL;
90 ret = idr_alloc_u32(idr, ptr, &id, end > 0 ? end - 1 : INT_MAX, gfp);
91 if (ret)
92 return ret;
94 return id;
96 EXPORT_SYMBOL_GPL(idr_alloc);
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.
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.
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.
117 * Return: The newly allocated ID, -ENOMEM if memory allocation failed,
118 * or -ENOSPC if no free IDs could be found.
120 int idr_alloc_cyclic(struct idr *idr, void *ptr, int start, int end, gfp_t gfp)
122 u32 id = idr->idr_next;
123 int err, max = end > 0 ? end - 1 : INT_MAX;
125 if ((int)id < start)
126 id = start;
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);
133 if (err)
134 return err;
136 idr->idr_next = id + 1;
137 return id;
139 EXPORT_SYMBOL(idr_alloc_cyclic);
142 * idr_remove() - Remove an ID from the IDR.
143 * @idr: IDR handle.
144 * @id: Pointer ID.
146 * Removes this ID from the IDR. If the ID was not previously in the IDR,
147 * this function returns %NULL.
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.
153 * Return: The pointer formerly associated with this ID.
155 void *idr_remove(struct idr *idr, unsigned long id)
157 return radix_tree_delete_item(&idr->idr_rt, id - idr->idr_base, NULL);
159 EXPORT_SYMBOL_GPL(idr_remove);
162 * idr_find() - Return pointer for given ID.
163 * @idr: IDR handle.
164 * @id: Pointer ID.
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.
170 * This function can be called under rcu_read_lock(), given that the leaf
171 * pointers lifetimes are correctly managed.
173 * Return: The pointer associated with this ID.
175 void *idr_find(const struct idr *idr, unsigned long id)
177 return radix_tree_lookup(&idr->idr_rt, id - idr->idr_base);
179 EXPORT_SYMBOL_GPL(idr_find);
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.
187 * The callback function will be called for each entry in @idr, passing
188 * the ID, the entry and @data.
190 * If @fn returns anything other than %0, the iteration stops and that
191 * value is returned from this function.
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.
198 int idr_for_each(const struct idr *idr,
199 int (*fn)(int id, void *p, void *data), void *data)
201 struct radix_tree_iter iter;
202 void __rcu **slot;
203 int base = idr->idr_base;
205 radix_tree_for_each_slot(slot, &idr->idr_rt, &iter, 0) {
206 int ret;
208 if (WARN_ON_ONCE(iter.index > INT_MAX))
209 break;
210 ret = fn(iter.index + base, rcu_dereference_raw(*slot), data);
211 if (ret)
212 return ret;
215 return 0;
217 EXPORT_SYMBOL(idr_for_each);
220 * idr_get_next() - Find next populated entry.
221 * @idr: IDR handle.
222 * @nextid: Pointer to an ID.
224 * Returns the next populated entry in the tree with an ID greater than
225 * or equal to the value pointed to by @nextid. On exit, @nextid is updated
226 * to the ID of the found value. To use in a loop, the value pointed to by
227 * nextid must be incremented by the user.
229 void *idr_get_next(struct idr *idr, int *nextid)
231 struct radix_tree_iter iter;
232 void __rcu **slot;
233 int base = idr->idr_base;
234 int id = *nextid;
236 id = (id < base) ? 0 : id - base;
237 slot = radix_tree_iter_find(&idr->idr_rt, &iter, id);
238 if (!slot)
239 return NULL;
240 id = iter.index + base;
242 if (WARN_ON_ONCE(id > INT_MAX))
243 return NULL;
245 *nextid = id;
246 return rcu_dereference_raw(*slot);
248 EXPORT_SYMBOL(idr_get_next);
251 * idr_get_next_ul() - Find next populated entry.
252 * @idr: IDR handle.
253 * @nextid: Pointer to an ID.
255 * Returns the next populated entry in the tree with an ID greater than
256 * or equal to the value pointed to by @nextid. On exit, @nextid is updated
257 * to the ID of the found value. To use in a loop, the value pointed to by
258 * nextid must be incremented by the user.
260 void *idr_get_next_ul(struct idr *idr, unsigned long *nextid)
262 struct radix_tree_iter iter;
263 void __rcu **slot;
264 unsigned long base = idr->idr_base;
265 unsigned long id = *nextid;
267 id = (id < base) ? 0 : id - base;
268 slot = radix_tree_iter_find(&idr->idr_rt, &iter, id);
269 if (!slot)
270 return NULL;
272 *nextid = iter.index + base;
273 return rcu_dereference_raw(*slot);
275 EXPORT_SYMBOL(idr_get_next_ul);
278 * idr_replace() - replace pointer for given ID.
279 * @idr: IDR handle.
280 * @ptr: New pointer to associate with the ID.
281 * @id: ID to change.
283 * Replace the pointer registered with an ID and return the old value.
284 * This function can be called under the RCU read lock concurrently with
285 * idr_alloc() and idr_remove() (as long as the ID being removed is not
286 * the one being replaced!).
288 * Returns: the old value on success. %-ENOENT indicates that @id was not
289 * found. %-EINVAL indicates that @ptr was not valid.
291 void *idr_replace(struct idr *idr, void *ptr, unsigned long id)
293 struct radix_tree_node *node;
294 void __rcu **slot = NULL;
295 void *entry;
297 if (WARN_ON_ONCE(radix_tree_is_internal_node(ptr)))
298 return ERR_PTR(-EINVAL);
299 id -= idr->idr_base;
301 entry = __radix_tree_lookup(&idr->idr_rt, id, &node, &slot);
302 if (!slot || radix_tree_tag_get(&idr->idr_rt, id, IDR_FREE))
303 return ERR_PTR(-ENOENT);
305 __radix_tree_replace(&idr->idr_rt, node, slot, ptr, NULL);
307 return entry;
309 EXPORT_SYMBOL(idr_replace);
312 * DOC: IDA description
314 * The IDA is an ID allocator which does not provide the ability to
315 * associate an ID with a pointer. As such, it only needs to store one
316 * bit per ID, and so is more space efficient than an IDR. To use an IDA,
317 * define it using DEFINE_IDA() (or embed a &struct ida in a data structure,
318 * then initialise it using ida_init()). To allocate a new ID, call
319 * ida_simple_get(). To free an ID, call ida_simple_remove().
321 * If you have more complex locking requirements, use a loop around
322 * ida_pre_get() and ida_get_new() to allocate a new ID. Then use
323 * ida_remove() to free an ID. You must make sure that ida_get_new() and
324 * ida_remove() cannot be called at the same time as each other for the
325 * same IDA.
327 * You can also use ida_get_new_above() if you need an ID to be allocated
328 * above a particular number. ida_destroy() can be used to dispose of an
329 * IDA without needing to free the individual IDs in it. You can use
330 * ida_is_empty() to find out whether the IDA has any IDs currently allocated.
332 * IDs are currently limited to the range [0-INT_MAX]. If this is an awkward
333 * limitation, it should be quite straightforward to raise the maximum.
337 * Developer's notes:
339 * The IDA uses the functionality provided by the IDR & radix tree to store
340 * bitmaps in each entry. The IDR_FREE tag means there is at least one bit
341 * free, unlike the IDR where it means at least one entry is free.
343 * I considered telling the radix tree that each slot is an order-10 node
344 * and storing the bit numbers in the radix tree, but the radix tree can't
345 * allow a single multiorder entry at index 0, which would significantly
346 * increase memory consumption for the IDA. So instead we divide the index
347 * by the number of bits in the leaf bitmap before doing a radix tree lookup.
349 * As an optimisation, if there are only a few low bits set in any given
350 * leaf, instead of allocating a 128-byte bitmap, we use the 'exceptional
351 * entry' functionality of the radix tree to store BITS_PER_LONG - 2 bits
352 * directly in the entry. By being really tricksy, we could store
353 * BITS_PER_LONG - 1 bits, but there're diminishing returns after optimising
354 * for 0-3 allocated IDs.
356 * We allow the radix tree 'exceptional' count to get out of date. Nothing
357 * in the IDA nor the radix tree code checks it. If it becomes important
358 * to maintain an accurate exceptional count, switch the rcu_assign_pointer()
359 * calls to radix_tree_iter_replace() which will correct the exceptional
360 * count.
362 * The IDA always requires a lock to alloc/free. If we add a 'test_bit'
363 * equivalent, it will still need locking. Going to RCU lookup would require
364 * using RCU to free bitmaps, and that's not trivial without embedding an
365 * RCU head in the bitmap, which adds a 2-pointer overhead to each 128-byte
366 * bitmap, which is excessive.
369 #define IDA_MAX (0x80000000U / IDA_BITMAP_BITS - 1)
372 * ida_get_new_above - allocate new ID above or equal to a start id
373 * @ida: ida handle
374 * @start: id to start search at
375 * @id: pointer to the allocated handle
377 * Allocate new ID above or equal to @start. It should be called
378 * with any required locks to ensure that concurrent calls to
379 * ida_get_new_above() / ida_get_new() / ida_remove() are not allowed.
380 * Consider using ida_simple_get() if you do not have complex locking
381 * requirements.
383 * If memory is required, it will return %-EAGAIN, you should unlock
384 * and go back to the ida_pre_get() call. If the ida is full, it will
385 * return %-ENOSPC. On success, it will return 0.
387 * @id returns a value in the range @start ... %0x7fffffff.
389 int ida_get_new_above(struct ida *ida, int start, int *id)
391 struct radix_tree_root *root = &ida->ida_rt;
392 void __rcu **slot;
393 struct radix_tree_iter iter;
394 struct ida_bitmap *bitmap;
395 unsigned long index;
396 unsigned bit, ebit;
397 int new;
399 index = start / IDA_BITMAP_BITS;
400 bit = start % IDA_BITMAP_BITS;
401 ebit = bit + RADIX_TREE_EXCEPTIONAL_SHIFT;
403 slot = radix_tree_iter_init(&iter, index);
404 for (;;) {
405 if (slot)
406 slot = radix_tree_next_slot(slot, &iter,
407 RADIX_TREE_ITER_TAGGED);
408 if (!slot) {
409 slot = idr_get_free(root, &iter, GFP_NOWAIT, IDA_MAX);
410 if (IS_ERR(slot)) {
411 if (slot == ERR_PTR(-ENOMEM))
412 return -EAGAIN;
413 return PTR_ERR(slot);
416 if (iter.index > index) {
417 bit = 0;
418 ebit = RADIX_TREE_EXCEPTIONAL_SHIFT;
420 new = iter.index * IDA_BITMAP_BITS;
421 bitmap = rcu_dereference_raw(*slot);
422 if (radix_tree_exception(bitmap)) {
423 unsigned long tmp = (unsigned long)bitmap;
424 ebit = find_next_zero_bit(&tmp, BITS_PER_LONG, ebit);
425 if (ebit < BITS_PER_LONG) {
426 tmp |= 1UL << ebit;
427 rcu_assign_pointer(*slot, (void *)tmp);
428 *id = new + ebit - RADIX_TREE_EXCEPTIONAL_SHIFT;
429 return 0;
431 bitmap = this_cpu_xchg(ida_bitmap, NULL);
432 if (!bitmap)
433 return -EAGAIN;
434 memset(bitmap, 0, sizeof(*bitmap));
435 bitmap->bitmap[0] = tmp >> RADIX_TREE_EXCEPTIONAL_SHIFT;
436 rcu_assign_pointer(*slot, bitmap);
439 if (bitmap) {
440 bit = find_next_zero_bit(bitmap->bitmap,
441 IDA_BITMAP_BITS, bit);
442 new += bit;
443 if (new < 0)
444 return -ENOSPC;
445 if (bit == IDA_BITMAP_BITS)
446 continue;
448 __set_bit(bit, bitmap->bitmap);
449 if (bitmap_full(bitmap->bitmap, IDA_BITMAP_BITS))
450 radix_tree_iter_tag_clear(root, &iter,
451 IDR_FREE);
452 } else {
453 new += bit;
454 if (new < 0)
455 return -ENOSPC;
456 if (ebit < BITS_PER_LONG) {
457 bitmap = (void *)((1UL << ebit) |
458 RADIX_TREE_EXCEPTIONAL_ENTRY);
459 radix_tree_iter_replace(root, &iter, slot,
460 bitmap);
461 *id = new;
462 return 0;
464 bitmap = this_cpu_xchg(ida_bitmap, NULL);
465 if (!bitmap)
466 return -EAGAIN;
467 memset(bitmap, 0, sizeof(*bitmap));
468 __set_bit(bit, bitmap->bitmap);
469 radix_tree_iter_replace(root, &iter, slot, bitmap);
472 *id = new;
473 return 0;
476 EXPORT_SYMBOL(ida_get_new_above);
479 * ida_remove - Free the given ID
480 * @ida: ida handle
481 * @id: ID to free
483 * This function should not be called at the same time as ida_get_new_above().
485 void ida_remove(struct ida *ida, int id)
487 unsigned long index = id / IDA_BITMAP_BITS;
488 unsigned offset = id % IDA_BITMAP_BITS;
489 struct ida_bitmap *bitmap;
490 unsigned long *btmp;
491 struct radix_tree_iter iter;
492 void __rcu **slot;
494 slot = radix_tree_iter_lookup(&ida->ida_rt, &iter, index);
495 if (!slot)
496 goto err;
498 bitmap = rcu_dereference_raw(*slot);
499 if (radix_tree_exception(bitmap)) {
500 btmp = (unsigned long *)slot;
501 offset += RADIX_TREE_EXCEPTIONAL_SHIFT;
502 if (offset >= BITS_PER_LONG)
503 goto err;
504 } else {
505 btmp = bitmap->bitmap;
507 if (!test_bit(offset, btmp))
508 goto err;
510 __clear_bit(offset, btmp);
511 radix_tree_iter_tag_set(&ida->ida_rt, &iter, IDR_FREE);
512 if (radix_tree_exception(bitmap)) {
513 if (rcu_dereference_raw(*slot) ==
514 (void *)RADIX_TREE_EXCEPTIONAL_ENTRY)
515 radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
516 } else if (bitmap_empty(btmp, IDA_BITMAP_BITS)) {
517 kfree(bitmap);
518 radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
520 return;
521 err:
522 WARN(1, "ida_remove called for id=%d which is not allocated.\n", id);
524 EXPORT_SYMBOL(ida_remove);
527 * ida_destroy - Free the contents of an ida
528 * @ida: ida handle
530 * Calling this function releases all resources associated with an IDA. When
531 * this call returns, the IDA is empty and can be reused or freed. The caller
532 * should not allow ida_remove() or ida_get_new_above() to be called at the
533 * same time.
535 void ida_destroy(struct ida *ida)
537 struct radix_tree_iter iter;
538 void __rcu **slot;
540 radix_tree_for_each_slot(slot, &ida->ida_rt, &iter, 0) {
541 struct ida_bitmap *bitmap = rcu_dereference_raw(*slot);
542 if (!radix_tree_exception(bitmap))
543 kfree(bitmap);
544 radix_tree_iter_delete(&ida->ida_rt, &iter, slot);
547 EXPORT_SYMBOL(ida_destroy);
550 * ida_simple_get - get a new id.
551 * @ida: the (initialized) ida.
552 * @start: the minimum id (inclusive, < 0x8000000)
553 * @end: the maximum id (exclusive, < 0x8000000 or 0)
554 * @gfp_mask: memory allocation flags
556 * Allocates an id in the range start <= id < end, or returns -ENOSPC.
557 * On memory allocation failure, returns -ENOMEM.
559 * Compared to ida_get_new_above() this function does its own locking, and
560 * should be used unless there are special requirements.
562 * Use ida_simple_remove() to get rid of an id.
564 int ida_simple_get(struct ida *ida, unsigned int start, unsigned int end,
565 gfp_t gfp_mask)
567 int ret, id;
568 unsigned int max;
569 unsigned long flags;
571 BUG_ON((int)start < 0);
572 BUG_ON((int)end < 0);
574 if (end == 0)
575 max = 0x80000000;
576 else {
577 BUG_ON(end < start);
578 max = end - 1;
581 again:
582 if (!ida_pre_get(ida, gfp_mask))
583 return -ENOMEM;
585 spin_lock_irqsave(&simple_ida_lock, flags);
586 ret = ida_get_new_above(ida, start, &id);
587 if (!ret) {
588 if (id > max) {
589 ida_remove(ida, id);
590 ret = -ENOSPC;
591 } else {
592 ret = id;
595 spin_unlock_irqrestore(&simple_ida_lock, flags);
597 if (unlikely(ret == -EAGAIN))
598 goto again;
600 return ret;
602 EXPORT_SYMBOL(ida_simple_get);
605 * ida_simple_remove - remove an allocated id.
606 * @ida: the (initialized) ida.
607 * @id: the id returned by ida_simple_get.
609 * Use to release an id allocated with ida_simple_get().
611 * Compared to ida_remove() this function does its own locking, and should be
612 * used unless there are special requirements.
614 void ida_simple_remove(struct ida *ida, unsigned int id)
616 unsigned long flags;
618 BUG_ON((int)id < 0);
619 spin_lock_irqsave(&simple_ida_lock, flags);
620 ida_remove(ida, id);
621 spin_unlock_irqrestore(&simple_ida_lock, flags);
623 EXPORT_SYMBOL(ida_simple_remove);