]>
Commit | Line | Data |
---|---|---|
96872bc2 LT |
1 | /* |
2 | * name-hash.c | |
3 | * | |
4 | * Hashing names in the index state | |
5 | * | |
6 | * Copyright (C) 2008 Linus Torvalds | |
7 | */ | |
96872bc2 | 8 | #include "cache.h" |
32a8f510 | 9 | #include "environment.h" |
f394e093 | 10 | #include "gettext.h" |
07642942 | 11 | #include "thread-utils.h" |
74ea5c95 | 12 | #include "trace.h" |
6a9372f4 | 13 | #include "trace2.h" |
4589bca8 | 14 | #include "sparse-index.h" |
96872bc2 | 15 | |
2092678c | 16 | struct dir_entry { |
e05881a4 | 17 | struct hashmap_entry ent; |
2092678c | 18 | struct dir_entry *parent; |
2092678c KB |
19 | int nr; |
20 | unsigned int namelen; | |
41284eb0 | 21 | char name[FLEX_ARRAY]; |
2092678c KB |
22 | }; |
23 | ||
5cf88fd8 | 24 | static int dir_entry_cmp(const void *cmp_data UNUSED, |
939af16e EW |
25 | const struct hashmap_entry *eptr, |
26 | const struct hashmap_entry *entry_or_key, | |
56a14ea7 | 27 | const void *keydata) |
e05881a4 | 28 | { |
939af16e | 29 | const struct dir_entry *e1, *e2; |
56a14ea7 SB |
30 | const char *name = keydata; |
31 | ||
939af16e EW |
32 | e1 = container_of(eptr, const struct dir_entry, ent); |
33 | e2 = container_of(entry_or_key, const struct dir_entry, ent); | |
34 | ||
41284eb0 DT |
35 | return e1->namelen != e2->namelen || strncasecmp(e1->name, |
36 | name ? name : e2->name, e1->namelen); | |
e05881a4 KB |
37 | } |
38 | ||
846df809 JH |
39 | static struct dir_entry *find_dir_entry__hash(struct index_state *istate, |
40 | const char *name, unsigned int namelen, unsigned int hash) | |
2092678c | 41 | { |
e05881a4 | 42 | struct dir_entry key; |
d22245a2 | 43 | hashmap_entry_init(&key.ent, hash); |
e05881a4 | 44 | key.namelen = namelen; |
404ab78e | 45 | return hashmap_get_entry(&istate->dir_hash, &key, ent, name); |
2092678c KB |
46 | } |
47 | ||
846df809 JH |
48 | static struct dir_entry *find_dir_entry(struct index_state *istate, |
49 | const char *name, unsigned int namelen) | |
50 | { | |
51 | return find_dir_entry__hash(istate, name, namelen, memihash(name, namelen)); | |
52 | } | |
53 | ||
2092678c KB |
54 | static struct dir_entry *hash_dir_entry(struct index_state *istate, |
55 | struct cache_entry *ce, int namelen) | |
5102c617 JJ |
56 | { |
57 | /* | |
58 | * Throw each directory component in the hash for quick lookup | |
d28eec26 | 59 | * during a git status. Directory components are stored without their |
5102c617 | 60 | * closing slash. Despite submodules being a directory, they never |
d28eec26 | 61 | * reach this point, because they are stored |
2092678c | 62 | * in index_state.name_hash (as ordinary cache_entries). |
5102c617 | 63 | */ |
2092678c KB |
64 | struct dir_entry *dir; |
65 | ||
66 | /* get length of parent directory */ | |
67 | while (namelen > 0 && !is_dir_sep(ce->name[namelen - 1])) | |
68 | namelen--; | |
69 | if (namelen <= 0) | |
70 | return NULL; | |
d28eec26 | 71 | namelen--; |
2092678c KB |
72 | |
73 | /* lookup existing entry for that directory */ | |
74 | dir = find_dir_entry(istate, ce->name, namelen); | |
75 | if (!dir) { | |
76 | /* not found, create it and add to hash table */ | |
96ffc06f | 77 | FLEX_ALLOC_MEM(dir, name, ce->name, namelen); |
d22245a2 | 78 | hashmap_entry_init(&dir->ent, memihash(ce->name, namelen)); |
2092678c | 79 | dir->namelen = namelen; |
b94e5c1d | 80 | hashmap_add(&istate->dir_hash, &dir->ent); |
2092678c KB |
81 | |
82 | /* recursively add missing parent directories */ | |
d28eec26 | 83 | dir->parent = hash_dir_entry(istate, ce, namelen); |
5102c617 | 84 | } |
2092678c KB |
85 | return dir; |
86 | } | |
87 | ||
88 | static void add_dir_entry(struct index_state *istate, struct cache_entry *ce) | |
89 | { | |
90 | /* Add reference to the directory entry (and parents if 0). */ | |
91 | struct dir_entry *dir = hash_dir_entry(istate, ce, ce_namelen(ce)); | |
92 | while (dir && !(dir->nr++)) | |
93 | dir = dir->parent; | |
94 | } | |
95 | ||
96 | static void remove_dir_entry(struct index_state *istate, struct cache_entry *ce) | |
97 | { | |
98 | /* | |
1c8cca19 KB |
99 | * Release reference to the directory entry. If 0, remove and continue |
100 | * with parent directory. | |
2092678c KB |
101 | */ |
102 | struct dir_entry *dir = hash_dir_entry(istate, ce, ce_namelen(ce)); | |
1c8cca19 KB |
103 | while (dir && !(--dir->nr)) { |
104 | struct dir_entry *parent = dir->parent; | |
28ee7941 | 105 | hashmap_remove(&istate->dir_hash, &dir->ent, NULL); |
1c8cca19 KB |
106 | free(dir); |
107 | dir = parent; | |
108 | } | |
5102c617 JJ |
109 | } |
110 | ||
96872bc2 LT |
111 | static void hash_index_entry(struct index_state *istate, struct cache_entry *ce) |
112 | { | |
96872bc2 LT |
113 | if (ce->ce_flags & CE_HASHED) |
114 | return; | |
115 | ce->ce_flags |= CE_HASHED; | |
5f116695 DS |
116 | |
117 | if (!S_ISSPARSEDIR(ce->ce_mode)) { | |
118 | hashmap_entry_init(&ce->ent, memihash(ce->name, ce_namelen(ce))); | |
119 | hashmap_add(&istate->name_hash, &ce->ent); | |
120 | } | |
5102c617 | 121 | |
419a597f | 122 | if (ignore_case) |
2092678c | 123 | add_dir_entry(istate, ce); |
96872bc2 LT |
124 | } |
125 | ||
5cf88fd8 | 126 | static int cache_entry_cmp(const void *cmp_data UNUSED, |
939af16e EW |
127 | const struct hashmap_entry *eptr, |
128 | const struct hashmap_entry *entry_or_key, | |
7663cdc8 | 129 | const void *remove) |
419a597f | 130 | { |
939af16e EW |
131 | const struct cache_entry *ce1, *ce2; |
132 | ||
133 | ce1 = container_of(eptr, const struct cache_entry, ent); | |
134 | ce2 = container_of(entry_or_key, const struct cache_entry, ent); | |
135 | ||
419a597f KB |
136 | /* |
137 | * For remove_name_hash, find the exact entry (pointer equality); for | |
7b359ea6 | 138 | * index_file_exists, find all entries with matching hash code and |
419a597f KB |
139 | * decide whether the entry matches in same_name. |
140 | */ | |
141 | return remove ? !(ce1 == ce2) : 0; | |
142 | } | |
143 | ||
846df809 JH |
144 | static int lazy_try_threaded = 1; |
145 | static int lazy_nr_dir_threads; | |
146 | ||
846df809 JH |
147 | /* |
148 | * Set a minimum number of cache_entries that we will handle per | |
15beaaa3 | 149 | * thread and use that to decide how many threads to run (up to |
846df809 JH |
150 | * the number on the system). |
151 | * | |
152 | * For guidance setting the lower per-thread bound, see: | |
153 | * t/helper/test-lazy-init-name-hash --analyze | |
154 | */ | |
155 | #define LAZY_THREAD_COST (2000) | |
156 | ||
157 | /* | |
158 | * We use n mutexes to guard n partitions of the "istate->dir_hash" | |
159 | * hashtable. Since "find" and "insert" operations will hash to a | |
160 | * particular bucket and modify/search a single chain, we can say | |
161 | * that "all chains mod n" are guarded by the same mutex -- rather | |
162 | * than having a single mutex to guard the entire table. (This does | |
163 | * require that we disable "rehashing" on the hashtable.) | |
164 | * | |
165 | * So, a larger value here decreases the probability of a collision | |
166 | * and the time that each thread must wait for the mutex. | |
167 | */ | |
168 | #define LAZY_MAX_MUTEX (32) | |
169 | ||
170 | static pthread_mutex_t *lazy_dir_mutex_array; | |
171 | ||
172 | /* | |
173 | * An array of lazy_entry items is used by the n threads in | |
174 | * the directory parse (first) phase to (lock-free) store the | |
175 | * intermediate results. These values are then referenced by | |
176 | * the 2 threads in the second phase. | |
177 | */ | |
178 | struct lazy_entry { | |
179 | struct dir_entry *dir; | |
180 | unsigned int hash_dir; | |
181 | unsigned int hash_name; | |
182 | }; | |
183 | ||
184 | /* | |
185 | * Decide if we want to use threads (if available) to load | |
186 | * the hash tables. We set "lazy_nr_dir_threads" to zero when | |
187 | * it is not worth it. | |
188 | */ | |
189 | static int lookup_lazy_params(struct index_state *istate) | |
190 | { | |
191 | int nr_cpus; | |
192 | ||
193 | lazy_nr_dir_threads = 0; | |
194 | ||
195 | if (!lazy_try_threaded) | |
196 | return 0; | |
197 | ||
198 | /* | |
199 | * If we are respecting case, just use the original | |
200 | * code to build the "istate->name_hash". We don't | |
201 | * need the complexity here. | |
202 | */ | |
203 | if (!ignore_case) | |
204 | return 0; | |
205 | ||
206 | nr_cpus = online_cpus(); | |
207 | if (nr_cpus < 2) | |
208 | return 0; | |
209 | ||
210 | if (istate->cache_nr < 2 * LAZY_THREAD_COST) | |
211 | return 0; | |
96872bc2 | 212 | |
846df809 JH |
213 | if (istate->cache_nr < nr_cpus * LAZY_THREAD_COST) |
214 | nr_cpus = istate->cache_nr / LAZY_THREAD_COST; | |
215 | lazy_nr_dir_threads = nr_cpus; | |
216 | return lazy_nr_dir_threads; | |
217 | } | |
218 | ||
219 | /* | |
220 | * Initialize n mutexes for use when searching and inserting | |
221 | * into "istate->dir_hash". All "dir" threads are trying | |
222 | * to insert partial pathnames into the hash as they iterate | |
223 | * over their portions of the index, so lock contention is | |
224 | * high. | |
225 | * | |
226 | * However, the hashmap is going to put items into bucket | |
227 | * chains based on their hash values. Use that to create n | |
228 | * mutexes and lock on mutex[bucket(hash) % n]. This will | |
77363a51 | 229 | * decrease the collision rate by (hopefully) a factor of n. |
846df809 JH |
230 | */ |
231 | static void init_dir_mutex(void) | |
232 | { | |
233 | int j; | |
234 | ||
ca56dadb | 235 | CALLOC_ARRAY(lazy_dir_mutex_array, LAZY_MAX_MUTEX); |
846df809 JH |
236 | |
237 | for (j = 0; j < LAZY_MAX_MUTEX; j++) | |
238 | init_recursive_mutex(&lazy_dir_mutex_array[j]); | |
239 | } | |
240 | ||
241 | static void cleanup_dir_mutex(void) | |
242 | { | |
243 | int j; | |
244 | ||
245 | for (j = 0; j < LAZY_MAX_MUTEX; j++) | |
246 | pthread_mutex_destroy(&lazy_dir_mutex_array[j]); | |
247 | ||
248 | free(lazy_dir_mutex_array); | |
249 | } | |
250 | ||
251 | static void lock_dir_mutex(int j) | |
252 | { | |
253 | pthread_mutex_lock(&lazy_dir_mutex_array[j]); | |
254 | } | |
255 | ||
256 | static void unlock_dir_mutex(int j) | |
257 | { | |
258 | pthread_mutex_unlock(&lazy_dir_mutex_array[j]); | |
259 | } | |
260 | ||
261 | static inline int compute_dir_lock_nr( | |
262 | const struct hashmap *map, | |
263 | unsigned int hash) | |
264 | { | |
265 | return hashmap_bucket(map, hash) % LAZY_MAX_MUTEX; | |
266 | } | |
267 | ||
268 | static struct dir_entry *hash_dir_entry_with_parent_and_prefix( | |
269 | struct index_state *istate, | |
270 | struct dir_entry *parent, | |
271 | struct strbuf *prefix) | |
272 | { | |
273 | struct dir_entry *dir; | |
274 | unsigned int hash; | |
275 | int lock_nr; | |
276 | ||
277 | /* | |
278 | * Either we have a parent directory and path with slash(es) | |
279 | * or the directory is an immediate child of the root directory. | |
280 | */ | |
281 | assert((parent != NULL) ^ (strchr(prefix->buf, '/') == NULL)); | |
282 | ||
283 | if (parent) | |
284 | hash = memihash_cont(parent->ent.hash, | |
285 | prefix->buf + parent->namelen, | |
286 | prefix->len - parent->namelen); | |
287 | else | |
288 | hash = memihash(prefix->buf, prefix->len); | |
289 | ||
290 | lock_nr = compute_dir_lock_nr(&istate->dir_hash, hash); | |
291 | lock_dir_mutex(lock_nr); | |
292 | ||
293 | dir = find_dir_entry__hash(istate, prefix->buf, prefix->len, hash); | |
294 | if (!dir) { | |
295 | FLEX_ALLOC_MEM(dir, name, prefix->buf, prefix->len); | |
d22245a2 | 296 | hashmap_entry_init(&dir->ent, hash); |
846df809 JH |
297 | dir->namelen = prefix->len; |
298 | dir->parent = parent; | |
b94e5c1d | 299 | hashmap_add(&istate->dir_hash, &dir->ent); |
846df809 JH |
300 | |
301 | if (parent) { | |
302 | unlock_dir_mutex(lock_nr); | |
303 | ||
304 | /* All I really need here is an InterlockedIncrement(&(parent->nr)) */ | |
305 | lock_nr = compute_dir_lock_nr(&istate->dir_hash, parent->ent.hash); | |
306 | lock_dir_mutex(lock_nr); | |
307 | parent->nr++; | |
308 | } | |
309 | } | |
310 | ||
311 | unlock_dir_mutex(lock_nr); | |
312 | ||
313 | return dir; | |
314 | } | |
315 | ||
316 | /* | |
317 | * handle_range_1() and handle_range_dir() are derived from | |
318 | * clear_ce_flags_1() and clear_ce_flags_dir() in unpack-trees.c | |
319 | * and handle the iteration over the entire array of index entries. | |
320 | * They use recursion for adjacent entries in the same parent | |
321 | * directory. | |
322 | */ | |
323 | static int handle_range_1( | |
324 | struct index_state *istate, | |
325 | int k_start, | |
326 | int k_end, | |
327 | struct dir_entry *parent, | |
328 | struct strbuf *prefix, | |
329 | struct lazy_entry *lazy_entries); | |
330 | ||
331 | static int handle_range_dir( | |
332 | struct index_state *istate, | |
333 | int k_start, | |
334 | int k_end, | |
335 | struct dir_entry *parent, | |
336 | struct strbuf *prefix, | |
337 | struct lazy_entry *lazy_entries, | |
338 | struct dir_entry **dir_new_out) | |
339 | { | |
340 | int rc, k; | |
341 | int input_prefix_len = prefix->len; | |
342 | struct dir_entry *dir_new; | |
343 | ||
344 | dir_new = hash_dir_entry_with_parent_and_prefix(istate, parent, prefix); | |
345 | ||
346 | strbuf_addch(prefix, '/'); | |
347 | ||
348 | /* | |
349 | * Scan forward in the index array for index entries having the same | |
350 | * path prefix (that are also in this directory). | |
351 | */ | |
2a1bd45b KW |
352 | if (k_start + 1 >= k_end) |
353 | k = k_end; | |
354 | else if (strncmp(istate->cache[k_start + 1]->name, prefix->buf, prefix->len) > 0) | |
846df809 JH |
355 | k = k_start + 1; |
356 | else if (strncmp(istate->cache[k_end - 1]->name, prefix->buf, prefix->len) == 0) | |
357 | k = k_end; | |
358 | else { | |
359 | int begin = k_start; | |
360 | int end = k_end; | |
568a05c5 | 361 | assert(begin >= 0); |
846df809 | 362 | while (begin < end) { |
568a05c5 | 363 | int mid = begin + ((end - begin) >> 1); |
846df809 JH |
364 | int cmp = strncmp(istate->cache[mid]->name, prefix->buf, prefix->len); |
365 | if (cmp == 0) /* mid has same prefix; look in second part */ | |
366 | begin = mid + 1; | |
367 | else if (cmp > 0) /* mid is past group; look in first part */ | |
368 | end = mid; | |
369 | else | |
370 | die("cache entry out of order"); | |
371 | } | |
372 | k = begin; | |
373 | } | |
374 | ||
375 | /* | |
376 | * Recurse and process what we can of this subset [k_start, k). | |
377 | */ | |
378 | rc = handle_range_1(istate, k_start, k, dir_new, prefix, lazy_entries); | |
379 | ||
380 | strbuf_setlen(prefix, input_prefix_len); | |
381 | ||
382 | *dir_new_out = dir_new; | |
383 | return rc; | |
384 | } | |
385 | ||
386 | static int handle_range_1( | |
387 | struct index_state *istate, | |
388 | int k_start, | |
389 | int k_end, | |
390 | struct dir_entry *parent, | |
391 | struct strbuf *prefix, | |
392 | struct lazy_entry *lazy_entries) | |
393 | { | |
394 | int input_prefix_len = prefix->len; | |
395 | int k = k_start; | |
396 | ||
397 | while (k < k_end) { | |
398 | struct cache_entry *ce_k = istate->cache[k]; | |
399 | const char *name, *slash; | |
400 | ||
401 | if (prefix->len && strncmp(ce_k->name, prefix->buf, prefix->len)) | |
402 | break; | |
403 | ||
404 | name = ce_k->name + prefix->len; | |
405 | slash = strchr(name, '/'); | |
406 | ||
407 | if (slash) { | |
408 | int len = slash - name; | |
409 | int processed; | |
410 | struct dir_entry *dir_new; | |
411 | ||
412 | strbuf_add(prefix, name, len); | |
413 | processed = handle_range_dir(istate, k, k_end, parent, prefix, lazy_entries, &dir_new); | |
414 | if (processed) { | |
415 | k += processed; | |
416 | strbuf_setlen(prefix, input_prefix_len); | |
417 | continue; | |
418 | } | |
419 | ||
420 | strbuf_addch(prefix, '/'); | |
421 | processed = handle_range_1(istate, k, k_end, dir_new, prefix, lazy_entries); | |
422 | k += processed; | |
423 | strbuf_setlen(prefix, input_prefix_len); | |
424 | continue; | |
425 | } | |
426 | ||
427 | /* | |
428 | * It is too expensive to take a lock to insert "ce_k" | |
429 | * into "istate->name_hash" and increment the ref-count | |
430 | * on the "parent" dir. So we defer actually updating | |
431 | * permanent data structures until phase 2 (where we | |
432 | * can change the locking requirements) and simply | |
433 | * accumulate our current results into the lazy_entries | |
434 | * data array). | |
435 | * | |
436 | * We do not need to lock the lazy_entries array because | |
437 | * we have exclusive access to the cells in the range | |
438 | * [k_start,k_end) that this thread was given. | |
439 | */ | |
440 | lazy_entries[k].dir = parent; | |
441 | if (parent) { | |
442 | lazy_entries[k].hash_name = memihash_cont( | |
443 | parent->ent.hash, | |
444 | ce_k->name + parent->namelen, | |
445 | ce_namelen(ce_k) - parent->namelen); | |
446 | lazy_entries[k].hash_dir = parent->ent.hash; | |
447 | } else { | |
448 | lazy_entries[k].hash_name = memihash(ce_k->name, ce_namelen(ce_k)); | |
449 | } | |
450 | ||
451 | k++; | |
452 | } | |
453 | ||
454 | return k - k_start; | |
455 | } | |
456 | ||
457 | struct lazy_dir_thread_data { | |
458 | pthread_t pthread; | |
459 | struct index_state *istate; | |
460 | struct lazy_entry *lazy_entries; | |
461 | int k_start; | |
462 | int k_end; | |
463 | }; | |
464 | ||
465 | static void *lazy_dir_thread_proc(void *_data) | |
466 | { | |
467 | struct lazy_dir_thread_data *d = _data; | |
468 | struct strbuf prefix = STRBUF_INIT; | |
469 | handle_range_1(d->istate, d->k_start, d->k_end, NULL, &prefix, d->lazy_entries); | |
470 | strbuf_release(&prefix); | |
471 | return NULL; | |
472 | } | |
473 | ||
474 | struct lazy_name_thread_data { | |
475 | pthread_t pthread; | |
476 | struct index_state *istate; | |
477 | struct lazy_entry *lazy_entries; | |
478 | }; | |
479 | ||
480 | static void *lazy_name_thread_proc(void *_data) | |
481 | { | |
482 | struct lazy_name_thread_data *d = _data; | |
483 | int k; | |
484 | ||
485 | for (k = 0; k < d->istate->cache_nr; k++) { | |
486 | struct cache_entry *ce_k = d->istate->cache[k]; | |
487 | ce_k->ce_flags |= CE_HASHED; | |
d22245a2 | 488 | hashmap_entry_init(&ce_k->ent, d->lazy_entries[k].hash_name); |
b94e5c1d | 489 | hashmap_add(&d->istate->name_hash, &ce_k->ent); |
846df809 JH |
490 | } |
491 | ||
492 | return NULL; | |
493 | } | |
494 | ||
495 | static inline void lazy_update_dir_ref_counts( | |
496 | struct index_state *istate, | |
497 | struct lazy_entry *lazy_entries) | |
498 | { | |
499 | int k; | |
500 | ||
501 | for (k = 0; k < istate->cache_nr; k++) { | |
502 | if (lazy_entries[k].dir) | |
503 | lazy_entries[k].dir->nr++; | |
504 | } | |
505 | } | |
506 | ||
507 | static void threaded_lazy_init_name_hash( | |
508 | struct index_state *istate) | |
509 | { | |
2179045f | 510 | int err; |
846df809 JH |
511 | int nr_each; |
512 | int k_start; | |
513 | int t; | |
514 | struct lazy_entry *lazy_entries; | |
515 | struct lazy_dir_thread_data *td_dir; | |
516 | struct lazy_name_thread_data *td_name; | |
517 | ||
07642942 NTND |
518 | if (!HAVE_THREADS) |
519 | return; | |
520 | ||
846df809 JH |
521 | k_start = 0; |
522 | nr_each = DIV_ROUND_UP(istate->cache_nr, lazy_nr_dir_threads); | |
523 | ||
ca56dadb RS |
524 | CALLOC_ARRAY(lazy_entries, istate->cache_nr); |
525 | CALLOC_ARRAY(td_dir, lazy_nr_dir_threads); | |
526 | CALLOC_ARRAY(td_name, 1); | |
846df809 JH |
527 | |
528 | init_dir_mutex(); | |
529 | ||
530 | /* | |
531 | * Phase 1: | |
532 | * Build "istate->dir_hash" using n "dir" threads (and a read-only index). | |
533 | */ | |
534 | for (t = 0; t < lazy_nr_dir_threads; t++) { | |
535 | struct lazy_dir_thread_data *td_dir_t = td_dir + t; | |
536 | td_dir_t->istate = istate; | |
537 | td_dir_t->lazy_entries = lazy_entries; | |
538 | td_dir_t->k_start = k_start; | |
539 | k_start += nr_each; | |
540 | if (k_start > istate->cache_nr) | |
541 | k_start = istate->cache_nr; | |
542 | td_dir_t->k_end = k_start; | |
2179045f NTND |
543 | err = pthread_create(&td_dir_t->pthread, NULL, lazy_dir_thread_proc, td_dir_t); |
544 | if (err) | |
545 | die(_("unable to create lazy_dir thread: %s"), strerror(err)); | |
846df809 JH |
546 | } |
547 | for (t = 0; t < lazy_nr_dir_threads; t++) { | |
548 | struct lazy_dir_thread_data *td_dir_t = td_dir + t; | |
549 | if (pthread_join(td_dir_t->pthread, NULL)) | |
550 | die("unable to join lazy_dir_thread"); | |
551 | } | |
552 | ||
553 | /* | |
554 | * Phase 2: | |
555 | * Iterate over all index entries and add them to the "istate->name_hash" | |
556 | * using a single "name" background thread. | |
557 | * (Testing showed it wasn't worth running more than 1 thread for this.) | |
558 | * | |
559 | * Meanwhile, finish updating the parent directory ref-counts for each | |
560 | * index entry using the current thread. (This step is very fast and | |
561 | * doesn't need threading.) | |
562 | */ | |
563 | td_name->istate = istate; | |
564 | td_name->lazy_entries = lazy_entries; | |
2179045f NTND |
565 | err = pthread_create(&td_name->pthread, NULL, lazy_name_thread_proc, td_name); |
566 | if (err) | |
567 | die(_("unable to create lazy_name thread: %s"), strerror(err)); | |
846df809 JH |
568 | |
569 | lazy_update_dir_ref_counts(istate, lazy_entries); | |
570 | ||
2179045f NTND |
571 | err = pthread_join(td_name->pthread, NULL); |
572 | if (err) | |
573 | die(_("unable to join lazy_name thread: %s"), strerror(err)); | |
846df809 JH |
574 | |
575 | cleanup_dir_mutex(); | |
576 | ||
577 | free(td_name); | |
578 | free(td_dir); | |
579 | free(lazy_entries); | |
580 | } | |
581 | ||
846df809 JH |
582 | static void lazy_init_name_hash(struct index_state *istate) |
583 | { | |
ca54d9ba | 584 | |
96872bc2 LT |
585 | if (istate->name_hash_initialized) |
586 | return; | |
c46c406a | 587 | trace_performance_enter(); |
6a9372f4 | 588 | trace2_region_enter("index", "name-hash-init", istate->repo); |
56a14ea7 SB |
589 | hashmap_init(&istate->name_hash, cache_entry_cmp, NULL, istate->cache_nr); |
590 | hashmap_init(&istate->dir_hash, dir_entry_cmp, NULL, istate->cache_nr); | |
846df809 JH |
591 | |
592 | if (lookup_lazy_params(istate)) { | |
8b604d19 JH |
593 | /* |
594 | * Disable item counting and automatic rehashing because | |
595 | * we do per-chain (mod n) locking rather than whole hashmap | |
596 | * locking and we need to prevent the table-size from changing | |
597 | * and bucket items from being redistributed. | |
598 | */ | |
599 | hashmap_disable_item_counting(&istate->dir_hash); | |
846df809 | 600 | threaded_lazy_init_name_hash(istate); |
8b604d19 | 601 | hashmap_enable_item_counting(&istate->dir_hash); |
846df809 JH |
602 | } else { |
603 | int nr; | |
604 | for (nr = 0; nr < istate->cache_nr; nr++) | |
605 | hash_index_entry(istate, istate->cache[nr]); | |
606 | } | |
607 | ||
96872bc2 | 608 | istate->name_hash_initialized = 1; |
6a9372f4 | 609 | trace2_region_leave("index", "name-hash-init", istate->repo); |
c46c406a | 610 | trace_performance_leave("initialize name hash"); |
96872bc2 LT |
611 | } |
612 | ||
846df809 JH |
613 | /* |
614 | * A test routine for t/helper/ sources. | |
615 | * | |
616 | * Returns the number of threads used or 0 when | |
617 | * the non-threaded code path was used. | |
618 | * | |
619 | * Requesting threading WILL NOT override guards | |
620 | * in lookup_lazy_params(). | |
621 | */ | |
622 | int test_lazy_init_name_hash(struct index_state *istate, int try_threaded) | |
623 | { | |
624 | lazy_nr_dir_threads = 0; | |
625 | lazy_try_threaded = try_threaded; | |
626 | ||
627 | lazy_init_name_hash(istate); | |
628 | ||
629 | return lazy_nr_dir_threads; | |
630 | } | |
631 | ||
96872bc2 LT |
632 | void add_name_hash(struct index_state *istate, struct cache_entry *ce) |
633 | { | |
96872bc2 LT |
634 | if (istate->name_hash_initialized) |
635 | hash_index_entry(istate, ce); | |
636 | } | |
637 | ||
2092678c KB |
638 | void remove_name_hash(struct index_state *istate, struct cache_entry *ce) |
639 | { | |
419a597f KB |
640 | if (!istate->name_hash_initialized || !(ce->ce_flags & CE_HASHED)) |
641 | return; | |
642 | ce->ce_flags &= ~CE_HASHED; | |
28ee7941 | 643 | hashmap_remove(&istate->name_hash, &ce->ent, ce); |
2092678c | 644 | |
419a597f KB |
645 | if (ignore_case) |
646 | remove_dir_entry(istate, ce); | |
2092678c KB |
647 | } |
648 | ||
cd2fef59 LT |
649 | static int slow_same_name(const char *name1, int len1, const char *name2, int len2) |
650 | { | |
651 | if (len1 != len2) | |
652 | return 0; | |
653 | ||
654 | while (len1) { | |
655 | unsigned char c1 = *name1++; | |
656 | unsigned char c2 = *name2++; | |
657 | len1--; | |
658 | if (c1 != c2) { | |
659 | c1 = toupper(c1); | |
660 | c2 = toupper(c2); | |
661 | if (c1 != c2) | |
662 | return 0; | |
663 | } | |
664 | } | |
665 | return 1; | |
666 | } | |
667 | ||
668 | static int same_name(const struct cache_entry *ce, const char *name, int namelen, int icase) | |
669 | { | |
670 | int len = ce_namelen(ce); | |
671 | ||
672 | /* | |
673 | * Always do exact compare, even if we want a case-ignoring comparison; | |
674 | * we do the quick exact one first, because it will be the common case. | |
675 | */ | |
be99ec97 | 676 | if (len == namelen && !memcmp(name, ce->name, len)) |
cd2fef59 LT |
677 | return 1; |
678 | ||
5102c617 JJ |
679 | if (!icase) |
680 | return 0; | |
681 | ||
2092678c | 682 | return slow_same_name(name, namelen, ce->name, len); |
cd2fef59 LT |
683 | } |
684 | ||
41284eb0 | 685 | int index_dir_exists(struct index_state *istate, const char *name, int namelen) |
db5360f3 | 686 | { |
db5360f3 ES |
687 | struct dir_entry *dir; |
688 | ||
689 | lazy_init_name_hash(istate); | |
4589bca8 | 690 | expand_to_path(istate, name, namelen, 0); |
db5360f3 | 691 | dir = find_dir_entry(istate, name, namelen); |
41284eb0 DT |
692 | return dir && dir->nr; |
693 | } | |
db5360f3 | 694 | |
41284eb0 DT |
695 | void adjust_dirname_case(struct index_state *istate, char *name) |
696 | { | |
697 | const char *startPtr = name; | |
698 | const char *ptr = startPtr; | |
db5360f3 | 699 | |
41284eb0 | 700 | lazy_init_name_hash(istate); |
4589bca8 | 701 | expand_to_path(istate, name, strlen(name), 0); |
41284eb0 DT |
702 | while (*ptr) { |
703 | while (*ptr && *ptr != '/') | |
704 | ptr++; | |
705 | ||
706 | if (*ptr == '/') { | |
707 | struct dir_entry *dir; | |
708 | ||
c95525e9 | 709 | dir = find_dir_entry(istate, name, ptr - name); |
41284eb0 DT |
710 | if (dir) { |
711 | memcpy((void *)startPtr, dir->name + (startPtr - name), ptr - startPtr); | |
c95525e9 | 712 | startPtr = ptr + 1; |
41284eb0 | 713 | } |
c95525e9 | 714 | ptr++; |
41284eb0 DT |
715 | } |
716 | } | |
db5360f3 ES |
717 | } |
718 | ||
719 | struct cache_entry *index_file_exists(struct index_state *istate, const char *name, int namelen, int icase) | |
96872bc2 | 720 | { |
96872bc2 | 721 | struct cache_entry *ce; |
f0e63c41 | 722 | unsigned int hash = memihash(name, namelen); |
96872bc2 LT |
723 | |
724 | lazy_init_name_hash(istate); | |
4589bca8 | 725 | expand_to_path(istate, name, namelen, icase); |
96872bc2 | 726 | |
f0e63c41 EW |
727 | ce = hashmap_get_entry_from_hash(&istate->name_hash, hash, NULL, |
728 | struct cache_entry, ent); | |
23dee69f | 729 | hashmap_for_each_entry_from(&istate->name_hash, ce, ent) { |
419a597f KB |
730 | if (same_name(ce, name, namelen, icase)) |
731 | return ce; | |
96872bc2 | 732 | } |
df292c79 | 733 | return NULL; |
96872bc2 | 734 | } |
2092678c | 735 | |
2092678c KB |
736 | void free_name_hash(struct index_state *istate) |
737 | { | |
738 | if (!istate->name_hash_initialized) | |
739 | return; | |
740 | istate->name_hash_initialized = 0; | |
2092678c | 741 | |
6da1a258 EN |
742 | hashmap_clear(&istate->name_hash); |
743 | hashmap_clear_and_free(&istate->dir_hash, struct dir_entry, ent); | |
2092678c | 744 | } |