forked from cheekyguy/memcached
/
flat_storage.c
1755 lines (1449 loc) · 64.3 KB
/
flat_storage.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/* -*- Mode: C; tab-width: 4; c-basic-offset: 4; indent-tabs-mode: nil -*- */
/**
* The flat allocator mmaps the entire region as limited by settings.maxbytes.
* When first started, we do not initialize the entire region to prevent
* unnecessary page allocation. As we need additional memory, we will
* initialize (and page in) additional memory.
*/
#include <assert.h>
#include <stdlib.h>
#include <sys/mman.h>
#include "generic.h"
#if defined(USE_FLAT_ALLOCATOR)
#define FLAT_STORAGE_MODULE
#include "assoc.h"
#include "flat_storage.h"
#include "memcached.h"
#include "stats.h"
typedef enum {
COALESCE_NO_PROGRESS, /* no progress was made in coalescing a block */
COALESCE_LARGE_CHUNK_FORMED, /* a large chunk was formed */
COALESCE_FORWARD_PROGRESS, /* a large chunk was not formed, but
* forward progress was made. */
} coalesce_progress_t;
flat_storage_info_t fsi;
/** forward declarations */
static void free_list_push(chunk_t* chunk, chunk_type_t chunk_type, bool try_merge);
static chunk_t* free_list_pop(chunk_type_t chunk_type);
static void break_large_chunk(chunk_t* chunk);
static void unbreak_large_chunk(large_chunk_t* lc, bool mandatory);
static void item_free(item *it);
/**
* flat storage code
*/
void flat_storage_init(size_t maxbytes) {
intptr_t addr;
always_assert(fsi.initialized == false);
always_assert(maxbytes % LARGE_CHUNK_SZ == 0);
always_assert(maxbytes % FLAT_STORAGE_INCREMENT_DELTA == 0);
fsi.mmap_start = mmap(NULL,
maxbytes + LARGE_CHUNK_SZ - 1, /* alloc extra to
* ensure we can align
* our buffers. */
PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANON,
-1,
0);
if (fsi.mmap_start == MAP_FAILED) {
fprintf(stderr, "failed to mmap memory\n");
exit(EXIT_FAILURE);
}
/* ensure the alignment of mmap'ed region. */
addr = (intptr_t) fsi.mmap_start;
addr = ((addr + LARGE_CHUNK_SZ - 1) / LARGE_CHUNK_SZ) * LARGE_CHUNK_SZ;
fsi.flat_storage_start = (void*) addr;
fsi.uninitialized_start = fsi.flat_storage_start;
fsi.unused_memory = maxbytes;
fsi.large_free_list = NULL_CHUNKPTR;
fsi.large_free_list_sz = 0;
fsi.small_free_list = NULL_CHUNKPTR;
fsi.small_free_list_sz = 0;
fsi.large_lru_head = NULL_CHUNKPTR;
fsi.large_lru_tail = NULL_CHUNKPTR;
fsi.small_lru_head = NULL_CHUNKPTR;
fsi.small_lru_tail = NULL_CHUNKPTR;
/* shouldn't fail here.... right? */
flat_storage_alloc();
always_assert(fsi.large_free_list_sz != 0);
fsi.initialized = 1;
}
void item_init(void) {
/* run a bunch of always_asserts to make sure that there are no inconsistent or
* inherently wrong implementation details. */
always_assert(LARGE_CHUNK_SZ >= SMALL_CHUNK_SZ);
always_assert(LARGE_CHUNK_SZ >= CHUNK_ADDRESSING_SZ);
always_assert( (LARGE_CHUNK_SZ % CHUNK_ADDRESSING_SZ) == 0 );
always_assert( (LARGE_CHUNK_SZ / SMALL_CHUNK_SZ) <=
(LARGE_CHUNK_SZ / CHUNK_ADDRESSING_SZ) );
always_assert(LARGE_TITLE_CHUNK_DATA_SZ >= KEY_MAX_LENGTH);
always_assert(SMALL_CHUNKS_PER_LARGE_CHUNK >= 2);
/* make sure that the size of the structure is what they're supposed to be. */
always_assert(sizeof(large_chunk_t) == LARGE_CHUNK_SZ);
always_assert(sizeof(large_title_chunk_t) + LARGE_CHUNK_TAIL_SZ == LARGE_CHUNK_SZ);
always_assert(sizeof(large_body_chunk_t) + LARGE_CHUNK_TAIL_SZ == LARGE_CHUNK_SZ);
always_assert(sizeof(small_chunk_t) == SMALL_CHUNK_SZ);
always_assert(sizeof(small_title_chunk_t) + SMALL_CHUNK_TAIL_SZ == SMALL_CHUNK_SZ);
always_assert(sizeof(small_body_chunk_t) + SMALL_CHUNK_TAIL_SZ == SMALL_CHUNK_SZ);
/* make sure that the fields line up in item */
always_assert( &(((item*) 0)->empty_header.h_next) == &(((item*) 0)->large_title.h_next) );
always_assert( &(((item*) 0)->empty_header.h_next) == &(((item*) 0)->small_title.h_next) );
always_assert( &(((item*) 0)->empty_header.next) == &(((item*) 0)->large_title.next) );
always_assert( &(((item*) 0)->empty_header.next) == &(((item*) 0)->small_title.next) );
always_assert( &(((item*) 0)->empty_header.prev) == &(((item*) 0)->large_title.prev) );
always_assert( &(((item*) 0)->empty_header.prev) == &(((item*) 0)->small_title.prev) );
always_assert( &(((item*) 0)->empty_header.next_chunk) == &(((item*) 0)->large_title.next_chunk) );
always_assert( &(((item*) 0)->empty_header.next_chunk) == &(((item*) 0)->small_title.next_chunk) );
always_assert( &(((item*) 0)->empty_header.time) == &(((item*) 0)->large_title.time) );
always_assert( &(((item*) 0)->empty_header.time) == &(((item*) 0)->small_title.time) );
always_assert( &(((item*) 0)->empty_header.exptime) == &(((item*) 0)->large_title.exptime) );
always_assert( &(((item*) 0)->empty_header.exptime) == &(((item*) 0)->small_title.exptime) );
always_assert( &(((item*) 0)->empty_header.nbytes) == &(((item*) 0)->large_title.nbytes) );
always_assert( &(((item*) 0)->empty_header.nbytes) == &(((item*) 0)->small_title.nbytes) );
always_assert( &(((item*) 0)->empty_header.refcount) == &(((item*) 0)->large_title.refcount) );
always_assert( &(((item*) 0)->empty_header.refcount) == &(((item*) 0)->small_title.refcount) );
always_assert( &(((item*) 0)->empty_header.nkey) == &(((item*) 0)->large_title.nkey) );
always_assert( &(((item*) 0)->empty_header.nkey) == &(((item*) 0)->small_title.nkey) );
/* make sure that the casting functions in flat_storage.h are sane. */
always_assert( (void*) &(((item*) 0)->small_title) == ((void*) 0));
always_assert( (void*) &(((item*) 0)->large_title) == ((void*) 0));
/* ensure that the first piece of a broken large chunk is the same address
* as the large_title_chunk / large_body_chunk. this is so that the
* chunkptrs map correctly. */
always_assert( (intptr_t) &(((large_chunk_t*) 0)->lc_broken.lbc[0]) ==
(intptr_t) &(((large_chunk_t*) 0)->lc_title) );
always_assert(FLAT_STORAGE_INCREMENT_DELTA % LARGE_CHUNK_SZ == 0);
}
static inline size_t __fss_MIN(size_t a, size_t b) {
if (a < b) {
return a;
} else {
return b;
}
}
static inline size_t __fss_MAX(size_t a, size_t b) {
if (a > b) {
return a;
} else {
return b;
}
}
/* Get the next CAS id for a new item. */
static uint64_t get_cas_id() {
static uint64_t cas_id = 0;
return ++cas_id;
}
/* initialize at least nbytes more memory and add them as large chunks to the
* free list. */
STATIC bool flat_storage_alloc(void) {
large_chunk_t* initialize_end;
if (FLAT_STORAGE_INCREMENT_DELTA > fsi.unused_memory) {
return false;
}
initialize_end = fsi.uninitialized_start + (FLAT_STORAGE_INCREMENT_DELTA / LARGE_CHUNK_SZ);
stats.item_storage_allocated += FLAT_STORAGE_INCREMENT_DELTA;
/* initialize the large chunks. */
for (;
fsi.uninitialized_start < initialize_end;
fsi.unused_memory -= LARGE_CHUNK_SZ) {
large_chunk_t* chunk = fsi.uninitialized_start;
fsi.uninitialized_start ++;
chunk->flags = LARGE_CHUNK_INITIALIZED;
free_list_push( (chunk_t*) chunk, LARGE_CHUNK, false);
}
return true;
}
/* pushes the chunk onto the free list. the chunk must be initialized but not
* used. afterwards, the flags will be set to INITIALIZED | FREE. if
* try_merge is true and chunk_type is SMALL_CHUNK, it will try to coalesce the
* parent block. */
static void free_list_push(chunk_t* chunk, chunk_type_t chunk_type,
bool try_merge) {
/* can't merge large chunks */
assert(! (try_merge == true && chunk_type == LARGE_CHUNK) );
switch (chunk_type) {
case SMALL_CHUNK:
{
large_chunk_t* pc;
assert( SMALL_CHUNK_INITIALIZED ==
chunk->sc.flags );
/* adjust the allocated count for the parent chunk */
pc = get_parent_chunk(&(chunk->sc));
assert(pc->lc_broken.small_chunks_allocated > 0);
fsi.stats.broken_chunk_histogram[pc->lc_broken.small_chunks_allocated] --; /* STATS: update */
pc->lc_broken.small_chunks_allocated --;
fsi.stats.broken_chunk_histogram[pc->lc_broken.small_chunks_allocated] ++; /* STATS: update */
/* add ourselves to the free list.
*
* ttung NOTE: if small chunk fragmentation is severe (i.e., we
* spend too much time migrating data in
* coalesce_free_small_chunks(..)), then we can employ strategies
* to keep small chunks with its siblings on the free list. */
if (fsi.small_free_list != NULL_CHUNKPTR) {
chunk_t* old_head;
old_head = (chunk_t*) fsi.small_free_list;
old_head->sc.sc_free.prev_next = &chunk->sc.sc_free.next;
chunk->sc.sc_free.next = fsi.small_free_list;
} else {
chunk->sc.sc_free.next = NULL_CHUNKPTR;
}
chunk->sc.sc_free.prev_next = &fsi.small_free_list;
fsi.small_free_list = &(chunk->sc);
fsi.small_free_list_sz ++;
chunk->sc.flags = (SMALL_CHUNK_INITIALIZED | SMALL_CHUNK_FREE);
if (try_merge) {
unbreak_large_chunk(pc, false);
}
}
break;
case LARGE_CHUNK:
assert( LARGE_CHUNK_INITIALIZED ==
chunk->lc.flags );
if (fsi.large_free_list != NULL_CHUNKPTR) {
chunk_t* old_head;
old_head = (chunk_t*) fsi.large_free_list;
chunk->lc.lc_free.next = fsi.large_free_list;
} else {
chunk->lc.lc_free.next = NULL_CHUNKPTR;
}
fsi.large_free_list = &(chunk->lc);
fsi.large_free_list_sz ++;
chunk->lc.flags = (LARGE_CHUNK_INITIALIZED | LARGE_CHUNK_FREE);
break;
}
}
/* gets the first chunk from the free list. if no chunk is available, then
* NULL is returned. */
static chunk_t* free_list_pop(chunk_type_t chunk_type) {
chunk_t* retval;
switch (chunk_type) {
case SMALL_CHUNK:
{
large_chunk_t* parent_chunk;
if (fsi.small_free_list_sz == 0) {
return NULL;
}
retval = (chunk_t*) fsi.small_free_list;
parent_chunk = get_parent_chunk(&(retval->sc));
fsi.stats.broken_chunk_histogram[parent_chunk->lc_broken.small_chunks_allocated] --; /* STATS: update */
parent_chunk->lc_broken.small_chunks_allocated ++;
fsi.stats.broken_chunk_histogram[parent_chunk->lc_broken.small_chunks_allocated] ++; /* STATS: update */
assert(parent_chunk->lc_broken.small_chunks_allocated <= SMALL_CHUNKS_PER_LARGE_CHUNK);
/* remove ourselves from the free list. */
fsi.small_free_list = retval->sc.sc_free.next;
if (fsi.small_free_list != NULL_CHUNKPTR) {
chunk_t* new_head;
new_head = (chunk_t*) fsi.small_free_list;
new_head->sc.sc_free.prev_next = &fsi.small_free_list;
}
fsi.small_free_list_sz --;
/* do some sanity checks on the chunk */
assert( (SMALL_CHUNK_INITIALIZED | SMALL_CHUNK_FREE) ==
retval->sc.flags );
/* unmark free flag */
retval->sc.flags &= (~SMALL_CHUNK_FREE);
return retval;
}
case LARGE_CHUNK:
if (fsi.large_free_list_sz == 0) {
return NULL;
}
retval = (chunk_t*) fsi.large_free_list;
fsi.large_free_list = retval->lc.lc_free.next;
if (fsi.large_free_list != NULL_CHUNKPTR) {
chunk_t* new_head;
new_head = (chunk_t*) fsi.large_free_list;
}
fsi.large_free_list_sz --;
/* do some sanity checks on the chunk */
assert( (LARGE_CHUNK_INITIALIZED | LARGE_CHUNK_FREE) ==
retval->lc.flags );
/* unmark free flag */
retval->lc.flags &= (~LARGE_CHUNK_FREE);
return retval;
}
return NULL;
}
/* this takes an unused large chunk and breaks it into small chunks. it adds
* the small chunks to the small chunk free list.
*/
static void break_large_chunk(chunk_t* chunk) {
int i;
assert( LARGE_CHUNK_INITIALIZED == chunk->lc.flags );
chunk->lc.flags |= LARGE_CHUNK_USED | LARGE_CHUNK_BROKEN;
/* this is a dummy kludge to ensure that free_list_push can go ahead and
* decrement the allocated count safely. */
chunk->lc.lc_broken.small_chunks_allocated = SMALL_CHUNKS_PER_LARGE_CHUNK;
fsi.stats.broken_chunk_histogram[SMALL_CHUNKS_PER_LARGE_CHUNK] ++; /* STATS: update */
for (i = SMALL_CHUNKS_PER_LARGE_CHUNK - 1; i >= 0; i --) {
small_chunk_t* small_chunk = &(chunk->lc.lc_broken.lbc[i]);
small_chunk->flags = SMALL_CHUNK_INITIALIZED;
free_list_push( (chunk_t*) small_chunk, SMALL_CHUNK, false);
}
chunk->lc.lc_broken.small_chunks_allocated = 0;
/* STATS: update */
fsi.stats.large_broken_chunks ++;
fsi.stats.break_events ++;
}
/* take a large broken chunk and unbreak it.
*/
static void unbreak_large_chunk(large_chunk_t* lc, bool mandatory) {
int i;
assert( (LARGE_CHUNK_INITIALIZED | LARGE_CHUNK_USED | LARGE_CHUNK_BROKEN) ==
lc->flags );
/* if the coalesce is not mandatory, then make sure there are no used
* chunks. */
if (! mandatory) {
if (lc->lc_broken.small_chunks_allocated != 0) {
return;
}
#if !defined(NDEBUG)
for (i = 0; i < SMALL_CHUNKS_PER_LARGE_CHUNK; i ++) {
small_chunk_t* small_chunk = &(lc->lc_broken.lbc[i]);
assert(small_chunk->flags & SMALL_CHUNK_INITIALIZED);
assert((small_chunk->flags & SMALL_CHUNK_USED) == 0);
}
#endif /* #if !defined(NDEBUG) */
} else {
assert(lc->lc_broken.small_chunks_allocated == 0);
}
for (i = 0; i < SMALL_CHUNKS_PER_LARGE_CHUNK; i ++) {
small_chunk_t* small_chunk = &(lc->lc_broken.lbc[i]);
small_chunk_t** prev_next;
/* some sanity checks */
assert(small_chunk->flags == (SMALL_CHUNK_INITIALIZED |
SMALL_CHUNK_FREE) ||
small_chunk->flags == (SMALL_CHUNK_INITIALIZED |
SMALL_CHUNK_COALESCE_PENDING));
/* remove this chunk from the free list */
if (small_chunk->flags & SMALL_CHUNK_FREE) {
prev_next = small_chunk->sc_free.prev_next;
assert(*prev_next == small_chunk);
*(prev_next) = small_chunk->sc_free.next;
if (small_chunk->sc_free.next != NULL_CHUNKPTR) {
small_chunk_t* next = small_chunk->sc_free.next;
next->sc_free.prev_next = prev_next;
}
small_chunk->flags = 0;
fsi.small_free_list_sz --;
}
}
lc->flags = LARGE_CHUNK_INITIALIZED;
free_list_push( (chunk_t*) lc, LARGE_CHUNK, false);
/* STATS: update */
fsi.stats.large_broken_chunks --;
fsi.stats.broken_chunk_histogram[0] --;
fsi.stats.unbreak_events ++;
}
/*
* gets the oldest item on the LRU with refcount == 0. if chunk_type is
* SMALL_CHUNK, the search will start at start instead of the LRU tail.
*/
STATIC item* get_lru_item(chunk_type_t chunk_type, small_title_chunk_t* start) {
switch (chunk_type) {
case SMALL_CHUNK:
{
small_chunk_t* small_chunk_iter;
int i;
if (start == NULL) {
small_chunk_iter = (small_chunk_t*) fsi.small_lru_tail;
} else {
small_chunk_iter = (small_chunk_t*) start;
}
for (i = 0;
i < SMALL_LRU_SEARCH_DEPTH && small_chunk_iter != NULL;
i ++) {
chunkptr_t prev_chunkptr;
chunk_t* prev;
small_title_chunk_t* small_title;
assert(small_chunk_iter->flags == (SMALL_CHUNK_INITIALIZED |
SMALL_CHUNK_USED |
SMALL_CHUNK_TITLE));
small_title = &small_chunk_iter->sc_title;
if (small_title->refcount == 0) {
/* it is silly that we have to make this typecast, but
* there's no other way to make this assignment without
* a cast, even though it ought to be possible. */
item* small_item = get_item_from_small_title(small_title);
return small_item;
}
prev_chunkptr = small_title->prev;
if (prev_chunkptr == NULL_CHUNKPTR) {
return NULL;
}
prev = get_chunk_address(prev_chunkptr);
small_chunk_iter = &(prev->sc);
}
}
break;
case LARGE_CHUNK:
{
large_chunk_t* large_chunk_iter;
int i;
for (i = 0, large_chunk_iter = (large_chunk_t*) fsi.large_lru_tail;
i < LARGE_LRU_SEARCH_DEPTH && large_chunk_iter != NULL;
i ++) {
chunkptr_t prev_chunkptr;
chunk_t* prev;
large_title_chunk_t* large_title;
assert(large_chunk_iter->flags == (LARGE_CHUNK_INITIALIZED |
LARGE_CHUNK_USED |
LARGE_CHUNK_TITLE));
large_title = &large_chunk_iter->lc_title;
if (large_title->refcount == 0) {
/* it is silly that we have to make this typecast, but
* there's no other way to make this assignment without
* a cast, even though it ought to be possible. */
item* large_item = get_item_from_large_title(large_title);
return large_item;
}
prev_chunkptr = large_title->prev;
if (prev_chunkptr == NULL_CHUNKPTR) {
return NULL;
}
prev = get_chunk_address(prev_chunkptr);
large_chunk_iter = &(prev->lc);
}
}
break;
}
return NULL;
}
static bool small_chunk_referenced(const small_chunk_t* sc) {
assert((sc->flags & SMALL_CHUNK_INITIALIZED) != 0);
if (sc->flags & SMALL_CHUNK_FREE) {
return false; /* free nodes count as refcount = 0. */
} else {
for (;
(sc->flags & SMALL_CHUNK_TITLE) == 0;
sc = &get_chunk_address(sc->sc_body.prev_chunk)->sc) {
assert((sc->flags & (SMALL_CHUNK_INITIALIZED | SMALL_CHUNK_USED)) ==
(SMALL_CHUNK_INITIALIZED | SMALL_CHUNK_USED));
}
assert((sc->flags & (SMALL_CHUNK_INITIALIZED | SMALL_CHUNK_USED | SMALL_CHUNK_TITLE)) ==
(SMALL_CHUNK_INITIALIZED | SMALL_CHUNK_USED | SMALL_CHUNK_TITLE));
return (sc->sc_title.refcount == 0) ? false : true;
}
}
static bool large_broken_chunk_referenced(const large_broken_chunk_t* lc) {
unsigned counter;
for (counter = 0;
counter < SMALL_CHUNKS_PER_LARGE_CHUNK;
counter ++) {
const small_chunk_t* iter = &(lc->lbc[counter]);
if (small_chunk_referenced(iter)) {
return true;
}
}
return false;
}
/*
* if search_depth is zero, then the search depth is not limited. if the search
* depth is non-zero, constrain search to the first search_depth items on the
* small free list
*/
static large_chunk_t* find_unreferenced_broken_chunk(size_t search_depth) {
small_chunk_t* small_chunk_iter;
unsigned counter;
for (counter = 0,
small_chunk_iter = fsi.small_free_list;
small_chunk_iter != NULL && (search_depth == 0 || counter < search_depth);
counter ++,
small_chunk_iter = small_chunk_iter->sc_free.next) {
large_chunk_t* lc = get_parent_chunk(small_chunk_iter);
large_broken_chunk_t* pc = &(lc->lc_broken);
if (large_broken_chunk_referenced(pc) == false) {
return lc;
}
}
return NULL;
}
/*
* coalesce small free chunks to form a single large free chunk. two things are
* needed to perform this operation:
* 1) at least SMALL_CHUNKS_PER_LARGE_CHUNK free small chunks.
* 2) a large broken chunk that has refcount == 0 so we can move items off of it.
*
* if there aren't at least SMALL_CHUNKS_PER_LARGE_CHUNK, free enough small
* chunks to satisfy that requirement. if we have accidentally freed up a large
* free chunk, then return immediately.
*
* once the first criteria is satisfied, search through the freelist to find a
* parent chunk that has refcount == 0. if that fails, search through the LRU
* (search limited by SMALL_LRU_SEARCH_DEPTH) for such a chunk.
*
* if during the freeing process, the timestamp on the LRU small item is newer
* than large_lru_item_timestamp, then stop.
*
* returns COALESCE_NO_PROGRESS if no forward progress was made.
* COALESCE_LARGE_CHUNK_FORMED if a large chunk was formed.
* COALESCE_FORWARD_PROGRESS if a large chunk was not formed, but
* forward progress was made.
*/
static coalesce_progress_t coalesce_free_small_chunks(rel_time_t large_lru_item_timestamp) {
size_t large_free_list_sz_pre = fsi.large_free_list_sz,
small_free_list_sz_pre = fsi.small_free_list_sz;
large_chunk_t* lc;
int i;
item* small_lru_item;
while (fsi.small_free_list_sz < SMALL_CHUNKS_PER_LARGE_CHUNK) {
small_lru_item = get_lru_item(SMALL_CHUNK, NULL);
if (small_lru_item == NULL ||
small_lru_item->small_title.time > large_lru_item_timestamp) {
/* nothing more to free or large LRU is newer, so go free that */
/* instead. */
return ( small_free_list_sz_pre == fsi.small_free_list_sz ) ?
COALESCE_NO_PROGRESS : COALESCE_FORWARD_PROGRESS;
}
assert(small_lru_item->empty_header.refcount == 0);
do_item_unlink(small_lru_item, UNLINK_MAYBE_EVICT);
if (fsi.large_free_list_sz > large_free_list_sz_pre) {
/* the prior free released an entire large chunk. */
return COALESCE_LARGE_CHUNK_FORMED;
}
}
lc = find_unreferenced_broken_chunk(0);
if (lc == NULL) {
/* we don't want to be stuck in an infinite loop if we can't find a
* large unreferenced chunk, so just report no progress. */
return COALESCE_NO_PROGRESS;
}
/* STATS: update */
fsi.stats.broken_chunk_histogram[lc->lc_broken.small_chunks_allocated] --;
fsi.stats.migrates += lc->lc_broken.small_chunks_allocated;
if (lc->lc_broken.small_chunks_allocated != 0) {
/* any free small chunks that belong to the same parent chunk should be
* removed from the free list. this is to ensure that we don't pick them
* up as replacement blocks. */
for (i = 0; i < SMALL_CHUNKS_PER_LARGE_CHUNK; i ++) {
small_chunk_t* iter = &(lc->lc_broken.lbc[i]);
if (iter->flags & SMALL_CHUNK_FREE) {
/* need to remove this from the free list */
small_chunk_t** prev_next;
prev_next = iter->sc_free.prev_next;
assert(*prev_next == iter);
*(prev_next) = iter->sc_free.next;
if (iter->sc_free.next != NULL_CHUNKPTR) {
small_chunk_t* next = iter->sc_free.next;
next->sc_free.prev_next = prev_next;
}
iter->flags &= ~(SMALL_CHUNK_FREE);
iter->flags |= SMALL_CHUNK_COALESCE_PENDING;
fsi.small_free_list_sz --;
}
}
for (i = 0; i < SMALL_CHUNKS_PER_LARGE_CHUNK; i ++) {
small_chunk_t* iter = &(lc->lc_broken.lbc[i]);
chunk_t* old_chunk = (chunk_t*) iter;
(void) old_chunk; /* when optimizing, old_chunk is not
* used. this is to quiesce the
* compiler warning. */
assert( (iter->flags & SMALL_CHUNK_INITIALIZED) ==
SMALL_CHUNK_INITIALIZED);
if (iter->flags & SMALL_CHUNK_USED) {
/* title block */
chunk_t* _replacement = free_list_pop(SMALL_CHUNK);
small_chunk_t* replacement;
chunkptr_t replacement_chunkptr;
assert(_replacement != NULL);
replacement = &(_replacement->sc);
assert(replacement->flags == (SMALL_CHUNK_INITIALIZED));
memcpy(replacement, iter, sizeof(small_chunk_t));
replacement_chunkptr = get_chunkptr(_replacement);
if (iter->flags & SMALL_CHUNK_TITLE) {
item* new_it;
chunk_t* next, * prev;
small_chunk_t* next_chunk;
new_it = get_item_from_small_title(&(replacement->sc_title));
/* edit the forward and backward links. */
if (replacement->sc_title.next != NULL_CHUNKPTR) {
next = get_chunk_address(replacement->sc_title.next);
assert(next->sc.sc_title.prev == get_chunkptr(old_chunk));
next->sc.sc_title.prev = replacement_chunkptr;
} else {
assert(fsi.small_lru_tail == get_item_from_small_title(&old_chunk->sc.sc_title));
fsi.small_lru_tail = get_item_from_small_title(&replacement->sc_title);
}
if (replacement->sc_title.prev != NULL_CHUNKPTR) {
prev = get_chunk_address(replacement->sc_title.prev);
assert(prev->sc.sc_title.next == get_chunkptr(old_chunk));
prev->sc.sc_title.next = replacement_chunkptr;
} else {
assert(fsi.small_lru_head == get_item_from_small_title(&old_chunk->sc.sc_title));
fsi.small_lru_head = get_item_from_small_title(&replacement->sc_title);
}
/* edit the next_chunk's prev_chunk link */
next_chunk = &(get_chunk_address(replacement->sc_title.next_chunk))->sc;
if (next_chunk != NULL) {
assert(next_chunk->sc_body.prev_chunk == get_chunkptr(old_chunk));
next_chunk->sc_body.prev_chunk = replacement_chunkptr;
}
/* update flags */
replacement->flags |= (SMALL_CHUNK_USED | SMALL_CHUNK_TITLE);
/* do the replacement in the mapping. */
assoc_update(new_it);
} else {
/* body block. this is more straightforward */
small_chunk_t* prev_chunk = &(get_chunk_address(replacement->sc_body.prev_chunk))->sc;
small_chunk_t* next_chunk = &(get_chunk_address(replacement->sc_body.next_chunk))->sc;;
/* update the previous block's next pointer */
if (prev_chunk->flags & SMALL_CHUNK_TITLE) {
prev_chunk->sc_title.next_chunk = replacement_chunkptr;
} else {
prev_chunk->sc_body.next_chunk = replacement_chunkptr;
}
/* edit the next_chunk's prev_chunk link */
if (next_chunk != NULL) {
assert(next_chunk->sc_body.prev_chunk == get_chunkptr(old_chunk));
next_chunk->sc_body.prev_chunk = replacement_chunkptr;
}
/* update flags */
replacement->flags |= (SMALL_CHUNK_USED);
}
/* don't push this onto the free list. if we do, we'll immediately
* pick it up when finding a replacement block. instead, just mark
* it coalesce-pending. */
iter->flags = SMALL_CHUNK_INITIALIZED | SMALL_CHUNK_COALESCE_PENDING;
/* decrement the number of blocks allocated */
lc->lc_broken.small_chunks_allocated --;
}
}
}
/* STATS: update */
fsi.stats.broken_chunk_histogram[0] ++;
unbreak_large_chunk(lc, true);
return COALESCE_LARGE_CHUNK_FORMED;
}
static bool flat_storage_lru_evict(chunk_type_t chunk_type, size_t nchunks) {
while (1) {
bool release_small = false, release_large = false;
rel_time_t * small_lru_item_timestamp, * large_lru_item_timestamp;
item* small_lru_item = NULL, * large_lru_item = NULL;
size_t nreleased; /* number of chunks freed as a result
* of the release. */
/* get the LRU items */
small_lru_item = get_lru_item(SMALL_CHUNK, NULL);
large_lru_item = get_lru_item(LARGE_CHUNK, NULL);
/* it is possible that get_lru_item(..) will return NULL, but we'll
* never use these pointers without checking the return values. this
* is just an elaborate way to circumvent the compiler's warning that
* large_lru_item_timestamp may be used uninitialized. */
small_lru_item_timestamp = &small_lru_item->small_title.time;
large_lru_item_timestamp = &large_lru_item->large_title.time;
/* have the items, figure out which item to release. */
if (small_lru_item == NULL &&
large_lru_item == NULL) {
return false;
} else if (small_lru_item == NULL && large_lru_item != NULL) {
release_large = true;
} else if (small_lru_item != NULL && large_lru_item == NULL) {
release_small = true;
} else {
/* tie goes to large items, because they're easier to deal with. in
* any case, this is extraordinarily unlikely. */
if (*small_lru_item_timestamp < *large_lru_item_timestamp) {
release_small = true;
} else {
release_large = true;
}
}
switch (chunk_type) {
case SMALL_CHUNK:
/* several possibilities exist:
* 1) small chunks exist in LRU but not in the large LRU.
* release the item at the tail. check if we've released
* enough to satisfy the request.
* 2) small chunks do not exist in LRU, but exist in large chunk
* LRU. release the item at the tail. break up enough to
* satisfy the request and see if we've released enough to
* satisfy the request.
* 3) both LRUs have items. if the item at the tail of the
* small LRU is accessed less recently than the item at the
* tail of the large LRU, release the item at the tail of the
* small LRU. decrement nchunks by the number of chunks
* we've freed up.
* 4) both LRUs have items. if the item at the tail of the
* large LRU is accessed less recently than the item at the
* tail of the small LRU, release the item at the tail of the
* large LRU. break up enough to satisfy the request and
* decrement nchunks.
*/
/* handles case 1 and 3 */
if (release_small) {
assert(small_lru_item->small_title.refcount == 0);
nreleased = chunks_in_item(small_lru_item);
do_item_unlink(small_lru_item, UNLINK_MAYBE_EVICT);
if (fsi.small_free_list_sz +
(fsi.large_free_list_sz * SMALL_CHUNKS_PER_LARGE_CHUNK) >=
nchunks) {
/* we have enough chunks for the allocation */
return true;
}
}
/* handles case 2 and 4 */
if (release_large) {
assert(large_lru_item->large_title.refcount == 0);
nreleased = chunks_in_item(large_lru_item);
do_item_unlink(large_lru_item, UNLINK_MAYBE_EVICT);
while (fsi.large_free_list_sz > 0 &&
fsi.small_free_list_sz < nchunks) {
chunk_t* chunk = free_list_pop(LARGE_CHUNK);
break_large_chunk(chunk);
}
if (fsi.small_free_list_sz >= nchunks) {
return true;
}
}
break;
case LARGE_CHUNK:
/* several possibilities exist:
* 1) small chunks exist in LRU but not in the large LRU.
* release the item at the tail. free small chunks until we
* have enough to unbreak a large chunk.
* 2) small chunks do not exist in LRU, but exist in large chunk
* LRU. release the item at the tail. check if we've
* released enough to satisfy the request.
* 3) both LRUs have items. if the item at the tail of the
* small LRU is accessed less recently than the item at the
* tail of the large LRU, release the item at the tail of the
* small LRU. continue freeing items until we have enough
* space to unbreak a large chunk OR if the large LRU is
* accessed less recently than the small LRU.
* 4) both LRUs have items. if the item at the tail of the
* large LRU is accessed less recently than the item at the
* tail of the small LRU, release the item at the tail of the
* large LRU.
*/
if (release_small) {
/* this is now painfully complicated. we need to coalesce
* free small chunks to free up space to unbreak a large
* chunk. if the large LRU has items, unbreak until the
* large LRU is accessed less recently than the small LRU.
* otherwise, just keep unbreaking. */
rel_time_t stop_time;
if (large_lru_item == NULL) {
stop_time = current_time;
} else {
stop_time = *large_lru_item_timestamp;
}
if (coalesce_free_small_chunks(stop_time) == COALESCE_NO_PROGRESS) {
return false;
}
if (fsi.large_free_list_sz >= nchunks) {
return true;
}
}
if (release_large) {
assert(large_lru_item->large_title.refcount == 0);
nreleased = chunks_in_item(large_lru_item);
do_item_unlink(large_lru_item, UNLINK_MAYBE_EVICT);
if (fsi.large_free_list_sz >= nchunks) {
return true;
}
}
break;
}
}
}
static int do_stamp_on_block(char* block_start, size_t block_offset, size_t block_sz,
const rel_time_t now, const struct in_addr addr) {
int retflags = 0;
assert(block_offset <= block_sz);
if (block_sz - block_offset >= sizeof(now)) {
memcpy(&block_start[block_offset], &now, sizeof(now));
retflags |= ITEM_HAS_TIMESTAMP;
block_offset += sizeof(now);
}
if (block_sz - block_offset >= sizeof(addr)) {
memcpy(&block_start[block_offset], &addr, sizeof(addr));
retflags |= ITEM_HAS_IP_ADDRESS;
block_offset += sizeof(addr);
}
assert(block_offset <= block_sz);
return retflags;
}
void item_memcpy_to(item* it, size_t offset, const void* src, size_t nbytes,
bool beyond_item_boundary) {
#define MEMCPY_TO_APPLIER(it, ptr, bytes) \
memcpy((ptr), src, bytes); \
src += bytes;
ITEM_WALK(it, offset, nbytes, beyond_item_boundary, MEMCPY_TO_APPLIER, );
#undef MEMCPY_TO_APPLIER
}
void item_memcpy_from(void* dst, const item* it, size_t offset, size_t nbytes,
bool beyond_item_boundary) {
#define MEMCPY_FROM_APPLIER(it, ptr, bytes) \
memcpy(dst, (ptr), bytes); \
dst += bytes;
ITEM_WALK(it, offset, nbytes, beyond_item_boundary, MEMCPY_FROM_APPLIER, const);
#undef MEMCPY_FROM_APPLIER
}
void do_try_item_stamp(item* it, rel_time_t now, const struct in_addr addr) {
int slack;
size_t offset = 0;
it->empty_header.it_flags &= ~(ITEM_HAS_TIMESTAMP | ITEM_HAS_IP_ADDRESS);
slack = item_slackspace(it);
/* timestamp gets priority */
if (slack >= sizeof(now)) {
rel_time_t now = current_time;
item_memcpy_to(it, it->empty_header.nbytes + offset, &now, sizeof(now), true);
it->empty_header.it_flags |= ITEM_HAS_TIMESTAMP;
slack -= sizeof(now);
offset += sizeof(now);
}
/* still enough space for the ip address? */
if (slack >= sizeof(addr)) {
/* enough space for both the timestamp and the ip address */
/* save the address */
item_memcpy_to(it, it->empty_header.nbytes + offset, &addr, sizeof(addr), true);
it->empty_header.it_flags |= ITEM_HAS_IP_ADDRESS;
slack -= sizeof(addr);
offset += sizeof(addr);
}
}
/* allocates one item capable of storing a key of size nkey and a value field of
* size nbytes. stores the key, flags, and exptime. the value field is not
* initialized. if there is insufficient memory, NULL is returned. */
item* do_item_alloc(char *key, const size_t nkey, const int flags, const rel_time_t exptime,
const size_t nbytes, const struct in_addr addr) {
if (item_size_ok(nkey, flags, nbytes) == false) {
return NULL;
}
if (is_large_chunk(nkey, nbytes)) {
/* allocate a large chunk */
/* try various strategies to get a free item:
* 1) free_list
* 2) flat_storage_alloc
* 3) if we have sufficient small free chunks to constitute a large free
* chunk, do so.