nfs4client: convert to idr_alloc()
[linux-2.6-block.git] / lib / idr.c
CommitLineData
1da177e4
LT
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 *
3219b3b7
ND
9 * Modified by Nadia Derbey to make it RCU safe.
10 *
e15ae2dd 11 * Small id to pointer translation service.
1da177e4 12 *
e15ae2dd 13 * It uses a radix tree like structure as a sparse array indexed
1da177e4 14 * by the id to obtain the pointer. The bitmap makes allocating
e15ae2dd 15 * a new id quick.
1da177e4
LT
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
e15ae2dd 22 * You can release ids at any time. When all ids are released, most of
125c4c70 23 * the memory is returned (we keep MAX_IDR_FREE) in a local pool so we
e15ae2dd 24 * don't need to go to the memory "store" during an id allocate, just
1da177e4
LT
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>
8bc3bcc9 32#include <linux/export.h>
1da177e4 33#endif
5806f07c 34#include <linux/err.h>
1da177e4
LT
35#include <linux/string.h>
36#include <linux/idr.h>
88eca020 37#include <linux/spinlock.h>
d5c7409f
TH
38#include <linux/percpu.h>
39#include <linux/hardirq.h>
1da177e4 40
e18b890b 41static struct kmem_cache *idr_layer_cache;
d5c7409f
TH
42static DEFINE_PER_CPU(struct idr_layer *, idr_preload_head);
43static DEFINE_PER_CPU(int, idr_preload_cnt);
88eca020 44static DEFINE_SPINLOCK(simple_ida_lock);
1da177e4 45
4ae53789 46static struct idr_layer *get_from_free_list(struct idr *idp)
1da177e4
LT
47{
48 struct idr_layer *p;
c259cc28 49 unsigned long flags;
1da177e4 50
c259cc28 51 spin_lock_irqsave(&idp->lock, flags);
1da177e4
LT
52 if ((p = idp->id_free)) {
53 idp->id_free = p->ary[0];
54 idp->id_free_cnt--;
55 p->ary[0] = NULL;
56 }
c259cc28 57 spin_unlock_irqrestore(&idp->lock, flags);
1da177e4
LT
58 return(p);
59}
60
d5c7409f
TH
61/**
62 * idr_layer_alloc - allocate a new idr_layer
63 * @gfp_mask: allocation mask
64 * @layer_idr: optional idr to allocate from
65 *
66 * If @layer_idr is %NULL, directly allocate one using @gfp_mask or fetch
67 * one from the per-cpu preload buffer. If @layer_idr is not %NULL, fetch
68 * an idr_layer from @idr->id_free.
69 *
70 * @layer_idr is to maintain backward compatibility with the old alloc
71 * interface - idr_pre_get() and idr_get_new*() - and will be removed
72 * together with per-pool preload buffer.
73 */
74static struct idr_layer *idr_layer_alloc(gfp_t gfp_mask, struct idr *layer_idr)
75{
76 struct idr_layer *new;
77
78 /* this is the old path, bypass to get_from_free_list() */
79 if (layer_idr)
80 return get_from_free_list(layer_idr);
81
82 /* try to allocate directly from kmem_cache */
83 new = kmem_cache_zalloc(idr_layer_cache, gfp_mask);
84 if (new)
85 return new;
86
87 /*
88 * Try to fetch one from the per-cpu preload buffer if in process
89 * context. See idr_preload() for details.
90 */
91 if (in_interrupt())
92 return NULL;
93
94 preempt_disable();
95 new = __this_cpu_read(idr_preload_head);
96 if (new) {
97 __this_cpu_write(idr_preload_head, new->ary[0]);
98 __this_cpu_dec(idr_preload_cnt);
99 new->ary[0] = NULL;
100 }
101 preempt_enable();
102 return new;
103}
104
cf481c20
ND
105static void idr_layer_rcu_free(struct rcu_head *head)
106{
107 struct idr_layer *layer;
108
109 layer = container_of(head, struct idr_layer, rcu_head);
110 kmem_cache_free(idr_layer_cache, layer);
111}
112
113static inline void free_layer(struct idr_layer *p)
114{
115 call_rcu(&p->rcu_head, idr_layer_rcu_free);
116}
117
1eec0056 118/* only called when idp->lock is held */
4ae53789 119static void __move_to_free_list(struct idr *idp, struct idr_layer *p)
1eec0056
SR
120{
121 p->ary[0] = idp->id_free;
122 idp->id_free = p;
123 idp->id_free_cnt++;
124}
125
4ae53789 126static void move_to_free_list(struct idr *idp, struct idr_layer *p)
1da177e4 127{
c259cc28
RD
128 unsigned long flags;
129
1da177e4
LT
130 /*
131 * Depends on the return element being zeroed.
132 */
c259cc28 133 spin_lock_irqsave(&idp->lock, flags);
4ae53789 134 __move_to_free_list(idp, p);
c259cc28 135 spin_unlock_irqrestore(&idp->lock, flags);
1da177e4
LT
136}
137
e33ac8bd
TH
138static void idr_mark_full(struct idr_layer **pa, int id)
139{
140 struct idr_layer *p = pa[0];
141 int l = 0;
142
143 __set_bit(id & IDR_MASK, &p->bitmap);
144 /*
145 * If this layer is full mark the bit in the layer above to
146 * show that this part of the radix tree is full. This may
147 * complete the layer above and require walking up the radix
148 * tree.
149 */
150 while (p->bitmap == IDR_FULL) {
151 if (!(p = pa[++l]))
152 break;
153 id = id >> IDR_BITS;
154 __set_bit((id & IDR_MASK), &p->bitmap);
155 }
156}
157
1da177e4 158/**
56083ab1 159 * idr_pre_get - reserve resources for idr allocation
1da177e4
LT
160 * @idp: idr handle
161 * @gfp_mask: memory allocation flags
162 *
066a9be6
NA
163 * This function should be called prior to calling the idr_get_new* functions.
164 * It preallocates enough memory to satisfy the worst possible allocation. The
165 * caller should pass in GFP_KERNEL if possible. This of course requires that
166 * no spinning locks be held.
1da177e4 167 *
56083ab1
RD
168 * If the system is REALLY out of memory this function returns %0,
169 * otherwise %1.
1da177e4 170 */
fd4f2df2 171int idr_pre_get(struct idr *idp, gfp_t gfp_mask)
1da177e4 172{
125c4c70 173 while (idp->id_free_cnt < MAX_IDR_FREE) {
1da177e4 174 struct idr_layer *new;
5b019e99 175 new = kmem_cache_zalloc(idr_layer_cache, gfp_mask);
e15ae2dd 176 if (new == NULL)
1da177e4 177 return (0);
4ae53789 178 move_to_free_list(idp, new);
1da177e4
LT
179 }
180 return 1;
181}
182EXPORT_SYMBOL(idr_pre_get);
183
12d1b439
TH
184/**
185 * sub_alloc - try to allocate an id without growing the tree depth
186 * @idp: idr handle
187 * @starting_id: id to start search at
188 * @id: pointer to the allocated handle
189 * @pa: idr_layer[MAX_IDR_LEVEL] used as backtrack buffer
d5c7409f
TH
190 * @gfp_mask: allocation mask for idr_layer_alloc()
191 * @layer_idr: optional idr passed to idr_layer_alloc()
12d1b439
TH
192 *
193 * Allocate an id in range [@starting_id, INT_MAX] from @idp without
194 * growing its depth. Returns
195 *
196 * the allocated id >= 0 if successful,
197 * -EAGAIN if the tree needs to grow for allocation to succeed,
198 * -ENOSPC if the id space is exhausted,
199 * -ENOMEM if more idr_layers need to be allocated.
200 */
d5c7409f
TH
201static int sub_alloc(struct idr *idp, int *starting_id, struct idr_layer **pa,
202 gfp_t gfp_mask, struct idr *layer_idr)
1da177e4
LT
203{
204 int n, m, sh;
205 struct idr_layer *p, *new;
7aae6dd8 206 int l, id, oid;
5ba25331 207 unsigned long bm;
1da177e4
LT
208
209 id = *starting_id;
7aae6dd8 210 restart:
1da177e4
LT
211 p = idp->top;
212 l = idp->layers;
213 pa[l--] = NULL;
214 while (1) {
215 /*
216 * We run around this while until we reach the leaf node...
217 */
218 n = (id >> (IDR_BITS*l)) & IDR_MASK;
219 bm = ~p->bitmap;
220 m = find_next_bit(&bm, IDR_SIZE, n);
221 if (m == IDR_SIZE) {
222 /* no space available go back to previous layer. */
223 l++;
7aae6dd8 224 oid = id;
e15ae2dd 225 id = (id | ((1 << (IDR_BITS * l)) - 1)) + 1;
7aae6dd8
TH
226
227 /* if already at the top layer, we need to grow */
d2e7276b 228 if (id >= 1 << (idp->layers * IDR_BITS)) {
1da177e4 229 *starting_id = id;
12d1b439 230 return -EAGAIN;
1da177e4 231 }
d2e7276b
TH
232 p = pa[l];
233 BUG_ON(!p);
7aae6dd8
TH
234
235 /* If we need to go up one layer, continue the
236 * loop; otherwise, restart from the top.
237 */
238 sh = IDR_BITS * (l + 1);
239 if (oid >> sh == id >> sh)
240 continue;
241 else
242 goto restart;
1da177e4
LT
243 }
244 if (m != n) {
245 sh = IDR_BITS*l;
246 id = ((id >> sh) ^ n ^ m) << sh;
247 }
125c4c70 248 if ((id >= MAX_IDR_BIT) || (id < 0))
12d1b439 249 return -ENOSPC;
1da177e4
LT
250 if (l == 0)
251 break;
252 /*
253 * Create the layer below if it is missing.
254 */
255 if (!p->ary[m]) {
d5c7409f 256 new = idr_layer_alloc(gfp_mask, layer_idr);
4ae53789 257 if (!new)
12d1b439 258 return -ENOMEM;
6ff2d39b 259 new->layer = l-1;
3219b3b7 260 rcu_assign_pointer(p->ary[m], new);
1da177e4
LT
261 p->count++;
262 }
263 pa[l--] = p;
264 p = p->ary[m];
265 }
e33ac8bd
TH
266
267 pa[l] = p;
268 return id;
1da177e4
LT
269}
270
e33ac8bd 271static int idr_get_empty_slot(struct idr *idp, int starting_id,
d5c7409f
TH
272 struct idr_layer **pa, gfp_t gfp_mask,
273 struct idr *layer_idr)
1da177e4
LT
274{
275 struct idr_layer *p, *new;
276 int layers, v, id;
c259cc28 277 unsigned long flags;
e15ae2dd 278
1da177e4
LT
279 id = starting_id;
280build_up:
281 p = idp->top;
282 layers = idp->layers;
283 if (unlikely(!p)) {
d5c7409f 284 if (!(p = idr_layer_alloc(gfp_mask, layer_idr)))
12d1b439 285 return -ENOMEM;
6ff2d39b 286 p->layer = 0;
1da177e4
LT
287 layers = 1;
288 }
289 /*
290 * Add a new layer to the top of the tree if the requested
291 * id is larger than the currently allocated space.
292 */
125c4c70 293 while ((layers < (MAX_IDR_LEVEL - 1)) && (id >= (1 << (layers*IDR_BITS)))) {
1da177e4 294 layers++;
711a49a0
MS
295 if (!p->count) {
296 /* special case: if the tree is currently empty,
297 * then we grow the tree by moving the top node
298 * upwards.
299 */
300 p->layer++;
1da177e4 301 continue;
711a49a0 302 }
d5c7409f 303 if (!(new = idr_layer_alloc(gfp_mask, layer_idr))) {
1da177e4
LT
304 /*
305 * The allocation failed. If we built part of
306 * the structure tear it down.
307 */
c259cc28 308 spin_lock_irqsave(&idp->lock, flags);
1da177e4
LT
309 for (new = p; p && p != idp->top; new = p) {
310 p = p->ary[0];
311 new->ary[0] = NULL;
312 new->bitmap = new->count = 0;
4ae53789 313 __move_to_free_list(idp, new);
1da177e4 314 }
c259cc28 315 spin_unlock_irqrestore(&idp->lock, flags);
12d1b439 316 return -ENOMEM;
1da177e4
LT
317 }
318 new->ary[0] = p;
319 new->count = 1;
6ff2d39b 320 new->layer = layers-1;
1da177e4
LT
321 if (p->bitmap == IDR_FULL)
322 __set_bit(0, &new->bitmap);
323 p = new;
324 }
3219b3b7 325 rcu_assign_pointer(idp->top, p);
1da177e4 326 idp->layers = layers;
d5c7409f 327 v = sub_alloc(idp, &id, pa, gfp_mask, layer_idr);
12d1b439 328 if (v == -EAGAIN)
1da177e4
LT
329 goto build_up;
330 return(v);
331}
332
3594eb28
TH
333/*
334 * @id and @pa are from a successful allocation from idr_get_empty_slot().
335 * Install the user pointer @ptr and mark the slot full.
336 */
337static void idr_fill_slot(void *ptr, int id, struct idr_layer **pa)
e33ac8bd 338{
3594eb28
TH
339 rcu_assign_pointer(pa[0]->ary[id & IDR_MASK], (struct idr_layer *)ptr);
340 pa[0]->count++;
341 idr_mark_full(pa, id);
e33ac8bd
TH
342}
343
1da177e4 344/**
7c657f2f 345 * idr_get_new_above - allocate new idr entry above or equal to a start id
1da177e4 346 * @idp: idr handle
94e2bd68 347 * @ptr: pointer you want associated with the id
ea24ea85 348 * @starting_id: id to start search at
1da177e4
LT
349 * @id: pointer to the allocated handle
350 *
351 * This is the allocate id function. It should be called with any
352 * required locks.
353 *
066a9be6 354 * If allocation from IDR's private freelist fails, idr_get_new_above() will
56083ab1 355 * return %-EAGAIN. The caller should retry the idr_pre_get() call to refill
066a9be6
NA
356 * IDR's preallocation and then retry the idr_get_new_above() call.
357 *
56083ab1 358 * If the idr is full idr_get_new_above() will return %-ENOSPC.
1da177e4 359 *
56083ab1 360 * @id returns a value in the range @starting_id ... %0x7fffffff
1da177e4
LT
361 */
362int idr_get_new_above(struct idr *idp, void *ptr, int starting_id, int *id)
363{
3594eb28 364 struct idr_layer *pa[MAX_IDR_LEVEL];
1da177e4 365 int rv;
e15ae2dd 366
d5c7409f 367 rv = idr_get_empty_slot(idp, starting_id, pa, 0, idp);
944ca05c 368 if (rv < 0)
12d1b439 369 return rv == -ENOMEM ? -EAGAIN : rv;
3594eb28
TH
370
371 idr_fill_slot(ptr, rv, pa);
1da177e4
LT
372 *id = rv;
373 return 0;
374}
375EXPORT_SYMBOL(idr_get_new_above);
376
d5c7409f
TH
377/**
378 * idr_preload - preload for idr_alloc()
379 * @gfp_mask: allocation mask to use for preloading
380 *
381 * Preload per-cpu layer buffer for idr_alloc(). Can only be used from
382 * process context and each idr_preload() invocation should be matched with
383 * idr_preload_end(). Note that preemption is disabled while preloaded.
384 *
385 * The first idr_alloc() in the preloaded section can be treated as if it
386 * were invoked with @gfp_mask used for preloading. This allows using more
387 * permissive allocation masks for idrs protected by spinlocks.
388 *
389 * For example, if idr_alloc() below fails, the failure can be treated as
390 * if idr_alloc() were called with GFP_KERNEL rather than GFP_NOWAIT.
391 *
392 * idr_preload(GFP_KERNEL);
393 * spin_lock(lock);
394 *
395 * id = idr_alloc(idr, ptr, start, end, GFP_NOWAIT);
396 *
397 * spin_unlock(lock);
398 * idr_preload_end();
399 * if (id < 0)
400 * error;
401 */
402void idr_preload(gfp_t gfp_mask)
403{
404 /*
405 * Consuming preload buffer from non-process context breaks preload
406 * allocation guarantee. Disallow usage from those contexts.
407 */
408 WARN_ON_ONCE(in_interrupt());
409 might_sleep_if(gfp_mask & __GFP_WAIT);
410
411 preempt_disable();
412
413 /*
414 * idr_alloc() is likely to succeed w/o full idr_layer buffer and
415 * return value from idr_alloc() needs to be checked for failure
416 * anyway. Silently give up if allocation fails. The caller can
417 * treat failures from idr_alloc() as if idr_alloc() were called
418 * with @gfp_mask which should be enough.
419 */
420 while (__this_cpu_read(idr_preload_cnt) < MAX_IDR_FREE) {
421 struct idr_layer *new;
422
423 preempt_enable();
424 new = kmem_cache_zalloc(idr_layer_cache, gfp_mask);
425 preempt_disable();
426 if (!new)
427 break;
428
429 /* link the new one to per-cpu preload list */
430 new->ary[0] = __this_cpu_read(idr_preload_head);
431 __this_cpu_write(idr_preload_head, new);
432 __this_cpu_inc(idr_preload_cnt);
433 }
434}
435EXPORT_SYMBOL(idr_preload);
436
437/**
438 * idr_alloc - allocate new idr entry
439 * @idr: the (initialized) idr
440 * @ptr: pointer to be associated with the new id
441 * @start: the minimum id (inclusive)
442 * @end: the maximum id (exclusive, <= 0 for max)
443 * @gfp_mask: memory allocation flags
444 *
445 * Allocate an id in [start, end) and associate it with @ptr. If no ID is
446 * available in the specified range, returns -ENOSPC. On memory allocation
447 * failure, returns -ENOMEM.
448 *
449 * Note that @end is treated as max when <= 0. This is to always allow
450 * using @start + N as @end as long as N is inside integer range.
451 *
452 * The user is responsible for exclusively synchronizing all operations
453 * which may modify @idr. However, read-only accesses such as idr_find()
454 * or iteration can be performed under RCU read lock provided the user
455 * destroys @ptr in RCU-safe way after removal from idr.
456 */
457int idr_alloc(struct idr *idr, void *ptr, int start, int end, gfp_t gfp_mask)
458{
459 int max = end > 0 ? end - 1 : INT_MAX; /* inclusive upper limit */
460 struct idr_layer *pa[MAX_IDR_LEVEL];
461 int id;
462
463 might_sleep_if(gfp_mask & __GFP_WAIT);
464
465 /* sanity checks */
466 if (WARN_ON_ONCE(start < 0))
467 return -EINVAL;
468 if (unlikely(max < start))
469 return -ENOSPC;
470
471 /* allocate id */
472 id = idr_get_empty_slot(idr, start, pa, gfp_mask, NULL);
473 if (unlikely(id < 0))
474 return id;
475 if (unlikely(id > max))
476 return -ENOSPC;
477
478 idr_fill_slot(ptr, id, pa);
479 return id;
480}
481EXPORT_SYMBOL_GPL(idr_alloc);
482
1da177e4
LT
483static void idr_remove_warning(int id)
484{
f098ad65
ND
485 printk(KERN_WARNING
486 "idr_remove called for id=%d which is not allocated.\n", id);
1da177e4
LT
487 dump_stack();
488}
489
490static void sub_remove(struct idr *idp, int shift, int id)
491{
492 struct idr_layer *p = idp->top;
125c4c70 493 struct idr_layer **pa[MAX_IDR_LEVEL];
1da177e4 494 struct idr_layer ***paa = &pa[0];
cf481c20 495 struct idr_layer *to_free;
1da177e4
LT
496 int n;
497
498 *paa = NULL;
499 *++paa = &idp->top;
500
501 while ((shift > 0) && p) {
502 n = (id >> shift) & IDR_MASK;
503 __clear_bit(n, &p->bitmap);
504 *++paa = &p->ary[n];
505 p = p->ary[n];
506 shift -= IDR_BITS;
507 }
508 n = id & IDR_MASK;
509 if (likely(p != NULL && test_bit(n, &p->bitmap))){
510 __clear_bit(n, &p->bitmap);
cf481c20
ND
511 rcu_assign_pointer(p->ary[n], NULL);
512 to_free = NULL;
1da177e4 513 while(*paa && ! --((**paa)->count)){
cf481c20
ND
514 if (to_free)
515 free_layer(to_free);
516 to_free = **paa;
1da177e4
LT
517 **paa-- = NULL;
518 }
e15ae2dd 519 if (!*paa)
1da177e4 520 idp->layers = 0;
cf481c20
ND
521 if (to_free)
522 free_layer(to_free);
e15ae2dd 523 } else
1da177e4 524 idr_remove_warning(id);
1da177e4
LT
525}
526
527/**
56083ab1 528 * idr_remove - remove the given id and free its slot
72fd4a35
RD
529 * @idp: idr handle
530 * @id: unique key
1da177e4
LT
531 */
532void idr_remove(struct idr *idp, int id)
533{
534 struct idr_layer *p;
cf481c20 535 struct idr_layer *to_free;
1da177e4
LT
536
537 /* Mask off upper bits we don't use for the search. */
125c4c70 538 id &= MAX_IDR_MASK;
1da177e4
LT
539
540 sub_remove(idp, (idp->layers - 1) * IDR_BITS, id);
e15ae2dd 541 if (idp->top && idp->top->count == 1 && (idp->layers > 1) &&
cf481c20
ND
542 idp->top->ary[0]) {
543 /*
544 * Single child at leftmost slot: we can shrink the tree.
545 * This level is not needed anymore since when layers are
546 * inserted, they are inserted at the top of the existing
547 * tree.
548 */
549 to_free = idp->top;
1da177e4 550 p = idp->top->ary[0];
cf481c20 551 rcu_assign_pointer(idp->top, p);
1da177e4 552 --idp->layers;
cf481c20
ND
553 to_free->bitmap = to_free->count = 0;
554 free_layer(to_free);
1da177e4 555 }
125c4c70 556 while (idp->id_free_cnt >= MAX_IDR_FREE) {
4ae53789 557 p = get_from_free_list(idp);
cf481c20
ND
558 /*
559 * Note: we don't call the rcu callback here, since the only
560 * layers that fall into the freelist are those that have been
561 * preallocated.
562 */
1da177e4 563 kmem_cache_free(idr_layer_cache, p);
1da177e4 564 }
af8e2a4c 565 return;
1da177e4
LT
566}
567EXPORT_SYMBOL(idr_remove);
568
fe6e24ec 569void __idr_remove_all(struct idr *idp)
23936cc0 570{
6ace06dc 571 int n, id, max;
2dcb22b3 572 int bt_mask;
23936cc0 573 struct idr_layer *p;
125c4c70 574 struct idr_layer *pa[MAX_IDR_LEVEL];
23936cc0
KH
575 struct idr_layer **paa = &pa[0];
576
577 n = idp->layers * IDR_BITS;
578 p = idp->top;
1b23336a 579 rcu_assign_pointer(idp->top, NULL);
23936cc0
KH
580 max = 1 << n;
581
582 id = 0;
6ace06dc 583 while (id < max) {
23936cc0
KH
584 while (n > IDR_BITS && p) {
585 n -= IDR_BITS;
586 *paa++ = p;
587 p = p->ary[(id >> n) & IDR_MASK];
588 }
589
2dcb22b3 590 bt_mask = id;
23936cc0 591 id += 1 << n;
2dcb22b3
ID
592 /* Get the highest bit that the above add changed from 0->1. */
593 while (n < fls(id ^ bt_mask)) {
cf481c20
ND
594 if (p)
595 free_layer(p);
23936cc0
KH
596 n += IDR_BITS;
597 p = *--paa;
598 }
599 }
23936cc0
KH
600 idp->layers = 0;
601}
fe6e24ec 602EXPORT_SYMBOL(__idr_remove_all);
23936cc0 603
8d3b3591
AM
604/**
605 * idr_destroy - release all cached layers within an idr tree
ea24ea85 606 * @idp: idr handle
9bb26bc1
TH
607 *
608 * Free all id mappings and all idp_layers. After this function, @idp is
609 * completely unused and can be freed / recycled. The caller is
610 * responsible for ensuring that no one else accesses @idp during or after
611 * idr_destroy().
612 *
613 * A typical clean-up sequence for objects stored in an idr tree will use
614 * idr_for_each() to free all objects, if necessay, then idr_destroy() to
615 * free up the id mappings and cached idr_layers.
8d3b3591
AM
616 */
617void idr_destroy(struct idr *idp)
618{
fe6e24ec 619 __idr_remove_all(idp);
9bb26bc1 620
8d3b3591 621 while (idp->id_free_cnt) {
4ae53789 622 struct idr_layer *p = get_from_free_list(idp);
8d3b3591
AM
623 kmem_cache_free(idr_layer_cache, p);
624 }
625}
626EXPORT_SYMBOL(idr_destroy);
627
1da177e4
LT
628/**
629 * idr_find - return pointer for given id
630 * @idp: idr handle
631 * @id: lookup key
632 *
633 * Return the pointer given the id it has been registered with. A %NULL
634 * return indicates that @id is not valid or you passed %NULL in
635 * idr_get_new().
636 *
f9c46d6e
ND
637 * This function can be called under rcu_read_lock(), given that the leaf
638 * pointers lifetimes are correctly managed.
1da177e4
LT
639 */
640void *idr_find(struct idr *idp, int id)
641{
642 int n;
643 struct idr_layer *p;
644
96be753a 645 p = rcu_dereference_raw(idp->top);
6ff2d39b
MS
646 if (!p)
647 return NULL;
648 n = (p->layer+1) * IDR_BITS;
1da177e4
LT
649
650 /* Mask off upper bits we don't use for the search. */
125c4c70 651 id &= MAX_IDR_MASK;
1da177e4
LT
652
653 if (id >= (1 << n))
654 return NULL;
6ff2d39b 655 BUG_ON(n == 0);
1da177e4
LT
656
657 while (n > 0 && p) {
658 n -= IDR_BITS;
6ff2d39b 659 BUG_ON(n != p->layer*IDR_BITS);
96be753a 660 p = rcu_dereference_raw(p->ary[(id >> n) & IDR_MASK]);
1da177e4
LT
661 }
662 return((void *)p);
663}
664EXPORT_SYMBOL(idr_find);
665
96d7fa42
KH
666/**
667 * idr_for_each - iterate through all stored pointers
668 * @idp: idr handle
669 * @fn: function to be called for each pointer
670 * @data: data passed back to callback function
671 *
672 * Iterate over the pointers registered with the given idr. The
673 * callback function will be called for each pointer currently
674 * registered, passing the id, the pointer and the data pointer passed
675 * to this function. It is not safe to modify the idr tree while in
676 * the callback, so functions such as idr_get_new and idr_remove are
677 * not allowed.
678 *
679 * We check the return of @fn each time. If it returns anything other
56083ab1 680 * than %0, we break out and return that value.
96d7fa42
KH
681 *
682 * The caller must serialize idr_for_each() vs idr_get_new() and idr_remove().
683 */
684int idr_for_each(struct idr *idp,
685 int (*fn)(int id, void *p, void *data), void *data)
686{
687 int n, id, max, error = 0;
688 struct idr_layer *p;
125c4c70 689 struct idr_layer *pa[MAX_IDR_LEVEL];
96d7fa42
KH
690 struct idr_layer **paa = &pa[0];
691
692 n = idp->layers * IDR_BITS;
96be753a 693 p = rcu_dereference_raw(idp->top);
96d7fa42
KH
694 max = 1 << n;
695
696 id = 0;
697 while (id < max) {
698 while (n > 0 && p) {
699 n -= IDR_BITS;
700 *paa++ = p;
96be753a 701 p = rcu_dereference_raw(p->ary[(id >> n) & IDR_MASK]);
96d7fa42
KH
702 }
703
704 if (p) {
705 error = fn(id, (void *)p, data);
706 if (error)
707 break;
708 }
709
710 id += 1 << n;
711 while (n < fls(id)) {
712 n += IDR_BITS;
713 p = *--paa;
714 }
715 }
716
717 return error;
718}
719EXPORT_SYMBOL(idr_for_each);
720
38460b48
KH
721/**
722 * idr_get_next - lookup next object of id to given id.
723 * @idp: idr handle
ea24ea85 724 * @nextidp: pointer to lookup key
38460b48
KH
725 *
726 * Returns pointer to registered object with id, which is next number to
1458ce16
NA
727 * given id. After being looked up, *@nextidp will be updated for the next
728 * iteration.
9f7de827
HD
729 *
730 * This function can be called under rcu_read_lock(), given that the leaf
731 * pointers lifetimes are correctly managed.
38460b48 732 */
38460b48
KH
733void *idr_get_next(struct idr *idp, int *nextidp)
734{
125c4c70 735 struct idr_layer *p, *pa[MAX_IDR_LEVEL];
38460b48
KH
736 struct idr_layer **paa = &pa[0];
737 int id = *nextidp;
738 int n, max;
739
740 /* find first ent */
94bfa3b6 741 p = rcu_dereference_raw(idp->top);
38460b48
KH
742 if (!p)
743 return NULL;
9f7de827
HD
744 n = (p->layer + 1) * IDR_BITS;
745 max = 1 << n;
38460b48
KH
746
747 while (id < max) {
748 while (n > 0 && p) {
749 n -= IDR_BITS;
750 *paa++ = p;
94bfa3b6 751 p = rcu_dereference_raw(p->ary[(id >> n) & IDR_MASK]);
38460b48
KH
752 }
753
754 if (p) {
755 *nextidp = id;
756 return p;
757 }
758
6cdae741
TH
759 /*
760 * Proceed to the next layer at the current level. Unlike
761 * idr_for_each(), @id isn't guaranteed to be aligned to
762 * layer boundary at this point and adding 1 << n may
763 * incorrectly skip IDs. Make sure we jump to the
764 * beginning of the next layer using round_up().
765 */
766 id = round_up(id + 1, 1 << n);
38460b48
KH
767 while (n < fls(id)) {
768 n += IDR_BITS;
769 p = *--paa;
770 }
771 }
772 return NULL;
773}
4d1ee80f 774EXPORT_SYMBOL(idr_get_next);
38460b48
KH
775
776
5806f07c
JM
777/**
778 * idr_replace - replace pointer for given id
779 * @idp: idr handle
780 * @ptr: pointer you want associated with the id
781 * @id: lookup key
782 *
783 * Replace the pointer registered with an id and return the old value.
56083ab1
RD
784 * A %-ENOENT return indicates that @id was not found.
785 * A %-EINVAL return indicates that @id was not within valid constraints.
5806f07c 786 *
cf481c20 787 * The caller must serialize with writers.
5806f07c
JM
788 */
789void *idr_replace(struct idr *idp, void *ptr, int id)
790{
791 int n;
792 struct idr_layer *p, *old_p;
793
5806f07c 794 p = idp->top;
6ff2d39b
MS
795 if (!p)
796 return ERR_PTR(-EINVAL);
797
798 n = (p->layer+1) * IDR_BITS;
5806f07c 799
125c4c70 800 id &= MAX_IDR_MASK;
5806f07c
JM
801
802 if (id >= (1 << n))
803 return ERR_PTR(-EINVAL);
804
805 n -= IDR_BITS;
806 while ((n > 0) && p) {
807 p = p->ary[(id >> n) & IDR_MASK];
808 n -= IDR_BITS;
809 }
810
811 n = id & IDR_MASK;
812 if (unlikely(p == NULL || !test_bit(n, &p->bitmap)))
813 return ERR_PTR(-ENOENT);
814
815 old_p = p->ary[n];
cf481c20 816 rcu_assign_pointer(p->ary[n], ptr);
5806f07c
JM
817
818 return old_p;
819}
820EXPORT_SYMBOL(idr_replace);
821
199f0ca5 822void __init idr_init_cache(void)
1da177e4 823{
199f0ca5 824 idr_layer_cache = kmem_cache_create("idr_layer_cache",
5b019e99 825 sizeof(struct idr_layer), 0, SLAB_PANIC, NULL);
1da177e4
LT
826}
827
828/**
829 * idr_init - initialize idr handle
830 * @idp: idr handle
831 *
832 * This function is use to set up the handle (@idp) that you will pass
833 * to the rest of the functions.
834 */
835void idr_init(struct idr *idp)
836{
1da177e4
LT
837 memset(idp, 0, sizeof(struct idr));
838 spin_lock_init(&idp->lock);
839}
840EXPORT_SYMBOL(idr_init);
72dba584
TH
841
842
56083ab1
RD
843/**
844 * DOC: IDA description
72dba584
TH
845 * IDA - IDR based ID allocator
846 *
56083ab1 847 * This is id allocator without id -> pointer translation. Memory
72dba584
TH
848 * usage is much lower than full blown idr because each id only
849 * occupies a bit. ida uses a custom leaf node which contains
850 * IDA_BITMAP_BITS slots.
851 *
852 * 2007-04-25 written by Tejun Heo <htejun@gmail.com>
853 */
854
855static void free_bitmap(struct ida *ida, struct ida_bitmap *bitmap)
856{
857 unsigned long flags;
858
859 if (!ida->free_bitmap) {
860 spin_lock_irqsave(&ida->idr.lock, flags);
861 if (!ida->free_bitmap) {
862 ida->free_bitmap = bitmap;
863 bitmap = NULL;
864 }
865 spin_unlock_irqrestore(&ida->idr.lock, flags);
866 }
867
868 kfree(bitmap);
869}
870
871/**
872 * ida_pre_get - reserve resources for ida allocation
873 * @ida: ida handle
874 * @gfp_mask: memory allocation flag
875 *
876 * This function should be called prior to locking and calling the
877 * following function. It preallocates enough memory to satisfy the
878 * worst possible allocation.
879 *
56083ab1
RD
880 * If the system is REALLY out of memory this function returns %0,
881 * otherwise %1.
72dba584
TH
882 */
883int ida_pre_get(struct ida *ida, gfp_t gfp_mask)
884{
885 /* allocate idr_layers */
886 if (!idr_pre_get(&ida->idr, gfp_mask))
887 return 0;
888
889 /* allocate free_bitmap */
890 if (!ida->free_bitmap) {
891 struct ida_bitmap *bitmap;
892
893 bitmap = kmalloc(sizeof(struct ida_bitmap), gfp_mask);
894 if (!bitmap)
895 return 0;
896
897 free_bitmap(ida, bitmap);
898 }
899
900 return 1;
901}
902EXPORT_SYMBOL(ida_pre_get);
903
904/**
905 * ida_get_new_above - allocate new ID above or equal to a start id
906 * @ida: ida handle
ea24ea85 907 * @starting_id: id to start search at
72dba584
TH
908 * @p_id: pointer to the allocated handle
909 *
e3816c54
WSH
910 * Allocate new ID above or equal to @starting_id. It should be called
911 * with any required locks.
72dba584 912 *
56083ab1 913 * If memory is required, it will return %-EAGAIN, you should unlock
72dba584 914 * and go back to the ida_pre_get() call. If the ida is full, it will
56083ab1 915 * return %-ENOSPC.
72dba584 916 *
56083ab1 917 * @p_id returns a value in the range @starting_id ... %0x7fffffff.
72dba584
TH
918 */
919int ida_get_new_above(struct ida *ida, int starting_id, int *p_id)
920{
125c4c70 921 struct idr_layer *pa[MAX_IDR_LEVEL];
72dba584
TH
922 struct ida_bitmap *bitmap;
923 unsigned long flags;
924 int idr_id = starting_id / IDA_BITMAP_BITS;
925 int offset = starting_id % IDA_BITMAP_BITS;
926 int t, id;
927
928 restart:
929 /* get vacant slot */
d5c7409f 930 t = idr_get_empty_slot(&ida->idr, idr_id, pa, 0, &ida->idr);
944ca05c 931 if (t < 0)
12d1b439 932 return t == -ENOMEM ? -EAGAIN : t;
72dba584 933
125c4c70 934 if (t * IDA_BITMAP_BITS >= MAX_IDR_BIT)
72dba584
TH
935 return -ENOSPC;
936
937 if (t != idr_id)
938 offset = 0;
939 idr_id = t;
940
941 /* if bitmap isn't there, create a new one */
942 bitmap = (void *)pa[0]->ary[idr_id & IDR_MASK];
943 if (!bitmap) {
944 spin_lock_irqsave(&ida->idr.lock, flags);
945 bitmap = ida->free_bitmap;
946 ida->free_bitmap = NULL;
947 spin_unlock_irqrestore(&ida->idr.lock, flags);
948
949 if (!bitmap)
950 return -EAGAIN;
951
952 memset(bitmap, 0, sizeof(struct ida_bitmap));
3219b3b7
ND
953 rcu_assign_pointer(pa[0]->ary[idr_id & IDR_MASK],
954 (void *)bitmap);
72dba584
TH
955 pa[0]->count++;
956 }
957
958 /* lookup for empty slot */
959 t = find_next_zero_bit(bitmap->bitmap, IDA_BITMAP_BITS, offset);
960 if (t == IDA_BITMAP_BITS) {
961 /* no empty slot after offset, continue to the next chunk */
962 idr_id++;
963 offset = 0;
964 goto restart;
965 }
966
967 id = idr_id * IDA_BITMAP_BITS + t;
125c4c70 968 if (id >= MAX_IDR_BIT)
72dba584
TH
969 return -ENOSPC;
970
971 __set_bit(t, bitmap->bitmap);
972 if (++bitmap->nr_busy == IDA_BITMAP_BITS)
973 idr_mark_full(pa, idr_id);
974
975 *p_id = id;
976
977 /* Each leaf node can handle nearly a thousand slots and the
978 * whole idea of ida is to have small memory foot print.
979 * Throw away extra resources one by one after each successful
980 * allocation.
981 */
982 if (ida->idr.id_free_cnt || ida->free_bitmap) {
4ae53789 983 struct idr_layer *p = get_from_free_list(&ida->idr);
72dba584
TH
984 if (p)
985 kmem_cache_free(idr_layer_cache, p);
986 }
987
988 return 0;
989}
990EXPORT_SYMBOL(ida_get_new_above);
991
72dba584
TH
992/**
993 * ida_remove - remove the given ID
994 * @ida: ida handle
995 * @id: ID to free
996 */
997void ida_remove(struct ida *ida, int id)
998{
999 struct idr_layer *p = ida->idr.top;
1000 int shift = (ida->idr.layers - 1) * IDR_BITS;
1001 int idr_id = id / IDA_BITMAP_BITS;
1002 int offset = id % IDA_BITMAP_BITS;
1003 int n;
1004 struct ida_bitmap *bitmap;
1005
1006 /* clear full bits while looking up the leaf idr_layer */
1007 while ((shift > 0) && p) {
1008 n = (idr_id >> shift) & IDR_MASK;
1009 __clear_bit(n, &p->bitmap);
1010 p = p->ary[n];
1011 shift -= IDR_BITS;
1012 }
1013
1014 if (p == NULL)
1015 goto err;
1016
1017 n = idr_id & IDR_MASK;
1018 __clear_bit(n, &p->bitmap);
1019
1020 bitmap = (void *)p->ary[n];
1021 if (!test_bit(offset, bitmap->bitmap))
1022 goto err;
1023
1024 /* update bitmap and remove it if empty */
1025 __clear_bit(offset, bitmap->bitmap);
1026 if (--bitmap->nr_busy == 0) {
1027 __set_bit(n, &p->bitmap); /* to please idr_remove() */
1028 idr_remove(&ida->idr, idr_id);
1029 free_bitmap(ida, bitmap);
1030 }
1031
1032 return;
1033
1034 err:
1035 printk(KERN_WARNING
1036 "ida_remove called for id=%d which is not allocated.\n", id);
1037}
1038EXPORT_SYMBOL(ida_remove);
1039
1040/**
1041 * ida_destroy - release all cached layers within an ida tree
ea24ea85 1042 * @ida: ida handle
72dba584
TH
1043 */
1044void ida_destroy(struct ida *ida)
1045{
1046 idr_destroy(&ida->idr);
1047 kfree(ida->free_bitmap);
1048}
1049EXPORT_SYMBOL(ida_destroy);
1050
88eca020
RR
1051/**
1052 * ida_simple_get - get a new id.
1053 * @ida: the (initialized) ida.
1054 * @start: the minimum id (inclusive, < 0x8000000)
1055 * @end: the maximum id (exclusive, < 0x8000000 or 0)
1056 * @gfp_mask: memory allocation flags
1057 *
1058 * Allocates an id in the range start <= id < end, or returns -ENOSPC.
1059 * On memory allocation failure, returns -ENOMEM.
1060 *
1061 * Use ida_simple_remove() to get rid of an id.
1062 */
1063int ida_simple_get(struct ida *ida, unsigned int start, unsigned int end,
1064 gfp_t gfp_mask)
1065{
1066 int ret, id;
1067 unsigned int max;
46cbc1d3 1068 unsigned long flags;
88eca020
RR
1069
1070 BUG_ON((int)start < 0);
1071 BUG_ON((int)end < 0);
1072
1073 if (end == 0)
1074 max = 0x80000000;
1075 else {
1076 BUG_ON(end < start);
1077 max = end - 1;
1078 }
1079
1080again:
1081 if (!ida_pre_get(ida, gfp_mask))
1082 return -ENOMEM;
1083
46cbc1d3 1084 spin_lock_irqsave(&simple_ida_lock, flags);
88eca020
RR
1085 ret = ida_get_new_above(ida, start, &id);
1086 if (!ret) {
1087 if (id > max) {
1088 ida_remove(ida, id);
1089 ret = -ENOSPC;
1090 } else {
1091 ret = id;
1092 }
1093 }
46cbc1d3 1094 spin_unlock_irqrestore(&simple_ida_lock, flags);
88eca020
RR
1095
1096 if (unlikely(ret == -EAGAIN))
1097 goto again;
1098
1099 return ret;
1100}
1101EXPORT_SYMBOL(ida_simple_get);
1102
1103/**
1104 * ida_simple_remove - remove an allocated id.
1105 * @ida: the (initialized) ida.
1106 * @id: the id returned by ida_simple_get.
1107 */
1108void ida_simple_remove(struct ida *ida, unsigned int id)
1109{
46cbc1d3
TH
1110 unsigned long flags;
1111
88eca020 1112 BUG_ON((int)id < 0);
46cbc1d3 1113 spin_lock_irqsave(&simple_ida_lock, flags);
88eca020 1114 ida_remove(ida, id);
46cbc1d3 1115 spin_unlock_irqrestore(&simple_ida_lock, flags);
88eca020
RR
1116}
1117EXPORT_SYMBOL(ida_simple_remove);
1118
72dba584
TH
1119/**
1120 * ida_init - initialize ida handle
1121 * @ida: ida handle
1122 *
1123 * This function is use to set up the handle (@ida) that you will pass
1124 * to the rest of the functions.
1125 */
1126void ida_init(struct ida *ida)
1127{
1128 memset(ida, 0, sizeof(struct ida));
1129 idr_init(&ida->idr);
1130
1131}
1132EXPORT_SYMBOL(ida_init);