xref: /linux/lib/idr.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
1 /*
2  * 2002-10-18  written by Jim Houston jim.houston@ccur.com
3  *	Copyright (C) 2002 by Concurrent Computer Corporation
4  *	Distributed under the GNU GPL license version 2.
5  *
6  * Modified by George Anzinger to reuse immediately and to use
7  * find bit instructions.  Also removed _irq on spinlocks.
8  *
9  * Modified by Nadia Derbey to make it RCU safe.
10  *
11  * Small id to pointer translation service.
12  *
13  * It uses a radix tree like structure as a sparse array indexed
14  * by the id to obtain the pointer.  The bitmap makes allocating
15  * a new id quick.
16  *
17  * You call it to allocate an id (an int) an associate with that id a
18  * pointer or what ever, we treat it as a (void *).  You can pass this
19  * id to a user for him to pass back at a later time.  You then pass
20  * that id to this code and it returns your pointer.
21 
22  * You can release ids at any time. When all ids are released, most of
23  * the memory is returned (we keep MAX_IDR_FREE) in a local pool so we
24  * don't need to go to the memory "store" during an id allocate, just
25  * so you don't need to be too concerned about locking and conflicts
26  * with the slab allocator.
27  */
28 
29 #ifndef TEST                        // to test in user space...
30 #include <linux/slab.h>
31 #include <linux/init.h>
32 #include <linux/export.h>
33 #endif
34 #include <linux/err.h>
35 #include <linux/string.h>
36 #include <linux/idr.h>
37 #include <linux/spinlock.h>
38 
39 static struct kmem_cache *idr_layer_cache;
40 static DEFINE_SPINLOCK(simple_ida_lock);
41 
42 static struct idr_layer *get_from_free_list(struct idr *idp)
43 {
44 	struct idr_layer *p;
45 	unsigned long flags;
46 
47 	spin_lock_irqsave(&idp->lock, flags);
48 	if ((p = idp->id_free)) {
49 		idp->id_free = p->ary[0];
50 		idp->id_free_cnt--;
51 		p->ary[0] = NULL;
52 	}
53 	spin_unlock_irqrestore(&idp->lock, flags);
54 	return(p);
55 }
56 
57 static void idr_layer_rcu_free(struct rcu_head *head)
58 {
59 	struct idr_layer *layer;
60 
61 	layer = container_of(head, struct idr_layer, rcu_head);
62 	kmem_cache_free(idr_layer_cache, layer);
63 }
64 
65 static inline void free_layer(struct idr_layer *p)
66 {
67 	call_rcu(&p->rcu_head, idr_layer_rcu_free);
68 }
69 
70 /* only called when idp->lock is held */
71 static void __move_to_free_list(struct idr *idp, struct idr_layer *p)
72 {
73 	p->ary[0] = idp->id_free;
74 	idp->id_free = p;
75 	idp->id_free_cnt++;
76 }
77 
78 static void move_to_free_list(struct idr *idp, struct idr_layer *p)
79 {
80 	unsigned long flags;
81 
82 	/*
83 	 * Depends on the return element being zeroed.
84 	 */
85 	spin_lock_irqsave(&idp->lock, flags);
86 	__move_to_free_list(idp, p);
87 	spin_unlock_irqrestore(&idp->lock, flags);
88 }
89 
90 static void idr_mark_full(struct idr_layer **pa, int id)
91 {
92 	struct idr_layer *p = pa[0];
93 	int l = 0;
94 
95 	__set_bit(id & IDR_MASK, &p->bitmap);
96 	/*
97 	 * If this layer is full mark the bit in the layer above to
98 	 * show that this part of the radix tree is full.  This may
99 	 * complete the layer above and require walking up the radix
100 	 * tree.
101 	 */
102 	while (p->bitmap == IDR_FULL) {
103 		if (!(p = pa[++l]))
104 			break;
105 		id = id >> IDR_BITS;
106 		__set_bit((id & IDR_MASK), &p->bitmap);
107 	}
108 }
109 
110 /**
111  * idr_pre_get - reserve resources for idr allocation
112  * @idp:	idr handle
113  * @gfp_mask:	memory allocation flags
114  *
115  * This function should be called prior to calling the idr_get_new* functions.
116  * It preallocates enough memory to satisfy the worst possible allocation. The
117  * caller should pass in GFP_KERNEL if possible.  This of course requires that
118  * no spinning locks be held.
119  *
120  * If the system is REALLY out of memory this function returns %0,
121  * otherwise %1.
122  */
123 int idr_pre_get(struct idr *idp, gfp_t gfp_mask)
124 {
125 	while (idp->id_free_cnt < MAX_IDR_FREE) {
126 		struct idr_layer *new;
127 		new = kmem_cache_zalloc(idr_layer_cache, gfp_mask);
128 		if (new == NULL)
129 			return (0);
130 		move_to_free_list(idp, new);
131 	}
132 	return 1;
133 }
134 EXPORT_SYMBOL(idr_pre_get);
135 
136 static int sub_alloc(struct idr *idp, int *starting_id, struct idr_layer **pa)
137 {
138 	int n, m, sh;
139 	struct idr_layer *p, *new;
140 	int l, id, oid;
141 	unsigned long bm;
142 
143 	id = *starting_id;
144  restart:
145 	p = idp->top;
146 	l = idp->layers;
147 	pa[l--] = NULL;
148 	while (1) {
149 		/*
150 		 * We run around this while until we reach the leaf node...
151 		 */
152 		n = (id >> (IDR_BITS*l)) & IDR_MASK;
153 		bm = ~p->bitmap;
154 		m = find_next_bit(&bm, IDR_SIZE, n);
155 		if (m == IDR_SIZE) {
156 			/* no space available go back to previous layer. */
157 			l++;
158 			oid = id;
159 			id = (id | ((1 << (IDR_BITS * l)) - 1)) + 1;
160 
161 			/* if already at the top layer, we need to grow */
162 			if (id >= 1 << (idp->layers * IDR_BITS)) {
163 				*starting_id = id;
164 				return IDR_NEED_TO_GROW;
165 			}
166 			p = pa[l];
167 			BUG_ON(!p);
168 
169 			/* If we need to go up one layer, continue the
170 			 * loop; otherwise, restart from the top.
171 			 */
172 			sh = IDR_BITS * (l + 1);
173 			if (oid >> sh == id >> sh)
174 				continue;
175 			else
176 				goto restart;
177 		}
178 		if (m != n) {
179 			sh = IDR_BITS*l;
180 			id = ((id >> sh) ^ n ^ m) << sh;
181 		}
182 		if ((id >= MAX_IDR_BIT) || (id < 0))
183 			return IDR_NOMORE_SPACE;
184 		if (l == 0)
185 			break;
186 		/*
187 		 * Create the layer below if it is missing.
188 		 */
189 		if (!p->ary[m]) {
190 			new = get_from_free_list(idp);
191 			if (!new)
192 				return -1;
193 			new->layer = l-1;
194 			rcu_assign_pointer(p->ary[m], new);
195 			p->count++;
196 		}
197 		pa[l--] = p;
198 		p = p->ary[m];
199 	}
200 
201 	pa[l] = p;
202 	return id;
203 }
204 
205 static int idr_get_empty_slot(struct idr *idp, int starting_id,
206 			      struct idr_layer **pa)
207 {
208 	struct idr_layer *p, *new;
209 	int layers, v, id;
210 	unsigned long flags;
211 
212 	id = starting_id;
213 build_up:
214 	p = idp->top;
215 	layers = idp->layers;
216 	if (unlikely(!p)) {
217 		if (!(p = get_from_free_list(idp)))
218 			return -1;
219 		p->layer = 0;
220 		layers = 1;
221 	}
222 	/*
223 	 * Add a new layer to the top of the tree if the requested
224 	 * id is larger than the currently allocated space.
225 	 */
226 	while ((layers < (MAX_IDR_LEVEL - 1)) && (id >= (1 << (layers*IDR_BITS)))) {
227 		layers++;
228 		if (!p->count) {
229 			/* special case: if the tree is currently empty,
230 			 * then we grow the tree by moving the top node
231 			 * upwards.
232 			 */
233 			p->layer++;
234 			continue;
235 		}
236 		if (!(new = get_from_free_list(idp))) {
237 			/*
238 			 * The allocation failed.  If we built part of
239 			 * the structure tear it down.
240 			 */
241 			spin_lock_irqsave(&idp->lock, flags);
242 			for (new = p; p && p != idp->top; new = p) {
243 				p = p->ary[0];
244 				new->ary[0] = NULL;
245 				new->bitmap = new->count = 0;
246 				__move_to_free_list(idp, new);
247 			}
248 			spin_unlock_irqrestore(&idp->lock, flags);
249 			return -1;
250 		}
251 		new->ary[0] = p;
252 		new->count = 1;
253 		new->layer = layers-1;
254 		if (p->bitmap == IDR_FULL)
255 			__set_bit(0, &new->bitmap);
256 		p = new;
257 	}
258 	rcu_assign_pointer(idp->top, p);
259 	idp->layers = layers;
260 	v = sub_alloc(idp, &id, pa);
261 	if (v == IDR_NEED_TO_GROW)
262 		goto build_up;
263 	return(v);
264 }
265 
266 static int idr_get_new_above_int(struct idr *idp, void *ptr, int starting_id)
267 {
268 	struct idr_layer *pa[MAX_IDR_LEVEL];
269 	int id;
270 
271 	id = idr_get_empty_slot(idp, starting_id, pa);
272 	if (id >= 0) {
273 		/*
274 		 * Successfully found an empty slot.  Install the user
275 		 * pointer and mark the slot full.
276 		 */
277 		rcu_assign_pointer(pa[0]->ary[id & IDR_MASK],
278 				(struct idr_layer *)ptr);
279 		pa[0]->count++;
280 		idr_mark_full(pa, id);
281 	}
282 
283 	return id;
284 }
285 
286 /**
287  * idr_get_new_above - allocate new idr entry above or equal to a start id
288  * @idp: idr handle
289  * @ptr: pointer you want associated with the id
290  * @starting_id: id to start search at
291  * @id: pointer to the allocated handle
292  *
293  * This is the allocate id function.  It should be called with any
294  * required locks.
295  *
296  * If allocation from IDR's private freelist fails, idr_get_new_above() will
297  * return %-EAGAIN.  The caller should retry the idr_pre_get() call to refill
298  * IDR's preallocation and then retry the idr_get_new_above() call.
299  *
300  * If the idr is full idr_get_new_above() will return %-ENOSPC.
301  *
302  * @id returns a value in the range @starting_id ... %0x7fffffff
303  */
304 int idr_get_new_above(struct idr *idp, void *ptr, int starting_id, int *id)
305 {
306 	int rv;
307 
308 	rv = idr_get_new_above_int(idp, ptr, starting_id);
309 	/*
310 	 * This is a cheap hack until the IDR code can be fixed to
311 	 * return proper error values.
312 	 */
313 	if (rv < 0)
314 		return _idr_rc_to_errno(rv);
315 	*id = rv;
316 	return 0;
317 }
318 EXPORT_SYMBOL(idr_get_new_above);
319 
320 /**
321  * idr_get_new - allocate new idr entry
322  * @idp: idr handle
323  * @ptr: pointer you want associated with the id
324  * @id: pointer to the allocated handle
325  *
326  * If allocation from IDR's private freelist fails, idr_get_new_above() will
327  * return %-EAGAIN.  The caller should retry the idr_pre_get() call to refill
328  * IDR's preallocation and then retry the idr_get_new_above() call.
329  *
330  * If the idr is full idr_get_new_above() will return %-ENOSPC.
331  *
332  * @id returns a value in the range %0 ... %0x7fffffff
333  */
334 int idr_get_new(struct idr *idp, void *ptr, int *id)
335 {
336 	int rv;
337 
338 	rv = idr_get_new_above_int(idp, ptr, 0);
339 	/*
340 	 * This is a cheap hack until the IDR code can be fixed to
341 	 * return proper error values.
342 	 */
343 	if (rv < 0)
344 		return _idr_rc_to_errno(rv);
345 	*id = rv;
346 	return 0;
347 }
348 EXPORT_SYMBOL(idr_get_new);
349 
350 static void idr_remove_warning(int id)
351 {
352 	printk(KERN_WARNING
353 		"idr_remove called for id=%d which is not allocated.\n", id);
354 	dump_stack();
355 }
356 
357 static void sub_remove(struct idr *idp, int shift, int id)
358 {
359 	struct idr_layer *p = idp->top;
360 	struct idr_layer **pa[MAX_IDR_LEVEL];
361 	struct idr_layer ***paa = &pa[0];
362 	struct idr_layer *to_free;
363 	int n;
364 
365 	*paa = NULL;
366 	*++paa = &idp->top;
367 
368 	while ((shift > 0) && p) {
369 		n = (id >> shift) & IDR_MASK;
370 		__clear_bit(n, &p->bitmap);
371 		*++paa = &p->ary[n];
372 		p = p->ary[n];
373 		shift -= IDR_BITS;
374 	}
375 	n = id & IDR_MASK;
376 	if (likely(p != NULL && test_bit(n, &p->bitmap))){
377 		__clear_bit(n, &p->bitmap);
378 		rcu_assign_pointer(p->ary[n], NULL);
379 		to_free = NULL;
380 		while(*paa && ! --((**paa)->count)){
381 			if (to_free)
382 				free_layer(to_free);
383 			to_free = **paa;
384 			**paa-- = NULL;
385 		}
386 		if (!*paa)
387 			idp->layers = 0;
388 		if (to_free)
389 			free_layer(to_free);
390 	} else
391 		idr_remove_warning(id);
392 }
393 
394 /**
395  * idr_remove - remove the given id and free its slot
396  * @idp: idr handle
397  * @id: unique key
398  */
399 void idr_remove(struct idr *idp, int id)
400 {
401 	struct idr_layer *p;
402 	struct idr_layer *to_free;
403 
404 	/* Mask off upper bits we don't use for the search. */
405 	id &= MAX_IDR_MASK;
406 
407 	sub_remove(idp, (idp->layers - 1) * IDR_BITS, id);
408 	if (idp->top && idp->top->count == 1 && (idp->layers > 1) &&
409 	    idp->top->ary[0]) {
410 		/*
411 		 * Single child at leftmost slot: we can shrink the tree.
412 		 * This level is not needed anymore since when layers are
413 		 * inserted, they are inserted at the top of the existing
414 		 * tree.
415 		 */
416 		to_free = idp->top;
417 		p = idp->top->ary[0];
418 		rcu_assign_pointer(idp->top, p);
419 		--idp->layers;
420 		to_free->bitmap = to_free->count = 0;
421 		free_layer(to_free);
422 	}
423 	while (idp->id_free_cnt >= MAX_IDR_FREE) {
424 		p = get_from_free_list(idp);
425 		/*
426 		 * Note: we don't call the rcu callback here, since the only
427 		 * layers that fall into the freelist are those that have been
428 		 * preallocated.
429 		 */
430 		kmem_cache_free(idr_layer_cache, p);
431 	}
432 	return;
433 }
434 EXPORT_SYMBOL(idr_remove);
435 
436 /**
437  * idr_remove_all - remove all ids from the given idr tree
438  * @idp: idr handle
439  *
440  * idr_destroy() only frees up unused, cached idp_layers, but this
441  * function will remove all id mappings and leave all idp_layers
442  * unused.
443  *
444  * A typical clean-up sequence for objects stored in an idr tree will
445  * use idr_for_each() to free all objects, if necessay, then
446  * idr_remove_all() to remove all ids, and idr_destroy() to free
447  * up the cached idr_layers.
448  */
449 void idr_remove_all(struct idr *idp)
450 {
451 	int n, id, max;
452 	int bt_mask;
453 	struct idr_layer *p;
454 	struct idr_layer *pa[MAX_IDR_LEVEL];
455 	struct idr_layer **paa = &pa[0];
456 
457 	n = idp->layers * IDR_BITS;
458 	p = idp->top;
459 	rcu_assign_pointer(idp->top, NULL);
460 	max = 1 << n;
461 
462 	id = 0;
463 	while (id < max) {
464 		while (n > IDR_BITS && p) {
465 			n -= IDR_BITS;
466 			*paa++ = p;
467 			p = p->ary[(id >> n) & IDR_MASK];
468 		}
469 
470 		bt_mask = id;
471 		id += 1 << n;
472 		/* Get the highest bit that the above add changed from 0->1. */
473 		while (n < fls(id ^ bt_mask)) {
474 			if (p)
475 				free_layer(p);
476 			n += IDR_BITS;
477 			p = *--paa;
478 		}
479 	}
480 	idp->layers = 0;
481 }
482 EXPORT_SYMBOL(idr_remove_all);
483 
484 /**
485  * idr_destroy - release all cached layers within an idr tree
486  * @idp: idr handle
487  */
488 void idr_destroy(struct idr *idp)
489 {
490 	while (idp->id_free_cnt) {
491 		struct idr_layer *p = get_from_free_list(idp);
492 		kmem_cache_free(idr_layer_cache, p);
493 	}
494 }
495 EXPORT_SYMBOL(idr_destroy);
496 
497 /**
498  * idr_find - return pointer for given id
499  * @idp: idr handle
500  * @id: lookup key
501  *
502  * Return the pointer given the id it has been registered with.  A %NULL
503  * return indicates that @id is not valid or you passed %NULL in
504  * idr_get_new().
505  *
506  * This function can be called under rcu_read_lock(), given that the leaf
507  * pointers lifetimes are correctly managed.
508  */
509 void *idr_find(struct idr *idp, int id)
510 {
511 	int n;
512 	struct idr_layer *p;
513 
514 	p = rcu_dereference_raw(idp->top);
515 	if (!p)
516 		return NULL;
517 	n = (p->layer+1) * IDR_BITS;
518 
519 	/* Mask off upper bits we don't use for the search. */
520 	id &= MAX_IDR_MASK;
521 
522 	if (id >= (1 << n))
523 		return NULL;
524 	BUG_ON(n == 0);
525 
526 	while (n > 0 && p) {
527 		n -= IDR_BITS;
528 		BUG_ON(n != p->layer*IDR_BITS);
529 		p = rcu_dereference_raw(p->ary[(id >> n) & IDR_MASK]);
530 	}
531 	return((void *)p);
532 }
533 EXPORT_SYMBOL(idr_find);
534 
535 /**
536  * idr_for_each - iterate through all stored pointers
537  * @idp: idr handle
538  * @fn: function to be called for each pointer
539  * @data: data passed back to callback function
540  *
541  * Iterate over the pointers registered with the given idr.  The
542  * callback function will be called for each pointer currently
543  * registered, passing the id, the pointer and the data pointer passed
544  * to this function.  It is not safe to modify the idr tree while in
545  * the callback, so functions such as idr_get_new and idr_remove are
546  * not allowed.
547  *
548  * We check the return of @fn each time. If it returns anything other
549  * than %0, we break out and return that value.
550  *
551  * The caller must serialize idr_for_each() vs idr_get_new() and idr_remove().
552  */
553 int idr_for_each(struct idr *idp,
554 		 int (*fn)(int id, void *p, void *data), void *data)
555 {
556 	int n, id, max, error = 0;
557 	struct idr_layer *p;
558 	struct idr_layer *pa[MAX_IDR_LEVEL];
559 	struct idr_layer **paa = &pa[0];
560 
561 	n = idp->layers * IDR_BITS;
562 	p = rcu_dereference_raw(idp->top);
563 	max = 1 << n;
564 
565 	id = 0;
566 	while (id < max) {
567 		while (n > 0 && p) {
568 			n -= IDR_BITS;
569 			*paa++ = p;
570 			p = rcu_dereference_raw(p->ary[(id >> n) & IDR_MASK]);
571 		}
572 
573 		if (p) {
574 			error = fn(id, (void *)p, data);
575 			if (error)
576 				break;
577 		}
578 
579 		id += 1 << n;
580 		while (n < fls(id)) {
581 			n += IDR_BITS;
582 			p = *--paa;
583 		}
584 	}
585 
586 	return error;
587 }
588 EXPORT_SYMBOL(idr_for_each);
589 
590 /**
591  * idr_get_next - lookup next object of id to given id.
592  * @idp: idr handle
593  * @nextidp:  pointer to lookup key
594  *
595  * Returns pointer to registered object with id, which is next number to
596  * given id. After being looked up, *@nextidp will be updated for the next
597  * iteration.
598  *
599  * This function can be called under rcu_read_lock(), given that the leaf
600  * pointers lifetimes are correctly managed.
601  */
602 void *idr_get_next(struct idr *idp, int *nextidp)
603 {
604 	struct idr_layer *p, *pa[MAX_IDR_LEVEL];
605 	struct idr_layer **paa = &pa[0];
606 	int id = *nextidp;
607 	int n, max;
608 
609 	/* find first ent */
610 	p = rcu_dereference_raw(idp->top);
611 	if (!p)
612 		return NULL;
613 	n = (p->layer + 1) * IDR_BITS;
614 	max = 1 << n;
615 
616 	while (id < max) {
617 		while (n > 0 && p) {
618 			n -= IDR_BITS;
619 			*paa++ = p;
620 			p = rcu_dereference_raw(p->ary[(id >> n) & IDR_MASK]);
621 		}
622 
623 		if (p) {
624 			*nextidp = id;
625 			return p;
626 		}
627 
628 		id += 1 << n;
629 		while (n < fls(id)) {
630 			n += IDR_BITS;
631 			p = *--paa;
632 		}
633 	}
634 	return NULL;
635 }
636 EXPORT_SYMBOL(idr_get_next);
637 
638 
639 /**
640  * idr_replace - replace pointer for given id
641  * @idp: idr handle
642  * @ptr: pointer you want associated with the id
643  * @id: lookup key
644  *
645  * Replace the pointer registered with an id and return the old value.
646  * A %-ENOENT return indicates that @id was not found.
647  * A %-EINVAL return indicates that @id was not within valid constraints.
648  *
649  * The caller must serialize with writers.
650  */
651 void *idr_replace(struct idr *idp, void *ptr, int id)
652 {
653 	int n;
654 	struct idr_layer *p, *old_p;
655 
656 	p = idp->top;
657 	if (!p)
658 		return ERR_PTR(-EINVAL);
659 
660 	n = (p->layer+1) * IDR_BITS;
661 
662 	id &= MAX_IDR_MASK;
663 
664 	if (id >= (1 << n))
665 		return ERR_PTR(-EINVAL);
666 
667 	n -= IDR_BITS;
668 	while ((n > 0) && p) {
669 		p = p->ary[(id >> n) & IDR_MASK];
670 		n -= IDR_BITS;
671 	}
672 
673 	n = id & IDR_MASK;
674 	if (unlikely(p == NULL || !test_bit(n, &p->bitmap)))
675 		return ERR_PTR(-ENOENT);
676 
677 	old_p = p->ary[n];
678 	rcu_assign_pointer(p->ary[n], ptr);
679 
680 	return old_p;
681 }
682 EXPORT_SYMBOL(idr_replace);
683 
684 void __init idr_init_cache(void)
685 {
686 	idr_layer_cache = kmem_cache_create("idr_layer_cache",
687 				sizeof(struct idr_layer), 0, SLAB_PANIC, NULL);
688 }
689 
690 /**
691  * idr_init - initialize idr handle
692  * @idp:	idr handle
693  *
694  * This function is use to set up the handle (@idp) that you will pass
695  * to the rest of the functions.
696  */
697 void idr_init(struct idr *idp)
698 {
699 	memset(idp, 0, sizeof(struct idr));
700 	spin_lock_init(&idp->lock);
701 }
702 EXPORT_SYMBOL(idr_init);
703 
704 
705 /**
706  * DOC: IDA description
707  * IDA - IDR based ID allocator
708  *
709  * This is id allocator without id -> pointer translation.  Memory
710  * usage is much lower than full blown idr because each id only
711  * occupies a bit.  ida uses a custom leaf node which contains
712  * IDA_BITMAP_BITS slots.
713  *
714  * 2007-04-25  written by Tejun Heo <htejun@gmail.com>
715  */
716 
717 static void free_bitmap(struct ida *ida, struct ida_bitmap *bitmap)
718 {
719 	unsigned long flags;
720 
721 	if (!ida->free_bitmap) {
722 		spin_lock_irqsave(&ida->idr.lock, flags);
723 		if (!ida->free_bitmap) {
724 			ida->free_bitmap = bitmap;
725 			bitmap = NULL;
726 		}
727 		spin_unlock_irqrestore(&ida->idr.lock, flags);
728 	}
729 
730 	kfree(bitmap);
731 }
732 
733 /**
734  * ida_pre_get - reserve resources for ida allocation
735  * @ida:	ida handle
736  * @gfp_mask:	memory allocation flag
737  *
738  * This function should be called prior to locking and calling the
739  * following function.  It preallocates enough memory to satisfy the
740  * worst possible allocation.
741  *
742  * If the system is REALLY out of memory this function returns %0,
743  * otherwise %1.
744  */
745 int ida_pre_get(struct ida *ida, gfp_t gfp_mask)
746 {
747 	/* allocate idr_layers */
748 	if (!idr_pre_get(&ida->idr, gfp_mask))
749 		return 0;
750 
751 	/* allocate free_bitmap */
752 	if (!ida->free_bitmap) {
753 		struct ida_bitmap *bitmap;
754 
755 		bitmap = kmalloc(sizeof(struct ida_bitmap), gfp_mask);
756 		if (!bitmap)
757 			return 0;
758 
759 		free_bitmap(ida, bitmap);
760 	}
761 
762 	return 1;
763 }
764 EXPORT_SYMBOL(ida_pre_get);
765 
766 /**
767  * ida_get_new_above - allocate new ID above or equal to a start id
768  * @ida:	ida handle
769  * @starting_id: id to start search at
770  * @p_id:	pointer to the allocated handle
771  *
772  * Allocate new ID above or equal to @starting_id.  It should be called
773  * with any required locks.
774  *
775  * If memory is required, it will return %-EAGAIN, you should unlock
776  * and go back to the ida_pre_get() call.  If the ida is full, it will
777  * return %-ENOSPC.
778  *
779  * @p_id returns a value in the range @starting_id ... %0x7fffffff.
780  */
781 int ida_get_new_above(struct ida *ida, int starting_id, int *p_id)
782 {
783 	struct idr_layer *pa[MAX_IDR_LEVEL];
784 	struct ida_bitmap *bitmap;
785 	unsigned long flags;
786 	int idr_id = starting_id / IDA_BITMAP_BITS;
787 	int offset = starting_id % IDA_BITMAP_BITS;
788 	int t, id;
789 
790  restart:
791 	/* get vacant slot */
792 	t = idr_get_empty_slot(&ida->idr, idr_id, pa);
793 	if (t < 0)
794 		return _idr_rc_to_errno(t);
795 
796 	if (t * IDA_BITMAP_BITS >= MAX_IDR_BIT)
797 		return -ENOSPC;
798 
799 	if (t != idr_id)
800 		offset = 0;
801 	idr_id = t;
802 
803 	/* if bitmap isn't there, create a new one */
804 	bitmap = (void *)pa[0]->ary[idr_id & IDR_MASK];
805 	if (!bitmap) {
806 		spin_lock_irqsave(&ida->idr.lock, flags);
807 		bitmap = ida->free_bitmap;
808 		ida->free_bitmap = NULL;
809 		spin_unlock_irqrestore(&ida->idr.lock, flags);
810 
811 		if (!bitmap)
812 			return -EAGAIN;
813 
814 		memset(bitmap, 0, sizeof(struct ida_bitmap));
815 		rcu_assign_pointer(pa[0]->ary[idr_id & IDR_MASK],
816 				(void *)bitmap);
817 		pa[0]->count++;
818 	}
819 
820 	/* lookup for empty slot */
821 	t = find_next_zero_bit(bitmap->bitmap, IDA_BITMAP_BITS, offset);
822 	if (t == IDA_BITMAP_BITS) {
823 		/* no empty slot after offset, continue to the next chunk */
824 		idr_id++;
825 		offset = 0;
826 		goto restart;
827 	}
828 
829 	id = idr_id * IDA_BITMAP_BITS + t;
830 	if (id >= MAX_IDR_BIT)
831 		return -ENOSPC;
832 
833 	__set_bit(t, bitmap->bitmap);
834 	if (++bitmap->nr_busy == IDA_BITMAP_BITS)
835 		idr_mark_full(pa, idr_id);
836 
837 	*p_id = id;
838 
839 	/* Each leaf node can handle nearly a thousand slots and the
840 	 * whole idea of ida is to have small memory foot print.
841 	 * Throw away extra resources one by one after each successful
842 	 * allocation.
843 	 */
844 	if (ida->idr.id_free_cnt || ida->free_bitmap) {
845 		struct idr_layer *p = get_from_free_list(&ida->idr);
846 		if (p)
847 			kmem_cache_free(idr_layer_cache, p);
848 	}
849 
850 	return 0;
851 }
852 EXPORT_SYMBOL(ida_get_new_above);
853 
854 /**
855  * ida_get_new - allocate new ID
856  * @ida:	idr handle
857  * @p_id:	pointer to the allocated handle
858  *
859  * Allocate new ID.  It should be called with any required locks.
860  *
861  * If memory is required, it will return %-EAGAIN, you should unlock
862  * and go back to the idr_pre_get() call.  If the idr is full, it will
863  * return %-ENOSPC.
864  *
865  * @p_id returns a value in the range %0 ... %0x7fffffff.
866  */
867 int ida_get_new(struct ida *ida, int *p_id)
868 {
869 	return ida_get_new_above(ida, 0, p_id);
870 }
871 EXPORT_SYMBOL(ida_get_new);
872 
873 /**
874  * ida_remove - remove the given ID
875  * @ida:	ida handle
876  * @id:		ID to free
877  */
878 void ida_remove(struct ida *ida, int id)
879 {
880 	struct idr_layer *p = ida->idr.top;
881 	int shift = (ida->idr.layers - 1) * IDR_BITS;
882 	int idr_id = id / IDA_BITMAP_BITS;
883 	int offset = id % IDA_BITMAP_BITS;
884 	int n;
885 	struct ida_bitmap *bitmap;
886 
887 	/* clear full bits while looking up the leaf idr_layer */
888 	while ((shift > 0) && p) {
889 		n = (idr_id >> shift) & IDR_MASK;
890 		__clear_bit(n, &p->bitmap);
891 		p = p->ary[n];
892 		shift -= IDR_BITS;
893 	}
894 
895 	if (p == NULL)
896 		goto err;
897 
898 	n = idr_id & IDR_MASK;
899 	__clear_bit(n, &p->bitmap);
900 
901 	bitmap = (void *)p->ary[n];
902 	if (!test_bit(offset, bitmap->bitmap))
903 		goto err;
904 
905 	/* update bitmap and remove it if empty */
906 	__clear_bit(offset, bitmap->bitmap);
907 	if (--bitmap->nr_busy == 0) {
908 		__set_bit(n, &p->bitmap);	/* to please idr_remove() */
909 		idr_remove(&ida->idr, idr_id);
910 		free_bitmap(ida, bitmap);
911 	}
912 
913 	return;
914 
915  err:
916 	printk(KERN_WARNING
917 	       "ida_remove called for id=%d which is not allocated.\n", id);
918 }
919 EXPORT_SYMBOL(ida_remove);
920 
921 /**
922  * ida_destroy - release all cached layers within an ida tree
923  * @ida:		ida handle
924  */
925 void ida_destroy(struct ida *ida)
926 {
927 	idr_destroy(&ida->idr);
928 	kfree(ida->free_bitmap);
929 }
930 EXPORT_SYMBOL(ida_destroy);
931 
932 /**
933  * ida_simple_get - get a new id.
934  * @ida: the (initialized) ida.
935  * @start: the minimum id (inclusive, < 0x8000000)
936  * @end: the maximum id (exclusive, < 0x8000000 or 0)
937  * @gfp_mask: memory allocation flags
938  *
939  * Allocates an id in the range start <= id < end, or returns -ENOSPC.
940  * On memory allocation failure, returns -ENOMEM.
941  *
942  * Use ida_simple_remove() to get rid of an id.
943  */
944 int ida_simple_get(struct ida *ida, unsigned int start, unsigned int end,
945 		   gfp_t gfp_mask)
946 {
947 	int ret, id;
948 	unsigned int max;
949 	unsigned long flags;
950 
951 	BUG_ON((int)start < 0);
952 	BUG_ON((int)end < 0);
953 
954 	if (end == 0)
955 		max = 0x80000000;
956 	else {
957 		BUG_ON(end < start);
958 		max = end - 1;
959 	}
960 
961 again:
962 	if (!ida_pre_get(ida, gfp_mask))
963 		return -ENOMEM;
964 
965 	spin_lock_irqsave(&simple_ida_lock, flags);
966 	ret = ida_get_new_above(ida, start, &id);
967 	if (!ret) {
968 		if (id > max) {
969 			ida_remove(ida, id);
970 			ret = -ENOSPC;
971 		} else {
972 			ret = id;
973 		}
974 	}
975 	spin_unlock_irqrestore(&simple_ida_lock, flags);
976 
977 	if (unlikely(ret == -EAGAIN))
978 		goto again;
979 
980 	return ret;
981 }
982 EXPORT_SYMBOL(ida_simple_get);
983 
984 /**
985  * ida_simple_remove - remove an allocated id.
986  * @ida: the (initialized) ida.
987  * @id: the id returned by ida_simple_get.
988  */
989 void ida_simple_remove(struct ida *ida, unsigned int id)
990 {
991 	unsigned long flags;
992 
993 	BUG_ON((int)id < 0);
994 	spin_lock_irqsave(&simple_ida_lock, flags);
995 	ida_remove(ida, id);
996 	spin_unlock_irqrestore(&simple_ida_lock, flags);
997 }
998 EXPORT_SYMBOL(ida_simple_remove);
999 
1000 /**
1001  * ida_init - initialize ida handle
1002  * @ida:	ida handle
1003  *
1004  * This function is use to set up the handle (@ida) that you will pass
1005  * to the rest of the functions.
1006  */
1007 void ida_init(struct ida *ida)
1008 {
1009 	memset(ida, 0, sizeof(struct ida));
1010 	idr_init(&ida->idr);
1011 
1012 }
1013 EXPORT_SYMBOL(ida_init);
1014