1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * SLOB Allocator: Simple List Of Blocks
4 *
5 * Matt Mackall <mpm@selenic.com> 12/30/03
6 *
7 * NUMA support by Paul Mundt, 2007.
8 *
9 * How SLOB works:
10 *
11 * The core of SLOB is a traditional K&R style heap allocator, with
12 * support for returning aligned objects. The granularity of this
13 * allocator is as little as 2 bytes, however typically most architectures
14 * will require 4 bytes on 32-bit and 8 bytes on 64-bit.
15 *
16 * The slob heap is a set of linked list of pages from alloc_pages(),
17 * and within each page, there is a singly-linked list of free blocks
18 * (slob_t). The heap is grown on demand. To reduce fragmentation,
19 * heap pages are segregated into three lists, with objects less than
20 * 256 bytes, objects less than 1024 bytes, and all other objects.
21 *
22 * Allocation from heap involves first searching for a page with
23 * sufficient free blocks (using a next-fit-like approach) followed by
24 * a first-fit scan of the page. Deallocation inserts objects back
25 * into the free list in address order, so this is effectively an
26 * address-ordered first fit.
27 *
28 * Above this is an implementation of kmalloc/kfree. Blocks returned
29 * from kmalloc are prepended with a 4-byte header with the kmalloc size.
30 * If kmalloc is asked for objects of PAGE_SIZE or larger, it calls
31 * alloc_pages() directly, allocating compound pages so the page order
32 * does not have to be separately tracked.
33 * These objects are detected in kfree() because PageSlab()
34 * is false for them.
35 *
36 * SLAB is emulated on top of SLOB by simply calling constructors and
37 * destructors for every SLAB allocation. Objects are returned with the
38 * 4-byte alignment unless the SLAB_HWCACHE_ALIGN flag is set, in which
39 * case the low-level allocator will fragment blocks to create the proper
40 * alignment. Again, objects of page-size or greater are allocated by
41 * calling alloc_pages(). As SLAB objects know their size, no separate
42 * size bookkeeping is necessary and there is essentially no allocation
43 * space overhead, and compound pages aren't needed for multi-page
44 * allocations.
45 *
46 * NUMA support in SLOB is fairly simplistic, pushing most of the real
47 * logic down to the page allocator, and simply doing the node accounting
48 * on the upper levels. In the event that a node id is explicitly
49 * provided, __alloc_pages_node() with the specified node id is used
50 * instead. The common case (or when the node id isn't explicitly provided)
51 * will default to the current node, as per numa_node_id().
52 *
53 * Node aware pages are still inserted in to the global freelist, and
54 * these are scanned for by matching against the node id encoded in the
55 * page flags. As a result, block allocations that can be satisfied from
56 * the freelist will only be done so on pages residing on the same node,
57 * in order to prevent random node placement.
58 */
59
60 #include <linux/kernel.h>
61 #include <linux/slab.h>
62
63 #include <linux/mm.h>
64 #include <linux/swap.h> /* struct reclaim_state */
65 #include <linux/cache.h>
66 #include <linux/init.h>
67 #include <linux/export.h>
68 #include <linux/rcupdate.h>
69 #include <linux/list.h>
70 #include <linux/kmemleak.h>
71
72 #include <trace/events/kmem.h>
73
74 #include <linux/atomic.h>
75
76 #include "slab.h"
77 /*
78 * slob_block has a field 'units', which indicates size of block if +ve,
79 * or offset of next block if -ve (in SLOB_UNITs).
80 *
81 * Free blocks of size 1 unit simply contain the offset of the next block.
82 * Those with larger size contain their size in the first SLOB_UNIT of
83 * memory, and the offset of the next free block in the second SLOB_UNIT.
84 */
85 #if PAGE_SIZE <= (32767 * 2)
86 typedef s16 slobidx_t;
87 #else
88 typedef s32 slobidx_t;
89 #endif
90
91 struct slob_block {
92 slobidx_t units;
93 };
94 typedef struct slob_block slob_t;
95
96 /*
97 * All partially free slob pages go on these lists.
98 */
99 #define SLOB_BREAK1 256
100 #define SLOB_BREAK2 1024
101 static LIST_HEAD(free_slob_small);
102 static LIST_HEAD(free_slob_medium);
103 static LIST_HEAD(free_slob_large);
104
105 /*
106 * slob_page_free: true for pages on free_slob_pages list.
107 */
slob_page_free(struct page * sp)108 static inline int slob_page_free(struct page *sp)
109 {
110 return PageSlobFree(sp);
111 }
112
set_slob_page_free(struct page * sp,struct list_head * list)113 static void set_slob_page_free(struct page *sp, struct list_head *list)
114 {
115 list_add(&sp->lru, list);
116 __SetPageSlobFree(sp);
117 }
118
clear_slob_page_free(struct page * sp)119 static inline void clear_slob_page_free(struct page *sp)
120 {
121 list_del(&sp->lru);
122 __ClearPageSlobFree(sp);
123 }
124
125 #define SLOB_UNIT sizeof(slob_t)
126 #define SLOB_UNITS(size) DIV_ROUND_UP(size, SLOB_UNIT)
127
128 /*
129 * struct slob_rcu is inserted at the tail of allocated slob blocks, which
130 * were created with a SLAB_TYPESAFE_BY_RCU slab. slob_rcu is used to free
131 * the block using call_rcu.
132 */
133 struct slob_rcu {
134 struct rcu_head head;
135 int size;
136 };
137
138 /*
139 * slob_lock protects all slob allocator structures.
140 */
141 static DEFINE_SPINLOCK(slob_lock);
142
143 /*
144 * Encode the given size and next info into a free slob block s.
145 */
set_slob(slob_t * s,slobidx_t size,slob_t * next)146 static void set_slob(slob_t *s, slobidx_t size, slob_t *next)
147 {
148 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
149 slobidx_t offset = next - base;
150
151 if (size > 1) {
152 s[0].units = size;
153 s[1].units = offset;
154 } else
155 s[0].units = -offset;
156 }
157
158 /*
159 * Return the size of a slob block.
160 */
slob_units(slob_t * s)161 static slobidx_t slob_units(slob_t *s)
162 {
163 if (s->units > 0)
164 return s->units;
165 return 1;
166 }
167
168 /*
169 * Return the next free slob block pointer after this one.
170 */
slob_next(slob_t * s)171 static slob_t *slob_next(slob_t *s)
172 {
173 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
174 slobidx_t next;
175
176 if (s[0].units < 0)
177 next = -s[0].units;
178 else
179 next = s[1].units;
180 return base+next;
181 }
182
183 /*
184 * Returns true if s is the last free block in its page.
185 */
slob_last(slob_t * s)186 static int slob_last(slob_t *s)
187 {
188 return !((unsigned long)slob_next(s) & ~PAGE_MASK);
189 }
190
slob_new_pages(gfp_t gfp,int order,int node)191 static void *slob_new_pages(gfp_t gfp, int order, int node)
192 {
193 void *page;
194
195 #ifdef CONFIG_NUMA
196 if (node != NUMA_NO_NODE)
197 page = __alloc_pages_node(node, gfp, order);
198 else
199 #endif
200 page = alloc_pages(gfp, order);
201
202 if (!page)
203 return NULL;
204
205 return page_address(page);
206 }
207
slob_free_pages(void * b,int order)208 static void slob_free_pages(void *b, int order)
209 {
210 if (current->reclaim_state)
211 current->reclaim_state->reclaimed_slab += 1 << order;
212 free_pages((unsigned long)b, order);
213 }
214
215 /*
216 * Allocate a slob block within a given slob_page sp.
217 */
slob_page_alloc(struct page * sp,size_t size,int align)218 static void *slob_page_alloc(struct page *sp, size_t size, int align)
219 {
220 slob_t *prev, *cur, *aligned = NULL;
221 int delta = 0, units = SLOB_UNITS(size);
222
223 for (prev = NULL, cur = sp->freelist; ; prev = cur, cur = slob_next(cur)) {
224 slobidx_t avail = slob_units(cur);
225
226 if (align) {
227 aligned = (slob_t *)ALIGN((unsigned long)cur, align);
228 delta = aligned - cur;
229 }
230 if (avail >= units + delta) { /* room enough? */
231 slob_t *next;
232
233 if (delta) { /* need to fragment head to align? */
234 next = slob_next(cur);
235 set_slob(aligned, avail - delta, next);
236 set_slob(cur, delta, aligned);
237 prev = cur;
238 cur = aligned;
239 avail = slob_units(cur);
240 }
241
242 next = slob_next(cur);
243 if (avail == units) { /* exact fit? unlink. */
244 if (prev)
245 set_slob(prev, slob_units(prev), next);
246 else
247 sp->freelist = next;
248 } else { /* fragment */
249 if (prev)
250 set_slob(prev, slob_units(prev), cur + units);
251 else
252 sp->freelist = cur + units;
253 set_slob(cur + units, avail - units, next);
254 }
255
256 sp->units -= units;
257 if (!sp->units)
258 clear_slob_page_free(sp);
259 return cur;
260 }
261 if (slob_last(cur))
262 return NULL;
263 }
264 }
265
266 /*
267 * slob_alloc: entry point into the slob allocator.
268 */
slob_alloc(size_t size,gfp_t gfp,int align,int node)269 static void *slob_alloc(size_t size, gfp_t gfp, int align, int node)
270 {
271 struct page *sp;
272 struct list_head *prev;
273 struct list_head *slob_list;
274 slob_t *b = NULL;
275 unsigned long flags;
276
277 if (size < SLOB_BREAK1)
278 slob_list = &free_slob_small;
279 else if (size < SLOB_BREAK2)
280 slob_list = &free_slob_medium;
281 else
282 slob_list = &free_slob_large;
283
284 spin_lock_irqsave(&slob_lock, flags);
285 /* Iterate through each partially free page, try to find room */
286 list_for_each_entry(sp, slob_list, lru) {
287 #ifdef CONFIG_NUMA
288 /*
289 * If there's a node specification, search for a partial
290 * page with a matching node id in the freelist.
291 */
292 if (node != NUMA_NO_NODE && page_to_nid(sp) != node)
293 continue;
294 #endif
295 /* Enough room on this page? */
296 if (sp->units < SLOB_UNITS(size))
297 continue;
298
299 /* Attempt to alloc */
300 prev = sp->lru.prev;
301 b = slob_page_alloc(sp, size, align);
302 if (!b)
303 continue;
304
305 /* Improve fragment distribution and reduce our average
306 * search time by starting our next search here. (see
307 * Knuth vol 1, sec 2.5, pg 449) */
308 if (prev != slob_list->prev &&
309 slob_list->next != prev->next)
310 list_move_tail(slob_list, prev->next);
311 break;
312 }
313 spin_unlock_irqrestore(&slob_lock, flags);
314
315 /* Not enough space: must allocate a new page */
316 if (!b) {
317 b = slob_new_pages(gfp & ~__GFP_ZERO, 0, node);
318 if (!b)
319 return NULL;
320 sp = virt_to_page(b);
321 __SetPageSlab(sp);
322
323 spin_lock_irqsave(&slob_lock, flags);
324 sp->units = SLOB_UNITS(PAGE_SIZE);
325 sp->freelist = b;
326 INIT_LIST_HEAD(&sp->lru);
327 set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE));
328 set_slob_page_free(sp, slob_list);
329 b = slob_page_alloc(sp, size, align);
330 BUG_ON(!b);
331 spin_unlock_irqrestore(&slob_lock, flags);
332 }
333 if (unlikely(gfp & __GFP_ZERO))
334 memset(b, 0, size);
335 return b;
336 }
337
338 /*
339 * slob_free: entry point into the slob allocator.
340 */
slob_free(void * block,int size)341 static void slob_free(void *block, int size)
342 {
343 struct page *sp;
344 slob_t *prev, *next, *b = (slob_t *)block;
345 slobidx_t units;
346 unsigned long flags;
347 struct list_head *slob_list;
348
349 if (unlikely(ZERO_OR_NULL_PTR(block)))
350 return;
351 BUG_ON(!size);
352
353 sp = virt_to_page(block);
354 units = SLOB_UNITS(size);
355
356 spin_lock_irqsave(&slob_lock, flags);
357
358 if (sp->units + units == SLOB_UNITS(PAGE_SIZE)) {
359 /* Go directly to page allocator. Do not pass slob allocator */
360 if (slob_page_free(sp))
361 clear_slob_page_free(sp);
362 spin_unlock_irqrestore(&slob_lock, flags);
363 __ClearPageSlab(sp);
364 page_mapcount_reset(sp);
365 slob_free_pages(b, 0);
366 return;
367 }
368
369 if (!slob_page_free(sp)) {
370 /* This slob page is about to become partially free. Easy! */
371 sp->units = units;
372 sp->freelist = b;
373 set_slob(b, units,
374 (void *)((unsigned long)(b +
375 SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK));
376 if (size < SLOB_BREAK1)
377 slob_list = &free_slob_small;
378 else if (size < SLOB_BREAK2)
379 slob_list = &free_slob_medium;
380 else
381 slob_list = &free_slob_large;
382 set_slob_page_free(sp, slob_list);
383 goto out;
384 }
385
386 /*
387 * Otherwise the page is already partially free, so find reinsertion
388 * point.
389 */
390 sp->units += units;
391
392 if (b < (slob_t *)sp->freelist) {
393 if (b + units == sp->freelist) {
394 units += slob_units(sp->freelist);
395 sp->freelist = slob_next(sp->freelist);
396 }
397 set_slob(b, units, sp->freelist);
398 sp->freelist = b;
399 } else {
400 prev = sp->freelist;
401 next = slob_next(prev);
402 while (b > next) {
403 prev = next;
404 next = slob_next(prev);
405 }
406
407 if (!slob_last(prev) && b + units == next) {
408 units += slob_units(next);
409 set_slob(b, units, slob_next(next));
410 } else
411 set_slob(b, units, next);
412
413 if (prev + slob_units(prev) == b) {
414 units = slob_units(b) + slob_units(prev);
415 set_slob(prev, units, slob_next(b));
416 } else
417 set_slob(prev, slob_units(prev), b);
418 }
419 out:
420 spin_unlock_irqrestore(&slob_lock, flags);
421 }
422
423 /*
424 * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend.
425 */
426
427 static __always_inline void *
__do_kmalloc_node(size_t size,gfp_t gfp,int node,unsigned long caller)428 __do_kmalloc_node(size_t size, gfp_t gfp, int node, unsigned long caller)
429 {
430 unsigned int *m;
431 int align = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
432 void *ret;
433
434 gfp &= gfp_allowed_mask;
435
436 fs_reclaim_acquire(gfp);
437 fs_reclaim_release(gfp);
438
439 if (size < PAGE_SIZE - align) {
440 if (!size)
441 return ZERO_SIZE_PTR;
442
443 m = slob_alloc(size + align, gfp, align, node);
444
445 if (!m)
446 return NULL;
447 *m = size;
448 ret = (void *)m + align;
449
450 trace_kmalloc_node(caller, ret,
451 size, size + align, gfp, node);
452 } else {
453 unsigned int order = get_order(size);
454
455 if (likely(order))
456 gfp |= __GFP_COMP;
457 ret = slob_new_pages(gfp, order, node);
458
459 trace_kmalloc_node(caller, ret,
460 size, PAGE_SIZE << order, gfp, node);
461 }
462
463 kmemleak_alloc(ret, size, 1, gfp);
464 return ret;
465 }
466
__kmalloc(size_t size,gfp_t gfp)467 void *__kmalloc(size_t size, gfp_t gfp)
468 {
469 return __do_kmalloc_node(size, gfp, NUMA_NO_NODE, _RET_IP_);
470 }
471 EXPORT_SYMBOL(__kmalloc);
472
__kmalloc_track_caller(size_t size,gfp_t gfp,unsigned long caller)473 void *__kmalloc_track_caller(size_t size, gfp_t gfp, unsigned long caller)
474 {
475 return __do_kmalloc_node(size, gfp, NUMA_NO_NODE, caller);
476 }
477
478 #ifdef CONFIG_NUMA
__kmalloc_node_track_caller(size_t size,gfp_t gfp,int node,unsigned long caller)479 void *__kmalloc_node_track_caller(size_t size, gfp_t gfp,
480 int node, unsigned long caller)
481 {
482 return __do_kmalloc_node(size, gfp, node, caller);
483 }
484 #endif
485
kfree(const void * block)486 void kfree(const void *block)
487 {
488 struct page *sp;
489
490 trace_kfree(_RET_IP_, block);
491
492 if (unlikely(ZERO_OR_NULL_PTR(block)))
493 return;
494 kmemleak_free(block);
495
496 sp = virt_to_page(block);
497 if (PageSlab(sp)) {
498 int align = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
499 unsigned int *m = (unsigned int *)(block - align);
500 slob_free(m, *m + align);
501 } else
502 __free_pages(sp, compound_order(sp));
503 }
504 EXPORT_SYMBOL(kfree);
505
506 /* can't use ksize for kmem_cache_alloc memory, only kmalloc */
ksize(const void * block)507 size_t ksize(const void *block)
508 {
509 struct page *sp;
510 int align;
511 unsigned int *m;
512
513 BUG_ON(!block);
514 if (unlikely(block == ZERO_SIZE_PTR))
515 return 0;
516
517 sp = virt_to_page(block);
518 if (unlikely(!PageSlab(sp)))
519 return PAGE_SIZE << compound_order(sp);
520
521 align = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
522 m = (unsigned int *)(block - align);
523 return SLOB_UNITS(*m) * SLOB_UNIT;
524 }
525 EXPORT_SYMBOL(ksize);
526
__kmem_cache_create(struct kmem_cache * c,slab_flags_t flags)527 int __kmem_cache_create(struct kmem_cache *c, slab_flags_t flags)
528 {
529 if (flags & SLAB_TYPESAFE_BY_RCU) {
530 /* leave room for rcu footer at the end of object */
531 c->size += sizeof(struct slob_rcu);
532 }
533 c->flags = flags;
534 return 0;
535 }
536
slob_alloc_node(struct kmem_cache * c,gfp_t flags,int node)537 static void *slob_alloc_node(struct kmem_cache *c, gfp_t flags, int node)
538 {
539 void *b;
540
541 flags &= gfp_allowed_mask;
542
543 fs_reclaim_acquire(flags);
544 fs_reclaim_release(flags);
545
546 if (c->size < PAGE_SIZE) {
547 b = slob_alloc(c->size, flags, c->align, node);
548 trace_kmem_cache_alloc_node(_RET_IP_, b, c->object_size,
549 SLOB_UNITS(c->size) * SLOB_UNIT,
550 flags, node);
551 } else {
552 b = slob_new_pages(flags, get_order(c->size), node);
553 trace_kmem_cache_alloc_node(_RET_IP_, b, c->object_size,
554 PAGE_SIZE << get_order(c->size),
555 flags, node);
556 }
557
558 if (b && c->ctor) {
559 WARN_ON_ONCE(flags & __GFP_ZERO);
560 c->ctor(b);
561 }
562
563 kmemleak_alloc_recursive(b, c->size, 1, c->flags, flags);
564 return b;
565 }
566
kmem_cache_alloc(struct kmem_cache * cachep,gfp_t flags)567 void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
568 {
569 return slob_alloc_node(cachep, flags, NUMA_NO_NODE);
570 }
571 EXPORT_SYMBOL(kmem_cache_alloc);
572
573 #ifdef CONFIG_NUMA
__kmalloc_node(size_t size,gfp_t gfp,int node)574 void *__kmalloc_node(size_t size, gfp_t gfp, int node)
575 {
576 return __do_kmalloc_node(size, gfp, node, _RET_IP_);
577 }
578 EXPORT_SYMBOL(__kmalloc_node);
579
kmem_cache_alloc_node(struct kmem_cache * cachep,gfp_t gfp,int node)580 void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t gfp, int node)
581 {
582 return slob_alloc_node(cachep, gfp, node);
583 }
584 EXPORT_SYMBOL(kmem_cache_alloc_node);
585 #endif
586
__kmem_cache_free(void * b,int size)587 static void __kmem_cache_free(void *b, int size)
588 {
589 if (size < PAGE_SIZE)
590 slob_free(b, size);
591 else
592 slob_free_pages(b, get_order(size));
593 }
594
kmem_rcu_free(struct rcu_head * head)595 static void kmem_rcu_free(struct rcu_head *head)
596 {
597 struct slob_rcu *slob_rcu = (struct slob_rcu *)head;
598 void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu));
599
600 __kmem_cache_free(b, slob_rcu->size);
601 }
602
kmem_cache_free(struct kmem_cache * c,void * b)603 void kmem_cache_free(struct kmem_cache *c, void *b)
604 {
605 kmemleak_free_recursive(b, c->flags);
606 if (unlikely(c->flags & SLAB_TYPESAFE_BY_RCU)) {
607 struct slob_rcu *slob_rcu;
608 slob_rcu = b + (c->size - sizeof(struct slob_rcu));
609 slob_rcu->size = c->size;
610 call_rcu(&slob_rcu->head, kmem_rcu_free);
611 } else {
612 __kmem_cache_free(b, c->size);
613 }
614
615 trace_kmem_cache_free(_RET_IP_, b);
616 }
617 EXPORT_SYMBOL(kmem_cache_free);
618
kmem_cache_free_bulk(struct kmem_cache * s,size_t size,void ** p)619 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
620 {
621 __kmem_cache_free_bulk(s, size, p);
622 }
623 EXPORT_SYMBOL(kmem_cache_free_bulk);
624
kmem_cache_alloc_bulk(struct kmem_cache * s,gfp_t flags,size_t size,void ** p)625 int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
626 void **p)
627 {
628 return __kmem_cache_alloc_bulk(s, flags, size, p);
629 }
630 EXPORT_SYMBOL(kmem_cache_alloc_bulk);
631
__kmem_cache_shutdown(struct kmem_cache * c)632 int __kmem_cache_shutdown(struct kmem_cache *c)
633 {
634 /* No way to check for remaining objects */
635 return 0;
636 }
637
__kmem_cache_release(struct kmem_cache * c)638 void __kmem_cache_release(struct kmem_cache *c)
639 {
640 }
641
__kmem_cache_shrink(struct kmem_cache * d)642 int __kmem_cache_shrink(struct kmem_cache *d)
643 {
644 return 0;
645 }
646
647 struct kmem_cache kmem_cache_boot = {
648 .name = "kmem_cache",
649 .size = sizeof(struct kmem_cache),
650 .flags = SLAB_PANIC,
651 .align = ARCH_KMALLOC_MINALIGN,
652 };
653
kmem_cache_init(void)654 void __init kmem_cache_init(void)
655 {
656 kmem_cache = &kmem_cache_boot;
657 slab_state = UP;
658 }
659
kmem_cache_init_late(void)660 void __init kmem_cache_init_late(void)
661 {
662 slab_state = FULL;
663 }
664
