]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
a7334b09 | 2 | |
60918275 | 3 | #include <errno.h> |
11c3a366 | 4 | #include <stdint.h> |
d4510856 | 5 | #include <stdlib.h> |
60918275 | 6 | |
b5efdb8a | 7 | #include "alloc-util.h" |
556c7bae | 8 | #include "fileio.h" |
b4f60743 | 9 | #include "hashmap.h" |
60918275 | 10 | #include "macro.h" |
0a970718 | 11 | #include "memory-util.h" |
b3dcf58e | 12 | #include "mempool.h" |
f5947a5e | 13 | #include "missing_syscall.h" |
d4510856 | 14 | #include "process-util.h" |
3df3e884 | 15 | #include "random-util.h" |
d4510856 LP |
16 | #include "set.h" |
17 | #include "siphash24.h" | |
556c7bae | 18 | #include "string-util.h" |
d4510856 | 19 | #include "strv.h" |
60918275 | 20 | |
349cc4a5 | 21 | #if ENABLE_DEBUG_HASHMAP |
3d4db144 | 22 | #include <pthread.h> |
2eec67ac TA |
23 | #include "list.h" |
24 | #endif | |
25 | ||
89439d4f MS |
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 { | |
60918275 | 80 | const void *key; |
89439d4f MS |
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; | |
60918275 | 88 | void *value; |
60918275 LP |
89 | }; |
90 | ||
89439d4f MS |
91 | struct ordered_hashmap_entry { |
92 | struct plain_hashmap_entry p; | |
93 | unsigned iterate_next, iterate_previous; | |
94 | }; | |
60918275 | 95 | |
89439d4f MS |
96 | struct set_entry { |
97 | struct hashmap_base_entry b; | |
98 | }; | |
45fa9e29 | 99 | |
89439d4f MS |
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) | |
39c2a6f1 | 107 | |
89439d4f MS |
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]; | |
60918275 LP |
118 | }; |
119 | ||
89439d4f MS |
120 | /* Distance from Initial Bucket */ |
121 | typedef uint8_t dib_raw_t; | |
3ef11dcf ZJS |
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 */ | |
89439d4f MS |
126 | |
127 | #define DIB_FREE UINT_MAX | |
128 | ||
349cc4a5 | 129 | #if ENABLE_DEBUG_HASHMAP |
89439d4f MS |
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 */ | |
39c2a6f1 LP |
143 | }; |
144 | ||
89439d4f MS |
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); | |
4f1b3061 | 147 | static pthread_mutex_t hashmap_debug_list_mutex = PTHREAD_MUTEX_INITIALIZER; |
55825de5 | 148 | #endif |
39c2a6f1 | 149 | |
89439d4f MS |
150 | enum HashmapType { |
151 | HASHMAP_TYPE_PLAIN, | |
152 | HASHMAP_TYPE_ORDERED, | |
153 | HASHMAP_TYPE_SET, | |
154 | _HASHMAP_TYPE_MAX | |
155 | }; | |
39c2a6f1 | 156 | |
89439d4f | 157 | struct _packed_ indirect_storage { |
1a39bc8c | 158 | void *storage; /* where buckets and DIBs are stored */ |
89439d4f MS |
159 | uint8_t hash_key[HASH_KEY_SIZE]; /* hash key; changes during resize */ |
160 | ||
161 | unsigned n_entries; /* number of stored entries */ | |
162 | unsigned n_buckets; /* number of buckets */ | |
163 | ||
164 | unsigned idx_lowest_entry; /* Index below which all buckets are free. | |
165 | Makes "while(hashmap_steal_first())" loops | |
166 | O(n) instead of O(n^2) for unordered hashmaps. */ | |
167 | uint8_t _pad[3]; /* padding for the whole HashmapBase */ | |
168 | /* The bitfields in HashmapBase complete the alignment of the whole thing. */ | |
169 | }; | |
170 | ||
171 | struct direct_storage { | |
172 | /* This gives us 39 bytes on 64bit, or 35 bytes on 32bit. | |
173 | * That's room for 4 set_entries + 4 DIB bytes + 3 unused bytes on 64bit, | |
174 | * or 7 set_entries + 7 DIB bytes + 0 unused bytes on 32bit. */ | |
1a39bc8c | 175 | uint8_t storage[sizeof(struct indirect_storage)]; |
89439d4f MS |
176 | }; |
177 | ||
178 | #define DIRECT_BUCKETS(entry_t) \ | |
179 | (sizeof(struct direct_storage) / (sizeof(entry_t) + sizeof(dib_raw_t))) | |
180 | ||
181 | /* We should be able to store at least one entry directly. */ | |
182 | assert_cc(DIRECT_BUCKETS(struct ordered_hashmap_entry) >= 1); | |
183 | ||
184 | /* We have 3 bits for n_direct_entries. */ | |
185 | assert_cc(DIRECT_BUCKETS(struct set_entry) < (1 << 3)); | |
186 | ||
187 | /* Hashmaps with directly stored entries all use this shared hash key. | |
188 | * It's no big deal if the key is guessed, because there can be only | |
189 | * a handful of directly stored entries in a hashmap. When a hashmap | |
190 | * outgrows direct storage, it gets its own key for indirect storage. */ | |
191 | static uint8_t shared_hash_key[HASH_KEY_SIZE]; | |
192 | static bool shared_hash_key_initialized; | |
193 | ||
194 | /* Fields that all hashmap/set types must have */ | |
195 | struct HashmapBase { | |
196 | const struct hash_ops *hash_ops; /* hash and compare ops to use */ | |
197 | ||
198 | union _packed_ { | |
199 | struct indirect_storage indirect; /* if has_indirect */ | |
200 | struct direct_storage direct; /* if !has_indirect */ | |
201 | }; | |
202 | ||
203 | enum HashmapType type:2; /* HASHMAP_TYPE_* */ | |
204 | bool has_indirect:1; /* whether indirect storage is used */ | |
205 | unsigned n_direct_entries:3; /* Number of entries in direct storage. | |
206 | * Only valid if !has_indirect. */ | |
207 | bool from_pool:1; /* whether was allocated from mempool */ | |
45ea84d8 VC |
208 | bool dirty:1; /* whether dirtied since last iterated_cache_get() */ |
209 | bool cached:1; /* whether this hashmap is being cached */ | |
55825de5 ZJS |
210 | |
211 | #if ENABLE_DEBUG_HASHMAP | |
212 | struct hashmap_debug_info debug; | |
213 | #endif | |
89439d4f MS |
214 | }; |
215 | ||
216 | /* Specific hash types | |
217 | * HashmapBase must be at the beginning of each hashmap struct. */ | |
218 | ||
219 | struct Hashmap { | |
220 | struct HashmapBase b; | |
221 | }; | |
222 | ||
223 | struct OrderedHashmap { | |
224 | struct HashmapBase b; | |
225 | unsigned iterate_list_head, iterate_list_tail; | |
226 | }; | |
227 | ||
228 | struct Set { | |
229 | struct HashmapBase b; | |
230 | }; | |
231 | ||
45ea84d8 VC |
232 | typedef struct CacheMem { |
233 | const void **ptr; | |
234 | size_t n_populated, n_allocated; | |
235 | bool active:1; | |
236 | } CacheMem; | |
237 | ||
238 | struct IteratedCache { | |
239 | HashmapBase *hashmap; | |
240 | CacheMem keys, values; | |
241 | }; | |
242 | ||
89439d4f MS |
243 | DEFINE_MEMPOOL(hashmap_pool, Hashmap, 8); |
244 | DEFINE_MEMPOOL(ordered_hashmap_pool, OrderedHashmap, 8); | |
245 | /* No need for a separate Set pool */ | |
246 | assert_cc(sizeof(Hashmap) == sizeof(Set)); | |
247 | ||
248 | struct hashmap_type_info { | |
249 | size_t head_size; | |
250 | size_t entry_size; | |
251 | struct mempool *mempool; | |
252 | unsigned n_direct_buckets; | |
253 | }; | |
254 | ||
43874aa7 | 255 | static _used_ const struct hashmap_type_info hashmap_type_info[_HASHMAP_TYPE_MAX] = { |
89439d4f MS |
256 | [HASHMAP_TYPE_PLAIN] = { |
257 | .head_size = sizeof(Hashmap), | |
258 | .entry_size = sizeof(struct plain_hashmap_entry), | |
259 | .mempool = &hashmap_pool, | |
260 | .n_direct_buckets = DIRECT_BUCKETS(struct plain_hashmap_entry), | |
261 | }, | |
262 | [HASHMAP_TYPE_ORDERED] = { | |
263 | .head_size = sizeof(OrderedHashmap), | |
264 | .entry_size = sizeof(struct ordered_hashmap_entry), | |
265 | .mempool = &ordered_hashmap_pool, | |
266 | .n_direct_buckets = DIRECT_BUCKETS(struct ordered_hashmap_entry), | |
267 | }, | |
268 | [HASHMAP_TYPE_SET] = { | |
269 | .head_size = sizeof(Set), | |
270 | .entry_size = sizeof(struct set_entry), | |
271 | .mempool = &hashmap_pool, | |
272 | .n_direct_buckets = DIRECT_BUCKETS(struct set_entry), | |
273 | }, | |
274 | }; | |
39c2a6f1 | 275 | |
d18cb393 | 276 | #if VALGRIND |
d34dae18 | 277 | _destructor_ static void cleanup_pools(void) { |
556c7bae ZJS |
278 | _cleanup_free_ char *t = NULL; |
279 | int r; | |
280 | ||
281 | /* Be nice to valgrind */ | |
282 | ||
283 | /* The pool is only allocated by the main thread, but the memory can | |
284 | * be passed to other threads. Let's clean up if we are the main thread | |
285 | * and no other threads are live. */ | |
31c9d74d FS |
286 | /* We build our own is_main_thread() here, which doesn't use C11 |
287 | * TLS based caching of the result. That's because valgrind apparently | |
288 | * doesn't like malloc() (which C11 TLS internally uses) to be called | |
289 | * from a GCC destructors. */ | |
290 | if (getpid() != gettid()) | |
556c7bae ZJS |
291 | return; |
292 | ||
293 | r = get_proc_field("/proc/self/status", "Threads", WHITESPACE, &t); | |
294 | if (r < 0 || !streq(t, "1")) | |
295 | return; | |
296 | ||
297 | mempool_drop(&hashmap_pool); | |
298 | mempool_drop(&ordered_hashmap_pool); | |
299 | } | |
300 | #endif | |
301 | ||
89439d4f MS |
302 | static unsigned n_buckets(HashmapBase *h) { |
303 | return h->has_indirect ? h->indirect.n_buckets | |
304 | : hashmap_type_info[h->type].n_direct_buckets; | |
305 | } | |
306 | ||
307 | static unsigned n_entries(HashmapBase *h) { | |
308 | return h->has_indirect ? h->indirect.n_entries | |
309 | : h->n_direct_entries; | |
310 | } | |
311 | ||
312 | static void n_entries_inc(HashmapBase *h) { | |
313 | if (h->has_indirect) | |
314 | h->indirect.n_entries++; | |
315 | else | |
316 | h->n_direct_entries++; | |
317 | } | |
318 | ||
319 | static void n_entries_dec(HashmapBase *h) { | |
320 | if (h->has_indirect) | |
321 | h->indirect.n_entries--; | |
322 | else | |
323 | h->n_direct_entries--; | |
324 | } | |
325 | ||
1a39bc8c | 326 | static void *storage_ptr(HashmapBase *h) { |
89439d4f MS |
327 | return h->has_indirect ? h->indirect.storage |
328 | : h->direct.storage; | |
329 | } | |
330 | ||
331 | static uint8_t *hash_key(HashmapBase *h) { | |
332 | return h->has_indirect ? h->indirect.hash_key | |
333 | : shared_hash_key; | |
334 | } | |
335 | ||
336 | static unsigned base_bucket_hash(HashmapBase *h, const void *p) { | |
b826ab58 | 337 | struct siphash state; |
0cb3c286 | 338 | uint64_t hash; |
b826ab58 | 339 | |
0cb3c286 | 340 | siphash24_init(&state, hash_key(h)); |
b826ab58 TG |
341 | |
342 | h->hash_ops->hash(p, &state); | |
343 | ||
933f9cae | 344 | hash = siphash24_finalize(&state); |
0cb3c286 TG |
345 | |
346 | return (unsigned) (hash % n_buckets(h)); | |
9bf3b535 | 347 | } |
89439d4f | 348 | #define bucket_hash(h, p) base_bucket_hash(HASHMAP_BASE(h), p) |
9bf3b535 | 349 | |
a1e92eee | 350 | static void base_set_dirty(HashmapBase *h) { |
84dcca75 VC |
351 | h->dirty = true; |
352 | } | |
353 | #define hashmap_set_dirty(h) base_set_dirty(HASHMAP_BASE(h)) | |
354 | ||
9bf3b535 LP |
355 | static void get_hash_key(uint8_t hash_key[HASH_KEY_SIZE], bool reuse_is_ok) { |
356 | static uint8_t current[HASH_KEY_SIZE]; | |
357 | static bool current_initialized = false; | |
358 | ||
359 | /* Returns a hash function key to use. In order to keep things | |
360 | * fast we will not generate a new key each time we allocate a | |
361 | * new hash table. Instead, we'll just reuse the most recently | |
362 | * generated one, except if we never generated one or when we | |
363 | * are rehashing an entire hash table because we reached a | |
364 | * fill level */ | |
365 | ||
366 | if (!current_initialized || !reuse_is_ok) { | |
367 | random_bytes(current, sizeof(current)); | |
368 | current_initialized = true; | |
369 | } | |
370 | ||
371 | memcpy(hash_key, current, sizeof(current)); | |
a3b6fafe LP |
372 | } |
373 | ||
89439d4f MS |
374 | static struct hashmap_base_entry *bucket_at(HashmapBase *h, unsigned idx) { |
375 | return (struct hashmap_base_entry*) | |
1a39bc8c | 376 | ((uint8_t*) storage_ptr(h) + idx * hashmap_type_info[h->type].entry_size); |
89439d4f MS |
377 | } |
378 | ||
379 | static struct plain_hashmap_entry *plain_bucket_at(Hashmap *h, unsigned idx) { | |
380 | return (struct plain_hashmap_entry*) bucket_at(HASHMAP_BASE(h), idx); | |
381 | } | |
382 | ||
383 | static struct ordered_hashmap_entry *ordered_bucket_at(OrderedHashmap *h, unsigned idx) { | |
384 | return (struct ordered_hashmap_entry*) bucket_at(HASHMAP_BASE(h), idx); | |
385 | } | |
39c2a6f1 | 386 | |
89439d4f MS |
387 | static struct set_entry *set_bucket_at(Set *h, unsigned idx) { |
388 | return (struct set_entry*) bucket_at(HASHMAP_BASE(h), idx); | |
389 | } | |
39c2a6f1 | 390 | |
89439d4f MS |
391 | static struct ordered_hashmap_entry *bucket_at_swap(struct swap_entries *swap, unsigned idx) { |
392 | return &swap->e[idx - _IDX_SWAP_BEGIN]; | |
393 | } | |
39c2a6f1 | 394 | |
89439d4f MS |
395 | /* Returns a pointer to the bucket at index idx. |
396 | * Understands real indexes and swap indexes, hence "_virtual". */ | |
397 | static struct hashmap_base_entry *bucket_at_virtual(HashmapBase *h, struct swap_entries *swap, | |
398 | unsigned idx) { | |
399 | if (idx < _IDX_SWAP_BEGIN) | |
400 | return bucket_at(h, idx); | |
401 | ||
402 | if (idx < _IDX_SWAP_END) | |
403 | return &bucket_at_swap(swap, idx)->p.b; | |
404 | ||
405 | assert_not_reached("Invalid index"); | |
406 | } | |
407 | ||
408 | static dib_raw_t *dib_raw_ptr(HashmapBase *h) { | |
409 | return (dib_raw_t*) | |
1a39bc8c | 410 | ((uint8_t*) storage_ptr(h) + hashmap_type_info[h->type].entry_size * n_buckets(h)); |
89439d4f MS |
411 | } |
412 | ||
413 | static unsigned bucket_distance(HashmapBase *h, unsigned idx, unsigned from) { | |
414 | return idx >= from ? idx - from | |
415 | : n_buckets(h) + idx - from; | |
416 | } | |
417 | ||
418 | static unsigned bucket_calculate_dib(HashmapBase *h, unsigned idx, dib_raw_t raw_dib) { | |
419 | unsigned initial_bucket; | |
420 | ||
421 | if (raw_dib == DIB_RAW_FREE) | |
422 | return DIB_FREE; | |
423 | ||
424 | if (_likely_(raw_dib < DIB_RAW_OVERFLOW)) | |
425 | return raw_dib; | |
426 | ||
427 | /* | |
428 | * Having an overflow DIB value is very unlikely. The hash function | |
429 | * would have to be bad. For example, in a table of size 2^24 filled | |
430 | * to load factor 0.9 the maximum observed DIB is only about 60. | |
431 | * In theory (assuming I used Maxima correctly), for an infinite size | |
432 | * hash table with load factor 0.8 the probability of a given entry | |
433 | * having DIB > 40 is 1.9e-8. | |
434 | * This returns the correct DIB value by recomputing the hash value in | |
435 | * the unlikely case. XXX Hitting this case could be a hint to rehash. | |
436 | */ | |
437 | initial_bucket = bucket_hash(h, bucket_at(h, idx)->key); | |
438 | return bucket_distance(h, idx, initial_bucket); | |
439 | } | |
440 | ||
441 | static void bucket_set_dib(HashmapBase *h, unsigned idx, unsigned dib) { | |
442 | dib_raw_ptr(h)[idx] = dib != DIB_FREE ? MIN(dib, DIB_RAW_OVERFLOW) : DIB_RAW_FREE; | |
443 | } | |
444 | ||
445 | static unsigned skip_free_buckets(HashmapBase *h, unsigned idx) { | |
446 | dib_raw_t *dibs; | |
447 | ||
448 | dibs = dib_raw_ptr(h); | |
449 | ||
450 | for ( ; idx < n_buckets(h); idx++) | |
451 | if (dibs[idx] != DIB_RAW_FREE) | |
452 | return idx; | |
453 | ||
454 | return IDX_NIL; | |
455 | } | |
456 | ||
457 | static void bucket_mark_free(HashmapBase *h, unsigned idx) { | |
eccaf899 | 458 | memzero(bucket_at(h, idx), hashmap_type_info[h->type].entry_size); |
89439d4f MS |
459 | bucket_set_dib(h, idx, DIB_FREE); |
460 | } | |
461 | ||
462 | static void bucket_move_entry(HashmapBase *h, struct swap_entries *swap, | |
463 | unsigned from, unsigned to) { | |
464 | struct hashmap_base_entry *e_from, *e_to; | |
465 | ||
466 | assert(from != to); | |
39c2a6f1 | 467 | |
89439d4f MS |
468 | e_from = bucket_at_virtual(h, swap, from); |
469 | e_to = bucket_at_virtual(h, swap, to); | |
470 | ||
471 | memcpy(e_to, e_from, hashmap_type_info[h->type].entry_size); | |
472 | ||
473 | if (h->type == HASHMAP_TYPE_ORDERED) { | |
474 | OrderedHashmap *lh = (OrderedHashmap*) h; | |
475 | struct ordered_hashmap_entry *le, *le_to; | |
476 | ||
477 | le_to = (struct ordered_hashmap_entry*) e_to; | |
478 | ||
479 | if (le_to->iterate_next != IDX_NIL) { | |
480 | le = (struct ordered_hashmap_entry*) | |
481 | bucket_at_virtual(h, swap, le_to->iterate_next); | |
482 | le->iterate_previous = to; | |
483 | } | |
484 | ||
485 | if (le_to->iterate_previous != IDX_NIL) { | |
486 | le = (struct ordered_hashmap_entry*) | |
487 | bucket_at_virtual(h, swap, le_to->iterate_previous); | |
488 | le->iterate_next = to; | |
489 | } | |
490 | ||
491 | if (lh->iterate_list_head == from) | |
492 | lh->iterate_list_head = to; | |
493 | if (lh->iterate_list_tail == from) | |
494 | lh->iterate_list_tail = to; | |
39c2a6f1 | 495 | } |
89439d4f | 496 | } |
60918275 | 497 | |
89439d4f MS |
498 | static unsigned next_idx(HashmapBase *h, unsigned idx) { |
499 | return (idx + 1U) % n_buckets(h); | |
500 | } | |
60918275 | 501 | |
89439d4f MS |
502 | static unsigned prev_idx(HashmapBase *h, unsigned idx) { |
503 | return (n_buckets(h) + idx - 1U) % n_buckets(h); | |
504 | } | |
60918275 | 505 | |
89439d4f MS |
506 | static void *entry_value(HashmapBase *h, struct hashmap_base_entry *e) { |
507 | switch (h->type) { | |
45fa9e29 | 508 | |
89439d4f MS |
509 | case HASHMAP_TYPE_PLAIN: |
510 | case HASHMAP_TYPE_ORDERED: | |
511 | return ((struct plain_hashmap_entry*)e)->value; | |
39c2a6f1 | 512 | |
89439d4f MS |
513 | case HASHMAP_TYPE_SET: |
514 | return (void*) e->key; | |
a3b6fafe | 515 | |
89439d4f MS |
516 | default: |
517 | assert_not_reached("Unknown hashmap type"); | |
518 | } | |
60918275 LP |
519 | } |
520 | ||
89439d4f MS |
521 | static void base_remove_entry(HashmapBase *h, unsigned idx) { |
522 | unsigned left, right, prev, dib; | |
523 | dib_raw_t raw_dib, *dibs; | |
45fa9e29 | 524 | |
89439d4f MS |
525 | dibs = dib_raw_ptr(h); |
526 | assert(dibs[idx] != DIB_RAW_FREE); | |
034c6ed7 | 527 | |
349cc4a5 | 528 | #if ENABLE_DEBUG_HASHMAP |
89439d4f MS |
529 | h->debug.rem_count++; |
530 | h->debug.last_rem_idx = idx; | |
531 | #endif | |
034c6ed7 | 532 | |
89439d4f MS |
533 | left = idx; |
534 | /* Find the stop bucket ("right"). It is either free or has DIB == 0. */ | |
535 | for (right = next_idx(h, left); ; right = next_idx(h, right)) { | |
536 | raw_dib = dibs[right]; | |
4c701096 | 537 | if (IN_SET(raw_dib, 0, DIB_RAW_FREE)) |
89439d4f MS |
538 | break; |
539 | ||
540 | /* The buckets are not supposed to be all occupied and with DIB > 0. | |
541 | * That would mean we could make everyone better off by shifting them | |
542 | * backward. This scenario is impossible. */ | |
543 | assert(left != right); | |
544 | } | |
034c6ed7 | 545 | |
89439d4f MS |
546 | if (h->type == HASHMAP_TYPE_ORDERED) { |
547 | OrderedHashmap *lh = (OrderedHashmap*) h; | |
548 | struct ordered_hashmap_entry *le = ordered_bucket_at(lh, idx); | |
549 | ||
550 | if (le->iterate_next != IDX_NIL) | |
551 | ordered_bucket_at(lh, le->iterate_next)->iterate_previous = le->iterate_previous; | |
552 | else | |
553 | lh->iterate_list_tail = le->iterate_previous; | |
554 | ||
555 | if (le->iterate_previous != IDX_NIL) | |
556 | ordered_bucket_at(lh, le->iterate_previous)->iterate_next = le->iterate_next; | |
557 | else | |
558 | lh->iterate_list_head = le->iterate_next; | |
559 | } | |
560 | ||
561 | /* Now shift all buckets in the interval (left, right) one step backwards */ | |
562 | for (prev = left, left = next_idx(h, left); left != right; | |
563 | prev = left, left = next_idx(h, left)) { | |
564 | dib = bucket_calculate_dib(h, left, dibs[left]); | |
565 | assert(dib != 0); | |
566 | bucket_move_entry(h, NULL, left, prev); | |
567 | bucket_set_dib(h, prev, dib - 1); | |
568 | } | |
569 | ||
570 | bucket_mark_free(h, prev); | |
571 | n_entries_dec(h); | |
84dcca75 | 572 | base_set_dirty(h); |
034c6ed7 | 573 | } |
89439d4f MS |
574 | #define remove_entry(h, idx) base_remove_entry(HASHMAP_BASE(h), idx) |
575 | ||
576 | static unsigned hashmap_iterate_in_insertion_order(OrderedHashmap *h, Iterator *i) { | |
577 | struct ordered_hashmap_entry *e; | |
578 | unsigned idx; | |
034c6ed7 | 579 | |
101d8e63 | 580 | assert(h); |
89439d4f MS |
581 | assert(i); |
582 | ||
583 | if (i->idx == IDX_NIL) | |
584 | goto at_end; | |
585 | ||
586 | if (i->idx == IDX_FIRST && h->iterate_list_head == IDX_NIL) | |
587 | goto at_end; | |
588 | ||
589 | if (i->idx == IDX_FIRST) { | |
590 | idx = h->iterate_list_head; | |
591 | e = ordered_bucket_at(h, idx); | |
101d8e63 | 592 | } else { |
89439d4f MS |
593 | idx = i->idx; |
594 | e = ordered_bucket_at(h, idx); | |
595 | /* | |
596 | * We allow removing the current entry while iterating, but removal may cause | |
597 | * a backward shift. The next entry may thus move one bucket to the left. | |
598 | * To detect when it happens, we remember the key pointer of the entry we were | |
599 | * going to iterate next. If it does not match, there was a backward shift. | |
600 | */ | |
601 | if (e->p.b.key != i->next_key) { | |
602 | idx = prev_idx(HASHMAP_BASE(h), idx); | |
603 | e = ordered_bucket_at(h, idx); | |
604 | } | |
605 | assert(e->p.b.key == i->next_key); | |
101d8e63 | 606 | } |
101d8e63 | 607 | |
349cc4a5 | 608 | #if ENABLE_DEBUG_HASHMAP |
89439d4f MS |
609 | i->prev_idx = idx; |
610 | #endif | |
611 | ||
612 | if (e->iterate_next != IDX_NIL) { | |
613 | struct ordered_hashmap_entry *n; | |
614 | i->idx = e->iterate_next; | |
615 | n = ordered_bucket_at(h, i->idx); | |
616 | i->next_key = n->p.b.key; | |
617 | } else | |
618 | i->idx = IDX_NIL; | |
619 | ||
620 | return idx; | |
621 | ||
622 | at_end: | |
623 | i->idx = IDX_NIL; | |
624 | return IDX_NIL; | |
101d8e63 LP |
625 | } |
626 | ||
89439d4f MS |
627 | static unsigned hashmap_iterate_in_internal_order(HashmapBase *h, Iterator *i) { |
628 | unsigned idx; | |
629 | ||
60918275 | 630 | assert(h); |
89439d4f | 631 | assert(i); |
60918275 | 632 | |
89439d4f MS |
633 | if (i->idx == IDX_NIL) |
634 | goto at_end; | |
60918275 | 635 | |
89439d4f MS |
636 | if (i->idx == IDX_FIRST) { |
637 | /* fast forward to the first occupied bucket */ | |
638 | if (h->has_indirect) { | |
639 | i->idx = skip_free_buckets(h, h->indirect.idx_lowest_entry); | |
640 | h->indirect.idx_lowest_entry = i->idx; | |
641 | } else | |
642 | i->idx = skip_free_buckets(h, 0); | |
643 | ||
644 | if (i->idx == IDX_NIL) | |
645 | goto at_end; | |
646 | } else { | |
647 | struct hashmap_base_entry *e; | |
648 | ||
649 | assert(i->idx > 0); | |
60918275 | 650 | |
89439d4f MS |
651 | e = bucket_at(h, i->idx); |
652 | /* | |
653 | * We allow removing the current entry while iterating, but removal may cause | |
654 | * a backward shift. The next entry may thus move one bucket to the left. | |
655 | * To detect when it happens, we remember the key pointer of the entry we were | |
656 | * going to iterate next. If it does not match, there was a backward shift. | |
657 | */ | |
658 | if (e->key != i->next_key) | |
659 | e = bucket_at(h, --i->idx); | |
60918275 | 660 | |
89439d4f MS |
661 | assert(e->key == i->next_key); |
662 | } | |
663 | ||
664 | idx = i->idx; | |
349cc4a5 | 665 | #if ENABLE_DEBUG_HASHMAP |
89439d4f MS |
666 | i->prev_idx = idx; |
667 | #endif | |
668 | ||
669 | i->idx = skip_free_buckets(h, i->idx + 1); | |
670 | if (i->idx != IDX_NIL) | |
671 | i->next_key = bucket_at(h, i->idx)->key; | |
101d8e63 | 672 | else |
89439d4f MS |
673 | i->idx = IDX_NIL; |
674 | ||
675 | return idx; | |
60918275 | 676 | |
89439d4f MS |
677 | at_end: |
678 | i->idx = IDX_NIL; | |
679 | return IDX_NIL; | |
60918275 LP |
680 | } |
681 | ||
89439d4f MS |
682 | static unsigned hashmap_iterate_entry(HashmapBase *h, Iterator *i) { |
683 | if (!h) { | |
684 | i->idx = IDX_NIL; | |
685 | return IDX_NIL; | |
686 | } | |
101d8e63 | 687 | |
349cc4a5 | 688 | #if ENABLE_DEBUG_HASHMAP |
89439d4f MS |
689 | if (i->idx == IDX_FIRST) { |
690 | i->put_count = h->debug.put_count; | |
691 | i->rem_count = h->debug.rem_count; | |
692 | } else { | |
693 | /* While iterating, must not add any new entries */ | |
694 | assert(i->put_count == h->debug.put_count); | |
695 | /* ... or remove entries other than the current one */ | |
696 | assert(i->rem_count == h->debug.rem_count || | |
697 | (i->rem_count == h->debug.rem_count - 1 && | |
698 | i->prev_idx == h->debug.last_rem_idx)); | |
699 | /* Reset our removals counter */ | |
700 | i->rem_count = h->debug.rem_count; | |
701 | } | |
702 | #endif | |
101d8e63 | 703 | |
89439d4f MS |
704 | return h->type == HASHMAP_TYPE_ORDERED ? hashmap_iterate_in_insertion_order((OrderedHashmap*) h, i) |
705 | : hashmap_iterate_in_internal_order(h, i); | |
706 | } | |
39c2a6f1 | 707 | |
138f49e4 | 708 | bool _hashmap_iterate(HashmapBase *h, Iterator *i, void **value, const void **key) { |
89439d4f MS |
709 | struct hashmap_base_entry *e; |
710 | void *data; | |
711 | unsigned idx; | |
712 | ||
713 | idx = hashmap_iterate_entry(h, i); | |
714 | if (idx == IDX_NIL) { | |
8927b1da DH |
715 | if (value) |
716 | *value = NULL; | |
89439d4f MS |
717 | if (key) |
718 | *key = NULL; | |
719 | ||
8927b1da | 720 | return false; |
89439d4f MS |
721 | } |
722 | ||
723 | e = bucket_at(h, idx); | |
724 | data = entry_value(h, e); | |
8927b1da DH |
725 | if (value) |
726 | *value = data; | |
89439d4f MS |
727 | if (key) |
728 | *key = e->key; | |
729 | ||
8927b1da | 730 | return true; |
101d8e63 LP |
731 | } |
732 | ||
3c4e3031 | 733 | bool set_iterate(const Set *s, Iterator *i, void **value) { |
138f49e4 | 734 | return _hashmap_iterate(HASHMAP_BASE((Set*) s), i, value, NULL); |
89439d4f | 735 | } |
60918275 | 736 | |
89439d4f MS |
737 | #define HASHMAP_FOREACH_IDX(idx, h, i) \ |
738 | for ((i) = ITERATOR_FIRST, (idx) = hashmap_iterate_entry((h), &(i)); \ | |
739 | (idx != IDX_NIL); \ | |
740 | (idx) = hashmap_iterate_entry((h), &(i))) | |
741 | ||
138f49e4 | 742 | IteratedCache *_hashmap_iterated_cache_new(HashmapBase *h) { |
45ea84d8 VC |
743 | IteratedCache *cache; |
744 | ||
745 | assert(h); | |
746 | assert(!h->cached); | |
747 | ||
748 | if (h->cached) | |
749 | return NULL; | |
750 | ||
751 | cache = new0(IteratedCache, 1); | |
752 | if (!cache) | |
753 | return NULL; | |
754 | ||
755 | cache->hashmap = h; | |
756 | h->cached = true; | |
757 | ||
758 | return cache; | |
759 | } | |
760 | ||
89439d4f MS |
761 | static void reset_direct_storage(HashmapBase *h) { |
762 | const struct hashmap_type_info *hi = &hashmap_type_info[h->type]; | |
763 | void *p; | |
764 | ||
765 | assert(!h->has_indirect); | |
766 | ||
767 | p = mempset(h->direct.storage, 0, hi->entry_size * hi->n_direct_buckets); | |
768 | memset(p, DIB_RAW_INIT, sizeof(dib_raw_t) * hi->n_direct_buckets); | |
769 | } | |
770 | ||
add74e89 | 771 | static struct HashmapBase *hashmap_base_new(const struct hash_ops *hash_ops, enum HashmapType type HASHMAP_DEBUG_PARAMS) { |
89439d4f MS |
772 | HashmapBase *h; |
773 | const struct hashmap_type_info *hi = &hashmap_type_info[type]; | |
b4f60743 | 774 | bool up; |
89439d4f | 775 | |
7c48ea02 | 776 | up = mempool_enabled(); |
67f3c402 | 777 | |
b4f60743 | 778 | h = up ? mempool_alloc0_tile(hi->mempool) : malloc0(hi->head_size); |
60918275 | 779 | if (!h) |
89439d4f MS |
780 | return NULL; |
781 | ||
782 | h->type = type; | |
b4f60743 | 783 | h->from_pool = up; |
70b400d9 | 784 | h->hash_ops = hash_ops ?: &trivial_hash_ops; |
89439d4f MS |
785 | |
786 | if (type == HASHMAP_TYPE_ORDERED) { | |
787 | OrderedHashmap *lh = (OrderedHashmap*)h; | |
788 | lh->iterate_list_head = lh->iterate_list_tail = IDX_NIL; | |
789 | } | |
790 | ||
791 | reset_direct_storage(h); | |
60918275 | 792 | |
89439d4f MS |
793 | if (!shared_hash_key_initialized) { |
794 | random_bytes(shared_hash_key, sizeof(shared_hash_key)); | |
795 | shared_hash_key_initialized= true; | |
796 | } | |
797 | ||
349cc4a5 | 798 | #if ENABLE_DEBUG_HASHMAP |
89439d4f MS |
799 | h->debug.func = func; |
800 | h->debug.file = file; | |
801 | h->debug.line = line; | |
4f1b3061 TG |
802 | assert_se(pthread_mutex_lock(&hashmap_debug_list_mutex) == 0); |
803 | LIST_PREPEND(debug_list, hashmap_debug_list, &h->debug); | |
804 | assert_se(pthread_mutex_unlock(&hashmap_debug_list_mutex) == 0); | |
89439d4f MS |
805 | #endif |
806 | ||
807 | return h; | |
808 | } | |
60918275 | 809 | |
138f49e4 | 810 | Hashmap *_hashmap_new(const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { |
add74e89 | 811 | return (Hashmap*) hashmap_base_new(hash_ops, HASHMAP_TYPE_PLAIN HASHMAP_DEBUG_PASS_ARGS); |
89439d4f MS |
812 | } |
813 | ||
138f49e4 | 814 | OrderedHashmap *_ordered_hashmap_new(const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { |
add74e89 | 815 | return (OrderedHashmap*) hashmap_base_new(hash_ops, HASHMAP_TYPE_ORDERED HASHMAP_DEBUG_PASS_ARGS); |
89439d4f MS |
816 | } |
817 | ||
138f49e4 | 818 | Set *_set_new(const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { |
add74e89 | 819 | return (Set*) hashmap_base_new(hash_ops, HASHMAP_TYPE_SET HASHMAP_DEBUG_PASS_ARGS); |
89439d4f MS |
820 | } |
821 | ||
822 | static int hashmap_base_ensure_allocated(HashmapBase **h, const struct hash_ops *hash_ops, | |
add74e89 | 823 | enum HashmapType type HASHMAP_DEBUG_PARAMS) { |
89439d4f MS |
824 | HashmapBase *q; |
825 | ||
826 | assert(h); | |
827 | ||
828 | if (*h) | |
829 | return 0; | |
830 | ||
add74e89 | 831 | q = hashmap_base_new(hash_ops, type HASHMAP_DEBUG_PASS_ARGS); |
89439d4f MS |
832 | if (!q) |
833 | return -ENOMEM; | |
834 | ||
835 | *h = q; | |
9ff7c5b0 | 836 | return 1; |
89439d4f MS |
837 | } |
838 | ||
138f49e4 | 839 | int _hashmap_ensure_allocated(Hashmap **h, const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { |
add74e89 | 840 | return hashmap_base_ensure_allocated((HashmapBase**)h, hash_ops, HASHMAP_TYPE_PLAIN HASHMAP_DEBUG_PASS_ARGS); |
89439d4f MS |
841 | } |
842 | ||
138f49e4 | 843 | int _ordered_hashmap_ensure_allocated(OrderedHashmap **h, const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { |
add74e89 | 844 | return hashmap_base_ensure_allocated((HashmapBase**)h, hash_ops, HASHMAP_TYPE_ORDERED HASHMAP_DEBUG_PASS_ARGS); |
89439d4f MS |
845 | } |
846 | ||
138f49e4 | 847 | int _set_ensure_allocated(Set **s, const struct hash_ops *hash_ops HASHMAP_DEBUG_PARAMS) { |
add74e89 | 848 | return hashmap_base_ensure_allocated((HashmapBase**)s, hash_ops, HASHMAP_TYPE_SET HASHMAP_DEBUG_PASS_ARGS); |
89439d4f MS |
849 | } |
850 | ||
851 | static void hashmap_free_no_clear(HashmapBase *h) { | |
852 | assert(!h->has_indirect); | |
ee05335f | 853 | assert(h->n_direct_entries == 0); |
89439d4f | 854 | |
349cc4a5 | 855 | #if ENABLE_DEBUG_HASHMAP |
4f1b3061 | 856 | assert_se(pthread_mutex_lock(&hashmap_debug_list_mutex) == 0); |
89439d4f | 857 | LIST_REMOVE(debug_list, hashmap_debug_list, &h->debug); |
4f1b3061 | 858 | assert_se(pthread_mutex_unlock(&hashmap_debug_list_mutex) == 0); |
89439d4f | 859 | #endif |
45fa9e29 | 860 | |
205c085b LP |
861 | if (h->from_pool) { |
862 | /* Ensure that the object didn't get migrated between threads. */ | |
863 | assert_se(is_main_thread()); | |
89439d4f | 864 | mempool_free_tile(hashmap_type_info[h->type].mempool, h); |
205c085b | 865 | } else |
39c2a6f1 | 866 | free(h); |
60918275 LP |
867 | } |
868 | ||
138f49e4 | 869 | HashmapBase *_hashmap_free(HashmapBase *h, free_func_t default_free_key, free_func_t default_free_value) { |
cfe561a4 | 870 | if (h) { |
138f49e4 | 871 | _hashmap_clear(h, default_free_key, default_free_value); |
cfe561a4 DH |
872 | hashmap_free_no_clear(h); |
873 | } | |
89439d4f | 874 | |
cfe561a4 | 875 | return NULL; |
89439d4f MS |
876 | } |
877 | ||
138f49e4 | 878 | void _hashmap_clear(HashmapBase *h, free_func_t default_free_key, free_func_t default_free_value) { |
59a5cda7 YW |
879 | free_func_t free_key, free_value; |
880 | if (!h) | |
881 | return; | |
67f3c402 | 882 | |
59a5cda7 YW |
883 | free_key = h->hash_ops->free_key ?: default_free_key; |
884 | free_value = h->hash_ops->free_value ?: default_free_value; | |
67f3c402 | 885 | |
59a5cda7 | 886 | if (free_key || free_value) { |
449ddb2d | 887 | |
c380b84d LP |
888 | /* If destructor calls are defined, let's destroy things defensively: let's take the item out of the |
889 | * hash table, and only then call the destructor functions. If these destructors then try to unregister | |
890 | * themselves from our hash table a second time, the entry is already gone. */ | |
891 | ||
138f49e4 | 892 | while (_hashmap_size(h) > 0) { |
ca323715 TH |
893 | void *k = NULL; |
894 | void *v; | |
c380b84d | 895 | |
138f49e4 | 896 | v = _hashmap_first_key_and_value(h, true, &k); |
fabe5c0e | 897 | |
59a5cda7 | 898 | if (free_key) |
c380b84d | 899 | free_key(k); |
fabe5c0e | 900 | |
59a5cda7 | 901 | if (free_value) |
c380b84d | 902 | free_value(v); |
59a5cda7 | 903 | } |
cfe561a4 | 904 | } |
fabe5c0e | 905 | |
89439d4f MS |
906 | if (h->has_indirect) { |
907 | free(h->indirect.storage); | |
908 | h->has_indirect = false; | |
909 | } | |
910 | ||
911 | h->n_direct_entries = 0; | |
912 | reset_direct_storage(h); | |
913 | ||
914 | if (h->type == HASHMAP_TYPE_ORDERED) { | |
915 | OrderedHashmap *lh = (OrderedHashmap*) h; | |
916 | lh->iterate_list_head = lh->iterate_list_tail = IDX_NIL; | |
917 | } | |
84dcca75 VC |
918 | |
919 | base_set_dirty(h); | |
11dd41ce LP |
920 | } |
921 | ||
89439d4f MS |
922 | static int resize_buckets(HashmapBase *h, unsigned entries_add); |
923 | ||
924 | /* | |
925 | * Finds an empty bucket to put an entry into, starting the scan at 'idx'. | |
926 | * Performs Robin Hood swaps as it goes. The entry to put must be placed | |
927 | * by the caller into swap slot IDX_PUT. | |
928 | * If used for in-place resizing, may leave a displaced entry in swap slot | |
929 | * IDX_PUT. Caller must rehash it next. | |
930 | * Returns: true if it left a displaced entry to rehash next in IDX_PUT, | |
931 | * false otherwise. | |
932 | */ | |
933 | static bool hashmap_put_robin_hood(HashmapBase *h, unsigned idx, | |
934 | struct swap_entries *swap) { | |
935 | dib_raw_t raw_dib, *dibs; | |
936 | unsigned dib, distance; | |
937 | ||
349cc4a5 | 938 | #if ENABLE_DEBUG_HASHMAP |
89439d4f MS |
939 | h->debug.put_count++; |
940 | #endif | |
941 | ||
942 | dibs = dib_raw_ptr(h); | |
943 | ||
944 | for (distance = 0; ; distance++) { | |
945 | raw_dib = dibs[idx]; | |
3742095b | 946 | if (IN_SET(raw_dib, DIB_RAW_FREE, DIB_RAW_REHASH)) { |
89439d4f MS |
947 | if (raw_dib == DIB_RAW_REHASH) |
948 | bucket_move_entry(h, swap, idx, IDX_TMP); | |
949 | ||
950 | if (h->has_indirect && h->indirect.idx_lowest_entry > idx) | |
951 | h->indirect.idx_lowest_entry = idx; | |
60918275 | 952 | |
89439d4f MS |
953 | bucket_set_dib(h, idx, distance); |
954 | bucket_move_entry(h, swap, IDX_PUT, idx); | |
955 | if (raw_dib == DIB_RAW_REHASH) { | |
956 | bucket_move_entry(h, swap, IDX_TMP, IDX_PUT); | |
957 | return true; | |
958 | } | |
60918275 | 959 | |
89439d4f MS |
960 | return false; |
961 | } | |
962 | ||
963 | dib = bucket_calculate_dib(h, idx, raw_dib); | |
964 | ||
965 | if (dib < distance) { | |
966 | /* Found a wealthier entry. Go Robin Hood! */ | |
89439d4f MS |
967 | bucket_set_dib(h, idx, distance); |
968 | ||
969 | /* swap the entries */ | |
970 | bucket_move_entry(h, swap, idx, IDX_TMP); | |
971 | bucket_move_entry(h, swap, IDX_PUT, idx); | |
972 | bucket_move_entry(h, swap, IDX_TMP, IDX_PUT); | |
973 | ||
974 | distance = dib; | |
975 | } | |
976 | ||
977 | idx = next_idx(h, idx); | |
978 | } | |
60918275 LP |
979 | } |
980 | ||
89439d4f MS |
981 | /* |
982 | * Puts an entry into a hashmap, boldly - no check whether key already exists. | |
983 | * The caller must place the entry (only its key and value, not link indexes) | |
984 | * in swap slot IDX_PUT. | |
985 | * Caller must ensure: the key does not exist yet in the hashmap. | |
986 | * that resize is not needed if !may_resize. | |
987 | * Returns: 1 if entry was put successfully. | |
988 | * -ENOMEM if may_resize==true and resize failed with -ENOMEM. | |
989 | * Cannot return -ENOMEM if !may_resize. | |
990 | */ | |
991 | static int hashmap_base_put_boldly(HashmapBase *h, unsigned idx, | |
992 | struct swap_entries *swap, bool may_resize) { | |
993 | struct ordered_hashmap_entry *new_entry; | |
994 | int r; | |
995 | ||
996 | assert(idx < n_buckets(h)); | |
997 | ||
998 | new_entry = bucket_at_swap(swap, IDX_PUT); | |
999 | ||
1000 | if (may_resize) { | |
1001 | r = resize_buckets(h, 1); | |
1002 | if (r < 0) | |
1003 | return r; | |
1004 | if (r > 0) | |
1005 | idx = bucket_hash(h, new_entry->p.b.key); | |
1006 | } | |
1007 | assert(n_entries(h) < n_buckets(h)); | |
1008 | ||
1009 | if (h->type == HASHMAP_TYPE_ORDERED) { | |
1010 | OrderedHashmap *lh = (OrderedHashmap*) h; | |
1011 | ||
1012 | new_entry->iterate_next = IDX_NIL; | |
1013 | new_entry->iterate_previous = lh->iterate_list_tail; | |
1014 | ||
1015 | if (lh->iterate_list_tail != IDX_NIL) { | |
1016 | struct ordered_hashmap_entry *old_tail; | |
1017 | ||
1018 | old_tail = ordered_bucket_at(lh, lh->iterate_list_tail); | |
1019 | assert(old_tail->iterate_next == IDX_NIL); | |
1020 | old_tail->iterate_next = IDX_PUT; | |
1021 | } | |
1022 | ||
1023 | lh->iterate_list_tail = IDX_PUT; | |
1024 | if (lh->iterate_list_head == IDX_NIL) | |
1025 | lh->iterate_list_head = IDX_PUT; | |
1026 | } | |
1027 | ||
1028 | assert_se(hashmap_put_robin_hood(h, idx, swap) == false); | |
1029 | ||
1030 | n_entries_inc(h); | |
349cc4a5 | 1031 | #if ENABLE_DEBUG_HASHMAP |
89439d4f MS |
1032 | h->debug.max_entries = MAX(h->debug.max_entries, n_entries(h)); |
1033 | #endif | |
1034 | ||
84dcca75 VC |
1035 | base_set_dirty(h); |
1036 | ||
89439d4f MS |
1037 | return 1; |
1038 | } | |
1039 | #define hashmap_put_boldly(h, idx, swap, may_resize) \ | |
1040 | hashmap_base_put_boldly(HASHMAP_BASE(h), idx, swap, may_resize) | |
1041 | ||
1042 | /* | |
1043 | * Returns 0 if resize is not needed. | |
f131770b | 1044 | * 1 if successfully resized. |
89439d4f MS |
1045 | * -ENOMEM on allocation failure. |
1046 | */ | |
1047 | static int resize_buckets(HashmapBase *h, unsigned entries_add) { | |
1048 | struct swap_entries swap; | |
1a39bc8c | 1049 | void *new_storage; |
89439d4f MS |
1050 | dib_raw_t *old_dibs, *new_dibs; |
1051 | const struct hashmap_type_info *hi; | |
1052 | unsigned idx, optimal_idx; | |
1053 | unsigned old_n_buckets, new_n_buckets, n_rehashed, new_n_entries; | |
1054 | uint8_t new_shift; | |
1055 | bool rehash_next; | |
45fa9e29 LP |
1056 | |
1057 | assert(h); | |
1058 | ||
89439d4f MS |
1059 | hi = &hashmap_type_info[h->type]; |
1060 | new_n_entries = n_entries(h) + entries_add; | |
e4c691b5 MS |
1061 | |
1062 | /* overflow? */ | |
89439d4f | 1063 | if (_unlikely_(new_n_entries < entries_add)) |
e4c691b5 MS |
1064 | return -ENOMEM; |
1065 | ||
89439d4f MS |
1066 | /* For direct storage we allow 100% load, because it's tiny. */ |
1067 | if (!h->has_indirect && new_n_entries <= hi->n_direct_buckets) | |
9700d698 | 1068 | return 0; |
45fa9e29 | 1069 | |
89439d4f MS |
1070 | /* |
1071 | * Load factor = n/m = 1 - (1/INV_KEEP_FREE). | |
1072 | * From it follows: m = n + n/(INV_KEEP_FREE - 1) | |
1073 | */ | |
1074 | new_n_buckets = new_n_entries + new_n_entries / (INV_KEEP_FREE - 1); | |
1075 | /* overflow? */ | |
1076 | if (_unlikely_(new_n_buckets < new_n_entries)) | |
9700d698 | 1077 | return -ENOMEM; |
45fa9e29 | 1078 | |
89439d4f MS |
1079 | if (_unlikely_(new_n_buckets > UINT_MAX / (hi->entry_size + sizeof(dib_raw_t)))) |
1080 | return -ENOMEM; | |
a3b6fafe | 1081 | |
89439d4f | 1082 | old_n_buckets = n_buckets(h); |
45fa9e29 | 1083 | |
89439d4f MS |
1084 | if (_likely_(new_n_buckets <= old_n_buckets)) |
1085 | return 0; | |
45fa9e29 | 1086 | |
89439d4f MS |
1087 | new_shift = log2u_round_up(MAX( |
1088 | new_n_buckets * (hi->entry_size + sizeof(dib_raw_t)), | |
1089 | 2 * sizeof(struct direct_storage))); | |
45fa9e29 | 1090 | |
89439d4f MS |
1091 | /* Realloc storage (buckets and DIB array). */ |
1092 | new_storage = realloc(h->has_indirect ? h->indirect.storage : NULL, | |
1093 | 1U << new_shift); | |
1094 | if (!new_storage) | |
1095 | return -ENOMEM; | |
45fa9e29 | 1096 | |
89439d4f MS |
1097 | /* Must upgrade direct to indirect storage. */ |
1098 | if (!h->has_indirect) { | |
1099 | memcpy(new_storage, h->direct.storage, | |
1100 | old_n_buckets * (hi->entry_size + sizeof(dib_raw_t))); | |
1101 | h->indirect.n_entries = h->n_direct_entries; | |
1102 | h->indirect.idx_lowest_entry = 0; | |
1103 | h->n_direct_entries = 0; | |
1104 | } | |
45fa9e29 | 1105 | |
89439d4f MS |
1106 | /* Get a new hash key. If we've just upgraded to indirect storage, |
1107 | * allow reusing a previously generated key. It's still a different key | |
1108 | * from the shared one that we used for direct storage. */ | |
1109 | get_hash_key(h->indirect.hash_key, !h->has_indirect); | |
1110 | ||
1111 | h->has_indirect = true; | |
1112 | h->indirect.storage = new_storage; | |
1113 | h->indirect.n_buckets = (1U << new_shift) / | |
1114 | (hi->entry_size + sizeof(dib_raw_t)); | |
1115 | ||
1a39bc8c | 1116 | old_dibs = (dib_raw_t*)((uint8_t*) new_storage + hi->entry_size * old_n_buckets); |
89439d4f MS |
1117 | new_dibs = dib_raw_ptr(h); |
1118 | ||
1119 | /* | |
1120 | * Move the DIB array to the new place, replacing valid DIB values with | |
1121 | * DIB_RAW_REHASH to indicate all of the used buckets need rehashing. | |
1122 | * Note: Overlap is not possible, because we have at least doubled the | |
1123 | * number of buckets and dib_raw_t is smaller than any entry type. | |
1124 | */ | |
1125 | for (idx = 0; idx < old_n_buckets; idx++) { | |
1126 | assert(old_dibs[idx] != DIB_RAW_REHASH); | |
1127 | new_dibs[idx] = old_dibs[idx] == DIB_RAW_FREE ? DIB_RAW_FREE | |
1128 | : DIB_RAW_REHASH; | |
45fa9e29 LP |
1129 | } |
1130 | ||
89439d4f | 1131 | /* Zero the area of newly added entries (including the old DIB area) */ |
eccaf899 | 1132 | memzero(bucket_at(h, old_n_buckets), |
89439d4f | 1133 | (n_buckets(h) - old_n_buckets) * hi->entry_size); |
45fa9e29 | 1134 | |
89439d4f MS |
1135 | /* The upper half of the new DIB array needs initialization */ |
1136 | memset(&new_dibs[old_n_buckets], DIB_RAW_INIT, | |
1137 | (n_buckets(h) - old_n_buckets) * sizeof(dib_raw_t)); | |
9bf3b535 | 1138 | |
89439d4f MS |
1139 | /* Rehash entries that need it */ |
1140 | n_rehashed = 0; | |
1141 | for (idx = 0; idx < old_n_buckets; idx++) { | |
1142 | if (new_dibs[idx] != DIB_RAW_REHASH) | |
1143 | continue; | |
45fa9e29 | 1144 | |
89439d4f | 1145 | optimal_idx = bucket_hash(h, bucket_at(h, idx)->key); |
45fa9e29 | 1146 | |
89439d4f MS |
1147 | /* |
1148 | * Not much to do if by luck the entry hashes to its current | |
1149 | * location. Just set its DIB. | |
1150 | */ | |
1151 | if (optimal_idx == idx) { | |
1152 | new_dibs[idx] = 0; | |
1153 | n_rehashed++; | |
1154 | continue; | |
1155 | } | |
1156 | ||
1157 | new_dibs[idx] = DIB_RAW_FREE; | |
1158 | bucket_move_entry(h, &swap, idx, IDX_PUT); | |
1159 | /* bucket_move_entry does not clear the source */ | |
eccaf899 | 1160 | memzero(bucket_at(h, idx), hi->entry_size); |
89439d4f MS |
1161 | |
1162 | do { | |
1163 | /* | |
1164 | * Find the new bucket for the current entry. This may make | |
1165 | * another entry homeless and load it into IDX_PUT. | |
1166 | */ | |
1167 | rehash_next = hashmap_put_robin_hood(h, optimal_idx, &swap); | |
1168 | n_rehashed++; | |
1169 | ||
1170 | /* Did the current entry displace another one? */ | |
1171 | if (rehash_next) | |
1172 | optimal_idx = bucket_hash(h, bucket_at_swap(&swap, IDX_PUT)->p.b.key); | |
1173 | } while (rehash_next); | |
1174 | } | |
60918275 | 1175 | |
89439d4f | 1176 | assert(n_rehashed == n_entries(h)); |
60918275 | 1177 | |
89439d4f MS |
1178 | return 1; |
1179 | } | |
45fa9e29 | 1180 | |
89439d4f MS |
1181 | /* |
1182 | * Finds an entry with a matching key | |
1183 | * Returns: index of the found entry, or IDX_NIL if not found. | |
1184 | */ | |
1185 | static unsigned base_bucket_scan(HashmapBase *h, unsigned idx, const void *key) { | |
1186 | struct hashmap_base_entry *e; | |
1187 | unsigned dib, distance; | |
1188 | dib_raw_t *dibs = dib_raw_ptr(h); | |
39c2a6f1 | 1189 | |
89439d4f | 1190 | assert(idx < n_buckets(h)); |
60918275 | 1191 | |
89439d4f MS |
1192 | for (distance = 0; ; distance++) { |
1193 | if (dibs[idx] == DIB_RAW_FREE) | |
1194 | return IDX_NIL; | |
60918275 | 1195 | |
89439d4f | 1196 | dib = bucket_calculate_dib(h, idx, dibs[idx]); |
60918275 | 1197 | |
89439d4f MS |
1198 | if (dib < distance) |
1199 | return IDX_NIL; | |
1200 | if (dib == distance) { | |
1201 | e = bucket_at(h, idx); | |
1202 | if (h->hash_ops->compare(e->key, key) == 0) | |
1203 | return idx; | |
1204 | } | |
1205 | ||
1206 | idx = next_idx(h, idx); | |
1207 | } | |
60918275 | 1208 | } |
89439d4f | 1209 | #define bucket_scan(h, idx, key) base_bucket_scan(HASHMAP_BASE(h), idx, key) |
60918275 | 1210 | |
923041cb | 1211 | int hashmap_put(Hashmap *h, const void *key, void *value) { |
89439d4f MS |
1212 | struct swap_entries swap; |
1213 | struct plain_hashmap_entry *e; | |
1214 | unsigned hash, idx; | |
923041cb MS |
1215 | |
1216 | assert(h); | |
1217 | ||
1218 | hash = bucket_hash(h, key); | |
89439d4f MS |
1219 | idx = bucket_scan(h, hash, key); |
1220 | if (idx != IDX_NIL) { | |
1221 | e = plain_bucket_at(h, idx); | |
923041cb MS |
1222 | if (e->value == value) |
1223 | return 0; | |
1224 | return -EEXIST; | |
1225 | } | |
1226 | ||
89439d4f MS |
1227 | e = &bucket_at_swap(&swap, IDX_PUT)->p; |
1228 | e->b.key = key; | |
1229 | e->value = value; | |
1230 | return hashmap_put_boldly(h, hash, &swap, true); | |
1231 | } | |
1232 | ||
1233 | int set_put(Set *s, const void *key) { | |
1234 | struct swap_entries swap; | |
1235 | struct hashmap_base_entry *e; | |
1236 | unsigned hash, idx; | |
1237 | ||
1238 | assert(s); | |
1239 | ||
1240 | hash = bucket_hash(s, key); | |
1241 | idx = bucket_scan(s, hash, key); | |
1242 | if (idx != IDX_NIL) | |
1243 | return 0; | |
1244 | ||
1245 | e = &bucket_at_swap(&swap, IDX_PUT)->p.b; | |
1246 | e->key = key; | |
1247 | return hashmap_put_boldly(s, hash, &swap, true); | |
923041cb MS |
1248 | } |
1249 | ||
0f9ccd95 ZJS |
1250 | int _set_ensure_put(Set **s, const struct hash_ops *hash_ops, const void *key HASHMAP_DEBUG_PARAMS) { |
1251 | int r; | |
1252 | ||
1253 | r = _set_ensure_allocated(s, hash_ops HASHMAP_DEBUG_PASS_ARGS); | |
1254 | if (r < 0) | |
1255 | return r; | |
1256 | ||
1257 | return set_put(*s, key); | |
1258 | } | |
1259 | ||
fcc1d031 ZJS |
1260 | int _set_ensure_consume(Set **s, const struct hash_ops *hash_ops, void *key HASHMAP_DEBUG_PARAMS) { |
1261 | int r; | |
1262 | ||
1263 | r = _set_ensure_put(s, hash_ops, key HASHMAP_DEBUG_PASS_ARGS); | |
1264 | if (r <= 0) { | |
1265 | if (hash_ops && hash_ops->free_key) | |
1266 | hash_ops->free_key(key); | |
1267 | else | |
1268 | free(key); | |
1269 | } | |
1270 | ||
1271 | return r; | |
1272 | } | |
1273 | ||
3158713e | 1274 | int hashmap_replace(Hashmap *h, const void *key, void *value) { |
89439d4f MS |
1275 | struct swap_entries swap; |
1276 | struct plain_hashmap_entry *e; | |
1277 | unsigned hash, idx; | |
3158713e LP |
1278 | |
1279 | assert(h); | |
1280 | ||
a3b6fafe | 1281 | hash = bucket_hash(h, key); |
89439d4f MS |
1282 | idx = bucket_scan(h, hash, key); |
1283 | if (idx != IDX_NIL) { | |
1284 | e = plain_bucket_at(h, idx); | |
349cc4a5 | 1285 | #if ENABLE_DEBUG_HASHMAP |
89439d4f MS |
1286 | /* Although the key is equal, the key pointer may have changed, |
1287 | * and this would break our assumption for iterating. So count | |
1288 | * this operation as incompatible with iteration. */ | |
1289 | if (e->b.key != key) { | |
1290 | h->b.debug.put_count++; | |
1291 | h->b.debug.rem_count++; | |
1292 | h->b.debug.last_rem_idx = idx; | |
1293 | } | |
1294 | #endif | |
1295 | e->b.key = key; | |
3158713e | 1296 | e->value = value; |
84dcca75 VC |
1297 | hashmap_set_dirty(h); |
1298 | ||
3158713e LP |
1299 | return 0; |
1300 | } | |
1301 | ||
89439d4f MS |
1302 | e = &bucket_at_swap(&swap, IDX_PUT)->p; |
1303 | e->b.key = key; | |
1304 | e->value = value; | |
1305 | return hashmap_put_boldly(h, hash, &swap, true); | |
3158713e LP |
1306 | } |
1307 | ||
d99ae53a | 1308 | int hashmap_update(Hashmap *h, const void *key, void *value) { |
89439d4f MS |
1309 | struct plain_hashmap_entry *e; |
1310 | unsigned hash, idx; | |
d99ae53a LP |
1311 | |
1312 | assert(h); | |
1313 | ||
a3b6fafe | 1314 | hash = bucket_hash(h, key); |
89439d4f MS |
1315 | idx = bucket_scan(h, hash, key); |
1316 | if (idx == IDX_NIL) | |
d99ae53a LP |
1317 | return -ENOENT; |
1318 | ||
89439d4f | 1319 | e = plain_bucket_at(h, idx); |
d99ae53a | 1320 | e->value = value; |
84dcca75 VC |
1321 | hashmap_set_dirty(h); |
1322 | ||
d99ae53a LP |
1323 | return 0; |
1324 | } | |
1325 | ||
138f49e4 | 1326 | void *_hashmap_get(HashmapBase *h, const void *key) { |
89439d4f MS |
1327 | struct hashmap_base_entry *e; |
1328 | unsigned hash, idx; | |
60918275 LP |
1329 | |
1330 | if (!h) | |
1331 | return NULL; | |
1332 | ||
a3b6fafe | 1333 | hash = bucket_hash(h, key); |
89439d4f MS |
1334 | idx = bucket_scan(h, hash, key); |
1335 | if (idx == IDX_NIL) | |
60918275 LP |
1336 | return NULL; |
1337 | ||
89439d4f MS |
1338 | e = bucket_at(h, idx); |
1339 | return entry_value(h, e); | |
60918275 LP |
1340 | } |
1341 | ||
89439d4f MS |
1342 | void *hashmap_get2(Hashmap *h, const void *key, void **key2) { |
1343 | struct plain_hashmap_entry *e; | |
1344 | unsigned hash, idx; | |
d99ae53a LP |
1345 | |
1346 | if (!h) | |
1347 | return NULL; | |
1348 | ||
a3b6fafe | 1349 | hash = bucket_hash(h, key); |
89439d4f MS |
1350 | idx = bucket_scan(h, hash, key); |
1351 | if (idx == IDX_NIL) | |
d99ae53a LP |
1352 | return NULL; |
1353 | ||
89439d4f | 1354 | e = plain_bucket_at(h, idx); |
d99ae53a | 1355 | if (key2) |
89439d4f | 1356 | *key2 = (void*) e->b.key; |
d99ae53a LP |
1357 | |
1358 | return e->value; | |
1359 | } | |
1360 | ||
138f49e4 | 1361 | bool _hashmap_contains(HashmapBase *h, const void *key) { |
96342de6 | 1362 | unsigned hash; |
96342de6 LN |
1363 | |
1364 | if (!h) | |
1365 | return false; | |
1366 | ||
a3b6fafe | 1367 | hash = bucket_hash(h, key); |
89439d4f | 1368 | return bucket_scan(h, hash, key) != IDX_NIL; |
96342de6 LN |
1369 | } |
1370 | ||
138f49e4 | 1371 | void *_hashmap_remove(HashmapBase *h, const void *key) { |
89439d4f MS |
1372 | struct hashmap_base_entry *e; |
1373 | unsigned hash, idx; | |
60918275 LP |
1374 | void *data; |
1375 | ||
1376 | if (!h) | |
1377 | return NULL; | |
1378 | ||
a3b6fafe | 1379 | hash = bucket_hash(h, key); |
89439d4f MS |
1380 | idx = bucket_scan(h, hash, key); |
1381 | if (idx == IDX_NIL) | |
60918275 LP |
1382 | return NULL; |
1383 | ||
89439d4f MS |
1384 | e = bucket_at(h, idx); |
1385 | data = entry_value(h, e); | |
1386 | remove_entry(h, idx); | |
60918275 LP |
1387 | |
1388 | return data; | |
1389 | } | |
1390 | ||
89439d4f MS |
1391 | void *hashmap_remove2(Hashmap *h, const void *key, void **rkey) { |
1392 | struct plain_hashmap_entry *e; | |
1393 | unsigned hash, idx; | |
c582a3b3 LP |
1394 | void *data; |
1395 | ||
1396 | if (!h) { | |
1397 | if (rkey) | |
1398 | *rkey = NULL; | |
1399 | return NULL; | |
1400 | } | |
1401 | ||
1402 | hash = bucket_hash(h, key); | |
89439d4f MS |
1403 | idx = bucket_scan(h, hash, key); |
1404 | if (idx == IDX_NIL) { | |
c582a3b3 LP |
1405 | if (rkey) |
1406 | *rkey = NULL; | |
1407 | return NULL; | |
1408 | } | |
1409 | ||
89439d4f | 1410 | e = plain_bucket_at(h, idx); |
c582a3b3 LP |
1411 | data = e->value; |
1412 | if (rkey) | |
89439d4f | 1413 | *rkey = (void*) e->b.key; |
c582a3b3 | 1414 | |
89439d4f | 1415 | remove_entry(h, idx); |
c582a3b3 LP |
1416 | |
1417 | return data; | |
1418 | } | |
1419 | ||
101d8e63 | 1420 | int hashmap_remove_and_put(Hashmap *h, const void *old_key, const void *new_key, void *value) { |
89439d4f MS |
1421 | struct swap_entries swap; |
1422 | struct plain_hashmap_entry *e; | |
1423 | unsigned old_hash, new_hash, idx; | |
101d8e63 LP |
1424 | |
1425 | if (!h) | |
1426 | return -ENOENT; | |
1427 | ||
a3b6fafe | 1428 | old_hash = bucket_hash(h, old_key); |
89439d4f MS |
1429 | idx = bucket_scan(h, old_hash, old_key); |
1430 | if (idx == IDX_NIL) | |
101d8e63 LP |
1431 | return -ENOENT; |
1432 | ||
a3b6fafe | 1433 | new_hash = bucket_hash(h, new_key); |
89439d4f | 1434 | if (bucket_scan(h, new_hash, new_key) != IDX_NIL) |
101d8e63 LP |
1435 | return -EEXIST; |
1436 | ||
89439d4f | 1437 | remove_entry(h, idx); |
101d8e63 | 1438 | |
89439d4f MS |
1439 | e = &bucket_at_swap(&swap, IDX_PUT)->p; |
1440 | e->b.key = new_key; | |
101d8e63 | 1441 | e->value = value; |
89439d4f MS |
1442 | assert_se(hashmap_put_boldly(h, new_hash, &swap, false) == 1); |
1443 | ||
1444 | return 0; | |
1445 | } | |
1446 | ||
1447 | int set_remove_and_put(Set *s, const void *old_key, const void *new_key) { | |
1448 | struct swap_entries swap; | |
1449 | struct hashmap_base_entry *e; | |
1450 | unsigned old_hash, new_hash, idx; | |
101d8e63 | 1451 | |
89439d4f MS |
1452 | if (!s) |
1453 | return -ENOENT; | |
1454 | ||
1455 | old_hash = bucket_hash(s, old_key); | |
1456 | idx = bucket_scan(s, old_hash, old_key); | |
1457 | if (idx == IDX_NIL) | |
1458 | return -ENOENT; | |
1459 | ||
1460 | new_hash = bucket_hash(s, new_key); | |
1461 | if (bucket_scan(s, new_hash, new_key) != IDX_NIL) | |
1462 | return -EEXIST; | |
1463 | ||
1464 | remove_entry(s, idx); | |
1465 | ||
1466 | e = &bucket_at_swap(&swap, IDX_PUT)->p.b; | |
1467 | e->key = new_key; | |
1468 | assert_se(hashmap_put_boldly(s, new_hash, &swap, false) == 1); | |
101d8e63 LP |
1469 | |
1470 | return 0; | |
1471 | } | |
1472 | ||
8fe914ec | 1473 | int hashmap_remove_and_replace(Hashmap *h, const void *old_key, const void *new_key, void *value) { |
89439d4f MS |
1474 | struct swap_entries swap; |
1475 | struct plain_hashmap_entry *e; | |
1476 | unsigned old_hash, new_hash, idx_old, idx_new; | |
8fe914ec LP |
1477 | |
1478 | if (!h) | |
1479 | return -ENOENT; | |
1480 | ||
a3b6fafe | 1481 | old_hash = bucket_hash(h, old_key); |
89439d4f MS |
1482 | idx_old = bucket_scan(h, old_hash, old_key); |
1483 | if (idx_old == IDX_NIL) | |
8fe914ec LP |
1484 | return -ENOENT; |
1485 | ||
89439d4f | 1486 | old_key = bucket_at(HASHMAP_BASE(h), idx_old)->key; |
8fe914ec | 1487 | |
89439d4f MS |
1488 | new_hash = bucket_hash(h, new_key); |
1489 | idx_new = bucket_scan(h, new_hash, new_key); | |
1490 | if (idx_new != IDX_NIL) | |
1491 | if (idx_old != idx_new) { | |
1492 | remove_entry(h, idx_new); | |
1493 | /* Compensate for a possible backward shift. */ | |
1494 | if (old_key != bucket_at(HASHMAP_BASE(h), idx_old)->key) | |
1495 | idx_old = prev_idx(HASHMAP_BASE(h), idx_old); | |
1496 | assert(old_key == bucket_at(HASHMAP_BASE(h), idx_old)->key); | |
1497 | } | |
1498 | ||
1499 | remove_entry(h, idx_old); | |
1500 | ||
1501 | e = &bucket_at_swap(&swap, IDX_PUT)->p; | |
1502 | e->b.key = new_key; | |
8fe914ec | 1503 | e->value = value; |
89439d4f | 1504 | assert_se(hashmap_put_boldly(h, new_hash, &swap, false) == 1); |
8fe914ec LP |
1505 | |
1506 | return 0; | |
1507 | } | |
1508 | ||
138f49e4 | 1509 | void *_hashmap_remove_value(HashmapBase *h, const void *key, void *value) { |
c380b84d | 1510 | struct hashmap_base_entry *e; |
89439d4f | 1511 | unsigned hash, idx; |
3158713e LP |
1512 | |
1513 | if (!h) | |
1514 | return NULL; | |
1515 | ||
a3b6fafe | 1516 | hash = bucket_hash(h, key); |
89439d4f MS |
1517 | idx = bucket_scan(h, hash, key); |
1518 | if (idx == IDX_NIL) | |
3158713e LP |
1519 | return NULL; |
1520 | ||
c380b84d LP |
1521 | e = bucket_at(h, idx); |
1522 | if (entry_value(h, e) != value) | |
3158713e LP |
1523 | return NULL; |
1524 | ||
89439d4f | 1525 | remove_entry(h, idx); |
3158713e LP |
1526 | |
1527 | return value; | |
1528 | } | |
1529 | ||
89439d4f MS |
1530 | static unsigned find_first_entry(HashmapBase *h) { |
1531 | Iterator i = ITERATOR_FIRST; | |
60918275 | 1532 | |
89439d4f MS |
1533 | if (!h || !n_entries(h)) |
1534 | return IDX_NIL; | |
60918275 | 1535 | |
89439d4f | 1536 | return hashmap_iterate_entry(h, &i); |
60918275 LP |
1537 | } |
1538 | ||
138f49e4 | 1539 | void *_hashmap_first_key_and_value(HashmapBase *h, bool remove, void **ret_key) { |
89439d4f | 1540 | struct hashmap_base_entry *e; |
7ef670c3 | 1541 | void *key, *data; |
89439d4f | 1542 | unsigned idx; |
60918275 | 1543 | |
89439d4f | 1544 | idx = find_first_entry(h); |
51c682df TH |
1545 | if (idx == IDX_NIL) { |
1546 | if (ret_key) | |
1547 | *ret_key = NULL; | |
60918275 | 1548 | return NULL; |
51c682df | 1549 | } |
60918275 | 1550 | |
89439d4f | 1551 | e = bucket_at(h, idx); |
7ef670c3 | 1552 | key = (void*) e->key; |
89439d4f | 1553 | data = entry_value(h, e); |
60918275 | 1554 | |
7ef670c3 YW |
1555 | if (remove) |
1556 | remove_entry(h, idx); | |
60918275 | 1557 | |
7ef670c3 YW |
1558 | if (ret_key) |
1559 | *ret_key = key; | |
22be093f | 1560 | |
7ef670c3 | 1561 | return data; |
22be093f LP |
1562 | } |
1563 | ||
138f49e4 | 1564 | unsigned _hashmap_size(HashmapBase *h) { |
60918275 LP |
1565 | if (!h) |
1566 | return 0; | |
1567 | ||
89439d4f | 1568 | return n_entries(h); |
60918275 LP |
1569 | } |
1570 | ||
138f49e4 | 1571 | unsigned _hashmap_buckets(HashmapBase *h) { |
45fa9e29 LP |
1572 | if (!h) |
1573 | return 0; | |
1574 | ||
89439d4f | 1575 | return n_buckets(h); |
45fa9e29 LP |
1576 | } |
1577 | ||
138f49e4 | 1578 | int _hashmap_merge(Hashmap *h, Hashmap *other) { |
89439d4f MS |
1579 | Iterator i; |
1580 | unsigned idx; | |
60918275 | 1581 | |
89439d4f | 1582 | assert(h); |
60918275 | 1583 | |
89439d4f MS |
1584 | HASHMAP_FOREACH_IDX(idx, HASHMAP_BASE(other), i) { |
1585 | struct plain_hashmap_entry *pe = plain_bucket_at(other, idx); | |
1586 | int r; | |
91cdde8a | 1587 | |
89439d4f MS |
1588 | r = hashmap_put(h, pe->b.key, pe->value); |
1589 | if (r < 0 && r != -EEXIST) | |
1590 | return r; | |
1591 | } | |
91cdde8a | 1592 | |
89439d4f MS |
1593 | return 0; |
1594 | } | |
91cdde8a | 1595 | |
89439d4f MS |
1596 | int set_merge(Set *s, Set *other) { |
1597 | Iterator i; | |
1598 | unsigned idx; | |
91cdde8a | 1599 | |
89439d4f MS |
1600 | assert(s); |
1601 | ||
1602 | HASHMAP_FOREACH_IDX(idx, HASHMAP_BASE(other), i) { | |
1603 | struct set_entry *se = set_bucket_at(other, idx); | |
91cdde8a LP |
1604 | int r; |
1605 | ||
89439d4f MS |
1606 | r = set_put(s, se->b.key); |
1607 | if (r < 0) | |
a3b6fafe | 1608 | return r; |
91cdde8a LP |
1609 | } |
1610 | ||
1611 | return 0; | |
1612 | } | |
1613 | ||
138f49e4 | 1614 | int _hashmap_reserve(HashmapBase *h, unsigned entries_add) { |
e4c691b5 MS |
1615 | int r; |
1616 | ||
1617 | assert(h); | |
1618 | ||
1619 | r = resize_buckets(h, entries_add); | |
1620 | if (r < 0) | |
1621 | return r; | |
1622 | ||
1623 | return 0; | |
1624 | } | |
1625 | ||
89439d4f MS |
1626 | /* |
1627 | * The same as hashmap_merge(), but every new item from other is moved to h. | |
1628 | * Keys already in h are skipped and stay in other. | |
1629 | * Returns: 0 on success. | |
1630 | * -ENOMEM on alloc failure, in which case no move has been done. | |
1631 | */ | |
138f49e4 | 1632 | int _hashmap_move(HashmapBase *h, HashmapBase *other) { |
89439d4f MS |
1633 | struct swap_entries swap; |
1634 | struct hashmap_base_entry *e, *n; | |
1635 | Iterator i; | |
1636 | unsigned idx; | |
1637 | int r; | |
101d8e63 LP |
1638 | |
1639 | assert(h); | |
1640 | ||
101d8e63 | 1641 | if (!other) |
7ad63f57 | 1642 | return 0; |
101d8e63 | 1643 | |
89439d4f MS |
1644 | assert(other->type == h->type); |
1645 | ||
1646 | /* | |
1647 | * This reserves buckets for the worst case, where none of other's | |
1648 | * entries are yet present in h. This is preferable to risking | |
1649 | * an allocation failure in the middle of the moving and having to | |
1650 | * rollback or return a partial result. | |
1651 | */ | |
1652 | r = resize_buckets(h, n_entries(other)); | |
1653 | if (r < 0) | |
1654 | return r; | |
101d8e63 | 1655 | |
89439d4f MS |
1656 | HASHMAP_FOREACH_IDX(idx, other, i) { |
1657 | unsigned h_hash; | |
101d8e63 | 1658 | |
89439d4f | 1659 | e = bucket_at(other, idx); |
a3b6fafe | 1660 | h_hash = bucket_hash(h, e->key); |
89439d4f | 1661 | if (bucket_scan(h, h_hash, e->key) != IDX_NIL) |
101d8e63 LP |
1662 | continue; |
1663 | ||
89439d4f MS |
1664 | n = &bucket_at_swap(&swap, IDX_PUT)->p.b; |
1665 | n->key = e->key; | |
1666 | if (h->type != HASHMAP_TYPE_SET) | |
1667 | ((struct plain_hashmap_entry*) n)->value = | |
1668 | ((struct plain_hashmap_entry*) e)->value; | |
1669 | assert_se(hashmap_put_boldly(h, h_hash, &swap, false) == 1); | |
1670 | ||
1671 | remove_entry(other, idx); | |
101d8e63 | 1672 | } |
7ad63f57 MS |
1673 | |
1674 | return 0; | |
101d8e63 LP |
1675 | } |
1676 | ||
138f49e4 | 1677 | int _hashmap_move_one(HashmapBase *h, HashmapBase *other, const void *key) { |
89439d4f MS |
1678 | struct swap_entries swap; |
1679 | unsigned h_hash, other_hash, idx; | |
1680 | struct hashmap_base_entry *e, *n; | |
1681 | int r; | |
101d8e63 | 1682 | |
101d8e63 LP |
1683 | assert(h); |
1684 | ||
a3b6fafe | 1685 | h_hash = bucket_hash(h, key); |
89439d4f | 1686 | if (bucket_scan(h, h_hash, key) != IDX_NIL) |
101d8e63 LP |
1687 | return -EEXIST; |
1688 | ||
bf3d3e2b MS |
1689 | if (!other) |
1690 | return -ENOENT; | |
1691 | ||
89439d4f MS |
1692 | assert(other->type == h->type); |
1693 | ||
a3b6fafe | 1694 | other_hash = bucket_hash(other, key); |
89439d4f MS |
1695 | idx = bucket_scan(other, other_hash, key); |
1696 | if (idx == IDX_NIL) | |
101d8e63 LP |
1697 | return -ENOENT; |
1698 | ||
89439d4f MS |
1699 | e = bucket_at(other, idx); |
1700 | ||
1701 | n = &bucket_at_swap(&swap, IDX_PUT)->p.b; | |
1702 | n->key = e->key; | |
1703 | if (h->type != HASHMAP_TYPE_SET) | |
1704 | ((struct plain_hashmap_entry*) n)->value = | |
1705 | ((struct plain_hashmap_entry*) e)->value; | |
1706 | r = hashmap_put_boldly(h, h_hash, &swap, true); | |
1707 | if (r < 0) | |
1708 | return r; | |
101d8e63 | 1709 | |
89439d4f | 1710 | remove_entry(other, idx); |
101d8e63 LP |
1711 | return 0; |
1712 | } | |
1713 | ||
add74e89 | 1714 | HashmapBase *_hashmap_copy(HashmapBase *h HASHMAP_DEBUG_PARAMS) { |
89439d4f MS |
1715 | HashmapBase *copy; |
1716 | int r; | |
91cdde8a LP |
1717 | |
1718 | assert(h); | |
1719 | ||
add74e89 | 1720 | copy = hashmap_base_new(h->hash_ops, h->type HASHMAP_DEBUG_PASS_ARGS); |
45fa9e29 | 1721 | if (!copy) |
91cdde8a LP |
1722 | return NULL; |
1723 | ||
89439d4f MS |
1724 | switch (h->type) { |
1725 | case HASHMAP_TYPE_PLAIN: | |
1726 | case HASHMAP_TYPE_ORDERED: | |
1727 | r = hashmap_merge((Hashmap*)copy, (Hashmap*)h); | |
1728 | break; | |
1729 | case HASHMAP_TYPE_SET: | |
1730 | r = set_merge((Set*)copy, (Set*)h); | |
1731 | break; | |
1732 | default: | |
1733 | assert_not_reached("Unknown hashmap type"); | |
1734 | } | |
1735 | ||
add74e89 ZJS |
1736 | if (r < 0) |
1737 | return _hashmap_free(copy, false, false); | |
91cdde8a LP |
1738 | |
1739 | return copy; | |
1740 | } | |
db1413d7 | 1741 | |
138f49e4 | 1742 | char **_hashmap_get_strv(HashmapBase *h) { |
db1413d7 | 1743 | char **sv; |
89439d4f MS |
1744 | Iterator i; |
1745 | unsigned idx, n; | |
db1413d7 | 1746 | |
89439d4f | 1747 | sv = new(char*, n_entries(h)+1); |
729e3769 | 1748 | if (!sv) |
db1413d7 KS |
1749 | return NULL; |
1750 | ||
1751 | n = 0; | |
89439d4f MS |
1752 | HASHMAP_FOREACH_IDX(idx, h, i) |
1753 | sv[n++] = entry_value(h, bucket_at(h, idx)); | |
db1413d7 KS |
1754 | sv[n] = NULL; |
1755 | ||
1756 | return sv; | |
1757 | } | |
3c1668da | 1758 | |
89439d4f MS |
1759 | void *ordered_hashmap_next(OrderedHashmap *h, const void *key) { |
1760 | struct ordered_hashmap_entry *e; | |
1761 | unsigned hash, idx; | |
3c1668da | 1762 | |
3c1668da LP |
1763 | if (!h) |
1764 | return NULL; | |
1765 | ||
a3b6fafe | 1766 | hash = bucket_hash(h, key); |
89439d4f MS |
1767 | idx = bucket_scan(h, hash, key); |
1768 | if (idx == IDX_NIL) | |
3c1668da LP |
1769 | return NULL; |
1770 | ||
89439d4f MS |
1771 | e = ordered_bucket_at(h, idx); |
1772 | if (e->iterate_next == IDX_NIL) | |
3c1668da | 1773 | return NULL; |
89439d4f MS |
1774 | return ordered_bucket_at(h, e->iterate_next)->p.value; |
1775 | } | |
3c1668da | 1776 | |
89439d4f MS |
1777 | int set_consume(Set *s, void *value) { |
1778 | int r; | |
1779 | ||
d97c5aea LP |
1780 | assert(s); |
1781 | assert(value); | |
1782 | ||
89439d4f | 1783 | r = set_put(s, value); |
575ccc1b | 1784 | if (r <= 0) |
89439d4f MS |
1785 | free(value); |
1786 | ||
1787 | return r; | |
1788 | } | |
1789 | ||
b8b46b1c | 1790 | int _hashmap_put_strdup(Hashmap **h, const char *k, const char *v HASHMAP_DEBUG_PARAMS) { |
87da8784 ZJS |
1791 | int r; |
1792 | ||
b8b46b1c | 1793 | r = _hashmap_ensure_allocated(h, &string_hash_ops_free_free HASHMAP_DEBUG_PASS_ARGS); |
87da8784 ZJS |
1794 | if (r < 0) |
1795 | return r; | |
1796 | ||
1797 | _cleanup_free_ char *kdup = NULL, *vdup = NULL; | |
25b3e2a8 | 1798 | |
87da8784 | 1799 | kdup = strdup(k); |
25b3e2a8 | 1800 | if (!kdup) |
87da8784 ZJS |
1801 | return -ENOMEM; |
1802 | ||
25b3e2a8 ZJS |
1803 | if (v) { |
1804 | vdup = strdup(v); | |
1805 | if (!vdup) | |
1806 | return -ENOMEM; | |
1807 | } | |
1808 | ||
87da8784 ZJS |
1809 | r = hashmap_put(*h, kdup, vdup); |
1810 | if (r < 0) { | |
25b3e2a8 | 1811 | if (r == -EEXIST && streq_ptr(v, hashmap_get(*h, kdup))) |
87da8784 ZJS |
1812 | return 0; |
1813 | return r; | |
1814 | } | |
1815 | ||
25b3e2a8 ZJS |
1816 | /* 0 with non-null vdup would mean vdup is already in the hashmap, which cannot be */ |
1817 | assert(vdup == NULL || r > 0); | |
1818 | if (r > 0) | |
1819 | kdup = vdup = NULL; | |
87da8784 | 1820 | |
25b3e2a8 | 1821 | return r; |
87da8784 ZJS |
1822 | } |
1823 | ||
b8b46b1c | 1824 | int _set_put_strdup(Set **s, const char *p HASHMAP_DEBUG_PARAMS) { |
89439d4f | 1825 | char *c; |
be327321 | 1826 | int r; |
89439d4f MS |
1827 | |
1828 | assert(s); | |
1829 | assert(p); | |
1830 | ||
b8b46b1c | 1831 | r = _set_ensure_allocated(s, &string_hash_ops_free HASHMAP_DEBUG_PASS_ARGS); |
be327321 ZJS |
1832 | if (r < 0) |
1833 | return r; | |
1834 | ||
1835 | if (set_contains(*s, (char*) p)) | |
454f0f86 LP |
1836 | return 0; |
1837 | ||
89439d4f MS |
1838 | c = strdup(p); |
1839 | if (!c) | |
1840 | return -ENOMEM; | |
1841 | ||
be327321 | 1842 | return set_consume(*s, c); |
89439d4f MS |
1843 | } |
1844 | ||
b8b46b1c | 1845 | int _set_put_strdupv(Set **s, char **l HASHMAP_DEBUG_PARAMS) { |
89439d4f MS |
1846 | int n = 0, r; |
1847 | char **i; | |
1848 | ||
d97c5aea LP |
1849 | assert(s); |
1850 | ||
89439d4f | 1851 | STRV_FOREACH(i, l) { |
b8b46b1c | 1852 | r = _set_put_strdup(s, *i HASHMAP_DEBUG_PASS_ARGS); |
89439d4f MS |
1853 | if (r < 0) |
1854 | return r; | |
1855 | ||
1856 | n += r; | |
1857 | } | |
1858 | ||
1859 | return n; | |
3c1668da | 1860 | } |
d97c5aea LP |
1861 | |
1862 | int set_put_strsplit(Set *s, const char *v, const char *separators, ExtractFlags flags) { | |
1863 | const char *p = v; | |
1864 | int r; | |
1865 | ||
1866 | assert(s); | |
1867 | assert(v); | |
1868 | ||
1869 | for (;;) { | |
1870 | char *word; | |
1871 | ||
1872 | r = extract_first_word(&p, &word, separators, flags); | |
1873 | if (r <= 0) | |
1874 | return r; | |
1875 | ||
1876 | r = set_consume(s, word); | |
1877 | if (r < 0) | |
1878 | return r; | |
1879 | } | |
1880 | } | |
45ea84d8 VC |
1881 | |
1882 | /* expand the cachemem if needed, return true if newly (re)activated. */ | |
1883 | static int cachemem_maintain(CacheMem *mem, unsigned size) { | |
45ea84d8 VC |
1884 | assert(mem); |
1885 | ||
1886 | if (!GREEDY_REALLOC(mem->ptr, mem->n_allocated, size)) { | |
1887 | if (size > 0) | |
1888 | return -ENOMEM; | |
1889 | } | |
1890 | ||
afbbc068 ZJS |
1891 | if (!mem->active) { |
1892 | mem->active = true; | |
1893 | return true; | |
1894 | } | |
45ea84d8 | 1895 | |
afbbc068 | 1896 | return false; |
45ea84d8 VC |
1897 | } |
1898 | ||
1899 | int iterated_cache_get(IteratedCache *cache, const void ***res_keys, const void ***res_values, unsigned *res_n_entries) { | |
1900 | bool sync_keys = false, sync_values = false; | |
1901 | unsigned size; | |
1902 | int r; | |
1903 | ||
1904 | assert(cache); | |
1905 | assert(cache->hashmap); | |
1906 | ||
1907 | size = n_entries(cache->hashmap); | |
1908 | ||
1909 | if (res_keys) { | |
1910 | r = cachemem_maintain(&cache->keys, size); | |
1911 | if (r < 0) | |
1912 | return r; | |
1913 | ||
1914 | sync_keys = r; | |
1915 | } else | |
1916 | cache->keys.active = false; | |
1917 | ||
1918 | if (res_values) { | |
1919 | r = cachemem_maintain(&cache->values, size); | |
1920 | if (r < 0) | |
1921 | return r; | |
1922 | ||
1923 | sync_values = r; | |
1924 | } else | |
1925 | cache->values.active = false; | |
1926 | ||
1927 | if (cache->hashmap->dirty) { | |
1928 | if (cache->keys.active) | |
1929 | sync_keys = true; | |
1930 | if (cache->values.active) | |
1931 | sync_values = true; | |
1932 | ||
1933 | cache->hashmap->dirty = false; | |
1934 | } | |
1935 | ||
1936 | if (sync_keys || sync_values) { | |
1937 | unsigned i, idx; | |
1938 | Iterator iter; | |
1939 | ||
1940 | i = 0; | |
1941 | HASHMAP_FOREACH_IDX(idx, cache->hashmap, iter) { | |
1942 | struct hashmap_base_entry *e; | |
1943 | ||
1944 | e = bucket_at(cache->hashmap, idx); | |
1945 | ||
1946 | if (sync_keys) | |
1947 | cache->keys.ptr[i] = e->key; | |
1948 | if (sync_values) | |
1949 | cache->values.ptr[i] = entry_value(cache->hashmap, e); | |
1950 | i++; | |
1951 | } | |
1952 | } | |
1953 | ||
1954 | if (res_keys) | |
1955 | *res_keys = cache->keys.ptr; | |
1956 | if (res_values) | |
1957 | *res_values = cache->values.ptr; | |
1958 | if (res_n_entries) | |
1959 | *res_n_entries = size; | |
1960 | ||
1961 | return 0; | |
1962 | } | |
1963 | ||
1964 | IteratedCache *iterated_cache_free(IteratedCache *cache) { | |
1965 | if (cache) { | |
1966 | free(cache->keys.ptr); | |
1967 | free(cache->values.ptr); | |
45ea84d8 VC |
1968 | } |
1969 | ||
b61658fd | 1970 | return mfree(cache); |
45ea84d8 | 1971 | } |