]> git.proxmox.com Git - rustc.git/blob - src/jemalloc/src/jemalloc.c
Imported Upstream version 1.9.0+dfsg1
[rustc.git] / src / jemalloc / src / jemalloc.c
1 #define JEMALLOC_C_
2 #include "jemalloc/internal/jemalloc_internal.h"
3
4 /******************************************************************************/
5 /* Data. */
6
7 /* Runtime configuration options. */
8 const char *je_malloc_conf JEMALLOC_ATTR(weak);
9 bool opt_abort =
10 #ifdef JEMALLOC_DEBUG
11 true
12 #else
13 false
14 #endif
15 ;
16 const char *opt_junk =
17 #if (defined(JEMALLOC_DEBUG) && defined(JEMALLOC_FILL))
18 "true"
19 #else
20 "false"
21 #endif
22 ;
23 bool opt_junk_alloc =
24 #if (defined(JEMALLOC_DEBUG) && defined(JEMALLOC_FILL))
25 true
26 #else
27 false
28 #endif
29 ;
30 bool opt_junk_free =
31 #if (defined(JEMALLOC_DEBUG) && defined(JEMALLOC_FILL))
32 true
33 #else
34 false
35 #endif
36 ;
37
38 size_t opt_quarantine = ZU(0);
39 bool opt_redzone = false;
40 bool opt_utrace = false;
41 bool opt_xmalloc = false;
42 bool opt_zero = false;
43 unsigned opt_narenas = 0;
44
45 /* Initialized to true if the process is running inside Valgrind. */
46 bool in_valgrind;
47
48 unsigned ncpus;
49
50 /* Protects arenas initialization. */
51 static malloc_mutex_t arenas_lock;
52 /*
53 * Arenas that are used to service external requests. Not all elements of the
54 * arenas array are necessarily used; arenas are created lazily as needed.
55 *
56 * arenas[0..narenas_auto) are used for automatic multiplexing of threads and
57 * arenas. arenas[narenas_auto..narenas_total) are only used if the application
58 * takes some action to create them and allocate from them.
59 */
60 arena_t **arenas;
61 static unsigned narenas_total; /* Use narenas_total_*(). */
62 static arena_t *a0; /* arenas[0]; read-only after initialization. */
63 static unsigned narenas_auto; /* Read-only after initialization. */
64
65 typedef enum {
66 malloc_init_uninitialized = 3,
67 malloc_init_a0_initialized = 2,
68 malloc_init_recursible = 1,
69 malloc_init_initialized = 0 /* Common case --> jnz. */
70 } malloc_init_t;
71 static malloc_init_t malloc_init_state = malloc_init_uninitialized;
72
73 /* 0 should be the common case. Set to true to trigger initialization. */
74 static bool malloc_slow = true;
75
76 /* When malloc_slow != 0, set the corresponding bits for sanity check. */
77 enum {
78 flag_opt_junk_alloc = (1U),
79 flag_opt_junk_free = (1U << 1),
80 flag_opt_quarantine = (1U << 2),
81 flag_opt_zero = (1U << 3),
82 flag_opt_utrace = (1U << 4),
83 flag_in_valgrind = (1U << 5),
84 flag_opt_xmalloc = (1U << 6)
85 };
86 static uint8_t malloc_slow_flags;
87
88 /* Last entry for overflow detection only. */
89 JEMALLOC_ALIGNED(CACHELINE)
90 const size_t index2size_tab[NSIZES+1] = {
91 #define SC(index, lg_grp, lg_delta, ndelta, bin, lg_delta_lookup) \
92 ((ZU(1)<<lg_grp) + (ZU(ndelta)<<lg_delta)),
93 SIZE_CLASSES
94 #undef SC
95 ZU(0)
96 };
97
98 JEMALLOC_ALIGNED(CACHELINE)
99 const uint8_t size2index_tab[] = {
100 #if LG_TINY_MIN == 0
101 #warning "Dangerous LG_TINY_MIN"
102 #define S2B_0(i) i,
103 #elif LG_TINY_MIN == 1
104 #warning "Dangerous LG_TINY_MIN"
105 #define S2B_1(i) i,
106 #elif LG_TINY_MIN == 2
107 #warning "Dangerous LG_TINY_MIN"
108 #define S2B_2(i) i,
109 #elif LG_TINY_MIN == 3
110 #define S2B_3(i) i,
111 #elif LG_TINY_MIN == 4
112 #define S2B_4(i) i,
113 #elif LG_TINY_MIN == 5
114 #define S2B_5(i) i,
115 #elif LG_TINY_MIN == 6
116 #define S2B_6(i) i,
117 #elif LG_TINY_MIN == 7
118 #define S2B_7(i) i,
119 #elif LG_TINY_MIN == 8
120 #define S2B_8(i) i,
121 #elif LG_TINY_MIN == 9
122 #define S2B_9(i) i,
123 #elif LG_TINY_MIN == 10
124 #define S2B_10(i) i,
125 #elif LG_TINY_MIN == 11
126 #define S2B_11(i) i,
127 #else
128 #error "Unsupported LG_TINY_MIN"
129 #endif
130 #if LG_TINY_MIN < 1
131 #define S2B_1(i) S2B_0(i) S2B_0(i)
132 #endif
133 #if LG_TINY_MIN < 2
134 #define S2B_2(i) S2B_1(i) S2B_1(i)
135 #endif
136 #if LG_TINY_MIN < 3
137 #define S2B_3(i) S2B_2(i) S2B_2(i)
138 #endif
139 #if LG_TINY_MIN < 4
140 #define S2B_4(i) S2B_3(i) S2B_3(i)
141 #endif
142 #if LG_TINY_MIN < 5
143 #define S2B_5(i) S2B_4(i) S2B_4(i)
144 #endif
145 #if LG_TINY_MIN < 6
146 #define S2B_6(i) S2B_5(i) S2B_5(i)
147 #endif
148 #if LG_TINY_MIN < 7
149 #define S2B_7(i) S2B_6(i) S2B_6(i)
150 #endif
151 #if LG_TINY_MIN < 8
152 #define S2B_8(i) S2B_7(i) S2B_7(i)
153 #endif
154 #if LG_TINY_MIN < 9
155 #define S2B_9(i) S2B_8(i) S2B_8(i)
156 #endif
157 #if LG_TINY_MIN < 10
158 #define S2B_10(i) S2B_9(i) S2B_9(i)
159 #endif
160 #if LG_TINY_MIN < 11
161 #define S2B_11(i) S2B_10(i) S2B_10(i)
162 #endif
163 #define S2B_no(i)
164 #define SC(index, lg_grp, lg_delta, ndelta, bin, lg_delta_lookup) \
165 S2B_##lg_delta_lookup(index)
166 SIZE_CLASSES
167 #undef S2B_3
168 #undef S2B_4
169 #undef S2B_5
170 #undef S2B_6
171 #undef S2B_7
172 #undef S2B_8
173 #undef S2B_9
174 #undef S2B_10
175 #undef S2B_11
176 #undef S2B_no
177 #undef SC
178 };
179
180 #ifdef JEMALLOC_THREADED_INIT
181 /* Used to let the initializing thread recursively allocate. */
182 # define NO_INITIALIZER ((unsigned long)0)
183 # define INITIALIZER pthread_self()
184 # define IS_INITIALIZER (malloc_initializer == pthread_self())
185 static pthread_t malloc_initializer = NO_INITIALIZER;
186 #else
187 # define NO_INITIALIZER false
188 # define INITIALIZER true
189 # define IS_INITIALIZER malloc_initializer
190 static bool malloc_initializer = NO_INITIALIZER;
191 #endif
192
193 /* Used to avoid initialization races. */
194 #ifdef _WIN32
195 #if _WIN32_WINNT >= 0x0600
196 static malloc_mutex_t init_lock = SRWLOCK_INIT;
197 #else
198 static malloc_mutex_t init_lock;
199 static bool init_lock_initialized = false;
200
201 JEMALLOC_ATTR(constructor)
202 static void WINAPI
203 _init_init_lock(void)
204 {
205
206 /* If another constructor in the same binary is using mallctl to
207 * e.g. setup chunk hooks, it may end up running before this one,
208 * and malloc_init_hard will crash trying to lock the uninitialized
209 * lock. So we force an initialization of the lock in
210 * malloc_init_hard as well. We don't try to care about atomicity
211 * of the accessed to the init_lock_initialized boolean, since it
212 * really only matters early in the process creation, before any
213 * separate thread normally starts doing anything. */
214 if (!init_lock_initialized)
215 malloc_mutex_init(&init_lock);
216 init_lock_initialized = true;
217 }
218
219 #ifdef _MSC_VER
220 # pragma section(".CRT$XCU", read)
221 JEMALLOC_SECTION(".CRT$XCU") JEMALLOC_ATTR(used)
222 static const void (WINAPI *init_init_lock)(void) = _init_init_lock;
223 #endif
224 #endif
225 #else
226 static malloc_mutex_t init_lock = MALLOC_MUTEX_INITIALIZER;
227 #endif
228
229 typedef struct {
230 void *p; /* Input pointer (as in realloc(p, s)). */
231 size_t s; /* Request size. */
232 void *r; /* Result pointer. */
233 } malloc_utrace_t;
234
235 #ifdef JEMALLOC_UTRACE
236 # define UTRACE(a, b, c) do { \
237 if (unlikely(opt_utrace)) { \
238 int utrace_serrno = errno; \
239 malloc_utrace_t ut; \
240 ut.p = (a); \
241 ut.s = (b); \
242 ut.r = (c); \
243 utrace(&ut, sizeof(ut)); \
244 errno = utrace_serrno; \
245 } \
246 } while (0)
247 #else
248 # define UTRACE(a, b, c)
249 #endif
250
251 /******************************************************************************/
252 /*
253 * Function prototypes for static functions that are referenced prior to
254 * definition.
255 */
256
257 static bool malloc_init_hard_a0(void);
258 static bool malloc_init_hard(void);
259
260 /******************************************************************************/
261 /*
262 * Begin miscellaneous support functions.
263 */
264
265 JEMALLOC_ALWAYS_INLINE_C bool
266 malloc_initialized(void)
267 {
268
269 return (malloc_init_state == malloc_init_initialized);
270 }
271
272 JEMALLOC_ALWAYS_INLINE_C void
273 malloc_thread_init(void)
274 {
275
276 /*
277 * TSD initialization can't be safely done as a side effect of
278 * deallocation, because it is possible for a thread to do nothing but
279 * deallocate its TLS data via free(), in which case writing to TLS
280 * would cause write-after-free memory corruption. The quarantine
281 * facility *only* gets used as a side effect of deallocation, so make
282 * a best effort attempt at initializing its TSD by hooking all
283 * allocation events.
284 */
285 if (config_fill && unlikely(opt_quarantine))
286 quarantine_alloc_hook();
287 }
288
289 JEMALLOC_ALWAYS_INLINE_C bool
290 malloc_init_a0(void)
291 {
292
293 if (unlikely(malloc_init_state == malloc_init_uninitialized))
294 return (malloc_init_hard_a0());
295 return (false);
296 }
297
298 JEMALLOC_ALWAYS_INLINE_C bool
299 malloc_init(void)
300 {
301
302 if (unlikely(!malloc_initialized()) && malloc_init_hard())
303 return (true);
304 malloc_thread_init();
305
306 return (false);
307 }
308
309 /*
310 * The a0*() functions are used instead of i[mcd]alloc() in situations that
311 * cannot tolerate TLS variable access.
312 */
313
314 static void *
315 a0ialloc(size_t size, bool zero, bool is_metadata)
316 {
317
318 if (unlikely(malloc_init_a0()))
319 return (NULL);
320
321 return (iallocztm(NULL, size, size2index(size), zero, false,
322 is_metadata, arena_get(0, false), true));
323 }
324
325 static void
326 a0idalloc(void *ptr, bool is_metadata)
327 {
328
329 idalloctm(NULL, ptr, false, is_metadata, true);
330 }
331
332 void *
333 a0malloc(size_t size)
334 {
335
336 return (a0ialloc(size, false, true));
337 }
338
339 void
340 a0dalloc(void *ptr)
341 {
342
343 a0idalloc(ptr, true);
344 }
345
346 /*
347 * FreeBSD's libc uses the bootstrap_*() functions in bootstrap-senstive
348 * situations that cannot tolerate TLS variable access (TLS allocation and very
349 * early internal data structure initialization).
350 */
351
352 void *
353 bootstrap_malloc(size_t size)
354 {
355
356 if (unlikely(size == 0))
357 size = 1;
358
359 return (a0ialloc(size, false, false));
360 }
361
362 void *
363 bootstrap_calloc(size_t num, size_t size)
364 {
365 size_t num_size;
366
367 num_size = num * size;
368 if (unlikely(num_size == 0)) {
369 assert(num == 0 || size == 0);
370 num_size = 1;
371 }
372
373 return (a0ialloc(num_size, true, false));
374 }
375
376 void
377 bootstrap_free(void *ptr)
378 {
379
380 if (unlikely(ptr == NULL))
381 return;
382
383 a0idalloc(ptr, false);
384 }
385
386 static void
387 arena_set(unsigned ind, arena_t *arena)
388 {
389
390 atomic_write_p((void **)&arenas[ind], arena);
391 }
392
393 static void
394 narenas_total_set(unsigned narenas)
395 {
396
397 atomic_write_u(&narenas_total, narenas);
398 }
399
400 static void
401 narenas_total_inc(void)
402 {
403
404 atomic_add_u(&narenas_total, 1);
405 }
406
407 unsigned
408 narenas_total_get(void)
409 {
410
411 return (atomic_read_u(&narenas_total));
412 }
413
414 /* Create a new arena and insert it into the arenas array at index ind. */
415 static arena_t *
416 arena_init_locked(unsigned ind)
417 {
418 arena_t *arena;
419
420 assert(ind <= narenas_total_get());
421 if (ind > MALLOCX_ARENA_MAX)
422 return (NULL);
423 if (ind == narenas_total_get())
424 narenas_total_inc();
425
426 /*
427 * Another thread may have already initialized arenas[ind] if it's an
428 * auto arena.
429 */
430 arena = arena_get(ind, false);
431 if (arena != NULL) {
432 assert(ind < narenas_auto);
433 return (arena);
434 }
435
436 /* Actually initialize the arena. */
437 arena = arena_new(ind);
438 arena_set(ind, arena);
439 return (arena);
440 }
441
442 arena_t *
443 arena_init(unsigned ind)
444 {
445 arena_t *arena;
446
447 malloc_mutex_lock(&arenas_lock);
448 arena = arena_init_locked(ind);
449 malloc_mutex_unlock(&arenas_lock);
450 return (arena);
451 }
452
453 static void
454 arena_bind(tsd_t *tsd, unsigned ind)
455 {
456 arena_t *arena;
457
458 arena = arena_get(ind, false);
459 arena_nthreads_inc(arena);
460
461 if (tsd_nominal(tsd))
462 tsd_arena_set(tsd, arena);
463 }
464
465 void
466 arena_migrate(tsd_t *tsd, unsigned oldind, unsigned newind)
467 {
468 arena_t *oldarena, *newarena;
469
470 oldarena = arena_get(oldind, false);
471 newarena = arena_get(newind, false);
472 arena_nthreads_dec(oldarena);
473 arena_nthreads_inc(newarena);
474 tsd_arena_set(tsd, newarena);
475 }
476
477 static void
478 arena_unbind(tsd_t *tsd, unsigned ind)
479 {
480 arena_t *arena;
481
482 arena = arena_get(ind, false);
483 arena_nthreads_dec(arena);
484 tsd_arena_set(tsd, NULL);
485 }
486
487 arena_tdata_t *
488 arena_tdata_get_hard(tsd_t *tsd, unsigned ind)
489 {
490 arena_tdata_t *tdata, *arenas_tdata_old;
491 arena_tdata_t *arenas_tdata = tsd_arenas_tdata_get(tsd);
492 unsigned narenas_tdata_old, i;
493 unsigned narenas_tdata = tsd_narenas_tdata_get(tsd);
494 unsigned narenas_actual = narenas_total_get();
495
496 /*
497 * Dissociate old tdata array (and set up for deallocation upon return)
498 * if it's too small.
499 */
500 if (arenas_tdata != NULL && narenas_tdata < narenas_actual) {
501 arenas_tdata_old = arenas_tdata;
502 narenas_tdata_old = narenas_tdata;
503 arenas_tdata = NULL;
504 narenas_tdata = 0;
505 tsd_arenas_tdata_set(tsd, arenas_tdata);
506 tsd_narenas_tdata_set(tsd, narenas_tdata);
507 } else {
508 arenas_tdata_old = NULL;
509 narenas_tdata_old = 0;
510 }
511
512 /* Allocate tdata array if it's missing. */
513 if (arenas_tdata == NULL) {
514 bool *arenas_tdata_bypassp = tsd_arenas_tdata_bypassp_get(tsd);
515 narenas_tdata = (ind < narenas_actual) ? narenas_actual : ind+1;
516
517 if (tsd_nominal(tsd) && !*arenas_tdata_bypassp) {
518 *arenas_tdata_bypassp = true;
519 arenas_tdata = (arena_tdata_t *)a0malloc(
520 sizeof(arena_tdata_t) * narenas_tdata);
521 *arenas_tdata_bypassp = false;
522 }
523 if (arenas_tdata == NULL) {
524 tdata = NULL;
525 goto label_return;
526 }
527 assert(tsd_nominal(tsd) && !*arenas_tdata_bypassp);
528 tsd_arenas_tdata_set(tsd, arenas_tdata);
529 tsd_narenas_tdata_set(tsd, narenas_tdata);
530 }
531
532 /*
533 * Copy to tdata array. It's possible that the actual number of arenas
534 * has increased since narenas_total_get() was called above, but that
535 * causes no correctness issues unless two threads concurrently execute
536 * the arenas.extend mallctl, which we trust mallctl synchronization to
537 * prevent.
538 */
539
540 /* Copy/initialize tickers. */
541 for (i = 0; i < narenas_actual; i++) {
542 if (i < narenas_tdata_old) {
543 ticker_copy(&arenas_tdata[i].decay_ticker,
544 &arenas_tdata_old[i].decay_ticker);
545 } else {
546 ticker_init(&arenas_tdata[i].decay_ticker,
547 DECAY_NTICKS_PER_UPDATE);
548 }
549 }
550 if (narenas_tdata > narenas_actual) {
551 memset(&arenas_tdata[narenas_actual], 0, sizeof(arena_tdata_t)
552 * (narenas_tdata - narenas_actual));
553 }
554
555 /* Read the refreshed tdata array. */
556 tdata = &arenas_tdata[ind];
557 label_return:
558 if (arenas_tdata_old != NULL)
559 a0dalloc(arenas_tdata_old);
560 return (tdata);
561 }
562
563 /* Slow path, called only by arena_choose(). */
564 arena_t *
565 arena_choose_hard(tsd_t *tsd)
566 {
567 arena_t *ret;
568
569 if (narenas_auto > 1) {
570 unsigned i, choose, first_null;
571
572 choose = 0;
573 first_null = narenas_auto;
574 malloc_mutex_lock(&arenas_lock);
575 assert(arena_get(0, false) != NULL);
576 for (i = 1; i < narenas_auto; i++) {
577 if (arena_get(i, false) != NULL) {
578 /*
579 * Choose the first arena that has the lowest
580 * number of threads assigned to it.
581 */
582 if (arena_nthreads_get(arena_get(i, false)) <
583 arena_nthreads_get(arena_get(choose,
584 false)))
585 choose = i;
586 } else if (first_null == narenas_auto) {
587 /*
588 * Record the index of the first uninitialized
589 * arena, in case all extant arenas are in use.
590 *
591 * NB: It is possible for there to be
592 * discontinuities in terms of initialized
593 * versus uninitialized arenas, due to the
594 * "thread.arena" mallctl.
595 */
596 first_null = i;
597 }
598 }
599
600 if (arena_nthreads_get(arena_get(choose, false)) == 0
601 || first_null == narenas_auto) {
602 /*
603 * Use an unloaded arena, or the least loaded arena if
604 * all arenas are already initialized.
605 */
606 ret = arena_get(choose, false);
607 } else {
608 /* Initialize a new arena. */
609 choose = first_null;
610 ret = arena_init_locked(choose);
611 if (ret == NULL) {
612 malloc_mutex_unlock(&arenas_lock);
613 return (NULL);
614 }
615 }
616 arena_bind(tsd, choose);
617 malloc_mutex_unlock(&arenas_lock);
618 } else {
619 ret = arena_get(0, false);
620 arena_bind(tsd, 0);
621 }
622
623 return (ret);
624 }
625
626 void
627 thread_allocated_cleanup(tsd_t *tsd)
628 {
629
630 /* Do nothing. */
631 }
632
633 void
634 thread_deallocated_cleanup(tsd_t *tsd)
635 {
636
637 /* Do nothing. */
638 }
639
640 void
641 arena_cleanup(tsd_t *tsd)
642 {
643 arena_t *arena;
644
645 arena = tsd_arena_get(tsd);
646 if (arena != NULL)
647 arena_unbind(tsd, arena->ind);
648 }
649
650 void
651 arenas_tdata_cleanup(tsd_t *tsd)
652 {
653 arena_tdata_t *arenas_tdata;
654
655 /* Prevent tsd->arenas_tdata from being (re)created. */
656 *tsd_arenas_tdata_bypassp_get(tsd) = true;
657
658 arenas_tdata = tsd_arenas_tdata_get(tsd);
659 if (arenas_tdata != NULL) {
660 tsd_arenas_tdata_set(tsd, NULL);
661 a0dalloc(arenas_tdata);
662 }
663 }
664
665 void
666 narenas_tdata_cleanup(tsd_t *tsd)
667 {
668
669 /* Do nothing. */
670 }
671
672 void
673 arenas_tdata_bypass_cleanup(tsd_t *tsd)
674 {
675
676 /* Do nothing. */
677 }
678
679 static void
680 stats_print_atexit(void)
681 {
682
683 if (config_tcache && config_stats) {
684 unsigned narenas, i;
685
686 /*
687 * Merge stats from extant threads. This is racy, since
688 * individual threads do not lock when recording tcache stats
689 * events. As a consequence, the final stats may be slightly
690 * out of date by the time they are reported, if other threads
691 * continue to allocate.
692 */
693 for (i = 0, narenas = narenas_total_get(); i < narenas; i++) {
694 arena_t *arena = arena_get(i, false);
695 if (arena != NULL) {
696 tcache_t *tcache;
697
698 /*
699 * tcache_stats_merge() locks bins, so if any
700 * code is introduced that acquires both arena
701 * and bin locks in the opposite order,
702 * deadlocks may result.
703 */
704 malloc_mutex_lock(&arena->lock);
705 ql_foreach(tcache, &arena->tcache_ql, link) {
706 tcache_stats_merge(tcache, arena);
707 }
708 malloc_mutex_unlock(&arena->lock);
709 }
710 }
711 }
712 je_malloc_stats_print(NULL, NULL, NULL);
713 }
714
715 /*
716 * End miscellaneous support functions.
717 */
718 /******************************************************************************/
719 /*
720 * Begin initialization functions.
721 */
722
723 #ifndef JEMALLOC_HAVE_SECURE_GETENV
724 static char *
725 secure_getenv(const char *name)
726 {
727
728 # ifdef JEMALLOC_HAVE_ISSETUGID
729 if (issetugid() != 0)
730 return (NULL);
731 # endif
732 return (getenv(name));
733 }
734 #endif
735
736 static unsigned
737 malloc_ncpus(void)
738 {
739 long result;
740
741 #ifdef _WIN32
742 SYSTEM_INFO si;
743 GetSystemInfo(&si);
744 result = si.dwNumberOfProcessors;
745 #else
746 result = sysconf(_SC_NPROCESSORS_ONLN);
747 #endif
748 return ((result == -1) ? 1 : (unsigned)result);
749 }
750
751 static bool
752 malloc_conf_next(char const **opts_p, char const **k_p, size_t *klen_p,
753 char const **v_p, size_t *vlen_p)
754 {
755 bool accept;
756 const char *opts = *opts_p;
757
758 *k_p = opts;
759
760 for (accept = false; !accept;) {
761 switch (*opts) {
762 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
763 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
764 case 'M': case 'N': case 'O': case 'P': case 'Q': case 'R':
765 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
766 case 'Y': case 'Z':
767 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
768 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
769 case 'm': case 'n': case 'o': case 'p': case 'q': case 'r':
770 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
771 case 'y': case 'z':
772 case '0': case '1': case '2': case '3': case '4': case '5':
773 case '6': case '7': case '8': case '9':
774 case '_':
775 opts++;
776 break;
777 case ':':
778 opts++;
779 *klen_p = (uintptr_t)opts - 1 - (uintptr_t)*k_p;
780 *v_p = opts;
781 accept = true;
782 break;
783 case '\0':
784 if (opts != *opts_p) {
785 malloc_write("<jemalloc>: Conf string ends "
786 "with key\n");
787 }
788 return (true);
789 default:
790 malloc_write("<jemalloc>: Malformed conf string\n");
791 return (true);
792 }
793 }
794
795 for (accept = false; !accept;) {
796 switch (*opts) {
797 case ',':
798 opts++;
799 /*
800 * Look ahead one character here, because the next time
801 * this function is called, it will assume that end of
802 * input has been cleanly reached if no input remains,
803 * but we have optimistically already consumed the
804 * comma if one exists.
805 */
806 if (*opts == '\0') {
807 malloc_write("<jemalloc>: Conf string ends "
808 "with comma\n");
809 }
810 *vlen_p = (uintptr_t)opts - 1 - (uintptr_t)*v_p;
811 accept = true;
812 break;
813 case '\0':
814 *vlen_p = (uintptr_t)opts - (uintptr_t)*v_p;
815 accept = true;
816 break;
817 default:
818 opts++;
819 break;
820 }
821 }
822
823 *opts_p = opts;
824 return (false);
825 }
826
827 static void
828 malloc_conf_error(const char *msg, const char *k, size_t klen, const char *v,
829 size_t vlen)
830 {
831
832 malloc_printf("<jemalloc>: %s: %.*s:%.*s\n", msg, (int)klen, k,
833 (int)vlen, v);
834 }
835
836 static void
837 malloc_slow_flag_init(void)
838 {
839 /*
840 * Combine the runtime options into malloc_slow for fast path. Called
841 * after processing all the options.
842 */
843 malloc_slow_flags |= (opt_junk_alloc ? flag_opt_junk_alloc : 0)
844 | (opt_junk_free ? flag_opt_junk_free : 0)
845 | (opt_quarantine ? flag_opt_quarantine : 0)
846 | (opt_zero ? flag_opt_zero : 0)
847 | (opt_utrace ? flag_opt_utrace : 0)
848 | (opt_xmalloc ? flag_opt_xmalloc : 0);
849
850 if (config_valgrind)
851 malloc_slow_flags |= (in_valgrind ? flag_in_valgrind : 0);
852
853 malloc_slow = (malloc_slow_flags != 0);
854 }
855
856 static void
857 malloc_conf_init(void)
858 {
859 unsigned i;
860 char buf[PATH_MAX + 1];
861 const char *opts, *k, *v;
862 size_t klen, vlen;
863
864 /*
865 * Automatically configure valgrind before processing options. The
866 * valgrind option remains in jemalloc 3.x for compatibility reasons.
867 */
868 if (config_valgrind) {
869 in_valgrind = (RUNNING_ON_VALGRIND != 0) ? true : false;
870 if (config_fill && unlikely(in_valgrind)) {
871 opt_junk = "false";
872 opt_junk_alloc = false;
873 opt_junk_free = false;
874 assert(!opt_zero);
875 opt_quarantine = JEMALLOC_VALGRIND_QUARANTINE_DEFAULT;
876 opt_redzone = true;
877 }
878 if (config_tcache && unlikely(in_valgrind))
879 opt_tcache = false;
880 }
881
882 for (i = 0; i < 4; i++) {
883 /* Get runtime configuration. */
884 switch (i) {
885 case 0:
886 opts = config_malloc_conf;
887 break;
888 case 1:
889 if (je_malloc_conf != NULL) {
890 /*
891 * Use options that were compiled into the
892 * program.
893 */
894 opts = je_malloc_conf;
895 } else {
896 /* No configuration specified. */
897 buf[0] = '\0';
898 opts = buf;
899 }
900 break;
901 case 2: {
902 ssize_t linklen = 0;
903 #ifndef _WIN32
904 int saved_errno = errno;
905 const char *linkname =
906 # ifdef JEMALLOC_PREFIX
907 "/etc/"JEMALLOC_PREFIX"malloc.conf"
908 # else
909 "/etc/malloc.conf"
910 # endif
911 ;
912
913 /*
914 * Try to use the contents of the "/etc/malloc.conf"
915 * symbolic link's name.
916 */
917 linklen = readlink(linkname, buf, sizeof(buf) - 1);
918 if (linklen == -1) {
919 /* No configuration specified. */
920 linklen = 0;
921 /* Restore errno. */
922 set_errno(saved_errno);
923 }
924 #endif
925 buf[linklen] = '\0';
926 opts = buf;
927 break;
928 } case 3: {
929 const char *envname =
930 #ifdef JEMALLOC_PREFIX
931 JEMALLOC_CPREFIX"MALLOC_CONF"
932 #else
933 "MALLOC_CONF"
934 #endif
935 ;
936
937 if ((opts = secure_getenv(envname)) != NULL) {
938 /*
939 * Do nothing; opts is already initialized to
940 * the value of the MALLOC_CONF environment
941 * variable.
942 */
943 } else {
944 /* No configuration specified. */
945 buf[0] = '\0';
946 opts = buf;
947 }
948 break;
949 } default:
950 not_reached();
951 buf[0] = '\0';
952 opts = buf;
953 }
954
955 while (*opts != '\0' && !malloc_conf_next(&opts, &k, &klen, &v,
956 &vlen)) {
957 #define CONF_MATCH(n) \
958 (sizeof(n)-1 == klen && strncmp(n, k, klen) == 0)
959 #define CONF_MATCH_VALUE(n) \
960 (sizeof(n)-1 == vlen && strncmp(n, v, vlen) == 0)
961 #define CONF_HANDLE_BOOL(o, n, cont) \
962 if (CONF_MATCH(n)) { \
963 if (CONF_MATCH_VALUE("true")) \
964 o = true; \
965 else if (CONF_MATCH_VALUE("false")) \
966 o = false; \
967 else { \
968 malloc_conf_error( \
969 "Invalid conf value", \
970 k, klen, v, vlen); \
971 } \
972 if (cont) \
973 continue; \
974 }
975 #define CONF_HANDLE_T_U(t, o, n, min, max, clip) \
976 if (CONF_MATCH(n)) { \
977 uintmax_t um; \
978 char *end; \
979 \
980 set_errno(0); \
981 um = malloc_strtoumax(v, &end, 0); \
982 if (get_errno() != 0 || (uintptr_t)end -\
983 (uintptr_t)v != vlen) { \
984 malloc_conf_error( \
985 "Invalid conf value", \
986 k, klen, v, vlen); \
987 } else if (clip) { \
988 if ((min) != 0 && um < (min)) \
989 o = (t)(min); \
990 else if (um > (max)) \
991 o = (t)(max); \
992 else \
993 o = (t)um; \
994 } else { \
995 if (((min) != 0 && um < (min)) \
996 || um > (max)) { \
997 malloc_conf_error( \
998 "Out-of-range " \
999 "conf value", \
1000 k, klen, v, vlen); \
1001 } else \
1002 o = (t)um; \
1003 } \
1004 continue; \
1005 }
1006 #define CONF_HANDLE_UNSIGNED(o, n, min, max, clip) \
1007 CONF_HANDLE_T_U(unsigned, o, n, min, max, clip)
1008 #define CONF_HANDLE_SIZE_T(o, n, min, max, clip) \
1009 CONF_HANDLE_T_U(size_t, o, n, min, max, clip)
1010 #define CONF_HANDLE_SSIZE_T(o, n, min, max) \
1011 if (CONF_MATCH(n)) { \
1012 long l; \
1013 char *end; \
1014 \
1015 set_errno(0); \
1016 l = strtol(v, &end, 0); \
1017 if (get_errno() != 0 || (uintptr_t)end -\
1018 (uintptr_t)v != vlen) { \
1019 malloc_conf_error( \
1020 "Invalid conf value", \
1021 k, klen, v, vlen); \
1022 } else if (l < (ssize_t)(min) || l > \
1023 (ssize_t)(max)) { \
1024 malloc_conf_error( \
1025 "Out-of-range conf value", \
1026 k, klen, v, vlen); \
1027 } else \
1028 o = l; \
1029 continue; \
1030 }
1031 #define CONF_HANDLE_CHAR_P(o, n, d) \
1032 if (CONF_MATCH(n)) { \
1033 size_t cpylen = (vlen <= \
1034 sizeof(o)-1) ? vlen : \
1035 sizeof(o)-1; \
1036 strncpy(o, v, cpylen); \
1037 o[cpylen] = '\0'; \
1038 continue; \
1039 }
1040
1041 CONF_HANDLE_BOOL(opt_abort, "abort", true)
1042 /*
1043 * Chunks always require at least one header page,
1044 * as many as 2^(LG_SIZE_CLASS_GROUP+1) data pages, and
1045 * possibly an additional page in the presence of
1046 * redzones. In order to simplify options processing,
1047 * use a conservative bound that accommodates all these
1048 * constraints.
1049 */
1050 CONF_HANDLE_SIZE_T(opt_lg_chunk, "lg_chunk", LG_PAGE +
1051 LG_SIZE_CLASS_GROUP + (config_fill ? 2 : 1),
1052 (sizeof(size_t) << 3) - 1, true)
1053 if (strncmp("dss", k, klen) == 0) {
1054 int i;
1055 bool match = false;
1056 for (i = 0; i < dss_prec_limit; i++) {
1057 if (strncmp(dss_prec_names[i], v, vlen)
1058 == 0) {
1059 if (chunk_dss_prec_set(i)) {
1060 malloc_conf_error(
1061 "Error setting dss",
1062 k, klen, v, vlen);
1063 } else {
1064 opt_dss =
1065 dss_prec_names[i];
1066 match = true;
1067 break;
1068 }
1069 }
1070 }
1071 if (!match) {
1072 malloc_conf_error("Invalid conf value",
1073 k, klen, v, vlen);
1074 }
1075 continue;
1076 }
1077 CONF_HANDLE_UNSIGNED(opt_narenas, "narenas", 1,
1078 UINT_MAX, false)
1079 if (strncmp("purge", k, klen) == 0) {
1080 int i;
1081 bool match = false;
1082 for (i = 0; i < purge_mode_limit; i++) {
1083 if (strncmp(purge_mode_names[i], v,
1084 vlen) == 0) {
1085 opt_purge = (purge_mode_t)i;
1086 match = true;
1087 break;
1088 }
1089 }
1090 if (!match) {
1091 malloc_conf_error("Invalid conf value",
1092 k, klen, v, vlen);
1093 }
1094 continue;
1095 }
1096 CONF_HANDLE_SSIZE_T(opt_lg_dirty_mult, "lg_dirty_mult",
1097 -1, (sizeof(size_t) << 3) - 1)
1098 CONF_HANDLE_SSIZE_T(opt_decay_time, "decay_time", -1,
1099 NSTIME_SEC_MAX);
1100 CONF_HANDLE_BOOL(opt_stats_print, "stats_print", true)
1101 if (config_fill) {
1102 if (CONF_MATCH("junk")) {
1103 if (CONF_MATCH_VALUE("true")) {
1104 opt_junk = "true";
1105 opt_junk_alloc = opt_junk_free =
1106 true;
1107 } else if (CONF_MATCH_VALUE("false")) {
1108 opt_junk = "false";
1109 opt_junk_alloc = opt_junk_free =
1110 false;
1111 } else if (CONF_MATCH_VALUE("alloc")) {
1112 opt_junk = "alloc";
1113 opt_junk_alloc = true;
1114 opt_junk_free = false;
1115 } else if (CONF_MATCH_VALUE("free")) {
1116 opt_junk = "free";
1117 opt_junk_alloc = false;
1118 opt_junk_free = true;
1119 } else {
1120 malloc_conf_error(
1121 "Invalid conf value", k,
1122 klen, v, vlen);
1123 }
1124 continue;
1125 }
1126 CONF_HANDLE_SIZE_T(opt_quarantine, "quarantine",
1127 0, SIZE_T_MAX, false)
1128 CONF_HANDLE_BOOL(opt_redzone, "redzone", true)
1129 CONF_HANDLE_BOOL(opt_zero, "zero", true)
1130 }
1131 if (config_utrace) {
1132 CONF_HANDLE_BOOL(opt_utrace, "utrace", true)
1133 }
1134 if (config_xmalloc) {
1135 CONF_HANDLE_BOOL(opt_xmalloc, "xmalloc", true)
1136 }
1137 if (config_tcache) {
1138 CONF_HANDLE_BOOL(opt_tcache, "tcache",
1139 !config_valgrind || !in_valgrind)
1140 if (CONF_MATCH("tcache")) {
1141 assert(config_valgrind && in_valgrind);
1142 if (opt_tcache) {
1143 opt_tcache = false;
1144 malloc_conf_error(
1145 "tcache cannot be enabled "
1146 "while running inside Valgrind",
1147 k, klen, v, vlen);
1148 }
1149 continue;
1150 }
1151 CONF_HANDLE_SSIZE_T(opt_lg_tcache_max,
1152 "lg_tcache_max", -1,
1153 (sizeof(size_t) << 3) - 1)
1154 }
1155 if (config_prof) {
1156 CONF_HANDLE_BOOL(opt_prof, "prof", true)
1157 CONF_HANDLE_CHAR_P(opt_prof_prefix,
1158 "prof_prefix", "jeprof")
1159 CONF_HANDLE_BOOL(opt_prof_active, "prof_active",
1160 true)
1161 CONF_HANDLE_BOOL(opt_prof_thread_active_init,
1162 "prof_thread_active_init", true)
1163 CONF_HANDLE_SIZE_T(opt_lg_prof_sample,
1164 "lg_prof_sample", 0,
1165 (sizeof(uint64_t) << 3) - 1, true)
1166 CONF_HANDLE_BOOL(opt_prof_accum, "prof_accum",
1167 true)
1168 CONF_HANDLE_SSIZE_T(opt_lg_prof_interval,
1169 "lg_prof_interval", -1,
1170 (sizeof(uint64_t) << 3) - 1)
1171 CONF_HANDLE_BOOL(opt_prof_gdump, "prof_gdump",
1172 true)
1173 CONF_HANDLE_BOOL(opt_prof_final, "prof_final",
1174 true)
1175 CONF_HANDLE_BOOL(opt_prof_leak, "prof_leak",
1176 true)
1177 }
1178 malloc_conf_error("Invalid conf pair", k, klen, v,
1179 vlen);
1180 #undef CONF_MATCH
1181 #undef CONF_HANDLE_BOOL
1182 #undef CONF_HANDLE_SIZE_T
1183 #undef CONF_HANDLE_SSIZE_T
1184 #undef CONF_HANDLE_CHAR_P
1185 }
1186 }
1187 }
1188
1189 /* init_lock must be held. */
1190 static bool
1191 malloc_init_hard_needed(void)
1192 {
1193
1194 if (malloc_initialized() || (IS_INITIALIZER && malloc_init_state ==
1195 malloc_init_recursible)) {
1196 /*
1197 * Another thread initialized the allocator before this one
1198 * acquired init_lock, or this thread is the initializing
1199 * thread, and it is recursively allocating.
1200 */
1201 return (false);
1202 }
1203 #ifdef JEMALLOC_THREADED_INIT
1204 if (malloc_initializer != NO_INITIALIZER && !IS_INITIALIZER) {
1205 /* Busy-wait until the initializing thread completes. */
1206 do {
1207 malloc_mutex_unlock(&init_lock);
1208 CPU_SPINWAIT;
1209 malloc_mutex_lock(&init_lock);
1210 } while (!malloc_initialized());
1211 return (false);
1212 }
1213 #endif
1214 return (true);
1215 }
1216
1217 /* init_lock must be held. */
1218 static bool
1219 malloc_init_hard_a0_locked(void)
1220 {
1221
1222 malloc_initializer = INITIALIZER;
1223
1224 if (config_prof)
1225 prof_boot0();
1226 malloc_conf_init();
1227 if (opt_stats_print) {
1228 /* Print statistics at exit. */
1229 if (atexit(stats_print_atexit) != 0) {
1230 malloc_write("<jemalloc>: Error in atexit()\n");
1231 if (opt_abort)
1232 abort();
1233 }
1234 }
1235 if (base_boot())
1236 return (true);
1237 if (chunk_boot())
1238 return (true);
1239 if (ctl_boot())
1240 return (true);
1241 if (config_prof)
1242 prof_boot1();
1243 if (arena_boot())
1244 return (true);
1245 if (config_tcache && tcache_boot())
1246 return (true);
1247 if (malloc_mutex_init(&arenas_lock))
1248 return (true);
1249 /*
1250 * Create enough scaffolding to allow recursive allocation in
1251 * malloc_ncpus().
1252 */
1253 narenas_auto = 1;
1254 narenas_total_set(narenas_auto);
1255 arenas = &a0;
1256 memset(arenas, 0, sizeof(arena_t *) * narenas_auto);
1257 /*
1258 * Initialize one arena here. The rest are lazily created in
1259 * arena_choose_hard().
1260 */
1261 if (arena_init(0) == NULL)
1262 return (true);
1263 malloc_init_state = malloc_init_a0_initialized;
1264 return (false);
1265 }
1266
1267 static bool
1268 malloc_init_hard_a0(void)
1269 {
1270 bool ret;
1271
1272 malloc_mutex_lock(&init_lock);
1273 ret = malloc_init_hard_a0_locked();
1274 malloc_mutex_unlock(&init_lock);
1275 return (ret);
1276 }
1277
1278 /*
1279 * Initialize data structures which may trigger recursive allocation.
1280 *
1281 * init_lock must be held.
1282 */
1283 static bool
1284 malloc_init_hard_recursible(void)
1285 {
1286 bool ret = false;
1287
1288 malloc_init_state = malloc_init_recursible;
1289 malloc_mutex_unlock(&init_lock);
1290
1291 /* LinuxThreads' pthread_setspecific() allocates. */
1292 if (malloc_tsd_boot0()) {
1293 ret = true;
1294 goto label_return;
1295 }
1296
1297 ncpus = malloc_ncpus();
1298
1299 #if (!defined(JEMALLOC_MUTEX_INIT_CB) && !defined(JEMALLOC_ZONE) \
1300 && !defined(_WIN32) && !defined(__native_client__))
1301 /* LinuxThreads' pthread_atfork() allocates. */
1302 if (pthread_atfork(jemalloc_prefork, jemalloc_postfork_parent,
1303 jemalloc_postfork_child) != 0) {
1304 ret = true;
1305 malloc_write("<jemalloc>: Error in pthread_atfork()\n");
1306 if (opt_abort)
1307 abort();
1308 }
1309 #endif
1310
1311 label_return:
1312 malloc_mutex_lock(&init_lock);
1313 return (ret);
1314 }
1315
1316 /* init_lock must be held. */
1317 static bool
1318 malloc_init_hard_finish(void)
1319 {
1320
1321 if (mutex_boot())
1322 return (true);
1323
1324 if (opt_narenas == 0) {
1325 /*
1326 * For SMP systems, create more than one arena per CPU by
1327 * default.
1328 */
1329 if (ncpus > 1)
1330 opt_narenas = ncpus << 2;
1331 else
1332 opt_narenas = 1;
1333 }
1334 narenas_auto = opt_narenas;
1335 /*
1336 * Limit the number of arenas to the indexing range of MALLOCX_ARENA().
1337 */
1338 if (narenas_auto > MALLOCX_ARENA_MAX) {
1339 narenas_auto = MALLOCX_ARENA_MAX;
1340 malloc_printf("<jemalloc>: Reducing narenas to limit (%d)\n",
1341 narenas_auto);
1342 }
1343 narenas_total_set(narenas_auto);
1344
1345 /* Allocate and initialize arenas. */
1346 arenas = (arena_t **)base_alloc(sizeof(arena_t *) *
1347 (MALLOCX_ARENA_MAX+1));
1348 if (arenas == NULL)
1349 return (true);
1350 /* Copy the pointer to the one arena that was already initialized. */
1351 arena_set(0, a0);
1352
1353 malloc_init_state = malloc_init_initialized;
1354 malloc_slow_flag_init();
1355
1356 return (false);
1357 }
1358
1359 static bool
1360 malloc_init_hard(void)
1361 {
1362
1363 #if defined(_WIN32) && _WIN32_WINNT < 0x0600
1364 _init_init_lock();
1365 #endif
1366 malloc_mutex_lock(&init_lock);
1367 if (!malloc_init_hard_needed()) {
1368 malloc_mutex_unlock(&init_lock);
1369 return (false);
1370 }
1371
1372 if (malloc_init_state != malloc_init_a0_initialized &&
1373 malloc_init_hard_a0_locked()) {
1374 malloc_mutex_unlock(&init_lock);
1375 return (true);
1376 }
1377
1378 if (malloc_init_hard_recursible()) {
1379 malloc_mutex_unlock(&init_lock);
1380 return (true);
1381 }
1382
1383 if (config_prof && prof_boot2()) {
1384 malloc_mutex_unlock(&init_lock);
1385 return (true);
1386 }
1387
1388 if (malloc_init_hard_finish()) {
1389 malloc_mutex_unlock(&init_lock);
1390 return (true);
1391 }
1392
1393 malloc_mutex_unlock(&init_lock);
1394 malloc_tsd_boot1();
1395 return (false);
1396 }
1397
1398 /*
1399 * End initialization functions.
1400 */
1401 /******************************************************************************/
1402 /*
1403 * Begin malloc(3)-compatible functions.
1404 */
1405
1406 static void *
1407 imalloc_prof_sample(tsd_t *tsd, size_t usize, szind_t ind,
1408 prof_tctx_t *tctx, bool slow_path)
1409 {
1410 void *p;
1411
1412 if (tctx == NULL)
1413 return (NULL);
1414 if (usize <= SMALL_MAXCLASS) {
1415 szind_t ind_large = size2index(LARGE_MINCLASS);
1416 p = imalloc(tsd, LARGE_MINCLASS, ind_large, slow_path);
1417 if (p == NULL)
1418 return (NULL);
1419 arena_prof_promoted(p, usize);
1420 } else
1421 p = imalloc(tsd, usize, ind, slow_path);
1422
1423 return (p);
1424 }
1425
1426 JEMALLOC_ALWAYS_INLINE_C void *
1427 imalloc_prof(tsd_t *tsd, size_t usize, szind_t ind, bool slow_path)
1428 {
1429 void *p;
1430 prof_tctx_t *tctx;
1431
1432 tctx = prof_alloc_prep(tsd, usize, prof_active_get_unlocked(), true);
1433 if (unlikely((uintptr_t)tctx != (uintptr_t)1U))
1434 p = imalloc_prof_sample(tsd, usize, ind, tctx, slow_path);
1435 else
1436 p = imalloc(tsd, usize, ind, slow_path);
1437 if (unlikely(p == NULL)) {
1438 prof_alloc_rollback(tsd, tctx, true);
1439 return (NULL);
1440 }
1441 prof_malloc(p, usize, tctx);
1442
1443 return (p);
1444 }
1445
1446 JEMALLOC_ALWAYS_INLINE_C void *
1447 imalloc_body(size_t size, tsd_t **tsd, size_t *usize, bool slow_path)
1448 {
1449 szind_t ind;
1450
1451 if (slow_path && unlikely(malloc_init()))
1452 return (NULL);
1453 *tsd = tsd_fetch();
1454 ind = size2index(size);
1455 if (unlikely(ind >= NSIZES))
1456 return (NULL);
1457
1458 if (config_stats || (config_prof && opt_prof) || (slow_path &&
1459 config_valgrind && unlikely(in_valgrind))) {
1460 *usize = index2size(ind);
1461 assert(*usize > 0 && *usize <= HUGE_MAXCLASS);
1462 }
1463
1464 if (config_prof && opt_prof)
1465 return (imalloc_prof(*tsd, *usize, ind, slow_path));
1466
1467 return (imalloc(*tsd, size, ind, slow_path));
1468 }
1469
1470 JEMALLOC_ALWAYS_INLINE_C void
1471 imalloc_post_check(void *ret, tsd_t *tsd, size_t usize, bool slow_path)
1472 {
1473 if (unlikely(ret == NULL)) {
1474 if (slow_path && config_xmalloc && unlikely(opt_xmalloc)) {
1475 malloc_write("<jemalloc>: Error in malloc(): "
1476 "out of memory\n");
1477 abort();
1478 }
1479 set_errno(ENOMEM);
1480 }
1481 if (config_stats && likely(ret != NULL)) {
1482 assert(usize == isalloc(ret, config_prof));
1483 *tsd_thread_allocatedp_get(tsd) += usize;
1484 }
1485 }
1486
1487 JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN
1488 void JEMALLOC_NOTHROW *
1489 JEMALLOC_ATTR(malloc) JEMALLOC_ALLOC_SIZE(1)
1490 je_malloc(size_t size)
1491 {
1492 void *ret;
1493 tsd_t *tsd;
1494 size_t usize JEMALLOC_CC_SILENCE_INIT(0);
1495
1496 if (size == 0)
1497 size = 1;
1498
1499 if (likely(!malloc_slow)) {
1500 /*
1501 * imalloc_body() is inlined so that fast and slow paths are
1502 * generated separately with statically known slow_path.
1503 */
1504 ret = imalloc_body(size, &tsd, &usize, false);
1505 imalloc_post_check(ret, tsd, usize, false);
1506 } else {
1507 ret = imalloc_body(size, &tsd, &usize, true);
1508 imalloc_post_check(ret, tsd, usize, true);
1509 UTRACE(0, size, ret);
1510 JEMALLOC_VALGRIND_MALLOC(ret != NULL, ret, usize, false);
1511 }
1512
1513 return (ret);
1514 }
1515
1516 static void *
1517 imemalign_prof_sample(tsd_t *tsd, size_t alignment, size_t usize,
1518 prof_tctx_t *tctx)
1519 {
1520 void *p;
1521
1522 if (tctx == NULL)
1523 return (NULL);
1524 if (usize <= SMALL_MAXCLASS) {
1525 assert(sa2u(LARGE_MINCLASS, alignment) == LARGE_MINCLASS);
1526 p = ipalloc(tsd, LARGE_MINCLASS, alignment, false);
1527 if (p == NULL)
1528 return (NULL);
1529 arena_prof_promoted(p, usize);
1530 } else
1531 p = ipalloc(tsd, usize, alignment, false);
1532
1533 return (p);
1534 }
1535
1536 JEMALLOC_ALWAYS_INLINE_C void *
1537 imemalign_prof(tsd_t *tsd, size_t alignment, size_t usize)
1538 {
1539 void *p;
1540 prof_tctx_t *tctx;
1541
1542 tctx = prof_alloc_prep(tsd, usize, prof_active_get_unlocked(), true);
1543 if (unlikely((uintptr_t)tctx != (uintptr_t)1U))
1544 p = imemalign_prof_sample(tsd, alignment, usize, tctx);
1545 else
1546 p = ipalloc(tsd, usize, alignment, false);
1547 if (unlikely(p == NULL)) {
1548 prof_alloc_rollback(tsd, tctx, true);
1549 return (NULL);
1550 }
1551 prof_malloc(p, usize, tctx);
1552
1553 return (p);
1554 }
1555
1556 JEMALLOC_ATTR(nonnull(1))
1557 static int
1558 imemalign(void **memptr, size_t alignment, size_t size, size_t min_alignment)
1559 {
1560 int ret;
1561 tsd_t *tsd;
1562 size_t usize;
1563 void *result;
1564
1565 assert(min_alignment != 0);
1566
1567 if (unlikely(malloc_init())) {
1568 result = NULL;
1569 goto label_oom;
1570 }
1571 tsd = tsd_fetch();
1572 if (size == 0)
1573 size = 1;
1574
1575 /* Make sure that alignment is a large enough power of 2. */
1576 if (unlikely(((alignment - 1) & alignment) != 0
1577 || (alignment < min_alignment))) {
1578 if (config_xmalloc && unlikely(opt_xmalloc)) {
1579 malloc_write("<jemalloc>: Error allocating "
1580 "aligned memory: invalid alignment\n");
1581 abort();
1582 }
1583 result = NULL;
1584 ret = EINVAL;
1585 goto label_return;
1586 }
1587
1588 usize = sa2u(size, alignment);
1589 if (unlikely(usize == 0 || usize > HUGE_MAXCLASS)) {
1590 result = NULL;
1591 goto label_oom;
1592 }
1593
1594 if (config_prof && opt_prof)
1595 result = imemalign_prof(tsd, alignment, usize);
1596 else
1597 result = ipalloc(tsd, usize, alignment, false);
1598 if (unlikely(result == NULL))
1599 goto label_oom;
1600 assert(((uintptr_t)result & (alignment - 1)) == ZU(0));
1601
1602 *memptr = result;
1603 ret = 0;
1604 label_return:
1605 if (config_stats && likely(result != NULL)) {
1606 assert(usize == isalloc(result, config_prof));
1607 *tsd_thread_allocatedp_get(tsd) += usize;
1608 }
1609 UTRACE(0, size, result);
1610 return (ret);
1611 label_oom:
1612 assert(result == NULL);
1613 if (config_xmalloc && unlikely(opt_xmalloc)) {
1614 malloc_write("<jemalloc>: Error allocating aligned memory: "
1615 "out of memory\n");
1616 abort();
1617 }
1618 ret = ENOMEM;
1619 goto label_return;
1620 }
1621
1622 JEMALLOC_EXPORT int JEMALLOC_NOTHROW
1623 JEMALLOC_ATTR(nonnull(1))
1624 je_posix_memalign(void **memptr, size_t alignment, size_t size)
1625 {
1626 int ret = imemalign(memptr, alignment, size, sizeof(void *));
1627 JEMALLOC_VALGRIND_MALLOC(ret == 0, *memptr, isalloc(*memptr,
1628 config_prof), false);
1629 return (ret);
1630 }
1631
1632 JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN
1633 void JEMALLOC_NOTHROW *
1634 JEMALLOC_ATTR(malloc) JEMALLOC_ALLOC_SIZE(2)
1635 je_aligned_alloc(size_t alignment, size_t size)
1636 {
1637 void *ret;
1638 int err;
1639
1640 if (unlikely((err = imemalign(&ret, alignment, size, 1)) != 0)) {
1641 ret = NULL;
1642 set_errno(err);
1643 }
1644 JEMALLOC_VALGRIND_MALLOC(err == 0, ret, isalloc(ret, config_prof),
1645 false);
1646 return (ret);
1647 }
1648
1649 static void *
1650 icalloc_prof_sample(tsd_t *tsd, size_t usize, szind_t ind, prof_tctx_t *tctx)
1651 {
1652 void *p;
1653
1654 if (tctx == NULL)
1655 return (NULL);
1656 if (usize <= SMALL_MAXCLASS) {
1657 szind_t ind_large = size2index(LARGE_MINCLASS);
1658 p = icalloc(tsd, LARGE_MINCLASS, ind_large);
1659 if (p == NULL)
1660 return (NULL);
1661 arena_prof_promoted(p, usize);
1662 } else
1663 p = icalloc(tsd, usize, ind);
1664
1665 return (p);
1666 }
1667
1668 JEMALLOC_ALWAYS_INLINE_C void *
1669 icalloc_prof(tsd_t *tsd, size_t usize, szind_t ind)
1670 {
1671 void *p;
1672 prof_tctx_t *tctx;
1673
1674 tctx = prof_alloc_prep(tsd, usize, prof_active_get_unlocked(), true);
1675 if (unlikely((uintptr_t)tctx != (uintptr_t)1U))
1676 p = icalloc_prof_sample(tsd, usize, ind, tctx);
1677 else
1678 p = icalloc(tsd, usize, ind);
1679 if (unlikely(p == NULL)) {
1680 prof_alloc_rollback(tsd, tctx, true);
1681 return (NULL);
1682 }
1683 prof_malloc(p, usize, tctx);
1684
1685 return (p);
1686 }
1687
1688 JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN
1689 void JEMALLOC_NOTHROW *
1690 JEMALLOC_ATTR(malloc) JEMALLOC_ALLOC_SIZE2(1, 2)
1691 je_calloc(size_t num, size_t size)
1692 {
1693 void *ret;
1694 tsd_t *tsd;
1695 size_t num_size;
1696 szind_t ind;
1697 size_t usize JEMALLOC_CC_SILENCE_INIT(0);
1698
1699 if (unlikely(malloc_init())) {
1700 num_size = 0;
1701 ret = NULL;
1702 goto label_return;
1703 }
1704 tsd = tsd_fetch();
1705
1706 num_size = num * size;
1707 if (unlikely(num_size == 0)) {
1708 if (num == 0 || size == 0)
1709 num_size = 1;
1710 else {
1711 ret = NULL;
1712 goto label_return;
1713 }
1714 /*
1715 * Try to avoid division here. We know that it isn't possible to
1716 * overflow during multiplication if neither operand uses any of the
1717 * most significant half of the bits in a size_t.
1718 */
1719 } else if (unlikely(((num | size) & (SIZE_T_MAX << (sizeof(size_t) <<
1720 2))) && (num_size / size != num))) {
1721 /* size_t overflow. */
1722 ret = NULL;
1723 goto label_return;
1724 }
1725
1726 ind = size2index(num_size);
1727 if (unlikely(ind >= NSIZES)) {
1728 ret = NULL;
1729 goto label_return;
1730 }
1731 if (config_prof && opt_prof) {
1732 usize = index2size(ind);
1733 ret = icalloc_prof(tsd, usize, ind);
1734 } else {
1735 if (config_stats || (config_valgrind && unlikely(in_valgrind)))
1736 usize = index2size(ind);
1737 ret = icalloc(tsd, num_size, ind);
1738 }
1739
1740 label_return:
1741 if (unlikely(ret == NULL)) {
1742 if (config_xmalloc && unlikely(opt_xmalloc)) {
1743 malloc_write("<jemalloc>: Error in calloc(): out of "
1744 "memory\n");
1745 abort();
1746 }
1747 set_errno(ENOMEM);
1748 }
1749 if (config_stats && likely(ret != NULL)) {
1750 assert(usize == isalloc(ret, config_prof));
1751 *tsd_thread_allocatedp_get(tsd) += usize;
1752 }
1753 UTRACE(0, num_size, ret);
1754 JEMALLOC_VALGRIND_MALLOC(ret != NULL, ret, usize, true);
1755 return (ret);
1756 }
1757
1758 static void *
1759 irealloc_prof_sample(tsd_t *tsd, void *old_ptr, size_t old_usize, size_t usize,
1760 prof_tctx_t *tctx)
1761 {
1762 void *p;
1763
1764 if (tctx == NULL)
1765 return (NULL);
1766 if (usize <= SMALL_MAXCLASS) {
1767 p = iralloc(tsd, old_ptr, old_usize, LARGE_MINCLASS, 0, false);
1768 if (p == NULL)
1769 return (NULL);
1770 arena_prof_promoted(p, usize);
1771 } else
1772 p = iralloc(tsd, old_ptr, old_usize, usize, 0, false);
1773
1774 return (p);
1775 }
1776
1777 JEMALLOC_ALWAYS_INLINE_C void *
1778 irealloc_prof(tsd_t *tsd, void *old_ptr, size_t old_usize, size_t usize)
1779 {
1780 void *p;
1781 bool prof_active;
1782 prof_tctx_t *old_tctx, *tctx;
1783
1784 prof_active = prof_active_get_unlocked();
1785 old_tctx = prof_tctx_get(old_ptr);
1786 tctx = prof_alloc_prep(tsd, usize, prof_active, true);
1787 if (unlikely((uintptr_t)tctx != (uintptr_t)1U))
1788 p = irealloc_prof_sample(tsd, old_ptr, old_usize, usize, tctx);
1789 else
1790 p = iralloc(tsd, old_ptr, old_usize, usize, 0, false);
1791 if (unlikely(p == NULL)) {
1792 prof_alloc_rollback(tsd, tctx, true);
1793 return (NULL);
1794 }
1795 prof_realloc(tsd, p, usize, tctx, prof_active, true, old_ptr, old_usize,
1796 old_tctx);
1797
1798 return (p);
1799 }
1800
1801 JEMALLOC_INLINE_C void
1802 ifree(tsd_t *tsd, void *ptr, tcache_t *tcache, bool slow_path)
1803 {
1804 size_t usize;
1805 UNUSED size_t rzsize JEMALLOC_CC_SILENCE_INIT(0);
1806
1807 assert(ptr != NULL);
1808 assert(malloc_initialized() || IS_INITIALIZER);
1809
1810 if (config_prof && opt_prof) {
1811 usize = isalloc(ptr, config_prof);
1812 prof_free(tsd, ptr, usize);
1813 } else if (config_stats || config_valgrind)
1814 usize = isalloc(ptr, config_prof);
1815 if (config_stats)
1816 *tsd_thread_deallocatedp_get(tsd) += usize;
1817
1818 if (likely(!slow_path))
1819 iqalloc(tsd, ptr, tcache, false);
1820 else {
1821 if (config_valgrind && unlikely(in_valgrind))
1822 rzsize = p2rz(ptr);
1823 iqalloc(tsd, ptr, tcache, true);
1824 JEMALLOC_VALGRIND_FREE(ptr, rzsize);
1825 }
1826 }
1827
1828 JEMALLOC_INLINE_C void
1829 isfree(tsd_t *tsd, void *ptr, size_t usize, tcache_t *tcache)
1830 {
1831 UNUSED size_t rzsize JEMALLOC_CC_SILENCE_INIT(0);
1832
1833 assert(ptr != NULL);
1834 assert(malloc_initialized() || IS_INITIALIZER);
1835
1836 if (config_prof && opt_prof)
1837 prof_free(tsd, ptr, usize);
1838 if (config_stats)
1839 *tsd_thread_deallocatedp_get(tsd) += usize;
1840 if (config_valgrind && unlikely(in_valgrind))
1841 rzsize = p2rz(ptr);
1842 isqalloc(tsd, ptr, usize, tcache);
1843 JEMALLOC_VALGRIND_FREE(ptr, rzsize);
1844 }
1845
1846 JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN
1847 void JEMALLOC_NOTHROW *
1848 JEMALLOC_ALLOC_SIZE(2)
1849 je_realloc(void *ptr, size_t size)
1850 {
1851 void *ret;
1852 tsd_t *tsd JEMALLOC_CC_SILENCE_INIT(NULL);
1853 size_t usize JEMALLOC_CC_SILENCE_INIT(0);
1854 size_t old_usize = 0;
1855 UNUSED size_t old_rzsize JEMALLOC_CC_SILENCE_INIT(0);
1856
1857 if (unlikely(size == 0)) {
1858 if (ptr != NULL) {
1859 /* realloc(ptr, 0) is equivalent to free(ptr). */
1860 UTRACE(ptr, 0, 0);
1861 tsd = tsd_fetch();
1862 ifree(tsd, ptr, tcache_get(tsd, false), true);
1863 return (NULL);
1864 }
1865 size = 1;
1866 }
1867
1868 if (likely(ptr != NULL)) {
1869 assert(malloc_initialized() || IS_INITIALIZER);
1870 malloc_thread_init();
1871 tsd = tsd_fetch();
1872
1873 old_usize = isalloc(ptr, config_prof);
1874 if (config_valgrind && unlikely(in_valgrind))
1875 old_rzsize = config_prof ? p2rz(ptr) : u2rz(old_usize);
1876
1877 if (config_prof && opt_prof) {
1878 usize = s2u(size);
1879 ret = unlikely(usize == 0 || usize > HUGE_MAXCLASS) ?
1880 NULL : irealloc_prof(tsd, ptr, old_usize, usize);
1881 } else {
1882 if (config_stats || (config_valgrind &&
1883 unlikely(in_valgrind)))
1884 usize = s2u(size);
1885 ret = iralloc(tsd, ptr, old_usize, size, 0, false);
1886 }
1887 } else {
1888 /* realloc(NULL, size) is equivalent to malloc(size). */
1889 if (likely(!malloc_slow))
1890 ret = imalloc_body(size, &tsd, &usize, false);
1891 else
1892 ret = imalloc_body(size, &tsd, &usize, true);
1893 }
1894
1895 if (unlikely(ret == NULL)) {
1896 if (config_xmalloc && unlikely(opt_xmalloc)) {
1897 malloc_write("<jemalloc>: Error in realloc(): "
1898 "out of memory\n");
1899 abort();
1900 }
1901 set_errno(ENOMEM);
1902 }
1903 if (config_stats && likely(ret != NULL)) {
1904 assert(usize == isalloc(ret, config_prof));
1905 *tsd_thread_allocatedp_get(tsd) += usize;
1906 *tsd_thread_deallocatedp_get(tsd) += old_usize;
1907 }
1908 UTRACE(ptr, size, ret);
1909 JEMALLOC_VALGRIND_REALLOC(true, ret, usize, true, ptr, old_usize,
1910 old_rzsize, true, false);
1911 return (ret);
1912 }
1913
1914 JEMALLOC_EXPORT void JEMALLOC_NOTHROW
1915 je_free(void *ptr)
1916 {
1917
1918 UTRACE(ptr, 0, 0);
1919 if (likely(ptr != NULL)) {
1920 tsd_t *tsd = tsd_fetch();
1921 if (likely(!malloc_slow))
1922 ifree(tsd, ptr, tcache_get(tsd, false), false);
1923 else
1924 ifree(tsd, ptr, tcache_get(tsd, false), true);
1925 }
1926 }
1927
1928 /*
1929 * End malloc(3)-compatible functions.
1930 */
1931 /******************************************************************************/
1932 /*
1933 * Begin non-standard override functions.
1934 */
1935
1936 #ifdef JEMALLOC_OVERRIDE_MEMALIGN
1937 JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN
1938 void JEMALLOC_NOTHROW *
1939 JEMALLOC_ATTR(malloc)
1940 je_memalign(size_t alignment, size_t size)
1941 {
1942 void *ret JEMALLOC_CC_SILENCE_INIT(NULL);
1943 if (unlikely(imemalign(&ret, alignment, size, 1) != 0))
1944 ret = NULL;
1945 JEMALLOC_VALGRIND_MALLOC(ret != NULL, ret, size, false);
1946 return (ret);
1947 }
1948 #endif
1949
1950 #ifdef JEMALLOC_OVERRIDE_VALLOC
1951 JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN
1952 void JEMALLOC_NOTHROW *
1953 JEMALLOC_ATTR(malloc)
1954 je_valloc(size_t size)
1955 {
1956 void *ret JEMALLOC_CC_SILENCE_INIT(NULL);
1957 if (unlikely(imemalign(&ret, PAGE, size, 1) != 0))
1958 ret = NULL;
1959 JEMALLOC_VALGRIND_MALLOC(ret != NULL, ret, size, false);
1960 return (ret);
1961 }
1962 #endif
1963
1964 /*
1965 * is_malloc(je_malloc) is some macro magic to detect if jemalloc_defs.h has
1966 * #define je_malloc malloc
1967 */
1968 #define malloc_is_malloc 1
1969 #define is_malloc_(a) malloc_is_ ## a
1970 #define is_malloc(a) is_malloc_(a)
1971
1972 #if ((is_malloc(je_malloc) == 1) && defined(JEMALLOC_GLIBC_MALLOC_HOOK))
1973 /*
1974 * glibc provides the RTLD_DEEPBIND flag for dlopen which can make it possible
1975 * to inconsistently reference libc's malloc(3)-compatible functions
1976 * (https://bugzilla.mozilla.org/show_bug.cgi?id=493541).
1977 *
1978 * These definitions interpose hooks in glibc. The functions are actually
1979 * passed an extra argument for the caller return address, which will be
1980 * ignored.
1981 */
1982 JEMALLOC_EXPORT void (*__free_hook)(void *ptr) = je_free;
1983 JEMALLOC_EXPORT void *(*__malloc_hook)(size_t size) = je_malloc;
1984 JEMALLOC_EXPORT void *(*__realloc_hook)(void *ptr, size_t size) = je_realloc;
1985 # ifdef JEMALLOC_GLIBC_MEMALIGN_HOOK
1986 JEMALLOC_EXPORT void *(*__memalign_hook)(size_t alignment, size_t size) =
1987 je_memalign;
1988 # endif
1989 #endif
1990
1991 /*
1992 * End non-standard override functions.
1993 */
1994 /******************************************************************************/
1995 /*
1996 * Begin non-standard functions.
1997 */
1998
1999 JEMALLOC_ALWAYS_INLINE_C bool
2000 imallocx_flags_decode_hard(tsd_t *tsd, size_t size, int flags, size_t *usize,
2001 size_t *alignment, bool *zero, tcache_t **tcache, arena_t **arena)
2002 {
2003
2004 if ((flags & MALLOCX_LG_ALIGN_MASK) == 0) {
2005 *alignment = 0;
2006 *usize = s2u(size);
2007 } else {
2008 *alignment = MALLOCX_ALIGN_GET_SPECIFIED(flags);
2009 *usize = sa2u(size, *alignment);
2010 }
2011 if (unlikely(*usize == 0 || *usize > HUGE_MAXCLASS))
2012 return (true);
2013 *zero = MALLOCX_ZERO_GET(flags);
2014 if ((flags & MALLOCX_TCACHE_MASK) != 0) {
2015 if ((flags & MALLOCX_TCACHE_MASK) == MALLOCX_TCACHE_NONE)
2016 *tcache = NULL;
2017 else
2018 *tcache = tcaches_get(tsd, MALLOCX_TCACHE_GET(flags));
2019 } else
2020 *tcache = tcache_get(tsd, true);
2021 if ((flags & MALLOCX_ARENA_MASK) != 0) {
2022 unsigned arena_ind = MALLOCX_ARENA_GET(flags);
2023 *arena = arena_get(arena_ind, true);
2024 if (unlikely(*arena == NULL))
2025 return (true);
2026 } else
2027 *arena = NULL;
2028 return (false);
2029 }
2030
2031 JEMALLOC_ALWAYS_INLINE_C bool
2032 imallocx_flags_decode(tsd_t *tsd, size_t size, int flags, size_t *usize,
2033 size_t *alignment, bool *zero, tcache_t **tcache, arena_t **arena)
2034 {
2035
2036 if (likely(flags == 0)) {
2037 *usize = s2u(size);
2038 if (unlikely(*usize == 0 || *usize > HUGE_MAXCLASS))
2039 return (true);
2040 *alignment = 0;
2041 *zero = false;
2042 *tcache = tcache_get(tsd, true);
2043 *arena = NULL;
2044 return (false);
2045 } else {
2046 return (imallocx_flags_decode_hard(tsd, size, flags, usize,
2047 alignment, zero, tcache, arena));
2048 }
2049 }
2050
2051 JEMALLOC_ALWAYS_INLINE_C void *
2052 imallocx_flags(tsd_t *tsd, size_t usize, size_t alignment, bool zero,
2053 tcache_t *tcache, arena_t *arena)
2054 {
2055 szind_t ind;
2056
2057 if (unlikely(alignment != 0))
2058 return (ipalloct(tsd, usize, alignment, zero, tcache, arena));
2059 ind = size2index(usize);
2060 assert(ind < NSIZES);
2061 if (unlikely(zero))
2062 return (icalloct(tsd, usize, ind, tcache, arena));
2063 return (imalloct(tsd, usize, ind, tcache, arena));
2064 }
2065
2066 static void *
2067 imallocx_prof_sample(tsd_t *tsd, size_t usize, size_t alignment, bool zero,
2068 tcache_t *tcache, arena_t *arena)
2069 {
2070 void *p;
2071
2072 if (usize <= SMALL_MAXCLASS) {
2073 assert(((alignment == 0) ? s2u(LARGE_MINCLASS) :
2074 sa2u(LARGE_MINCLASS, alignment)) == LARGE_MINCLASS);
2075 p = imallocx_flags(tsd, LARGE_MINCLASS, alignment, zero, tcache,
2076 arena);
2077 if (p == NULL)
2078 return (NULL);
2079 arena_prof_promoted(p, usize);
2080 } else
2081 p = imallocx_flags(tsd, usize, alignment, zero, tcache, arena);
2082
2083 return (p);
2084 }
2085
2086 JEMALLOC_ALWAYS_INLINE_C void *
2087 imallocx_prof(tsd_t *tsd, size_t size, int flags, size_t *usize)
2088 {
2089 void *p;
2090 size_t alignment;
2091 bool zero;
2092 tcache_t *tcache;
2093 arena_t *arena;
2094 prof_tctx_t *tctx;
2095
2096 if (unlikely(imallocx_flags_decode(tsd, size, flags, usize, &alignment,
2097 &zero, &tcache, &arena)))
2098 return (NULL);
2099 tctx = prof_alloc_prep(tsd, *usize, prof_active_get_unlocked(), true);
2100 if (likely((uintptr_t)tctx == (uintptr_t)1U))
2101 p = imallocx_flags(tsd, *usize, alignment, zero, tcache, arena);
2102 else if ((uintptr_t)tctx > (uintptr_t)1U) {
2103 p = imallocx_prof_sample(tsd, *usize, alignment, zero, tcache,
2104 arena);
2105 } else
2106 p = NULL;
2107 if (unlikely(p == NULL)) {
2108 prof_alloc_rollback(tsd, tctx, true);
2109 return (NULL);
2110 }
2111 prof_malloc(p, *usize, tctx);
2112
2113 assert(alignment == 0 || ((uintptr_t)p & (alignment - 1)) == ZU(0));
2114 return (p);
2115 }
2116
2117 JEMALLOC_ALWAYS_INLINE_C void *
2118 imallocx_no_prof(tsd_t *tsd, size_t size, int flags, size_t *usize)
2119 {
2120 void *p;
2121 size_t alignment;
2122 bool zero;
2123 tcache_t *tcache;
2124 arena_t *arena;
2125
2126 if (likely(flags == 0)) {
2127 szind_t ind = size2index(size);
2128 if (unlikely(ind >= NSIZES))
2129 return (NULL);
2130 if (config_stats || (config_valgrind &&
2131 unlikely(in_valgrind))) {
2132 *usize = index2size(ind);
2133 assert(*usize > 0 && *usize <= HUGE_MAXCLASS);
2134 }
2135 return (imalloc(tsd, size, ind, true));
2136 }
2137
2138 if (unlikely(imallocx_flags_decode_hard(tsd, size, flags, usize,
2139 &alignment, &zero, &tcache, &arena)))
2140 return (NULL);
2141 p = imallocx_flags(tsd, *usize, alignment, zero, tcache, arena);
2142 assert(alignment == 0 || ((uintptr_t)p & (alignment - 1)) == ZU(0));
2143 return (p);
2144 }
2145
2146 JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN
2147 void JEMALLOC_NOTHROW *
2148 JEMALLOC_ATTR(malloc) JEMALLOC_ALLOC_SIZE(1)
2149 je_mallocx(size_t size, int flags)
2150 {
2151 tsd_t *tsd;
2152 void *p;
2153 size_t usize;
2154
2155 assert(size != 0);
2156
2157 if (unlikely(malloc_init()))
2158 goto label_oom;
2159 tsd = tsd_fetch();
2160
2161 if (config_prof && opt_prof)
2162 p = imallocx_prof(tsd, size, flags, &usize);
2163 else
2164 p = imallocx_no_prof(tsd, size, flags, &usize);
2165 if (unlikely(p == NULL))
2166 goto label_oom;
2167
2168 if (config_stats) {
2169 assert(usize == isalloc(p, config_prof));
2170 *tsd_thread_allocatedp_get(tsd) += usize;
2171 }
2172 UTRACE(0, size, p);
2173 JEMALLOC_VALGRIND_MALLOC(true, p, usize, MALLOCX_ZERO_GET(flags));
2174 return (p);
2175 label_oom:
2176 if (config_xmalloc && unlikely(opt_xmalloc)) {
2177 malloc_write("<jemalloc>: Error in mallocx(): out of memory\n");
2178 abort();
2179 }
2180 UTRACE(0, size, 0);
2181 return (NULL);
2182 }
2183
2184 static void *
2185 irallocx_prof_sample(tsd_t *tsd, void *old_ptr, size_t old_usize,
2186 size_t usize, size_t alignment, bool zero, tcache_t *tcache, arena_t *arena,
2187 prof_tctx_t *tctx)
2188 {
2189 void *p;
2190
2191 if (tctx == NULL)
2192 return (NULL);
2193 if (usize <= SMALL_MAXCLASS) {
2194 p = iralloct(tsd, old_ptr, old_usize, LARGE_MINCLASS, alignment,
2195 zero, tcache, arena);
2196 if (p == NULL)
2197 return (NULL);
2198 arena_prof_promoted(p, usize);
2199 } else {
2200 p = iralloct(tsd, old_ptr, old_usize, usize, alignment, zero,
2201 tcache, arena);
2202 }
2203
2204 return (p);
2205 }
2206
2207 JEMALLOC_ALWAYS_INLINE_C void *
2208 irallocx_prof(tsd_t *tsd, void *old_ptr, size_t old_usize, size_t size,
2209 size_t alignment, size_t *usize, bool zero, tcache_t *tcache,
2210 arena_t *arena)
2211 {
2212 void *p;
2213 bool prof_active;
2214 prof_tctx_t *old_tctx, *tctx;
2215
2216 prof_active = prof_active_get_unlocked();
2217 old_tctx = prof_tctx_get(old_ptr);
2218 tctx = prof_alloc_prep(tsd, *usize, prof_active, true);
2219 if (unlikely((uintptr_t)tctx != (uintptr_t)1U)) {
2220 p = irallocx_prof_sample(tsd, old_ptr, old_usize, *usize,
2221 alignment, zero, tcache, arena, tctx);
2222 } else {
2223 p = iralloct(tsd, old_ptr, old_usize, size, alignment, zero,
2224 tcache, arena);
2225 }
2226 if (unlikely(p == NULL)) {
2227 prof_alloc_rollback(tsd, tctx, true);
2228 return (NULL);
2229 }
2230
2231 if (p == old_ptr && alignment != 0) {
2232 /*
2233 * The allocation did not move, so it is possible that the size
2234 * class is smaller than would guarantee the requested
2235 * alignment, and that the alignment constraint was
2236 * serendipitously satisfied. Additionally, old_usize may not
2237 * be the same as the current usize because of in-place large
2238 * reallocation. Therefore, query the actual value of usize.
2239 */
2240 *usize = isalloc(p, config_prof);
2241 }
2242 prof_realloc(tsd, p, *usize, tctx, prof_active, true, old_ptr,
2243 old_usize, old_tctx);
2244
2245 return (p);
2246 }
2247
2248 JEMALLOC_EXPORT JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN
2249 void JEMALLOC_NOTHROW *
2250 JEMALLOC_ALLOC_SIZE(2)
2251 je_rallocx(void *ptr, size_t size, int flags)
2252 {
2253 void *p;
2254 tsd_t *tsd;
2255 size_t usize;
2256 size_t old_usize;
2257 UNUSED size_t old_rzsize JEMALLOC_CC_SILENCE_INIT(0);
2258 size_t alignment = MALLOCX_ALIGN_GET(flags);
2259 bool zero = flags & MALLOCX_ZERO;
2260 arena_t *arena;
2261 tcache_t *tcache;
2262
2263 assert(ptr != NULL);
2264 assert(size != 0);
2265 assert(malloc_initialized() || IS_INITIALIZER);
2266 malloc_thread_init();
2267 tsd = tsd_fetch();
2268
2269 if (unlikely((flags & MALLOCX_ARENA_MASK) != 0)) {
2270 unsigned arena_ind = MALLOCX_ARENA_GET(flags);
2271 arena = arena_get(arena_ind, true);
2272 if (unlikely(arena == NULL))
2273 goto label_oom;
2274 } else
2275 arena = NULL;
2276
2277 if (unlikely((flags & MALLOCX_TCACHE_MASK) != 0)) {
2278 if ((flags & MALLOCX_TCACHE_MASK) == MALLOCX_TCACHE_NONE)
2279 tcache = NULL;
2280 else
2281 tcache = tcaches_get(tsd, MALLOCX_TCACHE_GET(flags));
2282 } else
2283 tcache = tcache_get(tsd, true);
2284
2285 old_usize = isalloc(ptr, config_prof);
2286 if (config_valgrind && unlikely(in_valgrind))
2287 old_rzsize = u2rz(old_usize);
2288
2289 if (config_prof && opt_prof) {
2290 usize = (alignment == 0) ? s2u(size) : sa2u(size, alignment);
2291 if (unlikely(usize == 0 || usize > HUGE_MAXCLASS))
2292 goto label_oom;
2293 p = irallocx_prof(tsd, ptr, old_usize, size, alignment, &usize,
2294 zero, tcache, arena);
2295 if (unlikely(p == NULL))
2296 goto label_oom;
2297 } else {
2298 p = iralloct(tsd, ptr, old_usize, size, alignment, zero,
2299 tcache, arena);
2300 if (unlikely(p == NULL))
2301 goto label_oom;
2302 if (config_stats || (config_valgrind && unlikely(in_valgrind)))
2303 usize = isalloc(p, config_prof);
2304 }
2305 assert(alignment == 0 || ((uintptr_t)p & (alignment - 1)) == ZU(0));
2306
2307 if (config_stats) {
2308 *tsd_thread_allocatedp_get(tsd) += usize;
2309 *tsd_thread_deallocatedp_get(tsd) += old_usize;
2310 }
2311 UTRACE(ptr, size, p);
2312 JEMALLOC_VALGRIND_REALLOC(true, p, usize, false, ptr, old_usize,
2313 old_rzsize, false, zero);
2314 return (p);
2315 label_oom:
2316 if (config_xmalloc && unlikely(opt_xmalloc)) {
2317 malloc_write("<jemalloc>: Error in rallocx(): out of memory\n");
2318 abort();
2319 }
2320 UTRACE(ptr, size, 0);
2321 return (NULL);
2322 }
2323
2324 JEMALLOC_ALWAYS_INLINE_C size_t
2325 ixallocx_helper(tsd_t *tsd, void *ptr, size_t old_usize, size_t size,
2326 size_t extra, size_t alignment, bool zero)
2327 {
2328 size_t usize;
2329
2330 if (ixalloc(tsd, ptr, old_usize, size, extra, alignment, zero))
2331 return (old_usize);
2332 usize = isalloc(ptr, config_prof);
2333
2334 return (usize);
2335 }
2336
2337 static size_t
2338 ixallocx_prof_sample(tsd_t *tsd, void *ptr, size_t old_usize, size_t size,
2339 size_t extra, size_t alignment, bool zero, prof_tctx_t *tctx)
2340 {
2341 size_t usize;
2342
2343 if (tctx == NULL)
2344 return (old_usize);
2345 usize = ixallocx_helper(tsd, ptr, old_usize, size, extra, alignment,
2346 zero);
2347
2348 return (usize);
2349 }
2350
2351 JEMALLOC_ALWAYS_INLINE_C size_t
2352 ixallocx_prof(tsd_t *tsd, void *ptr, size_t old_usize, size_t size,
2353 size_t extra, size_t alignment, bool zero)
2354 {
2355 size_t usize_max, usize;
2356 bool prof_active;
2357 prof_tctx_t *old_tctx, *tctx;
2358
2359 prof_active = prof_active_get_unlocked();
2360 old_tctx = prof_tctx_get(ptr);
2361 /*
2362 * usize isn't knowable before ixalloc() returns when extra is non-zero.
2363 * Therefore, compute its maximum possible value and use that in
2364 * prof_alloc_prep() to decide whether to capture a backtrace.
2365 * prof_realloc() will use the actual usize to decide whether to sample.
2366 */
2367 if (alignment == 0) {
2368 usize_max = s2u(size+extra);
2369 assert(usize_max > 0 && usize_max <= HUGE_MAXCLASS);
2370 } else {
2371 usize_max = sa2u(size+extra, alignment);
2372 if (unlikely(usize_max == 0 || usize_max > HUGE_MAXCLASS)) {
2373 /*
2374 * usize_max is out of range, and chances are that
2375 * allocation will fail, but use the maximum possible
2376 * value and carry on with prof_alloc_prep(), just in
2377 * case allocation succeeds.
2378 */
2379 usize_max = HUGE_MAXCLASS;
2380 }
2381 }
2382 tctx = prof_alloc_prep(tsd, usize_max, prof_active, false);
2383
2384 if (unlikely((uintptr_t)tctx != (uintptr_t)1U)) {
2385 usize = ixallocx_prof_sample(tsd, ptr, old_usize, size, extra,
2386 alignment, zero, tctx);
2387 } else {
2388 usize = ixallocx_helper(tsd, ptr, old_usize, size, extra,
2389 alignment, zero);
2390 }
2391 if (usize == old_usize) {
2392 prof_alloc_rollback(tsd, tctx, false);
2393 return (usize);
2394 }
2395 prof_realloc(tsd, ptr, usize, tctx, prof_active, false, ptr, old_usize,
2396 old_tctx);
2397
2398 return (usize);
2399 }
2400
2401 JEMALLOC_EXPORT size_t JEMALLOC_NOTHROW
2402 je_xallocx(void *ptr, size_t size, size_t extra, int flags)
2403 {
2404 tsd_t *tsd;
2405 size_t usize, old_usize;
2406 UNUSED size_t old_rzsize JEMALLOC_CC_SILENCE_INIT(0);
2407 size_t alignment = MALLOCX_ALIGN_GET(flags);
2408 bool zero = flags & MALLOCX_ZERO;
2409
2410 assert(ptr != NULL);
2411 assert(size != 0);
2412 assert(SIZE_T_MAX - size >= extra);
2413 assert(malloc_initialized() || IS_INITIALIZER);
2414 malloc_thread_init();
2415 tsd = tsd_fetch();
2416
2417 old_usize = isalloc(ptr, config_prof);
2418
2419 /*
2420 * The API explicitly absolves itself of protecting against (size +
2421 * extra) numerical overflow, but we may need to clamp extra to avoid
2422 * exceeding HUGE_MAXCLASS.
2423 *
2424 * Ordinarily, size limit checking is handled deeper down, but here we
2425 * have to check as part of (size + extra) clamping, since we need the
2426 * clamped value in the above helper functions.
2427 */
2428 if (unlikely(size > HUGE_MAXCLASS)) {
2429 usize = old_usize;
2430 goto label_not_resized;
2431 }
2432 if (unlikely(HUGE_MAXCLASS - size < extra))
2433 extra = HUGE_MAXCLASS - size;
2434
2435 if (config_valgrind && unlikely(in_valgrind))
2436 old_rzsize = u2rz(old_usize);
2437
2438 if (config_prof && opt_prof) {
2439 usize = ixallocx_prof(tsd, ptr, old_usize, size, extra,
2440 alignment, zero);
2441 } else {
2442 usize = ixallocx_helper(tsd, ptr, old_usize, size, extra,
2443 alignment, zero);
2444 }
2445 if (unlikely(usize == old_usize))
2446 goto label_not_resized;
2447
2448 if (config_stats) {
2449 *tsd_thread_allocatedp_get(tsd) += usize;
2450 *tsd_thread_deallocatedp_get(tsd) += old_usize;
2451 }
2452 JEMALLOC_VALGRIND_REALLOC(false, ptr, usize, false, ptr, old_usize,
2453 old_rzsize, false, zero);
2454 label_not_resized:
2455 UTRACE(ptr, size, ptr);
2456 return (usize);
2457 }
2458
2459 JEMALLOC_EXPORT size_t JEMALLOC_NOTHROW
2460 JEMALLOC_ATTR(pure)
2461 je_sallocx(const void *ptr, int flags)
2462 {
2463 size_t usize;
2464
2465 assert(malloc_initialized() || IS_INITIALIZER);
2466 malloc_thread_init();
2467
2468 if (config_ivsalloc)
2469 usize = ivsalloc(ptr, config_prof);
2470 else
2471 usize = isalloc(ptr, config_prof);
2472
2473 return (usize);
2474 }
2475
2476 JEMALLOC_EXPORT void JEMALLOC_NOTHROW
2477 je_dallocx(void *ptr, int flags)
2478 {
2479 tsd_t *tsd;
2480 tcache_t *tcache;
2481
2482 assert(ptr != NULL);
2483 assert(malloc_initialized() || IS_INITIALIZER);
2484
2485 tsd = tsd_fetch();
2486 if (unlikely((flags & MALLOCX_TCACHE_MASK) != 0)) {
2487 if ((flags & MALLOCX_TCACHE_MASK) == MALLOCX_TCACHE_NONE)
2488 tcache = NULL;
2489 else
2490 tcache = tcaches_get(tsd, MALLOCX_TCACHE_GET(flags));
2491 } else
2492 tcache = tcache_get(tsd, false);
2493
2494 UTRACE(ptr, 0, 0);
2495 ifree(tsd_fetch(), ptr, tcache, true);
2496 }
2497
2498 JEMALLOC_ALWAYS_INLINE_C size_t
2499 inallocx(size_t size, int flags)
2500 {
2501 size_t usize;
2502
2503 if (likely((flags & MALLOCX_LG_ALIGN_MASK) == 0))
2504 usize = s2u(size);
2505 else
2506 usize = sa2u(size, MALLOCX_ALIGN_GET_SPECIFIED(flags));
2507 return (usize);
2508 }
2509
2510 JEMALLOC_EXPORT void JEMALLOC_NOTHROW
2511 je_sdallocx(void *ptr, size_t size, int flags)
2512 {
2513 tsd_t *tsd;
2514 tcache_t *tcache;
2515 size_t usize;
2516
2517 assert(ptr != NULL);
2518 assert(malloc_initialized() || IS_INITIALIZER);
2519 usize = inallocx(size, flags);
2520 assert(usize == isalloc(ptr, config_prof));
2521
2522 tsd = tsd_fetch();
2523 if (unlikely((flags & MALLOCX_TCACHE_MASK) != 0)) {
2524 if ((flags & MALLOCX_TCACHE_MASK) == MALLOCX_TCACHE_NONE)
2525 tcache = NULL;
2526 else
2527 tcache = tcaches_get(tsd, MALLOCX_TCACHE_GET(flags));
2528 } else
2529 tcache = tcache_get(tsd, false);
2530
2531 UTRACE(ptr, 0, 0);
2532 isfree(tsd, ptr, usize, tcache);
2533 }
2534
2535 JEMALLOC_EXPORT size_t JEMALLOC_NOTHROW
2536 JEMALLOC_ATTR(pure)
2537 je_nallocx(size_t size, int flags)
2538 {
2539 size_t usize;
2540
2541 assert(size != 0);
2542
2543 if (unlikely(malloc_init()))
2544 return (0);
2545
2546 usize = inallocx(size, flags);
2547 if (unlikely(usize > HUGE_MAXCLASS))
2548 return (0);
2549
2550 return (usize);
2551 }
2552
2553 JEMALLOC_EXPORT int JEMALLOC_NOTHROW
2554 je_mallctl(const char *name, void *oldp, size_t *oldlenp, void *newp,
2555 size_t newlen)
2556 {
2557
2558 if (unlikely(malloc_init()))
2559 return (EAGAIN);
2560
2561 return (ctl_byname(name, oldp, oldlenp, newp, newlen));
2562 }
2563
2564 JEMALLOC_EXPORT int JEMALLOC_NOTHROW
2565 je_mallctlnametomib(const char *name, size_t *mibp, size_t *miblenp)
2566 {
2567
2568 if (unlikely(malloc_init()))
2569 return (EAGAIN);
2570
2571 return (ctl_nametomib(name, mibp, miblenp));
2572 }
2573
2574 JEMALLOC_EXPORT int JEMALLOC_NOTHROW
2575 je_mallctlbymib(const size_t *mib, size_t miblen, void *oldp, size_t *oldlenp,
2576 void *newp, size_t newlen)
2577 {
2578
2579 if (unlikely(malloc_init()))
2580 return (EAGAIN);
2581
2582 return (ctl_bymib(mib, miblen, oldp, oldlenp, newp, newlen));
2583 }
2584
2585 JEMALLOC_EXPORT void JEMALLOC_NOTHROW
2586 je_malloc_stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
2587 const char *opts)
2588 {
2589
2590 stats_print(write_cb, cbopaque, opts);
2591 }
2592
2593 JEMALLOC_EXPORT size_t JEMALLOC_NOTHROW
2594 je_malloc_usable_size(JEMALLOC_USABLE_SIZE_CONST void *ptr)
2595 {
2596 size_t ret;
2597
2598 assert(malloc_initialized() || IS_INITIALIZER);
2599 malloc_thread_init();
2600
2601 if (config_ivsalloc)
2602 ret = ivsalloc(ptr, config_prof);
2603 else
2604 ret = (ptr == NULL) ? 0 : isalloc(ptr, config_prof);
2605
2606 return (ret);
2607 }
2608
2609 /*
2610 * End non-standard functions.
2611 */
2612 /******************************************************************************/
2613 /*
2614 * The following functions are used by threading libraries for protection of
2615 * malloc during fork().
2616 */
2617
2618 /*
2619 * If an application creates a thread before doing any allocation in the main
2620 * thread, then calls fork(2) in the main thread followed by memory allocation
2621 * in the child process, a race can occur that results in deadlock within the
2622 * child: the main thread may have forked while the created thread had
2623 * partially initialized the allocator. Ordinarily jemalloc prevents
2624 * fork/malloc races via the following functions it registers during
2625 * initialization using pthread_atfork(), but of course that does no good if
2626 * the allocator isn't fully initialized at fork time. The following library
2627 * constructor is a partial solution to this problem. It may still be possible
2628 * to trigger the deadlock described above, but doing so would involve forking
2629 * via a library constructor that runs before jemalloc's runs.
2630 */
2631 JEMALLOC_ATTR(constructor)
2632 static void
2633 jemalloc_constructor(void)
2634 {
2635
2636 malloc_init();
2637 }
2638
2639 #ifndef JEMALLOC_MUTEX_INIT_CB
2640 void
2641 jemalloc_prefork(void)
2642 #else
2643 JEMALLOC_EXPORT void
2644 _malloc_prefork(void)
2645 #endif
2646 {
2647 unsigned i, narenas;
2648
2649 #ifdef JEMALLOC_MUTEX_INIT_CB
2650 if (!malloc_initialized())
2651 return;
2652 #endif
2653 assert(malloc_initialized());
2654
2655 /* Acquire all mutexes in a safe order. */
2656 ctl_prefork();
2657 prof_prefork();
2658 malloc_mutex_prefork(&arenas_lock);
2659 for (i = 0, narenas = narenas_total_get(); i < narenas; i++) {
2660 arena_t *arena;
2661
2662 if ((arena = arena_get(i, false)) != NULL)
2663 arena_prefork(arena);
2664 }
2665 chunk_prefork();
2666 base_prefork();
2667 }
2668
2669 #ifndef JEMALLOC_MUTEX_INIT_CB
2670 void
2671 jemalloc_postfork_parent(void)
2672 #else
2673 JEMALLOC_EXPORT void
2674 _malloc_postfork(void)
2675 #endif
2676 {
2677 unsigned i, narenas;
2678
2679 #ifdef JEMALLOC_MUTEX_INIT_CB
2680 if (!malloc_initialized())
2681 return;
2682 #endif
2683 assert(malloc_initialized());
2684
2685 /* Release all mutexes, now that fork() has completed. */
2686 base_postfork_parent();
2687 chunk_postfork_parent();
2688 for (i = 0, narenas = narenas_total_get(); i < narenas; i++) {
2689 arena_t *arena;
2690
2691 if ((arena = arena_get(i, false)) != NULL)
2692 arena_postfork_parent(arena);
2693 }
2694 malloc_mutex_postfork_parent(&arenas_lock);
2695 prof_postfork_parent();
2696 ctl_postfork_parent();
2697 }
2698
2699 void
2700 jemalloc_postfork_child(void)
2701 {
2702 unsigned i, narenas;
2703
2704 assert(malloc_initialized());
2705
2706 /* Release all mutexes, now that fork() has completed. */
2707 base_postfork_child();
2708 chunk_postfork_child();
2709 for (i = 0, narenas = narenas_total_get(); i < narenas; i++) {
2710 arena_t *arena;
2711
2712 if ((arena = arena_get(i, false)) != NULL)
2713 arena_postfork_child(arena);
2714 }
2715 malloc_mutex_postfork_child(&arenas_lock);
2716 prof_postfork_child();
2717 ctl_postfork_child();
2718 }
2719
2720 /******************************************************************************/