summaryrefslogtreecommitdiff
path: root/src/basic/hashmap.c
blob: fbc8d397a2be412179dd1301b9d40fc672819f93 (plain)
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
/* SPDX-License-Identifier: LGPL-2.1+ */
/***
  This file is part of systemd.

  Copyright 2010 Lennart Poettering
  Copyright 2014 Michal Schmidt

  systemd is free software; you can redistribute it and/or modify it
  under the terms of the GNU Lesser General Public License as published by
  the Free Software Foundation; either version 2.1 of the License, or
  (at your option) any later version.

  systemd is distributed in the hope that it will be useful, but
  WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  Lesser General Public License for more details.

  You should have received a copy of the GNU Lesser General Public License
  along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/

#include <errno.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>

#include "alloc-util.h"
#include "hashmap.h"
#include "fileio.h"
#include "macro.h"
#include "mempool.h"
#include "process-util.h"
#include "random-util.h"
#include "set.h"
#include "siphash24.h"
#include "string-util.h"
#include "strv.h"
#include "util.h"

#if ENABLE_DEBUG_HASHMAP
#include <pthread.h>
#include "list.h"
#endif

/*
 * Implementation of hashmaps.
 * Addressing: open
 *   - uses less RAM compared to closed addressing (chaining), because
 *     our entries are small (especially in Sets, which tend to contain
 *     the majority of entries in systemd).
 * Collision resolution: Robin Hood
 *   - tends to equalize displacement of entries from their optimal buckets.
 * Probe sequence: linear
 *   - though theoretically worse than random probing/uniform hashing/double
 *     hashing, it is good for cache locality.
 *
 * References:
 * Celis, P. 1986. Robin Hood Hashing.
 * Ph.D. Dissertation. University of Waterloo, Waterloo, Ont., Canada, Canada.
 * https://cs.uwaterloo.ca/research/tr/1986/CS-86-14.pdf
 * - The results are derived for random probing. Suggests deletion with
 *   tombstones and two mean-centered search methods. None of that works
 *   well for linear probing.
 *
 * Janson, S. 2005. Individual displacements for linear probing hashing with different insertion policies.
 * ACM Trans. Algorithms 1, 2 (October 2005), 177-213.
 * DOI=10.1145/1103963.1103964 http://doi.acm.org/10.1145/1103963.1103964
 * http://www.math.uu.se/~svante/papers/sj157.pdf
 * - Applies to Robin Hood with linear probing. Contains remarks on
 *   the unsuitability of mean-centered search with linear probing.
 *
 * Viola, A. 2005. Exact distribution of individual displacements in linear probing hashing.
 * ACM Trans. Algorithms 1, 2 (October 2005), 214-242.
 * DOI=10.1145/1103963.1103965 http://doi.acm.org/10.1145/1103963.1103965
 * - Similar to Janson. Note that Viola writes about C_{m,n} (number of probes
 *   in a successful search), and Janson writes about displacement. C = d + 1.
 *
 * Goossaert, E. 2013. Robin Hood hashing: backward shift deletion.
 * http://codecapsule.com/2013/11/17/robin-hood-hashing-backward-shift-deletion/
 * - Explanation of backward shift deletion with pictures.
 *
 * Khuong, P. 2013. The Other Robin Hood Hashing.
 * http://www.pvk.ca/Blog/2013/11/26/the-other-robin-hood-hashing/
 * - Short summary of random vs. linear probing, and tombstones vs. backward shift.
 */

/*
 * XXX Ideas for improvement:
 * For unordered hashmaps, randomize iteration order, similarly to Perl:
 * http://blog.booking.com/hardening-perls-hash-function.html
 */

/* INV_KEEP_FREE = 1 / (1 - max_load_factor)
 * e.g. 1 / (1 - 0.8) = 5 ... keep one fifth of the buckets free. */
#define INV_KEEP_FREE            5U

/* Fields common to entries of all hashmap/set types */
struct hashmap_base_entry {
        const void *key;
};

/* Entry types for specific hashmap/set types
 * hashmap_base_entry must be at the beginning of each entry struct. */

struct plain_hashmap_entry {
        struct hashmap_base_entry b;
        void *value;
};

struct ordered_hashmap_entry {
        struct plain_hashmap_entry p;
        unsigned iterate_next, iterate_previous;
};

struct set_entry {
        struct hashmap_base_entry b;
};

/* In several functions it is advantageous to have the hash table extended
 * virtually by a couple of additional buckets. We reserve special index values
 * for these "swap" buckets. */
#define _IDX_SWAP_BEGIN     (UINT_MAX - 3)
#define IDX_PUT             (_IDX_SWAP_BEGIN + 0)
#define IDX_TMP             (_IDX_SWAP_BEGIN + 1)
#define _IDX_SWAP_END       (_IDX_SWAP_BEGIN + 2)

#define IDX_FIRST           (UINT_MAX - 1) /* special index for freshly initialized iterators */
#define IDX_NIL             UINT_MAX       /* special index value meaning "none" or "end" */

assert_cc(IDX_FIRST == _IDX_SWAP_END);
assert_cc(IDX_FIRST == _IDX_ITERATOR_FIRST);

/* Storage space for the "swap" buckets.
 * All entry types can fit into a ordered_hashmap_entry. */
struct swap_entries {
        struct ordered_hashmap_entry e[_IDX_SWAP_END - _IDX_SWAP_BEGIN];
};

/* Distance from Initial Bucket */
typedef uint8_t dib_raw_t;
#define DIB_RAW_OVERFLOW ((dib_raw_t)0xfdU)   /* indicates DIB value is greater than representable */
#define DIB_RAW_REHASH   ((dib_raw_t)0xfeU)   /* entry yet to be rehashed during in-place resize */
#define DIB_RAW_FREE     ((dib_raw_t)0xffU)   /* a free bucket */
#define DIB_RAW_INIT     ((char)DIB_RAW_FREE) /* a byte to memset a DIB store with when initializing */

#define DIB_FREE UINT_MAX

#if ENABLE_DEBUG_HASHMAP
struct hashmap_debug_info {
        LIST_FIELDS(struct hashmap_debug_info, debug_list);
        unsigned max_entries;  /* high watermark of n_entries */

        /* who allocated this hashmap */
        int line;
        const char *file;
        const char *func;

        /* fields to detect modification while iterating */
        unsigned put_count;    /* counts puts into the hashmap */
        unsigned rem_count;    /* counts removals from hashmap */
        unsigned last_rem_idx; /* remembers last removal index */
};

/* Tracks all existing hashmaps. Get at it from gdb. See sd_dump_hashmaps.py */
static LIST_HEAD(struct hashmap_debug_info, hashmap_debug_list);
static pthread_mutex_t hashmap_debug_list_mutex = PTHREAD_MUTEX_INITIALIZER;

#define HASHMAP_DEBUG_FIELDS struct hashmap_debug_info debug;

#else /* !ENABLE_DEBUG_HASHMAP */
#define HASHMAP_DEBUG_FIELDS
#endif /* ENABLE_DEBUG_HASHMAP */

enum HashmapType {
        HASHMAP_TYPE_PLAIN,
        HASHMAP_TYPE_ORDERED,
        HASHMAP_TYPE_SET,
        _HASHMAP_TYPE_MAX
};

struct _packed_ indirect_storage {
        void *storage;                     /* where buckets and DIBs are stored */
        uint8_t  hash_key[HASH_KEY_SIZE];  /* hash key; changes during resize */

        unsigned n_entries;                /* number of stored entries */
        unsigned n_buckets;                /* number of buckets */

        unsigned idx_lowest_entry;         /* Index below which all buckets are free.
                                              Makes "while(hashmap_steal_first())" loops
                                              O(n) instead of O(n^2) for unordered hashmaps. */
        uint8_t  _pad[3];                  /* padding for the whole HashmapBase */
        /* The bitfields in HashmapBase complete the alignment of the whole thing. */
};

struct direct_storage {
        /* This gives us 39 bytes on 64bit, or 35 bytes on 32bit.
         * That's room for 4 set_entries + 4 DIB bytes + 3 unused bytes on 64bit,
         *              or 7 set_entries + 7 DIB bytes + 0 unused bytes on 32bit. */
        uint8_t storage[sizeof(struct indirect_storage)];
};

#define DIRECT_BUCKETS(entry_t) \
        (sizeof(struct direct_storage) / (sizeof(entry_t) + sizeof(dib_raw_t)))

/* We should be able to store at least one entry directly. */
assert_cc(DIRECT_BUCKETS(struct ordered_hashmap_entry) >= 1);

/* We have 3 bits for n_direct_entries. */
assert_cc(DIRECT_BUCKETS(struct set_entry) < (1 << 3));

/* Hashmaps with directly stored entries all use this shared hash key.
 * It's no big deal if the key is guessed, because there can be only
 * a handful of directly stored entries in a hashmap. When a hashmap
 * outgrows direct storage, it gets its own key for indirect storage. */
static uint8_t shared_hash_key[HASH_KEY_SIZE];
static bool shared_hash_key_initialized;

/* Fields that all hashmap/set types must have */
struct HashmapBase {
        const struct hash_ops *hash_ops;  /* hash and compare ops to use */

        union _packed_ {
                struct indirect_storage indirect; /* if  has_indirect */
                struct direct_storage direct;     /* if !has_indirect */
        };

        enum HashmapType type:2;     /* HASHMAP_TYPE_* */
        bool has_indirect:1;         /* whether indirect storage is used */
        unsigned n_direct_entries:3; /* Number of entries in direct storage.
                                      * Only valid if !has_indirect. */
        bool from_pool:1;            /* whether was allocated from mempool */
        bool dirty:1;                /* whether dirtied since cache sync */
        HASHMAP_DEBUG_FIELDS         /* optional hashmap_debug_info */
};

/* Specific hash types
 * HashmapBase must be at the beginning of each hashmap struct. */

struct Hashmap {
        struct HashmapBase b;
};

struct OrderedHashmap {
        struct HashmapBase b;
        unsigned iterate_list_head, iterate_list_tail;
};

struct Set {
        struct HashmapBase b;
};

DEFINE_MEMPOOL(hashmap_pool,         Hashmap,        8);
DEFINE_MEMPOOL(ordered_hashmap_pool, OrderedHashmap, 8);
/* No need for a separate Set pool */
assert_cc(sizeof(Hashmap) == sizeof(Set));

struct hashmap_type_info {
        size_t head_size;
        size_t entry_size;
        struct mempool *mempool;
        unsigned n_direct_buckets;
};

static const struct hashmap_type_info hashmap_type_info[_HASHMAP_TYPE_MAX] = {
        [HASHMAP_TYPE_PLAIN] = {
                .head_size        = sizeof(Hashmap),
                .entry_size       = sizeof(struct plain_hashmap_entry),
                .mempool          = &hashmap_pool,
                .n_direct_buckets = DIRECT_BUCKETS(struct plain_hashmap_entry),
        },
        [HASHMAP_TYPE_ORDERED] = {
                .head_size        = sizeof(OrderedHashmap),
                .entry_size       = sizeof(struct ordered_hashmap_entry),
                .mempool          = &ordered_hashmap_pool,
                .n_direct_buckets = DIRECT_BUCKETS(struct ordered_hashmap_entry),
        },
        [HASHMAP_TYPE_SET] = {
                .head_size        = sizeof(Set),
                .entry_size       = sizeof(struct set_entry),
                .mempool          = &hashmap_pool,
                .n_direct_buckets = DIRECT_BUCKETS(struct set_entry),
        },
};

#ifdef VALGRIND
__attribute__((destructor)) static void cleanup_pools(void) {
        _cleanup_free_ char *t = NULL;
        int r;

        /* Be nice to valgrind */

        /* The pool is only allocated by the main thread, but the memory can
         * be passed to other threads. Let's clean up if we are the main thread
         * and no other threads are live. */
        if (!is_main_thread())
                return;

        r = get_proc_field("/proc/self/status", "Threads", WHITESPACE, &t);
        if (r < 0 || !streq(t, "1"))
                return;

        mempool_drop(&hashmap_pool);
        mempool_drop(&ordered_hashmap_pool);
}
#endif

static unsigned n_buckets(HashmapBase *h) {
        return h->has_indirect ? h->indirect.n_buckets
                               : hashmap_type_info[h->type].n_direct_buckets;
}

static unsigned n_entries(HashmapBase *h) {
        return h->has_indirect ? h->indirect.n_entries
                               : h->n_direct_entries;
}

static void n_entries_inc(HashmapBase *h) {
        if (h->has_indirect)
                h->indirect.n_entries++;
        else
                h->n_direct_entries++;
}

static void n_entries_dec(HashmapBase *h) {
        if (h->has_indirect)
                h->indirect.n_entries--;
        else
                h->n_direct_entries--;
}

static void *storage_ptr(HashmapBase *h) {
        return h->has_indirect ? h->indirect.storage
                               : h->direct.storage;
}

static uint8_t *hash_key(HashmapBase *h) {
        return h->has_indirect ? h->indirect.hash_key
                               : shared_hash_key;
}

static unsigned base_bucket_hash(HashmapBase *h, const void *p) {
        struct siphash state;
        uint64_t hash;

        siphash24_init(&state, hash_key(h));

        h->hash_ops->hash(p, &state);

        hash = siphash24_finalize(&state);

        return (unsigned) (hash % n_buckets(h));
}
#define bucket_hash(h, p) base_bucket_hash(HASHMAP_BASE(h), p)

static inline void base_set_dirty(HashmapBase *h) {
        h->dirty = true;
}
#define hashmap_set_dirty(h) base_set_dirty(HASHMAP_BASE(h))

static void get_hash_key(uint8_t hash_key[HASH_KEY_SIZE], bool reuse_is_ok) {
        static uint8_t current[HASH_KEY_SIZE];
        static bool current_initialized = false;

        /* Returns a hash function key to use. In order to keep things
         * fast we will not generate a new key each time we allocate a
         * new hash table. Instead, we'll just reuse the most recently
         * generated one, except if we never generated one or when we
         * are rehashing an entire hash table because we reached a
         * fill level */

        if (!current_initialized || !reuse_is_ok) {
                random_bytes(current, sizeof(current));
                current_initialized = true;
        }

        memcpy(hash_key, current, sizeof(current));
}

static struct hashmap_base_entry *bucket_at(HashmapBase *h, unsigned idx) {
        return (struct hashmap_base_entry*)
                ((uint8_t*) storage_ptr(h) + idx * hashmap_type_info[h->type].entry_size);
}

static struct plain_hashmap_entry *plain_bucket_at(Hashmap *h, unsigned idx) {
        return (struct plain_hashmap_entry*) bucket_at(HASHMAP_BASE(h), idx);
}

static struct ordered_hashmap_entry *ordered_bucket_at(OrderedHashmap *h, unsigned idx) {
        return (struct ordered_hashmap_entry*) bucket_at(HASHMAP_BASE(h), idx);
}

static struct set_entry *set_bucket_at(Set *h, unsigned idx) {
        return (struct set_entry*) bucket_at(HASHMAP_BASE(h), idx);
}

static struct ordered_hashmap_entry *bucket_at_swap(struct swap_entries *swap, unsigned idx) {
        return &swap->e[idx - _IDX_SWAP_BEGIN];
}

/* Returns a pointer to the bucket at index idx.
 * Understands real indexes and swap indexes, hence "_virtual". */
static struct hashmap_base_entry *bucket_at_virtual(HashmapBase *h, struct swap_entries *swap,
                                                    unsigned idx) {
        if (idx < _IDX_SWAP_BEGIN)
                return bucket_at(h, idx);

        if (idx < _IDX_SWAP_END)
                return &bucket_at_swap(swap, idx)->p.b;

        assert_not_reached("Invalid index");
}

static dib_raw_t *dib_raw_ptr(HashmapBase *h) {
        return (dib_raw_t*)
                ((uint8_t*) storage_ptr(h) + hashmap_type_info[h->type].entry_size * n_buckets(h));
}

static unsigned bucket_distance(HashmapBase *h, unsigned idx, unsigned from) {
        return idx >= from ? idx - from
                           : n_buckets(h) + idx - from;
}

static unsigned bucket_calculate_dib(HashmapBase *h, unsigned idx, dib_raw_t raw_dib) {
        unsigned initial_bucket;

        if (raw_dib == DIB_RAW_FREE)
                return DIB_FREE;

        if (_likely_(raw_dib < DIB_RAW_OVERFLOW))
                return raw_dib;

        /*
         * Having an overflow DIB value is very unlikely. The hash function
         * would have to be bad. For example, in a table of size 2^24 filled
         * to load factor 0.9 the maximum observed DIB is only about 60.
         * In theory (assuming I used Maxima correctly), for an infinite size
         * hash table with load factor 0.8 the probability of a given entry
         * having DIB > 40 is 1.9e-8.
         * This returns the correct DIB value by recomputing the hash value in
         * the unlikely case. XXX Hitting this case could be a hint to rehash.
         */
        initial_bucket = bucket_hash(h, bucket_at(h, idx)->key);
        return bucket_distance(h, idx, initial_bucket);
}

static void bucket_set_dib(HashmapBase *h, unsigned idx, unsigned dib) {
        dib_raw_ptr(h)[idx] = dib != DIB_FREE ? MIN(dib, DIB_RAW_OVERFLOW) : DIB_RAW_FREE;
}

static unsigned skip_free_buckets(HashmapBase *h, unsigned idx) {
        dib_raw_t *dibs;

        dibs = dib_raw_ptr(h);

        for ( ; idx < n_buckets(h); idx++)
                if (dibs[idx] != DIB_RAW_FREE)
                        return idx;

        return IDX_NIL;
}

static void bucket_mark_free(HashmapBase *h, unsigned idx) {
        memzero(bucket_at(h, idx), hashmap_type_info[h->type].entry_size);
        bucket_set_dib(h, idx, DIB_FREE);
}

static void bucket_move_entry(HashmapBase *h, struct swap_entries *swap,
                              unsigned from, unsigned to) {
        struct hashmap_base_entry *e_from, *e_to;

        assert(from != to);

        e_from = bucket_at_virtual(h, swap, from);
        e_to   = bucket_at_virtual(h, swap, to);

        memcpy(e_to, e_from, hashmap_type_info[h->type].entry_size);

        if (h->type == HASHMAP_TYPE_ORDERED) {
                OrderedHashmap *lh = (OrderedHashmap*) h;
                struct ordered_hashmap_entry *le, *le_to;

                le_to = (struct ordered_hashmap_entry*) e_to;

                if (le_to->iterate_next != IDX_NIL) {
                        le = (struct ordered_hashmap_entry*)
                             bucket_at_virtual(h, swap, le_to->iterate_next);
                        le->iterate_previous = to;
                }

                if (le_to->iterate_previous != IDX_NIL) {
                        le = (struct ordered_hashmap_entry*)
                             bucket_at_virtual(h, swap, le_to->iterate_previous);
                        le->iterate_next = to;
                }

                if (lh->iterate_list_head == from)
                        lh->iterate_list_head = to;
                if (lh->iterate_list_tail == from)
                        lh->iterate_list_tail = to;
        }
}

static unsigned next_idx(HashmapBase *h, unsigned idx) {
        return (idx + 1U) % n_buckets(h);
}

static unsigned prev_idx(HashmapBase *h, unsigned idx) {
        return (n_buckets(h) + idx - 1U) % n_buckets(h);
}

static void *entry_value(HashmapBase *h, struct hashmap_base_entry *e) {
        switch (h->type) {

        case HASHMAP_TYPE_PLAIN:
        case HASHMAP_TYPE_ORDERED:
                return ((struct plain_hashmap_entry*)e)->value;

        case HASHMAP_TYPE_SET:
                return (void*) e->key;

        default:
                assert_not_reached("Unknown hashmap type");
        }
}

static void base_remove_entry(HashmapBase *h, unsigned idx) {
        unsigned left, right, prev, dib;
        dib_raw_t raw_dib, *dibs;

        dibs = dib_raw_ptr(h);
        assert(dibs[idx] != DIB_RAW_FREE);

#if ENABLE_DEBUG_HASHMAP
        h->debug.rem_count++;
        h->debug.last_rem_idx = idx;
#endif

        left = idx;
        /* Find the stop bucket ("right"). It is either free or has DIB == 0. */
        for (right = next_idx(h, left); ; right = next_idx(h, right)) {
                raw_dib = dibs[right];
                if (IN_SET(raw_dib, 0, DIB_RAW_FREE))
                        break;

                /* The buckets are not supposed to be all occupied and with DIB > 0.
                 * That would mean we could make everyone better off by shifting them
                 * backward. This scenario is impossible. */
                assert(left != right);
        }

        if (h->type == HASHMAP_TYPE_ORDERED) {
                OrderedHashmap *lh = (OrderedHashmap*) h;
                struct ordered_hashmap_entry *le = ordered_bucket_at(lh, idx);

                if (le->iterate_next != IDX_NIL)
                        ordered_bucket_at(lh, le->iterate_next)->iterate_previous = le->iterate_previous;
                else
                        lh->iterate_list_tail = le->iterate_previous;

                if (le->iterate_previous != IDX_NIL)
                        ordered_bucket_at(lh, le->iterate_previous)->iterate_next = le->iterate_next;
                else
                        lh->iterate_list_head = le->iterate_next;
        }

        /* Now shift all buckets in the interval (left, right) one step backwards */
        for (prev = left, left = next_idx(h, left); left != right;
             prev = left, left = next_idx(h, left)) {
                dib = bucket_calculate_dib(h, left, dibs[left]);
                assert(dib != 0);
                bucket_move_entry(h, NULL, left, prev);
                bucket_set_dib(h, prev, dib - 1);
        }

        bucket_mark_free(h, prev);
        n_entries_dec(h);
        base_set_dirty(h);
}
#define remove_entry(h, idx) base_remove_entry(HASHMAP_BASE(h), idx)

static unsigned hashmap_iterate_in_insertion_order(OrderedHashmap *h, Iterator *i) {
        struct ordered_hashmap_entry *e;
        unsigned idx;

        assert(h);
        assert(i);

        if (i->idx == IDX_NIL)
                goto at_end;

        if (i->idx == IDX_FIRST && h->iterate_list_head == IDX_NIL)
                goto at_end;

        if (i->idx == IDX_FIRST) {
                idx = h->iterate_list_head;
                e = ordered_bucket_at(h, idx);
        } else {
                idx = i->idx;
                e = ordered_bucket_at(h, idx);
                /*
                 * We allow removing the current entry while iterating, but removal may cause
                 * a backward shift. The next entry may thus move one bucket to the left.
                 * To detect when it happens, we remember the key pointer of the entry we were
                 * going to iterate next. If it does not match, there was a backward shift.
                 */
                if (e->p.b.key != i->next_key) {
                        idx = prev_idx(HASHMAP_BASE(h), idx);
                        e = ordered_bucket_at(h, idx);
                }
                assert(e->p.b.key == i->next_key);
        }

#if ENABLE_DEBUG_HASHMAP
        i->prev_idx = idx;
#endif

        if (e->iterate_next != IDX_NIL) {
                struct ordered_hashmap_entry *n;
                i->idx = e->iterate_next;
                n = ordered_bucket_at(h, i->idx);
                i->next_key = n->p.b.key;
        } else
                i->idx = IDX_NIL;

        return idx;

at_end:
        i->idx = IDX_NIL;
        return IDX_NIL;
}

static unsigned hashmap_iterate_in_internal_order(HashmapBase *h, Iterator *i) {
        unsigned idx;

        assert(h);
        assert(i);

        if (i->idx == IDX_NIL)
                goto at_end;

        if (i->idx == IDX_FIRST) {
                /* fast forward to the first occupied bucket */
                if (h->has_indirect) {
                        i->idx = skip_free_buckets(h, h->indirect.idx_lowest_entry);
                        h->indirect.idx_lowest_entry = i->idx;
                } else
                        i->idx = skip_free_buckets(h, 0);

                if (i->idx == IDX_NIL)
                        goto at_end;
        } else {
                struct hashmap_base_entry *e;

                assert(i->idx > 0);

                e = bucket_at(h, i->idx);
                /*
                 * We allow removing the current entry while iterating, but removal may cause
                 * a backward shift. The next entry may thus move one bucket to the left.
                 * To detect when it happens, we remember the key pointer of the entry we were
                 * going to iterate next. If it does not match, there was a backward shift.
                 */
                if (e->key != i->next_key)
                        e = bucket_at(h, --i->idx);

                assert(e->key == i->next_key);
        }

        idx = i->idx;
#if ENABLE_DEBUG_HASHMAP
        i->prev_idx = idx;
#endif

        i->idx = skip_free_buckets(h, i->idx + 1);
        if (i->idx != IDX_NIL)
                i->next_key = bucket_at(h, i->idx)->key;
        else
                i->idx = IDX_NIL;

        return idx;

at_end:
        i->idx = IDX_NIL;
        return IDX_NIL;
}

static unsigned hashmap_iterate_entry(HashmapBase *h, Iterator *i) {
        if (!h) {
                i->idx = IDX_NIL;
                return IDX_NIL;
        }

#if ENABLE_DEBUG_HASHMAP
        if (i->idx == IDX_FIRST) {
                i->put_count = h->debug.put_count;
                i->rem_count = h->debug.rem_count;
        } else {
                /* While iterating, must not add any new entries */
                assert(i->put_count == h->debug.put_count);
                /* ... or remove entries other than the current one */
                assert(i->rem_count == h->debug.rem_count ||
                       (i->rem_count == h->debug.rem_count - 1 &&
                        i->prev_idx == h->debug.last_rem_idx));
                /* Reset our removals counter */
                i->rem_count = h->debug.rem_count;
        }
#endif

        return h->type == HASHMAP_TYPE_ORDERED ? hashmap_iterate_in_insertion_order((OrderedHashmap*) h, i)
                                               : hashmap_iterate_in_internal_order(h, i);
}

bool internal_hashmap_iterate(HashmapBase *h, Iterator *i, void **value, const void **key) {
        struct hashmap_base_entry *e;
        void *data;
        unsigned idx;

        idx = hashmap_iterate_entry(h, i);
        if (idx == IDX_NIL) {
                if (value)
                        *value = NULL;
                if (key)
                        *key = NULL;

                return false;
        }

        e = bucket_at(h, idx);
        data = entry_value(h, e);
        if (value)
                *value = data;
        if (key)
                *key = e->key;

        return true;
}

bool set_iterate(Set *s, Iterator *i, void **value) {
        return internal_hashmap_iterate(HASHMAP_BASE(s), i, value, NULL);
}

#define HASHMAP_FOREACH_IDX(idx, h, i) \
        for ((i) = ITERATOR_FIRST, (idx) = hashmap_iterate_entry((h), &(i)); \
             (idx != IDX_NIL); \
             (idx) = hashmap_iterate_entry((h), &(i)))

static void reset_direct_storage(HashmapBase *h) {
        const struct hashmap_type_info *hi = &hashmap_type_info[h->type];
        void *p;

        assert(!h->has_indirect);

        p = mempset(h->direct.storage, 0, hi->entry_size * hi->n_direct_buckets);
        memset(p, DIB_RAW_INIT, sizeof(dib_raw_t) * hi->n_direct_buckets);
}

static struct HashmapBase *hashmap_base_new(const struct hash_ops *hash_ops, enum HashmapType type HASHMAP_DEBUG_PARAMS) {
        HashmapBase *h;
        const struct hashmap_type_info *hi = &hashmap_type_info[type];
        bool use_pool;

        use_pool = is_main_thread();

        h = use_pool ? mempool_alloc0_tile(hi->mempool) : malloc0(hi->head_size);

        if (!h)
                return NULL;

        h->type = type;
        h->from_pool = use_pool;
        h->hash_ops = hash_ops ? hash_ops : &trivial_hash_ops;

        if (type == HASHMAP_TYPE_ORDERED) {
                OrderedHashmap *lh = (OrderedHashmap*)h;
                lh->iterate_list_head = lh->iterate_list_tail = IDX_NIL;
        }

        reset_direct_storage(h);

        if (!shared_hash_key_initialized) {
                random_bytes(shared_hash_key, sizeof(shared_hash_key));
                shared_hash_key_initialized= true;
        }

#if ENABLE_DEBUG_HASHMAP
        h->debug.func = func;
        h->debug.file = file;
        h->debug.line = line;
        assert_se(pthread_mutex_lock(&hashmap_debug_list_mutex) == 0);
        LIST_PREPEND(debug_list, hashmap_debug_list, &h->debug);
        assert_se(pthread_mutex_unlock(&hashmap_debug_list_mutex) == 0);
#endif

        return h;
}

Hashmap *internal_hashmap_new(const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
        return (Hashmap*)        hashmap_base_new(hash_ops, HASHMAP_TYPE_PLAIN HASHMAP_DEBUG_PASS_ARGS);
}

OrderedHashmap *internal_ordered_hashmap_new(const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
        return (OrderedHashmap*) hashmap_base_new(hash_ops, HASHMAP_TYPE_ORDERED HASHMAP_DEBUG_PASS_ARGS);
}

Set *internal_set_new(const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
        return (Set*)            hashmap_base_new(hash_ops, HASHMAP_TYPE_SET HASHMAP_DEBUG_PASS_ARGS);
}

static int hashmap_base_ensure_allocated(HashmapBase **h, const struct hash_ops *hash_ops,
                                         enum HashmapType type HASHMAP_DEBUG_PARAMS) {
        HashmapBase *q;

        assert(h);

        if (*h)
                return 0;

        q = hashmap_base_new(hash_ops, type HASHMAP_DEBUG_PASS_ARGS);
        if (!q)
                return -ENOMEM;

        *h = q;
        return 0;
}

int internal_hashmap_ensure_allocated(Hashmap **h, const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
        return hashmap_base_ensure_allocated((HashmapBase**)h, hash_ops, HASHMAP_TYPE_PLAIN HASHMAP_DEBUG_PASS_ARGS);
}

int internal_ordered_hashmap_ensure_allocated(OrderedHashmap **h, const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
        return hashmap_base_ensure_allocated((HashmapBase**)h, hash_ops, HASHMAP_TYPE_ORDERED HASHMAP_DEBUG_PASS_ARGS);
}

int internal_set_ensure_allocated(Set **s, const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
        return hashmap_base_ensure_allocated((HashmapBase**)s, hash_ops, HASHMAP_TYPE_SET HASHMAP_DEBUG_PASS_ARGS);
}

static void hashmap_free_no_clear(HashmapBase *h) {
        assert(!h->has_indirect);
        assert(!h->n_direct_entries);

#if ENABLE_DEBUG_HASHMAP
        assert_se(pthread_mutex_lock(&hashmap_debug_list_mutex) == 0);
        LIST_REMOVE(debug_list, hashmap_debug_list, &h->debug);
        assert_se(pthread_mutex_unlock(&hashmap_debug_list_mutex) == 0);
#endif

        if (h->from_pool)
                mempool_free_tile(hashmap_type_info[h->type].mempool, h);
        else
                free(h);
}

HashmapBase *internal_hashmap_free(HashmapBase *h) {

        /* Free the hashmap, but nothing in it */

        if (h) {
                internal_hashmap_clear(h);
                hashmap_free_no_clear(h);
        }

        return NULL;
}

HashmapBase *internal_hashmap_free_free(HashmapBase *h) {

        /* Free the hashmap and all data objects in it, but not the
         * keys */

        if (h) {
                internal_hashmap_clear_free(h);
                hashmap_free_no_clear(h);
        }

        return NULL;
}

Hashmap *hashmap_free_free_free(Hashmap *h) {

        /* Free the hashmap and all data and key objects in it */

        if (h) {
                hashmap_clear_free_free(h);
                hashmap_free_no_clear(HASHMAP_BASE(h));
        }

        return NULL;
}

void internal_hashmap_clear(HashmapBase *h) {
        if (!h)
                return;

        if (h->has_indirect) {
                free(h->indirect.storage);
                h->has_indirect = false;
        }

        h->n_direct_entries = 0;
        reset_direct_storage(h);

        if (h->type == HASHMAP_TYPE_ORDERED) {
                OrderedHashmap *lh = (OrderedHashmap*) h;
                lh->iterate_list_head = lh->iterate_list_tail = IDX_NIL;
        }

        base_set_dirty(h);
}

void internal_hashmap_clear_free(HashmapBase *h) {
        unsigned idx;

        if (!h)
                return;

        for (idx = skip_free_buckets(h, 0); idx != IDX_NIL;
             idx = skip_free_buckets(h, idx + 1))
                free(entry_value(h, bucket_at(h, idx)));

        internal_hashmap_clear(h);
}

void hashmap_clear_free_free(Hashmap *h) {
        unsigned idx;

        if (!h)
                return;

        for (idx = skip_free_buckets(HASHMAP_BASE(h), 0); idx != IDX_NIL;
             idx = skip_free_buckets(HASHMAP_BASE(h), idx + 1)) {
                struct plain_hashmap_entry *e = plain_bucket_at(h, idx);
                free((void*)e->b.key);
                free(e->value);
        }

        internal_hashmap_clear(HASHMAP_BASE(h));
}

static int resize_buckets(HashmapBase *h, unsigned entries_add);

/*
 * Finds an empty bucket to put an entry into, starting the scan at 'idx'.
 * Performs Robin Hood swaps as it goes. The entry to put must be placed
 * by the caller into swap slot IDX_PUT.
 * If used for in-place resizing, may leave a displaced entry in swap slot
 * IDX_PUT. Caller must rehash it next.
 * Returns: true if it left a displaced entry to rehash next in IDX_PUT,
 *          false otherwise.
 */
static bool hashmap_put_robin_hood(HashmapBase *h, unsigned idx,
                                   struct swap_entries *swap) {
        dib_raw_t raw_dib, *dibs;
        unsigned dib, distance;

#if ENABLE_DEBUG_HASHMAP
        h->debug.put_count++;
#endif

        dibs = dib_raw_ptr(h);

        for (distance = 0; ; distance++) {
                raw_dib = dibs[idx];
                if (IN_SET(raw_dib, DIB_RAW_FREE, DIB_RAW_REHASH)) {
                        if (raw_dib == DIB_RAW_REHASH)
                                bucket_move_entry(h, swap, idx, IDX_TMP);

                        if (h->has_indirect && h->indirect.idx_lowest_entry > idx)
                                h->indirect.idx_lowest_entry = idx;

                        bucket_set_dib(h, idx, distance);
                        bucket_move_entry(h, swap, IDX_PUT, idx);
                        if (raw_dib == DIB_RAW_REHASH) {
                                bucket_move_entry(h, swap, IDX_TMP, IDX_PUT);
                                return true;
                        }

                        return false;
                }

                dib = bucket_calculate_dib(h, idx, raw_dib);

                if (dib < distance) {
                        /* Found a wealthier entry. Go Robin Hood! */
                        bucket_set_dib(h, idx, distance);

                        /* swap the entries */
                        bucket_move_entry(h, swap, idx, IDX_TMP);
                        bucket_move_entry(h, swap, IDX_PUT, idx);
                        bucket_move_entry(h, swap, IDX_TMP, IDX_PUT);

                        distance = dib;
                }

                idx = next_idx(h, idx);
        }
}

/*
 * Puts an entry into a hashmap, boldly - no check whether key already exists.
 * The caller must place the entry (only its key and value, not link indexes)
 * in swap slot IDX_PUT.
 * Caller must ensure: the key does not exist yet in the hashmap.
 *                     that resize is not needed if !may_resize.
 * Returns: 1 if entry was put successfully.
 *          -ENOMEM if may_resize==true and resize failed with -ENOMEM.
 *          Cannot return -ENOMEM if !may_resize.
 */
static int hashmap_base_put_boldly(HashmapBase *h, unsigned idx,
                                   struct swap_entries *swap, bool may_resize) {
        struct ordered_hashmap_entry *new_entry;
        int r;

        assert(idx < n_buckets(h));

        new_entry = bucket_at_swap(swap, IDX_PUT);

        if (may_resize) {
                r = resize_buckets(h, 1);
                if (r < 0)
                        return r;
                if (r > 0)
                        idx = bucket_hash(h, new_entry->p.b.key);
        }
        assert(n_entries(h) < n_buckets(h));

        if (h->type == HASHMAP_TYPE_ORDERED) {
                OrderedHashmap *lh = (OrderedHashmap*) h;

                new_entry->iterate_next = IDX_NIL;
                new_entry->iterate_previous = lh->iterate_list_tail;

                if (lh->iterate_list_tail != IDX_NIL) {
                        struct ordered_hashmap_entry *old_tail;

                        old_tail = ordered_bucket_at(lh, lh->iterate_list_tail);
                        assert(old_tail->iterate_next == IDX_NIL);
                        old_tail->iterate_next = IDX_PUT;
                }

                lh->iterate_list_tail = IDX_PUT;
                if (lh->iterate_list_head == IDX_NIL)
                        lh->iterate_list_head = IDX_PUT;
        }

        assert_se(hashmap_put_robin_hood(h, idx, swap) == false);

        n_entries_inc(h);
#if ENABLE_DEBUG_HASHMAP
        h->debug.max_entries = MAX(h->debug.max_entries, n_entries(h));
#endif

        base_set_dirty(h);

        return 1;
}
#define hashmap_put_boldly(h, idx, swap, may_resize) \
        hashmap_base_put_boldly(HASHMAP_BASE(h), idx, swap, may_resize)

/*
 * Returns 0 if resize is not needed.
 *         1 if successfully resized.
 *         -ENOMEM on allocation failure.
 */
static int resize_buckets(HashmapBase *h, unsigned entries_add) {
        struct swap_entries swap;
        void *new_storage;
        dib_raw_t *old_dibs, *new_dibs;
        const struct hashmap_type_info *hi;
        unsigned idx, optimal_idx;
        unsigned old_n_buckets, new_n_buckets, n_rehashed, new_n_entries;
        uint8_t new_shift;
        bool rehash_next;

        assert(h);

        hi = &hashmap_type_info[h->type];
        new_n_entries = n_entries(h) + entries_add;

        /* overflow? */
        if (_unlikely_(new_n_entries < entries_add))
                return -ENOMEM;

        /* For direct storage we allow 100% load, because it's tiny. */
        if (!h->has_indirect && new_n_entries <= hi->n_direct_buckets)
                return 0;

        /*
         * Load factor = n/m = 1 - (1/INV_KEEP_FREE).
         * From it follows: m = n + n/(INV_KEEP_FREE - 1)
         */
        new_n_buckets = new_n_entries + new_n_entries / (INV_KEEP_FREE - 1);
        /* overflow? */
        if (_unlikely_(new_n_buckets < new_n_entries))
                return -ENOMEM;

        if (_unlikely_(new_n_buckets > UINT_MAX / (hi->entry_size + sizeof(dib_raw_t))))
                return -ENOMEM;

        old_n_buckets = n_buckets(h);

        if (_likely_(new_n_buckets <= old_n_buckets))
                return 0;

        new_shift = log2u_round_up(MAX(
                        new_n_buckets * (hi->entry_size + sizeof(dib_raw_t)),
                        2 * sizeof(struct direct_storage)));

        /* Realloc storage (buckets and DIB array). */
        new_storage = realloc(h->has_indirect ? h->indirect.storage : NULL,
                              1U << new_shift);
        if (!new_storage)
                return -ENOMEM;

        /* Must upgrade direct to indirect storage. */
        if (!h->has_indirect) {
                memcpy(new_storage, h->direct.storage,
                       old_n_buckets * (hi->entry_size + sizeof(dib_raw_t)));
                h->indirect.n_entries = h->n_direct_entries;
                h->indirect.idx_lowest_entry = 0;
                h->n_direct_entries = 0;
        }

        /* Get a new hash key. If we've just upgraded to indirect storage,
         * allow reusing a previously generated key. It's still a different key
         * from the shared one that we used for direct storage. */
        get_hash_key(h->indirect.hash_key, !h->has_indirect);

        h->has_indirect = true;
        h->indirect.storage = new_storage;
        h->indirect.n_buckets = (1U << new_shift) /
                                (hi->entry_size + sizeof(dib_raw_t));

        old_dibs = (dib_raw_t*)((uint8_t*) new_storage + hi->entry_size * old_n_buckets);
        new_dibs = dib_raw_ptr(h);

        /*
         * Move the DIB array to the new place, replacing valid DIB values with
         * DIB_RAW_REHASH to indicate all of the used buckets need rehashing.
         * Note: Overlap is not possible, because we have at least doubled the
         * number of buckets and dib_raw_t is smaller than any entry type.
         */
        for (idx = 0; idx < old_n_buckets; idx++) {
                assert(old_dibs[idx] != DIB_RAW_REHASH);
                new_dibs[idx] = old_dibs[idx] == DIB_RAW_FREE ? DIB_RAW_FREE
                                                              : DIB_RAW_REHASH;
        }

        /* Zero the area of newly added entries (including the old DIB area) */
        memzero(bucket_at(h, old_n_buckets),
               (n_buckets(h) - old_n_buckets) * hi->entry_size);

        /* The upper half of the new DIB array needs initialization */
        memset(&new_dibs[old_n_buckets], DIB_RAW_INIT,
               (n_buckets(h) - old_n_buckets) * sizeof(dib_raw_t));

        /* Rehash entries that need it */
        n_rehashed = 0;
        for (idx = 0; idx < old_n_buckets; idx++) {
                if (new_dibs[idx] != DIB_RAW_REHASH)
                        continue;

                optimal_idx = bucket_hash(h, bucket_at(h, idx)->key);

                /*
                 * Not much to do if by luck the entry hashes to its current
                 * location. Just set its DIB.
                 */
                if (optimal_idx == idx) {
                        new_dibs[idx] = 0;
                        n_rehashed++;
                        continue;
                }

                new_dibs[idx] = DIB_RAW_FREE;
                bucket_move_entry(h, &swap, idx, IDX_PUT);
                /* bucket_move_entry does not clear the source */
                memzero(bucket_at(h, idx), hi->entry_size);

                do {
                        /*
                         * Find the new bucket for the current entry. This may make
                         * another entry homeless and load it into IDX_PUT.
                         */
                        rehash_next = hashmap_put_robin_hood(h, optimal_idx, &swap);
                        n_rehashed++;

                        /* Did the current entry displace another one? */
                        if (rehash_next)
                                optimal_idx = bucket_hash(h, bucket_at_swap(&swap, IDX_PUT)->p.b.key);
                } while (rehash_next);
        }

        assert(n_rehashed == n_entries(h));

        return 1;
}

/*
 * Finds an entry with a matching key
 * Returns: index of the found entry, or IDX_NIL if not found.
 */
static unsigned base_bucket_scan(HashmapBase *h, unsigned idx, const void *key) {
        struct hashmap_base_entry *e;
        unsigned dib, distance;
        dib_raw_t *dibs = dib_raw_ptr(h);

        assert(idx < n_buckets(h));

        for (distance = 0; ; distance++) {
                if (dibs[idx] == DIB_RAW_FREE)
                        return IDX_NIL;

                dib = bucket_calculate_dib(h, idx, dibs[idx]);

                if (dib < distance)
                        return IDX_NIL;
                if (dib == distance) {
                        e = bucket_at(h, idx);
                        if (h->hash_ops->compare(e->key, key) == 0)
                                return idx;
                }

                idx = next_idx(h, idx);
        }
}
#define bucket_scan(h, idx, key) base_bucket_scan(HASHMAP_BASE(h), idx, key)

int hashmap_put(Hashmap *h, const void *key, void *value) {
        struct swap_entries swap;
        struct plain_hashmap_entry *e;
        unsigned hash, idx;

        assert(h);

        hash = bucket_hash(h, key);
        idx = bucket_scan(h, hash, key);
        if (idx != IDX_NIL) {
                e = plain_bucket_at(h, idx);
                if (e->value == value)
                        return 0;
                return -EEXIST;
        }

        e = &bucket_at_swap(&swap, IDX_PUT)->p;
        e->b.key = key;
        e->value = value;
        return hashmap_put_boldly(h, hash, &swap, true);
}

int set_put(Set *s, const void *key) {
        struct swap_entries swap;
        struct hashmap_base_entry *e;
        unsigned hash, idx;

        assert(s);

        hash = bucket_hash(s, key);
        idx = bucket_scan(s, hash, key);
        if (idx != IDX_NIL)
                return 0;

        e = &bucket_at_swap(&swap, IDX_PUT)->p.b;
        e->key = key;
        return hashmap_put_boldly(s, hash, &swap, true);
}

int hashmap_replace(Hashmap *h, const void *key, void *value) {
        struct swap_entries swap;
        struct plain_hashmap_entry *e;
        unsigned hash, idx;

        assert(h);

        hash = bucket_hash(h, key);
        idx = bucket_scan(h, hash, key);
        if (idx != IDX_NIL) {
                e = plain_bucket_at(h, idx);
#if ENABLE_DEBUG_HASHMAP
                /* Although the key is equal, the key pointer may have changed,
                 * and this would break our assumption for iterating. So count
                 * this operation as incompatible with iteration. */
                if (e->b.key != key) {
                        h->b.debug.put_count++;
                        h->b.debug.rem_count++;
                        h->b.debug.last_rem_idx = idx;
                }
#endif
                e->b.key = key;
                e->value = value;
                hashmap_set_dirty(h);

                return 0;
        }

        e = &bucket_at_swap(&swap, IDX_PUT)->p;
        e->b.key = key;
        e->value = value;
        return hashmap_put_boldly(h, hash, &swap, true);
}

int hashmap_update(Hashmap *h, const void *key, void *value) {
        struct plain_hashmap_entry *e;
        unsigned hash, idx;

        assert(h);

        hash = bucket_hash(h, key);
        idx = bucket_scan(h, hash, key);
        if (idx == IDX_NIL)
                return -ENOENT;

        e = plain_bucket_at(h, idx);
        e->value = value;
        hashmap_set_dirty(h);

        return 0;
}

void *internal_hashmap_get(HashmapBase *h, const void *key) {
        struct hashmap_base_entry *e;
        unsigned hash, idx;

        if (!h)
                return NULL;

        hash = bucket_hash(h, key);
        idx = bucket_scan(h, hash, key);
        if (idx == IDX_NIL)
                return NULL;

        e = bucket_at(h, idx);
        return entry_value(h, e);
}

void *hashmap_get2(Hashmap *h, const void *key, void **key2) {
        struct plain_hashmap_entry *e;
        unsigned hash, idx;

        if (!h)
                return NULL;

        hash = bucket_hash(h, key);
        idx = bucket_scan(h, hash, key);
        if (idx == IDX_NIL)
                return NULL;

        e = plain_bucket_at(h, idx);
        if (key2)
                *key2 = (void*) e->b.key;

        return e->value;
}

bool internal_hashmap_contains(HashmapBase *h, const void *key) {
        unsigned hash;

        if (!h)
                return false;

        hash = bucket_hash(h, key);
        return bucket_scan(h, hash, key) != IDX_NIL;
}

void *internal_hashmap_remove(HashmapBase *h, const void *key) {
        struct hashmap_base_entry *e;
        unsigned hash, idx;
        void *data;

        if (!h)
                return NULL;

        hash = bucket_hash(h, key);
        idx = bucket_scan(h, hash, key);
        if (idx == IDX_NIL)
                return NULL;

        e = bucket_at(h, idx);
        data = entry_value(h, e);
        remove_entry(h, idx);

        return data;
}

void *hashmap_remove2(Hashmap *h, const void *key, void **rkey) {
        struct plain_hashmap_entry *e;
        unsigned hash, idx;
        void *data;

        if (!h) {
                if (rkey)
                        *rkey = NULL;
                return NULL;
        }

        hash = bucket_hash(h, key);
        idx = bucket_scan(h, hash, key);
        if (idx == IDX_NIL) {
                if (rkey)
                        *rkey = NULL;
                return NULL;
        }

        e = plain_bucket_at(h, idx);
        data = e->value;
        if (rkey)
                *rkey = (void*) e->b.key;

        remove_entry(h, idx);

        return data;
}

int hashmap_remove_and_put(Hashmap *h, const void *old_key, const void *new_key, void *value) {
        struct swap_entries swap;
        struct plain_hashmap_entry *e;
        unsigned old_hash, new_hash, idx;

        if (!h)
                return -ENOENT;

        old_hash = bucket_hash(h, old_key);
        idx = bucket_scan(h, old_hash, old_key);
        if (idx == IDX_NIL)
                return -ENOENT;

        new_hash = bucket_hash(h, new_key);
        if (bucket_scan(h, new_hash, new_key) != IDX_NIL)
                return -EEXIST;

        remove_entry(h, idx);

        e = &bucket_at_swap(&swap, IDX_PUT)->p;
        e->b.key = new_key;
        e->value = value;
        assert_se(hashmap_put_boldly(h, new_hash, &swap, false) == 1);

        return 0;
}

#if 0 /// UNNEEDED by elogind
int set_remove_and_put(Set *s, const void *old_key, const void *new_key) {
        struct swap_entries swap;
        struct hashmap_base_entry *e;
        unsigned old_hash, new_hash, idx;

        if (!s)
                return -ENOENT;

        old_hash = bucket_hash(s, old_key);
        idx = bucket_scan(s, old_hash, old_key);
        if (idx == IDX_NIL)
                return -ENOENT;

        new_hash = bucket_hash(s, new_key);
        if (bucket_scan(s, new_hash, new_key) != IDX_NIL)
                return -EEXIST;

        remove_entry(s, idx);

        e = &bucket_at_swap(&swap, IDX_PUT)->p.b;
        e->key = new_key;
        assert_se(hashmap_put_boldly(s, new_hash, &swap, false) == 1);

        return 0;
}
#endif // 0

int hashmap_remove_and_replace(Hashmap *h, const void *old_key, const void *new_key, void *value) {
        struct swap_entries swap;
        struct plain_hashmap_entry *e;
        unsigned old_hash, new_hash, idx_old, idx_new;

        if (!h)
                return -ENOENT;

        old_hash = bucket_hash(h, old_key);
        idx_old = bucket_scan(h, old_hash, old_key);
        if (idx_old == IDX_NIL)
                return -ENOENT;

        old_key = bucket_at(HASHMAP_BASE(h), idx_old)->key;

        new_hash = bucket_hash(h, new_key);
        idx_new = bucket_scan(h, new_hash, new_key);
        if (idx_new != IDX_NIL)
                if (idx_old != idx_new) {
                        remove_entry(h, idx_new);
                        /* Compensate for a possible backward shift. */
                        if (old_key != bucket_at(HASHMAP_BASE(h), idx_old)->key)
                                idx_old = prev_idx(HASHMAP_BASE(h), idx_old);
                        assert(old_key == bucket_at(HASHMAP_BASE(h), idx_old)->key);
                }

        remove_entry(h, idx_old);

        e = &bucket_at_swap(&swap, IDX_PUT)->p;
        e->b.key = new_key;
        e->value = value;
        assert_se(hashmap_put_boldly(h, new_hash, &swap, false) == 1);

        return 0;
}

void *hashmap_remove_value(Hashmap *h, const void *key, void *value) {
        struct plain_hashmap_entry *e;
        unsigned hash, idx;

        if (!h)
                return NULL;

        hash = bucket_hash(h, key);
        idx = bucket_scan(h, hash, key);
        if (idx == IDX_NIL)
                return NULL;

        e = plain_bucket_at(h, idx);
        if (e->value != value)
                return NULL;

        remove_entry(h, idx);

        return value;
}

static unsigned find_first_entry(HashmapBase *h) {
        Iterator i = ITERATOR_FIRST;

        if (!h || !n_entries(h))
                return IDX_NIL;

        return hashmap_iterate_entry(h, &i);
}

void *internal_hashmap_first(HashmapBase *h) {
        unsigned idx;

        idx = find_first_entry(h);
        if (idx == IDX_NIL)
                return NULL;

        return entry_value(h, bucket_at(h, idx));
}

void *internal_hashmap_first_key(HashmapBase *h) {
        struct hashmap_base_entry *e;
        unsigned idx;

        idx = find_first_entry(h);
        if (idx == IDX_NIL)
                return NULL;

        e = bucket_at(h, idx);
        return (void*) e->key;
}

void *internal_hashmap_steal_first(HashmapBase *h) {
        struct hashmap_base_entry *e;
        void *data;
        unsigned idx;

        idx = find_first_entry(h);
        if (idx == IDX_NIL)
                return NULL;

        e = bucket_at(h, idx);
        data = entry_value(h, e);
        remove_entry(h, idx);

        return data;
}

void *internal_hashmap_steal_first_key(HashmapBase *h) {
        struct hashmap_base_entry *e;
        void *key;
        unsigned idx;

        idx = find_first_entry(h);
        if (idx == IDX_NIL)
                return NULL;

        e = bucket_at(h, idx);
        key = (void*) e->key;
        remove_entry(h, idx);

        return key;
}

unsigned internal_hashmap_size(HashmapBase *h) {

        if (!h)
                return 0;

        return n_entries(h);
}

unsigned internal_hashmap_buckets(HashmapBase *h) {

        if (!h)
                return 0;

        return n_buckets(h);
}

int internal_hashmap_merge(Hashmap *h, Hashmap *other) {
        Iterator i;
        unsigned idx;

        assert(h);

        HASHMAP_FOREACH_IDX(idx, HASHMAP_BASE(other), i) {
                struct plain_hashmap_entry *pe = plain_bucket_at(other, idx);
                int r;

                r = hashmap_put(h, pe->b.key, pe->value);
                if (r < 0 && r != -EEXIST)
                        return r;
        }

        return 0;
}

int set_merge(Set *s, Set *other) {
        Iterator i;
        unsigned idx;

        assert(s);

        HASHMAP_FOREACH_IDX(idx, HASHMAP_BASE(other), i) {
                struct set_entry *se = set_bucket_at(other, idx);
                int r;

                r = set_put(s, se->b.key);
                if (r < 0)
                        return r;
        }

        return 0;
}

int internal_hashmap_reserve(HashmapBase *h, unsigned entries_add) {
        int r;

        assert(h);

        r = resize_buckets(h, entries_add);
        if (r < 0)
                return r;

        return 0;
}

/*
 * The same as hashmap_merge(), but every new item from other is moved to h.
 * Keys already in h are skipped and stay in other.
 * Returns: 0 on success.
 *          -ENOMEM on alloc failure, in which case no move has been done.
 */
int internal_hashmap_move(HashmapBase *h, HashmapBase *other) {
        struct swap_entries swap;
        struct hashmap_base_entry *e, *n;
        Iterator i;
        unsigned idx;
        int r;

        assert(h);

        if (!other)
                return 0;

        assert(other->type == h->type);

        /*
         * This reserves buckets for the worst case, where none of other's
         * entries are yet present in h. This is preferable to risking
         * an allocation failure in the middle of the moving and having to
         * rollback or return a partial result.
         */
        r = resize_buckets(h, n_entries(other));
        if (r < 0)
                return r;

        HASHMAP_FOREACH_IDX(idx, other, i) {
                unsigned h_hash;

                e = bucket_at(other, idx);
                h_hash = bucket_hash(h, e->key);
                if (bucket_scan(h, h_hash, e->key) != IDX_NIL)
                        continue;

                n = &bucket_at_swap(&swap, IDX_PUT)->p.b;
                n->key = e->key;
                if (h->type != HASHMAP_TYPE_SET)
                        ((struct plain_hashmap_entry*) n)->value =
                                ((struct plain_hashmap_entry*) e)->value;
                assert_se(hashmap_put_boldly(h, h_hash, &swap, false) == 1);

                remove_entry(other, idx);
        }

        return 0;
}

int internal_hashmap_move_one(HashmapBase *h, HashmapBase *other, const void *key) {
        struct swap_entries swap;
        unsigned h_hash, other_hash, idx;
        struct hashmap_base_entry *e, *n;
        int r;

        assert(h);

        h_hash = bucket_hash(h, key);
        if (bucket_scan(h, h_hash, key) != IDX_NIL)
                return -EEXIST;

        if (!other)
                return -ENOENT;

        assert(other->type == h->type);

        other_hash = bucket_hash(other, key);
        idx = bucket_scan(other, other_hash, key);
        if (idx == IDX_NIL)
                return -ENOENT;

        e = bucket_at(other, idx);

        n = &bucket_at_swap(&swap, IDX_PUT)->p.b;
        n->key = e->key;
        if (h->type != HASHMAP_TYPE_SET)
                ((struct plain_hashmap_entry*) n)->value =
                        ((struct plain_hashmap_entry*) e)->value;
        r = hashmap_put_boldly(h, h_hash, &swap, true);
        if (r < 0)
                return r;

        remove_entry(other, idx);
        return 0;
}

HashmapBase *internal_hashmap_copy(HashmapBase *h) {
        HashmapBase *copy;
        int r;

        assert(h);

        copy = hashmap_base_new(h->hash_ops, h->type  HASHMAP_DEBUG_SRC_ARGS);
        if (!copy)
                return NULL;

        switch (h->type) {
        case HASHMAP_TYPE_PLAIN:
        case HASHMAP_TYPE_ORDERED:
                r = hashmap_merge((Hashmap*)copy, (Hashmap*)h);
                break;
        case HASHMAP_TYPE_SET:
                r = set_merge((Set*)copy, (Set*)h);
                break;
        default:
                assert_not_reached("Unknown hashmap type");
        }

        if (r < 0) {
                internal_hashmap_free(copy);
                return NULL;
        }

        return copy;
}

char **internal_hashmap_get_strv(HashmapBase *h) {
        char **sv;
        Iterator i;
        unsigned idx, n;

        sv = new(char*, n_entries(h)+1);
        if (!sv)
                return NULL;

        n = 0;
        HASHMAP_FOREACH_IDX(idx, h, i)
                sv[n++] = entry_value(h, bucket_at(h, idx));
        sv[n] = NULL;

        return sv;
}

void *ordered_hashmap_next(OrderedHashmap *h, const void *key) {
        struct ordered_hashmap_entry *e;
        unsigned hash, idx;

        if (!h)
                return NULL;

        hash = bucket_hash(h, key);
        idx = bucket_scan(h, hash, key);
        if (idx == IDX_NIL)
                return NULL;

        e = ordered_bucket_at(h, idx);
        if (e->iterate_next == IDX_NIL)
                return NULL;
        return ordered_bucket_at(h, e->iterate_next)->p.value;
}

int set_consume(Set *s, void *value) {
        int r;

        assert(s);
        assert(value);

        r = set_put(s, value);
        if (r <= 0)
                free(value);

        return r;
}

int set_put_strdup(Set *s, const char *p) {
        char *c;

        assert(s);
        assert(p);

        if (set_contains(s, (char*) p))
                return 0;

        c = strdup(p);
        if (!c)
                return -ENOMEM;

        return set_consume(s, c);
}

#if 0 /// UNNEEDED by elogind
int set_put_strdupv(Set *s, char **l) {
        int n = 0, r;
        char **i;

        assert(s);

        STRV_FOREACH(i, l) {
                r = set_put_strdup(s, *i);
                if (r < 0)
                        return r;

                n += r;
        }

        return n;
}

int set_put_strsplit(Set *s, const char *v, const char *separators, ExtractFlags flags) {
        const char *p = v;
        int r;

        assert(s);
        assert(v);

        for (;;) {
                char *word;

                r = extract_first_word(&p, &word, separators, flags);
                if (r <= 0)
                        return r;

                r = set_consume(s, word);
                if (r < 0)
                        return r;
        }
}
#endif // 0