1 | #define JEMALLOC_C_ |
2 | #include "jemalloc/internal/jemalloc_preamble.h" |
3 | #include "jemalloc/internal/jemalloc_internal_includes.h" |
4 | |
5 | #include "jemalloc/internal/assert.h" |
6 | #include "jemalloc/internal/atomic.h" |
7 | #include "jemalloc/internal/ctl.h" |
8 | #include "jemalloc/internal/extent_dss.h" |
9 | #include "jemalloc/internal/extent_mmap.h" |
10 | #include "jemalloc/internal/hook.h" |
11 | #include "jemalloc/internal/jemalloc_internal_types.h" |
12 | #include "jemalloc/internal/log.h" |
13 | #include "jemalloc/internal/malloc_io.h" |
14 | #include "jemalloc/internal/mutex.h" |
15 | #include "jemalloc/internal/rtree.h" |
16 | #include "jemalloc/internal/safety_check.h" |
17 | #include "jemalloc/internal/sc.h" |
18 | #include "jemalloc/internal/spin.h" |
19 | #include "jemalloc/internal/sz.h" |
20 | #include "jemalloc/internal/ticker.h" |
21 | #include "jemalloc/internal/util.h" |
22 | |
23 | /******************************************************************************/ |
24 | /* Data. */ |
25 | |
26 | /* Runtime configuration options. */ |
27 | const char *je_malloc_conf |
28 | #ifndef _WIN32 |
29 | JEMALLOC_ATTR(weak) |
30 | #endif |
31 | ; |
32 | bool opt_abort = |
33 | #ifdef JEMALLOC_DEBUG |
34 | true |
35 | #else |
36 | false |
37 | #endif |
38 | ; |
39 | bool opt_abort_conf = |
40 | #ifdef JEMALLOC_DEBUG |
41 | true |
42 | #else |
43 | false |
44 | #endif |
45 | ; |
46 | /* Intentionally default off, even with debug builds. */ |
47 | bool opt_confirm_conf = false; |
48 | const char *opt_junk = |
49 | #if (defined(JEMALLOC_DEBUG) && defined(JEMALLOC_FILL)) |
50 | "true" |
51 | #else |
52 | "false" |
53 | #endif |
54 | ; |
55 | bool opt_junk_alloc = |
56 | #if (defined(JEMALLOC_DEBUG) && defined(JEMALLOC_FILL)) |
57 | true |
58 | #else |
59 | false |
60 | #endif |
61 | ; |
62 | bool opt_junk_free = |
63 | #if (defined(JEMALLOC_DEBUG) && defined(JEMALLOC_FILL)) |
64 | true |
65 | #else |
66 | false |
67 | #endif |
68 | ; |
69 | |
70 | bool opt_utrace = false; |
71 | bool opt_xmalloc = false; |
72 | bool opt_zero = false; |
73 | unsigned opt_narenas = 0; |
74 | |
75 | unsigned ncpus; |
76 | |
77 | /* Protects arenas initialization. */ |
78 | malloc_mutex_t arenas_lock; |
79 | /* |
80 | * Arenas that are used to service external requests. Not all elements of the |
81 | * arenas array are necessarily used; arenas are created lazily as needed. |
82 | * |
83 | * arenas[0..narenas_auto) are used for automatic multiplexing of threads and |
84 | * arenas. arenas[narenas_auto..narenas_total) are only used if the application |
85 | * takes some action to create them and allocate from them. |
86 | * |
87 | * Points to an arena_t. |
88 | */ |
89 | JEMALLOC_ALIGNED(CACHELINE) |
90 | atomic_p_t arenas[MALLOCX_ARENA_LIMIT]; |
91 | static atomic_u_t narenas_total; /* Use narenas_total_*(). */ |
92 | /* Below three are read-only after initialization. */ |
93 | static arena_t *a0; /* arenas[0]. */ |
94 | unsigned narenas_auto; |
95 | unsigned manual_arena_base; |
96 | |
97 | typedef enum { |
98 | malloc_init_uninitialized = 3, |
99 | malloc_init_a0_initialized = 2, |
100 | malloc_init_recursible = 1, |
101 | malloc_init_initialized = 0 /* Common case --> jnz. */ |
102 | } malloc_init_t; |
103 | static malloc_init_t malloc_init_state = malloc_init_uninitialized; |
104 | |
105 | /* False should be the common case. Set to true to trigger initialization. */ |
106 | bool malloc_slow = true; |
107 | |
108 | /* When malloc_slow is true, set the corresponding bits for sanity check. */ |
109 | enum { |
110 | flag_opt_junk_alloc = (1U), |
111 | flag_opt_junk_free = (1U << 1), |
112 | flag_opt_zero = (1U << 2), |
113 | flag_opt_utrace = (1U << 3), |
114 | flag_opt_xmalloc = (1U << 4) |
115 | }; |
116 | static uint8_t malloc_slow_flags; |
117 | |
118 | #ifdef JEMALLOC_THREADED_INIT |
119 | /* Used to let the initializing thread recursively allocate. */ |
120 | # define NO_INITIALIZER ((unsigned long)0) |
121 | # define INITIALIZER pthread_self() |
122 | # define IS_INITIALIZER (malloc_initializer == pthread_self()) |
123 | static pthread_t malloc_initializer = NO_INITIALIZER; |
124 | #else |
125 | # define NO_INITIALIZER false |
126 | # define INITIALIZER true |
127 | # define IS_INITIALIZER malloc_initializer |
128 | static bool malloc_initializer = NO_INITIALIZER; |
129 | #endif |
130 | |
131 | /* Used to avoid initialization races. */ |
132 | #ifdef _WIN32 |
133 | #if _WIN32_WINNT >= 0x0600 |
134 | static malloc_mutex_t init_lock = SRWLOCK_INIT; |
135 | #else |
136 | static malloc_mutex_t init_lock; |
137 | static bool init_lock_initialized = false; |
138 | |
139 | JEMALLOC_ATTR(constructor) |
140 | static void WINAPI |
141 | _init_init_lock(void) { |
142 | /* |
143 | * If another constructor in the same binary is using mallctl to e.g. |
144 | * set up extent hooks, it may end up running before this one, and |
145 | * malloc_init_hard will crash trying to lock the uninitialized lock. So |
146 | * we force an initialization of the lock in malloc_init_hard as well. |
147 | * We don't try to care about atomicity of the accessed to the |
148 | * init_lock_initialized boolean, since it really only matters early in |
149 | * the process creation, before any separate thread normally starts |
150 | * doing anything. |
151 | */ |
152 | if (!init_lock_initialized) { |
153 | malloc_mutex_init(&init_lock, "init" , WITNESS_RANK_INIT, |
154 | malloc_mutex_rank_exclusive); |
155 | } |
156 | init_lock_initialized = true; |
157 | } |
158 | |
159 | #ifdef _MSC_VER |
160 | # pragma section(".CRT$XCU", read) |
161 | JEMALLOC_SECTION(".CRT$XCU" ) JEMALLOC_ATTR(used) |
162 | static const void (WINAPI *init_init_lock)(void) = _init_init_lock; |
163 | #endif |
164 | #endif |
165 | #else |
166 | static malloc_mutex_t init_lock = MALLOC_MUTEX_INITIALIZER; |
167 | #endif |
168 | |
169 | typedef struct { |
170 | void *p; /* Input pointer (as in realloc(p, s)). */ |
171 | size_t s; /* Request size. */ |
172 | void *r; /* Result pointer. */ |
173 | } malloc_utrace_t; |
174 | |
175 | #ifdef JEMALLOC_UTRACE |
176 | # define UTRACE(a, b, c) do { \ |
177 | if (unlikely(opt_utrace)) { \ |
178 | int utrace_serrno = errno; \ |
179 | malloc_utrace_t ut; \ |
180 | ut.p = (a); \ |
181 | ut.s = (b); \ |
182 | ut.r = (c); \ |
183 | utrace(&ut, sizeof(ut)); \ |
184 | errno = utrace_serrno; \ |
185 | } \ |
186 | } while (0) |
187 | #else |
188 | # define UTRACE(a, b, c) |
189 | #endif |
190 | |
191 | /* Whether encountered any invalid config options. */ |
192 | static bool had_conf_error = false; |
193 | |
194 | /******************************************************************************/ |
195 | /* |
196 | * Function prototypes for static functions that are referenced prior to |
197 | * definition. |
198 | */ |
199 | |
200 | static bool malloc_init_hard_a0(void); |
201 | static bool malloc_init_hard(void); |
202 | |
203 | /******************************************************************************/ |
204 | /* |
205 | * Begin miscellaneous support functions. |
206 | */ |
207 | |
208 | bool |
209 | malloc_initialized(void) { |
210 | return (malloc_init_state == malloc_init_initialized); |
211 | } |
212 | |
213 | JEMALLOC_ALWAYS_INLINE bool |
214 | malloc_init_a0(void) { |
215 | if (unlikely(malloc_init_state == malloc_init_uninitialized)) { |
216 | return malloc_init_hard_a0(); |
217 | } |
218 | return false; |
219 | } |
220 | |
221 | JEMALLOC_ALWAYS_INLINE bool |
222 | malloc_init(void) { |
223 | if (unlikely(!malloc_initialized()) && malloc_init_hard()) { |
224 | return true; |
225 | } |
226 | return false; |
227 | } |
228 | |
229 | /* |
230 | * The a0*() functions are used instead of i{d,}alloc() in situations that |
231 | * cannot tolerate TLS variable access. |
232 | */ |
233 | |
234 | static void * |
235 | a0ialloc(size_t size, bool zero, bool is_internal) { |
236 | if (unlikely(malloc_init_a0())) { |
237 | return NULL; |
238 | } |
239 | |
240 | return iallocztm(TSDN_NULL, size, sz_size2index(size), zero, NULL, |
241 | is_internal, arena_get(TSDN_NULL, 0, true), true); |
242 | } |
243 | |
244 | static void |
245 | a0idalloc(void *ptr, bool is_internal) { |
246 | idalloctm(TSDN_NULL, ptr, NULL, NULL, is_internal, true); |
247 | } |
248 | |
249 | void * |
250 | a0malloc(size_t size) { |
251 | return a0ialloc(size, false, true); |
252 | } |
253 | |
254 | void |
255 | a0dalloc(void *ptr) { |
256 | a0idalloc(ptr, true); |
257 | } |
258 | |
259 | /* |
260 | * FreeBSD's libc uses the bootstrap_*() functions in bootstrap-senstive |
261 | * situations that cannot tolerate TLS variable access (TLS allocation and very |
262 | * early internal data structure initialization). |
263 | */ |
264 | |
265 | void * |
266 | bootstrap_malloc(size_t size) { |
267 | if (unlikely(size == 0)) { |
268 | size = 1; |
269 | } |
270 | |
271 | return a0ialloc(size, false, false); |
272 | } |
273 | |
274 | void * |
275 | bootstrap_calloc(size_t num, size_t size) { |
276 | size_t num_size; |
277 | |
278 | num_size = num * size; |
279 | if (unlikely(num_size == 0)) { |
280 | assert(num == 0 || size == 0); |
281 | num_size = 1; |
282 | } |
283 | |
284 | return a0ialloc(num_size, true, false); |
285 | } |
286 | |
287 | void |
288 | bootstrap_free(void *ptr) { |
289 | if (unlikely(ptr == NULL)) { |
290 | return; |
291 | } |
292 | |
293 | a0idalloc(ptr, false); |
294 | } |
295 | |
296 | void |
297 | arena_set(unsigned ind, arena_t *arena) { |
298 | atomic_store_p(&arenas[ind], arena, ATOMIC_RELEASE); |
299 | } |
300 | |
301 | static void |
302 | narenas_total_set(unsigned narenas) { |
303 | atomic_store_u(&narenas_total, narenas, ATOMIC_RELEASE); |
304 | } |
305 | |
306 | static void |
307 | narenas_total_inc(void) { |
308 | atomic_fetch_add_u(&narenas_total, 1, ATOMIC_RELEASE); |
309 | } |
310 | |
311 | unsigned |
312 | narenas_total_get(void) { |
313 | return atomic_load_u(&narenas_total, ATOMIC_ACQUIRE); |
314 | } |
315 | |
316 | /* Create a new arena and insert it into the arenas array at index ind. */ |
317 | static arena_t * |
318 | arena_init_locked(tsdn_t *tsdn, unsigned ind, extent_hooks_t *extent_hooks) { |
319 | arena_t *arena; |
320 | |
321 | assert(ind <= narenas_total_get()); |
322 | if (ind >= MALLOCX_ARENA_LIMIT) { |
323 | return NULL; |
324 | } |
325 | if (ind == narenas_total_get()) { |
326 | narenas_total_inc(); |
327 | } |
328 | |
329 | /* |
330 | * Another thread may have already initialized arenas[ind] if it's an |
331 | * auto arena. |
332 | */ |
333 | arena = arena_get(tsdn, ind, false); |
334 | if (arena != NULL) { |
335 | assert(arena_is_auto(arena)); |
336 | return arena; |
337 | } |
338 | |
339 | /* Actually initialize the arena. */ |
340 | arena = arena_new(tsdn, ind, extent_hooks); |
341 | |
342 | return arena; |
343 | } |
344 | |
345 | static void |
346 | arena_new_create_background_thread(tsdn_t *tsdn, unsigned ind) { |
347 | if (ind == 0) { |
348 | return; |
349 | } |
350 | /* |
351 | * Avoid creating a new background thread just for the huge arena, which |
352 | * purges eagerly by default. |
353 | */ |
354 | if (have_background_thread && !arena_is_huge(ind)) { |
355 | if (background_thread_create(tsdn_tsd(tsdn), ind)) { |
356 | malloc_printf("<jemalloc>: error in background thread " |
357 | "creation for arena %u. Abort.\n" , ind); |
358 | abort(); |
359 | } |
360 | } |
361 | } |
362 | |
363 | arena_t * |
364 | arena_init(tsdn_t *tsdn, unsigned ind, extent_hooks_t *extent_hooks) { |
365 | arena_t *arena; |
366 | |
367 | malloc_mutex_lock(tsdn, &arenas_lock); |
368 | arena = arena_init_locked(tsdn, ind, extent_hooks); |
369 | malloc_mutex_unlock(tsdn, &arenas_lock); |
370 | |
371 | arena_new_create_background_thread(tsdn, ind); |
372 | |
373 | return arena; |
374 | } |
375 | |
376 | static void |
377 | arena_bind(tsd_t *tsd, unsigned ind, bool internal) { |
378 | arena_t *arena = arena_get(tsd_tsdn(tsd), ind, false); |
379 | arena_nthreads_inc(arena, internal); |
380 | |
381 | if (internal) { |
382 | tsd_iarena_set(tsd, arena); |
383 | } else { |
384 | tsd_arena_set(tsd, arena); |
385 | unsigned shard = atomic_fetch_add_u(&arena->binshard_next, 1, |
386 | ATOMIC_RELAXED); |
387 | tsd_binshards_t *bins = tsd_binshardsp_get(tsd); |
388 | for (unsigned i = 0; i < SC_NBINS; i++) { |
389 | assert(bin_infos[i].n_shards > 0 && |
390 | bin_infos[i].n_shards <= BIN_SHARDS_MAX); |
391 | bins->binshard[i] = shard % bin_infos[i].n_shards; |
392 | } |
393 | } |
394 | } |
395 | |
396 | void |
397 | arena_migrate(tsd_t *tsd, unsigned oldind, unsigned newind) { |
398 | arena_t *oldarena, *newarena; |
399 | |
400 | oldarena = arena_get(tsd_tsdn(tsd), oldind, false); |
401 | newarena = arena_get(tsd_tsdn(tsd), newind, false); |
402 | arena_nthreads_dec(oldarena, false); |
403 | arena_nthreads_inc(newarena, false); |
404 | tsd_arena_set(tsd, newarena); |
405 | } |
406 | |
407 | static void |
408 | arena_unbind(tsd_t *tsd, unsigned ind, bool internal) { |
409 | arena_t *arena; |
410 | |
411 | arena = arena_get(tsd_tsdn(tsd), ind, false); |
412 | arena_nthreads_dec(arena, internal); |
413 | |
414 | if (internal) { |
415 | tsd_iarena_set(tsd, NULL); |
416 | } else { |
417 | tsd_arena_set(tsd, NULL); |
418 | } |
419 | } |
420 | |
421 | arena_tdata_t * |
422 | arena_tdata_get_hard(tsd_t *tsd, unsigned ind) { |
423 | arena_tdata_t *tdata, *arenas_tdata_old; |
424 | arena_tdata_t *arenas_tdata = tsd_arenas_tdata_get(tsd); |
425 | unsigned narenas_tdata_old, i; |
426 | unsigned narenas_tdata = tsd_narenas_tdata_get(tsd); |
427 | unsigned narenas_actual = narenas_total_get(); |
428 | |
429 | /* |
430 | * Dissociate old tdata array (and set up for deallocation upon return) |
431 | * if it's too small. |
432 | */ |
433 | if (arenas_tdata != NULL && narenas_tdata < narenas_actual) { |
434 | arenas_tdata_old = arenas_tdata; |
435 | narenas_tdata_old = narenas_tdata; |
436 | arenas_tdata = NULL; |
437 | narenas_tdata = 0; |
438 | tsd_arenas_tdata_set(tsd, arenas_tdata); |
439 | tsd_narenas_tdata_set(tsd, narenas_tdata); |
440 | } else { |
441 | arenas_tdata_old = NULL; |
442 | narenas_tdata_old = 0; |
443 | } |
444 | |
445 | /* Allocate tdata array if it's missing. */ |
446 | if (arenas_tdata == NULL) { |
447 | bool *arenas_tdata_bypassp = tsd_arenas_tdata_bypassp_get(tsd); |
448 | narenas_tdata = (ind < narenas_actual) ? narenas_actual : ind+1; |
449 | |
450 | if (tsd_nominal(tsd) && !*arenas_tdata_bypassp) { |
451 | *arenas_tdata_bypassp = true; |
452 | arenas_tdata = (arena_tdata_t *)a0malloc( |
453 | sizeof(arena_tdata_t) * narenas_tdata); |
454 | *arenas_tdata_bypassp = false; |
455 | } |
456 | if (arenas_tdata == NULL) { |
457 | tdata = NULL; |
458 | goto label_return; |
459 | } |
460 | assert(tsd_nominal(tsd) && !*arenas_tdata_bypassp); |
461 | tsd_arenas_tdata_set(tsd, arenas_tdata); |
462 | tsd_narenas_tdata_set(tsd, narenas_tdata); |
463 | } |
464 | |
465 | /* |
466 | * Copy to tdata array. It's possible that the actual number of arenas |
467 | * has increased since narenas_total_get() was called above, but that |
468 | * causes no correctness issues unless two threads concurrently execute |
469 | * the arenas.create mallctl, which we trust mallctl synchronization to |
470 | * prevent. |
471 | */ |
472 | |
473 | /* Copy/initialize tickers. */ |
474 | for (i = 0; i < narenas_actual; i++) { |
475 | if (i < narenas_tdata_old) { |
476 | ticker_copy(&arenas_tdata[i].decay_ticker, |
477 | &arenas_tdata_old[i].decay_ticker); |
478 | } else { |
479 | ticker_init(&arenas_tdata[i].decay_ticker, |
480 | DECAY_NTICKS_PER_UPDATE); |
481 | } |
482 | } |
483 | if (narenas_tdata > narenas_actual) { |
484 | memset(&arenas_tdata[narenas_actual], 0, sizeof(arena_tdata_t) |
485 | * (narenas_tdata - narenas_actual)); |
486 | } |
487 | |
488 | /* Read the refreshed tdata array. */ |
489 | tdata = &arenas_tdata[ind]; |
490 | label_return: |
491 | if (arenas_tdata_old != NULL) { |
492 | a0dalloc(arenas_tdata_old); |
493 | } |
494 | return tdata; |
495 | } |
496 | |
497 | /* Slow path, called only by arena_choose(). */ |
498 | arena_t * |
499 | arena_choose_hard(tsd_t *tsd, bool internal) { |
500 | arena_t *ret JEMALLOC_CC_SILENCE_INIT(NULL); |
501 | |
502 | if (have_percpu_arena && PERCPU_ARENA_ENABLED(opt_percpu_arena)) { |
503 | unsigned choose = percpu_arena_choose(); |
504 | ret = arena_get(tsd_tsdn(tsd), choose, true); |
505 | assert(ret != NULL); |
506 | arena_bind(tsd, arena_ind_get(ret), false); |
507 | arena_bind(tsd, arena_ind_get(ret), true); |
508 | |
509 | return ret; |
510 | } |
511 | |
512 | if (narenas_auto > 1) { |
513 | unsigned i, j, choose[2], first_null; |
514 | bool is_new_arena[2]; |
515 | |
516 | /* |
517 | * Determine binding for both non-internal and internal |
518 | * allocation. |
519 | * |
520 | * choose[0]: For application allocation. |
521 | * choose[1]: For internal metadata allocation. |
522 | */ |
523 | |
524 | for (j = 0; j < 2; j++) { |
525 | choose[j] = 0; |
526 | is_new_arena[j] = false; |
527 | } |
528 | |
529 | first_null = narenas_auto; |
530 | malloc_mutex_lock(tsd_tsdn(tsd), &arenas_lock); |
531 | assert(arena_get(tsd_tsdn(tsd), 0, false) != NULL); |
532 | for (i = 1; i < narenas_auto; i++) { |
533 | if (arena_get(tsd_tsdn(tsd), i, false) != NULL) { |
534 | /* |
535 | * Choose the first arena that has the lowest |
536 | * number of threads assigned to it. |
537 | */ |
538 | for (j = 0; j < 2; j++) { |
539 | if (arena_nthreads_get(arena_get( |
540 | tsd_tsdn(tsd), i, false), !!j) < |
541 | arena_nthreads_get(arena_get( |
542 | tsd_tsdn(tsd), choose[j], false), |
543 | !!j)) { |
544 | choose[j] = i; |
545 | } |
546 | } |
547 | } else if (first_null == narenas_auto) { |
548 | /* |
549 | * Record the index of the first uninitialized |
550 | * arena, in case all extant arenas are in use. |
551 | * |
552 | * NB: It is possible for there to be |
553 | * discontinuities in terms of initialized |
554 | * versus uninitialized arenas, due to the |
555 | * "thread.arena" mallctl. |
556 | */ |
557 | first_null = i; |
558 | } |
559 | } |
560 | |
561 | for (j = 0; j < 2; j++) { |
562 | if (arena_nthreads_get(arena_get(tsd_tsdn(tsd), |
563 | choose[j], false), !!j) == 0 || first_null == |
564 | narenas_auto) { |
565 | /* |
566 | * Use an unloaded arena, or the least loaded |
567 | * arena if all arenas are already initialized. |
568 | */ |
569 | if (!!j == internal) { |
570 | ret = arena_get(tsd_tsdn(tsd), |
571 | choose[j], false); |
572 | } |
573 | } else { |
574 | arena_t *arena; |
575 | |
576 | /* Initialize a new arena. */ |
577 | choose[j] = first_null; |
578 | arena = arena_init_locked(tsd_tsdn(tsd), |
579 | choose[j], |
580 | (extent_hooks_t *)&extent_hooks_default); |
581 | if (arena == NULL) { |
582 | malloc_mutex_unlock(tsd_tsdn(tsd), |
583 | &arenas_lock); |
584 | return NULL; |
585 | } |
586 | is_new_arena[j] = true; |
587 | if (!!j == internal) { |
588 | ret = arena; |
589 | } |
590 | } |
591 | arena_bind(tsd, choose[j], !!j); |
592 | } |
593 | malloc_mutex_unlock(tsd_tsdn(tsd), &arenas_lock); |
594 | |
595 | for (j = 0; j < 2; j++) { |
596 | if (is_new_arena[j]) { |
597 | assert(choose[j] > 0); |
598 | arena_new_create_background_thread( |
599 | tsd_tsdn(tsd), choose[j]); |
600 | } |
601 | } |
602 | |
603 | } else { |
604 | ret = arena_get(tsd_tsdn(tsd), 0, false); |
605 | arena_bind(tsd, 0, false); |
606 | arena_bind(tsd, 0, true); |
607 | } |
608 | |
609 | return ret; |
610 | } |
611 | |
612 | void |
613 | iarena_cleanup(tsd_t *tsd) { |
614 | arena_t *iarena; |
615 | |
616 | iarena = tsd_iarena_get(tsd); |
617 | if (iarena != NULL) { |
618 | arena_unbind(tsd, arena_ind_get(iarena), true); |
619 | } |
620 | } |
621 | |
622 | void |
623 | arena_cleanup(tsd_t *tsd) { |
624 | arena_t *arena; |
625 | |
626 | arena = tsd_arena_get(tsd); |
627 | if (arena != NULL) { |
628 | arena_unbind(tsd, arena_ind_get(arena), false); |
629 | } |
630 | } |
631 | |
632 | void |
633 | arenas_tdata_cleanup(tsd_t *tsd) { |
634 | arena_tdata_t *arenas_tdata; |
635 | |
636 | /* Prevent tsd->arenas_tdata from being (re)created. */ |
637 | *tsd_arenas_tdata_bypassp_get(tsd) = true; |
638 | |
639 | arenas_tdata = tsd_arenas_tdata_get(tsd); |
640 | if (arenas_tdata != NULL) { |
641 | tsd_arenas_tdata_set(tsd, NULL); |
642 | a0dalloc(arenas_tdata); |
643 | } |
644 | } |
645 | |
646 | static void |
647 | stats_print_atexit(void) { |
648 | if (config_stats) { |
649 | tsdn_t *tsdn; |
650 | unsigned narenas, i; |
651 | |
652 | tsdn = tsdn_fetch(); |
653 | |
654 | /* |
655 | * Merge stats from extant threads. This is racy, since |
656 | * individual threads do not lock when recording tcache stats |
657 | * events. As a consequence, the final stats may be slightly |
658 | * out of date by the time they are reported, if other threads |
659 | * continue to allocate. |
660 | */ |
661 | for (i = 0, narenas = narenas_total_get(); i < narenas; i++) { |
662 | arena_t *arena = arena_get(tsdn, i, false); |
663 | if (arena != NULL) { |
664 | tcache_t *tcache; |
665 | |
666 | malloc_mutex_lock(tsdn, &arena->tcache_ql_mtx); |
667 | ql_foreach(tcache, &arena->tcache_ql, link) { |
668 | tcache_stats_merge(tsdn, tcache, arena); |
669 | } |
670 | malloc_mutex_unlock(tsdn, |
671 | &arena->tcache_ql_mtx); |
672 | } |
673 | } |
674 | } |
675 | je_malloc_stats_print(NULL, NULL, opt_stats_print_opts); |
676 | } |
677 | |
678 | /* |
679 | * Ensure that we don't hold any locks upon entry to or exit from allocator |
680 | * code (in a "broad" sense that doesn't count a reentrant allocation as an |
681 | * entrance or exit). |
682 | */ |
683 | JEMALLOC_ALWAYS_INLINE void |
684 | check_entry_exit_locking(tsdn_t *tsdn) { |
685 | if (!config_debug) { |
686 | return; |
687 | } |
688 | if (tsdn_null(tsdn)) { |
689 | return; |
690 | } |
691 | tsd_t *tsd = tsdn_tsd(tsdn); |
692 | /* |
693 | * It's possible we hold locks at entry/exit if we're in a nested |
694 | * allocation. |
695 | */ |
696 | int8_t reentrancy_level = tsd_reentrancy_level_get(tsd); |
697 | if (reentrancy_level != 0) { |
698 | return; |
699 | } |
700 | witness_assert_lockless(tsdn_witness_tsdp_get(tsdn)); |
701 | } |
702 | |
703 | /* |
704 | * End miscellaneous support functions. |
705 | */ |
706 | /******************************************************************************/ |
707 | /* |
708 | * Begin initialization functions. |
709 | */ |
710 | |
711 | static char * |
712 | jemalloc_secure_getenv(const char *name) { |
713 | #ifdef JEMALLOC_HAVE_SECURE_GETENV |
714 | return secure_getenv(name); |
715 | #else |
716 | # ifdef JEMALLOC_HAVE_ISSETUGID |
717 | if (issetugid() != 0) { |
718 | return NULL; |
719 | } |
720 | # endif |
721 | return getenv(name); |
722 | #endif |
723 | } |
724 | |
725 | static unsigned |
726 | malloc_ncpus(void) { |
727 | long result; |
728 | |
729 | #ifdef _WIN32 |
730 | SYSTEM_INFO si; |
731 | GetSystemInfo(&si); |
732 | result = si.dwNumberOfProcessors; |
733 | #elif defined(JEMALLOC_GLIBC_MALLOC_HOOK) && defined(CPU_COUNT) |
734 | /* |
735 | * glibc >= 2.6 has the CPU_COUNT macro. |
736 | * |
737 | * glibc's sysconf() uses isspace(). glibc allocates for the first time |
738 | * *before* setting up the isspace tables. Therefore we need a |
739 | * different method to get the number of CPUs. |
740 | */ |
741 | { |
742 | cpu_set_t set; |
743 | |
744 | pthread_getaffinity_np(pthread_self(), sizeof(set), &set); |
745 | result = CPU_COUNT(&set); |
746 | } |
747 | #else |
748 | result = sysconf(_SC_NPROCESSORS_ONLN); |
749 | #endif |
750 | return ((result == -1) ? 1 : (unsigned)result); |
751 | } |
752 | |
753 | static void |
754 | init_opt_stats_print_opts(const char *v, size_t vlen) { |
755 | size_t opts_len = strlen(opt_stats_print_opts); |
756 | assert(opts_len <= stats_print_tot_num_options); |
757 | |
758 | for (size_t i = 0; i < vlen; i++) { |
759 | switch (v[i]) { |
760 | #define OPTION(o, v, d, s) case o: break; |
761 | STATS_PRINT_OPTIONS |
762 | #undef OPTION |
763 | default: continue; |
764 | } |
765 | |
766 | if (strchr(opt_stats_print_opts, v[i]) != NULL) { |
767 | /* Ignore repeated. */ |
768 | continue; |
769 | } |
770 | |
771 | opt_stats_print_opts[opts_len++] = v[i]; |
772 | opt_stats_print_opts[opts_len] = '\0'; |
773 | assert(opts_len <= stats_print_tot_num_options); |
774 | } |
775 | assert(opts_len == strlen(opt_stats_print_opts)); |
776 | } |
777 | |
778 | /* Reads the next size pair in a multi-sized option. */ |
779 | static bool |
780 | malloc_conf_multi_sizes_next(const char **slab_size_segment_cur, |
781 | size_t *vlen_left, size_t *slab_start, size_t *slab_end, size_t *new_size) { |
782 | const char *cur = *slab_size_segment_cur; |
783 | char *end; |
784 | uintmax_t um; |
785 | |
786 | set_errno(0); |
787 | |
788 | /* First number, then '-' */ |
789 | um = malloc_strtoumax(cur, &end, 0); |
790 | if (get_errno() != 0 || *end != '-') { |
791 | return true; |
792 | } |
793 | *slab_start = (size_t)um; |
794 | cur = end + 1; |
795 | |
796 | /* Second number, then ':' */ |
797 | um = malloc_strtoumax(cur, &end, 0); |
798 | if (get_errno() != 0 || *end != ':') { |
799 | return true; |
800 | } |
801 | *slab_end = (size_t)um; |
802 | cur = end + 1; |
803 | |
804 | /* Last number */ |
805 | um = malloc_strtoumax(cur, &end, 0); |
806 | if (get_errno() != 0) { |
807 | return true; |
808 | } |
809 | *new_size = (size_t)um; |
810 | |
811 | /* Consume the separator if there is one. */ |
812 | if (*end == '|') { |
813 | end++; |
814 | } |
815 | |
816 | *vlen_left -= end - *slab_size_segment_cur; |
817 | *slab_size_segment_cur = end; |
818 | |
819 | return false; |
820 | } |
821 | |
822 | static bool |
823 | malloc_conf_next(char const **opts_p, char const **k_p, size_t *klen_p, |
824 | char const **v_p, size_t *vlen_p) { |
825 | bool accept; |
826 | const char *opts = *opts_p; |
827 | |
828 | *k_p = opts; |
829 | |
830 | for (accept = false; !accept;) { |
831 | switch (*opts) { |
832 | case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': |
833 | case 'G': case 'H': case 'I': case 'J': case 'K': case 'L': |
834 | case 'M': case 'N': case 'O': case 'P': case 'Q': case 'R': |
835 | case 'S': case 'T': case 'U': case 'V': case 'W': case 'X': |
836 | case 'Y': case 'Z': |
837 | case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': |
838 | case 'g': case 'h': case 'i': case 'j': case 'k': case 'l': |
839 | case 'm': case 'n': case 'o': case 'p': case 'q': case 'r': |
840 | case 's': case 't': case 'u': case 'v': case 'w': case 'x': |
841 | case 'y': case 'z': |
842 | case '0': case '1': case '2': case '3': case '4': case '5': |
843 | case '6': case '7': case '8': case '9': |
844 | case '_': |
845 | opts++; |
846 | break; |
847 | case ':': |
848 | opts++; |
849 | *klen_p = (uintptr_t)opts - 1 - (uintptr_t)*k_p; |
850 | *v_p = opts; |
851 | accept = true; |
852 | break; |
853 | case '\0': |
854 | if (opts != *opts_p) { |
855 | malloc_write("<jemalloc>: Conf string ends " |
856 | "with key\n" ); |
857 | } |
858 | return true; |
859 | default: |
860 | malloc_write("<jemalloc>: Malformed conf string\n" ); |
861 | return true; |
862 | } |
863 | } |
864 | |
865 | for (accept = false; !accept;) { |
866 | switch (*opts) { |
867 | case ',': |
868 | opts++; |
869 | /* |
870 | * Look ahead one character here, because the next time |
871 | * this function is called, it will assume that end of |
872 | * input has been cleanly reached if no input remains, |
873 | * but we have optimistically already consumed the |
874 | * comma if one exists. |
875 | */ |
876 | if (*opts == '\0') { |
877 | malloc_write("<jemalloc>: Conf string ends " |
878 | "with comma\n" ); |
879 | } |
880 | *vlen_p = (uintptr_t)opts - 1 - (uintptr_t)*v_p; |
881 | accept = true; |
882 | break; |
883 | case '\0': |
884 | *vlen_p = (uintptr_t)opts - (uintptr_t)*v_p; |
885 | accept = true; |
886 | break; |
887 | default: |
888 | opts++; |
889 | break; |
890 | } |
891 | } |
892 | |
893 | *opts_p = opts; |
894 | return false; |
895 | } |
896 | |
897 | static void |
898 | malloc_abort_invalid_conf(void) { |
899 | assert(opt_abort_conf); |
900 | malloc_printf("<jemalloc>: Abort (abort_conf:true) on invalid conf " |
901 | "value (see above).\n" ); |
902 | abort(); |
903 | } |
904 | |
905 | static void |
906 | malloc_conf_error(const char *msg, const char *k, size_t klen, const char *v, |
907 | size_t vlen) { |
908 | malloc_printf("<jemalloc>: %s: %.*s:%.*s\n" , msg, (int)klen, k, |
909 | (int)vlen, v); |
910 | /* If abort_conf is set, error out after processing all options. */ |
911 | const char *experimental = "experimental_" ; |
912 | if (strncmp(k, experimental, strlen(experimental)) == 0) { |
913 | /* However, tolerate experimental features. */ |
914 | return; |
915 | } |
916 | had_conf_error = true; |
917 | } |
918 | |
919 | static void |
920 | malloc_slow_flag_init(void) { |
921 | /* |
922 | * Combine the runtime options into malloc_slow for fast path. Called |
923 | * after processing all the options. |
924 | */ |
925 | malloc_slow_flags |= (opt_junk_alloc ? flag_opt_junk_alloc : 0) |
926 | | (opt_junk_free ? flag_opt_junk_free : 0) |
927 | | (opt_zero ? flag_opt_zero : 0) |
928 | | (opt_utrace ? flag_opt_utrace : 0) |
929 | | (opt_xmalloc ? flag_opt_xmalloc : 0); |
930 | |
931 | malloc_slow = (malloc_slow_flags != 0); |
932 | } |
933 | |
934 | /* Number of sources for initializing malloc_conf */ |
935 | #define MALLOC_CONF_NSOURCES 4 |
936 | |
937 | static const char * |
938 | obtain_malloc_conf(unsigned which_source, char buf[PATH_MAX + 1]) { |
939 | if (config_debug) { |
940 | static unsigned read_source = 0; |
941 | /* |
942 | * Each source should only be read once, to minimize # of |
943 | * syscalls on init. |
944 | */ |
945 | assert(read_source++ == which_source); |
946 | } |
947 | assert(which_source < MALLOC_CONF_NSOURCES); |
948 | |
949 | const char *ret; |
950 | switch (which_source) { |
951 | case 0: |
952 | ret = config_malloc_conf; |
953 | break; |
954 | case 1: |
955 | if (je_malloc_conf != NULL) { |
956 | /* Use options that were compiled into the program. */ |
957 | ret = je_malloc_conf; |
958 | } else { |
959 | /* No configuration specified. */ |
960 | ret = NULL; |
961 | } |
962 | break; |
963 | case 2: { |
964 | ssize_t linklen = 0; |
965 | #ifndef _WIN32 |
966 | int saved_errno = errno; |
967 | const char *linkname = |
968 | # ifdef JEMALLOC_PREFIX |
969 | "/etc/" JEMALLOC_PREFIX"malloc.conf" |
970 | # else |
971 | "/etc/malloc.conf" |
972 | # endif |
973 | ; |
974 | |
975 | /* |
976 | * Try to use the contents of the "/etc/malloc.conf" symbolic |
977 | * link's name. |
978 | */ |
979 | #ifndef JEMALLOC_READLINKAT |
980 | linklen = readlink(linkname, buf, PATH_MAX); |
981 | #else |
982 | linklen = readlinkat(AT_FDCWD, linkname, buf, PATH_MAX); |
983 | #endif |
984 | if (linklen == -1) { |
985 | /* No configuration specified. */ |
986 | linklen = 0; |
987 | /* Restore errno. */ |
988 | set_errno(saved_errno); |
989 | } |
990 | #endif |
991 | buf[linklen] = '\0'; |
992 | ret = buf; |
993 | break; |
994 | } case 3: { |
995 | const char *envname = |
996 | #ifdef JEMALLOC_PREFIX |
997 | JEMALLOC_CPREFIX"MALLOC_CONF" |
998 | #else |
999 | "MALLOC_CONF" |
1000 | #endif |
1001 | ; |
1002 | |
1003 | if ((ret = jemalloc_secure_getenv(envname)) != NULL) { |
1004 | /* |
1005 | * Do nothing; opts is already initialized to the value |
1006 | * of the MALLOC_CONF environment variable. |
1007 | */ |
1008 | } else { |
1009 | /* No configuration specified. */ |
1010 | ret = NULL; |
1011 | } |
1012 | break; |
1013 | } default: |
1014 | not_reached(); |
1015 | ret = NULL; |
1016 | } |
1017 | return ret; |
1018 | } |
1019 | |
1020 | static void |
1021 | malloc_conf_init_helper(sc_data_t *sc_data, unsigned bin_shard_sizes[SC_NBINS], |
1022 | bool initial_call, const char *opts_cache[MALLOC_CONF_NSOURCES], |
1023 | char buf[PATH_MAX + 1]) { |
1024 | static const char *opts_explain[MALLOC_CONF_NSOURCES] = { |
1025 | "string specified via --with-malloc-conf" , |
1026 | "string pointed to by the global variable malloc_conf" , |
1027 | "\"name\" of the file referenced by the symbolic link named " |
1028 | "/etc/malloc.conf" , |
1029 | "value of the environment variable MALLOC_CONF" |
1030 | }; |
1031 | unsigned i; |
1032 | const char *opts, *k, *v; |
1033 | size_t klen, vlen; |
1034 | |
1035 | for (i = 0; i < MALLOC_CONF_NSOURCES; i++) { |
1036 | /* Get runtime configuration. */ |
1037 | if (initial_call) { |
1038 | opts_cache[i] = obtain_malloc_conf(i, buf); |
1039 | } |
1040 | opts = opts_cache[i]; |
1041 | if (!initial_call && opt_confirm_conf) { |
1042 | malloc_printf( |
1043 | "<jemalloc>: malloc_conf #%u (%s): \"%s\"\n" , |
1044 | i + 1, opts_explain[i], opts != NULL ? opts : "" ); |
1045 | } |
1046 | if (opts == NULL) { |
1047 | continue; |
1048 | } |
1049 | |
1050 | while (*opts != '\0' && !malloc_conf_next(&opts, &k, &klen, &v, |
1051 | &vlen)) { |
1052 | |
1053 | #define CONF_ERROR(msg, k, klen, v, vlen) \ |
1054 | if (!initial_call) { \ |
1055 | malloc_conf_error( \ |
1056 | msg, k, klen, v, vlen); \ |
1057 | cur_opt_valid = false; \ |
1058 | } |
1059 | #define CONF_CONTINUE { \ |
1060 | if (!initial_call && opt_confirm_conf \ |
1061 | && cur_opt_valid) { \ |
1062 | malloc_printf("<jemalloc>: -- " \ |
1063 | "Set conf value: %.*s:%.*s" \ |
1064 | "\n", (int)klen, k, \ |
1065 | (int)vlen, v); \ |
1066 | } \ |
1067 | continue; \ |
1068 | } |
1069 | #define CONF_MATCH(n) \ |
1070 | (sizeof(n)-1 == klen && strncmp(n, k, klen) == 0) |
1071 | #define CONF_MATCH_VALUE(n) \ |
1072 | (sizeof(n)-1 == vlen && strncmp(n, v, vlen) == 0) |
1073 | #define CONF_HANDLE_BOOL(o, n) \ |
1074 | if (CONF_MATCH(n)) { \ |
1075 | if (CONF_MATCH_VALUE("true")) { \ |
1076 | o = true; \ |
1077 | } else if (CONF_MATCH_VALUE("false")) { \ |
1078 | o = false; \ |
1079 | } else { \ |
1080 | CONF_ERROR("Invalid conf value",\ |
1081 | k, klen, v, vlen); \ |
1082 | } \ |
1083 | CONF_CONTINUE; \ |
1084 | } |
1085 | /* |
1086 | * One of the CONF_MIN macros below expands, in one of the use points, |
1087 | * to "unsigned integer < 0", which is always false, triggering the |
1088 | * GCC -Wtype-limits warning, which we disable here and re-enable below. |
1089 | */ |
1090 | JEMALLOC_DIAGNOSTIC_PUSH |
1091 | JEMALLOC_DIAGNOSTIC_IGNORE_TYPE_LIMITS |
1092 | |
1093 | #define CONF_DONT_CHECK_MIN(um, min) false |
1094 | #define CONF_CHECK_MIN(um, min) ((um) < (min)) |
1095 | #define CONF_DONT_CHECK_MAX(um, max) false |
1096 | #define CONF_CHECK_MAX(um, max) ((um) > (max)) |
1097 | #define CONF_HANDLE_T_U(t, o, n, min, max, check_min, check_max, clip) \ |
1098 | if (CONF_MATCH(n)) { \ |
1099 | uintmax_t um; \ |
1100 | char *end; \ |
1101 | \ |
1102 | set_errno(0); \ |
1103 | um = malloc_strtoumax(v, &end, 0); \ |
1104 | if (get_errno() != 0 || (uintptr_t)end -\ |
1105 | (uintptr_t)v != vlen) { \ |
1106 | CONF_ERROR("Invalid conf value",\ |
1107 | k, klen, v, vlen); \ |
1108 | } else if (clip) { \ |
1109 | if (check_min(um, (t)(min))) { \ |
1110 | o = (t)(min); \ |
1111 | } else if ( \ |
1112 | check_max(um, (t)(max))) { \ |
1113 | o = (t)(max); \ |
1114 | } else { \ |
1115 | o = (t)um; \ |
1116 | } \ |
1117 | } else { \ |
1118 | if (check_min(um, (t)(min)) || \ |
1119 | check_max(um, (t)(max))) { \ |
1120 | CONF_ERROR( \ |
1121 | "Out-of-range " \ |
1122 | "conf value", \ |
1123 | k, klen, v, vlen); \ |
1124 | } else { \ |
1125 | o = (t)um; \ |
1126 | } \ |
1127 | } \ |
1128 | CONF_CONTINUE; \ |
1129 | } |
1130 | #define CONF_HANDLE_UNSIGNED(o, n, min, max, check_min, check_max, \ |
1131 | clip) \ |
1132 | CONF_HANDLE_T_U(unsigned, o, n, min, max, \ |
1133 | check_min, check_max, clip) |
1134 | #define CONF_HANDLE_SIZE_T(o, n, min, max, check_min, check_max, clip) \ |
1135 | CONF_HANDLE_T_U(size_t, o, n, min, max, \ |
1136 | check_min, check_max, clip) |
1137 | #define CONF_HANDLE_SSIZE_T(o, n, min, max) \ |
1138 | if (CONF_MATCH(n)) { \ |
1139 | long l; \ |
1140 | char *end; \ |
1141 | \ |
1142 | set_errno(0); \ |
1143 | l = strtol(v, &end, 0); \ |
1144 | if (get_errno() != 0 || (uintptr_t)end -\ |
1145 | (uintptr_t)v != vlen) { \ |
1146 | CONF_ERROR("Invalid conf value",\ |
1147 | k, klen, v, vlen); \ |
1148 | } else if (l < (ssize_t)(min) || l > \ |
1149 | (ssize_t)(max)) { \ |
1150 | CONF_ERROR( \ |
1151 | "Out-of-range conf value", \ |
1152 | k, klen, v, vlen); \ |
1153 | } else { \ |
1154 | o = l; \ |
1155 | } \ |
1156 | CONF_CONTINUE; \ |
1157 | } |
1158 | #define CONF_HANDLE_CHAR_P(o, n, d) \ |
1159 | if (CONF_MATCH(n)) { \ |
1160 | size_t cpylen = (vlen <= \ |
1161 | sizeof(o)-1) ? vlen : \ |
1162 | sizeof(o)-1; \ |
1163 | strncpy(o, v, cpylen); \ |
1164 | o[cpylen] = '\0'; \ |
1165 | CONF_CONTINUE; \ |
1166 | } |
1167 | |
1168 | bool cur_opt_valid = true; |
1169 | |
1170 | CONF_HANDLE_BOOL(opt_confirm_conf, "confirm_conf" ) |
1171 | if (initial_call) { |
1172 | continue; |
1173 | } |
1174 | |
1175 | CONF_HANDLE_BOOL(opt_abort, "abort" ) |
1176 | CONF_HANDLE_BOOL(opt_abort_conf, "abort_conf" ) |
1177 | if (strncmp("metadata_thp" , k, klen) == 0) { |
1178 | int i; |
1179 | bool match = false; |
1180 | for (i = 0; i < metadata_thp_mode_limit; i++) { |
1181 | if (strncmp(metadata_thp_mode_names[i], |
1182 | v, vlen) == 0) { |
1183 | opt_metadata_thp = i; |
1184 | match = true; |
1185 | break; |
1186 | } |
1187 | } |
1188 | if (!match) { |
1189 | CONF_ERROR("Invalid conf value" , |
1190 | k, klen, v, vlen); |
1191 | } |
1192 | CONF_CONTINUE; |
1193 | } |
1194 | CONF_HANDLE_BOOL(opt_retain, "retain" ) |
1195 | if (strncmp("dss" , k, klen) == 0) { |
1196 | int i; |
1197 | bool match = false; |
1198 | for (i = 0; i < dss_prec_limit; i++) { |
1199 | if (strncmp(dss_prec_names[i], v, vlen) |
1200 | == 0) { |
1201 | if (extent_dss_prec_set(i)) { |
1202 | CONF_ERROR( |
1203 | "Error setting dss" , |
1204 | k, klen, v, vlen); |
1205 | } else { |
1206 | opt_dss = |
1207 | dss_prec_names[i]; |
1208 | match = true; |
1209 | break; |
1210 | } |
1211 | } |
1212 | } |
1213 | if (!match) { |
1214 | CONF_ERROR("Invalid conf value" , |
1215 | k, klen, v, vlen); |
1216 | } |
1217 | CONF_CONTINUE; |
1218 | } |
1219 | CONF_HANDLE_UNSIGNED(opt_narenas, "narenas" , 1, |
1220 | UINT_MAX, CONF_CHECK_MIN, CONF_DONT_CHECK_MAX, |
1221 | false) |
1222 | if (CONF_MATCH("bin_shards" )) { |
1223 | const char *bin_shards_segment_cur = v; |
1224 | size_t vlen_left = vlen; |
1225 | do { |
1226 | size_t size_start; |
1227 | size_t size_end; |
1228 | size_t nshards; |
1229 | bool err = malloc_conf_multi_sizes_next( |
1230 | &bin_shards_segment_cur, &vlen_left, |
1231 | &size_start, &size_end, &nshards); |
1232 | if (err || bin_update_shard_size( |
1233 | bin_shard_sizes, size_start, |
1234 | size_end, nshards)) { |
1235 | CONF_ERROR( |
1236 | "Invalid settings for " |
1237 | "bin_shards" , k, klen, v, |
1238 | vlen); |
1239 | break; |
1240 | } |
1241 | } while (vlen_left > 0); |
1242 | CONF_CONTINUE; |
1243 | } |
1244 | CONF_HANDLE_SSIZE_T(opt_dirty_decay_ms, |
1245 | "dirty_decay_ms" , -1, NSTIME_SEC_MAX * KQU(1000) < |
1246 | QU(SSIZE_MAX) ? NSTIME_SEC_MAX * KQU(1000) : |
1247 | SSIZE_MAX); |
1248 | CONF_HANDLE_SSIZE_T(opt_muzzy_decay_ms, |
1249 | "muzzy_decay_ms" , -1, NSTIME_SEC_MAX * KQU(1000) < |
1250 | QU(SSIZE_MAX) ? NSTIME_SEC_MAX * KQU(1000) : |
1251 | SSIZE_MAX); |
1252 | CONF_HANDLE_BOOL(opt_stats_print, "stats_print" ) |
1253 | if (CONF_MATCH("stats_print_opts" )) { |
1254 | init_opt_stats_print_opts(v, vlen); |
1255 | CONF_CONTINUE; |
1256 | } |
1257 | if (config_fill) { |
1258 | if (CONF_MATCH("junk" )) { |
1259 | if (CONF_MATCH_VALUE("true" )) { |
1260 | opt_junk = "true" ; |
1261 | opt_junk_alloc = opt_junk_free = |
1262 | true; |
1263 | } else if (CONF_MATCH_VALUE("false" )) { |
1264 | opt_junk = "false" ; |
1265 | opt_junk_alloc = opt_junk_free = |
1266 | false; |
1267 | } else if (CONF_MATCH_VALUE("alloc" )) { |
1268 | opt_junk = "alloc" ; |
1269 | opt_junk_alloc = true; |
1270 | opt_junk_free = false; |
1271 | } else if (CONF_MATCH_VALUE("free" )) { |
1272 | opt_junk = "free" ; |
1273 | opt_junk_alloc = false; |
1274 | opt_junk_free = true; |
1275 | } else { |
1276 | CONF_ERROR( |
1277 | "Invalid conf value" , |
1278 | k, klen, v, vlen); |
1279 | } |
1280 | CONF_CONTINUE; |
1281 | } |
1282 | CONF_HANDLE_BOOL(opt_zero, "zero" ) |
1283 | } |
1284 | if (config_utrace) { |
1285 | CONF_HANDLE_BOOL(opt_utrace, "utrace" ) |
1286 | } |
1287 | if (config_xmalloc) { |
1288 | CONF_HANDLE_BOOL(opt_xmalloc, "xmalloc" ) |
1289 | } |
1290 | CONF_HANDLE_BOOL(opt_tcache, "tcache" ) |
1291 | CONF_HANDLE_SSIZE_T(opt_lg_tcache_max, "lg_tcache_max" , |
1292 | -1, (sizeof(size_t) << 3) - 1) |
1293 | |
1294 | /* |
1295 | * The runtime option of oversize_threshold remains |
1296 | * undocumented. It may be tweaked in the next major |
1297 | * release (6.0). The default value 8M is rather |
1298 | * conservative / safe. Tuning it further down may |
1299 | * improve fragmentation a bit more, but may also cause |
1300 | * contention on the huge arena. |
1301 | */ |
1302 | CONF_HANDLE_SIZE_T(opt_oversize_threshold, |
1303 | "oversize_threshold" , 0, SC_LARGE_MAXCLASS, |
1304 | CONF_DONT_CHECK_MIN, CONF_CHECK_MAX, false) |
1305 | CONF_HANDLE_SIZE_T(opt_lg_extent_max_active_fit, |
1306 | "lg_extent_max_active_fit" , 0, |
1307 | (sizeof(size_t) << 3), CONF_DONT_CHECK_MIN, |
1308 | CONF_CHECK_MAX, false) |
1309 | |
1310 | if (strncmp("percpu_arena" , k, klen) == 0) { |
1311 | bool match = false; |
1312 | for (int i = percpu_arena_mode_names_base; i < |
1313 | percpu_arena_mode_names_limit; i++) { |
1314 | if (strncmp(percpu_arena_mode_names[i], |
1315 | v, vlen) == 0) { |
1316 | if (!have_percpu_arena) { |
1317 | CONF_ERROR( |
1318 | "No getcpu support" , |
1319 | k, klen, v, vlen); |
1320 | } |
1321 | opt_percpu_arena = i; |
1322 | match = true; |
1323 | break; |
1324 | } |
1325 | } |
1326 | if (!match) { |
1327 | CONF_ERROR("Invalid conf value" , |
1328 | k, klen, v, vlen); |
1329 | } |
1330 | CONF_CONTINUE; |
1331 | } |
1332 | CONF_HANDLE_BOOL(opt_background_thread, |
1333 | "background_thread" ); |
1334 | CONF_HANDLE_SIZE_T(opt_max_background_threads, |
1335 | "max_background_threads" , 1, |
1336 | opt_max_background_threads, |
1337 | CONF_CHECK_MIN, CONF_CHECK_MAX, |
1338 | true); |
1339 | if (CONF_MATCH("slab_sizes" )) { |
1340 | bool err; |
1341 | const char *slab_size_segment_cur = v; |
1342 | size_t vlen_left = vlen; |
1343 | do { |
1344 | size_t slab_start; |
1345 | size_t slab_end; |
1346 | size_t pgs; |
1347 | err = malloc_conf_multi_sizes_next( |
1348 | &slab_size_segment_cur, |
1349 | &vlen_left, &slab_start, &slab_end, |
1350 | &pgs); |
1351 | if (!err) { |
1352 | sc_data_update_slab_size( |
1353 | sc_data, slab_start, |
1354 | slab_end, (int)pgs); |
1355 | } else { |
1356 | CONF_ERROR("Invalid settings " |
1357 | "for slab_sizes" , |
1358 | k, klen, v, vlen); |
1359 | } |
1360 | } while (!err && vlen_left > 0); |
1361 | CONF_CONTINUE; |
1362 | } |
1363 | if (config_prof) { |
1364 | CONF_HANDLE_BOOL(opt_prof, "prof" ) |
1365 | CONF_HANDLE_CHAR_P(opt_prof_prefix, |
1366 | "prof_prefix" , "jeprof" ) |
1367 | CONF_HANDLE_BOOL(opt_prof_active, "prof_active" ) |
1368 | CONF_HANDLE_BOOL(opt_prof_thread_active_init, |
1369 | "prof_thread_active_init" ) |
1370 | CONF_HANDLE_SIZE_T(opt_lg_prof_sample, |
1371 | "lg_prof_sample" , 0, (sizeof(uint64_t) << 3) |
1372 | - 1, CONF_DONT_CHECK_MIN, CONF_CHECK_MAX, |
1373 | true) |
1374 | CONF_HANDLE_BOOL(opt_prof_accum, "prof_accum" ) |
1375 | CONF_HANDLE_SSIZE_T(opt_lg_prof_interval, |
1376 | "lg_prof_interval" , -1, |
1377 | (sizeof(uint64_t) << 3) - 1) |
1378 | CONF_HANDLE_BOOL(opt_prof_gdump, "prof_gdump" ) |
1379 | CONF_HANDLE_BOOL(opt_prof_final, "prof_final" ) |
1380 | CONF_HANDLE_BOOL(opt_prof_leak, "prof_leak" ) |
1381 | CONF_HANDLE_BOOL(opt_prof_log, "prof_log" ) |
1382 | } |
1383 | if (config_log) { |
1384 | if (CONF_MATCH("log" )) { |
1385 | size_t cpylen = ( |
1386 | vlen <= sizeof(log_var_names) ? |
1387 | vlen : sizeof(log_var_names) - 1); |
1388 | strncpy(log_var_names, v, cpylen); |
1389 | log_var_names[cpylen] = '\0'; |
1390 | CONF_CONTINUE; |
1391 | } |
1392 | } |
1393 | if (CONF_MATCH("thp" )) { |
1394 | bool match = false; |
1395 | for (int i = 0; i < thp_mode_names_limit; i++) { |
1396 | if (strncmp(thp_mode_names[i],v, vlen) |
1397 | == 0) { |
1398 | if (!have_madvise_huge) { |
1399 | CONF_ERROR( |
1400 | "No THP support" , |
1401 | k, klen, v, vlen); |
1402 | } |
1403 | opt_thp = i; |
1404 | match = true; |
1405 | break; |
1406 | } |
1407 | } |
1408 | if (!match) { |
1409 | CONF_ERROR("Invalid conf value" , |
1410 | k, klen, v, vlen); |
1411 | } |
1412 | CONF_CONTINUE; |
1413 | } |
1414 | CONF_ERROR("Invalid conf pair" , k, klen, v, vlen); |
1415 | #undef CONF_ERROR |
1416 | #undef CONF_CONTINUE |
1417 | #undef CONF_MATCH |
1418 | #undef CONF_MATCH_VALUE |
1419 | #undef CONF_HANDLE_BOOL |
1420 | #undef CONF_DONT_CHECK_MIN |
1421 | #undef CONF_CHECK_MIN |
1422 | #undef CONF_DONT_CHECK_MAX |
1423 | #undef CONF_CHECK_MAX |
1424 | #undef CONF_HANDLE_T_U |
1425 | #undef CONF_HANDLE_UNSIGNED |
1426 | #undef CONF_HANDLE_SIZE_T |
1427 | #undef CONF_HANDLE_SSIZE_T |
1428 | #undef CONF_HANDLE_CHAR_P |
1429 | /* Re-enable diagnostic "-Wtype-limits" */ |
1430 | JEMALLOC_DIAGNOSTIC_POP |
1431 | } |
1432 | if (opt_abort_conf && had_conf_error) { |
1433 | malloc_abort_invalid_conf(); |
1434 | } |
1435 | } |
1436 | atomic_store_b(&log_init_done, true, ATOMIC_RELEASE); |
1437 | } |
1438 | |
1439 | static void |
1440 | malloc_conf_init(sc_data_t *sc_data, unsigned bin_shard_sizes[SC_NBINS]) { |
1441 | const char *opts_cache[MALLOC_CONF_NSOURCES] = {NULL, NULL, NULL, NULL}; |
1442 | char buf[PATH_MAX + 1]; |
1443 | |
1444 | /* The first call only set the confirm_conf option and opts_cache */ |
1445 | malloc_conf_init_helper(NULL, NULL, true, opts_cache, buf); |
1446 | malloc_conf_init_helper(sc_data, bin_shard_sizes, false, opts_cache, |
1447 | NULL); |
1448 | } |
1449 | |
1450 | #undef MALLOC_CONF_NSOURCES |
1451 | |
1452 | static bool |
1453 | malloc_init_hard_needed(void) { |
1454 | if (malloc_initialized() || (IS_INITIALIZER && malloc_init_state == |
1455 | malloc_init_recursible)) { |
1456 | /* |
1457 | * Another thread initialized the allocator before this one |
1458 | * acquired init_lock, or this thread is the initializing |
1459 | * thread, and it is recursively allocating. |
1460 | */ |
1461 | return false; |
1462 | } |
1463 | #ifdef JEMALLOC_THREADED_INIT |
1464 | if (malloc_initializer != NO_INITIALIZER && !IS_INITIALIZER) { |
1465 | /* Busy-wait until the initializing thread completes. */ |
1466 | spin_t spinner = SPIN_INITIALIZER; |
1467 | do { |
1468 | malloc_mutex_unlock(TSDN_NULL, &init_lock); |
1469 | spin_adaptive(&spinner); |
1470 | malloc_mutex_lock(TSDN_NULL, &init_lock); |
1471 | } while (!malloc_initialized()); |
1472 | return false; |
1473 | } |
1474 | #endif |
1475 | return true; |
1476 | } |
1477 | |
1478 | static bool |
1479 | malloc_init_hard_a0_locked() { |
1480 | malloc_initializer = INITIALIZER; |
1481 | |
1482 | JEMALLOC_DIAGNOSTIC_PUSH |
1483 | JEMALLOC_DIAGNOSTIC_IGNORE_MISSING_STRUCT_FIELD_INITIALIZERS |
1484 | sc_data_t sc_data = {0}; |
1485 | JEMALLOC_DIAGNOSTIC_POP |
1486 | |
1487 | /* |
1488 | * Ordering here is somewhat tricky; we need sc_boot() first, since that |
1489 | * determines what the size classes will be, and then |
1490 | * malloc_conf_init(), since any slab size tweaking will need to be done |
1491 | * before sz_boot and bin_boot, which assume that the values they read |
1492 | * out of sc_data_global are final. |
1493 | */ |
1494 | sc_boot(&sc_data); |
1495 | unsigned bin_shard_sizes[SC_NBINS]; |
1496 | bin_shard_sizes_boot(bin_shard_sizes); |
1497 | /* |
1498 | * prof_boot0 only initializes opt_prof_prefix. We need to do it before |
1499 | * we parse malloc_conf options, in case malloc_conf parsing overwrites |
1500 | * it. |
1501 | */ |
1502 | if (config_prof) { |
1503 | prof_boot0(); |
1504 | } |
1505 | malloc_conf_init(&sc_data, bin_shard_sizes); |
1506 | sz_boot(&sc_data); |
1507 | bin_boot(&sc_data, bin_shard_sizes); |
1508 | |
1509 | if (opt_stats_print) { |
1510 | /* Print statistics at exit. */ |
1511 | if (atexit(stats_print_atexit) != 0) { |
1512 | malloc_write("<jemalloc>: Error in atexit()\n" ); |
1513 | if (opt_abort) { |
1514 | abort(); |
1515 | } |
1516 | } |
1517 | } |
1518 | if (pages_boot()) { |
1519 | return true; |
1520 | } |
1521 | if (base_boot(TSDN_NULL)) { |
1522 | return true; |
1523 | } |
1524 | if (extent_boot()) { |
1525 | return true; |
1526 | } |
1527 | if (ctl_boot()) { |
1528 | return true; |
1529 | } |
1530 | if (config_prof) { |
1531 | prof_boot1(); |
1532 | } |
1533 | arena_boot(&sc_data); |
1534 | if (tcache_boot(TSDN_NULL)) { |
1535 | return true; |
1536 | } |
1537 | if (malloc_mutex_init(&arenas_lock, "arenas" , WITNESS_RANK_ARENAS, |
1538 | malloc_mutex_rank_exclusive)) { |
1539 | return true; |
1540 | } |
1541 | hook_boot(); |
1542 | /* |
1543 | * Create enough scaffolding to allow recursive allocation in |
1544 | * malloc_ncpus(). |
1545 | */ |
1546 | narenas_auto = 1; |
1547 | manual_arena_base = narenas_auto + 1; |
1548 | memset(arenas, 0, sizeof(arena_t *) * narenas_auto); |
1549 | /* |
1550 | * Initialize one arena here. The rest are lazily created in |
1551 | * arena_choose_hard(). |
1552 | */ |
1553 | if (arena_init(TSDN_NULL, 0, (extent_hooks_t *)&extent_hooks_default) |
1554 | == NULL) { |
1555 | return true; |
1556 | } |
1557 | a0 = arena_get(TSDN_NULL, 0, false); |
1558 | malloc_init_state = malloc_init_a0_initialized; |
1559 | |
1560 | return false; |
1561 | } |
1562 | |
1563 | static bool |
1564 | malloc_init_hard_a0(void) { |
1565 | bool ret; |
1566 | |
1567 | malloc_mutex_lock(TSDN_NULL, &init_lock); |
1568 | ret = malloc_init_hard_a0_locked(); |
1569 | malloc_mutex_unlock(TSDN_NULL, &init_lock); |
1570 | return ret; |
1571 | } |
1572 | |
1573 | /* Initialize data structures which may trigger recursive allocation. */ |
1574 | static bool |
1575 | malloc_init_hard_recursible(void) { |
1576 | malloc_init_state = malloc_init_recursible; |
1577 | |
1578 | ncpus = malloc_ncpus(); |
1579 | |
1580 | #if (defined(JEMALLOC_HAVE_PTHREAD_ATFORK) && !defined(JEMALLOC_MUTEX_INIT_CB) \ |
1581 | && !defined(JEMALLOC_ZONE) && !defined(_WIN32) && \ |
1582 | !defined(__native_client__)) |
1583 | /* LinuxThreads' pthread_atfork() allocates. */ |
1584 | if (pthread_atfork(jemalloc_prefork, jemalloc_postfork_parent, |
1585 | jemalloc_postfork_child) != 0) { |
1586 | malloc_write("<jemalloc>: Error in pthread_atfork()\n" ); |
1587 | if (opt_abort) { |
1588 | abort(); |
1589 | } |
1590 | return true; |
1591 | } |
1592 | #endif |
1593 | |
1594 | if (background_thread_boot0()) { |
1595 | return true; |
1596 | } |
1597 | |
1598 | return false; |
1599 | } |
1600 | |
1601 | static unsigned |
1602 | malloc_narenas_default(void) { |
1603 | assert(ncpus > 0); |
1604 | /* |
1605 | * For SMP systems, create more than one arena per CPU by |
1606 | * default. |
1607 | */ |
1608 | if (ncpus > 1) { |
1609 | return ncpus << 2; |
1610 | } else { |
1611 | return 1; |
1612 | } |
1613 | } |
1614 | |
1615 | static percpu_arena_mode_t |
1616 | percpu_arena_as_initialized(percpu_arena_mode_t mode) { |
1617 | assert(!malloc_initialized()); |
1618 | assert(mode <= percpu_arena_disabled); |
1619 | |
1620 | if (mode != percpu_arena_disabled) { |
1621 | mode += percpu_arena_mode_enabled_base; |
1622 | } |
1623 | |
1624 | return mode; |
1625 | } |
1626 | |
1627 | static bool |
1628 | malloc_init_narenas(void) { |
1629 | assert(ncpus > 0); |
1630 | |
1631 | if (opt_percpu_arena != percpu_arena_disabled) { |
1632 | if (!have_percpu_arena || malloc_getcpu() < 0) { |
1633 | opt_percpu_arena = percpu_arena_disabled; |
1634 | malloc_printf("<jemalloc>: perCPU arena getcpu() not " |
1635 | "available. Setting narenas to %u.\n" , opt_narenas ? |
1636 | opt_narenas : malloc_narenas_default()); |
1637 | if (opt_abort) { |
1638 | abort(); |
1639 | } |
1640 | } else { |
1641 | if (ncpus >= MALLOCX_ARENA_LIMIT) { |
1642 | malloc_printf("<jemalloc>: narenas w/ percpu" |
1643 | "arena beyond limit (%d)\n" , ncpus); |
1644 | if (opt_abort) { |
1645 | abort(); |
1646 | } |
1647 | return true; |
1648 | } |
1649 | /* NB: opt_percpu_arena isn't fully initialized yet. */ |
1650 | if (percpu_arena_as_initialized(opt_percpu_arena) == |
1651 | per_phycpu_arena && ncpus % 2 != 0) { |
1652 | malloc_printf("<jemalloc>: invalid " |
1653 | "configuration -- per physical CPU arena " |
1654 | "with odd number (%u) of CPUs (no hyper " |
1655 | "threading?).\n" , ncpus); |
1656 | if (opt_abort) |
1657 | abort(); |
1658 | } |
1659 | unsigned n = percpu_arena_ind_limit( |
1660 | percpu_arena_as_initialized(opt_percpu_arena)); |
1661 | if (opt_narenas < n) { |
1662 | /* |
1663 | * If narenas is specified with percpu_arena |
1664 | * enabled, actual narenas is set as the greater |
1665 | * of the two. percpu_arena_choose will be free |
1666 | * to use any of the arenas based on CPU |
1667 | * id. This is conservative (at a small cost) |
1668 | * but ensures correctness. |
1669 | * |
1670 | * If for some reason the ncpus determined at |
1671 | * boot is not the actual number (e.g. because |
1672 | * of affinity setting from numactl), reserving |
1673 | * narenas this way provides a workaround for |
1674 | * percpu_arena. |
1675 | */ |
1676 | opt_narenas = n; |
1677 | } |
1678 | } |
1679 | } |
1680 | if (opt_narenas == 0) { |
1681 | opt_narenas = malloc_narenas_default(); |
1682 | } |
1683 | assert(opt_narenas > 0); |
1684 | |
1685 | narenas_auto = opt_narenas; |
1686 | /* |
1687 | * Limit the number of arenas to the indexing range of MALLOCX_ARENA(). |
1688 | */ |
1689 | if (narenas_auto >= MALLOCX_ARENA_LIMIT) { |
1690 | narenas_auto = MALLOCX_ARENA_LIMIT - 1; |
1691 | malloc_printf("<jemalloc>: Reducing narenas to limit (%d)\n" , |
1692 | narenas_auto); |
1693 | } |
1694 | narenas_total_set(narenas_auto); |
1695 | if (arena_init_huge()) { |
1696 | narenas_total_inc(); |
1697 | } |
1698 | manual_arena_base = narenas_total_get(); |
1699 | |
1700 | return false; |
1701 | } |
1702 | |
1703 | static void |
1704 | malloc_init_percpu(void) { |
1705 | opt_percpu_arena = percpu_arena_as_initialized(opt_percpu_arena); |
1706 | } |
1707 | |
1708 | static bool |
1709 | malloc_init_hard_finish(void) { |
1710 | if (malloc_mutex_boot()) { |
1711 | return true; |
1712 | } |
1713 | |
1714 | malloc_init_state = malloc_init_initialized; |
1715 | malloc_slow_flag_init(); |
1716 | |
1717 | return false; |
1718 | } |
1719 | |
1720 | static void |
1721 | malloc_init_hard_cleanup(tsdn_t *tsdn, bool reentrancy_set) { |
1722 | malloc_mutex_assert_owner(tsdn, &init_lock); |
1723 | malloc_mutex_unlock(tsdn, &init_lock); |
1724 | if (reentrancy_set) { |
1725 | assert(!tsdn_null(tsdn)); |
1726 | tsd_t *tsd = tsdn_tsd(tsdn); |
1727 | assert(tsd_reentrancy_level_get(tsd) > 0); |
1728 | post_reentrancy(tsd); |
1729 | } |
1730 | } |
1731 | |
1732 | static bool |
1733 | malloc_init_hard(void) { |
1734 | tsd_t *tsd; |
1735 | |
1736 | #if defined(_WIN32) && _WIN32_WINNT < 0x0600 |
1737 | _init_init_lock(); |
1738 | #endif |
1739 | malloc_mutex_lock(TSDN_NULL, &init_lock); |
1740 | |
1741 | #define UNLOCK_RETURN(tsdn, ret, reentrancy) \ |
1742 | malloc_init_hard_cleanup(tsdn, reentrancy); \ |
1743 | return ret; |
1744 | |
1745 | if (!malloc_init_hard_needed()) { |
1746 | UNLOCK_RETURN(TSDN_NULL, false, false) |
1747 | } |
1748 | |
1749 | if (malloc_init_state != malloc_init_a0_initialized && |
1750 | malloc_init_hard_a0_locked()) { |
1751 | UNLOCK_RETURN(TSDN_NULL, true, false) |
1752 | } |
1753 | |
1754 | malloc_mutex_unlock(TSDN_NULL, &init_lock); |
1755 | /* Recursive allocation relies on functional tsd. */ |
1756 | tsd = malloc_tsd_boot0(); |
1757 | if (tsd == NULL) { |
1758 | return true; |
1759 | } |
1760 | if (malloc_init_hard_recursible()) { |
1761 | return true; |
1762 | } |
1763 | |
1764 | malloc_mutex_lock(tsd_tsdn(tsd), &init_lock); |
1765 | /* Set reentrancy level to 1 during init. */ |
1766 | pre_reentrancy(tsd, NULL); |
1767 | /* Initialize narenas before prof_boot2 (for allocation). */ |
1768 | if (malloc_init_narenas() || background_thread_boot1(tsd_tsdn(tsd))) { |
1769 | UNLOCK_RETURN(tsd_tsdn(tsd), true, true) |
1770 | } |
1771 | if (config_prof && prof_boot2(tsd)) { |
1772 | UNLOCK_RETURN(tsd_tsdn(tsd), true, true) |
1773 | } |
1774 | |
1775 | malloc_init_percpu(); |
1776 | |
1777 | if (malloc_init_hard_finish()) { |
1778 | UNLOCK_RETURN(tsd_tsdn(tsd), true, true) |
1779 | } |
1780 | post_reentrancy(tsd); |
1781 | malloc_mutex_unlock(tsd_tsdn(tsd), &init_lock); |
1782 | |
1783 | witness_assert_lockless(witness_tsd_tsdn( |
1784 | tsd_witness_tsdp_get_unsafe(tsd))); |
1785 | malloc_tsd_boot1(); |
1786 | /* Update TSD after tsd_boot1. */ |
1787 | tsd = tsd_fetch(); |
1788 | if (opt_background_thread) { |
1789 | assert(have_background_thread); |
1790 | /* |
1791 | * Need to finish init & unlock first before creating background |
1792 | * threads (pthread_create depends on malloc). ctl_init (which |
1793 | * sets isthreaded) needs to be called without holding any lock. |
1794 | */ |
1795 | background_thread_ctl_init(tsd_tsdn(tsd)); |
1796 | if (background_thread_create(tsd, 0)) { |
1797 | return true; |
1798 | } |
1799 | } |
1800 | #undef UNLOCK_RETURN |
1801 | return false; |
1802 | } |
1803 | |
1804 | /* |
1805 | * End initialization functions. |
1806 | */ |
1807 | /******************************************************************************/ |
1808 | /* |
1809 | * Begin allocation-path internal functions and data structures. |
1810 | */ |
1811 | |
1812 | /* |
1813 | * Settings determined by the documented behavior of the allocation functions. |
1814 | */ |
1815 | typedef struct static_opts_s static_opts_t; |
1816 | struct static_opts_s { |
1817 | /* Whether or not allocation size may overflow. */ |
1818 | bool may_overflow; |
1819 | |
1820 | /* |
1821 | * Whether or not allocations (with alignment) of size 0 should be |
1822 | * treated as size 1. |
1823 | */ |
1824 | bool bump_empty_aligned_alloc; |
1825 | /* |
1826 | * Whether to assert that allocations are not of size 0 (after any |
1827 | * bumping). |
1828 | */ |
1829 | bool assert_nonempty_alloc; |
1830 | |
1831 | /* |
1832 | * Whether or not to modify the 'result' argument to malloc in case of |
1833 | * error. |
1834 | */ |
1835 | bool null_out_result_on_error; |
1836 | /* Whether to set errno when we encounter an error condition. */ |
1837 | bool set_errno_on_error; |
1838 | |
1839 | /* |
1840 | * The minimum valid alignment for functions requesting aligned storage. |
1841 | */ |
1842 | size_t min_alignment; |
1843 | |
1844 | /* The error string to use if we oom. */ |
1845 | const char *oom_string; |
1846 | /* The error string to use if the passed-in alignment is invalid. */ |
1847 | const char *invalid_alignment_string; |
1848 | |
1849 | /* |
1850 | * False if we're configured to skip some time-consuming operations. |
1851 | * |
1852 | * This isn't really a malloc "behavior", but it acts as a useful |
1853 | * summary of several other static (or at least, static after program |
1854 | * initialization) options. |
1855 | */ |
1856 | bool slow; |
1857 | /* |
1858 | * Return size. |
1859 | */ |
1860 | bool usize; |
1861 | }; |
1862 | |
1863 | JEMALLOC_ALWAYS_INLINE void |
1864 | static_opts_init(static_opts_t *static_opts) { |
1865 | static_opts->may_overflow = false; |
1866 | static_opts->bump_empty_aligned_alloc = false; |
1867 | static_opts->assert_nonempty_alloc = false; |
1868 | static_opts->null_out_result_on_error = false; |
1869 | static_opts->set_errno_on_error = false; |
1870 | static_opts->min_alignment = 0; |
1871 | static_opts->oom_string = "" ; |
1872 | static_opts->invalid_alignment_string = "" ; |
1873 | static_opts->slow = false; |
1874 | static_opts->usize = false; |
1875 | } |
1876 | |
1877 | /* |
1878 | * These correspond to the macros in jemalloc/jemalloc_macros.h. Broadly, we |
1879 | * should have one constant here per magic value there. Note however that the |
1880 | * representations need not be related. |
1881 | */ |
1882 | #define TCACHE_IND_NONE ((unsigned)-1) |
1883 | #define TCACHE_IND_AUTOMATIC ((unsigned)-2) |
1884 | #define ARENA_IND_AUTOMATIC ((unsigned)-1) |
1885 | |
1886 | typedef struct dynamic_opts_s dynamic_opts_t; |
1887 | struct dynamic_opts_s { |
1888 | void **result; |
1889 | size_t usize; |
1890 | size_t num_items; |
1891 | size_t item_size; |
1892 | size_t alignment; |
1893 | bool zero; |
1894 | unsigned tcache_ind; |
1895 | unsigned arena_ind; |
1896 | }; |
1897 | |
1898 | JEMALLOC_ALWAYS_INLINE void |
1899 | dynamic_opts_init(dynamic_opts_t *dynamic_opts) { |
1900 | dynamic_opts->result = NULL; |
1901 | dynamic_opts->usize = 0; |
1902 | dynamic_opts->num_items = 0; |
1903 | dynamic_opts->item_size = 0; |
1904 | dynamic_opts->alignment = 0; |
1905 | dynamic_opts->zero = false; |
1906 | dynamic_opts->tcache_ind = TCACHE_IND_AUTOMATIC; |
1907 | dynamic_opts->arena_ind = ARENA_IND_AUTOMATIC; |
1908 | } |
1909 | |
1910 | /* ind is ignored if dopts->alignment > 0. */ |
1911 | JEMALLOC_ALWAYS_INLINE void * |
1912 | imalloc_no_sample(static_opts_t *sopts, dynamic_opts_t *dopts, tsd_t *tsd, |
1913 | size_t size, size_t usize, szind_t ind) { |
1914 | tcache_t *tcache; |
1915 | arena_t *arena; |
1916 | |
1917 | /* Fill in the tcache. */ |
1918 | if (dopts->tcache_ind == TCACHE_IND_AUTOMATIC) { |
1919 | if (likely(!sopts->slow)) { |
1920 | /* Getting tcache ptr unconditionally. */ |
1921 | tcache = tsd_tcachep_get(tsd); |
1922 | assert(tcache == tcache_get(tsd)); |
1923 | } else { |
1924 | tcache = tcache_get(tsd); |
1925 | } |
1926 | } else if (dopts->tcache_ind == TCACHE_IND_NONE) { |
1927 | tcache = NULL; |
1928 | } else { |
1929 | tcache = tcaches_get(tsd, dopts->tcache_ind); |
1930 | } |
1931 | |
1932 | /* Fill in the arena. */ |
1933 | if (dopts->arena_ind == ARENA_IND_AUTOMATIC) { |
1934 | /* |
1935 | * In case of automatic arena management, we defer arena |
1936 | * computation until as late as we can, hoping to fill the |
1937 | * allocation out of the tcache. |
1938 | */ |
1939 | arena = NULL; |
1940 | } else { |
1941 | arena = arena_get(tsd_tsdn(tsd), dopts->arena_ind, true); |
1942 | } |
1943 | |
1944 | if (unlikely(dopts->alignment != 0)) { |
1945 | return ipalloct(tsd_tsdn(tsd), usize, dopts->alignment, |
1946 | dopts->zero, tcache, arena); |
1947 | } |
1948 | |
1949 | return iallocztm(tsd_tsdn(tsd), size, ind, dopts->zero, tcache, false, |
1950 | arena, sopts->slow); |
1951 | } |
1952 | |
1953 | JEMALLOC_ALWAYS_INLINE void * |
1954 | imalloc_sample(static_opts_t *sopts, dynamic_opts_t *dopts, tsd_t *tsd, |
1955 | size_t usize, szind_t ind) { |
1956 | void *ret; |
1957 | |
1958 | /* |
1959 | * For small allocations, sampling bumps the usize. If so, we allocate |
1960 | * from the ind_large bucket. |
1961 | */ |
1962 | szind_t ind_large; |
1963 | size_t bumped_usize = usize; |
1964 | |
1965 | if (usize <= SC_SMALL_MAXCLASS) { |
1966 | assert(((dopts->alignment == 0) ? |
1967 | sz_s2u(SC_LARGE_MINCLASS) : |
1968 | sz_sa2u(SC_LARGE_MINCLASS, dopts->alignment)) |
1969 | == SC_LARGE_MINCLASS); |
1970 | ind_large = sz_size2index(SC_LARGE_MINCLASS); |
1971 | bumped_usize = sz_s2u(SC_LARGE_MINCLASS); |
1972 | ret = imalloc_no_sample(sopts, dopts, tsd, bumped_usize, |
1973 | bumped_usize, ind_large); |
1974 | if (unlikely(ret == NULL)) { |
1975 | return NULL; |
1976 | } |
1977 | arena_prof_promote(tsd_tsdn(tsd), ret, usize); |
1978 | } else { |
1979 | ret = imalloc_no_sample(sopts, dopts, tsd, usize, usize, ind); |
1980 | } |
1981 | |
1982 | return ret; |
1983 | } |
1984 | |
1985 | /* |
1986 | * Returns true if the allocation will overflow, and false otherwise. Sets |
1987 | * *size to the product either way. |
1988 | */ |
1989 | JEMALLOC_ALWAYS_INLINE bool |
1990 | compute_size_with_overflow(bool may_overflow, dynamic_opts_t *dopts, |
1991 | size_t *size) { |
1992 | /* |
1993 | * This function is just num_items * item_size, except that we may have |
1994 | * to check for overflow. |
1995 | */ |
1996 | |
1997 | if (!may_overflow) { |
1998 | assert(dopts->num_items == 1); |
1999 | *size = dopts->item_size; |
2000 | return false; |
2001 | } |
2002 | |
2003 | /* A size_t with its high-half bits all set to 1. */ |
2004 | static const size_t high_bits = SIZE_T_MAX << (sizeof(size_t) * 8 / 2); |
2005 | |
2006 | *size = dopts->item_size * dopts->num_items; |
2007 | |
2008 | if (unlikely(*size == 0)) { |
2009 | return (dopts->num_items != 0 && dopts->item_size != 0); |
2010 | } |
2011 | |
2012 | /* |
2013 | * We got a non-zero size, but we don't know if we overflowed to get |
2014 | * there. To avoid having to do a divide, we'll be clever and note that |
2015 | * if both A and B can be represented in N/2 bits, then their product |
2016 | * can be represented in N bits (without the possibility of overflow). |
2017 | */ |
2018 | if (likely((high_bits & (dopts->num_items | dopts->item_size)) == 0)) { |
2019 | return false; |
2020 | } |
2021 | if (likely(*size / dopts->item_size == dopts->num_items)) { |
2022 | return false; |
2023 | } |
2024 | return true; |
2025 | } |
2026 | |
2027 | JEMALLOC_ALWAYS_INLINE int |
2028 | imalloc_body(static_opts_t *sopts, dynamic_opts_t *dopts, tsd_t *tsd) { |
2029 | /* Where the actual allocated memory will live. */ |
2030 | void *allocation = NULL; |
2031 | /* Filled in by compute_size_with_overflow below. */ |
2032 | size_t size = 0; |
2033 | /* |
2034 | * For unaligned allocations, we need only ind. For aligned |
2035 | * allocations, or in case of stats or profiling we need usize. |
2036 | * |
2037 | * These are actually dead stores, in that their values are reset before |
2038 | * any branch on their value is taken. Sometimes though, it's |
2039 | * convenient to pass them as arguments before this point. To avoid |
2040 | * undefined behavior then, we initialize them with dummy stores. |
2041 | */ |
2042 | szind_t ind = 0; |
2043 | size_t usize = 0; |
2044 | |
2045 | /* Reentrancy is only checked on slow path. */ |
2046 | int8_t reentrancy_level; |
2047 | |
2048 | /* Compute the amount of memory the user wants. */ |
2049 | if (unlikely(compute_size_with_overflow(sopts->may_overflow, dopts, |
2050 | &size))) { |
2051 | goto label_oom; |
2052 | } |
2053 | |
2054 | if (unlikely(dopts->alignment < sopts->min_alignment |
2055 | || (dopts->alignment & (dopts->alignment - 1)) != 0)) { |
2056 | goto label_invalid_alignment; |
2057 | } |
2058 | |
2059 | /* This is the beginning of the "core" algorithm. */ |
2060 | |
2061 | if (dopts->alignment == 0) { |
2062 | ind = sz_size2index(size); |
2063 | if (unlikely(ind >= SC_NSIZES)) { |
2064 | goto label_oom; |
2065 | } |
2066 | if (config_stats || (config_prof && opt_prof) || sopts->usize) { |
2067 | usize = sz_index2size(ind); |
2068 | dopts->usize = usize; |
2069 | assert(usize > 0 && usize |
2070 | <= SC_LARGE_MAXCLASS); |
2071 | } |
2072 | } else { |
2073 | if (sopts->bump_empty_aligned_alloc) { |
2074 | if (unlikely(size == 0)) { |
2075 | size = 1; |
2076 | } |
2077 | } |
2078 | usize = sz_sa2u(size, dopts->alignment); |
2079 | dopts->usize = usize; |
2080 | if (unlikely(usize == 0 |
2081 | || usize > SC_LARGE_MAXCLASS)) { |
2082 | goto label_oom; |
2083 | } |
2084 | } |
2085 | /* Validate the user input. */ |
2086 | if (sopts->assert_nonempty_alloc) { |
2087 | assert (size != 0); |
2088 | } |
2089 | |
2090 | check_entry_exit_locking(tsd_tsdn(tsd)); |
2091 | |
2092 | /* |
2093 | * If we need to handle reentrancy, we can do it out of a |
2094 | * known-initialized arena (i.e. arena 0). |
2095 | */ |
2096 | reentrancy_level = tsd_reentrancy_level_get(tsd); |
2097 | if (sopts->slow && unlikely(reentrancy_level > 0)) { |
2098 | /* |
2099 | * We should never specify particular arenas or tcaches from |
2100 | * within our internal allocations. |
2101 | */ |
2102 | assert(dopts->tcache_ind == TCACHE_IND_AUTOMATIC || |
2103 | dopts->tcache_ind == TCACHE_IND_NONE); |
2104 | assert(dopts->arena_ind == ARENA_IND_AUTOMATIC); |
2105 | dopts->tcache_ind = TCACHE_IND_NONE; |
2106 | /* We know that arena 0 has already been initialized. */ |
2107 | dopts->arena_ind = 0; |
2108 | } |
2109 | |
2110 | /* If profiling is on, get our profiling context. */ |
2111 | if (config_prof && opt_prof) { |
2112 | /* |
2113 | * Note that if we're going down this path, usize must have been |
2114 | * initialized in the previous if statement. |
2115 | */ |
2116 | prof_tctx_t *tctx = prof_alloc_prep( |
2117 | tsd, usize, prof_active_get_unlocked(), true); |
2118 | |
2119 | alloc_ctx_t alloc_ctx; |
2120 | if (likely((uintptr_t)tctx == (uintptr_t)1U)) { |
2121 | alloc_ctx.slab = (usize |
2122 | <= SC_SMALL_MAXCLASS); |
2123 | allocation = imalloc_no_sample( |
2124 | sopts, dopts, tsd, usize, usize, ind); |
2125 | } else if ((uintptr_t)tctx > (uintptr_t)1U) { |
2126 | /* |
2127 | * Note that ind might still be 0 here. This is fine; |
2128 | * imalloc_sample ignores ind if dopts->alignment > 0. |
2129 | */ |
2130 | allocation = imalloc_sample( |
2131 | sopts, dopts, tsd, usize, ind); |
2132 | alloc_ctx.slab = false; |
2133 | } else { |
2134 | allocation = NULL; |
2135 | } |
2136 | |
2137 | if (unlikely(allocation == NULL)) { |
2138 | prof_alloc_rollback(tsd, tctx, true); |
2139 | goto label_oom; |
2140 | } |
2141 | prof_malloc(tsd_tsdn(tsd), allocation, usize, &alloc_ctx, tctx); |
2142 | } else { |
2143 | /* |
2144 | * If dopts->alignment > 0, then ind is still 0, but usize was |
2145 | * computed in the previous if statement. Down the positive |
2146 | * alignment path, imalloc_no_sample ignores ind and size |
2147 | * (relying only on usize). |
2148 | */ |
2149 | allocation = imalloc_no_sample(sopts, dopts, tsd, size, usize, |
2150 | ind); |
2151 | if (unlikely(allocation == NULL)) { |
2152 | goto label_oom; |
2153 | } |
2154 | } |
2155 | |
2156 | /* |
2157 | * Allocation has been done at this point. We still have some |
2158 | * post-allocation work to do though. |
2159 | */ |
2160 | assert(dopts->alignment == 0 |
2161 | || ((uintptr_t)allocation & (dopts->alignment - 1)) == ZU(0)); |
2162 | |
2163 | if (config_stats) { |
2164 | assert(usize == isalloc(tsd_tsdn(tsd), allocation)); |
2165 | *tsd_thread_allocatedp_get(tsd) += usize; |
2166 | } |
2167 | |
2168 | if (sopts->slow) { |
2169 | UTRACE(0, size, allocation); |
2170 | } |
2171 | |
2172 | /* Success! */ |
2173 | check_entry_exit_locking(tsd_tsdn(tsd)); |
2174 | *dopts->result = allocation; |
2175 | return 0; |
2176 | |
2177 | label_oom: |
2178 | if (unlikely(sopts->slow) && config_xmalloc && unlikely(opt_xmalloc)) { |
2179 | malloc_write(sopts->oom_string); |
2180 | abort(); |
2181 | } |
2182 | |
2183 | if (sopts->slow) { |
2184 | UTRACE(NULL, size, NULL); |
2185 | } |
2186 | |
2187 | check_entry_exit_locking(tsd_tsdn(tsd)); |
2188 | |
2189 | if (sopts->set_errno_on_error) { |
2190 | set_errno(ENOMEM); |
2191 | } |
2192 | |
2193 | if (sopts->null_out_result_on_error) { |
2194 | *dopts->result = NULL; |
2195 | } |
2196 | |
2197 | return ENOMEM; |
2198 | |
2199 | /* |
2200 | * This label is only jumped to by one goto; we move it out of line |
2201 | * anyways to avoid obscuring the non-error paths, and for symmetry with |
2202 | * the oom case. |
2203 | */ |
2204 | label_invalid_alignment: |
2205 | if (config_xmalloc && unlikely(opt_xmalloc)) { |
2206 | malloc_write(sopts->invalid_alignment_string); |
2207 | abort(); |
2208 | } |
2209 | |
2210 | if (sopts->set_errno_on_error) { |
2211 | set_errno(EINVAL); |
2212 | } |
2213 | |
2214 | if (sopts->slow) { |
2215 | UTRACE(NULL, size, NULL); |
2216 | } |
2217 | |
2218 | check_entry_exit_locking(tsd_tsdn(tsd)); |
2219 | |
2220 | if (sopts->null_out_result_on_error) { |
2221 | *dopts->result = NULL; |
2222 | } |
2223 | |
2224 | return EINVAL; |
2225 | } |
2226 | |
2227 | JEMALLOC_ALWAYS_INLINE bool |
2228 | imalloc_init_check(static_opts_t *sopts, dynamic_opts_t *dopts) { |
2229 | if (unlikely(!malloc_initialized()) && unlikely(malloc_init())) { |
2230 | if (config_xmalloc && unlikely(opt_xmalloc)) { |
2231 | malloc_write(sopts->oom_string); |
2232 | abort(); |
2233 | } |
2234 | UTRACE(NULL, dopts->num_items * dopts->item_size, NULL); |
2235 | set_errno(ENOMEM); |
2236 | *dopts->result = NULL; |
2237 | |
2238 | return false; |
2239 | } |
2240 | |
2241 | return true; |
2242 | } |
2243 | |
2244 | /* Returns the errno-style error code of the allocation. */ |
2245 | JEMALLOC_ALWAYS_INLINE int |
2246 | imalloc(static_opts_t *sopts, dynamic_opts_t *dopts) { |
2247 | if (tsd_get_allocates() && !imalloc_init_check(sopts, dopts)) { |
2248 | return ENOMEM; |
2249 | } |
2250 | |
2251 | /* We always need the tsd. Let's grab it right away. */ |
2252 | tsd_t *tsd = tsd_fetch(); |
2253 | assert(tsd); |
2254 | if (likely(tsd_fast(tsd))) { |
2255 | /* Fast and common path. */ |
2256 | tsd_assert_fast(tsd); |
2257 | sopts->slow = false; |
2258 | return imalloc_body(sopts, dopts, tsd); |
2259 | } else { |
2260 | if (!tsd_get_allocates() && !imalloc_init_check(sopts, dopts)) { |
2261 | return ENOMEM; |
2262 | } |
2263 | |
2264 | sopts->slow = true; |
2265 | return imalloc_body(sopts, dopts, tsd); |
2266 | } |
2267 | } |
2268 | |
2269 | JEMALLOC_NOINLINE |
2270 | void * |
2271 | malloc_default(size_t size) { |
2272 | void *ret; |
2273 | static_opts_t sopts; |
2274 | dynamic_opts_t dopts; |
2275 | |
2276 | LOG("core.malloc.entry" , "size: %zu" , size); |
2277 | |
2278 | static_opts_init(&sopts); |
2279 | dynamic_opts_init(&dopts); |
2280 | |
2281 | sopts.null_out_result_on_error = true; |
2282 | sopts.set_errno_on_error = true; |
2283 | sopts.oom_string = "<jemalloc>: Error in malloc(): out of memory\n" ; |
2284 | |
2285 | dopts.result = &ret; |
2286 | dopts.num_items = 1; |
2287 | dopts.item_size = size; |
2288 | |
2289 | imalloc(&sopts, &dopts); |
2290 | /* |
2291 | * Note that this branch gets optimized away -- it immediately follows |
2292 | * the check on tsd_fast that sets sopts.slow. |
2293 | */ |
2294 | if (sopts.slow) { |
2295 | uintptr_t args[3] = {size}; |
2296 | hook_invoke_alloc(hook_alloc_malloc, ret, (uintptr_t)ret, args); |
2297 | } |
2298 | |
2299 | LOG("core.malloc.exit" , "result: %p" , ret); |
2300 | |
2301 | return ret; |
2302 | } |
2303 | |
2304 | /******************************************************************************/ |
2305 | /* |
2306 | * Begin malloc(3)-compatible functions. |
2307 | */ |
2308 | |
2309 | /* |
2310 | * malloc() fastpath. |
2311 | * |
2312 | * Fastpath assumes size <= SC_LOOKUP_MAXCLASS, and that we hit |
2313 | * tcache. If either of these is false, we tail-call to the slowpath, |
2314 | * malloc_default(). Tail-calling is used to avoid any caller-saved |
2315 | * registers. |
2316 | * |
2317 | * fastpath supports ticker and profiling, both of which will also |
2318 | * tail-call to the slowpath if they fire. |
2319 | */ |
2320 | JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN |
2321 | void JEMALLOC_NOTHROW * |
2322 | JEMALLOC_ATTR(malloc) JEMALLOC_ALLOC_SIZE(1) |
2323 | je_malloc(size_t size) { |
2324 | LOG("core.malloc.entry" , "size: %zu" , size); |
2325 | |
2326 | if (tsd_get_allocates() && unlikely(!malloc_initialized())) { |
2327 | return malloc_default(size); |
2328 | } |
2329 | |
2330 | tsd_t *tsd = tsd_get(false); |
2331 | if (unlikely(!tsd || !tsd_fast(tsd) || (size > SC_LOOKUP_MAXCLASS))) { |
2332 | return malloc_default(size); |
2333 | } |
2334 | |
2335 | tcache_t *tcache = tsd_tcachep_get(tsd); |
2336 | |
2337 | if (unlikely(ticker_trytick(&tcache->gc_ticker))) { |
2338 | return malloc_default(size); |
2339 | } |
2340 | |
2341 | szind_t ind = sz_size2index_lookup(size); |
2342 | size_t usize; |
2343 | if (config_stats || config_prof) { |
2344 | usize = sz_index2size(ind); |
2345 | } |
2346 | /* Fast path relies on size being a bin. I.e. SC_LOOKUP_MAXCLASS < SC_SMALL_MAXCLASS */ |
2347 | assert(ind < SC_NBINS); |
2348 | assert(size <= SC_SMALL_MAXCLASS); |
2349 | |
2350 | if (config_prof) { |
2351 | int64_t bytes_until_sample = tsd_bytes_until_sample_get(tsd); |
2352 | bytes_until_sample -= usize; |
2353 | tsd_bytes_until_sample_set(tsd, bytes_until_sample); |
2354 | |
2355 | if (unlikely(bytes_until_sample < 0)) { |
2356 | /* |
2357 | * Avoid a prof_active check on the fastpath. |
2358 | * If prof_active is false, set bytes_until_sample to |
2359 | * a large value. If prof_active is set to true, |
2360 | * bytes_until_sample will be reset. |
2361 | */ |
2362 | if (!prof_active) { |
2363 | tsd_bytes_until_sample_set(tsd, SSIZE_MAX); |
2364 | } |
2365 | return malloc_default(size); |
2366 | } |
2367 | } |
2368 | |
2369 | cache_bin_t *bin = tcache_small_bin_get(tcache, ind); |
2370 | bool tcache_success; |
2371 | void* ret = cache_bin_alloc_easy(bin, &tcache_success); |
2372 | |
2373 | if (tcache_success) { |
2374 | if (config_stats) { |
2375 | *tsd_thread_allocatedp_get(tsd) += usize; |
2376 | bin->tstats.nrequests++; |
2377 | } |
2378 | if (config_prof) { |
2379 | tcache->prof_accumbytes += usize; |
2380 | } |
2381 | |
2382 | LOG("core.malloc.exit" , "result: %p" , ret); |
2383 | |
2384 | /* Fastpath success */ |
2385 | return ret; |
2386 | } |
2387 | |
2388 | return malloc_default(size); |
2389 | } |
2390 | |
2391 | JEMALLOC_EXPORT int JEMALLOC_NOTHROW |
2392 | JEMALLOC_ATTR(nonnull(1)) |
2393 | je_posix_memalign(void **memptr, size_t alignment, size_t size) { |
2394 | int ret; |
2395 | static_opts_t sopts; |
2396 | dynamic_opts_t dopts; |
2397 | |
2398 | LOG("core.posix_memalign.entry" , "mem ptr: %p, alignment: %zu, " |
2399 | "size: %zu" , memptr, alignment, size); |
2400 | |
2401 | static_opts_init(&sopts); |
2402 | dynamic_opts_init(&dopts); |
2403 | |
2404 | sopts.bump_empty_aligned_alloc = true; |
2405 | sopts.min_alignment = sizeof(void *); |
2406 | sopts.oom_string = |
2407 | "<jemalloc>: Error allocating aligned memory: out of memory\n" ; |
2408 | sopts.invalid_alignment_string = |
2409 | "<jemalloc>: Error allocating aligned memory: invalid alignment\n" ; |
2410 | |
2411 | dopts.result = memptr; |
2412 | dopts.num_items = 1; |
2413 | dopts.item_size = size; |
2414 | dopts.alignment = alignment; |
2415 | |
2416 | ret = imalloc(&sopts, &dopts); |
2417 | if (sopts.slow) { |
2418 | uintptr_t args[3] = {(uintptr_t)memptr, (uintptr_t)alignment, |
2419 | (uintptr_t)size}; |
2420 | hook_invoke_alloc(hook_alloc_posix_memalign, *memptr, |
2421 | (uintptr_t)ret, args); |
2422 | } |
2423 | |
2424 | LOG("core.posix_memalign.exit" , "result: %d, alloc ptr: %p" , ret, |
2425 | *memptr); |
2426 | |
2427 | return ret; |
2428 | } |
2429 | |
2430 | JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN |
2431 | void JEMALLOC_NOTHROW * |
2432 | JEMALLOC_ATTR(malloc) JEMALLOC_ALLOC_SIZE(2) |
2433 | je_aligned_alloc(size_t alignment, size_t size) { |
2434 | void *ret; |
2435 | |
2436 | static_opts_t sopts; |
2437 | dynamic_opts_t dopts; |
2438 | |
2439 | LOG("core.aligned_alloc.entry" , "alignment: %zu, size: %zu\n" , |
2440 | alignment, size); |
2441 | |
2442 | static_opts_init(&sopts); |
2443 | dynamic_opts_init(&dopts); |
2444 | |
2445 | sopts.bump_empty_aligned_alloc = true; |
2446 | sopts.null_out_result_on_error = true; |
2447 | sopts.set_errno_on_error = true; |
2448 | sopts.min_alignment = 1; |
2449 | sopts.oom_string = |
2450 | "<jemalloc>: Error allocating aligned memory: out of memory\n" ; |
2451 | sopts.invalid_alignment_string = |
2452 | "<jemalloc>: Error allocating aligned memory: invalid alignment\n" ; |
2453 | |
2454 | dopts.result = &ret; |
2455 | dopts.num_items = 1; |
2456 | dopts.item_size = size; |
2457 | dopts.alignment = alignment; |
2458 | |
2459 | imalloc(&sopts, &dopts); |
2460 | if (sopts.slow) { |
2461 | uintptr_t args[3] = {(uintptr_t)alignment, (uintptr_t)size}; |
2462 | hook_invoke_alloc(hook_alloc_aligned_alloc, ret, |
2463 | (uintptr_t)ret, args); |
2464 | } |
2465 | |
2466 | LOG("core.aligned_alloc.exit" , "result: %p" , ret); |
2467 | |
2468 | return ret; |
2469 | } |
2470 | |
2471 | JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN |
2472 | void JEMALLOC_NOTHROW * |
2473 | JEMALLOC_ATTR(malloc) JEMALLOC_ALLOC_SIZE2(1, 2) |
2474 | je_calloc(size_t num, size_t size) { |
2475 | void *ret; |
2476 | static_opts_t sopts; |
2477 | dynamic_opts_t dopts; |
2478 | |
2479 | LOG("core.calloc.entry" , "num: %zu, size: %zu\n" , num, size); |
2480 | |
2481 | static_opts_init(&sopts); |
2482 | dynamic_opts_init(&dopts); |
2483 | |
2484 | sopts.may_overflow = true; |
2485 | sopts.null_out_result_on_error = true; |
2486 | sopts.set_errno_on_error = true; |
2487 | sopts.oom_string = "<jemalloc>: Error in calloc(): out of memory\n" ; |
2488 | |
2489 | dopts.result = &ret; |
2490 | dopts.num_items = num; |
2491 | dopts.item_size = size; |
2492 | dopts.zero = true; |
2493 | |
2494 | imalloc(&sopts, &dopts); |
2495 | if (sopts.slow) { |
2496 | uintptr_t args[3] = {(uintptr_t)num, (uintptr_t)size}; |
2497 | hook_invoke_alloc(hook_alloc_calloc, ret, (uintptr_t)ret, args); |
2498 | } |
2499 | |
2500 | LOG("core.calloc.exit" , "result: %p" , ret); |
2501 | |
2502 | return ret; |
2503 | } |
2504 | |
2505 | static void * |
2506 | irealloc_prof_sample(tsd_t *tsd, void *old_ptr, size_t old_usize, size_t usize, |
2507 | prof_tctx_t *tctx, hook_ralloc_args_t *hook_args) { |
2508 | void *p; |
2509 | |
2510 | if (tctx == NULL) { |
2511 | return NULL; |
2512 | } |
2513 | if (usize <= SC_SMALL_MAXCLASS) { |
2514 | p = iralloc(tsd, old_ptr, old_usize, |
2515 | SC_LARGE_MINCLASS, 0, false, hook_args); |
2516 | if (p == NULL) { |
2517 | return NULL; |
2518 | } |
2519 | arena_prof_promote(tsd_tsdn(tsd), p, usize); |
2520 | } else { |
2521 | p = iralloc(tsd, old_ptr, old_usize, usize, 0, false, |
2522 | hook_args); |
2523 | } |
2524 | |
2525 | return p; |
2526 | } |
2527 | |
2528 | JEMALLOC_ALWAYS_INLINE void * |
2529 | irealloc_prof(tsd_t *tsd, void *old_ptr, size_t old_usize, size_t usize, |
2530 | alloc_ctx_t *alloc_ctx, hook_ralloc_args_t *hook_args) { |
2531 | void *p; |
2532 | bool prof_active; |
2533 | prof_tctx_t *old_tctx, *tctx; |
2534 | |
2535 | prof_active = prof_active_get_unlocked(); |
2536 | old_tctx = prof_tctx_get(tsd_tsdn(tsd), old_ptr, alloc_ctx); |
2537 | tctx = prof_alloc_prep(tsd, usize, prof_active, true); |
2538 | if (unlikely((uintptr_t)tctx != (uintptr_t)1U)) { |
2539 | p = irealloc_prof_sample(tsd, old_ptr, old_usize, usize, tctx, |
2540 | hook_args); |
2541 | } else { |
2542 | p = iralloc(tsd, old_ptr, old_usize, usize, 0, false, |
2543 | hook_args); |
2544 | } |
2545 | if (unlikely(p == NULL)) { |
2546 | prof_alloc_rollback(tsd, tctx, true); |
2547 | return NULL; |
2548 | } |
2549 | prof_realloc(tsd, p, usize, tctx, prof_active, true, old_ptr, old_usize, |
2550 | old_tctx); |
2551 | |
2552 | return p; |
2553 | } |
2554 | |
2555 | JEMALLOC_ALWAYS_INLINE void |
2556 | ifree(tsd_t *tsd, void *ptr, tcache_t *tcache, bool slow_path) { |
2557 | if (!slow_path) { |
2558 | tsd_assert_fast(tsd); |
2559 | } |
2560 | check_entry_exit_locking(tsd_tsdn(tsd)); |
2561 | if (tsd_reentrancy_level_get(tsd) != 0) { |
2562 | assert(slow_path); |
2563 | } |
2564 | |
2565 | assert(ptr != NULL); |
2566 | assert(malloc_initialized() || IS_INITIALIZER); |
2567 | |
2568 | alloc_ctx_t alloc_ctx; |
2569 | rtree_ctx_t *rtree_ctx = tsd_rtree_ctx(tsd); |
2570 | rtree_szind_slab_read(tsd_tsdn(tsd), &extents_rtree, rtree_ctx, |
2571 | (uintptr_t)ptr, true, &alloc_ctx.szind, &alloc_ctx.slab); |
2572 | assert(alloc_ctx.szind != SC_NSIZES); |
2573 | |
2574 | size_t usize; |
2575 | if (config_prof && opt_prof) { |
2576 | usize = sz_index2size(alloc_ctx.szind); |
2577 | prof_free(tsd, ptr, usize, &alloc_ctx); |
2578 | } else if (config_stats) { |
2579 | usize = sz_index2size(alloc_ctx.szind); |
2580 | } |
2581 | if (config_stats) { |
2582 | *tsd_thread_deallocatedp_get(tsd) += usize; |
2583 | } |
2584 | |
2585 | if (likely(!slow_path)) { |
2586 | idalloctm(tsd_tsdn(tsd), ptr, tcache, &alloc_ctx, false, |
2587 | false); |
2588 | } else { |
2589 | idalloctm(tsd_tsdn(tsd), ptr, tcache, &alloc_ctx, false, |
2590 | true); |
2591 | } |
2592 | } |
2593 | |
2594 | JEMALLOC_ALWAYS_INLINE void |
2595 | isfree(tsd_t *tsd, void *ptr, size_t usize, tcache_t *tcache, bool slow_path) { |
2596 | if (!slow_path) { |
2597 | tsd_assert_fast(tsd); |
2598 | } |
2599 | check_entry_exit_locking(tsd_tsdn(tsd)); |
2600 | if (tsd_reentrancy_level_get(tsd) != 0) { |
2601 | assert(slow_path); |
2602 | } |
2603 | |
2604 | assert(ptr != NULL); |
2605 | assert(malloc_initialized() || IS_INITIALIZER); |
2606 | |
2607 | alloc_ctx_t alloc_ctx, *ctx; |
2608 | if (!config_cache_oblivious && ((uintptr_t)ptr & PAGE_MASK) != 0) { |
2609 | /* |
2610 | * When cache_oblivious is disabled and ptr is not page aligned, |
2611 | * the allocation was not sampled -- usize can be used to |
2612 | * determine szind directly. |
2613 | */ |
2614 | alloc_ctx.szind = sz_size2index(usize); |
2615 | alloc_ctx.slab = true; |
2616 | ctx = &alloc_ctx; |
2617 | if (config_debug) { |
2618 | alloc_ctx_t dbg_ctx; |
2619 | rtree_ctx_t *rtree_ctx = tsd_rtree_ctx(tsd); |
2620 | rtree_szind_slab_read(tsd_tsdn(tsd), &extents_rtree, |
2621 | rtree_ctx, (uintptr_t)ptr, true, &dbg_ctx.szind, |
2622 | &dbg_ctx.slab); |
2623 | assert(dbg_ctx.szind == alloc_ctx.szind); |
2624 | assert(dbg_ctx.slab == alloc_ctx.slab); |
2625 | } |
2626 | } else if (config_prof && opt_prof) { |
2627 | rtree_ctx_t *rtree_ctx = tsd_rtree_ctx(tsd); |
2628 | rtree_szind_slab_read(tsd_tsdn(tsd), &extents_rtree, rtree_ctx, |
2629 | (uintptr_t)ptr, true, &alloc_ctx.szind, &alloc_ctx.slab); |
2630 | assert(alloc_ctx.szind == sz_size2index(usize)); |
2631 | ctx = &alloc_ctx; |
2632 | } else { |
2633 | ctx = NULL; |
2634 | } |
2635 | |
2636 | if (config_prof && opt_prof) { |
2637 | prof_free(tsd, ptr, usize, ctx); |
2638 | } |
2639 | if (config_stats) { |
2640 | *tsd_thread_deallocatedp_get(tsd) += usize; |
2641 | } |
2642 | |
2643 | if (likely(!slow_path)) { |
2644 | isdalloct(tsd_tsdn(tsd), ptr, usize, tcache, ctx, false); |
2645 | } else { |
2646 | isdalloct(tsd_tsdn(tsd), ptr, usize, tcache, ctx, true); |
2647 | } |
2648 | } |
2649 | |
2650 | JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN |
2651 | void JEMALLOC_NOTHROW * |
2652 | JEMALLOC_ALLOC_SIZE(2) |
2653 | je_realloc(void *ptr, size_t arg_size) { |
2654 | void *ret; |
2655 | tsdn_t *tsdn JEMALLOC_CC_SILENCE_INIT(NULL); |
2656 | size_t usize JEMALLOC_CC_SILENCE_INIT(0); |
2657 | size_t old_usize = 0; |
2658 | size_t size = arg_size; |
2659 | |
2660 | LOG("core.realloc.entry" , "ptr: %p, size: %zu\n" , ptr, size); |
2661 | |
2662 | if (unlikely(size == 0)) { |
2663 | if (ptr != NULL) { |
2664 | /* realloc(ptr, 0) is equivalent to free(ptr). */ |
2665 | UTRACE(ptr, 0, 0); |
2666 | tcache_t *tcache; |
2667 | tsd_t *tsd = tsd_fetch(); |
2668 | if (tsd_reentrancy_level_get(tsd) == 0) { |
2669 | tcache = tcache_get(tsd); |
2670 | } else { |
2671 | tcache = NULL; |
2672 | } |
2673 | |
2674 | uintptr_t args[3] = {(uintptr_t)ptr, size}; |
2675 | hook_invoke_dalloc(hook_dalloc_realloc, ptr, args); |
2676 | |
2677 | ifree(tsd, ptr, tcache, true); |
2678 | |
2679 | LOG("core.realloc.exit" , "result: %p" , NULL); |
2680 | return NULL; |
2681 | } |
2682 | size = 1; |
2683 | } |
2684 | |
2685 | if (likely(ptr != NULL)) { |
2686 | assert(malloc_initialized() || IS_INITIALIZER); |
2687 | tsd_t *tsd = tsd_fetch(); |
2688 | |
2689 | check_entry_exit_locking(tsd_tsdn(tsd)); |
2690 | |
2691 | |
2692 | hook_ralloc_args_t hook_args = {true, {(uintptr_t)ptr, |
2693 | (uintptr_t)arg_size, 0, 0}}; |
2694 | |
2695 | alloc_ctx_t alloc_ctx; |
2696 | rtree_ctx_t *rtree_ctx = tsd_rtree_ctx(tsd); |
2697 | rtree_szind_slab_read(tsd_tsdn(tsd), &extents_rtree, rtree_ctx, |
2698 | (uintptr_t)ptr, true, &alloc_ctx.szind, &alloc_ctx.slab); |
2699 | assert(alloc_ctx.szind != SC_NSIZES); |
2700 | old_usize = sz_index2size(alloc_ctx.szind); |
2701 | assert(old_usize == isalloc(tsd_tsdn(tsd), ptr)); |
2702 | if (config_prof && opt_prof) { |
2703 | usize = sz_s2u(size); |
2704 | if (unlikely(usize == 0 |
2705 | || usize > SC_LARGE_MAXCLASS)) { |
2706 | ret = NULL; |
2707 | } else { |
2708 | ret = irealloc_prof(tsd, ptr, old_usize, usize, |
2709 | &alloc_ctx, &hook_args); |
2710 | } |
2711 | } else { |
2712 | if (config_stats) { |
2713 | usize = sz_s2u(size); |
2714 | } |
2715 | ret = iralloc(tsd, ptr, old_usize, size, 0, false, |
2716 | &hook_args); |
2717 | } |
2718 | tsdn = tsd_tsdn(tsd); |
2719 | } else { |
2720 | /* realloc(NULL, size) is equivalent to malloc(size). */ |
2721 | static_opts_t sopts; |
2722 | dynamic_opts_t dopts; |
2723 | |
2724 | static_opts_init(&sopts); |
2725 | dynamic_opts_init(&dopts); |
2726 | |
2727 | sopts.null_out_result_on_error = true; |
2728 | sopts.set_errno_on_error = true; |
2729 | sopts.oom_string = |
2730 | "<jemalloc>: Error in realloc(): out of memory\n" ; |
2731 | |
2732 | dopts.result = &ret; |
2733 | dopts.num_items = 1; |
2734 | dopts.item_size = size; |
2735 | |
2736 | imalloc(&sopts, &dopts); |
2737 | if (sopts.slow) { |
2738 | uintptr_t args[3] = {(uintptr_t)ptr, arg_size}; |
2739 | hook_invoke_alloc(hook_alloc_realloc, ret, |
2740 | (uintptr_t)ret, args); |
2741 | } |
2742 | |
2743 | return ret; |
2744 | } |
2745 | |
2746 | if (unlikely(ret == NULL)) { |
2747 | if (config_xmalloc && unlikely(opt_xmalloc)) { |
2748 | malloc_write("<jemalloc>: Error in realloc(): " |
2749 | "out of memory\n" ); |
2750 | abort(); |
2751 | } |
2752 | set_errno(ENOMEM); |
2753 | } |
2754 | if (config_stats && likely(ret != NULL)) { |
2755 | tsd_t *tsd; |
2756 | |
2757 | assert(usize == isalloc(tsdn, ret)); |
2758 | tsd = tsdn_tsd(tsdn); |
2759 | *tsd_thread_allocatedp_get(tsd) += usize; |
2760 | *tsd_thread_deallocatedp_get(tsd) += old_usize; |
2761 | } |
2762 | UTRACE(ptr, size, ret); |
2763 | check_entry_exit_locking(tsdn); |
2764 | |
2765 | LOG("core.realloc.exit" , "result: %p" , ret); |
2766 | return ret; |
2767 | } |
2768 | |
2769 | JEMALLOC_NOINLINE |
2770 | void |
2771 | free_default(void *ptr) { |
2772 | UTRACE(ptr, 0, 0); |
2773 | if (likely(ptr != NULL)) { |
2774 | /* |
2775 | * We avoid setting up tsd fully (e.g. tcache, arena binding) |
2776 | * based on only free() calls -- other activities trigger the |
2777 | * minimal to full transition. This is because free() may |
2778 | * happen during thread shutdown after tls deallocation: if a |
2779 | * thread never had any malloc activities until then, a |
2780 | * fully-setup tsd won't be destructed properly. |
2781 | */ |
2782 | tsd_t *tsd = tsd_fetch_min(); |
2783 | check_entry_exit_locking(tsd_tsdn(tsd)); |
2784 | |
2785 | tcache_t *tcache; |
2786 | if (likely(tsd_fast(tsd))) { |
2787 | tsd_assert_fast(tsd); |
2788 | /* Unconditionally get tcache ptr on fast path. */ |
2789 | tcache = tsd_tcachep_get(tsd); |
2790 | ifree(tsd, ptr, tcache, false); |
2791 | } else { |
2792 | if (likely(tsd_reentrancy_level_get(tsd) == 0)) { |
2793 | tcache = tcache_get(tsd); |
2794 | } else { |
2795 | tcache = NULL; |
2796 | } |
2797 | uintptr_t args_raw[3] = {(uintptr_t)ptr}; |
2798 | hook_invoke_dalloc(hook_dalloc_free, ptr, args_raw); |
2799 | ifree(tsd, ptr, tcache, true); |
2800 | } |
2801 | check_entry_exit_locking(tsd_tsdn(tsd)); |
2802 | } |
2803 | } |
2804 | |
2805 | JEMALLOC_ALWAYS_INLINE |
2806 | bool free_fastpath(void *ptr, size_t size, bool size_hint) { |
2807 | tsd_t *tsd = tsd_get(false); |
2808 | if (unlikely(!tsd || !tsd_fast(tsd))) { |
2809 | return false; |
2810 | } |
2811 | |
2812 | tcache_t *tcache = tsd_tcachep_get(tsd); |
2813 | |
2814 | alloc_ctx_t alloc_ctx; |
2815 | /* |
2816 | * If !config_cache_oblivious, we can check PAGE alignment to |
2817 | * detect sampled objects. Otherwise addresses are |
2818 | * randomized, and we have to look it up in the rtree anyway. |
2819 | * See also isfree(). |
2820 | */ |
2821 | if (!size_hint || config_cache_oblivious) { |
2822 | rtree_ctx_t *rtree_ctx = tsd_rtree_ctx(tsd); |
2823 | bool res = rtree_szind_slab_read_fast(tsd_tsdn(tsd), &extents_rtree, |
2824 | rtree_ctx, (uintptr_t)ptr, |
2825 | &alloc_ctx.szind, &alloc_ctx.slab); |
2826 | |
2827 | /* Note: profiled objects will have alloc_ctx.slab set */ |
2828 | if (!res || !alloc_ctx.slab) { |
2829 | return false; |
2830 | } |
2831 | assert(alloc_ctx.szind != SC_NSIZES); |
2832 | } else { |
2833 | /* |
2834 | * Check for both sizes that are too large, and for sampled objects. |
2835 | * Sampled objects are always page-aligned. The sampled object check |
2836 | * will also check for null ptr. |
2837 | */ |
2838 | if (size > SC_LOOKUP_MAXCLASS || (((uintptr_t)ptr & PAGE_MASK) == 0)) { |
2839 | return false; |
2840 | } |
2841 | alloc_ctx.szind = sz_size2index_lookup(size); |
2842 | } |
2843 | |
2844 | if (unlikely(ticker_trytick(&tcache->gc_ticker))) { |
2845 | return false; |
2846 | } |
2847 | |
2848 | cache_bin_t *bin = tcache_small_bin_get(tcache, alloc_ctx.szind); |
2849 | cache_bin_info_t *bin_info = &tcache_bin_info[alloc_ctx.szind]; |
2850 | if (!cache_bin_dalloc_easy(bin, bin_info, ptr)) { |
2851 | return false; |
2852 | } |
2853 | |
2854 | if (config_stats) { |
2855 | size_t usize = sz_index2size(alloc_ctx.szind); |
2856 | *tsd_thread_deallocatedp_get(tsd) += usize; |
2857 | } |
2858 | |
2859 | return true; |
2860 | } |
2861 | |
2862 | JEMALLOC_EXPORT void JEMALLOC_NOTHROW |
2863 | je_free(void *ptr) { |
2864 | LOG("core.free.entry" , "ptr: %p" , ptr); |
2865 | |
2866 | if (!free_fastpath(ptr, 0, false)) { |
2867 | free_default(ptr); |
2868 | } |
2869 | |
2870 | LOG("core.free.exit" , "" ); |
2871 | } |
2872 | |
2873 | /* |
2874 | * End malloc(3)-compatible functions. |
2875 | */ |
2876 | /******************************************************************************/ |
2877 | /* |
2878 | * Begin non-standard override functions. |
2879 | */ |
2880 | |
2881 | #ifdef JEMALLOC_OVERRIDE_MEMALIGN |
2882 | JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN |
2883 | void JEMALLOC_NOTHROW * |
2884 | JEMALLOC_ATTR(malloc) |
2885 | je_memalign(size_t alignment, size_t size) { |
2886 | void *ret; |
2887 | static_opts_t sopts; |
2888 | dynamic_opts_t dopts; |
2889 | |
2890 | LOG("core.memalign.entry" , "alignment: %zu, size: %zu\n" , alignment, |
2891 | size); |
2892 | |
2893 | static_opts_init(&sopts); |
2894 | dynamic_opts_init(&dopts); |
2895 | |
2896 | sopts.min_alignment = 1; |
2897 | sopts.oom_string = |
2898 | "<jemalloc>: Error allocating aligned memory: out of memory\n" ; |
2899 | sopts.invalid_alignment_string = |
2900 | "<jemalloc>: Error allocating aligned memory: invalid alignment\n" ; |
2901 | sopts.null_out_result_on_error = true; |
2902 | |
2903 | dopts.result = &ret; |
2904 | dopts.num_items = 1; |
2905 | dopts.item_size = size; |
2906 | dopts.alignment = alignment; |
2907 | |
2908 | imalloc(&sopts, &dopts); |
2909 | if (sopts.slow) { |
2910 | uintptr_t args[3] = {alignment, size}; |
2911 | hook_invoke_alloc(hook_alloc_memalign, ret, (uintptr_t)ret, |
2912 | args); |
2913 | } |
2914 | |
2915 | LOG("core.memalign.exit" , "result: %p" , ret); |
2916 | return ret; |
2917 | } |
2918 | #endif |
2919 | |
2920 | #ifdef JEMALLOC_OVERRIDE_VALLOC |
2921 | JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN |
2922 | void JEMALLOC_NOTHROW * |
2923 | JEMALLOC_ATTR(malloc) |
2924 | je_valloc(size_t size) { |
2925 | void *ret; |
2926 | |
2927 | static_opts_t sopts; |
2928 | dynamic_opts_t dopts; |
2929 | |
2930 | LOG("core.valloc.entry" , "size: %zu\n" , size); |
2931 | |
2932 | static_opts_init(&sopts); |
2933 | dynamic_opts_init(&dopts); |
2934 | |
2935 | sopts.null_out_result_on_error = true; |
2936 | sopts.min_alignment = PAGE; |
2937 | sopts.oom_string = |
2938 | "<jemalloc>: Error allocating aligned memory: out of memory\n" ; |
2939 | sopts.invalid_alignment_string = |
2940 | "<jemalloc>: Error allocating aligned memory: invalid alignment\n" ; |
2941 | |
2942 | dopts.result = &ret; |
2943 | dopts.num_items = 1; |
2944 | dopts.item_size = size; |
2945 | dopts.alignment = PAGE; |
2946 | |
2947 | imalloc(&sopts, &dopts); |
2948 | if (sopts.slow) { |
2949 | uintptr_t args[3] = {size}; |
2950 | hook_invoke_alloc(hook_alloc_valloc, ret, (uintptr_t)ret, args); |
2951 | } |
2952 | |
2953 | LOG("core.valloc.exit" , "result: %p\n" , ret); |
2954 | return ret; |
2955 | } |
2956 | #endif |
2957 | |
2958 | #if defined(JEMALLOC_IS_MALLOC) && defined(JEMALLOC_GLIBC_MALLOC_HOOK) |
2959 | /* |
2960 | * glibc provides the RTLD_DEEPBIND flag for dlopen which can make it possible |
2961 | * to inconsistently reference libc's malloc(3)-compatible functions |
2962 | * (https://bugzilla.mozilla.org/show_bug.cgi?id=493541). |
2963 | * |
2964 | * These definitions interpose hooks in glibc. The functions are actually |
2965 | * passed an extra argument for the caller return address, which will be |
2966 | * ignored. |
2967 | */ |
2968 | JEMALLOC_EXPORT void (*__free_hook)(void *ptr) = je_free; |
2969 | JEMALLOC_EXPORT void *(*__malloc_hook)(size_t size) = je_malloc; |
2970 | JEMALLOC_EXPORT void *(*__realloc_hook)(void *ptr, size_t size) = je_realloc; |
2971 | # ifdef JEMALLOC_GLIBC_MEMALIGN_HOOK |
2972 | JEMALLOC_EXPORT void *(*__memalign_hook)(size_t alignment, size_t size) = |
2973 | je_memalign; |
2974 | # endif |
2975 | |
2976 | # ifdef CPU_COUNT |
2977 | /* |
2978 | * To enable static linking with glibc, the libc specific malloc interface must |
2979 | * be implemented also, so none of glibc's malloc.o functions are added to the |
2980 | * link. |
2981 | */ |
2982 | # define ALIAS(je_fn) __attribute__((alias (#je_fn), used)) |
2983 | /* To force macro expansion of je_ prefix before stringification. */ |
2984 | # define PREALIAS(je_fn) ALIAS(je_fn) |
2985 | # ifdef JEMALLOC_OVERRIDE___LIBC_CALLOC |
2986 | void *__libc_calloc(size_t n, size_t size) PREALIAS(je_calloc); |
2987 | # endif |
2988 | # ifdef JEMALLOC_OVERRIDE___LIBC_FREE |
2989 | void __libc_free(void* ptr) PREALIAS(je_free); |
2990 | # endif |
2991 | # ifdef JEMALLOC_OVERRIDE___LIBC_MALLOC |
2992 | void *__libc_malloc(size_t size) PREALIAS(je_malloc); |
2993 | # endif |
2994 | # ifdef JEMALLOC_OVERRIDE___LIBC_MEMALIGN |
2995 | void *__libc_memalign(size_t align, size_t s) PREALIAS(je_memalign); |
2996 | # endif |
2997 | # ifdef JEMALLOC_OVERRIDE___LIBC_REALLOC |
2998 | void *__libc_realloc(void* ptr, size_t size) PREALIAS(je_realloc); |
2999 | # endif |
3000 | # ifdef JEMALLOC_OVERRIDE___LIBC_VALLOC |
3001 | void *__libc_valloc(size_t size) PREALIAS(je_valloc); |
3002 | # endif |
3003 | # ifdef JEMALLOC_OVERRIDE___POSIX_MEMALIGN |
3004 | int __posix_memalign(void** r, size_t a, size_t s) PREALIAS(je_posix_memalign); |
3005 | # endif |
3006 | # undef PREALIAS |
3007 | # undef ALIAS |
3008 | # endif |
3009 | #endif |
3010 | |
3011 | /* |
3012 | * End non-standard override functions. |
3013 | */ |
3014 | /******************************************************************************/ |
3015 | /* |
3016 | * Begin non-standard functions. |
3017 | */ |
3018 | |
3019 | #ifdef JEMALLOC_EXPERIMENTAL_SMALLOCX_API |
3020 | |
3021 | #define JEMALLOC_SMALLOCX_CONCAT_HELPER(x, y) x ## y |
3022 | #define JEMALLOC_SMALLOCX_CONCAT_HELPER2(x, y) \ |
3023 | JEMALLOC_SMALLOCX_CONCAT_HELPER(x, y) |
3024 | |
3025 | typedef struct { |
3026 | void *ptr; |
3027 | size_t size; |
3028 | } smallocx_return_t; |
3029 | |
3030 | JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN |
3031 | smallocx_return_t JEMALLOC_NOTHROW |
3032 | /* |
3033 | * The attribute JEMALLOC_ATTR(malloc) cannot be used due to: |
3034 | * - https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86488 |
3035 | */ |
3036 | JEMALLOC_SMALLOCX_CONCAT_HELPER2(je_smallocx_, JEMALLOC_VERSION_GID_IDENT) |
3037 | (size_t size, int flags) { |
3038 | /* |
3039 | * Note: the attribute JEMALLOC_ALLOC_SIZE(1) cannot be |
3040 | * used here because it makes writing beyond the `size` |
3041 | * of the `ptr` undefined behavior, but the objective |
3042 | * of this function is to allow writing beyond `size` |
3043 | * up to `smallocx_return_t::size`. |
3044 | */ |
3045 | smallocx_return_t ret; |
3046 | static_opts_t sopts; |
3047 | dynamic_opts_t dopts; |
3048 | |
3049 | LOG("core.smallocx.entry" , "size: %zu, flags: %d" , size, flags); |
3050 | |
3051 | static_opts_init(&sopts); |
3052 | dynamic_opts_init(&dopts); |
3053 | |
3054 | sopts.assert_nonempty_alloc = true; |
3055 | sopts.null_out_result_on_error = true; |
3056 | sopts.oom_string = "<jemalloc>: Error in mallocx(): out of memory\n" ; |
3057 | sopts.usize = true; |
3058 | |
3059 | dopts.result = &ret.ptr; |
3060 | dopts.num_items = 1; |
3061 | dopts.item_size = size; |
3062 | if (unlikely(flags != 0)) { |
3063 | if ((flags & MALLOCX_LG_ALIGN_MASK) != 0) { |
3064 | dopts.alignment = MALLOCX_ALIGN_GET_SPECIFIED(flags); |
3065 | } |
3066 | |
3067 | dopts.zero = MALLOCX_ZERO_GET(flags); |
3068 | |
3069 | if ((flags & MALLOCX_TCACHE_MASK) != 0) { |
3070 | if ((flags & MALLOCX_TCACHE_MASK) |
3071 | == MALLOCX_TCACHE_NONE) { |
3072 | dopts.tcache_ind = TCACHE_IND_NONE; |
3073 | } else { |
3074 | dopts.tcache_ind = MALLOCX_TCACHE_GET(flags); |
3075 | } |
3076 | } else { |
3077 | dopts.tcache_ind = TCACHE_IND_AUTOMATIC; |
3078 | } |
3079 | |
3080 | if ((flags & MALLOCX_ARENA_MASK) != 0) |
3081 | dopts.arena_ind = MALLOCX_ARENA_GET(flags); |
3082 | } |
3083 | |
3084 | imalloc(&sopts, &dopts); |
3085 | assert(dopts.usize == je_nallocx(size, flags)); |
3086 | ret.size = dopts.usize; |
3087 | |
3088 | LOG("core.smallocx.exit" , "result: %p, size: %zu" , ret.ptr, ret.size); |
3089 | return ret; |
3090 | } |
3091 | #undef JEMALLOC_SMALLOCX_CONCAT_HELPER |
3092 | #undef JEMALLOC_SMALLOCX_CONCAT_HELPER2 |
3093 | #endif |
3094 | |
3095 | JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN |
3096 | void JEMALLOC_NOTHROW * |
3097 | JEMALLOC_ATTR(malloc) JEMALLOC_ALLOC_SIZE(1) |
3098 | je_mallocx(size_t size, int flags) { |
3099 | void *ret; |
3100 | static_opts_t sopts; |
3101 | dynamic_opts_t dopts; |
3102 | |
3103 | LOG("core.mallocx.entry" , "size: %zu, flags: %d" , size, flags); |
3104 | |
3105 | static_opts_init(&sopts); |
3106 | dynamic_opts_init(&dopts); |
3107 | |
3108 | sopts.assert_nonempty_alloc = true; |
3109 | sopts.null_out_result_on_error = true; |
3110 | sopts.oom_string = "<jemalloc>: Error in mallocx(): out of memory\n" ; |
3111 | |
3112 | dopts.result = &ret; |
3113 | dopts.num_items = 1; |
3114 | dopts.item_size = size; |
3115 | if (unlikely(flags != 0)) { |
3116 | if ((flags & MALLOCX_LG_ALIGN_MASK) != 0) { |
3117 | dopts.alignment = MALLOCX_ALIGN_GET_SPECIFIED(flags); |
3118 | } |
3119 | |
3120 | dopts.zero = MALLOCX_ZERO_GET(flags); |
3121 | |
3122 | if ((flags & MALLOCX_TCACHE_MASK) != 0) { |
3123 | if ((flags & MALLOCX_TCACHE_MASK) |
3124 | == MALLOCX_TCACHE_NONE) { |
3125 | dopts.tcache_ind = TCACHE_IND_NONE; |
3126 | } else { |
3127 | dopts.tcache_ind = MALLOCX_TCACHE_GET(flags); |
3128 | } |
3129 | } else { |
3130 | dopts.tcache_ind = TCACHE_IND_AUTOMATIC; |
3131 | } |
3132 | |
3133 | if ((flags & MALLOCX_ARENA_MASK) != 0) |
3134 | dopts.arena_ind = MALLOCX_ARENA_GET(flags); |
3135 | } |
3136 | |
3137 | imalloc(&sopts, &dopts); |
3138 | if (sopts.slow) { |
3139 | uintptr_t args[3] = {size, flags}; |
3140 | hook_invoke_alloc(hook_alloc_mallocx, ret, (uintptr_t)ret, |
3141 | args); |
3142 | } |
3143 | |
3144 | LOG("core.mallocx.exit" , "result: %p" , ret); |
3145 | return ret; |
3146 | } |
3147 | |
3148 | static void * |
3149 | irallocx_prof_sample(tsdn_t *tsdn, void *old_ptr, size_t old_usize, |
3150 | size_t usize, size_t alignment, bool zero, tcache_t *tcache, arena_t *arena, |
3151 | prof_tctx_t *tctx, hook_ralloc_args_t *hook_args) { |
3152 | void *p; |
3153 | |
3154 | if (tctx == NULL) { |
3155 | return NULL; |
3156 | } |
3157 | if (usize <= SC_SMALL_MAXCLASS) { |
3158 | p = iralloct(tsdn, old_ptr, old_usize, |
3159 | SC_LARGE_MINCLASS, alignment, zero, tcache, |
3160 | arena, hook_args); |
3161 | if (p == NULL) { |
3162 | return NULL; |
3163 | } |
3164 | arena_prof_promote(tsdn, p, usize); |
3165 | } else { |
3166 | p = iralloct(tsdn, old_ptr, old_usize, usize, alignment, zero, |
3167 | tcache, arena, hook_args); |
3168 | } |
3169 | |
3170 | return p; |
3171 | } |
3172 | |
3173 | JEMALLOC_ALWAYS_INLINE void * |
3174 | irallocx_prof(tsd_t *tsd, void *old_ptr, size_t old_usize, size_t size, |
3175 | size_t alignment, size_t *usize, bool zero, tcache_t *tcache, |
3176 | arena_t *arena, alloc_ctx_t *alloc_ctx, hook_ralloc_args_t *hook_args) { |
3177 | void *p; |
3178 | bool prof_active; |
3179 | prof_tctx_t *old_tctx, *tctx; |
3180 | |
3181 | prof_active = prof_active_get_unlocked(); |
3182 | old_tctx = prof_tctx_get(tsd_tsdn(tsd), old_ptr, alloc_ctx); |
3183 | tctx = prof_alloc_prep(tsd, *usize, prof_active, false); |
3184 | if (unlikely((uintptr_t)tctx != (uintptr_t)1U)) { |
3185 | p = irallocx_prof_sample(tsd_tsdn(tsd), old_ptr, old_usize, |
3186 | *usize, alignment, zero, tcache, arena, tctx, hook_args); |
3187 | } else { |
3188 | p = iralloct(tsd_tsdn(tsd), old_ptr, old_usize, size, alignment, |
3189 | zero, tcache, arena, hook_args); |
3190 | } |
3191 | if (unlikely(p == NULL)) { |
3192 | prof_alloc_rollback(tsd, tctx, false); |
3193 | return NULL; |
3194 | } |
3195 | |
3196 | if (p == old_ptr && alignment != 0) { |
3197 | /* |
3198 | * The allocation did not move, so it is possible that the size |
3199 | * class is smaller than would guarantee the requested |
3200 | * alignment, and that the alignment constraint was |
3201 | * serendipitously satisfied. Additionally, old_usize may not |
3202 | * be the same as the current usize because of in-place large |
3203 | * reallocation. Therefore, query the actual value of usize. |
3204 | */ |
3205 | *usize = isalloc(tsd_tsdn(tsd), p); |
3206 | } |
3207 | prof_realloc(tsd, p, *usize, tctx, prof_active, false, old_ptr, |
3208 | old_usize, old_tctx); |
3209 | |
3210 | return p; |
3211 | } |
3212 | |
3213 | JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN |
3214 | void JEMALLOC_NOTHROW * |
3215 | JEMALLOC_ALLOC_SIZE(2) |
3216 | je_rallocx(void *ptr, size_t size, int flags) { |
3217 | void *p; |
3218 | tsd_t *tsd; |
3219 | size_t usize; |
3220 | size_t old_usize; |
3221 | size_t alignment = MALLOCX_ALIGN_GET(flags); |
3222 | bool zero = flags & MALLOCX_ZERO; |
3223 | arena_t *arena; |
3224 | tcache_t *tcache; |
3225 | |
3226 | LOG("core.rallocx.entry" , "ptr: %p, size: %zu, flags: %d" , ptr, |
3227 | size, flags); |
3228 | |
3229 | |
3230 | assert(ptr != NULL); |
3231 | assert(size != 0); |
3232 | assert(malloc_initialized() || IS_INITIALIZER); |
3233 | tsd = tsd_fetch(); |
3234 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3235 | |
3236 | if (unlikely((flags & MALLOCX_ARENA_MASK) != 0)) { |
3237 | unsigned arena_ind = MALLOCX_ARENA_GET(flags); |
3238 | arena = arena_get(tsd_tsdn(tsd), arena_ind, true); |
3239 | if (unlikely(arena == NULL)) { |
3240 | goto label_oom; |
3241 | } |
3242 | } else { |
3243 | arena = NULL; |
3244 | } |
3245 | |
3246 | if (unlikely((flags & MALLOCX_TCACHE_MASK) != 0)) { |
3247 | if ((flags & MALLOCX_TCACHE_MASK) == MALLOCX_TCACHE_NONE) { |
3248 | tcache = NULL; |
3249 | } else { |
3250 | tcache = tcaches_get(tsd, MALLOCX_TCACHE_GET(flags)); |
3251 | } |
3252 | } else { |
3253 | tcache = tcache_get(tsd); |
3254 | } |
3255 | |
3256 | alloc_ctx_t alloc_ctx; |
3257 | rtree_ctx_t *rtree_ctx = tsd_rtree_ctx(tsd); |
3258 | rtree_szind_slab_read(tsd_tsdn(tsd), &extents_rtree, rtree_ctx, |
3259 | (uintptr_t)ptr, true, &alloc_ctx.szind, &alloc_ctx.slab); |
3260 | assert(alloc_ctx.szind != SC_NSIZES); |
3261 | old_usize = sz_index2size(alloc_ctx.szind); |
3262 | assert(old_usize == isalloc(tsd_tsdn(tsd), ptr)); |
3263 | |
3264 | hook_ralloc_args_t hook_args = {false, {(uintptr_t)ptr, size, flags, |
3265 | 0}}; |
3266 | if (config_prof && opt_prof) { |
3267 | usize = (alignment == 0) ? |
3268 | sz_s2u(size) : sz_sa2u(size, alignment); |
3269 | if (unlikely(usize == 0 |
3270 | || usize > SC_LARGE_MAXCLASS)) { |
3271 | goto label_oom; |
3272 | } |
3273 | p = irallocx_prof(tsd, ptr, old_usize, size, alignment, &usize, |
3274 | zero, tcache, arena, &alloc_ctx, &hook_args); |
3275 | if (unlikely(p == NULL)) { |
3276 | goto label_oom; |
3277 | } |
3278 | } else { |
3279 | p = iralloct(tsd_tsdn(tsd), ptr, old_usize, size, alignment, |
3280 | zero, tcache, arena, &hook_args); |
3281 | if (unlikely(p == NULL)) { |
3282 | goto label_oom; |
3283 | } |
3284 | if (config_stats) { |
3285 | usize = isalloc(tsd_tsdn(tsd), p); |
3286 | } |
3287 | } |
3288 | assert(alignment == 0 || ((uintptr_t)p & (alignment - 1)) == ZU(0)); |
3289 | |
3290 | if (config_stats) { |
3291 | *tsd_thread_allocatedp_get(tsd) += usize; |
3292 | *tsd_thread_deallocatedp_get(tsd) += old_usize; |
3293 | } |
3294 | UTRACE(ptr, size, p); |
3295 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3296 | |
3297 | LOG("core.rallocx.exit" , "result: %p" , p); |
3298 | return p; |
3299 | label_oom: |
3300 | if (config_xmalloc && unlikely(opt_xmalloc)) { |
3301 | malloc_write("<jemalloc>: Error in rallocx(): out of memory\n" ); |
3302 | abort(); |
3303 | } |
3304 | UTRACE(ptr, size, 0); |
3305 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3306 | |
3307 | LOG("core.rallocx.exit" , "result: %p" , NULL); |
3308 | return NULL; |
3309 | } |
3310 | |
3311 | JEMALLOC_ALWAYS_INLINE size_t |
3312 | ixallocx_helper(tsdn_t *tsdn, void *ptr, size_t old_usize, size_t size, |
3313 | size_t , size_t alignment, bool zero) { |
3314 | size_t newsize; |
3315 | |
3316 | if (ixalloc(tsdn, ptr, old_usize, size, extra, alignment, zero, |
3317 | &newsize)) { |
3318 | return old_usize; |
3319 | } |
3320 | |
3321 | return newsize; |
3322 | } |
3323 | |
3324 | static size_t |
3325 | ixallocx_prof_sample(tsdn_t *tsdn, void *ptr, size_t old_usize, size_t size, |
3326 | size_t , size_t alignment, bool zero, prof_tctx_t *tctx) { |
3327 | size_t usize; |
3328 | |
3329 | if (tctx == NULL) { |
3330 | return old_usize; |
3331 | } |
3332 | usize = ixallocx_helper(tsdn, ptr, old_usize, size, extra, alignment, |
3333 | zero); |
3334 | |
3335 | return usize; |
3336 | } |
3337 | |
3338 | JEMALLOC_ALWAYS_INLINE size_t |
3339 | ixallocx_prof(tsd_t *tsd, void *ptr, size_t old_usize, size_t size, |
3340 | size_t , size_t alignment, bool zero, alloc_ctx_t *alloc_ctx) { |
3341 | size_t usize_max, usize; |
3342 | bool prof_active; |
3343 | prof_tctx_t *old_tctx, *tctx; |
3344 | |
3345 | prof_active = prof_active_get_unlocked(); |
3346 | old_tctx = prof_tctx_get(tsd_tsdn(tsd), ptr, alloc_ctx); |
3347 | /* |
3348 | * usize isn't knowable before ixalloc() returns when extra is non-zero. |
3349 | * Therefore, compute its maximum possible value and use that in |
3350 | * prof_alloc_prep() to decide whether to capture a backtrace. |
3351 | * prof_realloc() will use the actual usize to decide whether to sample. |
3352 | */ |
3353 | if (alignment == 0) { |
3354 | usize_max = sz_s2u(size+extra); |
3355 | assert(usize_max > 0 |
3356 | && usize_max <= SC_LARGE_MAXCLASS); |
3357 | } else { |
3358 | usize_max = sz_sa2u(size+extra, alignment); |
3359 | if (unlikely(usize_max == 0 |
3360 | || usize_max > SC_LARGE_MAXCLASS)) { |
3361 | /* |
3362 | * usize_max is out of range, and chances are that |
3363 | * allocation will fail, but use the maximum possible |
3364 | * value and carry on with prof_alloc_prep(), just in |
3365 | * case allocation succeeds. |
3366 | */ |
3367 | usize_max = SC_LARGE_MAXCLASS; |
3368 | } |
3369 | } |
3370 | tctx = prof_alloc_prep(tsd, usize_max, prof_active, false); |
3371 | |
3372 | if (unlikely((uintptr_t)tctx != (uintptr_t)1U)) { |
3373 | usize = ixallocx_prof_sample(tsd_tsdn(tsd), ptr, old_usize, |
3374 | size, extra, alignment, zero, tctx); |
3375 | } else { |
3376 | usize = ixallocx_helper(tsd_tsdn(tsd), ptr, old_usize, size, |
3377 | extra, alignment, zero); |
3378 | } |
3379 | if (usize == old_usize) { |
3380 | prof_alloc_rollback(tsd, tctx, false); |
3381 | return usize; |
3382 | } |
3383 | prof_realloc(tsd, ptr, usize, tctx, prof_active, false, ptr, old_usize, |
3384 | old_tctx); |
3385 | |
3386 | return usize; |
3387 | } |
3388 | |
3389 | JEMALLOC_EXPORT size_t JEMALLOC_NOTHROW |
3390 | je_xallocx(void *ptr, size_t size, size_t , int flags) { |
3391 | tsd_t *tsd; |
3392 | size_t usize, old_usize; |
3393 | size_t alignment = MALLOCX_ALIGN_GET(flags); |
3394 | bool zero = flags & MALLOCX_ZERO; |
3395 | |
3396 | LOG("core.xallocx.entry" , "ptr: %p, size: %zu, extra: %zu, " |
3397 | "flags: %d" , ptr, size, extra, flags); |
3398 | |
3399 | assert(ptr != NULL); |
3400 | assert(size != 0); |
3401 | assert(SIZE_T_MAX - size >= extra); |
3402 | assert(malloc_initialized() || IS_INITIALIZER); |
3403 | tsd = tsd_fetch(); |
3404 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3405 | |
3406 | alloc_ctx_t alloc_ctx; |
3407 | rtree_ctx_t *rtree_ctx = tsd_rtree_ctx(tsd); |
3408 | rtree_szind_slab_read(tsd_tsdn(tsd), &extents_rtree, rtree_ctx, |
3409 | (uintptr_t)ptr, true, &alloc_ctx.szind, &alloc_ctx.slab); |
3410 | assert(alloc_ctx.szind != SC_NSIZES); |
3411 | old_usize = sz_index2size(alloc_ctx.szind); |
3412 | assert(old_usize == isalloc(tsd_tsdn(tsd), ptr)); |
3413 | /* |
3414 | * The API explicitly absolves itself of protecting against (size + |
3415 | * extra) numerical overflow, but we may need to clamp extra to avoid |
3416 | * exceeding SC_LARGE_MAXCLASS. |
3417 | * |
3418 | * Ordinarily, size limit checking is handled deeper down, but here we |
3419 | * have to check as part of (size + extra) clamping, since we need the |
3420 | * clamped value in the above helper functions. |
3421 | */ |
3422 | if (unlikely(size > SC_LARGE_MAXCLASS)) { |
3423 | usize = old_usize; |
3424 | goto label_not_resized; |
3425 | } |
3426 | if (unlikely(SC_LARGE_MAXCLASS - size < extra)) { |
3427 | extra = SC_LARGE_MAXCLASS - size; |
3428 | } |
3429 | |
3430 | if (config_prof && opt_prof) { |
3431 | usize = ixallocx_prof(tsd, ptr, old_usize, size, extra, |
3432 | alignment, zero, &alloc_ctx); |
3433 | } else { |
3434 | usize = ixallocx_helper(tsd_tsdn(tsd), ptr, old_usize, size, |
3435 | extra, alignment, zero); |
3436 | } |
3437 | if (unlikely(usize == old_usize)) { |
3438 | goto label_not_resized; |
3439 | } |
3440 | |
3441 | if (config_stats) { |
3442 | *tsd_thread_allocatedp_get(tsd) += usize; |
3443 | *tsd_thread_deallocatedp_get(tsd) += old_usize; |
3444 | } |
3445 | label_not_resized: |
3446 | if (unlikely(!tsd_fast(tsd))) { |
3447 | uintptr_t args[4] = {(uintptr_t)ptr, size, extra, flags}; |
3448 | hook_invoke_expand(hook_expand_xallocx, ptr, old_usize, |
3449 | usize, (uintptr_t)usize, args); |
3450 | } |
3451 | |
3452 | UTRACE(ptr, size, ptr); |
3453 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3454 | |
3455 | LOG("core.xallocx.exit" , "result: %zu" , usize); |
3456 | return usize; |
3457 | } |
3458 | |
3459 | JEMALLOC_EXPORT size_t JEMALLOC_NOTHROW |
3460 | JEMALLOC_ATTR(pure) |
3461 | je_sallocx(const void *ptr, int flags) { |
3462 | size_t usize; |
3463 | tsdn_t *tsdn; |
3464 | |
3465 | LOG("core.sallocx.entry" , "ptr: %p, flags: %d" , ptr, flags); |
3466 | |
3467 | assert(malloc_initialized() || IS_INITIALIZER); |
3468 | assert(ptr != NULL); |
3469 | |
3470 | tsdn = tsdn_fetch(); |
3471 | check_entry_exit_locking(tsdn); |
3472 | |
3473 | if (config_debug || force_ivsalloc) { |
3474 | usize = ivsalloc(tsdn, ptr); |
3475 | assert(force_ivsalloc || usize != 0); |
3476 | } else { |
3477 | usize = isalloc(tsdn, ptr); |
3478 | } |
3479 | |
3480 | check_entry_exit_locking(tsdn); |
3481 | |
3482 | LOG("core.sallocx.exit" , "result: %zu" , usize); |
3483 | return usize; |
3484 | } |
3485 | |
3486 | JEMALLOC_EXPORT void JEMALLOC_NOTHROW |
3487 | je_dallocx(void *ptr, int flags) { |
3488 | LOG("core.dallocx.entry" , "ptr: %p, flags: %d" , ptr, flags); |
3489 | |
3490 | assert(ptr != NULL); |
3491 | assert(malloc_initialized() || IS_INITIALIZER); |
3492 | |
3493 | tsd_t *tsd = tsd_fetch(); |
3494 | bool fast = tsd_fast(tsd); |
3495 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3496 | |
3497 | tcache_t *tcache; |
3498 | if (unlikely((flags & MALLOCX_TCACHE_MASK) != 0)) { |
3499 | /* Not allowed to be reentrant and specify a custom tcache. */ |
3500 | assert(tsd_reentrancy_level_get(tsd) == 0); |
3501 | if ((flags & MALLOCX_TCACHE_MASK) == MALLOCX_TCACHE_NONE) { |
3502 | tcache = NULL; |
3503 | } else { |
3504 | tcache = tcaches_get(tsd, MALLOCX_TCACHE_GET(flags)); |
3505 | } |
3506 | } else { |
3507 | if (likely(fast)) { |
3508 | tcache = tsd_tcachep_get(tsd); |
3509 | assert(tcache == tcache_get(tsd)); |
3510 | } else { |
3511 | if (likely(tsd_reentrancy_level_get(tsd) == 0)) { |
3512 | tcache = tcache_get(tsd); |
3513 | } else { |
3514 | tcache = NULL; |
3515 | } |
3516 | } |
3517 | } |
3518 | |
3519 | UTRACE(ptr, 0, 0); |
3520 | if (likely(fast)) { |
3521 | tsd_assert_fast(tsd); |
3522 | ifree(tsd, ptr, tcache, false); |
3523 | } else { |
3524 | uintptr_t args_raw[3] = {(uintptr_t)ptr, flags}; |
3525 | hook_invoke_dalloc(hook_dalloc_dallocx, ptr, args_raw); |
3526 | ifree(tsd, ptr, tcache, true); |
3527 | } |
3528 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3529 | |
3530 | LOG("core.dallocx.exit" , "" ); |
3531 | } |
3532 | |
3533 | JEMALLOC_ALWAYS_INLINE size_t |
3534 | inallocx(tsdn_t *tsdn, size_t size, int flags) { |
3535 | check_entry_exit_locking(tsdn); |
3536 | |
3537 | size_t usize; |
3538 | if (likely((flags & MALLOCX_LG_ALIGN_MASK) == 0)) { |
3539 | usize = sz_s2u(size); |
3540 | } else { |
3541 | usize = sz_sa2u(size, MALLOCX_ALIGN_GET_SPECIFIED(flags)); |
3542 | } |
3543 | check_entry_exit_locking(tsdn); |
3544 | return usize; |
3545 | } |
3546 | |
3547 | JEMALLOC_NOINLINE void |
3548 | sdallocx_default(void *ptr, size_t size, int flags) { |
3549 | assert(ptr != NULL); |
3550 | assert(malloc_initialized() || IS_INITIALIZER); |
3551 | |
3552 | tsd_t *tsd = tsd_fetch(); |
3553 | bool fast = tsd_fast(tsd); |
3554 | size_t usize = inallocx(tsd_tsdn(tsd), size, flags); |
3555 | assert(usize == isalloc(tsd_tsdn(tsd), ptr)); |
3556 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3557 | |
3558 | tcache_t *tcache; |
3559 | if (unlikely((flags & MALLOCX_TCACHE_MASK) != 0)) { |
3560 | /* Not allowed to be reentrant and specify a custom tcache. */ |
3561 | assert(tsd_reentrancy_level_get(tsd) == 0); |
3562 | if ((flags & MALLOCX_TCACHE_MASK) == MALLOCX_TCACHE_NONE) { |
3563 | tcache = NULL; |
3564 | } else { |
3565 | tcache = tcaches_get(tsd, MALLOCX_TCACHE_GET(flags)); |
3566 | } |
3567 | } else { |
3568 | if (likely(fast)) { |
3569 | tcache = tsd_tcachep_get(tsd); |
3570 | assert(tcache == tcache_get(tsd)); |
3571 | } else { |
3572 | if (likely(tsd_reentrancy_level_get(tsd) == 0)) { |
3573 | tcache = tcache_get(tsd); |
3574 | } else { |
3575 | tcache = NULL; |
3576 | } |
3577 | } |
3578 | } |
3579 | |
3580 | UTRACE(ptr, 0, 0); |
3581 | if (likely(fast)) { |
3582 | tsd_assert_fast(tsd); |
3583 | isfree(tsd, ptr, usize, tcache, false); |
3584 | } else { |
3585 | uintptr_t args_raw[3] = {(uintptr_t)ptr, size, flags}; |
3586 | hook_invoke_dalloc(hook_dalloc_sdallocx, ptr, args_raw); |
3587 | isfree(tsd, ptr, usize, tcache, true); |
3588 | } |
3589 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3590 | |
3591 | } |
3592 | |
3593 | JEMALLOC_EXPORT void JEMALLOC_NOTHROW |
3594 | je_sdallocx(void *ptr, size_t size, int flags) { |
3595 | LOG("core.sdallocx.entry" , "ptr: %p, size: %zu, flags: %d" , ptr, |
3596 | size, flags); |
3597 | |
3598 | if (flags !=0 || !free_fastpath(ptr, size, true)) { |
3599 | sdallocx_default(ptr, size, flags); |
3600 | } |
3601 | |
3602 | LOG("core.sdallocx.exit" , "" ); |
3603 | } |
3604 | |
3605 | void JEMALLOC_NOTHROW |
3606 | je_sdallocx_noflags(void *ptr, size_t size) { |
3607 | LOG("core.sdallocx.entry" , "ptr: %p, size: %zu, flags: 0" , ptr, |
3608 | size); |
3609 | |
3610 | if (!free_fastpath(ptr, size, true)) { |
3611 | sdallocx_default(ptr, size, 0); |
3612 | } |
3613 | |
3614 | LOG("core.sdallocx.exit" , "" ); |
3615 | } |
3616 | |
3617 | JEMALLOC_EXPORT size_t JEMALLOC_NOTHROW |
3618 | JEMALLOC_ATTR(pure) |
3619 | je_nallocx(size_t size, int flags) { |
3620 | size_t usize; |
3621 | tsdn_t *tsdn; |
3622 | |
3623 | assert(size != 0); |
3624 | |
3625 | if (unlikely(malloc_init())) { |
3626 | LOG("core.nallocx.exit" , "result: %zu" , ZU(0)); |
3627 | return 0; |
3628 | } |
3629 | |
3630 | tsdn = tsdn_fetch(); |
3631 | check_entry_exit_locking(tsdn); |
3632 | |
3633 | usize = inallocx(tsdn, size, flags); |
3634 | if (unlikely(usize > SC_LARGE_MAXCLASS)) { |
3635 | LOG("core.nallocx.exit" , "result: %zu" , ZU(0)); |
3636 | return 0; |
3637 | } |
3638 | |
3639 | check_entry_exit_locking(tsdn); |
3640 | LOG("core.nallocx.exit" , "result: %zu" , usize); |
3641 | return usize; |
3642 | } |
3643 | |
3644 | JEMALLOC_EXPORT int JEMALLOC_NOTHROW |
3645 | je_mallctl(const char *name, void *oldp, size_t *oldlenp, void *newp, |
3646 | size_t newlen) { |
3647 | int ret; |
3648 | tsd_t *tsd; |
3649 | |
3650 | LOG("core.mallctl.entry" , "name: %s" , name); |
3651 | |
3652 | if (unlikely(malloc_init())) { |
3653 | LOG("core.mallctl.exit" , "result: %d" , EAGAIN); |
3654 | return EAGAIN; |
3655 | } |
3656 | |
3657 | tsd = tsd_fetch(); |
3658 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3659 | ret = ctl_byname(tsd, name, oldp, oldlenp, newp, newlen); |
3660 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3661 | |
3662 | LOG("core.mallctl.exit" , "result: %d" , ret); |
3663 | return ret; |
3664 | } |
3665 | |
3666 | JEMALLOC_EXPORT int JEMALLOC_NOTHROW |
3667 | je_mallctlnametomib(const char *name, size_t *mibp, size_t *miblenp) { |
3668 | int ret; |
3669 | |
3670 | LOG("core.mallctlnametomib.entry" , "name: %s" , name); |
3671 | |
3672 | if (unlikely(malloc_init())) { |
3673 | LOG("core.mallctlnametomib.exit" , "result: %d" , EAGAIN); |
3674 | return EAGAIN; |
3675 | } |
3676 | |
3677 | tsd_t *tsd = tsd_fetch(); |
3678 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3679 | ret = ctl_nametomib(tsd, name, mibp, miblenp); |
3680 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3681 | |
3682 | LOG("core.mallctlnametomib.exit" , "result: %d" , ret); |
3683 | return ret; |
3684 | } |
3685 | |
3686 | JEMALLOC_EXPORT int JEMALLOC_NOTHROW |
3687 | je_mallctlbymib(const size_t *mib, size_t miblen, void *oldp, size_t *oldlenp, |
3688 | void *newp, size_t newlen) { |
3689 | int ret; |
3690 | tsd_t *tsd; |
3691 | |
3692 | LOG("core.mallctlbymib.entry" , "" ); |
3693 | |
3694 | if (unlikely(malloc_init())) { |
3695 | LOG("core.mallctlbymib.exit" , "result: %d" , EAGAIN); |
3696 | return EAGAIN; |
3697 | } |
3698 | |
3699 | tsd = tsd_fetch(); |
3700 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3701 | ret = ctl_bymib(tsd, mib, miblen, oldp, oldlenp, newp, newlen); |
3702 | check_entry_exit_locking(tsd_tsdn(tsd)); |
3703 | LOG("core.mallctlbymib.exit" , "result: %d" , ret); |
3704 | return ret; |
3705 | } |
3706 | |
3707 | JEMALLOC_EXPORT void JEMALLOC_NOTHROW |
3708 | je_malloc_stats_print(void (*write_cb)(void *, const char *), void *cbopaque, |
3709 | const char *opts) { |
3710 | tsdn_t *tsdn; |
3711 | |
3712 | LOG("core.malloc_stats_print.entry" , "" ); |
3713 | |
3714 | tsdn = tsdn_fetch(); |
3715 | check_entry_exit_locking(tsdn); |
3716 | stats_print(write_cb, cbopaque, opts); |
3717 | check_entry_exit_locking(tsdn); |
3718 | LOG("core.malloc_stats_print.exit" , "" ); |
3719 | } |
3720 | |
3721 | JEMALLOC_EXPORT size_t JEMALLOC_NOTHROW |
3722 | je_malloc_usable_size(JEMALLOC_USABLE_SIZE_CONST void *ptr) { |
3723 | size_t ret; |
3724 | tsdn_t *tsdn; |
3725 | |
3726 | LOG("core.malloc_usable_size.entry" , "ptr: %p" , ptr); |
3727 | |
3728 | assert(malloc_initialized() || IS_INITIALIZER); |
3729 | |
3730 | tsdn = tsdn_fetch(); |
3731 | check_entry_exit_locking(tsdn); |
3732 | |
3733 | if (unlikely(ptr == NULL)) { |
3734 | ret = 0; |
3735 | } else { |
3736 | if (config_debug || force_ivsalloc) { |
3737 | ret = ivsalloc(tsdn, ptr); |
3738 | assert(force_ivsalloc || ret != 0); |
3739 | } else { |
3740 | ret = isalloc(tsdn, ptr); |
3741 | } |
3742 | } |
3743 | |
3744 | check_entry_exit_locking(tsdn); |
3745 | LOG("core.malloc_usable_size.exit" , "result: %zu" , ret); |
3746 | return ret; |
3747 | } |
3748 | |
3749 | /* |
3750 | * End non-standard functions. |
3751 | */ |
3752 | /******************************************************************************/ |
3753 | /* |
3754 | * The following functions are used by threading libraries for protection of |
3755 | * malloc during fork(). |
3756 | */ |
3757 | |
3758 | /* |
3759 | * If an application creates a thread before doing any allocation in the main |
3760 | * thread, then calls fork(2) in the main thread followed by memory allocation |
3761 | * in the child process, a race can occur that results in deadlock within the |
3762 | * child: the main thread may have forked while the created thread had |
3763 | * partially initialized the allocator. Ordinarily jemalloc prevents |
3764 | * fork/malloc races via the following functions it registers during |
3765 | * initialization using pthread_atfork(), but of course that does no good if |
3766 | * the allocator isn't fully initialized at fork time. The following library |
3767 | * constructor is a partial solution to this problem. It may still be possible |
3768 | * to trigger the deadlock described above, but doing so would involve forking |
3769 | * via a library constructor that runs before jemalloc's runs. |
3770 | */ |
3771 | #ifndef JEMALLOC_JET |
3772 | JEMALLOC_ATTR(constructor) |
3773 | static void |
3774 | jemalloc_constructor(void) { |
3775 | malloc_init(); |
3776 | } |
3777 | #endif |
3778 | |
3779 | #ifndef JEMALLOC_MUTEX_INIT_CB |
3780 | void |
3781 | jemalloc_prefork(void) |
3782 | #else |
3783 | JEMALLOC_EXPORT void |
3784 | _malloc_prefork(void) |
3785 | #endif |
3786 | { |
3787 | tsd_t *tsd; |
3788 | unsigned i, j, narenas; |
3789 | arena_t *arena; |
3790 | |
3791 | #ifdef JEMALLOC_MUTEX_INIT_CB |
3792 | if (!malloc_initialized()) { |
3793 | return; |
3794 | } |
3795 | #endif |
3796 | assert(malloc_initialized()); |
3797 | |
3798 | tsd = tsd_fetch(); |
3799 | |
3800 | narenas = narenas_total_get(); |
3801 | |
3802 | witness_prefork(tsd_witness_tsdp_get(tsd)); |
3803 | /* Acquire all mutexes in a safe order. */ |
3804 | ctl_prefork(tsd_tsdn(tsd)); |
3805 | tcache_prefork(tsd_tsdn(tsd)); |
3806 | malloc_mutex_prefork(tsd_tsdn(tsd), &arenas_lock); |
3807 | if (have_background_thread) { |
3808 | background_thread_prefork0(tsd_tsdn(tsd)); |
3809 | } |
3810 | prof_prefork0(tsd_tsdn(tsd)); |
3811 | if (have_background_thread) { |
3812 | background_thread_prefork1(tsd_tsdn(tsd)); |
3813 | } |
3814 | /* Break arena prefork into stages to preserve lock order. */ |
3815 | for (i = 0; i < 8; i++) { |
3816 | for (j = 0; j < narenas; j++) { |
3817 | if ((arena = arena_get(tsd_tsdn(tsd), j, false)) != |
3818 | NULL) { |
3819 | switch (i) { |
3820 | case 0: |
3821 | arena_prefork0(tsd_tsdn(tsd), arena); |
3822 | break; |
3823 | case 1: |
3824 | arena_prefork1(tsd_tsdn(tsd), arena); |
3825 | break; |
3826 | case 2: |
3827 | arena_prefork2(tsd_tsdn(tsd), arena); |
3828 | break; |
3829 | case 3: |
3830 | arena_prefork3(tsd_tsdn(tsd), arena); |
3831 | break; |
3832 | case 4: |
3833 | arena_prefork4(tsd_tsdn(tsd), arena); |
3834 | break; |
3835 | case 5: |
3836 | arena_prefork5(tsd_tsdn(tsd), arena); |
3837 | break; |
3838 | case 6: |
3839 | arena_prefork6(tsd_tsdn(tsd), arena); |
3840 | break; |
3841 | case 7: |
3842 | arena_prefork7(tsd_tsdn(tsd), arena); |
3843 | break; |
3844 | default: not_reached(); |
3845 | } |
3846 | } |
3847 | } |
3848 | } |
3849 | prof_prefork1(tsd_tsdn(tsd)); |
3850 | tsd_prefork(tsd); |
3851 | } |
3852 | |
3853 | #ifndef JEMALLOC_MUTEX_INIT_CB |
3854 | void |
3855 | jemalloc_postfork_parent(void) |
3856 | #else |
3857 | JEMALLOC_EXPORT void |
3858 | _malloc_postfork(void) |
3859 | #endif |
3860 | { |
3861 | tsd_t *tsd; |
3862 | unsigned i, narenas; |
3863 | |
3864 | #ifdef JEMALLOC_MUTEX_INIT_CB |
3865 | if (!malloc_initialized()) { |
3866 | return; |
3867 | } |
3868 | #endif |
3869 | assert(malloc_initialized()); |
3870 | |
3871 | tsd = tsd_fetch(); |
3872 | |
3873 | tsd_postfork_parent(tsd); |
3874 | |
3875 | witness_postfork_parent(tsd_witness_tsdp_get(tsd)); |
3876 | /* Release all mutexes, now that fork() has completed. */ |
3877 | for (i = 0, narenas = narenas_total_get(); i < narenas; i++) { |
3878 | arena_t *arena; |
3879 | |
3880 | if ((arena = arena_get(tsd_tsdn(tsd), i, false)) != NULL) { |
3881 | arena_postfork_parent(tsd_tsdn(tsd), arena); |
3882 | } |
3883 | } |
3884 | prof_postfork_parent(tsd_tsdn(tsd)); |
3885 | if (have_background_thread) { |
3886 | background_thread_postfork_parent(tsd_tsdn(tsd)); |
3887 | } |
3888 | malloc_mutex_postfork_parent(tsd_tsdn(tsd), &arenas_lock); |
3889 | tcache_postfork_parent(tsd_tsdn(tsd)); |
3890 | ctl_postfork_parent(tsd_tsdn(tsd)); |
3891 | } |
3892 | |
3893 | void |
3894 | jemalloc_postfork_child(void) { |
3895 | tsd_t *tsd; |
3896 | unsigned i, narenas; |
3897 | |
3898 | assert(malloc_initialized()); |
3899 | |
3900 | tsd = tsd_fetch(); |
3901 | |
3902 | tsd_postfork_child(tsd); |
3903 | |
3904 | witness_postfork_child(tsd_witness_tsdp_get(tsd)); |
3905 | /* Release all mutexes, now that fork() has completed. */ |
3906 | for (i = 0, narenas = narenas_total_get(); i < narenas; i++) { |
3907 | arena_t *arena; |
3908 | |
3909 | if ((arena = arena_get(tsd_tsdn(tsd), i, false)) != NULL) { |
3910 | arena_postfork_child(tsd_tsdn(tsd), arena); |
3911 | } |
3912 | } |
3913 | prof_postfork_child(tsd_tsdn(tsd)); |
3914 | if (have_background_thread) { |
3915 | background_thread_postfork_child(tsd_tsdn(tsd)); |
3916 | } |
3917 | malloc_mutex_postfork_child(tsd_tsdn(tsd), &arenas_lock); |
3918 | tcache_postfork_child(tsd_tsdn(tsd)); |
3919 | ctl_postfork_child(tsd_tsdn(tsd)); |
3920 | } |
3921 | |
3922 | /******************************************************************************/ |
3923 | |
3924 | /* Helps the application decide if a pointer is worth re-allocating in order to reduce fragmentation. |
3925 | * returns 1 if the allocation should be moved, and 0 if the allocation be kept. |
3926 | * If the application decides to re-allocate it should use MALLOCX_TCACHE_NONE when doing so. */ |
3927 | JEMALLOC_EXPORT int JEMALLOC_NOTHROW |
3928 | get_defrag_hint(void* ptr) { |
3929 | assert(ptr != NULL); |
3930 | return iget_defrag_hint(TSDN_NULL, ptr); |
3931 | } |
3932 | |