]>
Commit | Line | Data |
---|---|---|
52ad194e | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
663996b3 | 2 | |
663996b3 | 3 | #include <errno.h> |
4c89c718 | 4 | #include <stdint.h> |
db2df898 | 5 | #include <stdlib.h> |
4c89c718 | 6 | #include <string.h> |
663996b3 | 7 | |
db2df898 | 8 | #include "alloc-util.h" |
52ad194e | 9 | #include "fileio.h" |
6e866b33 | 10 | #include "hashmap.h" |
663996b3 | 11 | #include "macro.h" |
bb4f798a | 12 | #include "memory-util.h" |
5eef597e | 13 | #include "mempool.h" |
db2df898 | 14 | #include "process-util.h" |
e3bff60a | 15 | #include "random-util.h" |
db2df898 MP |
16 | #include "set.h" |
17 | #include "siphash24.h" | |
52ad194e | 18 | #include "string-util.h" |
db2df898 | 19 | #include "strv.h" |
e3bff60a | 20 | |
f5e65279 | 21 | #if ENABLE_DEBUG_HASHMAP |
4c89c718 | 22 | #include <pthread.h> |
e3bff60a MP |
23 | #include "list.h" |
24 | #endif | |
663996b3 | 25 | |
f47781d8 MP |
26 | /* |
27 | * Implementation of hashmaps. | |
28 | * Addressing: open | |
29 | * - uses less RAM compared to closed addressing (chaining), because | |
30 | * our entries are small (especially in Sets, which tend to contain | |
31 | * the majority of entries in systemd). | |
32 | * Collision resolution: Robin Hood | |
33 | * - tends to equalize displacement of entries from their optimal buckets. | |
34 | * Probe sequence: linear | |
35 | * - though theoretically worse than random probing/uniform hashing/double | |
36 | * hashing, it is good for cache locality. | |
37 | * | |
38 | * References: | |
39 | * Celis, P. 1986. Robin Hood Hashing. | |
40 | * Ph.D. Dissertation. University of Waterloo, Waterloo, Ont., Canada, Canada. | |
41 | * https://cs.uwaterloo.ca/research/tr/1986/CS-86-14.pdf | |
42 | * - The results are derived for random probing. Suggests deletion with | |
43 | * tombstones and two mean-centered search methods. None of that works | |
44 | * well for linear probing. | |
45 | * | |
46 | * Janson, S. 2005. Individual displacements for linear probing hashing with different insertion policies. | |
47 | * ACM Trans. Algorithms 1, 2 (October 2005), 177-213. | |
48 | * DOI=10.1145/1103963.1103964 http://doi.acm.org/10.1145/1103963.1103964 | |
49 | * http://www.math.uu.se/~svante/papers/sj157.pdf | |
50 | * - Applies to Robin Hood with linear probing. Contains remarks on | |
51 | * the unsuitability of mean-centered search with linear probing. | |
52 | * | |
53 | * Viola, A. 2005. Exact distribution of individual displacements in linear probing hashing. | |
54 | * ACM Trans. Algorithms 1, 2 (October 2005), 214-242. | |
55 | * DOI=10.1145/1103963.1103965 http://doi.acm.org/10.1145/1103963.1103965 | |
56 | * - Similar to Janson. Note that Viola writes about C_{m,n} (number of probes | |
57 | * in a successful search), and Janson writes about displacement. C = d + 1. | |
58 | * | |
59 | * Goossaert, E. 2013. Robin Hood hashing: backward shift deletion. | |
60 | * http://codecapsule.com/2013/11/17/robin-hood-hashing-backward-shift-deletion/ | |
61 | * - Explanation of backward shift deletion with pictures. | |
62 | * | |
63 | * Khuong, P. 2013. The Other Robin Hood Hashing. | |
64 | * http://www.pvk.ca/Blog/2013/11/26/the-other-robin-hood-hashing/ | |
65 | * - Short summary of random vs. linear probing, and tombstones vs. backward shift. | |
66 | */ | |
67 | ||
68 | /* | |
69 | * XXX Ideas for improvement: | |
70 | * For unordered hashmaps, randomize iteration order, similarly to Perl: | |
71 | * http://blog.booking.com/hardening-perls-hash-function.html | |
72 | */ | |
73 | ||
74 | /* INV_KEEP_FREE = 1 / (1 - max_load_factor) | |
75 | * e.g. 1 / (1 - 0.8) = 5 ... keep one fifth of the buckets free. */ | |
76 | #define INV_KEEP_FREE 5U | |
77 | ||
78 | /* Fields common to entries of all hashmap/set types */ | |
79 | struct hashmap_base_entry { | |
663996b3 | 80 | const void *key; |
f47781d8 MP |
81 | }; |
82 | ||
83 | /* Entry types for specific hashmap/set types | |
84 | * hashmap_base_entry must be at the beginning of each entry struct. */ | |
85 | ||
86 | struct plain_hashmap_entry { | |
87 | struct hashmap_base_entry b; | |
663996b3 | 88 | void *value; |
663996b3 MS |
89 | }; |
90 | ||
f47781d8 MP |
91 | struct ordered_hashmap_entry { |
92 | struct plain_hashmap_entry p; | |
93 | unsigned iterate_next, iterate_previous; | |
94 | }; | |
663996b3 | 95 | |
f47781d8 MP |
96 | struct set_entry { |
97 | struct hashmap_base_entry b; | |
98 | }; | |
663996b3 | 99 | |
f47781d8 MP |
100 | /* In several functions it is advantageous to have the hash table extended |
101 | * virtually by a couple of additional buckets. We reserve special index values | |
102 | * for these "swap" buckets. */ | |
103 | #define _IDX_SWAP_BEGIN (UINT_MAX - 3) | |
104 | #define IDX_PUT (_IDX_SWAP_BEGIN + 0) | |
105 | #define IDX_TMP (_IDX_SWAP_BEGIN + 1) | |
106 | #define _IDX_SWAP_END (_IDX_SWAP_BEGIN + 2) | |
14228c0d | 107 | |
f47781d8 MP |
108 | #define IDX_FIRST (UINT_MAX - 1) /* special index for freshly initialized iterators */ |
109 | #define IDX_NIL UINT_MAX /* special index value meaning "none" or "end" */ | |
110 | ||
111 | assert_cc(IDX_FIRST == _IDX_SWAP_END); | |
112 | assert_cc(IDX_FIRST == _IDX_ITERATOR_FIRST); | |
113 | ||
114 | /* Storage space for the "swap" buckets. | |
115 | * All entry types can fit into a ordered_hashmap_entry. */ | |
116 | struct swap_entries { | |
117 | struct ordered_hashmap_entry e[_IDX_SWAP_END - _IDX_SWAP_BEGIN]; | |
663996b3 MS |
118 | }; |
119 | ||
f47781d8 MP |
120 | /* Distance from Initial Bucket */ |
121 | typedef uint8_t dib_raw_t; | |
e3bff60a MP |
122 | #define DIB_RAW_OVERFLOW ((dib_raw_t)0xfdU) /* indicates DIB value is greater than representable */ |
123 | #define DIB_RAW_REHASH ((dib_raw_t)0xfeU) /* entry yet to be rehashed during in-place resize */ | |
124 | #define DIB_RAW_FREE ((dib_raw_t)0xffU) /* a free bucket */ | |
125 | #define DIB_RAW_INIT ((char)DIB_RAW_FREE) /* a byte to memset a DIB store with when initializing */ | |
f47781d8 MP |
126 | |
127 | #define DIB_FREE UINT_MAX | |
128 | ||
f5e65279 | 129 | #if ENABLE_DEBUG_HASHMAP |
f47781d8 MP |
130 | struct hashmap_debug_info { |
131 | LIST_FIELDS(struct hashmap_debug_info, debug_list); | |
132 | unsigned max_entries; /* high watermark of n_entries */ | |
133 | ||
134 | /* who allocated this hashmap */ | |
135 | int line; | |
136 | const char *file; | |
137 | const char *func; | |
138 | ||
139 | /* fields to detect modification while iterating */ | |
140 | unsigned put_count; /* counts puts into the hashmap */ | |
141 | unsigned rem_count; /* counts removals from hashmap */ | |
142 | unsigned last_rem_idx; /* remembers last removal index */ | |
663996b3 MS |
143 | }; |
144 | ||
f47781d8 MP |
145 | /* Tracks all existing hashmaps. Get at it from gdb. See sd_dump_hashmaps.py */ |
146 | static LIST_HEAD(struct hashmap_debug_info, hashmap_debug_list); | |
86f210e9 | 147 | static pthread_mutex_t hashmap_debug_list_mutex = PTHREAD_MUTEX_INITIALIZER; |
663996b3 | 148 | |
f47781d8 | 149 | #define HASHMAP_DEBUG_FIELDS struct hashmap_debug_info debug; |
663996b3 | 150 | |
e735f4d4 | 151 | #else /* !ENABLE_DEBUG_HASHMAP */ |
f47781d8 | 152 | #define HASHMAP_DEBUG_FIELDS |
e735f4d4 | 153 | #endif /* ENABLE_DEBUG_HASHMAP */ |
663996b3 | 154 | |
f47781d8 MP |
155 | enum HashmapType { |
156 | HASHMAP_TYPE_PLAIN, | |
157 | HASHMAP_TYPE_ORDERED, | |
158 | HASHMAP_TYPE_SET, | |
159 | _HASHMAP_TYPE_MAX | |
160 | }; | |
663996b3 | 161 | |
f47781d8 | 162 | struct _packed_ indirect_storage { |
aa27b158 | 163 | void *storage; /* where buckets and DIBs are stored */ |
f47781d8 MP |
164 | uint8_t hash_key[HASH_KEY_SIZE]; /* hash key; changes during resize */ |
165 | ||
166 | unsigned n_entries; /* number of stored entries */ | |
167 | unsigned n_buckets; /* number of buckets */ | |
168 | ||
169 | unsigned idx_lowest_entry; /* Index below which all buckets are free. | |
170 | Makes "while(hashmap_steal_first())" loops | |
171 | O(n) instead of O(n^2) for unordered hashmaps. */ | |
172 | uint8_t _pad[3]; /* padding for the whole HashmapBase */ | |
173 | /* The bitfields in HashmapBase complete the alignment of the whole thing. */ | |
174 | }; | |
175 | ||
176 | struct direct_storage { | |
177 | /* This gives us 39 bytes on 64bit, or 35 bytes on 32bit. | |
178 | * That's room for 4 set_entries + 4 DIB bytes + 3 unused bytes on 64bit, | |
179 | * or 7 set_entries + 7 DIB bytes + 0 unused bytes on 32bit. */ | |
aa27b158 | 180 | uint8_t storage[sizeof(struct indirect_storage)]; |
f47781d8 MP |
181 | }; |
182 | ||
183 | #define DIRECT_BUCKETS(entry_t) \ | |
184 | (sizeof(struct direct_storage) / (sizeof(entry_t) + sizeof(dib_raw_t))) | |
185 | ||
186 | /* We should be able to store at least one entry directly. */ | |
187 | assert_cc(DIRECT_BUCKETS(struct ordered_hashmap_entry) >= 1); | |
188 | ||
189 | /* We have 3 bits for n_direct_entries. */ | |
190 | assert_cc(DIRECT_BUCKETS(struct set_entry) < (1 << 3)); | |
191 | ||
192 | /* Hashmaps with directly stored entries all use this shared hash key. | |
193 | * It's no big deal if the key is guessed, because there can be only | |
194 | * a handful of directly stored entries in a hashmap. When a hashmap | |
195 | * outgrows direct storage, it gets its own key for indirect storage. */ | |
196 | static uint8_t shared_hash_key[HASH_KEY_SIZE]; | |
197 | static bool shared_hash_key_initialized; | |
198 | ||
199 | /* Fields that all hashmap/set types must have */ | |
200 | struct HashmapBase { | |
201 | const struct hash_ops *hash_ops; /* hash and compare ops to use */ | |
202 | ||
203 | union _packed_ { | |
204 | struct indirect_storage indirect; /* if has_indirect */ | |
205 | struct direct_storage direct; /* if !has_indirect */ | |
206 | }; | |
207 | ||
208 | enum HashmapType type:2; /* HASHMAP_TYPE_* */ | |
209 | bool has_indirect:1; /* whether indirect storage is used */ | |
210 | unsigned n_direct_entries:3; /* Number of entries in direct storage. | |
211 | * Only valid if !has_indirect. */ | |
212 | bool from_pool:1; /* whether was allocated from mempool */ | |
98393f85 MB |
213 | bool dirty:1; /* whether dirtied since last iterated_cache_get() */ |
214 | bool cached:1; /* whether this hashmap is being cached */ | |
f47781d8 MP |
215 | HASHMAP_DEBUG_FIELDS /* optional hashmap_debug_info */ |
216 | }; | |
217 | ||
218 | /* Specific hash types | |
219 | * HashmapBase must be at the beginning of each hashmap struct. */ | |
220 | ||
221 | struct Hashmap { | |
222 | struct HashmapBase b; | |
223 | }; | |
224 | ||
225 | struct OrderedHashmap { | |
226 | struct HashmapBase b; | |
227 | unsigned iterate_list_head, iterate_list_tail; | |
228 | }; | |
229 | ||
230 | struct Set { | |
231 | struct HashmapBase b; | |
232 | }; | |
233 | ||
98393f85 MB |
234 | typedef struct CacheMem { |
235 | const void **ptr; | |
236 | size_t n_populated, n_allocated; | |
237 | bool active:1; | |
238 | } CacheMem; | |
239 | ||
240 | struct IteratedCache { | |
241 | HashmapBase *hashmap; | |
242 | CacheMem keys, values; | |
243 | }; | |
244 | ||
f47781d8 MP |
245 | DEFINE_MEMPOOL(hashmap_pool, Hashmap, 8); |
246 | DEFINE_MEMPOOL(ordered_hashmap_pool, OrderedHashmap, 8); | |
247 | /* No need for a separate Set pool */ | |
248 | assert_cc(sizeof(Hashmap) == sizeof(Set)); | |
249 | ||
250 | struct hashmap_type_info { | |
251 | size_t head_size; | |
252 | size_t entry_size; | |
253 | struct mempool *mempool; | |
254 | unsigned n_direct_buckets; | |
255 | }; | |
256 | ||
257 | static const struct hashmap_type_info hashmap_type_info[_HASHMAP_TYPE_MAX] = { | |
258 | [HASHMAP_TYPE_PLAIN] = { | |
259 | .head_size = sizeof(Hashmap), | |
260 | .entry_size = sizeof(struct plain_hashmap_entry), | |
261 | .mempool = &hashmap_pool, | |
262 | .n_direct_buckets = DIRECT_BUCKETS(struct plain_hashmap_entry), | |
263 | }, | |
264 | [HASHMAP_TYPE_ORDERED] = { | |
265 | .head_size = sizeof(OrderedHashmap), | |
266 | .entry_size = sizeof(struct ordered_hashmap_entry), | |
267 | .mempool = &ordered_hashmap_pool, | |
268 | .n_direct_buckets = DIRECT_BUCKETS(struct ordered_hashmap_entry), | |
269 | }, | |
270 | [HASHMAP_TYPE_SET] = { | |
271 | .head_size = sizeof(Set), | |
272 | .entry_size = sizeof(struct set_entry), | |
273 | .mempool = &hashmap_pool, | |
274 | .n_direct_buckets = DIRECT_BUCKETS(struct set_entry), | |
275 | }, | |
276 | }; | |
663996b3 | 277 | |
b012e921 | 278 | #if VALGRIND |
6e866b33 | 279 | _destructor_ static void cleanup_pools(void) { |
52ad194e MB |
280 | _cleanup_free_ char *t = NULL; |
281 | int r; | |
282 | ||
283 | /* Be nice to valgrind */ | |
284 | ||
285 | /* The pool is only allocated by the main thread, but the memory can | |
286 | * be passed to other threads. Let's clean up if we are the main thread | |
287 | * and no other threads are live. */ | |
288 | if (!is_main_thread()) | |
289 | return; | |
290 | ||
291 | r = get_proc_field("/proc/self/status", "Threads", WHITESPACE, &t); | |
292 | if (r < 0 || !streq(t, "1")) | |
293 | return; | |
294 | ||
295 | mempool_drop(&hashmap_pool); | |
296 | mempool_drop(&ordered_hashmap_pool); | |
297 | } | |
298 | #endif | |
299 | ||
f47781d8 MP |
300 | static unsigned n_buckets(HashmapBase *h) { |
301 | return h->has_indirect ? h->indirect.n_buckets | |
302 | : hashmap_type_info[h->type].n_direct_buckets; | |
303 | } | |
304 | ||
305 | static unsigned n_entries(HashmapBase *h) { | |
306 | return h->has_indirect ? h->indirect.n_entries | |
307 | : h->n_direct_entries; | |
308 | } | |
309 | ||
310 | static void n_entries_inc(HashmapBase *h) { | |
311 | if (h->has_indirect) | |
312 | h->indirect.n_entries++; | |
313 | else | |
314 | h->n_direct_entries++; | |
315 | } | |
316 | ||
317 | static void n_entries_dec(HashmapBase *h) { | |
318 | if (h->has_indirect) | |
319 | h->indirect.n_entries--; | |
320 | else | |
321 | h->n_direct_entries--; | |
322 | } | |
323 | ||
aa27b158 | 324 | static void *storage_ptr(HashmapBase *h) { |
f47781d8 MP |
325 | return h->has_indirect ? h->indirect.storage |
326 | : h->direct.storage; | |
327 | } | |
328 | ||
329 | static uint8_t *hash_key(HashmapBase *h) { | |
330 | return h->has_indirect ? h->indirect.hash_key | |
331 | : shared_hash_key; | |
332 | } | |
333 | ||
334 | static unsigned base_bucket_hash(HashmapBase *h, const void *p) { | |
6300502b MP |
335 | struct siphash state; |
336 | uint64_t hash; | |
337 | ||
338 | siphash24_init(&state, hash_key(h)); | |
339 | ||
340 | h->hash_ops->hash(p, &state); | |
341 | ||
db2df898 | 342 | hash = siphash24_finalize(&state); |
6300502b MP |
343 | |
344 | return (unsigned) (hash % n_buckets(h)); | |
60f067b4 | 345 | } |
f47781d8 | 346 | #define bucket_hash(h, p) base_bucket_hash(HASHMAP_BASE(h), p) |
60f067b4 | 347 | |
7c20daf6 | 348 | static void base_set_dirty(HashmapBase *h) { |
98393f85 MB |
349 | h->dirty = true; |
350 | } | |
351 | #define hashmap_set_dirty(h) base_set_dirty(HASHMAP_BASE(h)) | |
352 | ||
60f067b4 JS |
353 | static void get_hash_key(uint8_t hash_key[HASH_KEY_SIZE], bool reuse_is_ok) { |
354 | static uint8_t current[HASH_KEY_SIZE]; | |
355 | static bool current_initialized = false; | |
356 | ||
357 | /* Returns a hash function key to use. In order to keep things | |
358 | * fast we will not generate a new key each time we allocate a | |
359 | * new hash table. Instead, we'll just reuse the most recently | |
360 | * generated one, except if we never generated one or when we | |
361 | * are rehashing an entire hash table because we reached a | |
362 | * fill level */ | |
363 | ||
364 | if (!current_initialized || !reuse_is_ok) { | |
365 | random_bytes(current, sizeof(current)); | |
366 | current_initialized = true; | |
367 | } | |
368 | ||
369 | memcpy(hash_key, current, sizeof(current)); | |
14228c0d MB |
370 | } |
371 | ||
f47781d8 MP |
372 | static struct hashmap_base_entry *bucket_at(HashmapBase *h, unsigned idx) { |
373 | return (struct hashmap_base_entry*) | |
aa27b158 | 374 | ((uint8_t*) storage_ptr(h) + idx * hashmap_type_info[h->type].entry_size); |
f47781d8 MP |
375 | } |
376 | ||
377 | static struct plain_hashmap_entry *plain_bucket_at(Hashmap *h, unsigned idx) { | |
378 | return (struct plain_hashmap_entry*) bucket_at(HASHMAP_BASE(h), idx); | |
379 | } | |
380 | ||
381 | static struct ordered_hashmap_entry *ordered_bucket_at(OrderedHashmap *h, unsigned idx) { | |
382 | return (struct ordered_hashmap_entry*) bucket_at(HASHMAP_BASE(h), idx); | |
383 | } | |
663996b3 | 384 | |
f47781d8 MP |
385 | static struct set_entry *set_bucket_at(Set *h, unsigned idx) { |
386 | return (struct set_entry*) bucket_at(HASHMAP_BASE(h), idx); | |
387 | } | |
663996b3 | 388 | |
f47781d8 MP |
389 | static struct ordered_hashmap_entry *bucket_at_swap(struct swap_entries *swap, unsigned idx) { |
390 | return &swap->e[idx - _IDX_SWAP_BEGIN]; | |
391 | } | |
663996b3 | 392 | |
f47781d8 MP |
393 | /* Returns a pointer to the bucket at index idx. |
394 | * Understands real indexes and swap indexes, hence "_virtual". */ | |
395 | static struct hashmap_base_entry *bucket_at_virtual(HashmapBase *h, struct swap_entries *swap, | |
396 | unsigned idx) { | |
397 | if (idx < _IDX_SWAP_BEGIN) | |
398 | return bucket_at(h, idx); | |
399 | ||
400 | if (idx < _IDX_SWAP_END) | |
401 | return &bucket_at_swap(swap, idx)->p.b; | |
402 | ||
403 | assert_not_reached("Invalid index"); | |
404 | } | |
405 | ||
406 | static dib_raw_t *dib_raw_ptr(HashmapBase *h) { | |
407 | return (dib_raw_t*) | |
aa27b158 | 408 | ((uint8_t*) storage_ptr(h) + hashmap_type_info[h->type].entry_size * n_buckets(h)); |
f47781d8 MP |
409 | } |
410 | ||
411 | static unsigned bucket_distance(HashmapBase *h, unsigned idx, unsigned from) { | |
412 | return idx >= from ? idx - from | |
413 | : n_buckets(h) + idx - from; | |
414 | } | |
415 | ||
416 | static unsigned bucket_calculate_dib(HashmapBase *h, unsigned idx, dib_raw_t raw_dib) { | |
417 | unsigned initial_bucket; | |
418 | ||
419 | if (raw_dib == DIB_RAW_FREE) | |
420 | return DIB_FREE; | |
421 | ||
422 | if (_likely_(raw_dib < DIB_RAW_OVERFLOW)) | |
423 | return raw_dib; | |
424 | ||
425 | /* | |
426 | * Having an overflow DIB value is very unlikely. The hash function | |
427 | * would have to be bad. For example, in a table of size 2^24 filled | |
428 | * to load factor 0.9 the maximum observed DIB is only about 60. | |
429 | * In theory (assuming I used Maxima correctly), for an infinite size | |
430 | * hash table with load factor 0.8 the probability of a given entry | |
431 | * having DIB > 40 is 1.9e-8. | |
432 | * This returns the correct DIB value by recomputing the hash value in | |
433 | * the unlikely case. XXX Hitting this case could be a hint to rehash. | |
434 | */ | |
435 | initial_bucket = bucket_hash(h, bucket_at(h, idx)->key); | |
436 | return bucket_distance(h, idx, initial_bucket); | |
437 | } | |
438 | ||
439 | static void bucket_set_dib(HashmapBase *h, unsigned idx, unsigned dib) { | |
440 | dib_raw_ptr(h)[idx] = dib != DIB_FREE ? MIN(dib, DIB_RAW_OVERFLOW) : DIB_RAW_FREE; | |
441 | } | |
442 | ||
443 | static unsigned skip_free_buckets(HashmapBase *h, unsigned idx) { | |
444 | dib_raw_t *dibs; | |
445 | ||
446 | dibs = dib_raw_ptr(h); | |
447 | ||
448 | for ( ; idx < n_buckets(h); idx++) | |
449 | if (dibs[idx] != DIB_RAW_FREE) | |
450 | return idx; | |
451 | ||
452 | return IDX_NIL; | |
453 | } | |
454 | ||
455 | static void bucket_mark_free(HashmapBase *h, unsigned idx) { | |
e735f4d4 | 456 | memzero(bucket_at(h, idx), hashmap_type_info[h->type].entry_size); |
f47781d8 MP |
457 | bucket_set_dib(h, idx, DIB_FREE); |
458 | } | |
459 | ||
460 | static void bucket_move_entry(HashmapBase *h, struct swap_entries *swap, | |
461 | unsigned from, unsigned to) { | |
462 | struct hashmap_base_entry *e_from, *e_to; | |
463 | ||
464 | assert(from != to); | |
663996b3 | 465 | |
f47781d8 MP |
466 | e_from = bucket_at_virtual(h, swap, from); |
467 | e_to = bucket_at_virtual(h, swap, to); | |
468 | ||
469 | memcpy(e_to, e_from, hashmap_type_info[h->type].entry_size); | |
470 | ||
471 | if (h->type == HASHMAP_TYPE_ORDERED) { | |
472 | OrderedHashmap *lh = (OrderedHashmap*) h; | |
473 | struct ordered_hashmap_entry *le, *le_to; | |
474 | ||
475 | le_to = (struct ordered_hashmap_entry*) e_to; | |
476 | ||
477 | if (le_to->iterate_next != IDX_NIL) { | |
478 | le = (struct ordered_hashmap_entry*) | |
479 | bucket_at_virtual(h, swap, le_to->iterate_next); | |
480 | le->iterate_previous = to; | |
481 | } | |
482 | ||
483 | if (le_to->iterate_previous != IDX_NIL) { | |
484 | le = (struct ordered_hashmap_entry*) | |
485 | bucket_at_virtual(h, swap, le_to->iterate_previous); | |
486 | le->iterate_next = to; | |
487 | } | |
488 | ||
489 | if (lh->iterate_list_head == from) | |
490 | lh->iterate_list_head = to; | |
491 | if (lh->iterate_list_tail == from) | |
492 | lh->iterate_list_tail = to; | |
663996b3 | 493 | } |
f47781d8 | 494 | } |
663996b3 | 495 | |
f47781d8 MP |
496 | static unsigned next_idx(HashmapBase *h, unsigned idx) { |
497 | return (idx + 1U) % n_buckets(h); | |
498 | } | |
663996b3 | 499 | |
f47781d8 MP |
500 | static unsigned prev_idx(HashmapBase *h, unsigned idx) { |
501 | return (n_buckets(h) + idx - 1U) % n_buckets(h); | |
502 | } | |
663996b3 | 503 | |
f47781d8 MP |
504 | static void *entry_value(HashmapBase *h, struct hashmap_base_entry *e) { |
505 | switch (h->type) { | |
14228c0d | 506 | |
f47781d8 MP |
507 | case HASHMAP_TYPE_PLAIN: |
508 | case HASHMAP_TYPE_ORDERED: | |
509 | return ((struct plain_hashmap_entry*)e)->value; | |
663996b3 | 510 | |
f47781d8 MP |
511 | case HASHMAP_TYPE_SET: |
512 | return (void*) e->key; | |
14228c0d | 513 | |
f47781d8 MP |
514 | default: |
515 | assert_not_reached("Unknown hashmap type"); | |
516 | } | |
663996b3 MS |
517 | } |
518 | ||
f47781d8 MP |
519 | static void base_remove_entry(HashmapBase *h, unsigned idx) { |
520 | unsigned left, right, prev, dib; | |
521 | dib_raw_t raw_dib, *dibs; | |
14228c0d | 522 | |
f47781d8 MP |
523 | dibs = dib_raw_ptr(h); |
524 | assert(dibs[idx] != DIB_RAW_FREE); | |
663996b3 | 525 | |
f5e65279 | 526 | #if ENABLE_DEBUG_HASHMAP |
f47781d8 MP |
527 | h->debug.rem_count++; |
528 | h->debug.last_rem_idx = idx; | |
529 | #endif | |
663996b3 | 530 | |
f47781d8 MP |
531 | left = idx; |
532 | /* Find the stop bucket ("right"). It is either free or has DIB == 0. */ | |
533 | for (right = next_idx(h, left); ; right = next_idx(h, right)) { | |
534 | raw_dib = dibs[right]; | |
f5e65279 | 535 | if (IN_SET(raw_dib, 0, DIB_RAW_FREE)) |
f47781d8 MP |
536 | break; |
537 | ||
538 | /* The buckets are not supposed to be all occupied and with DIB > 0. | |
539 | * That would mean we could make everyone better off by shifting them | |
540 | * backward. This scenario is impossible. */ | |
541 | assert(left != right); | |
542 | } | |
663996b3 | 543 | |
f47781d8 MP |
544 | if (h->type == HASHMAP_TYPE_ORDERED) { |
545 | OrderedHashmap *lh = (OrderedHashmap*) h; | |
546 | struct ordered_hashmap_entry *le = ordered_bucket_at(lh, idx); | |
547 | ||
548 | if (le->iterate_next != IDX_NIL) | |
549 | ordered_bucket_at(lh, le->iterate_next)->iterate_previous = le->iterate_previous; | |
550 | else | |
551 | lh->iterate_list_tail = le->iterate_previous; | |
552 | ||
553 | if (le->iterate_previous != IDX_NIL) | |
554 | ordered_bucket_at(lh, le->iterate_previous)->iterate_next = le->iterate_next; | |
555 | else | |
556 | lh->iterate_list_head = le->iterate_next; | |
557 | } | |
558 | ||
559 | /* Now shift all buckets in the interval (left, right) one step backwards */ | |
560 | for (prev = left, left = next_idx(h, left); left != right; | |
561 | prev = left, left = next_idx(h, left)) { | |
562 | dib = bucket_calculate_dib(h, left, dibs[left]); | |
563 | assert(dib != 0); | |
564 | bucket_move_entry(h, NULL, left, prev); | |
565 | bucket_set_dib(h, prev, dib - 1); | |
566 | } | |
567 | ||
568 | bucket_mark_free(h, prev); | |
569 | n_entries_dec(h); | |
98393f85 | 570 | base_set_dirty(h); |
663996b3 | 571 | } |
f47781d8 MP |
572 | #define remove_entry(h, idx) base_remove_entry(HASHMAP_BASE(h), idx) |
573 | ||
574 | static unsigned hashmap_iterate_in_insertion_order(OrderedHashmap *h, Iterator *i) { | |
575 | struct ordered_hashmap_entry *e; | |
576 | unsigned idx; | |
663996b3 | 577 | |
663996b3 | 578 | assert(h); |
f47781d8 MP |
579 | assert(i); |
580 | ||
581 | if (i->idx == IDX_NIL) | |
582 | goto at_end; | |
583 | ||
584 | if (i->idx == IDX_FIRST && h->iterate_list_head == IDX_NIL) | |
585 | goto at_end; | |
586 | ||
587 | if (i->idx == IDX_FIRST) { | |
588 | idx = h->iterate_list_head; | |
589 | e = ordered_bucket_at(h, idx); | |
663996b3 | 590 | } else { |
f47781d8 MP |
591 | idx = i->idx; |
592 | e = ordered_bucket_at(h, idx); | |
593 | /* | |
594 | * We allow removing the current entry while iterating, but removal may cause | |
595 | * a backward shift. The next entry may thus move one bucket to the left. | |
596 | * To detect when it happens, we remember the key pointer of the entry we were | |
597 | * going to iterate next. If it does not match, there was a backward shift. | |
598 | */ | |
599 | if (e->p.b.key != i->next_key) { | |
600 | idx = prev_idx(HASHMAP_BASE(h), idx); | |
601 | e = ordered_bucket_at(h, idx); | |
602 | } | |
603 | assert(e->p.b.key == i->next_key); | |
663996b3 | 604 | } |
663996b3 | 605 | |
f5e65279 | 606 | #if ENABLE_DEBUG_HASHMAP |
f47781d8 MP |
607 | i->prev_idx = idx; |
608 | #endif | |
609 | ||
610 | if (e->iterate_next != IDX_NIL) { | |
611 | struct ordered_hashmap_entry *n; | |
612 | i->idx = e->iterate_next; | |
613 | n = ordered_bucket_at(h, i->idx); | |
614 | i->next_key = n->p.b.key; | |
615 | } else | |
616 | i->idx = IDX_NIL; | |
617 | ||
618 | return idx; | |
619 | ||
620 | at_end: | |
621 | i->idx = IDX_NIL; | |
622 | return IDX_NIL; | |
663996b3 MS |
623 | } |
624 | ||
f47781d8 MP |
625 | static unsigned hashmap_iterate_in_internal_order(HashmapBase *h, Iterator *i) { |
626 | unsigned idx; | |
627 | ||
663996b3 | 628 | assert(h); |
f47781d8 | 629 | assert(i); |
663996b3 | 630 | |
f47781d8 MP |
631 | if (i->idx == IDX_NIL) |
632 | goto at_end; | |
663996b3 | 633 | |
f47781d8 MP |
634 | if (i->idx == IDX_FIRST) { |
635 | /* fast forward to the first occupied bucket */ | |
636 | if (h->has_indirect) { | |
637 | i->idx = skip_free_buckets(h, h->indirect.idx_lowest_entry); | |
638 | h->indirect.idx_lowest_entry = i->idx; | |
639 | } else | |
640 | i->idx = skip_free_buckets(h, 0); | |
641 | ||
642 | if (i->idx == IDX_NIL) | |
643 | goto at_end; | |
644 | } else { | |
645 | struct hashmap_base_entry *e; | |
646 | ||
647 | assert(i->idx > 0); | |
663996b3 | 648 | |
f47781d8 MP |
649 | e = bucket_at(h, i->idx); |
650 | /* | |
651 | * We allow removing the current entry while iterating, but removal may cause | |
652 | * a backward shift. The next entry may thus move one bucket to the left. | |
653 | * To detect when it happens, we remember the key pointer of the entry we were | |
654 | * going to iterate next. If it does not match, there was a backward shift. | |
655 | */ | |
656 | if (e->key != i->next_key) | |
657 | e = bucket_at(h, --i->idx); | |
663996b3 | 658 | |
f47781d8 MP |
659 | assert(e->key == i->next_key); |
660 | } | |
661 | ||
662 | idx = i->idx; | |
f5e65279 | 663 | #if ENABLE_DEBUG_HASHMAP |
f47781d8 MP |
664 | i->prev_idx = idx; |
665 | #endif | |
666 | ||
667 | i->idx = skip_free_buckets(h, i->idx + 1); | |
668 | if (i->idx != IDX_NIL) | |
669 | i->next_key = bucket_at(h, i->idx)->key; | |
663996b3 | 670 | else |
f47781d8 MP |
671 | i->idx = IDX_NIL; |
672 | ||
673 | return idx; | |
663996b3 | 674 | |
f47781d8 MP |
675 | at_end: |
676 | i->idx = IDX_NIL; | |
677 | return IDX_NIL; | |
663996b3 MS |
678 | } |
679 | ||
f47781d8 MP |
680 | static unsigned hashmap_iterate_entry(HashmapBase *h, Iterator *i) { |
681 | if (!h) { | |
682 | i->idx = IDX_NIL; | |
683 | return IDX_NIL; | |
684 | } | |
663996b3 | 685 | |
f5e65279 | 686 | #if ENABLE_DEBUG_HASHMAP |
f47781d8 MP |
687 | if (i->idx == IDX_FIRST) { |
688 | i->put_count = h->debug.put_count; | |
689 | i->rem_count = h->debug.rem_count; | |
690 | } else { | |
691 | /* While iterating, must not add any new entries */ | |
692 | assert(i->put_count == h->debug.put_count); | |
693 | /* ... or remove entries other than the current one */ | |
694 | assert(i->rem_count == h->debug.rem_count || | |
695 | (i->rem_count == h->debug.rem_count - 1 && | |
696 | i->prev_idx == h->debug.last_rem_idx)); | |
697 | /* Reset our removals counter */ | |
698 | i->rem_count = h->debug.rem_count; | |
699 | } | |
700 | #endif | |
663996b3 | 701 | |
f47781d8 MP |
702 | return h->type == HASHMAP_TYPE_ORDERED ? hashmap_iterate_in_insertion_order((OrderedHashmap*) h, i) |
703 | : hashmap_iterate_in_internal_order(h, i); | |
704 | } | |
663996b3 | 705 | |
86f210e9 | 706 | bool internal_hashmap_iterate(HashmapBase *h, Iterator *i, void **value, const void **key) { |
f47781d8 MP |
707 | struct hashmap_base_entry *e; |
708 | void *data; | |
709 | unsigned idx; | |
710 | ||
711 | idx = hashmap_iterate_entry(h, i); | |
712 | if (idx == IDX_NIL) { | |
86f210e9 MP |
713 | if (value) |
714 | *value = NULL; | |
f47781d8 MP |
715 | if (key) |
716 | *key = NULL; | |
717 | ||
86f210e9 | 718 | return false; |
f47781d8 MP |
719 | } |
720 | ||
721 | e = bucket_at(h, idx); | |
722 | data = entry_value(h, e); | |
86f210e9 MP |
723 | if (value) |
724 | *value = data; | |
f47781d8 MP |
725 | if (key) |
726 | *key = e->key; | |
727 | ||
86f210e9 | 728 | return true; |
f47781d8 MP |
729 | } |
730 | ||
86f210e9 MP |
731 | bool set_iterate(Set *s, Iterator *i, void **value) { |
732 | return internal_hashmap_iterate(HASHMAP_BASE(s), i, value, NULL); | |
663996b3 MS |
733 | } |
734 | ||
f47781d8 MP |
735 | #define HASHMAP_FOREACH_IDX(idx, h, i) \ |
736 | for ((i) = ITERATOR_FIRST, (idx) = hashmap_iterate_entry((h), &(i)); \ | |
737 | (idx != IDX_NIL); \ | |
738 | (idx) = hashmap_iterate_entry((h), &(i))) | |
663996b3 | 739 | |
98393f85 MB |
740 | IteratedCache *internal_hashmap_iterated_cache_new(HashmapBase *h) { |
741 | IteratedCache *cache; | |
742 | ||
743 | assert(h); | |
744 | assert(!h->cached); | |
745 | ||
746 | if (h->cached) | |
747 | return NULL; | |
748 | ||
749 | cache = new0(IteratedCache, 1); | |
750 | if (!cache) | |
751 | return NULL; | |
752 | ||
753 | cache->hashmap = h; | |
754 | h->cached = true; | |
755 | ||
756 | return cache; | |
757 | } | |
758 | ||
f47781d8 MP |
759 | static void reset_direct_storage(HashmapBase *h) { |
760 | const struct hashmap_type_info *hi = &hashmap_type_info[h->type]; | |
761 | void *p; | |
762 | ||
763 | assert(!h->has_indirect); | |
764 | ||
765 | p = mempset(h->direct.storage, 0, hi->entry_size * hi->n_direct_buckets); | |
766 | memset(p, DIB_RAW_INIT, sizeof(dib_raw_t) * hi->n_direct_buckets); | |
767 | } | |
768 | ||
e3bff60a | 769 | static struct HashmapBase *hashmap_base_new(const struct hash_ops *hash_ops, enum HashmapType type HASHMAP_DEBUG_PARAMS) { |
f47781d8 MP |
770 | HashmapBase *h; |
771 | const struct hashmap_type_info *hi = &hashmap_type_info[type]; | |
6e866b33 | 772 | bool up; |
f47781d8 | 773 | |
6e866b33 | 774 | up = mempool_enabled(); |
663996b3 | 775 | |
6e866b33 | 776 | h = up ? mempool_alloc0_tile(hi->mempool) : malloc0(hi->head_size); |
663996b3 | 777 | if (!h) |
f47781d8 MP |
778 | return NULL; |
779 | ||
780 | h->type = type; | |
6e866b33 MB |
781 | h->from_pool = up; |
782 | h->hash_ops = hash_ops ?: &trivial_hash_ops; | |
f47781d8 MP |
783 | |
784 | if (type == HASHMAP_TYPE_ORDERED) { | |
785 | OrderedHashmap *lh = (OrderedHashmap*)h; | |
786 | lh->iterate_list_head = lh->iterate_list_tail = IDX_NIL; | |
787 | } | |
788 | ||
789 | reset_direct_storage(h); | |
790 | ||
791 | if (!shared_hash_key_initialized) { | |
792 | random_bytes(shared_hash_key, sizeof(shared_hash_key)); | |
793 | shared_hash_key_initialized= true; | |
794 | } | |
795 | ||
f5e65279 | 796 | #if ENABLE_DEBUG_HASHMAP |
f47781d8 MP |
797 | h->debug.func = func; |
798 | h->debug.file = file; | |
799 | h->debug.line = line; | |
86f210e9 MP |
800 | assert_se(pthread_mutex_lock(&hashmap_debug_list_mutex) == 0); |
801 | LIST_PREPEND(debug_list, hashmap_debug_list, &h->debug); | |
802 | assert_se(pthread_mutex_unlock(&hashmap_debug_list_mutex) == 0); | |
f47781d8 MP |
803 | #endif |
804 | ||
805 | return h; | |
806 | } | |
807 | ||
808 | Hashmap *internal_hashmap_new(const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { | |
e3bff60a | 809 | return (Hashmap*) hashmap_base_new(hash_ops, HASHMAP_TYPE_PLAIN HASHMAP_DEBUG_PASS_ARGS); |
f47781d8 MP |
810 | } |
811 | ||
812 | OrderedHashmap *internal_ordered_hashmap_new(const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { | |
e3bff60a | 813 | return (OrderedHashmap*) hashmap_base_new(hash_ops, HASHMAP_TYPE_ORDERED HASHMAP_DEBUG_PASS_ARGS); |
f47781d8 MP |
814 | } |
815 | ||
816 | Set *internal_set_new(const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { | |
e3bff60a | 817 | return (Set*) hashmap_base_new(hash_ops, HASHMAP_TYPE_SET HASHMAP_DEBUG_PASS_ARGS); |
f47781d8 MP |
818 | } |
819 | ||
820 | static int hashmap_base_ensure_allocated(HashmapBase **h, const struct hash_ops *hash_ops, | |
e3bff60a | 821 | enum HashmapType type HASHMAP_DEBUG_PARAMS) { |
f47781d8 MP |
822 | HashmapBase *q; |
823 | ||
824 | assert(h); | |
825 | ||
826 | if (*h) | |
827 | return 0; | |
828 | ||
e3bff60a | 829 | q = hashmap_base_new(hash_ops, type HASHMAP_DEBUG_PASS_ARGS); |
f47781d8 MP |
830 | if (!q) |
831 | return -ENOMEM; | |
832 | ||
833 | *h = q; | |
834 | return 0; | |
835 | } | |
836 | ||
837 | int internal_hashmap_ensure_allocated(Hashmap **h, const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { | |
e3bff60a | 838 | return hashmap_base_ensure_allocated((HashmapBase**)h, hash_ops, HASHMAP_TYPE_PLAIN HASHMAP_DEBUG_PASS_ARGS); |
f47781d8 MP |
839 | } |
840 | ||
841 | int internal_ordered_hashmap_ensure_allocated(OrderedHashmap **h, const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { | |
e3bff60a | 842 | return hashmap_base_ensure_allocated((HashmapBase**)h, hash_ops, HASHMAP_TYPE_ORDERED HASHMAP_DEBUG_PASS_ARGS); |
f47781d8 MP |
843 | } |
844 | ||
845 | int internal_set_ensure_allocated(Set **s, const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { | |
e3bff60a | 846 | return hashmap_base_ensure_allocated((HashmapBase**)s, hash_ops, HASHMAP_TYPE_SET HASHMAP_DEBUG_PASS_ARGS); |
f47781d8 | 847 | } |
663996b3 | 848 | |
f47781d8 MP |
849 | static void hashmap_free_no_clear(HashmapBase *h) { |
850 | assert(!h->has_indirect); | |
6e866b33 | 851 | assert(h->n_direct_entries == 0); |
663996b3 | 852 | |
f5e65279 | 853 | #if ENABLE_DEBUG_HASHMAP |
86f210e9 | 854 | assert_se(pthread_mutex_lock(&hashmap_debug_list_mutex) == 0); |
f47781d8 | 855 | LIST_REMOVE(debug_list, hashmap_debug_list, &h->debug); |
86f210e9 | 856 | assert_se(pthread_mutex_unlock(&hashmap_debug_list_mutex) == 0); |
f47781d8 | 857 | #endif |
14228c0d | 858 | |
6e866b33 MB |
859 | if (h->from_pool) { |
860 | /* Ensure that the object didn't get migrated between threads. */ | |
861 | assert_se(is_main_thread()); | |
f47781d8 | 862 | mempool_free_tile(hashmap_type_info[h->type].mempool, h); |
6e866b33 | 863 | } else |
663996b3 MS |
864 | free(h); |
865 | } | |
866 | ||
6e866b33 | 867 | HashmapBase *internal_hashmap_free(HashmapBase *h, free_func_t default_free_key, free_func_t default_free_value) { |
e3bff60a | 868 | if (h) { |
6e866b33 | 869 | internal_hashmap_clear(h, default_free_key, default_free_value); |
e3bff60a MP |
870 | hashmap_free_no_clear(h); |
871 | } | |
f47781d8 | 872 | |
e3bff60a | 873 | return NULL; |
f47781d8 MP |
874 | } |
875 | ||
6e866b33 MB |
876 | void internal_hashmap_clear(HashmapBase *h, free_func_t default_free_key, free_func_t default_free_value) { |
877 | free_func_t free_key, free_value; | |
878 | if (!h) | |
879 | return; | |
663996b3 | 880 | |
6e866b33 MB |
881 | free_key = h->hash_ops->free_key ?: default_free_key; |
882 | free_value = h->hash_ops->free_value ?: default_free_value; | |
663996b3 | 883 | |
6e866b33 | 884 | if (free_key || free_value) { |
663996b3 | 885 | |
6e866b33 MB |
886 | /* If destructor calls are defined, let's destroy things defensively: let's take the item out of the |
887 | * hash table, and only then call the destructor functions. If these destructors then try to unregister | |
888 | * themselves from our hash table a second time, the entry is already gone. */ | |
663996b3 | 889 | |
6e866b33 | 890 | while (internal_hashmap_size(h) > 0) { |
7c20daf6 FS |
891 | void *k = NULL; |
892 | void *v; | |
663996b3 | 893 | |
6e866b33 | 894 | v = internal_hashmap_first_key_and_value(h, true, &k); |
663996b3 | 895 | |
6e866b33 MB |
896 | if (free_key) |
897 | free_key(k); | |
663996b3 | 898 | |
6e866b33 MB |
899 | if (free_value) |
900 | free_value(v); | |
901 | } | |
902 | } | |
663996b3 | 903 | |
f47781d8 MP |
904 | if (h->has_indirect) { |
905 | free(h->indirect.storage); | |
906 | h->has_indirect = false; | |
907 | } | |
908 | ||
909 | h->n_direct_entries = 0; | |
910 | reset_direct_storage(h); | |
911 | ||
912 | if (h->type == HASHMAP_TYPE_ORDERED) { | |
913 | OrderedHashmap *lh = (OrderedHashmap*) h; | |
914 | lh->iterate_list_head = lh->iterate_list_tail = IDX_NIL; | |
915 | } | |
98393f85 MB |
916 | |
917 | base_set_dirty(h); | |
663996b3 MS |
918 | } |
919 | ||
f47781d8 MP |
920 | static int resize_buckets(HashmapBase *h, unsigned entries_add); |
921 | ||
922 | /* | |
923 | * Finds an empty bucket to put an entry into, starting the scan at 'idx'. | |
924 | * Performs Robin Hood swaps as it goes. The entry to put must be placed | |
925 | * by the caller into swap slot IDX_PUT. | |
926 | * If used for in-place resizing, may leave a displaced entry in swap slot | |
927 | * IDX_PUT. Caller must rehash it next. | |
928 | * Returns: true if it left a displaced entry to rehash next in IDX_PUT, | |
929 | * false otherwise. | |
930 | */ | |
931 | static bool hashmap_put_robin_hood(HashmapBase *h, unsigned idx, | |
932 | struct swap_entries *swap) { | |
933 | dib_raw_t raw_dib, *dibs; | |
934 | unsigned dib, distance; | |
935 | ||
f5e65279 | 936 | #if ENABLE_DEBUG_HASHMAP |
f47781d8 MP |
937 | h->debug.put_count++; |
938 | #endif | |
939 | ||
940 | dibs = dib_raw_ptr(h); | |
941 | ||
942 | for (distance = 0; ; distance++) { | |
943 | raw_dib = dibs[idx]; | |
f5e65279 | 944 | if (IN_SET(raw_dib, DIB_RAW_FREE, DIB_RAW_REHASH)) { |
f47781d8 MP |
945 | if (raw_dib == DIB_RAW_REHASH) |
946 | bucket_move_entry(h, swap, idx, IDX_TMP); | |
663996b3 | 947 | |
f47781d8 MP |
948 | if (h->has_indirect && h->indirect.idx_lowest_entry > idx) |
949 | h->indirect.idx_lowest_entry = idx; | |
663996b3 | 950 | |
f47781d8 MP |
951 | bucket_set_dib(h, idx, distance); |
952 | bucket_move_entry(h, swap, IDX_PUT, idx); | |
953 | if (raw_dib == DIB_RAW_REHASH) { | |
954 | bucket_move_entry(h, swap, IDX_TMP, IDX_PUT); | |
955 | return true; | |
956 | } | |
957 | ||
958 | return false; | |
959 | } | |
960 | ||
961 | dib = bucket_calculate_dib(h, idx, raw_dib); | |
962 | ||
963 | if (dib < distance) { | |
964 | /* Found a wealthier entry. Go Robin Hood! */ | |
f47781d8 MP |
965 | bucket_set_dib(h, idx, distance); |
966 | ||
967 | /* swap the entries */ | |
968 | bucket_move_entry(h, swap, idx, IDX_TMP); | |
969 | bucket_move_entry(h, swap, IDX_PUT, idx); | |
970 | bucket_move_entry(h, swap, IDX_TMP, IDX_PUT); | |
971 | ||
972 | distance = dib; | |
973 | } | |
974 | ||
975 | idx = next_idx(h, idx); | |
976 | } | |
663996b3 MS |
977 | } |
978 | ||
f47781d8 MP |
979 | /* |
980 | * Puts an entry into a hashmap, boldly - no check whether key already exists. | |
981 | * The caller must place the entry (only its key and value, not link indexes) | |
982 | * in swap slot IDX_PUT. | |
983 | * Caller must ensure: the key does not exist yet in the hashmap. | |
984 | * that resize is not needed if !may_resize. | |
985 | * Returns: 1 if entry was put successfully. | |
986 | * -ENOMEM if may_resize==true and resize failed with -ENOMEM. | |
987 | * Cannot return -ENOMEM if !may_resize. | |
988 | */ | |
989 | static int hashmap_base_put_boldly(HashmapBase *h, unsigned idx, | |
990 | struct swap_entries *swap, bool may_resize) { | |
991 | struct ordered_hashmap_entry *new_entry; | |
992 | int r; | |
993 | ||
994 | assert(idx < n_buckets(h)); | |
995 | ||
996 | new_entry = bucket_at_swap(swap, IDX_PUT); | |
997 | ||
998 | if (may_resize) { | |
999 | r = resize_buckets(h, 1); | |
1000 | if (r < 0) | |
1001 | return r; | |
1002 | if (r > 0) | |
1003 | idx = bucket_hash(h, new_entry->p.b.key); | |
1004 | } | |
1005 | assert(n_entries(h) < n_buckets(h)); | |
1006 | ||
1007 | if (h->type == HASHMAP_TYPE_ORDERED) { | |
1008 | OrderedHashmap *lh = (OrderedHashmap*) h; | |
1009 | ||
1010 | new_entry->iterate_next = IDX_NIL; | |
1011 | new_entry->iterate_previous = lh->iterate_list_tail; | |
1012 | ||
1013 | if (lh->iterate_list_tail != IDX_NIL) { | |
1014 | struct ordered_hashmap_entry *old_tail; | |
1015 | ||
1016 | old_tail = ordered_bucket_at(lh, lh->iterate_list_tail); | |
1017 | assert(old_tail->iterate_next == IDX_NIL); | |
1018 | old_tail->iterate_next = IDX_PUT; | |
1019 | } | |
1020 | ||
1021 | lh->iterate_list_tail = IDX_PUT; | |
1022 | if (lh->iterate_list_head == IDX_NIL) | |
1023 | lh->iterate_list_head = IDX_PUT; | |
1024 | } | |
1025 | ||
1026 | assert_se(hashmap_put_robin_hood(h, idx, swap) == false); | |
1027 | ||
1028 | n_entries_inc(h); | |
f5e65279 | 1029 | #if ENABLE_DEBUG_HASHMAP |
f47781d8 MP |
1030 | h->debug.max_entries = MAX(h->debug.max_entries, n_entries(h)); |
1031 | #endif | |
1032 | ||
98393f85 MB |
1033 | base_set_dirty(h); |
1034 | ||
f47781d8 MP |
1035 | return 1; |
1036 | } | |
1037 | #define hashmap_put_boldly(h, idx, swap, may_resize) \ | |
1038 | hashmap_base_put_boldly(HASHMAP_BASE(h), idx, swap, may_resize) | |
1039 | ||
1040 | /* | |
1041 | * Returns 0 if resize is not needed. | |
e735f4d4 | 1042 | * 1 if successfully resized. |
f47781d8 MP |
1043 | * -ENOMEM on allocation failure. |
1044 | */ | |
1045 | static int resize_buckets(HashmapBase *h, unsigned entries_add) { | |
1046 | struct swap_entries swap; | |
aa27b158 | 1047 | void *new_storage; |
f47781d8 MP |
1048 | dib_raw_t *old_dibs, *new_dibs; |
1049 | const struct hashmap_type_info *hi; | |
1050 | unsigned idx, optimal_idx; | |
1051 | unsigned old_n_buckets, new_n_buckets, n_rehashed, new_n_entries; | |
1052 | uint8_t new_shift; | |
1053 | bool rehash_next; | |
14228c0d MB |
1054 | |
1055 | assert(h); | |
1056 | ||
f47781d8 MP |
1057 | hi = &hashmap_type_info[h->type]; |
1058 | new_n_entries = n_entries(h) + entries_add; | |
5eef597e MP |
1059 | |
1060 | /* overflow? */ | |
f47781d8 | 1061 | if (_unlikely_(new_n_entries < entries_add)) |
5eef597e MP |
1062 | return -ENOMEM; |
1063 | ||
f47781d8 MP |
1064 | /* For direct storage we allow 100% load, because it's tiny. */ |
1065 | if (!h->has_indirect && new_n_entries <= hi->n_direct_buckets) | |
5eef597e | 1066 | return 0; |
14228c0d | 1067 | |
f47781d8 MP |
1068 | /* |
1069 | * Load factor = n/m = 1 - (1/INV_KEEP_FREE). | |
1070 | * From it follows: m = n + n/(INV_KEEP_FREE - 1) | |
1071 | */ | |
1072 | new_n_buckets = new_n_entries + new_n_entries / (INV_KEEP_FREE - 1); | |
1073 | /* overflow? */ | |
1074 | if (_unlikely_(new_n_buckets < new_n_entries)) | |
5eef597e | 1075 | return -ENOMEM; |
14228c0d | 1076 | |
f47781d8 MP |
1077 | if (_unlikely_(new_n_buckets > UINT_MAX / (hi->entry_size + sizeof(dib_raw_t)))) |
1078 | return -ENOMEM; | |
14228c0d | 1079 | |
f47781d8 | 1080 | old_n_buckets = n_buckets(h); |
14228c0d | 1081 | |
f47781d8 MP |
1082 | if (_likely_(new_n_buckets <= old_n_buckets)) |
1083 | return 0; | |
14228c0d | 1084 | |
f47781d8 MP |
1085 | new_shift = log2u_round_up(MAX( |
1086 | new_n_buckets * (hi->entry_size + sizeof(dib_raw_t)), | |
1087 | 2 * sizeof(struct direct_storage))); | |
14228c0d | 1088 | |
f47781d8 MP |
1089 | /* Realloc storage (buckets and DIB array). */ |
1090 | new_storage = realloc(h->has_indirect ? h->indirect.storage : NULL, | |
1091 | 1U << new_shift); | |
1092 | if (!new_storage) | |
1093 | return -ENOMEM; | |
14228c0d | 1094 | |
f47781d8 MP |
1095 | /* Must upgrade direct to indirect storage. */ |
1096 | if (!h->has_indirect) { | |
1097 | memcpy(new_storage, h->direct.storage, | |
1098 | old_n_buckets * (hi->entry_size + sizeof(dib_raw_t))); | |
1099 | h->indirect.n_entries = h->n_direct_entries; | |
1100 | h->indirect.idx_lowest_entry = 0; | |
1101 | h->n_direct_entries = 0; | |
1102 | } | |
14228c0d | 1103 | |
f47781d8 MP |
1104 | /* Get a new hash key. If we've just upgraded to indirect storage, |
1105 | * allow reusing a previously generated key. It's still a different key | |
1106 | * from the shared one that we used for direct storage. */ | |
1107 | get_hash_key(h->indirect.hash_key, !h->has_indirect); | |
1108 | ||
1109 | h->has_indirect = true; | |
1110 | h->indirect.storage = new_storage; | |
1111 | h->indirect.n_buckets = (1U << new_shift) / | |
1112 | (hi->entry_size + sizeof(dib_raw_t)); | |
1113 | ||
aa27b158 | 1114 | old_dibs = (dib_raw_t*)((uint8_t*) new_storage + hi->entry_size * old_n_buckets); |
f47781d8 MP |
1115 | new_dibs = dib_raw_ptr(h); |
1116 | ||
1117 | /* | |
1118 | * Move the DIB array to the new place, replacing valid DIB values with | |
1119 | * DIB_RAW_REHASH to indicate all of the used buckets need rehashing. | |
1120 | * Note: Overlap is not possible, because we have at least doubled the | |
1121 | * number of buckets and dib_raw_t is smaller than any entry type. | |
1122 | */ | |
1123 | for (idx = 0; idx < old_n_buckets; idx++) { | |
1124 | assert(old_dibs[idx] != DIB_RAW_REHASH); | |
1125 | new_dibs[idx] = old_dibs[idx] == DIB_RAW_FREE ? DIB_RAW_FREE | |
1126 | : DIB_RAW_REHASH; | |
14228c0d MB |
1127 | } |
1128 | ||
f47781d8 | 1129 | /* Zero the area of newly added entries (including the old DIB area) */ |
e735f4d4 | 1130 | memzero(bucket_at(h, old_n_buckets), |
f47781d8 | 1131 | (n_buckets(h) - old_n_buckets) * hi->entry_size); |
14228c0d | 1132 | |
f47781d8 MP |
1133 | /* The upper half of the new DIB array needs initialization */ |
1134 | memset(&new_dibs[old_n_buckets], DIB_RAW_INIT, | |
1135 | (n_buckets(h) - old_n_buckets) * sizeof(dib_raw_t)); | |
60f067b4 | 1136 | |
f47781d8 MP |
1137 | /* Rehash entries that need it */ |
1138 | n_rehashed = 0; | |
1139 | for (idx = 0; idx < old_n_buckets; idx++) { | |
1140 | if (new_dibs[idx] != DIB_RAW_REHASH) | |
1141 | continue; | |
14228c0d | 1142 | |
f47781d8 | 1143 | optimal_idx = bucket_hash(h, bucket_at(h, idx)->key); |
14228c0d | 1144 | |
f47781d8 MP |
1145 | /* |
1146 | * Not much to do if by luck the entry hashes to its current | |
1147 | * location. Just set its DIB. | |
1148 | */ | |
1149 | if (optimal_idx == idx) { | |
1150 | new_dibs[idx] = 0; | |
1151 | n_rehashed++; | |
1152 | continue; | |
1153 | } | |
1154 | ||
1155 | new_dibs[idx] = DIB_RAW_FREE; | |
1156 | bucket_move_entry(h, &swap, idx, IDX_PUT); | |
1157 | /* bucket_move_entry does not clear the source */ | |
e735f4d4 | 1158 | memzero(bucket_at(h, idx), hi->entry_size); |
f47781d8 MP |
1159 | |
1160 | do { | |
1161 | /* | |
1162 | * Find the new bucket for the current entry. This may make | |
1163 | * another entry homeless and load it into IDX_PUT. | |
1164 | */ | |
1165 | rehash_next = hashmap_put_robin_hood(h, optimal_idx, &swap); | |
1166 | n_rehashed++; | |
1167 | ||
1168 | /* Did the current entry displace another one? */ | |
1169 | if (rehash_next) | |
1170 | optimal_idx = bucket_hash(h, bucket_at_swap(&swap, IDX_PUT)->p.b.key); | |
1171 | } while (rehash_next); | |
1172 | } | |
663996b3 | 1173 | |
f47781d8 | 1174 | assert(n_rehashed == n_entries(h)); |
663996b3 | 1175 | |
f47781d8 MP |
1176 | return 1; |
1177 | } | |
14228c0d | 1178 | |
f47781d8 MP |
1179 | /* |
1180 | * Finds an entry with a matching key | |
1181 | * Returns: index of the found entry, or IDX_NIL if not found. | |
1182 | */ | |
1183 | static unsigned base_bucket_scan(HashmapBase *h, unsigned idx, const void *key) { | |
1184 | struct hashmap_base_entry *e; | |
1185 | unsigned dib, distance; | |
1186 | dib_raw_t *dibs = dib_raw_ptr(h); | |
663996b3 | 1187 | |
f47781d8 | 1188 | assert(idx < n_buckets(h)); |
663996b3 | 1189 | |
f47781d8 MP |
1190 | for (distance = 0; ; distance++) { |
1191 | if (dibs[idx] == DIB_RAW_FREE) | |
1192 | return IDX_NIL; | |
663996b3 | 1193 | |
f47781d8 | 1194 | dib = bucket_calculate_dib(h, idx, dibs[idx]); |
663996b3 | 1195 | |
f47781d8 MP |
1196 | if (dib < distance) |
1197 | return IDX_NIL; | |
1198 | if (dib == distance) { | |
1199 | e = bucket_at(h, idx); | |
1200 | if (h->hash_ops->compare(e->key, key) == 0) | |
1201 | return idx; | |
1202 | } | |
1203 | ||
1204 | idx = next_idx(h, idx); | |
1205 | } | |
663996b3 | 1206 | } |
f47781d8 | 1207 | #define bucket_scan(h, idx, key) base_bucket_scan(HASHMAP_BASE(h), idx, key) |
663996b3 | 1208 | |
5eef597e | 1209 | int hashmap_put(Hashmap *h, const void *key, void *value) { |
f47781d8 MP |
1210 | struct swap_entries swap; |
1211 | struct plain_hashmap_entry *e; | |
1212 | unsigned hash, idx; | |
5eef597e MP |
1213 | |
1214 | assert(h); | |
1215 | ||
1216 | hash = bucket_hash(h, key); | |
f47781d8 MP |
1217 | idx = bucket_scan(h, hash, key); |
1218 | if (idx != IDX_NIL) { | |
1219 | e = plain_bucket_at(h, idx); | |
5eef597e MP |
1220 | if (e->value == value) |
1221 | return 0; | |
1222 | return -EEXIST; | |
1223 | } | |
1224 | ||
f47781d8 MP |
1225 | e = &bucket_at_swap(&swap, IDX_PUT)->p; |
1226 | e->b.key = key; | |
1227 | e->value = value; | |
1228 | return hashmap_put_boldly(h, hash, &swap, true); | |
1229 | } | |
1230 | ||
1231 | int set_put(Set *s, const void *key) { | |
1232 | struct swap_entries swap; | |
1233 | struct hashmap_base_entry *e; | |
1234 | unsigned hash, idx; | |
1235 | ||
1236 | assert(s); | |
1237 | ||
1238 | hash = bucket_hash(s, key); | |
1239 | idx = bucket_scan(s, hash, key); | |
1240 | if (idx != IDX_NIL) | |
1241 | return 0; | |
1242 | ||
1243 | e = &bucket_at_swap(&swap, IDX_PUT)->p.b; | |
1244 | e->key = key; | |
1245 | return hashmap_put_boldly(s, hash, &swap, true); | |
5eef597e MP |
1246 | } |
1247 | ||
663996b3 | 1248 | int hashmap_replace(Hashmap *h, const void *key, void *value) { |
f47781d8 MP |
1249 | struct swap_entries swap; |
1250 | struct plain_hashmap_entry *e; | |
1251 | unsigned hash, idx; | |
663996b3 MS |
1252 | |
1253 | assert(h); | |
1254 | ||
14228c0d | 1255 | hash = bucket_hash(h, key); |
f47781d8 MP |
1256 | idx = bucket_scan(h, hash, key); |
1257 | if (idx != IDX_NIL) { | |
1258 | e = plain_bucket_at(h, idx); | |
f5e65279 | 1259 | #if ENABLE_DEBUG_HASHMAP |
f47781d8 MP |
1260 | /* Although the key is equal, the key pointer may have changed, |
1261 | * and this would break our assumption for iterating. So count | |
1262 | * this operation as incompatible with iteration. */ | |
1263 | if (e->b.key != key) { | |
1264 | h->b.debug.put_count++; | |
1265 | h->b.debug.rem_count++; | |
1266 | h->b.debug.last_rem_idx = idx; | |
1267 | } | |
1268 | #endif | |
1269 | e->b.key = key; | |
663996b3 | 1270 | e->value = value; |
98393f85 MB |
1271 | hashmap_set_dirty(h); |
1272 | ||
663996b3 MS |
1273 | return 0; |
1274 | } | |
1275 | ||
f47781d8 MP |
1276 | e = &bucket_at_swap(&swap, IDX_PUT)->p; |
1277 | e->b.key = key; | |
1278 | e->value = value; | |
1279 | return hashmap_put_boldly(h, hash, &swap, true); | |
663996b3 MS |
1280 | } |
1281 | ||
1282 | int hashmap_update(Hashmap *h, const void *key, void *value) { | |
f47781d8 MP |
1283 | struct plain_hashmap_entry *e; |
1284 | unsigned hash, idx; | |
663996b3 MS |
1285 | |
1286 | assert(h); | |
1287 | ||
14228c0d | 1288 | hash = bucket_hash(h, key); |
f47781d8 MP |
1289 | idx = bucket_scan(h, hash, key); |
1290 | if (idx == IDX_NIL) | |
663996b3 MS |
1291 | return -ENOENT; |
1292 | ||
f47781d8 | 1293 | e = plain_bucket_at(h, idx); |
663996b3 | 1294 | e->value = value; |
98393f85 MB |
1295 | hashmap_set_dirty(h); |
1296 | ||
663996b3 MS |
1297 | return 0; |
1298 | } | |
1299 | ||
f47781d8 MP |
1300 | void *internal_hashmap_get(HashmapBase *h, const void *key) { |
1301 | struct hashmap_base_entry *e; | |
1302 | unsigned hash, idx; | |
663996b3 MS |
1303 | |
1304 | if (!h) | |
1305 | return NULL; | |
1306 | ||
14228c0d | 1307 | hash = bucket_hash(h, key); |
f47781d8 MP |
1308 | idx = bucket_scan(h, hash, key); |
1309 | if (idx == IDX_NIL) | |
663996b3 MS |
1310 | return NULL; |
1311 | ||
f47781d8 MP |
1312 | e = bucket_at(h, idx); |
1313 | return entry_value(h, e); | |
663996b3 MS |
1314 | } |
1315 | ||
f47781d8 MP |
1316 | void *hashmap_get2(Hashmap *h, const void *key, void **key2) { |
1317 | struct plain_hashmap_entry *e; | |
1318 | unsigned hash, idx; | |
663996b3 MS |
1319 | |
1320 | if (!h) | |
1321 | return NULL; | |
1322 | ||
14228c0d | 1323 | hash = bucket_hash(h, key); |
f47781d8 MP |
1324 | idx = bucket_scan(h, hash, key); |
1325 | if (idx == IDX_NIL) | |
663996b3 MS |
1326 | return NULL; |
1327 | ||
f47781d8 | 1328 | e = plain_bucket_at(h, idx); |
663996b3 | 1329 | if (key2) |
f47781d8 | 1330 | *key2 = (void*) e->b.key; |
663996b3 MS |
1331 | |
1332 | return e->value; | |
1333 | } | |
1334 | ||
f47781d8 | 1335 | bool internal_hashmap_contains(HashmapBase *h, const void *key) { |
663996b3 MS |
1336 | unsigned hash; |
1337 | ||
1338 | if (!h) | |
1339 | return false; | |
1340 | ||
14228c0d | 1341 | hash = bucket_hash(h, key); |
f47781d8 | 1342 | return bucket_scan(h, hash, key) != IDX_NIL; |
663996b3 MS |
1343 | } |
1344 | ||
f47781d8 MP |
1345 | void *internal_hashmap_remove(HashmapBase *h, const void *key) { |
1346 | struct hashmap_base_entry *e; | |
1347 | unsigned hash, idx; | |
663996b3 MS |
1348 | void *data; |
1349 | ||
1350 | if (!h) | |
1351 | return NULL; | |
1352 | ||
14228c0d | 1353 | hash = bucket_hash(h, key); |
f47781d8 MP |
1354 | idx = bucket_scan(h, hash, key); |
1355 | if (idx == IDX_NIL) | |
663996b3 MS |
1356 | return NULL; |
1357 | ||
f47781d8 MP |
1358 | e = bucket_at(h, idx); |
1359 | data = entry_value(h, e); | |
1360 | remove_entry(h, idx); | |
663996b3 MS |
1361 | |
1362 | return data; | |
1363 | } | |
1364 | ||
f47781d8 MP |
1365 | void *hashmap_remove2(Hashmap *h, const void *key, void **rkey) { |
1366 | struct plain_hashmap_entry *e; | |
1367 | unsigned hash, idx; | |
60f067b4 JS |
1368 | void *data; |
1369 | ||
1370 | if (!h) { | |
1371 | if (rkey) | |
1372 | *rkey = NULL; | |
1373 | return NULL; | |
1374 | } | |
1375 | ||
1376 | hash = bucket_hash(h, key); | |
f47781d8 MP |
1377 | idx = bucket_scan(h, hash, key); |
1378 | if (idx == IDX_NIL) { | |
60f067b4 JS |
1379 | if (rkey) |
1380 | *rkey = NULL; | |
1381 | return NULL; | |
1382 | } | |
1383 | ||
f47781d8 | 1384 | e = plain_bucket_at(h, idx); |
60f067b4 JS |
1385 | data = e->value; |
1386 | if (rkey) | |
f47781d8 | 1387 | *rkey = (void*) e->b.key; |
60f067b4 | 1388 | |
f47781d8 | 1389 | remove_entry(h, idx); |
60f067b4 JS |
1390 | |
1391 | return data; | |
1392 | } | |
1393 | ||
663996b3 | 1394 | int hashmap_remove_and_put(Hashmap *h, const void *old_key, const void *new_key, void *value) { |
f47781d8 MP |
1395 | struct swap_entries swap; |
1396 | struct plain_hashmap_entry *e; | |
1397 | unsigned old_hash, new_hash, idx; | |
663996b3 MS |
1398 | |
1399 | if (!h) | |
1400 | return -ENOENT; | |
1401 | ||
14228c0d | 1402 | old_hash = bucket_hash(h, old_key); |
f47781d8 MP |
1403 | idx = bucket_scan(h, old_hash, old_key); |
1404 | if (idx == IDX_NIL) | |
663996b3 MS |
1405 | return -ENOENT; |
1406 | ||
14228c0d | 1407 | new_hash = bucket_hash(h, new_key); |
f47781d8 | 1408 | if (bucket_scan(h, new_hash, new_key) != IDX_NIL) |
663996b3 MS |
1409 | return -EEXIST; |
1410 | ||
f47781d8 | 1411 | remove_entry(h, idx); |
663996b3 | 1412 | |
f47781d8 MP |
1413 | e = &bucket_at_swap(&swap, IDX_PUT)->p; |
1414 | e->b.key = new_key; | |
663996b3 | 1415 | e->value = value; |
f47781d8 MP |
1416 | assert_se(hashmap_put_boldly(h, new_hash, &swap, false) == 1); |
1417 | ||
1418 | return 0; | |
1419 | } | |
663996b3 | 1420 | |
f47781d8 MP |
1421 | int set_remove_and_put(Set *s, const void *old_key, const void *new_key) { |
1422 | struct swap_entries swap; | |
1423 | struct hashmap_base_entry *e; | |
1424 | unsigned old_hash, new_hash, idx; | |
1425 | ||
1426 | if (!s) | |
1427 | return -ENOENT; | |
1428 | ||
1429 | old_hash = bucket_hash(s, old_key); | |
1430 | idx = bucket_scan(s, old_hash, old_key); | |
1431 | if (idx == IDX_NIL) | |
1432 | return -ENOENT; | |
1433 | ||
1434 | new_hash = bucket_hash(s, new_key); | |
1435 | if (bucket_scan(s, new_hash, new_key) != IDX_NIL) | |
1436 | return -EEXIST; | |
1437 | ||
1438 | remove_entry(s, idx); | |
1439 | ||
1440 | e = &bucket_at_swap(&swap, IDX_PUT)->p.b; | |
1441 | e->key = new_key; | |
1442 | assert_se(hashmap_put_boldly(s, new_hash, &swap, false) == 1); | |
663996b3 MS |
1443 | |
1444 | return 0; | |
1445 | } | |
1446 | ||
1447 | int hashmap_remove_and_replace(Hashmap *h, const void *old_key, const void *new_key, void *value) { | |
f47781d8 MP |
1448 | struct swap_entries swap; |
1449 | struct plain_hashmap_entry *e; | |
1450 | unsigned old_hash, new_hash, idx_old, idx_new; | |
663996b3 MS |
1451 | |
1452 | if (!h) | |
1453 | return -ENOENT; | |
1454 | ||
14228c0d | 1455 | old_hash = bucket_hash(h, old_key); |
f47781d8 MP |
1456 | idx_old = bucket_scan(h, old_hash, old_key); |
1457 | if (idx_old == IDX_NIL) | |
663996b3 MS |
1458 | return -ENOENT; |
1459 | ||
f47781d8 | 1460 | old_key = bucket_at(HASHMAP_BASE(h), idx_old)->key; |
663996b3 | 1461 | |
f47781d8 MP |
1462 | new_hash = bucket_hash(h, new_key); |
1463 | idx_new = bucket_scan(h, new_hash, new_key); | |
1464 | if (idx_new != IDX_NIL) | |
1465 | if (idx_old != idx_new) { | |
1466 | remove_entry(h, idx_new); | |
1467 | /* Compensate for a possible backward shift. */ | |
1468 | if (old_key != bucket_at(HASHMAP_BASE(h), idx_old)->key) | |
1469 | idx_old = prev_idx(HASHMAP_BASE(h), idx_old); | |
1470 | assert(old_key == bucket_at(HASHMAP_BASE(h), idx_old)->key); | |
1471 | } | |
1472 | ||
1473 | remove_entry(h, idx_old); | |
1474 | ||
1475 | e = &bucket_at_swap(&swap, IDX_PUT)->p; | |
1476 | e->b.key = new_key; | |
663996b3 | 1477 | e->value = value; |
f47781d8 | 1478 | assert_se(hashmap_put_boldly(h, new_hash, &swap, false) == 1); |
663996b3 MS |
1479 | |
1480 | return 0; | |
1481 | } | |
1482 | ||
6e866b33 MB |
1483 | void *internal_hashmap_remove_value(HashmapBase *h, const void *key, void *value) { |
1484 | struct hashmap_base_entry *e; | |
f47781d8 | 1485 | unsigned hash, idx; |
663996b3 MS |
1486 | |
1487 | if (!h) | |
1488 | return NULL; | |
1489 | ||
14228c0d | 1490 | hash = bucket_hash(h, key); |
f47781d8 MP |
1491 | idx = bucket_scan(h, hash, key); |
1492 | if (idx == IDX_NIL) | |
663996b3 MS |
1493 | return NULL; |
1494 | ||
6e866b33 MB |
1495 | e = bucket_at(h, idx); |
1496 | if (entry_value(h, e) != value) | |
663996b3 MS |
1497 | return NULL; |
1498 | ||
f47781d8 | 1499 | remove_entry(h, idx); |
663996b3 MS |
1500 | |
1501 | return value; | |
1502 | } | |
1503 | ||
f47781d8 MP |
1504 | static unsigned find_first_entry(HashmapBase *h) { |
1505 | Iterator i = ITERATOR_FIRST; | |
663996b3 | 1506 | |
f47781d8 MP |
1507 | if (!h || !n_entries(h)) |
1508 | return IDX_NIL; | |
663996b3 | 1509 | |
f47781d8 | 1510 | return hashmap_iterate_entry(h, &i); |
663996b3 MS |
1511 | } |
1512 | ||
6e866b33 | 1513 | void *internal_hashmap_first_key_and_value(HashmapBase *h, bool remove, void **ret_key) { |
f47781d8 | 1514 | struct hashmap_base_entry *e; |
6e866b33 | 1515 | void *key, *data; |
f47781d8 | 1516 | unsigned idx; |
663996b3 | 1517 | |
f47781d8 | 1518 | idx = find_first_entry(h); |
7c20daf6 FS |
1519 | if (idx == IDX_NIL) { |
1520 | if (ret_key) | |
1521 | *ret_key = NULL; | |
663996b3 | 1522 | return NULL; |
7c20daf6 | 1523 | } |
663996b3 | 1524 | |
f47781d8 | 1525 | e = bucket_at(h, idx); |
6e866b33 | 1526 | key = (void*) e->key; |
f47781d8 | 1527 | data = entry_value(h, e); |
663996b3 | 1528 | |
6e866b33 MB |
1529 | if (remove) |
1530 | remove_entry(h, idx); | |
663996b3 | 1531 | |
6e866b33 MB |
1532 | if (ret_key) |
1533 | *ret_key = key; | |
663996b3 | 1534 | |
6e866b33 | 1535 | return data; |
663996b3 MS |
1536 | } |
1537 | ||
f47781d8 | 1538 | unsigned internal_hashmap_size(HashmapBase *h) { |
663996b3 MS |
1539 | if (!h) |
1540 | return 0; | |
1541 | ||
f47781d8 | 1542 | return n_entries(h); |
663996b3 MS |
1543 | } |
1544 | ||
f47781d8 | 1545 | unsigned internal_hashmap_buckets(HashmapBase *h) { |
14228c0d MB |
1546 | if (!h) |
1547 | return 0; | |
1548 | ||
f47781d8 | 1549 | return n_buckets(h); |
14228c0d MB |
1550 | } |
1551 | ||
f47781d8 MP |
1552 | int internal_hashmap_merge(Hashmap *h, Hashmap *other) { |
1553 | Iterator i; | |
1554 | unsigned idx; | |
663996b3 | 1555 | |
f47781d8 | 1556 | assert(h); |
663996b3 | 1557 | |
f47781d8 MP |
1558 | HASHMAP_FOREACH_IDX(idx, HASHMAP_BASE(other), i) { |
1559 | struct plain_hashmap_entry *pe = plain_bucket_at(other, idx); | |
1560 | int r; | |
663996b3 | 1561 | |
f47781d8 MP |
1562 | r = hashmap_put(h, pe->b.key, pe->value); |
1563 | if (r < 0 && r != -EEXIST) | |
1564 | return r; | |
1565 | } | |
663996b3 | 1566 | |
f47781d8 MP |
1567 | return 0; |
1568 | } | |
663996b3 | 1569 | |
f47781d8 MP |
1570 | int set_merge(Set *s, Set *other) { |
1571 | Iterator i; | |
1572 | unsigned idx; | |
1573 | ||
1574 | assert(s); | |
663996b3 | 1575 | |
f47781d8 MP |
1576 | HASHMAP_FOREACH_IDX(idx, HASHMAP_BASE(other), i) { |
1577 | struct set_entry *se = set_bucket_at(other, idx); | |
663996b3 MS |
1578 | int r; |
1579 | ||
f47781d8 MP |
1580 | r = set_put(s, se->b.key); |
1581 | if (r < 0) | |
14228c0d | 1582 | return r; |
663996b3 MS |
1583 | } |
1584 | ||
1585 | return 0; | |
1586 | } | |
1587 | ||
f47781d8 | 1588 | int internal_hashmap_reserve(HashmapBase *h, unsigned entries_add) { |
5eef597e MP |
1589 | int r; |
1590 | ||
1591 | assert(h); | |
1592 | ||
1593 | r = resize_buckets(h, entries_add); | |
1594 | if (r < 0) | |
1595 | return r; | |
1596 | ||
1597 | return 0; | |
1598 | } | |
1599 | ||
f47781d8 MP |
1600 | /* |
1601 | * The same as hashmap_merge(), but every new item from other is moved to h. | |
1602 | * Keys already in h are skipped and stay in other. | |
1603 | * Returns: 0 on success. | |
1604 | * -ENOMEM on alloc failure, in which case no move has been done. | |
1605 | */ | |
1606 | int internal_hashmap_move(HashmapBase *h, HashmapBase *other) { | |
1607 | struct swap_entries swap; | |
1608 | struct hashmap_base_entry *e, *n; | |
1609 | Iterator i; | |
1610 | unsigned idx; | |
1611 | int r; | |
663996b3 MS |
1612 | |
1613 | assert(h); | |
1614 | ||
663996b3 | 1615 | if (!other) |
5eef597e | 1616 | return 0; |
663996b3 | 1617 | |
f47781d8 MP |
1618 | assert(other->type == h->type); |
1619 | ||
1620 | /* | |
1621 | * This reserves buckets for the worst case, where none of other's | |
1622 | * entries are yet present in h. This is preferable to risking | |
1623 | * an allocation failure in the middle of the moving and having to | |
1624 | * rollback or return a partial result. | |
1625 | */ | |
1626 | r = resize_buckets(h, n_entries(other)); | |
1627 | if (r < 0) | |
1628 | return r; | |
663996b3 | 1629 | |
f47781d8 MP |
1630 | HASHMAP_FOREACH_IDX(idx, other, i) { |
1631 | unsigned h_hash; | |
663996b3 | 1632 | |
f47781d8 | 1633 | e = bucket_at(other, idx); |
14228c0d | 1634 | h_hash = bucket_hash(h, e->key); |
f47781d8 | 1635 | if (bucket_scan(h, h_hash, e->key) != IDX_NIL) |
663996b3 MS |
1636 | continue; |
1637 | ||
f47781d8 MP |
1638 | n = &bucket_at_swap(&swap, IDX_PUT)->p.b; |
1639 | n->key = e->key; | |
1640 | if (h->type != HASHMAP_TYPE_SET) | |
1641 | ((struct plain_hashmap_entry*) n)->value = | |
1642 | ((struct plain_hashmap_entry*) e)->value; | |
1643 | assert_se(hashmap_put_boldly(h, h_hash, &swap, false) == 1); | |
1644 | ||
1645 | remove_entry(other, idx); | |
663996b3 | 1646 | } |
5eef597e MP |
1647 | |
1648 | return 0; | |
663996b3 MS |
1649 | } |
1650 | ||
f47781d8 MP |
1651 | int internal_hashmap_move_one(HashmapBase *h, HashmapBase *other, const void *key) { |
1652 | struct swap_entries swap; | |
1653 | unsigned h_hash, other_hash, idx; | |
1654 | struct hashmap_base_entry *e, *n; | |
1655 | int r; | |
663996b3 | 1656 | |
663996b3 MS |
1657 | assert(h); |
1658 | ||
14228c0d | 1659 | h_hash = bucket_hash(h, key); |
f47781d8 | 1660 | if (bucket_scan(h, h_hash, key) != IDX_NIL) |
663996b3 MS |
1661 | return -EEXIST; |
1662 | ||
5eef597e MP |
1663 | if (!other) |
1664 | return -ENOENT; | |
1665 | ||
f47781d8 MP |
1666 | assert(other->type == h->type); |
1667 | ||
14228c0d | 1668 | other_hash = bucket_hash(other, key); |
f47781d8 MP |
1669 | idx = bucket_scan(other, other_hash, key); |
1670 | if (idx == IDX_NIL) | |
663996b3 MS |
1671 | return -ENOENT; |
1672 | ||
f47781d8 MP |
1673 | e = bucket_at(other, idx); |
1674 | ||
1675 | n = &bucket_at_swap(&swap, IDX_PUT)->p.b; | |
1676 | n->key = e->key; | |
1677 | if (h->type != HASHMAP_TYPE_SET) | |
1678 | ((struct plain_hashmap_entry*) n)->value = | |
1679 | ((struct plain_hashmap_entry*) e)->value; | |
1680 | r = hashmap_put_boldly(h, h_hash, &swap, true); | |
1681 | if (r < 0) | |
1682 | return r; | |
663996b3 | 1683 | |
f47781d8 | 1684 | remove_entry(other, idx); |
663996b3 MS |
1685 | return 0; |
1686 | } | |
1687 | ||
f47781d8 MP |
1688 | HashmapBase *internal_hashmap_copy(HashmapBase *h) { |
1689 | HashmapBase *copy; | |
1690 | int r; | |
663996b3 MS |
1691 | |
1692 | assert(h); | |
1693 | ||
f47781d8 | 1694 | copy = hashmap_base_new(h->hash_ops, h->type HASHMAP_DEBUG_SRC_ARGS); |
14228c0d | 1695 | if (!copy) |
663996b3 MS |
1696 | return NULL; |
1697 | ||
f47781d8 MP |
1698 | switch (h->type) { |
1699 | case HASHMAP_TYPE_PLAIN: | |
1700 | case HASHMAP_TYPE_ORDERED: | |
1701 | r = hashmap_merge((Hashmap*)copy, (Hashmap*)h); | |
1702 | break; | |
1703 | case HASHMAP_TYPE_SET: | |
1704 | r = set_merge((Set*)copy, (Set*)h); | |
1705 | break; | |
1706 | default: | |
1707 | assert_not_reached("Unknown hashmap type"); | |
1708 | } | |
1709 | ||
1710 | if (r < 0) { | |
6e866b33 | 1711 | internal_hashmap_free(copy, false, false); |
663996b3 MS |
1712 | return NULL; |
1713 | } | |
1714 | ||
1715 | return copy; | |
1716 | } | |
1717 | ||
f47781d8 | 1718 | char **internal_hashmap_get_strv(HashmapBase *h) { |
663996b3 | 1719 | char **sv; |
f47781d8 MP |
1720 | Iterator i; |
1721 | unsigned idx, n; | |
663996b3 | 1722 | |
f47781d8 | 1723 | sv = new(char*, n_entries(h)+1); |
663996b3 MS |
1724 | if (!sv) |
1725 | return NULL; | |
1726 | ||
1727 | n = 0; | |
f47781d8 MP |
1728 | HASHMAP_FOREACH_IDX(idx, h, i) |
1729 | sv[n++] = entry_value(h, bucket_at(h, idx)); | |
663996b3 MS |
1730 | sv[n] = NULL; |
1731 | ||
1732 | return sv; | |
1733 | } | |
1734 | ||
f47781d8 MP |
1735 | void *ordered_hashmap_next(OrderedHashmap *h, const void *key) { |
1736 | struct ordered_hashmap_entry *e; | |
1737 | unsigned hash, idx; | |
663996b3 | 1738 | |
663996b3 MS |
1739 | if (!h) |
1740 | return NULL; | |
1741 | ||
14228c0d | 1742 | hash = bucket_hash(h, key); |
f47781d8 MP |
1743 | idx = bucket_scan(h, hash, key); |
1744 | if (idx == IDX_NIL) | |
663996b3 MS |
1745 | return NULL; |
1746 | ||
f47781d8 MP |
1747 | e = ordered_bucket_at(h, idx); |
1748 | if (e->iterate_next == IDX_NIL) | |
663996b3 | 1749 | return NULL; |
f47781d8 MP |
1750 | return ordered_bucket_at(h, e->iterate_next)->p.value; |
1751 | } | |
663996b3 | 1752 | |
f47781d8 MP |
1753 | int set_consume(Set *s, void *value) { |
1754 | int r; | |
1755 | ||
5a920b42 MP |
1756 | assert(s); |
1757 | assert(value); | |
1758 | ||
f47781d8 MP |
1759 | r = set_put(s, value); |
1760 | if (r <= 0) | |
1761 | free(value); | |
1762 | ||
1763 | return r; | |
1764 | } | |
1765 | ||
1766 | int set_put_strdup(Set *s, const char *p) { | |
1767 | char *c; | |
f47781d8 MP |
1768 | |
1769 | assert(s); | |
1770 | assert(p); | |
1771 | ||
aa27b158 MP |
1772 | if (set_contains(s, (char*) p)) |
1773 | return 0; | |
1774 | ||
f47781d8 MP |
1775 | c = strdup(p); |
1776 | if (!c) | |
1777 | return -ENOMEM; | |
1778 | ||
aa27b158 | 1779 | return set_consume(s, c); |
f47781d8 MP |
1780 | } |
1781 | ||
1782 | int set_put_strdupv(Set *s, char **l) { | |
1783 | int n = 0, r; | |
1784 | char **i; | |
1785 | ||
5a920b42 MP |
1786 | assert(s); |
1787 | ||
f47781d8 MP |
1788 | STRV_FOREACH(i, l) { |
1789 | r = set_put_strdup(s, *i); | |
1790 | if (r < 0) | |
1791 | return r; | |
1792 | ||
1793 | n += r; | |
1794 | } | |
1795 | ||
1796 | return n; | |
663996b3 | 1797 | } |
5a920b42 MP |
1798 | |
1799 | int set_put_strsplit(Set *s, const char *v, const char *separators, ExtractFlags flags) { | |
1800 | const char *p = v; | |
1801 | int r; | |
1802 | ||
1803 | assert(s); | |
1804 | assert(v); | |
1805 | ||
1806 | for (;;) { | |
1807 | char *word; | |
1808 | ||
1809 | r = extract_first_word(&p, &word, separators, flags); | |
1810 | if (r <= 0) | |
1811 | return r; | |
1812 | ||
1813 | r = set_consume(s, word); | |
1814 | if (r < 0) | |
1815 | return r; | |
1816 | } | |
1817 | } | |
98393f85 MB |
1818 | |
1819 | /* expand the cachemem if needed, return true if newly (re)activated. */ | |
1820 | static int cachemem_maintain(CacheMem *mem, unsigned size) { | |
1821 | assert(mem); | |
1822 | ||
1823 | if (!GREEDY_REALLOC(mem->ptr, mem->n_allocated, size)) { | |
1824 | if (size > 0) | |
1825 | return -ENOMEM; | |
1826 | } | |
1827 | ||
1828 | if (!mem->active) { | |
1829 | mem->active = true; | |
1830 | return true; | |
1831 | } | |
1832 | ||
1833 | return false; | |
1834 | } | |
1835 | ||
1836 | int iterated_cache_get(IteratedCache *cache, const void ***res_keys, const void ***res_values, unsigned *res_n_entries) { | |
1837 | bool sync_keys = false, sync_values = false; | |
1838 | unsigned size; | |
1839 | int r; | |
1840 | ||
1841 | assert(cache); | |
1842 | assert(cache->hashmap); | |
1843 | ||
1844 | size = n_entries(cache->hashmap); | |
1845 | ||
1846 | if (res_keys) { | |
1847 | r = cachemem_maintain(&cache->keys, size); | |
1848 | if (r < 0) | |
1849 | return r; | |
1850 | ||
1851 | sync_keys = r; | |
1852 | } else | |
1853 | cache->keys.active = false; | |
1854 | ||
1855 | if (res_values) { | |
1856 | r = cachemem_maintain(&cache->values, size); | |
1857 | if (r < 0) | |
1858 | return r; | |
1859 | ||
1860 | sync_values = r; | |
1861 | } else | |
1862 | cache->values.active = false; | |
1863 | ||
1864 | if (cache->hashmap->dirty) { | |
1865 | if (cache->keys.active) | |
1866 | sync_keys = true; | |
1867 | if (cache->values.active) | |
1868 | sync_values = true; | |
1869 | ||
1870 | cache->hashmap->dirty = false; | |
1871 | } | |
1872 | ||
1873 | if (sync_keys || sync_values) { | |
1874 | unsigned i, idx; | |
1875 | Iterator iter; | |
1876 | ||
1877 | i = 0; | |
1878 | HASHMAP_FOREACH_IDX(idx, cache->hashmap, iter) { | |
1879 | struct hashmap_base_entry *e; | |
1880 | ||
1881 | e = bucket_at(cache->hashmap, idx); | |
1882 | ||
1883 | if (sync_keys) | |
1884 | cache->keys.ptr[i] = e->key; | |
1885 | if (sync_values) | |
1886 | cache->values.ptr[i] = entry_value(cache->hashmap, e); | |
1887 | i++; | |
1888 | } | |
1889 | } | |
1890 | ||
1891 | if (res_keys) | |
1892 | *res_keys = cache->keys.ptr; | |
1893 | if (res_values) | |
1894 | *res_values = cache->values.ptr; | |
1895 | if (res_n_entries) | |
1896 | *res_n_entries = size; | |
1897 | ||
1898 | return 0; | |
1899 | } | |
1900 | ||
1901 | IteratedCache *iterated_cache_free(IteratedCache *cache) { | |
1902 | if (cache) { | |
1903 | free(cache->keys.ptr); | |
1904 | free(cache->values.ptr); | |
98393f85 MB |
1905 | } |
1906 | ||
bb4f798a | 1907 | return mfree(cache); |
98393f85 | 1908 | } |