]>
Commit | Line | Data |
---|---|---|
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 | #include <linux/percpu.h> | |
39 | #include <linux/hardirq.h> | |
40 | ||
41 | #define MAX_IDR_SHIFT (sizeof(int) * 8 - 1) | |
42 | #define MAX_IDR_BIT (1U << MAX_IDR_SHIFT) | |
43 | ||
44 | /* Leave the possibility of an incomplete final layer */ | |
45 | #define MAX_IDR_LEVEL ((MAX_IDR_SHIFT + IDR_BITS - 1) / IDR_BITS) | |
46 | ||
47 | /* Number of id_layer structs to leave in free list */ | |
48 | #define MAX_IDR_FREE (MAX_IDR_LEVEL * 2) | |
49 | ||
50 | static struct kmem_cache *idr_layer_cache; | |
51 | static DEFINE_PER_CPU(struct idr_layer *, idr_preload_head); | |
52 | static DEFINE_PER_CPU(int, idr_preload_cnt); | |
53 | static DEFINE_SPINLOCK(simple_ida_lock); | |
54 | ||
55 | /* the maximum ID which can be allocated given idr->layers */ | |
56 | static int idr_max(int layers) | |
57 | { | |
58 | int bits = min_t(int, layers * IDR_BITS, MAX_IDR_SHIFT); | |
59 | ||
60 | return (1 << bits) - 1; | |
61 | } | |
62 | ||
63 | /* | |
64 | * Prefix mask for an idr_layer at @layer. For layer 0, the prefix mask is | |
65 | * all bits except for the lower IDR_BITS. For layer 1, 2 * IDR_BITS, and | |
66 | * so on. | |
67 | */ | |
68 | static int idr_layer_prefix_mask(int layer) | |
69 | { | |
70 | return ~idr_max(layer + 1); | |
71 | } | |
72 | ||
73 | static struct idr_layer *get_from_free_list(struct idr *idp) | |
74 | { | |
75 | struct idr_layer *p; | |
76 | unsigned long flags; | |
77 | ||
78 | spin_lock_irqsave(&idp->lock, flags); | |
79 | if ((p = idp->id_free)) { | |
80 | idp->id_free = p->ary[0]; | |
81 | idp->id_free_cnt--; | |
82 | p->ary[0] = NULL; | |
83 | } | |
84 | spin_unlock_irqrestore(&idp->lock, flags); | |
85 | return(p); | |
86 | } | |
87 | ||
88 | /** | |
89 | * idr_layer_alloc - allocate a new idr_layer | |
90 | * @gfp_mask: allocation mask | |
91 | * @layer_idr: optional idr to allocate from | |
92 | * | |
93 | * If @layer_idr is %NULL, directly allocate one using @gfp_mask or fetch | |
94 | * one from the per-cpu preload buffer. If @layer_idr is not %NULL, fetch | |
95 | * an idr_layer from @idr->id_free. | |
96 | * | |
97 | * @layer_idr is to maintain backward compatibility with the old alloc | |
98 | * interface - idr_pre_get() and idr_get_new*() - and will be removed | |
99 | * together with per-pool preload buffer. | |
100 | */ | |
101 | static struct idr_layer *idr_layer_alloc(gfp_t gfp_mask, struct idr *layer_idr) | |
102 | { | |
103 | struct idr_layer *new; | |
104 | ||
105 | /* this is the old path, bypass to get_from_free_list() */ | |
106 | if (layer_idr) | |
107 | return get_from_free_list(layer_idr); | |
108 | ||
109 | /* | |
110 | * Try to allocate directly from kmem_cache. We want to try this | |
111 | * before preload buffer; otherwise, non-preloading idr_alloc() | |
112 | * users will end up taking advantage of preloading ones. As the | |
113 | * following is allowed to fail for preloaded cases, suppress | |
114 | * warning this time. | |
115 | */ | |
116 | new = kmem_cache_zalloc(idr_layer_cache, gfp_mask | __GFP_NOWARN); | |
117 | if (new) | |
118 | return new; | |
119 | ||
120 | /* | |
121 | * Try to fetch one from the per-cpu preload buffer if in process | |
122 | * context. See idr_preload() for details. | |
123 | */ | |
124 | if (!in_interrupt()) { | |
125 | preempt_disable(); | |
126 | new = __this_cpu_read(idr_preload_head); | |
127 | if (new) { | |
128 | __this_cpu_write(idr_preload_head, new->ary[0]); | |
129 | __this_cpu_dec(idr_preload_cnt); | |
130 | new->ary[0] = NULL; | |
131 | } | |
132 | preempt_enable(); | |
133 | if (new) | |
134 | return new; | |
135 | } | |
136 | ||
137 | /* | |
138 | * Both failed. Try kmem_cache again w/o adding __GFP_NOWARN so | |
139 | * that memory allocation failure warning is printed as intended. | |
140 | */ | |
141 | return kmem_cache_zalloc(idr_layer_cache, gfp_mask); | |
142 | } | |
143 | ||
144 | static void idr_layer_rcu_free(struct rcu_head *head) | |
145 | { | |
146 | struct idr_layer *layer; | |
147 | ||
148 | layer = container_of(head, struct idr_layer, rcu_head); | |
149 | kmem_cache_free(idr_layer_cache, layer); | |
150 | } | |
151 | ||
152 | static inline void free_layer(struct idr *idr, struct idr_layer *p) | |
153 | { | |
154 | if (idr->hint && idr->hint == p) | |
155 | RCU_INIT_POINTER(idr->hint, NULL); | |
156 | call_rcu(&p->rcu_head, idr_layer_rcu_free); | |
157 | } | |
158 | ||
159 | /* only called when idp->lock is held */ | |
160 | static void __move_to_free_list(struct idr *idp, struct idr_layer *p) | |
161 | { | |
162 | p->ary[0] = idp->id_free; | |
163 | idp->id_free = p; | |
164 | idp->id_free_cnt++; | |
165 | } | |
166 | ||
167 | static void move_to_free_list(struct idr *idp, struct idr_layer *p) | |
168 | { | |
169 | unsigned long flags; | |
170 | ||
171 | /* | |
172 | * Depends on the return element being zeroed. | |
173 | */ | |
174 | spin_lock_irqsave(&idp->lock, flags); | |
175 | __move_to_free_list(idp, p); | |
176 | spin_unlock_irqrestore(&idp->lock, flags); | |
177 | } | |
178 | ||
179 | static void idr_mark_full(struct idr_layer **pa, int id) | |
180 | { | |
181 | struct idr_layer *p = pa[0]; | |
182 | int l = 0; | |
183 | ||
184 | __set_bit(id & IDR_MASK, p->bitmap); | |
185 | /* | |
186 | * If this layer is full mark the bit in the layer above to | |
187 | * show that this part of the radix tree is full. This may | |
188 | * complete the layer above and require walking up the radix | |
189 | * tree. | |
190 | */ | |
191 | while (bitmap_full(p->bitmap, IDR_SIZE)) { | |
192 | if (!(p = pa[++l])) | |
193 | break; | |
194 | id = id >> IDR_BITS; | |
195 | __set_bit((id & IDR_MASK), p->bitmap); | |
196 | } | |
197 | } | |
198 | ||
199 | int __idr_pre_get(struct idr *idp, gfp_t gfp_mask) | |
200 | { | |
201 | while (idp->id_free_cnt < MAX_IDR_FREE) { | |
202 | struct idr_layer *new; | |
203 | new = kmem_cache_zalloc(idr_layer_cache, gfp_mask); | |
204 | if (new == NULL) | |
205 | return (0); | |
206 | move_to_free_list(idp, new); | |
207 | } | |
208 | return 1; | |
209 | } | |
210 | EXPORT_SYMBOL(__idr_pre_get); | |
211 | ||
212 | /** | |
213 | * sub_alloc - try to allocate an id without growing the tree depth | |
214 | * @idp: idr handle | |
215 | * @starting_id: id to start search at | |
216 | * @pa: idr_layer[MAX_IDR_LEVEL] used as backtrack buffer | |
217 | * @gfp_mask: allocation mask for idr_layer_alloc() | |
218 | * @layer_idr: optional idr passed to idr_layer_alloc() | |
219 | * | |
220 | * Allocate an id in range [@starting_id, INT_MAX] from @idp without | |
221 | * growing its depth. Returns | |
222 | * | |
223 | * the allocated id >= 0 if successful, | |
224 | * -EAGAIN if the tree needs to grow for allocation to succeed, | |
225 | * -ENOSPC if the id space is exhausted, | |
226 | * -ENOMEM if more idr_layers need to be allocated. | |
227 | */ | |
228 | static int sub_alloc(struct idr *idp, int *starting_id, struct idr_layer **pa, | |
229 | gfp_t gfp_mask, struct idr *layer_idr) | |
230 | { | |
231 | int n, m, sh; | |
232 | struct idr_layer *p, *new; | |
233 | int l, id, oid; | |
234 | ||
235 | id = *starting_id; | |
236 | restart: | |
237 | p = idp->top; | |
238 | l = idp->layers; | |
239 | pa[l--] = NULL; | |
240 | while (1) { | |
241 | /* | |
242 | * We run around this while until we reach the leaf node... | |
243 | */ | |
244 | n = (id >> (IDR_BITS*l)) & IDR_MASK; | |
245 | m = find_next_zero_bit(p->bitmap, IDR_SIZE, n); | |
246 | if (m == IDR_SIZE) { | |
247 | /* no space available go back to previous layer. */ | |
248 | l++; | |
249 | oid = id; | |
250 | id = (id | ((1 << (IDR_BITS * l)) - 1)) + 1; | |
251 | ||
252 | /* if already at the top layer, we need to grow */ | |
253 | if (id >= 1 << (idp->layers * IDR_BITS)) { | |
254 | *starting_id = id; | |
255 | return -EAGAIN; | |
256 | } | |
257 | p = pa[l]; | |
258 | BUG_ON(!p); | |
259 | ||
260 | /* If we need to go up one layer, continue the | |
261 | * loop; otherwise, restart from the top. | |
262 | */ | |
263 | sh = IDR_BITS * (l + 1); | |
264 | if (oid >> sh == id >> sh) | |
265 | continue; | |
266 | else | |
267 | goto restart; | |
268 | } | |
269 | if (m != n) { | |
270 | sh = IDR_BITS*l; | |
271 | id = ((id >> sh) ^ n ^ m) << sh; | |
272 | } | |
273 | if ((id >= MAX_IDR_BIT) || (id < 0)) | |
274 | return -ENOSPC; | |
275 | if (l == 0) | |
276 | break; | |
277 | /* | |
278 | * Create the layer below if it is missing. | |
279 | */ | |
280 | if (!p->ary[m]) { | |
281 | new = idr_layer_alloc(gfp_mask, layer_idr); | |
282 | if (!new) | |
283 | return -ENOMEM; | |
284 | new->layer = l-1; | |
285 | new->prefix = id & idr_layer_prefix_mask(new->layer); | |
286 | rcu_assign_pointer(p->ary[m], new); | |
287 | p->count++; | |
288 | } | |
289 | pa[l--] = p; | |
290 | p = p->ary[m]; | |
291 | } | |
292 | ||
293 | pa[l] = p; | |
294 | return id; | |
295 | } | |
296 | ||
297 | static int idr_get_empty_slot(struct idr *idp, int starting_id, | |
298 | struct idr_layer **pa, gfp_t gfp_mask, | |
299 | struct idr *layer_idr) | |
300 | { | |
301 | struct idr_layer *p, *new; | |
302 | int layers, v, id; | |
303 | unsigned long flags; | |
304 | ||
305 | id = starting_id; | |
306 | build_up: | |
307 | p = idp->top; | |
308 | layers = idp->layers; | |
309 | if (unlikely(!p)) { | |
310 | if (!(p = idr_layer_alloc(gfp_mask, layer_idr))) | |
311 | return -ENOMEM; | |
312 | p->layer = 0; | |
313 | layers = 1; | |
314 | } | |
315 | /* | |
316 | * Add a new layer to the top of the tree if the requested | |
317 | * id is larger than the currently allocated space. | |
318 | */ | |
319 | while (id > idr_max(layers)) { | |
320 | layers++; | |
321 | if (!p->count) { | |
322 | /* special case: if the tree is currently empty, | |
323 | * then we grow the tree by moving the top node | |
324 | * upwards. | |
325 | */ | |
326 | p->layer++; | |
327 | WARN_ON_ONCE(p->prefix); | |
328 | continue; | |
329 | } | |
330 | if (!(new = idr_layer_alloc(gfp_mask, layer_idr))) { | |
331 | /* | |
332 | * The allocation failed. If we built part of | |
333 | * the structure tear it down. | |
334 | */ | |
335 | spin_lock_irqsave(&idp->lock, flags); | |
336 | for (new = p; p && p != idp->top; new = p) { | |
337 | p = p->ary[0]; | |
338 | new->ary[0] = NULL; | |
339 | new->count = 0; | |
340 | bitmap_clear(new->bitmap, 0, IDR_SIZE); | |
341 | __move_to_free_list(idp, new); | |
342 | } | |
343 | spin_unlock_irqrestore(&idp->lock, flags); | |
344 | return -ENOMEM; | |
345 | } | |
346 | new->ary[0] = p; | |
347 | new->count = 1; | |
348 | new->layer = layers-1; | |
349 | new->prefix = id & idr_layer_prefix_mask(new->layer); | |
350 | if (bitmap_full(p->bitmap, IDR_SIZE)) | |
351 | __set_bit(0, new->bitmap); | |
352 | p = new; | |
353 | } | |
354 | rcu_assign_pointer(idp->top, p); | |
355 | idp->layers = layers; | |
356 | v = sub_alloc(idp, &id, pa, gfp_mask, layer_idr); | |
357 | if (v == -EAGAIN) | |
358 | goto build_up; | |
359 | return(v); | |
360 | } | |
361 | ||
362 | /* | |
363 | * @id and @pa are from a successful allocation from idr_get_empty_slot(). | |
364 | * Install the user pointer @ptr and mark the slot full. | |
365 | */ | |
366 | static void idr_fill_slot(struct idr *idr, void *ptr, int id, | |
367 | struct idr_layer **pa) | |
368 | { | |
369 | /* update hint used for lookup, cleared from free_layer() */ | |
370 | rcu_assign_pointer(idr->hint, pa[0]); | |
371 | ||
372 | rcu_assign_pointer(pa[0]->ary[id & IDR_MASK], (struct idr_layer *)ptr); | |
373 | pa[0]->count++; | |
374 | idr_mark_full(pa, id); | |
375 | } | |
376 | ||
377 | int __idr_get_new_above(struct idr *idp, void *ptr, int starting_id, int *id) | |
378 | { | |
379 | struct idr_layer *pa[MAX_IDR_LEVEL + 1]; | |
380 | int rv; | |
381 | ||
382 | rv = idr_get_empty_slot(idp, starting_id, pa, 0, idp); | |
383 | if (rv < 0) | |
384 | return rv == -ENOMEM ? -EAGAIN : rv; | |
385 | ||
386 | idr_fill_slot(idp, ptr, rv, pa); | |
387 | *id = rv; | |
388 | return 0; | |
389 | } | |
390 | EXPORT_SYMBOL(__idr_get_new_above); | |
391 | ||
392 | /** | |
393 | * idr_preload - preload for idr_alloc() | |
394 | * @gfp_mask: allocation mask to use for preloading | |
395 | * | |
396 | * Preload per-cpu layer buffer for idr_alloc(). Can only be used from | |
397 | * process context and each idr_preload() invocation should be matched with | |
398 | * idr_preload_end(). Note that preemption is disabled while preloaded. | |
399 | * | |
400 | * The first idr_alloc() in the preloaded section can be treated as if it | |
401 | * were invoked with @gfp_mask used for preloading. This allows using more | |
402 | * permissive allocation masks for idrs protected by spinlocks. | |
403 | * | |
404 | * For example, if idr_alloc() below fails, the failure can be treated as | |
405 | * if idr_alloc() were called with GFP_KERNEL rather than GFP_NOWAIT. | |
406 | * | |
407 | * idr_preload(GFP_KERNEL); | |
408 | * spin_lock(lock); | |
409 | * | |
410 | * id = idr_alloc(idr, ptr, start, end, GFP_NOWAIT); | |
411 | * | |
412 | * spin_unlock(lock); | |
413 | * idr_preload_end(); | |
414 | * if (id < 0) | |
415 | * error; | |
416 | */ | |
417 | void idr_preload(gfp_t gfp_mask) | |
418 | { | |
419 | /* | |
420 | * Consuming preload buffer from non-process context breaks preload | |
421 | * allocation guarantee. Disallow usage from those contexts. | |
422 | */ | |
423 | WARN_ON_ONCE(in_interrupt()); | |
424 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
425 | ||
426 | preempt_disable(); | |
427 | ||
428 | /* | |
429 | * idr_alloc() is likely to succeed w/o full idr_layer buffer and | |
430 | * return value from idr_alloc() needs to be checked for failure | |
431 | * anyway. Silently give up if allocation fails. The caller can | |
432 | * treat failures from idr_alloc() as if idr_alloc() were called | |
433 | * with @gfp_mask which should be enough. | |
434 | */ | |
435 | while (__this_cpu_read(idr_preload_cnt) < MAX_IDR_FREE) { | |
436 | struct idr_layer *new; | |
437 | ||
438 | preempt_enable(); | |
439 | new = kmem_cache_zalloc(idr_layer_cache, gfp_mask); | |
440 | preempt_disable(); | |
441 | if (!new) | |
442 | break; | |
443 | ||
444 | /* link the new one to per-cpu preload list */ | |
445 | new->ary[0] = __this_cpu_read(idr_preload_head); | |
446 | __this_cpu_write(idr_preload_head, new); | |
447 | __this_cpu_inc(idr_preload_cnt); | |
448 | } | |
449 | } | |
450 | EXPORT_SYMBOL(idr_preload); | |
451 | ||
452 | /** | |
453 | * idr_alloc - allocate new idr entry | |
454 | * @idr: the (initialized) idr | |
455 | * @ptr: pointer to be associated with the new id | |
456 | * @start: the minimum id (inclusive) | |
457 | * @end: the maximum id (exclusive, <= 0 for max) | |
458 | * @gfp_mask: memory allocation flags | |
459 | * | |
460 | * Allocate an id in [start, end) and associate it with @ptr. If no ID is | |
461 | * available in the specified range, returns -ENOSPC. On memory allocation | |
462 | * failure, returns -ENOMEM. | |
463 | * | |
464 | * Note that @end is treated as max when <= 0. This is to always allow | |
465 | * using @start + N as @end as long as N is inside integer range. | |
466 | * | |
467 | * The user is responsible for exclusively synchronizing all operations | |
468 | * which may modify @idr. However, read-only accesses such as idr_find() | |
469 | * or iteration can be performed under RCU read lock provided the user | |
470 | * destroys @ptr in RCU-safe way after removal from idr. | |
471 | */ | |
472 | int idr_alloc(struct idr *idr, void *ptr, int start, int end, gfp_t gfp_mask) | |
473 | { | |
474 | int max = end > 0 ? end - 1 : INT_MAX; /* inclusive upper limit */ | |
475 | struct idr_layer *pa[MAX_IDR_LEVEL + 1]; | |
476 | int id; | |
477 | ||
478 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
479 | ||
480 | /* sanity checks */ | |
481 | if (WARN_ON_ONCE(start < 0)) | |
482 | return -EINVAL; | |
483 | if (unlikely(max < start)) | |
484 | return -ENOSPC; | |
485 | ||
486 | /* allocate id */ | |
487 | id = idr_get_empty_slot(idr, start, pa, gfp_mask, NULL); | |
488 | if (unlikely(id < 0)) | |
489 | return id; | |
490 | if (unlikely(id > max)) | |
491 | return -ENOSPC; | |
492 | ||
493 | idr_fill_slot(idr, ptr, id, pa); | |
494 | return id; | |
495 | } | |
496 | EXPORT_SYMBOL_GPL(idr_alloc); | |
497 | ||
498 | /** | |
499 | * idr_alloc_cyclic - allocate new idr entry in a cyclical fashion | |
500 | * @idr: the (initialized) idr | |
501 | * @ptr: pointer to be associated with the new id | |
502 | * @start: the minimum id (inclusive) | |
503 | * @end: the maximum id (exclusive, <= 0 for max) | |
504 | * @gfp_mask: memory allocation flags | |
505 | * | |
506 | * Essentially the same as idr_alloc, but prefers to allocate progressively | |
507 | * higher ids if it can. If the "cur" counter wraps, then it will start again | |
508 | * at the "start" end of the range and allocate one that has already been used. | |
509 | */ | |
510 | int idr_alloc_cyclic(struct idr *idr, void *ptr, int start, int end, | |
511 | gfp_t gfp_mask) | |
512 | { | |
513 | int id; | |
514 | ||
515 | id = idr_alloc(idr, ptr, max(start, idr->cur), end, gfp_mask); | |
516 | if (id == -ENOSPC) | |
517 | id = idr_alloc(idr, ptr, start, end, gfp_mask); | |
518 | ||
519 | if (likely(id >= 0)) | |
520 | idr->cur = id + 1; | |
521 | return id; | |
522 | } | |
523 | EXPORT_SYMBOL(idr_alloc_cyclic); | |
524 | ||
525 | static void idr_remove_warning(int id) | |
526 | { | |
527 | printk(KERN_WARNING | |
528 | "idr_remove called for id=%d which is not allocated.\n", id); | |
529 | dump_stack(); | |
530 | } | |
531 | ||
532 | static void sub_remove(struct idr *idp, int shift, int id) | |
533 | { | |
534 | struct idr_layer *p = idp->top; | |
535 | struct idr_layer **pa[MAX_IDR_LEVEL + 1]; | |
536 | struct idr_layer ***paa = &pa[0]; | |
537 | struct idr_layer *to_free; | |
538 | int n; | |
539 | ||
540 | *paa = NULL; | |
541 | *++paa = &idp->top; | |
542 | ||
543 | while ((shift > 0) && p) { | |
544 | n = (id >> shift) & IDR_MASK; | |
545 | __clear_bit(n, p->bitmap); | |
546 | *++paa = &p->ary[n]; | |
547 | p = p->ary[n]; | |
548 | shift -= IDR_BITS; | |
549 | } | |
550 | n = id & IDR_MASK; | |
551 | if (likely(p != NULL && test_bit(n, p->bitmap))) { | |
552 | __clear_bit(n, p->bitmap); | |
553 | rcu_assign_pointer(p->ary[n], NULL); | |
554 | to_free = NULL; | |
555 | while(*paa && ! --((**paa)->count)){ | |
556 | if (to_free) | |
557 | free_layer(idp, to_free); | |
558 | to_free = **paa; | |
559 | **paa-- = NULL; | |
560 | } | |
561 | if (!*paa) | |
562 | idp->layers = 0; | |
563 | if (to_free) | |
564 | free_layer(idp, to_free); | |
565 | } else | |
566 | idr_remove_warning(id); | |
567 | } | |
568 | ||
569 | /** | |
570 | * idr_remove - remove the given id and free its slot | |
571 | * @idp: idr handle | |
572 | * @id: unique key | |
573 | */ | |
574 | void idr_remove(struct idr *idp, int id) | |
575 | { | |
576 | struct idr_layer *p; | |
577 | struct idr_layer *to_free; | |
578 | ||
579 | if (id < 0) | |
580 | return; | |
581 | ||
582 | sub_remove(idp, (idp->layers - 1) * IDR_BITS, id); | |
583 | if (idp->top && idp->top->count == 1 && (idp->layers > 1) && | |
584 | idp->top->ary[0]) { | |
585 | /* | |
586 | * Single child at leftmost slot: we can shrink the tree. | |
587 | * This level is not needed anymore since when layers are | |
588 | * inserted, they are inserted at the top of the existing | |
589 | * tree. | |
590 | */ | |
591 | to_free = idp->top; | |
592 | p = idp->top->ary[0]; | |
593 | rcu_assign_pointer(idp->top, p); | |
594 | --idp->layers; | |
595 | to_free->count = 0; | |
596 | bitmap_clear(to_free->bitmap, 0, IDR_SIZE); | |
597 | free_layer(idp, to_free); | |
598 | } | |
599 | while (idp->id_free_cnt >= MAX_IDR_FREE) { | |
600 | p = get_from_free_list(idp); | |
601 | /* | |
602 | * Note: we don't call the rcu callback here, since the only | |
603 | * layers that fall into the freelist are those that have been | |
604 | * preallocated. | |
605 | */ | |
606 | kmem_cache_free(idr_layer_cache, p); | |
607 | } | |
608 | return; | |
609 | } | |
610 | EXPORT_SYMBOL(idr_remove); | |
611 | ||
612 | void __idr_remove_all(struct idr *idp) | |
613 | { | |
614 | int n, id, max; | |
615 | int bt_mask; | |
616 | struct idr_layer *p; | |
617 | struct idr_layer *pa[MAX_IDR_LEVEL + 1]; | |
618 | struct idr_layer **paa = &pa[0]; | |
619 | ||
620 | n = idp->layers * IDR_BITS; | |
621 | p = idp->top; | |
622 | rcu_assign_pointer(idp->top, NULL); | |
623 | max = idr_max(idp->layers); | |
624 | ||
625 | id = 0; | |
626 | while (id >= 0 && id <= max) { | |
627 | while (n > IDR_BITS && p) { | |
628 | n -= IDR_BITS; | |
629 | *paa++ = p; | |
630 | p = p->ary[(id >> n) & IDR_MASK]; | |
631 | } | |
632 | ||
633 | bt_mask = id; | |
634 | id += 1 << n; | |
635 | /* Get the highest bit that the above add changed from 0->1. */ | |
636 | while (n < fls(id ^ bt_mask)) { | |
637 | if (p) | |
638 | free_layer(idp, p); | |
639 | n += IDR_BITS; | |
640 | p = *--paa; | |
641 | } | |
642 | } | |
643 | idp->layers = 0; | |
644 | } | |
645 | EXPORT_SYMBOL(__idr_remove_all); | |
646 | ||
647 | /** | |
648 | * idr_destroy - release all cached layers within an idr tree | |
649 | * @idp: idr handle | |
650 | * | |
651 | * Free all id mappings and all idp_layers. After this function, @idp is | |
652 | * completely unused and can be freed / recycled. The caller is | |
653 | * responsible for ensuring that no one else accesses @idp during or after | |
654 | * idr_destroy(). | |
655 | * | |
656 | * A typical clean-up sequence for objects stored in an idr tree will use | |
657 | * idr_for_each() to free all objects, if necessay, then idr_destroy() to | |
658 | * free up the id mappings and cached idr_layers. | |
659 | */ | |
660 | void idr_destroy(struct idr *idp) | |
661 | { | |
662 | __idr_remove_all(idp); | |
663 | ||
664 | while (idp->id_free_cnt) { | |
665 | struct idr_layer *p = get_from_free_list(idp); | |
666 | kmem_cache_free(idr_layer_cache, p); | |
667 | } | |
668 | } | |
669 | EXPORT_SYMBOL(idr_destroy); | |
670 | ||
671 | void *idr_find_slowpath(struct idr *idp, int id) | |
672 | { | |
673 | int n; | |
674 | struct idr_layer *p; | |
675 | ||
676 | if (id < 0) | |
677 | return NULL; | |
678 | ||
679 | p = rcu_dereference_raw(idp->top); | |
680 | if (!p) | |
681 | return NULL; | |
682 | n = (p->layer+1) * IDR_BITS; | |
683 | ||
684 | if (id > idr_max(p->layer + 1)) | |
685 | return NULL; | |
686 | BUG_ON(n == 0); | |
687 | ||
688 | while (n > 0 && p) { | |
689 | n -= IDR_BITS; | |
690 | BUG_ON(n != p->layer*IDR_BITS); | |
691 | p = rcu_dereference_raw(p->ary[(id >> n) & IDR_MASK]); | |
692 | } | |
693 | return((void *)p); | |
694 | } | |
695 | EXPORT_SYMBOL(idr_find_slowpath); | |
696 | ||
697 | /** | |
698 | * idr_for_each - iterate through all stored pointers | |
699 | * @idp: idr handle | |
700 | * @fn: function to be called for each pointer | |
701 | * @data: data passed back to callback function | |
702 | * | |
703 | * Iterate over the pointers registered with the given idr. The | |
704 | * callback function will be called for each pointer currently | |
705 | * registered, passing the id, the pointer and the data pointer passed | |
706 | * to this function. It is not safe to modify the idr tree while in | |
707 | * the callback, so functions such as idr_get_new and idr_remove are | |
708 | * not allowed. | |
709 | * | |
710 | * We check the return of @fn each time. If it returns anything other | |
711 | * than %0, we break out and return that value. | |
712 | * | |
713 | * The caller must serialize idr_for_each() vs idr_get_new() and idr_remove(). | |
714 | */ | |
715 | int idr_for_each(struct idr *idp, | |
716 | int (*fn)(int id, void *p, void *data), void *data) | |
717 | { | |
718 | int n, id, max, error = 0; | |
719 | struct idr_layer *p; | |
720 | struct idr_layer *pa[MAX_IDR_LEVEL + 1]; | |
721 | struct idr_layer **paa = &pa[0]; | |
722 | ||
723 | n = idp->layers * IDR_BITS; | |
724 | p = rcu_dereference_raw(idp->top); | |
725 | max = idr_max(idp->layers); | |
726 | ||
727 | id = 0; | |
728 | while (id >= 0 && id <= max) { | |
729 | while (n > 0 && p) { | |
730 | n -= IDR_BITS; | |
731 | *paa++ = p; | |
732 | p = rcu_dereference_raw(p->ary[(id >> n) & IDR_MASK]); | |
733 | } | |
734 | ||
735 | if (p) { | |
736 | error = fn(id, (void *)p, data); | |
737 | if (error) | |
738 | break; | |
739 | } | |
740 | ||
741 | id += 1 << n; | |
742 | while (n < fls(id)) { | |
743 | n += IDR_BITS; | |
744 | p = *--paa; | |
745 | } | |
746 | } | |
747 | ||
748 | return error; | |
749 | } | |
750 | EXPORT_SYMBOL(idr_for_each); | |
751 | ||
752 | /** | |
753 | * idr_get_next - lookup next object of id to given id. | |
754 | * @idp: idr handle | |
755 | * @nextidp: pointer to lookup key | |
756 | * | |
757 | * Returns pointer to registered object with id, which is next number to | |
758 | * given id. After being looked up, *@nextidp will be updated for the next | |
759 | * iteration. | |
760 | * | |
761 | * This function can be called under rcu_read_lock(), given that the leaf | |
762 | * pointers lifetimes are correctly managed. | |
763 | */ | |
764 | void *idr_get_next(struct idr *idp, int *nextidp) | |
765 | { | |
766 | struct idr_layer *p, *pa[MAX_IDR_LEVEL + 1]; | |
767 | struct idr_layer **paa = &pa[0]; | |
768 | int id = *nextidp; | |
769 | int n, max; | |
770 | ||
771 | /* find first ent */ | |
772 | p = rcu_dereference_raw(idp->top); | |
773 | if (!p) | |
774 | return NULL; | |
775 | n = (p->layer + 1) * IDR_BITS; | |
776 | max = idr_max(p->layer + 1); | |
777 | ||
778 | while (id >= 0 && id <= max) { | |
779 | while (n > 0 && p) { | |
780 | n -= IDR_BITS; | |
781 | *paa++ = p; | |
782 | p = rcu_dereference_raw(p->ary[(id >> n) & IDR_MASK]); | |
783 | } | |
784 | ||
785 | if (p) { | |
786 | *nextidp = id; | |
787 | return p; | |
788 | } | |
789 | ||
790 | /* | |
791 | * Proceed to the next layer at the current level. Unlike | |
792 | * idr_for_each(), @id isn't guaranteed to be aligned to | |
793 | * layer boundary at this point and adding 1 << n may | |
794 | * incorrectly skip IDs. Make sure we jump to the | |
795 | * beginning of the next layer using round_up(). | |
796 | */ | |
797 | id = round_up(id + 1, 1 << n); | |
798 | while (n < fls(id)) { | |
799 | n += IDR_BITS; | |
800 | p = *--paa; | |
801 | } | |
802 | } | |
803 | return NULL; | |
804 | } | |
805 | EXPORT_SYMBOL(idr_get_next); | |
806 | ||
807 | ||
808 | /** | |
809 | * idr_replace - replace pointer for given id | |
810 | * @idp: idr handle | |
811 | * @ptr: pointer you want associated with the id | |
812 | * @id: lookup key | |
813 | * | |
814 | * Replace the pointer registered with an id and return the old value. | |
815 | * A %-ENOENT return indicates that @id was not found. | |
816 | * A %-EINVAL return indicates that @id was not within valid constraints. | |
817 | * | |
818 | * The caller must serialize with writers. | |
819 | */ | |
820 | void *idr_replace(struct idr *idp, void *ptr, int id) | |
821 | { | |
822 | int n; | |
823 | struct idr_layer *p, *old_p; | |
824 | ||
825 | if (id < 0) | |
826 | return ERR_PTR(-EINVAL); | |
827 | ||
828 | p = idp->top; | |
829 | if (!p) | |
830 | return ERR_PTR(-EINVAL); | |
831 | ||
832 | n = (p->layer+1) * IDR_BITS; | |
833 | ||
834 | if (id >= (1 << n)) | |
835 | return ERR_PTR(-EINVAL); | |
836 | ||
837 | n -= IDR_BITS; | |
838 | while ((n > 0) && p) { | |
839 | p = p->ary[(id >> n) & IDR_MASK]; | |
840 | n -= IDR_BITS; | |
841 | } | |
842 | ||
843 | n = id & IDR_MASK; | |
844 | if (unlikely(p == NULL || !test_bit(n, p->bitmap))) | |
845 | return ERR_PTR(-ENOENT); | |
846 | ||
847 | old_p = p->ary[n]; | |
848 | rcu_assign_pointer(p->ary[n], ptr); | |
849 | ||
850 | return old_p; | |
851 | } | |
852 | EXPORT_SYMBOL(idr_replace); | |
853 | ||
854 | void __init idr_init_cache(void) | |
855 | { | |
856 | idr_layer_cache = kmem_cache_create("idr_layer_cache", | |
857 | sizeof(struct idr_layer), 0, SLAB_PANIC, NULL); | |
858 | } | |
859 | ||
860 | /** | |
861 | * idr_init - initialize idr handle | |
862 | * @idp: idr handle | |
863 | * | |
864 | * This function is use to set up the handle (@idp) that you will pass | |
865 | * to the rest of the functions. | |
866 | */ | |
867 | void idr_init(struct idr *idp) | |
868 | { | |
869 | memset(idp, 0, sizeof(struct idr)); | |
870 | spin_lock_init(&idp->lock); | |
871 | } | |
872 | EXPORT_SYMBOL(idr_init); | |
873 | ||
874 | ||
875 | /** | |
876 | * DOC: IDA description | |
877 | * IDA - IDR based ID allocator | |
878 | * | |
879 | * This is id allocator without id -> pointer translation. Memory | |
880 | * usage is much lower than full blown idr because each id only | |
881 | * occupies a bit. ida uses a custom leaf node which contains | |
882 | * IDA_BITMAP_BITS slots. | |
883 | * | |
884 | * 2007-04-25 written by Tejun Heo <htejun@gmail.com> | |
885 | */ | |
886 | ||
887 | static void free_bitmap(struct ida *ida, struct ida_bitmap *bitmap) | |
888 | { | |
889 | unsigned long flags; | |
890 | ||
891 | if (!ida->free_bitmap) { | |
892 | spin_lock_irqsave(&ida->idr.lock, flags); | |
893 | if (!ida->free_bitmap) { | |
894 | ida->free_bitmap = bitmap; | |
895 | bitmap = NULL; | |
896 | } | |
897 | spin_unlock_irqrestore(&ida->idr.lock, flags); | |
898 | } | |
899 | ||
900 | kfree(bitmap); | |
901 | } | |
902 | ||
903 | /** | |
904 | * ida_pre_get - reserve resources for ida allocation | |
905 | * @ida: ida handle | |
906 | * @gfp_mask: memory allocation flag | |
907 | * | |
908 | * This function should be called prior to locking and calling the | |
909 | * following function. It preallocates enough memory to satisfy the | |
910 | * worst possible allocation. | |
911 | * | |
912 | * If the system is REALLY out of memory this function returns %0, | |
913 | * otherwise %1. | |
914 | */ | |
915 | int ida_pre_get(struct ida *ida, gfp_t gfp_mask) | |
916 | { | |
917 | /* allocate idr_layers */ | |
918 | if (!__idr_pre_get(&ida->idr, gfp_mask)) | |
919 | return 0; | |
920 | ||
921 | /* allocate free_bitmap */ | |
922 | if (!ida->free_bitmap) { | |
923 | struct ida_bitmap *bitmap; | |
924 | ||
925 | bitmap = kmalloc(sizeof(struct ida_bitmap), gfp_mask); | |
926 | if (!bitmap) | |
927 | return 0; | |
928 | ||
929 | free_bitmap(ida, bitmap); | |
930 | } | |
931 | ||
932 | return 1; | |
933 | } | |
934 | EXPORT_SYMBOL(ida_pre_get); | |
935 | ||
936 | /** | |
937 | * ida_get_new_above - allocate new ID above or equal to a start id | |
938 | * @ida: ida handle | |
939 | * @starting_id: id to start search at | |
940 | * @p_id: pointer to the allocated handle | |
941 | * | |
942 | * Allocate new ID above or equal to @starting_id. It should be called | |
943 | * with any required locks. | |
944 | * | |
945 | * If memory is required, it will return %-EAGAIN, you should unlock | |
946 | * and go back to the ida_pre_get() call. If the ida is full, it will | |
947 | * return %-ENOSPC. | |
948 | * | |
949 | * @p_id returns a value in the range @starting_id ... %0x7fffffff. | |
950 | */ | |
951 | int ida_get_new_above(struct ida *ida, int starting_id, int *p_id) | |
952 | { | |
953 | struct idr_layer *pa[MAX_IDR_LEVEL + 1]; | |
954 | struct ida_bitmap *bitmap; | |
955 | unsigned long flags; | |
956 | int idr_id = starting_id / IDA_BITMAP_BITS; | |
957 | int offset = starting_id % IDA_BITMAP_BITS; | |
958 | int t, id; | |
959 | ||
960 | restart: | |
961 | /* get vacant slot */ | |
962 | t = idr_get_empty_slot(&ida->idr, idr_id, pa, 0, &ida->idr); | |
963 | if (t < 0) | |
964 | return t == -ENOMEM ? -EAGAIN : t; | |
965 | ||
966 | if (t * IDA_BITMAP_BITS >= MAX_IDR_BIT) | |
967 | return -ENOSPC; | |
968 | ||
969 | if (t != idr_id) | |
970 | offset = 0; | |
971 | idr_id = t; | |
972 | ||
973 | /* if bitmap isn't there, create a new one */ | |
974 | bitmap = (void *)pa[0]->ary[idr_id & IDR_MASK]; | |
975 | if (!bitmap) { | |
976 | spin_lock_irqsave(&ida->idr.lock, flags); | |
977 | bitmap = ida->free_bitmap; | |
978 | ida->free_bitmap = NULL; | |
979 | spin_unlock_irqrestore(&ida->idr.lock, flags); | |
980 | ||
981 | if (!bitmap) | |
982 | return -EAGAIN; | |
983 | ||
984 | memset(bitmap, 0, sizeof(struct ida_bitmap)); | |
985 | rcu_assign_pointer(pa[0]->ary[idr_id & IDR_MASK], | |
986 | (void *)bitmap); | |
987 | pa[0]->count++; | |
988 | } | |
989 | ||
990 | /* lookup for empty slot */ | |
991 | t = find_next_zero_bit(bitmap->bitmap, IDA_BITMAP_BITS, offset); | |
992 | if (t == IDA_BITMAP_BITS) { | |
993 | /* no empty slot after offset, continue to the next chunk */ | |
994 | idr_id++; | |
995 | offset = 0; | |
996 | goto restart; | |
997 | } | |
998 | ||
999 | id = idr_id * IDA_BITMAP_BITS + t; | |
1000 | if (id >= MAX_IDR_BIT) | |
1001 | return -ENOSPC; | |
1002 | ||
1003 | __set_bit(t, bitmap->bitmap); | |
1004 | if (++bitmap->nr_busy == IDA_BITMAP_BITS) | |
1005 | idr_mark_full(pa, idr_id); | |
1006 | ||
1007 | *p_id = id; | |
1008 | ||
1009 | /* Each leaf node can handle nearly a thousand slots and the | |
1010 | * whole idea of ida is to have small memory foot print. | |
1011 | * Throw away extra resources one by one after each successful | |
1012 | * allocation. | |
1013 | */ | |
1014 | if (ida->idr.id_free_cnt || ida->free_bitmap) { | |
1015 | struct idr_layer *p = get_from_free_list(&ida->idr); | |
1016 | if (p) | |
1017 | kmem_cache_free(idr_layer_cache, p); | |
1018 | } | |
1019 | ||
1020 | return 0; | |
1021 | } | |
1022 | EXPORT_SYMBOL(ida_get_new_above); | |
1023 | ||
1024 | /** | |
1025 | * ida_remove - remove the given ID | |
1026 | * @ida: ida handle | |
1027 | * @id: ID to free | |
1028 | */ | |
1029 | void ida_remove(struct ida *ida, int id) | |
1030 | { | |
1031 | struct idr_layer *p = ida->idr.top; | |
1032 | int shift = (ida->idr.layers - 1) * IDR_BITS; | |
1033 | int idr_id = id / IDA_BITMAP_BITS; | |
1034 | int offset = id % IDA_BITMAP_BITS; | |
1035 | int n; | |
1036 | struct ida_bitmap *bitmap; | |
1037 | ||
1038 | /* clear full bits while looking up the leaf idr_layer */ | |
1039 | while ((shift > 0) && p) { | |
1040 | n = (idr_id >> shift) & IDR_MASK; | |
1041 | __clear_bit(n, p->bitmap); | |
1042 | p = p->ary[n]; | |
1043 | shift -= IDR_BITS; | |
1044 | } | |
1045 | ||
1046 | if (p == NULL) | |
1047 | goto err; | |
1048 | ||
1049 | n = idr_id & IDR_MASK; | |
1050 | __clear_bit(n, p->bitmap); | |
1051 | ||
1052 | bitmap = (void *)p->ary[n]; | |
1053 | if (!test_bit(offset, bitmap->bitmap)) | |
1054 | goto err; | |
1055 | ||
1056 | /* update bitmap and remove it if empty */ | |
1057 | __clear_bit(offset, bitmap->bitmap); | |
1058 | if (--bitmap->nr_busy == 0) { | |
1059 | __set_bit(n, p->bitmap); /* to please idr_remove() */ | |
1060 | idr_remove(&ida->idr, idr_id); | |
1061 | free_bitmap(ida, bitmap); | |
1062 | } | |
1063 | ||
1064 | return; | |
1065 | ||
1066 | err: | |
1067 | printk(KERN_WARNING | |
1068 | "ida_remove called for id=%d which is not allocated.\n", id); | |
1069 | } | |
1070 | EXPORT_SYMBOL(ida_remove); | |
1071 | ||
1072 | /** | |
1073 | * ida_destroy - release all cached layers within an ida tree | |
1074 | * @ida: ida handle | |
1075 | */ | |
1076 | void ida_destroy(struct ida *ida) | |
1077 | { | |
1078 | idr_destroy(&ida->idr); | |
1079 | kfree(ida->free_bitmap); | |
1080 | } | |
1081 | EXPORT_SYMBOL(ida_destroy); | |
1082 | ||
1083 | /** | |
1084 | * ida_simple_get - get a new id. | |
1085 | * @ida: the (initialized) ida. | |
1086 | * @start: the minimum id (inclusive, < 0x8000000) | |
1087 | * @end: the maximum id (exclusive, < 0x8000000 or 0) | |
1088 | * @gfp_mask: memory allocation flags | |
1089 | * | |
1090 | * Allocates an id in the range start <= id < end, or returns -ENOSPC. | |
1091 | * On memory allocation failure, returns -ENOMEM. | |
1092 | * | |
1093 | * Use ida_simple_remove() to get rid of an id. | |
1094 | */ | |
1095 | int ida_simple_get(struct ida *ida, unsigned int start, unsigned int end, | |
1096 | gfp_t gfp_mask) | |
1097 | { | |
1098 | int ret, id; | |
1099 | unsigned int max; | |
1100 | unsigned long flags; | |
1101 | ||
1102 | BUG_ON((int)start < 0); | |
1103 | BUG_ON((int)end < 0); | |
1104 | ||
1105 | if (end == 0) | |
1106 | max = 0x80000000; | |
1107 | else { | |
1108 | BUG_ON(end < start); | |
1109 | max = end - 1; | |
1110 | } | |
1111 | ||
1112 | again: | |
1113 | if (!ida_pre_get(ida, gfp_mask)) | |
1114 | return -ENOMEM; | |
1115 | ||
1116 | spin_lock_irqsave(&simple_ida_lock, flags); | |
1117 | ret = ida_get_new_above(ida, start, &id); | |
1118 | if (!ret) { | |
1119 | if (id > max) { | |
1120 | ida_remove(ida, id); | |
1121 | ret = -ENOSPC; | |
1122 | } else { | |
1123 | ret = id; | |
1124 | } | |
1125 | } | |
1126 | spin_unlock_irqrestore(&simple_ida_lock, flags); | |
1127 | ||
1128 | if (unlikely(ret == -EAGAIN)) | |
1129 | goto again; | |
1130 | ||
1131 | return ret; | |
1132 | } | |
1133 | EXPORT_SYMBOL(ida_simple_get); | |
1134 | ||
1135 | /** | |
1136 | * ida_simple_remove - remove an allocated id. | |
1137 | * @ida: the (initialized) ida. | |
1138 | * @id: the id returned by ida_simple_get. | |
1139 | */ | |
1140 | void ida_simple_remove(struct ida *ida, unsigned int id) | |
1141 | { | |
1142 | unsigned long flags; | |
1143 | ||
1144 | BUG_ON((int)id < 0); | |
1145 | spin_lock_irqsave(&simple_ida_lock, flags); | |
1146 | ida_remove(ida, id); | |
1147 | spin_unlock_irqrestore(&simple_ida_lock, flags); | |
1148 | } | |
1149 | EXPORT_SYMBOL(ida_simple_remove); | |
1150 | ||
1151 | /** | |
1152 | * ida_init - initialize ida handle | |
1153 | * @ida: ida handle | |
1154 | * | |
1155 | * This function is use to set up the handle (@ida) that you will pass | |
1156 | * to the rest of the functions. | |
1157 | */ | |
1158 | void ida_init(struct ida *ida) | |
1159 | { | |
1160 | memset(ida, 0, sizeof(struct ida)); | |
1161 | idr_init(&ida->idr); | |
1162 | ||
1163 | } | |
1164 | EXPORT_SYMBOL(ida_init); |