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