1/* Malloc implementation for multiple threads without lock contention.
2 Copyright (C) 2001-2017 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4 Contributed by Wolfram Gloger <wg@malloc.de>, 2001.
5
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public License as
8 published by the Free Software Foundation; either version 2.1 of the
9 License, or (at your option) any later version.
10
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; see the file COPYING.LIB. If
18 not, see <http://www.gnu.org/licenses/>. */
19
20#include <stdbool.h>
21
22#if HAVE_TUNABLES
23# define TUNABLE_NAMESPACE malloc
24#endif
25#include <elf/dl-tunables.h>
26
27/* Compile-time constants. */
28
29#define HEAP_MIN_SIZE (32 * 1024)
30#ifndef HEAP_MAX_SIZE
31# ifdef DEFAULT_MMAP_THRESHOLD_MAX
32# define HEAP_MAX_SIZE (2 * DEFAULT_MMAP_THRESHOLD_MAX)
33# else
34# define HEAP_MAX_SIZE (1024 * 1024) /* must be a power of two */
35# endif
36#endif
37
38/* HEAP_MIN_SIZE and HEAP_MAX_SIZE limit the size of mmap()ed heaps
39 that are dynamically created for multi-threaded programs. The
40 maximum size must be a power of two, for fast determination of
41 which heap belongs to a chunk. It should be much larger than the
42 mmap threshold, so that requests with a size just below that
43 threshold can be fulfilled without creating too many heaps. */
44
45/***************************************************************************/
46
47#define top(ar_ptr) ((ar_ptr)->top)
48
49/* A heap is a single contiguous memory region holding (coalesceable)
50 malloc_chunks. It is allocated with mmap() and always starts at an
51 address aligned to HEAP_MAX_SIZE. */
52
53typedef struct _heap_info
54{
55 mstate ar_ptr; /* Arena for this heap. */
56 struct _heap_info *prev; /* Previous heap. */
57 size_t size; /* Current size in bytes. */
58 size_t mprotect_size; /* Size in bytes that has been mprotected
59 PROT_READ|PROT_WRITE. */
60 /* Make sure the following data is properly aligned, particularly
61 that sizeof (heap_info) + 2 * SIZE_SZ is a multiple of
62 MALLOC_ALIGNMENT. */
63 char pad[-6 * SIZE_SZ & MALLOC_ALIGN_MASK];
64} heap_info;
65
66/* Get a compile-time error if the heap_info padding is not correct
67 to make alignment work as expected in sYSMALLOc. */
68extern int sanity_check_heap_info_alignment[(sizeof (heap_info)
69 + 2 * SIZE_SZ) % MALLOC_ALIGNMENT
70 ? -1 : 1];
71
72/* Thread specific data. */
73
74static __thread mstate thread_arena attribute_tls_model_ie;
75
76/* Arena free list. free_list_lock synchronizes access to the
77 free_list variable below, and the next_free and attached_threads
78 members of struct malloc_state objects. No other locks must be
79 acquired after free_list_lock has been acquired. */
80
81__libc_lock_define_initialized (static, free_list_lock);
82static size_t narenas = 1;
83static mstate free_list;
84
85/* list_lock prevents concurrent writes to the next member of struct
86 malloc_state objects.
87
88 Read access to the next member is supposed to synchronize with the
89 atomic_write_barrier and the write to the next member in
90 _int_new_arena. This suffers from data races; see the FIXME
91 comments in _int_new_arena and reused_arena.
92
93 list_lock also prevents concurrent forks. At the time list_lock is
94 acquired, no arena lock must have been acquired, but it is
95 permitted to acquire arena locks subsequently, while list_lock is
96 acquired. */
97__libc_lock_define_initialized (static, list_lock);
98
99/* Already initialized? */
100int __malloc_initialized = -1;
101
102/**************************************************************************/
103
104
105/* arena_get() acquires an arena and locks the corresponding mutex.
106 First, try the one last locked successfully by this thread. (This
107 is the common case and handled with a macro for speed.) Then, loop
108 once over the circularly linked list of arenas. If no arena is
109 readily available, create a new one. In this latter case, `size'
110 is just a hint as to how much memory will be required immediately
111 in the new arena. */
112
113#define arena_get(ptr, size) do { \
114 ptr = thread_arena; \
115 arena_lock (ptr, size); \
116 } while (0)
117
118#define arena_lock(ptr, size) do { \
119 if (ptr && !arena_is_corrupt (ptr)) \
120 __libc_lock_lock (ptr->mutex); \
121 else \
122 ptr = arena_get2 ((size), NULL); \
123 } while (0)
124
125/* find the heap and corresponding arena for a given ptr */
126
127#define heap_for_ptr(ptr) \
128 ((heap_info *) ((unsigned long) (ptr) & ~(HEAP_MAX_SIZE - 1)))
129#define arena_for_chunk(ptr) \
130 (chunk_main_arena (ptr) ? &main_arena : heap_for_ptr (ptr)->ar_ptr)
131
132
133/**************************************************************************/
134
135/* atfork support. */
136
137/* The following three functions are called around fork from a
138 multi-threaded process. We do not use the general fork handler
139 mechanism to make sure that our handlers are the last ones being
140 called, so that other fork handlers can use the malloc
141 subsystem. */
142
143void
144internal_function
145__malloc_fork_lock_parent (void)
146{
147 if (__malloc_initialized < 1)
148 return;
149
150 /* We do not acquire free_list_lock here because we completely
151 reconstruct free_list in __malloc_fork_unlock_child. */
152
153 __libc_lock_lock (list_lock);
154
155 for (mstate ar_ptr = &main_arena;; )
156 {
157 __libc_lock_lock (ar_ptr->mutex);
158 ar_ptr = ar_ptr->next;
159 if (ar_ptr == &main_arena)
160 break;
161 }
162}
163
164void
165internal_function
166__malloc_fork_unlock_parent (void)
167{
168 if (__malloc_initialized < 1)
169 return;
170
171 for (mstate ar_ptr = &main_arena;; )
172 {
173 __libc_lock_unlock (ar_ptr->mutex);
174 ar_ptr = ar_ptr->next;
175 if (ar_ptr == &main_arena)
176 break;
177 }
178 __libc_lock_unlock (list_lock);
179}
180
181void
182internal_function
183__malloc_fork_unlock_child (void)
184{
185 if (__malloc_initialized < 1)
186 return;
187
188 /* Push all arenas to the free list, except thread_arena, which is
189 attached to the current thread. */
190 __libc_lock_init (free_list_lock);
191 if (thread_arena != NULL)
192 thread_arena->attached_threads = 1;
193 free_list = NULL;
194 for (mstate ar_ptr = &main_arena;; )
195 {
196 __libc_lock_init (ar_ptr->mutex);
197 if (ar_ptr != thread_arena)
198 {
199 /* This arena is no longer attached to any thread. */
200 ar_ptr->attached_threads = 0;
201 ar_ptr->next_free = free_list;
202 free_list = ar_ptr;
203 }
204 ar_ptr = ar_ptr->next;
205 if (ar_ptr == &main_arena)
206 break;
207 }
208
209 __libc_lock_init (list_lock);
210}
211
212#if HAVE_TUNABLES
213static inline int do_set_mallopt_check (int32_t value);
214void
215TUNABLE_CALLBACK (set_mallopt_check) (tunable_val_t *valp)
216{
217 int32_t value = (int32_t) valp->numval;
218 do_set_mallopt_check (value);
219 if (check_action != 0)
220 __malloc_check_init ();
221}
222
223# define TUNABLE_CALLBACK_FNDECL(__name, __type) \
224static inline int do_ ## __name (__type value); \
225void \
226TUNABLE_CALLBACK (__name) (tunable_val_t *valp) \
227{ \
228 __type value = (__type) (valp)->numval; \
229 do_ ## __name (value); \
230}
231
232TUNABLE_CALLBACK_FNDECL (set_mmap_threshold, size_t)
233TUNABLE_CALLBACK_FNDECL (set_mmaps_max, int32_t)
234TUNABLE_CALLBACK_FNDECL (set_top_pad, size_t)
235TUNABLE_CALLBACK_FNDECL (set_perturb_byte, int32_t)
236TUNABLE_CALLBACK_FNDECL (set_trim_threshold, size_t)
237TUNABLE_CALLBACK_FNDECL (set_arena_max, size_t)
238TUNABLE_CALLBACK_FNDECL (set_arena_test, size_t)
239#if USE_TCACHE
240TUNABLE_CALLBACK_FNDECL (set_tcache_max, size_t)
241TUNABLE_CALLBACK_FNDECL (set_tcache_count, size_t)
242TUNABLE_CALLBACK_FNDECL (set_tcache_unsorted_limit, size_t)
243#endif
244#else
245/* Initialization routine. */
246#include <string.h>
247extern char **_environ;
248
249static char *
250internal_function
251next_env_entry (char ***position)
252{
253 char **current = *position;
254 char *result = NULL;
255
256 while (*current != NULL)
257 {
258 if (__builtin_expect ((*current)[0] == 'M', 0)
259 && (*current)[1] == 'A'
260 && (*current)[2] == 'L'
261 && (*current)[3] == 'L'
262 && (*current)[4] == 'O'
263 && (*current)[5] == 'C'
264 && (*current)[6] == '_')
265 {
266 result = &(*current)[7];
267
268 /* Save current position for next visit. */
269 *position = ++current;
270
271 break;
272 }
273
274 ++current;
275 }
276
277 return result;
278}
279#endif
280
281
282#ifdef SHARED
283static void *
284__failing_morecore (ptrdiff_t d)
285{
286 return (void *) MORECORE_FAILURE;
287}
288
289extern struct dl_open_hook *_dl_open_hook;
290libc_hidden_proto (_dl_open_hook);
291#endif
292
293static void
294ptmalloc_init (void)
295{
296 if (__malloc_initialized >= 0)
297 return;
298
299 __malloc_initialized = 0;
300
301#ifdef SHARED
302 /* In case this libc copy is in a non-default namespace, never use brk.
303 Likewise if dlopened from statically linked program. */
304 Dl_info di;
305 struct link_map *l;
306
307 if (_dl_open_hook != NULL
308 || (_dl_addr (ptmalloc_init, &di, &l, NULL) != 0
309 && l->l_ns != LM_ID_BASE))
310 __morecore = __failing_morecore;
311#endif
312
313 thread_arena = &main_arena;
314
315#if HAVE_TUNABLES
316 /* Ensure initialization/consolidation and do it under a lock so that a
317 thread attempting to use the arena in parallel waits on us till we
318 finish. */
319 __libc_lock_lock (main_arena.mutex);
320 malloc_consolidate (&main_arena);
321
322 TUNABLE_GET (check, int32_t, TUNABLE_CALLBACK (set_mallopt_check));
323 TUNABLE_GET (top_pad, size_t, TUNABLE_CALLBACK (set_top_pad));
324 TUNABLE_GET (perturb, int32_t, TUNABLE_CALLBACK (set_perturb_byte));
325 TUNABLE_GET (mmap_threshold, size_t, TUNABLE_CALLBACK (set_mmap_threshold));
326 TUNABLE_GET (trim_threshold, size_t, TUNABLE_CALLBACK (set_trim_threshold));
327 TUNABLE_GET (mmap_max, int32_t, TUNABLE_CALLBACK (set_mmaps_max));
328 TUNABLE_GET (arena_max, size_t, TUNABLE_CALLBACK (set_arena_max));
329 TUNABLE_GET (arena_test, size_t, TUNABLE_CALLBACK (set_arena_test));
330#if USE_TCACHE
331 TUNABLE_GET (tcache_max, size_t, TUNABLE_CALLBACK (set_tcache_max));
332 TUNABLE_GET (tcache_count, size_t, TUNABLE_CALLBACK (set_tcache_count));
333 TUNABLE_GET (tcache_unsorted_limit, size_t,
334 TUNABLE_CALLBACK (set_tcache_unsorted_limit));
335#endif
336 __libc_lock_unlock (main_arena.mutex);
337#else
338 const char *s = NULL;
339 if (__glibc_likely (_environ != NULL))
340 {
341 char **runp = _environ;
342 char *envline;
343
344 while (__builtin_expect ((envline = next_env_entry (&runp)) != NULL,
345 0))
346 {
347 size_t len = strcspn (envline, "=");
348
349 if (envline[len] != '=')
350 /* This is a "MALLOC_" variable at the end of the string
351 without a '=' character. Ignore it since otherwise we
352 will access invalid memory below. */
353 continue;
354
355 switch (len)
356 {
357 case 6:
358 if (memcmp (envline, "CHECK_", 6) == 0)
359 s = &envline[7];
360 break;
361 case 8:
362 if (!__builtin_expect (__libc_enable_secure, 0))
363 {
364 if (memcmp (envline, "TOP_PAD_", 8) == 0)
365 __libc_mallopt (M_TOP_PAD, atoi (&envline[9]));
366 else if (memcmp (envline, "PERTURB_", 8) == 0)
367 __libc_mallopt (M_PERTURB, atoi (&envline[9]));
368 }
369 break;
370 case 9:
371 if (!__builtin_expect (__libc_enable_secure, 0))
372 {
373 if (memcmp (envline, "MMAP_MAX_", 9) == 0)
374 __libc_mallopt (M_MMAP_MAX, atoi (&envline[10]));
375 else if (memcmp (envline, "ARENA_MAX", 9) == 0)
376 __libc_mallopt (M_ARENA_MAX, atoi (&envline[10]));
377 }
378 break;
379 case 10:
380 if (!__builtin_expect (__libc_enable_secure, 0))
381 {
382 if (memcmp (envline, "ARENA_TEST", 10) == 0)
383 __libc_mallopt (M_ARENA_TEST, atoi (&envline[11]));
384 }
385 break;
386 case 15:
387 if (!__builtin_expect (__libc_enable_secure, 0))
388 {
389 if (memcmp (envline, "TRIM_THRESHOLD_", 15) == 0)
390 __libc_mallopt (M_TRIM_THRESHOLD, atoi (&envline[16]));
391 else if (memcmp (envline, "MMAP_THRESHOLD_", 15) == 0)
392 __libc_mallopt (M_MMAP_THRESHOLD, atoi (&envline[16]));
393 }
394 break;
395 default:
396 break;
397 }
398 }
399 }
400 if (s && s[0])
401 {
402 __libc_mallopt (M_CHECK_ACTION, (int) (s[0] - '0'));
403 if (check_action != 0)
404 __malloc_check_init ();
405 }
406#endif
407
408#if HAVE_MALLOC_INIT_HOOK
409 void (*hook) (void) = atomic_forced_read (__malloc_initialize_hook);
410 if (hook != NULL)
411 (*hook)();
412#endif
413 __malloc_initialized = 1;
414}
415
416/* Managing heaps and arenas (for concurrent threads) */
417
418#if MALLOC_DEBUG > 1
419
420/* Print the complete contents of a single heap to stderr. */
421
422static void
423dump_heap (heap_info *heap)
424{
425 char *ptr;
426 mchunkptr p;
427
428 fprintf (stderr, "Heap %p, size %10lx:\n", heap, (long) heap->size);
429 ptr = (heap->ar_ptr != (mstate) (heap + 1)) ?
430 (char *) (heap + 1) : (char *) (heap + 1) + sizeof (struct malloc_state);
431 p = (mchunkptr) (((unsigned long) ptr + MALLOC_ALIGN_MASK) &
432 ~MALLOC_ALIGN_MASK);
433 for (;; )
434 {
435 fprintf (stderr, "chunk %p size %10lx", p, (long) p->size);
436 if (p == top (heap->ar_ptr))
437 {
438 fprintf (stderr, " (top)\n");
439 break;
440 }
441 else if (p->size == (0 | PREV_INUSE))
442 {
443 fprintf (stderr, " (fence)\n");
444 break;
445 }
446 fprintf (stderr, "\n");
447 p = next_chunk (p);
448 }
449}
450#endif /* MALLOC_DEBUG > 1 */
451
452/* If consecutive mmap (0, HEAP_MAX_SIZE << 1, ...) calls return decreasing
453 addresses as opposed to increasing, new_heap would badly fragment the
454 address space. In that case remember the second HEAP_MAX_SIZE part
455 aligned to HEAP_MAX_SIZE from last mmap (0, HEAP_MAX_SIZE << 1, ...)
456 call (if it is already aligned) and try to reuse it next time. We need
457 no locking for it, as kernel ensures the atomicity for us - worst case
458 we'll call mmap (addr, HEAP_MAX_SIZE, ...) for some value of addr in
459 multiple threads, but only one will succeed. */
460static char *aligned_heap_area;
461
462/* Create a new heap. size is automatically rounded up to a multiple
463 of the page size. */
464
465static heap_info *
466internal_function
467new_heap (size_t size, size_t top_pad)
468{
469 size_t pagesize = GLRO (dl_pagesize);
470 char *p1, *p2;
471 unsigned long ul;
472 heap_info *h;
473
474 if (size + top_pad < HEAP_MIN_SIZE)
475 size = HEAP_MIN_SIZE;
476 else if (size + top_pad <= HEAP_MAX_SIZE)
477 size += top_pad;
478 else if (size > HEAP_MAX_SIZE)
479 return 0;
480 else
481 size = HEAP_MAX_SIZE;
482 size = ALIGN_UP (size, pagesize);
483
484 /* A memory region aligned to a multiple of HEAP_MAX_SIZE is needed.
485 No swap space needs to be reserved for the following large
486 mapping (on Linux, this is the case for all non-writable mappings
487 anyway). */
488 p2 = MAP_FAILED;
489 if (aligned_heap_area)
490 {
491 p2 = (char *) MMAP (aligned_heap_area, HEAP_MAX_SIZE, PROT_NONE,
492 MAP_NORESERVE);
493 aligned_heap_area = NULL;
494 if (p2 != MAP_FAILED && ((unsigned long) p2 & (HEAP_MAX_SIZE - 1)))
495 {
496 __munmap (p2, HEAP_MAX_SIZE);
497 p2 = MAP_FAILED;
498 }
499 }
500 if (p2 == MAP_FAILED)
501 {
502 p1 = (char *) MMAP (0, HEAP_MAX_SIZE << 1, PROT_NONE, MAP_NORESERVE);
503 if (p1 != MAP_FAILED)
504 {
505 p2 = (char *) (((unsigned long) p1 + (HEAP_MAX_SIZE - 1))
506 & ~(HEAP_MAX_SIZE - 1));
507 ul = p2 - p1;
508 if (ul)
509 __munmap (p1, ul);
510 else
511 aligned_heap_area = p2 + HEAP_MAX_SIZE;
512 __munmap (p2 + HEAP_MAX_SIZE, HEAP_MAX_SIZE - ul);
513 }
514 else
515 {
516 /* Try to take the chance that an allocation of only HEAP_MAX_SIZE
517 is already aligned. */
518 p2 = (char *) MMAP (0, HEAP_MAX_SIZE, PROT_NONE, MAP_NORESERVE);
519 if (p2 == MAP_FAILED)
520 return 0;
521
522 if ((unsigned long) p2 & (HEAP_MAX_SIZE - 1))
523 {
524 __munmap (p2, HEAP_MAX_SIZE);
525 return 0;
526 }
527 }
528 }
529 if (__mprotect (p2, size, PROT_READ | PROT_WRITE) != 0)
530 {
531 __munmap (p2, HEAP_MAX_SIZE);
532 return 0;
533 }
534 h = (heap_info *) p2;
535 h->size = size;
536 h->mprotect_size = size;
537 LIBC_PROBE (memory_heap_new, 2, h, h->size);
538 return h;
539}
540
541/* Grow a heap. size is automatically rounded up to a
542 multiple of the page size. */
543
544static int
545grow_heap (heap_info *h, long diff)
546{
547 size_t pagesize = GLRO (dl_pagesize);
548 long new_size;
549
550 diff = ALIGN_UP (diff, pagesize);
551 new_size = (long) h->size + diff;
552 if ((unsigned long) new_size > (unsigned long) HEAP_MAX_SIZE)
553 return -1;
554
555 if ((unsigned long) new_size > h->mprotect_size)
556 {
557 if (__mprotect ((char *) h + h->mprotect_size,
558 (unsigned long) new_size - h->mprotect_size,
559 PROT_READ | PROT_WRITE) != 0)
560 return -2;
561
562 h->mprotect_size = new_size;
563 }
564
565 h->size = new_size;
566 LIBC_PROBE (memory_heap_more, 2, h, h->size);
567 return 0;
568}
569
570/* Shrink a heap. */
571
572static int
573shrink_heap (heap_info *h, long diff)
574{
575 long new_size;
576
577 new_size = (long) h->size - diff;
578 if (new_size < (long) sizeof (*h))
579 return -1;
580
581 /* Try to re-map the extra heap space freshly to save memory, and make it
582 inaccessible. See malloc-sysdep.h to know when this is true. */
583 if (__glibc_unlikely (check_may_shrink_heap ()))
584 {
585 if ((char *) MMAP ((char *) h + new_size, diff, PROT_NONE,
586 MAP_FIXED) == (char *) MAP_FAILED)
587 return -2;
588
589 h->mprotect_size = new_size;
590 }
591 else
592 __madvise ((char *) h + new_size, diff, MADV_DONTNEED);
593 /*fprintf(stderr, "shrink %p %08lx\n", h, new_size);*/
594
595 h->size = new_size;
596 LIBC_PROBE (memory_heap_less, 2, h, h->size);
597 return 0;
598}
599
600/* Delete a heap. */
601
602#define delete_heap(heap) \
603 do { \
604 if ((char *) (heap) + HEAP_MAX_SIZE == aligned_heap_area) \
605 aligned_heap_area = NULL; \
606 __munmap ((char *) (heap), HEAP_MAX_SIZE); \
607 } while (0)
608
609static int
610internal_function
611heap_trim (heap_info *heap, size_t pad)
612{
613 mstate ar_ptr = heap->ar_ptr;
614 unsigned long pagesz = GLRO (dl_pagesize);
615 mchunkptr top_chunk = top (ar_ptr), p, bck, fwd;
616 heap_info *prev_heap;
617 long new_size, top_size, top_area, extra, prev_size, misalign;
618
619 /* Can this heap go away completely? */
620 while (top_chunk == chunk_at_offset (heap, sizeof (*heap)))
621 {
622 prev_heap = heap->prev;
623 prev_size = prev_heap->size - (MINSIZE - 2 * SIZE_SZ);
624 p = chunk_at_offset (prev_heap, prev_size);
625 /* fencepost must be properly aligned. */
626 misalign = ((long) p) & MALLOC_ALIGN_MASK;
627 p = chunk_at_offset (prev_heap, prev_size - misalign);
628 assert (chunksize_nomask (p) == (0 | PREV_INUSE)); /* must be fencepost */
629 p = prev_chunk (p);
630 new_size = chunksize (p) + (MINSIZE - 2 * SIZE_SZ) + misalign;
631 assert (new_size > 0 && new_size < (long) (2 * MINSIZE));
632 if (!prev_inuse (p))
633 new_size += prev_size (p);
634 assert (new_size > 0 && new_size < HEAP_MAX_SIZE);
635 if (new_size + (HEAP_MAX_SIZE - prev_heap->size) < pad + MINSIZE + pagesz)
636 break;
637 ar_ptr->system_mem -= heap->size;
638 LIBC_PROBE (memory_heap_free, 2, heap, heap->size);
639 delete_heap (heap);
640 heap = prev_heap;
641 if (!prev_inuse (p)) /* consolidate backward */
642 {
643 p = prev_chunk (p);
644 unlink (ar_ptr, p, bck, fwd);
645 }
646 assert (((unsigned long) ((char *) p + new_size) & (pagesz - 1)) == 0);
647 assert (((char *) p + new_size) == ((char *) heap + heap->size));
648 top (ar_ptr) = top_chunk = p;
649 set_head (top_chunk, new_size | PREV_INUSE);
650 /*check_chunk(ar_ptr, top_chunk);*/
651 }
652
653 /* Uses similar logic for per-thread arenas as the main arena with systrim
654 and _int_free by preserving the top pad and rounding down to the nearest
655 page. */
656 top_size = chunksize (top_chunk);
657 if ((unsigned long)(top_size) <
658 (unsigned long)(mp_.trim_threshold))
659 return 0;
660
661 top_area = top_size - MINSIZE - 1;
662 if (top_area < 0 || (size_t) top_area <= pad)
663 return 0;
664
665 /* Release in pagesize units and round down to the nearest page. */
666 extra = ALIGN_DOWN(top_area - pad, pagesz);
667 if (extra == 0)
668 return 0;
669
670 /* Try to shrink. */
671 if (shrink_heap (heap, extra) != 0)
672 return 0;
673
674 ar_ptr->system_mem -= extra;
675
676 /* Success. Adjust top accordingly. */
677 set_head (top_chunk, (top_size - extra) | PREV_INUSE);
678 /*check_chunk(ar_ptr, top_chunk);*/
679 return 1;
680}
681
682/* Create a new arena with initial size "size". */
683
684/* If REPLACED_ARENA is not NULL, detach it from this thread. Must be
685 called while free_list_lock is held. */
686static void
687detach_arena (mstate replaced_arena)
688{
689 if (replaced_arena != NULL)
690 {
691 assert (replaced_arena->attached_threads > 0);
692 /* The current implementation only detaches from main_arena in
693 case of allocation failure. This means that it is likely not
694 beneficial to put the arena on free_list even if the
695 reference count reaches zero. */
696 --replaced_arena->attached_threads;
697 }
698}
699
700static mstate
701_int_new_arena (size_t size)
702{
703 mstate a;
704 heap_info *h;
705 char *ptr;
706 unsigned long misalign;
707
708 h = new_heap (size + (sizeof (*h) + sizeof (*a) + MALLOC_ALIGNMENT),
709 mp_.top_pad);
710 if (!h)
711 {
712 /* Maybe size is too large to fit in a single heap. So, just try
713 to create a minimally-sized arena and let _int_malloc() attempt
714 to deal with the large request via mmap_chunk(). */
715 h = new_heap (sizeof (*h) + sizeof (*a) + MALLOC_ALIGNMENT, mp_.top_pad);
716 if (!h)
717 return 0;
718 }
719 a = h->ar_ptr = (mstate) (h + 1);
720 malloc_init_state (a);
721 a->attached_threads = 1;
722 /*a->next = NULL;*/
723 a->system_mem = a->max_system_mem = h->size;
724
725 /* Set up the top chunk, with proper alignment. */
726 ptr = (char *) (a + 1);
727 misalign = (unsigned long) chunk2mem (ptr) & MALLOC_ALIGN_MASK;
728 if (misalign > 0)
729 ptr += MALLOC_ALIGNMENT - misalign;
730 top (a) = (mchunkptr) ptr;
731 set_head (top (a), (((char *) h + h->size) - ptr) | PREV_INUSE);
732
733 LIBC_PROBE (memory_arena_new, 2, a, size);
734 mstate replaced_arena = thread_arena;
735 thread_arena = a;
736 __libc_lock_init (a->mutex);
737
738 __libc_lock_lock (list_lock);
739
740 /* Add the new arena to the global list. */
741 a->next = main_arena.next;
742 /* FIXME: The barrier is an attempt to synchronize with read access
743 in reused_arena, which does not acquire list_lock while
744 traversing the list. */
745 atomic_write_barrier ();
746 main_arena.next = a;
747
748 __libc_lock_unlock (list_lock);
749
750 __libc_lock_lock (free_list_lock);
751 detach_arena (replaced_arena);
752 __libc_lock_unlock (free_list_lock);
753
754 /* Lock this arena. NB: Another thread may have been attached to
755 this arena because the arena is now accessible from the
756 main_arena.next list and could have been picked by reused_arena.
757 This can only happen for the last arena created (before the arena
758 limit is reached). At this point, some arena has to be attached
759 to two threads. We could acquire the arena lock before list_lock
760 to make it less likely that reused_arena picks this new arena,
761 but this could result in a deadlock with
762 __malloc_fork_lock_parent. */
763
764 __libc_lock_lock (a->mutex);
765
766 return a;
767}
768
769
770/* Remove an arena from free_list. */
771static mstate
772get_free_list (void)
773{
774 mstate replaced_arena = thread_arena;
775 mstate result = free_list;
776 if (result != NULL)
777 {
778 __libc_lock_lock (free_list_lock);
779 result = free_list;
780 if (result != NULL)
781 {
782 free_list = result->next_free;
783
784 /* The arena will be attached to this thread. */
785 assert (result->attached_threads == 0);
786 result->attached_threads = 1;
787
788 detach_arena (replaced_arena);
789 }
790 __libc_lock_unlock (free_list_lock);
791
792 if (result != NULL)
793 {
794 LIBC_PROBE (memory_arena_reuse_free_list, 1, result);
795 __libc_lock_lock (result->mutex);
796 thread_arena = result;
797 }
798 }
799
800 return result;
801}
802
803/* Remove the arena from the free list (if it is present).
804 free_list_lock must have been acquired by the caller. */
805static void
806remove_from_free_list (mstate arena)
807{
808 mstate *previous = &free_list;
809 for (mstate p = free_list; p != NULL; p = p->next_free)
810 {
811 assert (p->attached_threads == 0);
812 if (p == arena)
813 {
814 /* Remove the requested arena from the list. */
815 *previous = p->next_free;
816 break;
817 }
818 else
819 previous = &p->next_free;
820 }
821}
822
823/* Lock and return an arena that can be reused for memory allocation.
824 Avoid AVOID_ARENA as we have already failed to allocate memory in
825 it and it is currently locked. */
826static mstate
827reused_arena (mstate avoid_arena)
828{
829 mstate result;
830 /* FIXME: Access to next_to_use suffers from data races. */
831 static mstate next_to_use;
832 if (next_to_use == NULL)
833 next_to_use = &main_arena;
834
835 /* Iterate over all arenas (including those linked from
836 free_list). */
837 result = next_to_use;
838 do
839 {
840 if (!arena_is_corrupt (result) && !__libc_lock_trylock (result->mutex))
841 goto out;
842
843 /* FIXME: This is a data race, see _int_new_arena. */
844 result = result->next;
845 }
846 while (result != next_to_use);
847
848 /* Avoid AVOID_ARENA as we have already failed to allocate memory
849 in that arena and it is currently locked. */
850 if (result == avoid_arena)
851 result = result->next;
852
853 /* Make sure that the arena we get is not corrupted. */
854 mstate begin = result;
855 while (arena_is_corrupt (result) || result == avoid_arena)
856 {
857 result = result->next;
858 if (result == begin)
859 /* We looped around the arena list. We could not find any
860 arena that was either not corrupted or not the one we
861 wanted to avoid. */
862 return NULL;
863 }
864
865 /* No arena available without contention. Wait for the next in line. */
866 LIBC_PROBE (memory_arena_reuse_wait, 3, &result->mutex, result, avoid_arena);
867 __libc_lock_lock (result->mutex);
868
869out:
870 /* Attach the arena to the current thread. */
871 {
872 /* Update the arena thread attachment counters. */
873 mstate replaced_arena = thread_arena;
874 __libc_lock_lock (free_list_lock);
875 detach_arena (replaced_arena);
876
877 /* We may have picked up an arena on the free list. We need to
878 preserve the invariant that no arena on the free list has a
879 positive attached_threads counter (otherwise,
880 arena_thread_freeres cannot use the counter to determine if the
881 arena needs to be put on the free list). We unconditionally
882 remove the selected arena from the free list. The caller of
883 reused_arena checked the free list and observed it to be empty,
884 so the list is very short. */
885 remove_from_free_list (result);
886
887 ++result->attached_threads;
888
889 __libc_lock_unlock (free_list_lock);
890 }
891
892 LIBC_PROBE (memory_arena_reuse, 2, result, avoid_arena);
893 thread_arena = result;
894 next_to_use = result->next;
895
896 return result;
897}
898
899static mstate
900internal_function
901arena_get2 (size_t size, mstate avoid_arena)
902{
903 mstate a;
904
905 static size_t narenas_limit;
906
907 a = get_free_list ();
908 if (a == NULL)
909 {
910 /* Nothing immediately available, so generate a new arena. */
911 if (narenas_limit == 0)
912 {
913 if (mp_.arena_max != 0)
914 narenas_limit = mp_.arena_max;
915 else if (narenas > mp_.arena_test)
916 {
917 int n = __get_nprocs ();
918
919 if (n >= 1)
920 narenas_limit = NARENAS_FROM_NCORES (n);
921 else
922 /* We have no information about the system. Assume two
923 cores. */
924 narenas_limit = NARENAS_FROM_NCORES (2);
925 }
926 }
927 repeat:;
928 size_t n = narenas;
929 /* NB: the following depends on the fact that (size_t)0 - 1 is a
930 very large number and that the underflow is OK. If arena_max
931 is set the value of arena_test is irrelevant. If arena_test
932 is set but narenas is not yet larger or equal to arena_test
933 narenas_limit is 0. There is no possibility for narenas to
934 be too big for the test to always fail since there is not
935 enough address space to create that many arenas. */
936 if (__glibc_unlikely (n <= narenas_limit - 1))
937 {
938 if (catomic_compare_and_exchange_bool_acq (&narenas, n + 1, n))
939 goto repeat;
940 a = _int_new_arena (size);
941 if (__glibc_unlikely (a == NULL))
942 catomic_decrement (&narenas);
943 }
944 else
945 a = reused_arena (avoid_arena);
946 }
947 return a;
948}
949
950/* If we don't have the main arena, then maybe the failure is due to running
951 out of mmapped areas, so we can try allocating on the main arena.
952 Otherwise, it is likely that sbrk() has failed and there is still a chance
953 to mmap(), so try one of the other arenas. */
954static mstate
955arena_get_retry (mstate ar_ptr, size_t bytes)
956{
957 LIBC_PROBE (memory_arena_retry, 2, bytes, ar_ptr);
958 if (ar_ptr != &main_arena)
959 {
960 __libc_lock_unlock (ar_ptr->mutex);
961 /* Don't touch the main arena if it is corrupt. */
962 if (arena_is_corrupt (&main_arena))
963 return NULL;
964
965 ar_ptr = &main_arena;
966 __libc_lock_lock (ar_ptr->mutex);
967 }
968 else
969 {
970 __libc_lock_unlock (ar_ptr->mutex);
971 ar_ptr = arena_get2 (bytes, ar_ptr);
972 }
973
974 return ar_ptr;
975}
976
977static void __attribute__ ((section ("__libc_thread_freeres_fn")))
978arena_thread_freeres (void)
979{
980 mstate a = thread_arena;
981 thread_arena = NULL;
982
983 if (a != NULL)
984 {
985 __libc_lock_lock (free_list_lock);
986 /* If this was the last attached thread for this arena, put the
987 arena on the free list. */
988 assert (a->attached_threads > 0);
989 if (--a->attached_threads == 0)
990 {
991 a->next_free = free_list;
992 free_list = a;
993 }
994 __libc_lock_unlock (free_list_lock);
995 }
996}
997text_set_element (__libc_thread_subfreeres, arena_thread_freeres);
998
999/*
1000 * Local variables:
1001 * c-basic-offset: 2
1002 * End:
1003 */
1004