1/* Malloc implementation for multiple threads without lock contention.
2 Copyright (C) 1996-2018 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4 Contributed by Wolfram Gloger <wg@malloc.de>
5 and Doug Lea <dl@cs.oswego.edu>, 2001.
6
7 The GNU C Library is free software; you can redistribute it and/or
8 modify it under the terms of the GNU Lesser General Public License as
9 published by the Free Software Foundation; either version 2.1 of the
10 License, or (at your option) any later version.
11
12 The GNU C Library is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
16
17 You should have received a copy of the GNU Lesser General Public
18 License along with the GNU C Library; see the file COPYING.LIB. If
19 not, see <http://www.gnu.org/licenses/>. */
20
21/*
22 This is a version (aka ptmalloc2) of malloc/free/realloc written by
23 Doug Lea and adapted to multiple threads/arenas by Wolfram Gloger.
24
25 There have been substantial changes made after the integration into
26 glibc in all parts of the code. Do not look for much commonality
27 with the ptmalloc2 version.
28
29* Version ptmalloc2-20011215
30 based on:
31 VERSION 2.7.0 Sun Mar 11 14:14:06 2001 Doug Lea (dl at gee)
32
33* Quickstart
34
35 In order to compile this implementation, a Makefile is provided with
36 the ptmalloc2 distribution, which has pre-defined targets for some
37 popular systems (e.g. "make posix" for Posix threads). All that is
38 typically required with regard to compiler flags is the selection of
39 the thread package via defining one out of USE_PTHREADS, USE_THR or
40 USE_SPROC. Check the thread-m.h file for what effects this has.
41 Many/most systems will additionally require USE_TSD_DATA_HACK to be
42 defined, so this is the default for "make posix".
43
44* Why use this malloc?
45
46 This is not the fastest, most space-conserving, most portable, or
47 most tunable malloc ever written. However it is among the fastest
48 while also being among the most space-conserving, portable and tunable.
49 Consistent balance across these factors results in a good general-purpose
50 allocator for malloc-intensive programs.
51
52 The main properties of the algorithms are:
53 * For large (>= 512 bytes) requests, it is a pure best-fit allocator,
54 with ties normally decided via FIFO (i.e. least recently used).
55 * For small (<= 64 bytes by default) requests, it is a caching
56 allocator, that maintains pools of quickly recycled chunks.
57 * In between, and for combinations of large and small requests, it does
58 the best it can trying to meet both goals at once.
59 * For very large requests (>= 128KB by default), it relies on system
60 memory mapping facilities, if supported.
61
62 For a longer but slightly out of date high-level description, see
63 http://gee.cs.oswego.edu/dl/html/malloc.html
64
65 You may already by default be using a C library containing a malloc
66 that is based on some version of this malloc (for example in
67 linux). You might still want to use the one in this file in order to
68 customize settings or to avoid overheads associated with library
69 versions.
70
71* Contents, described in more detail in "description of public routines" below.
72
73 Standard (ANSI/SVID/...) functions:
74 malloc(size_t n);
75 calloc(size_t n_elements, size_t element_size);
76 free(void* p);
77 realloc(void* p, size_t n);
78 memalign(size_t alignment, size_t n);
79 valloc(size_t n);
80 mallinfo()
81 mallopt(int parameter_number, int parameter_value)
82
83 Additional functions:
84 independent_calloc(size_t n_elements, size_t size, void* chunks[]);
85 independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]);
86 pvalloc(size_t n);
87 malloc_trim(size_t pad);
88 malloc_usable_size(void* p);
89 malloc_stats();
90
91* Vital statistics:
92
93 Supported pointer representation: 4 or 8 bytes
94 Supported size_t representation: 4 or 8 bytes
95 Note that size_t is allowed to be 4 bytes even if pointers are 8.
96 You can adjust this by defining INTERNAL_SIZE_T
97
98 Alignment: 2 * sizeof(size_t) (default)
99 (i.e., 8 byte alignment with 4byte size_t). This suffices for
100 nearly all current machines and C compilers. However, you can
101 define MALLOC_ALIGNMENT to be wider than this if necessary.
102
103 Minimum overhead per allocated chunk: 4 or 8 bytes
104 Each malloced chunk has a hidden word of overhead holding size
105 and status information.
106
107 Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead)
108 8-byte ptrs: 24/32 bytes (including, 4/8 overhead)
109
110 When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte
111 ptrs but 4 byte size) or 24 (for 8/8) additional bytes are
112 needed; 4 (8) for a trailing size field and 8 (16) bytes for
113 free list pointers. Thus, the minimum allocatable size is
114 16/24/32 bytes.
115
116 Even a request for zero bytes (i.e., malloc(0)) returns a
117 pointer to something of the minimum allocatable size.
118
119 The maximum overhead wastage (i.e., number of extra bytes
120 allocated than were requested in malloc) is less than or equal
121 to the minimum size, except for requests >= mmap_threshold that
122 are serviced via mmap(), where the worst case wastage is 2 *
123 sizeof(size_t) bytes plus the remainder from a system page (the
124 minimal mmap unit); typically 4096 or 8192 bytes.
125
126 Maximum allocated size: 4-byte size_t: 2^32 minus about two pages
127 8-byte size_t: 2^64 minus about two pages
128
129 It is assumed that (possibly signed) size_t values suffice to
130 represent chunk sizes. `Possibly signed' is due to the fact
131 that `size_t' may be defined on a system as either a signed or
132 an unsigned type. The ISO C standard says that it must be
133 unsigned, but a few systems are known not to adhere to this.
134 Additionally, even when size_t is unsigned, sbrk (which is by
135 default used to obtain memory from system) accepts signed
136 arguments, and may not be able to handle size_t-wide arguments
137 with negative sign bit. Generally, values that would
138 appear as negative after accounting for overhead and alignment
139 are supported only via mmap(), which does not have this
140 limitation.
141
142 Requests for sizes outside the allowed range will perform an optional
143 failure action and then return null. (Requests may also
144 also fail because a system is out of memory.)
145
146 Thread-safety: thread-safe
147
148 Compliance: I believe it is compliant with the 1997 Single Unix Specification
149 Also SVID/XPG, ANSI C, and probably others as well.
150
151* Synopsis of compile-time options:
152
153 People have reported using previous versions of this malloc on all
154 versions of Unix, sometimes by tweaking some of the defines
155 below. It has been tested most extensively on Solaris and Linux.
156 People also report using it in stand-alone embedded systems.
157
158 The implementation is in straight, hand-tuned ANSI C. It is not
159 at all modular. (Sorry!) It uses a lot of macros. To be at all
160 usable, this code should be compiled using an optimizing compiler
161 (for example gcc -O3) that can simplify expressions and control
162 paths. (FAQ: some macros import variables as arguments rather than
163 declare locals because people reported that some debuggers
164 otherwise get confused.)
165
166 OPTION DEFAULT VALUE
167
168 Compilation Environment options:
169
170 HAVE_MREMAP 0
171
172 Changing default word sizes:
173
174 INTERNAL_SIZE_T size_t
175
176 Configuration and functionality options:
177
178 USE_PUBLIC_MALLOC_WRAPPERS NOT defined
179 USE_MALLOC_LOCK NOT defined
180 MALLOC_DEBUG NOT defined
181 REALLOC_ZERO_BYTES_FREES 1
182 TRIM_FASTBINS 0
183
184 Options for customizing MORECORE:
185
186 MORECORE sbrk
187 MORECORE_FAILURE -1
188 MORECORE_CONTIGUOUS 1
189 MORECORE_CANNOT_TRIM NOT defined
190 MORECORE_CLEARS 1
191 MMAP_AS_MORECORE_SIZE (1024 * 1024)
192
193 Tuning options that are also dynamically changeable via mallopt:
194
195 DEFAULT_MXFAST 64 (for 32bit), 128 (for 64bit)
196 DEFAULT_TRIM_THRESHOLD 128 * 1024
197 DEFAULT_TOP_PAD 0
198 DEFAULT_MMAP_THRESHOLD 128 * 1024
199 DEFAULT_MMAP_MAX 65536
200
201 There are several other #defined constants and macros that you
202 probably don't want to touch unless you are extending or adapting malloc. */
203
204/*
205 void* is the pointer type that malloc should say it returns
206*/
207
208#ifndef void
209#define void void
210#endif /*void*/
211
212#include <stddef.h> /* for size_t */
213#include <stdlib.h> /* for getenv(), abort() */
214#include <unistd.h> /* for __libc_enable_secure */
215
216#include <atomic.h>
217#include <_itoa.h>
218#include <bits/wordsize.h>
219#include <sys/sysinfo.h>
220
221#include <ldsodefs.h>
222
223#include <unistd.h>
224#include <stdio.h> /* needed for malloc_stats */
225#include <errno.h>
226#include <assert.h>
227
228#include <shlib-compat.h>
229
230/* For uintptr_t. */
231#include <stdint.h>
232
233/* For va_arg, va_start, va_end. */
234#include <stdarg.h>
235
236/* For MIN, MAX, powerof2. */
237#include <sys/param.h>
238
239/* For ALIGN_UP et. al. */
240#include <libc-pointer-arith.h>
241
242/* For DIAG_PUSH/POP_NEEDS_COMMENT et al. */
243#include <libc-diag.h>
244
245#include <malloc/malloc-internal.h>
246
247/* For SINGLE_THREAD_P. */
248#include <sysdep-cancel.h>
249
250/*
251 Debugging:
252
253 Because freed chunks may be overwritten with bookkeeping fields, this
254 malloc will often die when freed memory is overwritten by user
255 programs. This can be very effective (albeit in an annoying way)
256 in helping track down dangling pointers.
257
258 If you compile with -DMALLOC_DEBUG, a number of assertion checks are
259 enabled that will catch more memory errors. You probably won't be
260 able to make much sense of the actual assertion errors, but they
261 should help you locate incorrectly overwritten memory. The checking
262 is fairly extensive, and will slow down execution
263 noticeably. Calling malloc_stats or mallinfo with MALLOC_DEBUG set
264 will attempt to check every non-mmapped allocated and free chunk in
265 the course of computing the summmaries. (By nature, mmapped regions
266 cannot be checked very much automatically.)
267
268 Setting MALLOC_DEBUG may also be helpful if you are trying to modify
269 this code. The assertions in the check routines spell out in more
270 detail the assumptions and invariants underlying the algorithms.
271
272 Setting MALLOC_DEBUG does NOT provide an automated mechanism for
273 checking that all accesses to malloced memory stay within their
274 bounds. However, there are several add-ons and adaptations of this
275 or other mallocs available that do this.
276*/
277
278#ifndef MALLOC_DEBUG
279#define MALLOC_DEBUG 0
280#endif
281
282#ifndef NDEBUG
283# define __assert_fail(assertion, file, line, function) \
284 __malloc_assert(assertion, file, line, function)
285
286extern const char *__progname;
287
288static void
289__malloc_assert (const char *assertion, const char *file, unsigned int line,
290 const char *function)
291{
292 (void) __fxprintf (NULL, "%s%s%s:%u: %s%sAssertion `%s' failed.\n",
293 __progname, __progname[0] ? ": " : "",
294 file, line,
295 function ? function : "", function ? ": " : "",
296 assertion);
297 fflush (stderr);
298 abort ();
299}
300#endif
301
302#if USE_TCACHE
303/* We want 64 entries. This is an arbitrary limit, which tunables can reduce. */
304# define TCACHE_MAX_BINS 64
305# define MAX_TCACHE_SIZE tidx2usize (TCACHE_MAX_BINS-1)
306
307/* Only used to pre-fill the tunables. */
308# define tidx2usize(idx) (((size_t) idx) * MALLOC_ALIGNMENT + MINSIZE - SIZE_SZ)
309
310/* When "x" is from chunksize(). */
311# define csize2tidx(x) (((x) - MINSIZE + MALLOC_ALIGNMENT - 1) / MALLOC_ALIGNMENT)
312/* When "x" is a user-provided size. */
313# define usize2tidx(x) csize2tidx (request2size (x))
314
315/* With rounding and alignment, the bins are...
316 idx 0 bytes 0..24 (64-bit) or 0..12 (32-bit)
317 idx 1 bytes 25..40 or 13..20
318 idx 2 bytes 41..56 or 21..28
319 etc. */
320
321/* This is another arbitrary limit, which tunables can change. Each
322 tcache bin will hold at most this number of chunks. */
323# define TCACHE_FILL_COUNT 7
324#endif
325
326
327/*
328 REALLOC_ZERO_BYTES_FREES should be set if a call to
329 realloc with zero bytes should be the same as a call to free.
330 This is required by the C standard. Otherwise, since this malloc
331 returns a unique pointer for malloc(0), so does realloc(p, 0).
332*/
333
334#ifndef REALLOC_ZERO_BYTES_FREES
335#define REALLOC_ZERO_BYTES_FREES 1
336#endif
337
338/*
339 TRIM_FASTBINS controls whether free() of a very small chunk can
340 immediately lead to trimming. Setting to true (1) can reduce memory
341 footprint, but will almost always slow down programs that use a lot
342 of small chunks.
343
344 Define this only if you are willing to give up some speed to more
345 aggressively reduce system-level memory footprint when releasing
346 memory in programs that use many small chunks. You can get
347 essentially the same effect by setting MXFAST to 0, but this can
348 lead to even greater slowdowns in programs using many small chunks.
349 TRIM_FASTBINS is an in-between compile-time option, that disables
350 only those chunks bordering topmost memory from being placed in
351 fastbins.
352*/
353
354#ifndef TRIM_FASTBINS
355#define TRIM_FASTBINS 0
356#endif
357
358
359/* Definition for getting more memory from the OS. */
360#define MORECORE (*__morecore)
361#define MORECORE_FAILURE 0
362void * __default_morecore (ptrdiff_t);
363void *(*__morecore)(ptrdiff_t) = __default_morecore;
364
365
366#include <string.h>
367
368/*
369 MORECORE-related declarations. By default, rely on sbrk
370*/
371
372
373/*
374 MORECORE is the name of the routine to call to obtain more memory
375 from the system. See below for general guidance on writing
376 alternative MORECORE functions, as well as a version for WIN32 and a
377 sample version for pre-OSX macos.
378*/
379
380#ifndef MORECORE
381#define MORECORE sbrk
382#endif
383
384/*
385 MORECORE_FAILURE is the value returned upon failure of MORECORE
386 as well as mmap. Since it cannot be an otherwise valid memory address,
387 and must reflect values of standard sys calls, you probably ought not
388 try to redefine it.
389*/
390
391#ifndef MORECORE_FAILURE
392#define MORECORE_FAILURE (-1)
393#endif
394
395/*
396 If MORECORE_CONTIGUOUS is true, take advantage of fact that
397 consecutive calls to MORECORE with positive arguments always return
398 contiguous increasing addresses. This is true of unix sbrk. Even
399 if not defined, when regions happen to be contiguous, malloc will
400 permit allocations spanning regions obtained from different
401 calls. But defining this when applicable enables some stronger
402 consistency checks and space efficiencies.
403*/
404
405#ifndef MORECORE_CONTIGUOUS
406#define MORECORE_CONTIGUOUS 1
407#endif
408
409/*
410 Define MORECORE_CANNOT_TRIM if your version of MORECORE
411 cannot release space back to the system when given negative
412 arguments. This is generally necessary only if you are using
413 a hand-crafted MORECORE function that cannot handle negative arguments.
414*/
415
416/* #define MORECORE_CANNOT_TRIM */
417
418/* MORECORE_CLEARS (default 1)
419 The degree to which the routine mapped to MORECORE zeroes out
420 memory: never (0), only for newly allocated space (1) or always
421 (2). The distinction between (1) and (2) is necessary because on
422 some systems, if the application first decrements and then
423 increments the break value, the contents of the reallocated space
424 are unspecified.
425 */
426
427#ifndef MORECORE_CLEARS
428# define MORECORE_CLEARS 1
429#endif
430
431
432/*
433 MMAP_AS_MORECORE_SIZE is the minimum mmap size argument to use if
434 sbrk fails, and mmap is used as a backup. The value must be a
435 multiple of page size. This backup strategy generally applies only
436 when systems have "holes" in address space, so sbrk cannot perform
437 contiguous expansion, but there is still space available on system.
438 On systems for which this is known to be useful (i.e. most linux
439 kernels), this occurs only when programs allocate huge amounts of
440 memory. Between this, and the fact that mmap regions tend to be
441 limited, the size should be large, to avoid too many mmap calls and
442 thus avoid running out of kernel resources. */
443
444#ifndef MMAP_AS_MORECORE_SIZE
445#define MMAP_AS_MORECORE_SIZE (1024 * 1024)
446#endif
447
448/*
449 Define HAVE_MREMAP to make realloc() use mremap() to re-allocate
450 large blocks.
451*/
452
453#ifndef HAVE_MREMAP
454#define HAVE_MREMAP 0
455#endif
456
457/* We may need to support __malloc_initialize_hook for backwards
458 compatibility. */
459
460#if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_24)
461# define HAVE_MALLOC_INIT_HOOK 1
462#else
463# define HAVE_MALLOC_INIT_HOOK 0
464#endif
465
466
467/*
468 This version of malloc supports the standard SVID/XPG mallinfo
469 routine that returns a struct containing usage properties and
470 statistics. It should work on any SVID/XPG compliant system that has
471 a /usr/include/malloc.h defining struct mallinfo. (If you'd like to
472 install such a thing yourself, cut out the preliminary declarations
473 as described above and below and save them in a malloc.h file. But
474 there's no compelling reason to bother to do this.)
475
476 The main declaration needed is the mallinfo struct that is returned
477 (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a
478 bunch of fields that are not even meaningful in this version of
479 malloc. These fields are are instead filled by mallinfo() with
480 other numbers that might be of interest.
481*/
482
483
484/* ---------- description of public routines ------------ */
485
486/*
487 malloc(size_t n)
488 Returns a pointer to a newly allocated chunk of at least n bytes, or null
489 if no space is available. Additionally, on failure, errno is
490 set to ENOMEM on ANSI C systems.
491
492 If n is zero, malloc returns a minumum-sized chunk. (The minimum
493 size is 16 bytes on most 32bit systems, and 24 or 32 bytes on 64bit
494 systems.) On most systems, size_t is an unsigned type, so calls
495 with negative arguments are interpreted as requests for huge amounts
496 of space, which will often fail. The maximum supported value of n
497 differs across systems, but is in all cases less than the maximum
498 representable value of a size_t.
499*/
500void* __libc_malloc(size_t);
501libc_hidden_proto (__libc_malloc)
502
503/*
504 free(void* p)
505 Releases the chunk of memory pointed to by p, that had been previously
506 allocated using malloc or a related routine such as realloc.
507 It has no effect if p is null. It can have arbitrary (i.e., bad!)
508 effects if p has already been freed.
509
510 Unless disabled (using mallopt), freeing very large spaces will
511 when possible, automatically trigger operations that give
512 back unused memory to the system, thus reducing program footprint.
513*/
514void __libc_free(void*);
515libc_hidden_proto (__libc_free)
516
517/*
518 calloc(size_t n_elements, size_t element_size);
519 Returns a pointer to n_elements * element_size bytes, with all locations
520 set to zero.
521*/
522void* __libc_calloc(size_t, size_t);
523
524/*
525 realloc(void* p, size_t n)
526 Returns a pointer to a chunk of size n that contains the same data
527 as does chunk p up to the minimum of (n, p's size) bytes, or null
528 if no space is available.
529
530 The returned pointer may or may not be the same as p. The algorithm
531 prefers extending p when possible, otherwise it employs the
532 equivalent of a malloc-copy-free sequence.
533
534 If p is null, realloc is equivalent to malloc.
535
536 If space is not available, realloc returns null, errno is set (if on
537 ANSI) and p is NOT freed.
538
539 if n is for fewer bytes than already held by p, the newly unused
540 space is lopped off and freed if possible. Unless the #define
541 REALLOC_ZERO_BYTES_FREES is set, realloc with a size argument of
542 zero (re)allocates a minimum-sized chunk.
543
544 Large chunks that were internally obtained via mmap will always be
545 grown using malloc-copy-free sequences unless the system supports
546 MREMAP (currently only linux).
547
548 The old unix realloc convention of allowing the last-free'd chunk
549 to be used as an argument to realloc is not supported.
550*/
551void* __libc_realloc(void*, size_t);
552libc_hidden_proto (__libc_realloc)
553
554/*
555 memalign(size_t alignment, size_t n);
556 Returns a pointer to a newly allocated chunk of n bytes, aligned
557 in accord with the alignment argument.
558
559 The alignment argument should be a power of two. If the argument is
560 not a power of two, the nearest greater power is used.
561 8-byte alignment is guaranteed by normal malloc calls, so don't
562 bother calling memalign with an argument of 8 or less.
563
564 Overreliance on memalign is a sure way to fragment space.
565*/
566void* __libc_memalign(size_t, size_t);
567libc_hidden_proto (__libc_memalign)
568
569/*
570 valloc(size_t n);
571 Equivalent to memalign(pagesize, n), where pagesize is the page
572 size of the system. If the pagesize is unknown, 4096 is used.
573*/
574void* __libc_valloc(size_t);
575
576
577
578/*
579 mallopt(int parameter_number, int parameter_value)
580 Sets tunable parameters The format is to provide a
581 (parameter-number, parameter-value) pair. mallopt then sets the
582 corresponding parameter to the argument value if it can (i.e., so
583 long as the value is meaningful), and returns 1 if successful else
584 0. SVID/XPG/ANSI defines four standard param numbers for mallopt,
585 normally defined in malloc.h. Only one of these (M_MXFAST) is used
586 in this malloc. The others (M_NLBLKS, M_GRAIN, M_KEEP) don't apply,
587 so setting them has no effect. But this malloc also supports four
588 other options in mallopt. See below for details. Briefly, supported
589 parameters are as follows (listed defaults are for "typical"
590 configurations).
591
592 Symbol param # default allowed param values
593 M_MXFAST 1 64 0-80 (0 disables fastbins)
594 M_TRIM_THRESHOLD -1 128*1024 any (-1U disables trimming)
595 M_TOP_PAD -2 0 any
596 M_MMAP_THRESHOLD -3 128*1024 any (or 0 if no MMAP support)
597 M_MMAP_MAX -4 65536 any (0 disables use of mmap)
598*/
599int __libc_mallopt(int, int);
600libc_hidden_proto (__libc_mallopt)
601
602
603/*
604 mallinfo()
605 Returns (by copy) a struct containing various summary statistics:
606
607 arena: current total non-mmapped bytes allocated from system
608 ordblks: the number of free chunks
609 smblks: the number of fastbin blocks (i.e., small chunks that
610 have been freed but not use resused or consolidated)
611 hblks: current number of mmapped regions
612 hblkhd: total bytes held in mmapped regions
613 usmblks: always 0
614 fsmblks: total bytes held in fastbin blocks
615 uordblks: current total allocated space (normal or mmapped)
616 fordblks: total free space
617 keepcost: the maximum number of bytes that could ideally be released
618 back to system via malloc_trim. ("ideally" means that
619 it ignores page restrictions etc.)
620
621 Because these fields are ints, but internal bookkeeping may
622 be kept as longs, the reported values may wrap around zero and
623 thus be inaccurate.
624*/
625struct mallinfo __libc_mallinfo(void);
626
627
628/*
629 pvalloc(size_t n);
630 Equivalent to valloc(minimum-page-that-holds(n)), that is,
631 round up n to nearest pagesize.
632 */
633void* __libc_pvalloc(size_t);
634
635/*
636 malloc_trim(size_t pad);
637
638 If possible, gives memory back to the system (via negative
639 arguments to sbrk) if there is unused memory at the `high' end of
640 the malloc pool. You can call this after freeing large blocks of
641 memory to potentially reduce the system-level memory requirements
642 of a program. However, it cannot guarantee to reduce memory. Under
643 some allocation patterns, some large free blocks of memory will be
644 locked between two used chunks, so they cannot be given back to
645 the system.
646
647 The `pad' argument to malloc_trim represents the amount of free
648 trailing space to leave untrimmed. If this argument is zero,
649 only the minimum amount of memory to maintain internal data
650 structures will be left (one page or less). Non-zero arguments
651 can be supplied to maintain enough trailing space to service
652 future expected allocations without having to re-obtain memory
653 from the system.
654
655 Malloc_trim returns 1 if it actually released any memory, else 0.
656 On systems that do not support "negative sbrks", it will always
657 return 0.
658*/
659int __malloc_trim(size_t);
660
661/*
662 malloc_usable_size(void* p);
663
664 Returns the number of bytes you can actually use in
665 an allocated chunk, which may be more than you requested (although
666 often not) due to alignment and minimum size constraints.
667 You can use this many bytes without worrying about
668 overwriting other allocated objects. This is not a particularly great
669 programming practice. malloc_usable_size can be more useful in
670 debugging and assertions, for example:
671
672 p = malloc(n);
673 assert(malloc_usable_size(p) >= 256);
674
675*/
676size_t __malloc_usable_size(void*);
677
678/*
679 malloc_stats();
680 Prints on stderr the amount of space obtained from the system (both
681 via sbrk and mmap), the maximum amount (which may be more than
682 current if malloc_trim and/or munmap got called), and the current
683 number of bytes allocated via malloc (or realloc, etc) but not yet
684 freed. Note that this is the number of bytes allocated, not the
685 number requested. It will be larger than the number requested
686 because of alignment and bookkeeping overhead. Because it includes
687 alignment wastage as being in use, this figure may be greater than
688 zero even when no user-level chunks are allocated.
689
690 The reported current and maximum system memory can be inaccurate if
691 a program makes other calls to system memory allocation functions
692 (normally sbrk) outside of malloc.
693
694 malloc_stats prints only the most commonly interesting statistics.
695 More information can be obtained by calling mallinfo.
696
697*/
698void __malloc_stats(void);
699
700/*
701 malloc_get_state(void);
702
703 Returns the state of all malloc variables in an opaque data
704 structure.
705*/
706void* __malloc_get_state(void);
707
708/*
709 malloc_set_state(void* state);
710
711 Restore the state of all malloc variables from data obtained with
712 malloc_get_state().
713*/
714int __malloc_set_state(void*);
715
716/*
717 posix_memalign(void **memptr, size_t alignment, size_t size);
718
719 POSIX wrapper like memalign(), checking for validity of size.
720*/
721int __posix_memalign(void **, size_t, size_t);
722
723/* mallopt tuning options */
724
725/*
726 M_MXFAST is the maximum request size used for "fastbins", special bins
727 that hold returned chunks without consolidating their spaces. This
728 enables future requests for chunks of the same size to be handled
729 very quickly, but can increase fragmentation, and thus increase the
730 overall memory footprint of a program.
731
732 This malloc manages fastbins very conservatively yet still
733 efficiently, so fragmentation is rarely a problem for values less
734 than or equal to the default. The maximum supported value of MXFAST
735 is 80. You wouldn't want it any higher than this anyway. Fastbins
736 are designed especially for use with many small structs, objects or
737 strings -- the default handles structs/objects/arrays with sizes up
738 to 8 4byte fields, or small strings representing words, tokens,
739 etc. Using fastbins for larger objects normally worsens
740 fragmentation without improving speed.
741
742 M_MXFAST is set in REQUEST size units. It is internally used in
743 chunksize units, which adds padding and alignment. You can reduce
744 M_MXFAST to 0 to disable all use of fastbins. This causes the malloc
745 algorithm to be a closer approximation of fifo-best-fit in all cases,
746 not just for larger requests, but will generally cause it to be
747 slower.
748*/
749
750
751/* M_MXFAST is a standard SVID/XPG tuning option, usually listed in malloc.h */
752#ifndef M_MXFAST
753#define M_MXFAST 1
754#endif
755
756#ifndef DEFAULT_MXFAST
757#define DEFAULT_MXFAST (64 * SIZE_SZ / 4)
758#endif
759
760
761/*
762 M_TRIM_THRESHOLD is the maximum amount of unused top-most memory
763 to keep before releasing via malloc_trim in free().
764
765 Automatic trimming is mainly useful in long-lived programs.
766 Because trimming via sbrk can be slow on some systems, and can
767 sometimes be wasteful (in cases where programs immediately
768 afterward allocate more large chunks) the value should be high
769 enough so that your overall system performance would improve by
770 releasing this much memory.
771
772 The trim threshold and the mmap control parameters (see below)
773 can be traded off with one another. Trimming and mmapping are
774 two different ways of releasing unused memory back to the
775 system. Between these two, it is often possible to keep
776 system-level demands of a long-lived program down to a bare
777 minimum. For example, in one test suite of sessions measuring
778 the XF86 X server on Linux, using a trim threshold of 128K and a
779 mmap threshold of 192K led to near-minimal long term resource
780 consumption.
781
782 If you are using this malloc in a long-lived program, it should
783 pay to experiment with these values. As a rough guide, you
784 might set to a value close to the average size of a process
785 (program) running on your system. Releasing this much memory
786 would allow such a process to run in memory. Generally, it's
787 worth it to tune for trimming rather tham memory mapping when a
788 program undergoes phases where several large chunks are
789 allocated and released in ways that can reuse each other's
790 storage, perhaps mixed with phases where there are no such
791 chunks at all. And in well-behaved long-lived programs,
792 controlling release of large blocks via trimming versus mapping
793 is usually faster.
794
795 However, in most programs, these parameters serve mainly as
796 protection against the system-level effects of carrying around
797 massive amounts of unneeded memory. Since frequent calls to
798 sbrk, mmap, and munmap otherwise degrade performance, the default
799 parameters are set to relatively high values that serve only as
800 safeguards.
801
802 The trim value It must be greater than page size to have any useful
803 effect. To disable trimming completely, you can set to
804 (unsigned long)(-1)
805
806 Trim settings interact with fastbin (MXFAST) settings: Unless
807 TRIM_FASTBINS is defined, automatic trimming never takes place upon
808 freeing a chunk with size less than or equal to MXFAST. Trimming is
809 instead delayed until subsequent freeing of larger chunks. However,
810 you can still force an attempted trim by calling malloc_trim.
811
812 Also, trimming is not generally possible in cases where
813 the main arena is obtained via mmap.
814
815 Note that the trick some people use of mallocing a huge space and
816 then freeing it at program startup, in an attempt to reserve system
817 memory, doesn't have the intended effect under automatic trimming,
818 since that memory will immediately be returned to the system.
819*/
820
821#define M_TRIM_THRESHOLD -1
822
823#ifndef DEFAULT_TRIM_THRESHOLD
824#define DEFAULT_TRIM_THRESHOLD (128 * 1024)
825#endif
826
827/*
828 M_TOP_PAD is the amount of extra `padding' space to allocate or
829 retain whenever sbrk is called. It is used in two ways internally:
830
831 * When sbrk is called to extend the top of the arena to satisfy
832 a new malloc request, this much padding is added to the sbrk
833 request.
834
835 * When malloc_trim is called automatically from free(),
836 it is used as the `pad' argument.
837
838 In both cases, the actual amount of padding is rounded
839 so that the end of the arena is always a system page boundary.
840
841 The main reason for using padding is to avoid calling sbrk so
842 often. Having even a small pad greatly reduces the likelihood
843 that nearly every malloc request during program start-up (or
844 after trimming) will invoke sbrk, which needlessly wastes
845 time.
846
847 Automatic rounding-up to page-size units is normally sufficient
848 to avoid measurable overhead, so the default is 0. However, in
849 systems where sbrk is relatively slow, it can pay to increase
850 this value, at the expense of carrying around more memory than
851 the program needs.
852*/
853
854#define M_TOP_PAD -2
855
856#ifndef DEFAULT_TOP_PAD
857#define DEFAULT_TOP_PAD (0)
858#endif
859
860/*
861 MMAP_THRESHOLD_MAX and _MIN are the bounds on the dynamically
862 adjusted MMAP_THRESHOLD.
863*/
864
865#ifndef DEFAULT_MMAP_THRESHOLD_MIN
866#define DEFAULT_MMAP_THRESHOLD_MIN (128 * 1024)
867#endif
868
869#ifndef DEFAULT_MMAP_THRESHOLD_MAX
870 /* For 32-bit platforms we cannot increase the maximum mmap
871 threshold much because it is also the minimum value for the
872 maximum heap size and its alignment. Going above 512k (i.e., 1M
873 for new heaps) wastes too much address space. */
874# if __WORDSIZE == 32
875# define DEFAULT_MMAP_THRESHOLD_MAX (512 * 1024)
876# else
877# define DEFAULT_MMAP_THRESHOLD_MAX (4 * 1024 * 1024 * sizeof(long))
878# endif
879#endif
880
881/*
882 M_MMAP_THRESHOLD is the request size threshold for using mmap()
883 to service a request. Requests of at least this size that cannot
884 be allocated using already-existing space will be serviced via mmap.
885 (If enough normal freed space already exists it is used instead.)
886
887 Using mmap segregates relatively large chunks of memory so that
888 they can be individually obtained and released from the host
889 system. A request serviced through mmap is never reused by any
890 other request (at least not directly; the system may just so
891 happen to remap successive requests to the same locations).
892
893 Segregating space in this way has the benefits that:
894
895 1. Mmapped space can ALWAYS be individually released back
896 to the system, which helps keep the system level memory
897 demands of a long-lived program low.
898 2. Mapped memory can never become `locked' between
899 other chunks, as can happen with normally allocated chunks, which
900 means that even trimming via malloc_trim would not release them.
901 3. On some systems with "holes" in address spaces, mmap can obtain
902 memory that sbrk cannot.
903
904 However, it has the disadvantages that:
905
906 1. The space cannot be reclaimed, consolidated, and then
907 used to service later requests, as happens with normal chunks.
908 2. It can lead to more wastage because of mmap page alignment
909 requirements
910 3. It causes malloc performance to be more dependent on host
911 system memory management support routines which may vary in
912 implementation quality and may impose arbitrary
913 limitations. Generally, servicing a request via normal
914 malloc steps is faster than going through a system's mmap.
915
916 The advantages of mmap nearly always outweigh disadvantages for
917 "large" chunks, but the value of "large" varies across systems. The
918 default is an empirically derived value that works well in most
919 systems.
920
921
922 Update in 2006:
923 The above was written in 2001. Since then the world has changed a lot.
924 Memory got bigger. Applications got bigger. The virtual address space
925 layout in 32 bit linux changed.
926
927 In the new situation, brk() and mmap space is shared and there are no
928 artificial limits on brk size imposed by the kernel. What is more,
929 applications have started using transient allocations larger than the
930 128Kb as was imagined in 2001.
931
932 The price for mmap is also high now; each time glibc mmaps from the
933 kernel, the kernel is forced to zero out the memory it gives to the
934 application. Zeroing memory is expensive and eats a lot of cache and
935 memory bandwidth. This has nothing to do with the efficiency of the
936 virtual memory system, by doing mmap the kernel just has no choice but
937 to zero.
938
939 In 2001, the kernel had a maximum size for brk() which was about 800
940 megabytes on 32 bit x86, at that point brk() would hit the first
941 mmaped shared libaries and couldn't expand anymore. With current 2.6
942 kernels, the VA space layout is different and brk() and mmap
943 both can span the entire heap at will.
944
945 Rather than using a static threshold for the brk/mmap tradeoff,
946 we are now using a simple dynamic one. The goal is still to avoid
947 fragmentation. The old goals we kept are
948 1) try to get the long lived large allocations to use mmap()
949 2) really large allocations should always use mmap()
950 and we're adding now:
951 3) transient allocations should use brk() to avoid forcing the kernel
952 having to zero memory over and over again
953
954 The implementation works with a sliding threshold, which is by default
955 limited to go between 128Kb and 32Mb (64Mb for 64 bitmachines) and starts
956 out at 128Kb as per the 2001 default.
957
958 This allows us to satisfy requirement 1) under the assumption that long
959 lived allocations are made early in the process' lifespan, before it has
960 started doing dynamic allocations of the same size (which will
961 increase the threshold).
962
963 The upperbound on the threshold satisfies requirement 2)
964
965 The threshold goes up in value when the application frees memory that was
966 allocated with the mmap allocator. The idea is that once the application
967 starts freeing memory of a certain size, it's highly probable that this is
968 a size the application uses for transient allocations. This estimator
969 is there to satisfy the new third requirement.
970
971*/
972
973#define M_MMAP_THRESHOLD -3
974
975#ifndef DEFAULT_MMAP_THRESHOLD
976#define DEFAULT_MMAP_THRESHOLD DEFAULT_MMAP_THRESHOLD_MIN
977#endif
978
979/*
980 M_MMAP_MAX is the maximum number of requests to simultaneously
981 service using mmap. This parameter exists because
982 some systems have a limited number of internal tables for
983 use by mmap, and using more than a few of them may degrade
984 performance.
985
986 The default is set to a value that serves only as a safeguard.
987 Setting to 0 disables use of mmap for servicing large requests.
988*/
989
990#define M_MMAP_MAX -4
991
992#ifndef DEFAULT_MMAP_MAX
993#define DEFAULT_MMAP_MAX (65536)
994#endif
995
996#include <malloc.h>
997
998#ifndef RETURN_ADDRESS
999#define RETURN_ADDRESS(X_) (NULL)
1000#endif
1001
1002/* Forward declarations. */
1003struct malloc_chunk;
1004typedef struct malloc_chunk* mchunkptr;
1005
1006/* Internal routines. */
1007
1008static void* _int_malloc(mstate, size_t);
1009static void _int_free(mstate, mchunkptr, int);
1010static void* _int_realloc(mstate, mchunkptr, INTERNAL_SIZE_T,
1011 INTERNAL_SIZE_T);
1012static void* _int_memalign(mstate, size_t, size_t);
1013static void* _mid_memalign(size_t, size_t, void *);
1014
1015static void malloc_printerr(const char *str) __attribute__ ((noreturn));
1016
1017static void* mem2mem_check(void *p, size_t sz);
1018static void top_check(void);
1019static void munmap_chunk(mchunkptr p);
1020#if HAVE_MREMAP
1021static mchunkptr mremap_chunk(mchunkptr p, size_t new_size);
1022#endif
1023
1024static void* malloc_check(size_t sz, const void *caller);
1025static void free_check(void* mem, const void *caller);
1026static void* realloc_check(void* oldmem, size_t bytes,
1027 const void *caller);
1028static void* memalign_check(size_t alignment, size_t bytes,
1029 const void *caller);
1030
1031/* ------------------ MMAP support ------------------ */
1032
1033
1034#include <fcntl.h>
1035#include <sys/mman.h>
1036
1037#if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
1038# define MAP_ANONYMOUS MAP_ANON
1039#endif
1040
1041#ifndef MAP_NORESERVE
1042# define MAP_NORESERVE 0
1043#endif
1044
1045#define MMAP(addr, size, prot, flags) \
1046 __mmap((addr), (size), (prot), (flags)|MAP_ANONYMOUS|MAP_PRIVATE, -1, 0)
1047
1048
1049/*
1050 ----------------------- Chunk representations -----------------------
1051*/
1052
1053
1054/*
1055 This struct declaration is misleading (but accurate and necessary).
1056 It declares a "view" into memory allowing access to necessary
1057 fields at known offsets from a given base. See explanation below.
1058*/
1059
1060struct malloc_chunk {
1061
1062 INTERNAL_SIZE_T mchunk_prev_size; /* Size of previous chunk (if free). */
1063 INTERNAL_SIZE_T mchunk_size; /* Size in bytes, including overhead. */
1064
1065 struct malloc_chunk* fd; /* double links -- used only if free. */
1066 struct malloc_chunk* bk;
1067
1068 /* Only used for large blocks: pointer to next larger size. */
1069 struct malloc_chunk* fd_nextsize; /* double links -- used only if free. */
1070 struct malloc_chunk* bk_nextsize;
1071};
1072
1073
1074/*
1075 malloc_chunk details:
1076
1077 (The following includes lightly edited explanations by Colin Plumb.)
1078
1079 Chunks of memory are maintained using a `boundary tag' method as
1080 described in e.g., Knuth or Standish. (See the paper by Paul
1081 Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a
1082 survey of such techniques.) Sizes of free chunks are stored both
1083 in the front of each chunk and at the end. This makes
1084 consolidating fragmented chunks into bigger chunks very fast. The
1085 size fields also hold bits representing whether chunks are free or
1086 in use.
1087
1088 An allocated chunk looks like this:
1089
1090
1091 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1092 | Size of previous chunk, if unallocated (P clear) |
1093 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1094 | Size of chunk, in bytes |A|M|P|
1095 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1096 | User data starts here... .
1097 . .
1098 . (malloc_usable_size() bytes) .
1099 . |
1100nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1101 | (size of chunk, but used for application data) |
1102 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1103 | Size of next chunk, in bytes |A|0|1|
1104 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1105
1106 Where "chunk" is the front of the chunk for the purpose of most of
1107 the malloc code, but "mem" is the pointer that is returned to the
1108 user. "Nextchunk" is the beginning of the next contiguous chunk.
1109
1110 Chunks always begin on even word boundaries, so the mem portion
1111 (which is returned to the user) is also on an even word boundary, and
1112 thus at least double-word aligned.
1113
1114 Free chunks are stored in circular doubly-linked lists, and look like this:
1115
1116 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1117 | Size of previous chunk, if unallocated (P clear) |
1118 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1119 `head:' | Size of chunk, in bytes |A|0|P|
1120 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1121 | Forward pointer to next chunk in list |
1122 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1123 | Back pointer to previous chunk in list |
1124 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1125 | Unused space (may be 0 bytes long) .
1126 . .
1127 . |
1128nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1129 `foot:' | Size of chunk, in bytes |
1130 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1131 | Size of next chunk, in bytes |A|0|0|
1132 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1133
1134 The P (PREV_INUSE) bit, stored in the unused low-order bit of the
1135 chunk size (which is always a multiple of two words), is an in-use
1136 bit for the *previous* chunk. If that bit is *clear*, then the
1137 word before the current chunk size contains the previous chunk
1138 size, and can be used to find the front of the previous chunk.
1139 The very first chunk allocated always has this bit set,
1140 preventing access to non-existent (or non-owned) memory. If
1141 prev_inuse is set for any given chunk, then you CANNOT determine
1142 the size of the previous chunk, and might even get a memory
1143 addressing fault when trying to do so.
1144
1145 The A (NON_MAIN_ARENA) bit is cleared for chunks on the initial,
1146 main arena, described by the main_arena variable. When additional
1147 threads are spawned, each thread receives its own arena (up to a
1148 configurable limit, after which arenas are reused for multiple
1149 threads), and the chunks in these arenas have the A bit set. To
1150 find the arena for a chunk on such a non-main arena, heap_for_ptr
1151 performs a bit mask operation and indirection through the ar_ptr
1152 member of the per-heap header heap_info (see arena.c).
1153
1154 Note that the `foot' of the current chunk is actually represented
1155 as the prev_size of the NEXT chunk. This makes it easier to
1156 deal with alignments etc but can be very confusing when trying
1157 to extend or adapt this code.
1158
1159 The three exceptions to all this are:
1160
1161 1. The special chunk `top' doesn't bother using the
1162 trailing size field since there is no next contiguous chunk
1163 that would have to index off it. After initialization, `top'
1164 is forced to always exist. If it would become less than
1165 MINSIZE bytes long, it is replenished.
1166
1167 2. Chunks allocated via mmap, which have the second-lowest-order
1168 bit M (IS_MMAPPED) set in their size fields. Because they are
1169 allocated one-by-one, each must contain its own trailing size
1170 field. If the M bit is set, the other bits are ignored
1171 (because mmapped chunks are neither in an arena, nor adjacent
1172 to a freed chunk). The M bit is also used for chunks which
1173 originally came from a dumped heap via malloc_set_state in
1174 hooks.c.
1175
1176 3. Chunks in fastbins are treated as allocated chunks from the
1177 point of view of the chunk allocator. They are consolidated
1178 with their neighbors only in bulk, in malloc_consolidate.
1179*/
1180
1181/*
1182 ---------- Size and alignment checks and conversions ----------
1183*/
1184
1185/* conversion from malloc headers to user pointers, and back */
1186
1187#define chunk2mem(p) ((void*)((char*)(p) + 2*SIZE_SZ))
1188#define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*SIZE_SZ))
1189
1190/* The smallest possible chunk */
1191#define MIN_CHUNK_SIZE (offsetof(struct malloc_chunk, fd_nextsize))
1192
1193/* The smallest size we can malloc is an aligned minimal chunk */
1194
1195#define MINSIZE \
1196 (unsigned long)(((MIN_CHUNK_SIZE+MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK))
1197
1198/* Check if m has acceptable alignment */
1199
1200#define aligned_OK(m) (((unsigned long)(m) & MALLOC_ALIGN_MASK) == 0)
1201
1202#define misaligned_chunk(p) \
1203 ((uintptr_t)(MALLOC_ALIGNMENT == 2 * SIZE_SZ ? (p) : chunk2mem (p)) \
1204 & MALLOC_ALIGN_MASK)
1205
1206
1207/*
1208 Check if a request is so large that it would wrap around zero when
1209 padded and aligned. To simplify some other code, the bound is made
1210 low enough so that adding MINSIZE will also not wrap around zero.
1211 */
1212
1213#define REQUEST_OUT_OF_RANGE(req) \
1214 ((unsigned long) (req) >= \
1215 (unsigned long) (INTERNAL_SIZE_T) (-2 * MINSIZE))
1216
1217/* pad request bytes into a usable size -- internal version */
1218
1219#define request2size(req) \
1220 (((req) + SIZE_SZ + MALLOC_ALIGN_MASK < MINSIZE) ? \
1221 MINSIZE : \
1222 ((req) + SIZE_SZ + MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK)
1223
1224/* Same, except also perform an argument and result check. First, we check
1225 that the padding done by request2size didn't result in an integer
1226 overflow. Then we check (using REQUEST_OUT_OF_RANGE) that the resulting
1227 size isn't so large that a later alignment would lead to another integer
1228 overflow. */
1229#define checked_request2size(req, sz) \
1230({ \
1231 (sz) = request2size (req); \
1232 if (((sz) < (req)) \
1233 || REQUEST_OUT_OF_RANGE (sz)) \
1234 { \
1235 __set_errno (ENOMEM); \
1236 return 0; \
1237 } \
1238})
1239
1240/*
1241 --------------- Physical chunk operations ---------------
1242 */
1243
1244
1245/* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */
1246#define PREV_INUSE 0x1
1247
1248/* extract inuse bit of previous chunk */
1249#define prev_inuse(p) ((p)->mchunk_size & PREV_INUSE)
1250
1251
1252/* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */
1253#define IS_MMAPPED 0x2
1254
1255/* check for mmap()'ed chunk */
1256#define chunk_is_mmapped(p) ((p)->mchunk_size & IS_MMAPPED)
1257
1258
1259/* size field is or'ed with NON_MAIN_ARENA if the chunk was obtained
1260 from a non-main arena. This is only set immediately before handing
1261 the chunk to the user, if necessary. */
1262#define NON_MAIN_ARENA 0x4
1263
1264/* Check for chunk from main arena. */
1265#define chunk_main_arena(p) (((p)->mchunk_size & NON_MAIN_ARENA) == 0)
1266
1267/* Mark a chunk as not being on the main arena. */
1268#define set_non_main_arena(p) ((p)->mchunk_size |= NON_MAIN_ARENA)
1269
1270
1271/*
1272 Bits to mask off when extracting size
1273
1274 Note: IS_MMAPPED is intentionally not masked off from size field in
1275 macros for which mmapped chunks should never be seen. This should
1276 cause helpful core dumps to occur if it is tried by accident by
1277 people extending or adapting this malloc.
1278 */
1279#define SIZE_BITS (PREV_INUSE | IS_MMAPPED | NON_MAIN_ARENA)
1280
1281/* Get size, ignoring use bits */
1282#define chunksize(p) (chunksize_nomask (p) & ~(SIZE_BITS))
1283
1284/* Like chunksize, but do not mask SIZE_BITS. */
1285#define chunksize_nomask(p) ((p)->mchunk_size)
1286
1287/* Ptr to next physical malloc_chunk. */
1288#define next_chunk(p) ((mchunkptr) (((char *) (p)) + chunksize (p)))
1289
1290/* Size of the chunk below P. Only valid if !prev_inuse (P). */
1291#define prev_size(p) ((p)->mchunk_prev_size)
1292
1293/* Set the size of the chunk below P. Only valid if !prev_inuse (P). */
1294#define set_prev_size(p, sz) ((p)->mchunk_prev_size = (sz))
1295
1296/* Ptr to previous physical malloc_chunk. Only valid if !prev_inuse (P). */
1297#define prev_chunk(p) ((mchunkptr) (((char *) (p)) - prev_size (p)))
1298
1299/* Treat space at ptr + offset as a chunk */
1300#define chunk_at_offset(p, s) ((mchunkptr) (((char *) (p)) + (s)))
1301
1302/* extract p's inuse bit */
1303#define inuse(p) \
1304 ((((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size) & PREV_INUSE)
1305
1306/* set/clear chunk as being inuse without otherwise disturbing */
1307#define set_inuse(p) \
1308 ((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size |= PREV_INUSE
1309
1310#define clear_inuse(p) \
1311 ((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size &= ~(PREV_INUSE)
1312
1313
1314/* check/set/clear inuse bits in known places */
1315#define inuse_bit_at_offset(p, s) \
1316 (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size & PREV_INUSE)
1317
1318#define set_inuse_bit_at_offset(p, s) \
1319 (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size |= PREV_INUSE)
1320
1321#define clear_inuse_bit_at_offset(p, s) \
1322 (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size &= ~(PREV_INUSE))
1323
1324
1325/* Set size at head, without disturbing its use bit */
1326#define set_head_size(p, s) ((p)->mchunk_size = (((p)->mchunk_size & SIZE_BITS) | (s)))
1327
1328/* Set size/use field */
1329#define set_head(p, s) ((p)->mchunk_size = (s))
1330
1331/* Set size at footer (only when chunk is not in use) */
1332#define set_foot(p, s) (((mchunkptr) ((char *) (p) + (s)))->mchunk_prev_size = (s))
1333
1334
1335#pragma GCC poison mchunk_size
1336#pragma GCC poison mchunk_prev_size
1337
1338/*
1339 -------------------- Internal data structures --------------------
1340
1341 All internal state is held in an instance of malloc_state defined
1342 below. There are no other static variables, except in two optional
1343 cases:
1344 * If USE_MALLOC_LOCK is defined, the mALLOC_MUTEx declared above.
1345 * If mmap doesn't support MAP_ANONYMOUS, a dummy file descriptor
1346 for mmap.
1347
1348 Beware of lots of tricks that minimize the total bookkeeping space
1349 requirements. The result is a little over 1K bytes (for 4byte
1350 pointers and size_t.)
1351 */
1352
1353/*
1354 Bins
1355
1356 An array of bin headers for free chunks. Each bin is doubly
1357 linked. The bins are approximately proportionally (log) spaced.
1358 There are a lot of these bins (128). This may look excessive, but
1359 works very well in practice. Most bins hold sizes that are
1360 unusual as malloc request sizes, but are more usual for fragments
1361 and consolidated sets of chunks, which is what these bins hold, so
1362 they can be found quickly. All procedures maintain the invariant
1363 that no consolidated chunk physically borders another one, so each
1364 chunk in a list is known to be preceeded and followed by either
1365 inuse chunks or the ends of memory.
1366
1367 Chunks in bins are kept in size order, with ties going to the
1368 approximately least recently used chunk. Ordering isn't needed
1369 for the small bins, which all contain the same-sized chunks, but
1370 facilitates best-fit allocation for larger chunks. These lists
1371 are just sequential. Keeping them in order almost never requires
1372 enough traversal to warrant using fancier ordered data
1373 structures.
1374
1375 Chunks of the same size are linked with the most
1376 recently freed at the front, and allocations are taken from the
1377 back. This results in LRU (FIFO) allocation order, which tends
1378 to give each chunk an equal opportunity to be consolidated with
1379 adjacent freed chunks, resulting in larger free chunks and less
1380 fragmentation.
1381
1382 To simplify use in double-linked lists, each bin header acts
1383 as a malloc_chunk. This avoids special-casing for headers.
1384 But to conserve space and improve locality, we allocate
1385 only the fd/bk pointers of bins, and then use repositioning tricks
1386 to treat these as the fields of a malloc_chunk*.
1387 */
1388
1389typedef struct malloc_chunk *mbinptr;
1390
1391/* addressing -- note that bin_at(0) does not exist */
1392#define bin_at(m, i) \
1393 (mbinptr) (((char *) &((m)->bins[((i) - 1) * 2])) \
1394 - offsetof (struct malloc_chunk, fd))
1395
1396/* analog of ++bin */
1397#define next_bin(b) ((mbinptr) ((char *) (b) + (sizeof (mchunkptr) << 1)))
1398
1399/* Reminders about list directionality within bins */
1400#define first(b) ((b)->fd)
1401#define last(b) ((b)->bk)
1402
1403/* Take a chunk off a bin list */
1404#define unlink(AV, P, BK, FD) { \
1405 if (__builtin_expect (chunksize(P) != prev_size (next_chunk(P)), 0)) \
1406 malloc_printerr ("corrupted size vs. prev_size"); \
1407 FD = P->fd; \
1408 BK = P->bk; \
1409 if (__builtin_expect (FD->bk != P || BK->fd != P, 0)) \
1410 malloc_printerr ("corrupted double-linked list"); \
1411 else { \
1412 FD->bk = BK; \
1413 BK->fd = FD; \
1414 if (!in_smallbin_range (chunksize_nomask (P)) \
1415 && __builtin_expect (P->fd_nextsize != NULL, 0)) { \
1416 if (__builtin_expect (P->fd_nextsize->bk_nextsize != P, 0) \
1417 || __builtin_expect (P->bk_nextsize->fd_nextsize != P, 0)) \
1418 malloc_printerr ("corrupted double-linked list (not small)"); \
1419 if (FD->fd_nextsize == NULL) { \
1420 if (P->fd_nextsize == P) \
1421 FD->fd_nextsize = FD->bk_nextsize = FD; \
1422 else { \
1423 FD->fd_nextsize = P->fd_nextsize; \
1424 FD->bk_nextsize = P->bk_nextsize; \
1425 P->fd_nextsize->bk_nextsize = FD; \
1426 P->bk_nextsize->fd_nextsize = FD; \
1427 } \
1428 } else { \
1429 P->fd_nextsize->bk_nextsize = P->bk_nextsize; \
1430 P->bk_nextsize->fd_nextsize = P->fd_nextsize; \
1431 } \
1432 } \
1433 } \
1434}
1435
1436/*
1437 Indexing
1438
1439 Bins for sizes < 512 bytes contain chunks of all the same size, spaced
1440 8 bytes apart. Larger bins are approximately logarithmically spaced:
1441
1442 64 bins of size 8
1443 32 bins of size 64
1444 16 bins of size 512
1445 8 bins of size 4096
1446 4 bins of size 32768
1447 2 bins of size 262144
1448 1 bin of size what's left
1449
1450 There is actually a little bit of slop in the numbers in bin_index
1451 for the sake of speed. This makes no difference elsewhere.
1452
1453 The bins top out around 1MB because we expect to service large
1454 requests via mmap.
1455
1456 Bin 0 does not exist. Bin 1 is the unordered list; if that would be
1457 a valid chunk size the small bins are bumped up one.
1458 */
1459
1460#define NBINS 128
1461#define NSMALLBINS 64
1462#define SMALLBIN_WIDTH MALLOC_ALIGNMENT
1463#define SMALLBIN_CORRECTION (MALLOC_ALIGNMENT > 2 * SIZE_SZ)
1464#define MIN_LARGE_SIZE ((NSMALLBINS - SMALLBIN_CORRECTION) * SMALLBIN_WIDTH)
1465
1466#define in_smallbin_range(sz) \
1467 ((unsigned long) (sz) < (unsigned long) MIN_LARGE_SIZE)
1468
1469#define smallbin_index(sz) \
1470 ((SMALLBIN_WIDTH == 16 ? (((unsigned) (sz)) >> 4) : (((unsigned) (sz)) >> 3))\
1471 + SMALLBIN_CORRECTION)
1472
1473#define largebin_index_32(sz) \
1474 (((((unsigned long) (sz)) >> 6) <= 38) ? 56 + (((unsigned long) (sz)) >> 6) :\
1475 ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1476 ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1477 ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1478 ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1479 126)
1480
1481#define largebin_index_32_big(sz) \
1482 (((((unsigned long) (sz)) >> 6) <= 45) ? 49 + (((unsigned long) (sz)) >> 6) :\
1483 ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1484 ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1485 ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1486 ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1487 126)
1488
1489// XXX It remains to be seen whether it is good to keep the widths of
1490// XXX the buckets the same or whether it should be scaled by a factor
1491// XXX of two as well.
1492#define largebin_index_64(sz) \
1493 (((((unsigned long) (sz)) >> 6) <= 48) ? 48 + (((unsigned long) (sz)) >> 6) :\
1494 ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1495 ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1496 ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1497 ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1498 126)
1499
1500#define largebin_index(sz) \
1501 (SIZE_SZ == 8 ? largebin_index_64 (sz) \
1502 : MALLOC_ALIGNMENT == 16 ? largebin_index_32_big (sz) \
1503 : largebin_index_32 (sz))
1504
1505#define bin_index(sz) \
1506 ((in_smallbin_range (sz)) ? smallbin_index (sz) : largebin_index (sz))
1507
1508
1509/*
1510 Unsorted chunks
1511
1512 All remainders from chunk splits, as well as all returned chunks,
1513 are first placed in the "unsorted" bin. They are then placed
1514 in regular bins after malloc gives them ONE chance to be used before
1515 binning. So, basically, the unsorted_chunks list acts as a queue,
1516 with chunks being placed on it in free (and malloc_consolidate),
1517 and taken off (to be either used or placed in bins) in malloc.
1518
1519 The NON_MAIN_ARENA flag is never set for unsorted chunks, so it
1520 does not have to be taken into account in size comparisons.
1521 */
1522
1523/* The otherwise unindexable 1-bin is used to hold unsorted chunks. */
1524#define unsorted_chunks(M) (bin_at (M, 1))
1525
1526/*
1527 Top
1528
1529 The top-most available chunk (i.e., the one bordering the end of
1530 available memory) is treated specially. It is never included in
1531 any bin, is used only if no other chunk is available, and is
1532 released back to the system if it is very large (see
1533 M_TRIM_THRESHOLD). Because top initially
1534 points to its own bin with initial zero size, thus forcing
1535 extension on the first malloc request, we avoid having any special
1536 code in malloc to check whether it even exists yet. But we still
1537 need to do so when getting memory from system, so we make
1538 initial_top treat the bin as a legal but unusable chunk during the
1539 interval between initialization and the first call to
1540 sysmalloc. (This is somewhat delicate, since it relies on
1541 the 2 preceding words to be zero during this interval as well.)
1542 */
1543
1544/* Conveniently, the unsorted bin can be used as dummy top on first call */
1545#define initial_top(M) (unsorted_chunks (M))
1546
1547/*
1548 Binmap
1549
1550 To help compensate for the large number of bins, a one-level index
1551 structure is used for bin-by-bin searching. `binmap' is a
1552 bitvector recording whether bins are definitely empty so they can
1553 be skipped over during during traversals. The bits are NOT always
1554 cleared as soon as bins are empty, but instead only
1555 when they are noticed to be empty during traversal in malloc.
1556 */
1557
1558/* Conservatively use 32 bits per map word, even if on 64bit system */
1559#define BINMAPSHIFT 5
1560#define BITSPERMAP (1U << BINMAPSHIFT)
1561#define BINMAPSIZE (NBINS / BITSPERMAP)
1562
1563#define idx2block(i) ((i) >> BINMAPSHIFT)
1564#define idx2bit(i) ((1U << ((i) & ((1U << BINMAPSHIFT) - 1))))
1565
1566#define mark_bin(m, i) ((m)->binmap[idx2block (i)] |= idx2bit (i))
1567#define unmark_bin(m, i) ((m)->binmap[idx2block (i)] &= ~(idx2bit (i)))
1568#define get_binmap(m, i) ((m)->binmap[idx2block (i)] & idx2bit (i))
1569
1570/*
1571 Fastbins
1572
1573 An array of lists holding recently freed small chunks. Fastbins
1574 are not doubly linked. It is faster to single-link them, and
1575 since chunks are never removed from the middles of these lists,
1576 double linking is not necessary. Also, unlike regular bins, they
1577 are not even processed in FIFO order (they use faster LIFO) since
1578 ordering doesn't much matter in the transient contexts in which
1579 fastbins are normally used.
1580
1581 Chunks in fastbins keep their inuse bit set, so they cannot
1582 be consolidated with other free chunks. malloc_consolidate
1583 releases all chunks in fastbins and consolidates them with
1584 other free chunks.
1585 */
1586
1587typedef struct malloc_chunk *mfastbinptr;
1588#define fastbin(ar_ptr, idx) ((ar_ptr)->fastbinsY[idx])
1589
1590/* offset 2 to use otherwise unindexable first 2 bins */
1591#define fastbin_index(sz) \
1592 ((((unsigned int) (sz)) >> (SIZE_SZ == 8 ? 4 : 3)) - 2)
1593
1594
1595/* The maximum fastbin request size we support */
1596#define MAX_FAST_SIZE (80 * SIZE_SZ / 4)
1597
1598#define NFASTBINS (fastbin_index (request2size (MAX_FAST_SIZE)) + 1)
1599
1600/*
1601 FASTBIN_CONSOLIDATION_THRESHOLD is the size of a chunk in free()
1602 that triggers automatic consolidation of possibly-surrounding
1603 fastbin chunks. This is a heuristic, so the exact value should not
1604 matter too much. It is defined at half the default trim threshold as a
1605 compromise heuristic to only attempt consolidation if it is likely
1606 to lead to trimming. However, it is not dynamically tunable, since
1607 consolidation reduces fragmentation surrounding large chunks even
1608 if trimming is not used.
1609 */
1610
1611#define FASTBIN_CONSOLIDATION_THRESHOLD (65536UL)
1612
1613/*
1614 NONCONTIGUOUS_BIT indicates that MORECORE does not return contiguous
1615 regions. Otherwise, contiguity is exploited in merging together,
1616 when possible, results from consecutive MORECORE calls.
1617
1618 The initial value comes from MORECORE_CONTIGUOUS, but is
1619 changed dynamically if mmap is ever used as an sbrk substitute.
1620 */
1621
1622#define NONCONTIGUOUS_BIT (2U)
1623
1624#define contiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) == 0)
1625#define noncontiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) != 0)
1626#define set_noncontiguous(M) ((M)->flags |= NONCONTIGUOUS_BIT)
1627#define set_contiguous(M) ((M)->flags &= ~NONCONTIGUOUS_BIT)
1628
1629/* Maximum size of memory handled in fastbins. */
1630static INTERNAL_SIZE_T global_max_fast;
1631
1632/*
1633 Set value of max_fast.
1634 Use impossibly small value if 0.
1635 Precondition: there are no existing fastbin chunks in the main arena.
1636 Since do_check_malloc_state () checks this, we call malloc_consolidate ()
1637 before changing max_fast. Note other arenas will leak their fast bin
1638 entries if max_fast is reduced.
1639 */
1640
1641#define set_max_fast(s) \
1642 global_max_fast = (((s) == 0) \
1643 ? SMALLBIN_WIDTH : ((s + SIZE_SZ) & ~MALLOC_ALIGN_MASK))
1644
1645static inline INTERNAL_SIZE_T
1646get_max_fast (void)
1647{
1648 /* Tell the GCC optimizers that global_max_fast is never larger
1649 than MAX_FAST_SIZE. This avoids out-of-bounds array accesses in
1650 _int_malloc after constant propagation of the size parameter.
1651 (The code never executes because malloc preserves the
1652 global_max_fast invariant, but the optimizers may not recognize
1653 this.) */
1654 if (global_max_fast > MAX_FAST_SIZE)
1655 __builtin_unreachable ();
1656 return global_max_fast;
1657}
1658
1659/*
1660 ----------- Internal state representation and initialization -----------
1661 */
1662
1663/*
1664 have_fastchunks indicates that there are probably some fastbin chunks.
1665 It is set true on entering a chunk into any fastbin, and cleared early in
1666 malloc_consolidate. The value is approximate since it may be set when there
1667 are no fastbin chunks, or it may be clear even if there are fastbin chunks
1668 available. Given it's sole purpose is to reduce number of redundant calls to
1669 malloc_consolidate, it does not affect correctness. As a result we can safely
1670 use relaxed atomic accesses.
1671 */
1672
1673
1674struct malloc_state
1675{
1676 /* Serialize access. */
1677 __libc_lock_define (, mutex);
1678
1679 /* Flags (formerly in max_fast). */
1680 int flags;
1681
1682 /* Set if the fastbin chunks contain recently inserted free blocks. */
1683 /* Note this is a bool but not all targets support atomics on booleans. */
1684 int have_fastchunks;
1685
1686 /* Fastbins */
1687 mfastbinptr fastbinsY[NFASTBINS];
1688
1689 /* Base of the topmost chunk -- not otherwise kept in a bin */
1690 mchunkptr top;
1691
1692 /* The remainder from the most recent split of a small request */
1693 mchunkptr last_remainder;
1694
1695 /* Normal bins packed as described above */
1696 mchunkptr bins[NBINS * 2 - 2];
1697
1698 /* Bitmap of bins */
1699 unsigned int binmap[BINMAPSIZE];
1700
1701 /* Linked list */
1702 struct malloc_state *next;
1703
1704 /* Linked list for free arenas. Access to this field is serialized
1705 by free_list_lock in arena.c. */
1706 struct malloc_state *next_free;
1707
1708 /* Number of threads attached to this arena. 0 if the arena is on
1709 the free list. Access to this field is serialized by
1710 free_list_lock in arena.c. */
1711 INTERNAL_SIZE_T attached_threads;
1712
1713 /* Memory allocated from the system in this arena. */
1714 INTERNAL_SIZE_T system_mem;
1715 INTERNAL_SIZE_T max_system_mem;
1716};
1717
1718struct malloc_par
1719{
1720 /* Tunable parameters */
1721 unsigned long trim_threshold;
1722 INTERNAL_SIZE_T top_pad;
1723 INTERNAL_SIZE_T mmap_threshold;
1724 INTERNAL_SIZE_T arena_test;
1725 INTERNAL_SIZE_T arena_max;
1726
1727 /* Memory map support */
1728 int n_mmaps;
1729 int n_mmaps_max;
1730 int max_n_mmaps;
1731 /* the mmap_threshold is dynamic, until the user sets
1732 it manually, at which point we need to disable any
1733 dynamic behavior. */
1734 int no_dyn_threshold;
1735
1736 /* Statistics */
1737 INTERNAL_SIZE_T mmapped_mem;
1738 INTERNAL_SIZE_T max_mmapped_mem;
1739
1740 /* First address handed out by MORECORE/sbrk. */
1741 char *sbrk_base;
1742
1743#if USE_TCACHE
1744 /* Maximum number of buckets to use. */
1745 size_t tcache_bins;
1746 size_t tcache_max_bytes;
1747 /* Maximum number of chunks in each bucket. */
1748 size_t tcache_count;
1749 /* Maximum number of chunks to remove from the unsorted list, which
1750 aren't used to prefill the cache. */
1751 size_t tcache_unsorted_limit;
1752#endif
1753};
1754
1755/* There are several instances of this struct ("arenas") in this
1756 malloc. If you are adapting this malloc in a way that does NOT use
1757 a static or mmapped malloc_state, you MUST explicitly zero-fill it
1758 before using. This malloc relies on the property that malloc_state
1759 is initialized to all zeroes (as is true of C statics). */
1760
1761static struct malloc_state main_arena =
1762{
1763 .mutex = _LIBC_LOCK_INITIALIZER,
1764 .next = &main_arena,
1765 .attached_threads = 1
1766};
1767
1768/* These variables are used for undumping support. Chunked are marked
1769 as using mmap, but we leave them alone if they fall into this
1770 range. NB: The chunk size for these chunks only includes the
1771 initial size field (of SIZE_SZ bytes), there is no trailing size
1772 field (unlike with regular mmapped chunks). */
1773static mchunkptr dumped_main_arena_start; /* Inclusive. */
1774static mchunkptr dumped_main_arena_end; /* Exclusive. */
1775
1776/* True if the pointer falls into the dumped arena. Use this after
1777 chunk_is_mmapped indicates a chunk is mmapped. */
1778#define DUMPED_MAIN_ARENA_CHUNK(p) \
1779 ((p) >= dumped_main_arena_start && (p) < dumped_main_arena_end)
1780
1781/* There is only one instance of the malloc parameters. */
1782
1783static struct malloc_par mp_ =
1784{
1785 .top_pad = DEFAULT_TOP_PAD,
1786 .n_mmaps_max = DEFAULT_MMAP_MAX,
1787 .mmap_threshold = DEFAULT_MMAP_THRESHOLD,
1788 .trim_threshold = DEFAULT_TRIM_THRESHOLD,
1789#define NARENAS_FROM_NCORES(n) ((n) * (sizeof (long) == 4 ? 2 : 8))
1790 .arena_test = NARENAS_FROM_NCORES (1)
1791#if USE_TCACHE
1792 ,
1793 .tcache_count = TCACHE_FILL_COUNT,
1794 .tcache_bins = TCACHE_MAX_BINS,
1795 .tcache_max_bytes = tidx2usize (TCACHE_MAX_BINS-1),
1796 .tcache_unsorted_limit = 0 /* No limit. */
1797#endif
1798};
1799
1800/*
1801 Initialize a malloc_state struct.
1802
1803 This is called from ptmalloc_init () or from _int_new_arena ()
1804 when creating a new arena.
1805 */
1806
1807static void
1808malloc_init_state (mstate av)
1809{
1810 int i;
1811 mbinptr bin;
1812
1813 /* Establish circular links for normal bins */
1814 for (i = 1; i < NBINS; ++i)
1815 {
1816 bin = bin_at (av, i);
1817 bin->fd = bin->bk = bin;
1818 }
1819
1820#if MORECORE_CONTIGUOUS
1821 if (av != &main_arena)
1822#endif
1823 set_noncontiguous (av);
1824 if (av == &main_arena)
1825 set_max_fast (DEFAULT_MXFAST);
1826 atomic_store_relaxed (&av->have_fastchunks, false);
1827
1828 av->top = initial_top (av);
1829}
1830
1831/*
1832 Other internal utilities operating on mstates
1833 */
1834
1835static void *sysmalloc (INTERNAL_SIZE_T, mstate);
1836static int systrim (size_t, mstate);
1837static void malloc_consolidate (mstate);
1838
1839
1840/* -------------- Early definitions for debugging hooks ---------------- */
1841
1842/* Define and initialize the hook variables. These weak definitions must
1843 appear before any use of the variables in a function (arena.c uses one). */
1844#ifndef weak_variable
1845/* In GNU libc we want the hook variables to be weak definitions to
1846 avoid a problem with Emacs. */
1847# define weak_variable weak_function
1848#endif
1849
1850/* Forward declarations. */
1851static void *malloc_hook_ini (size_t sz,
1852 const void *caller) __THROW;
1853static void *realloc_hook_ini (void *ptr, size_t sz,
1854 const void *caller) __THROW;
1855static void *memalign_hook_ini (size_t alignment, size_t sz,
1856 const void *caller) __THROW;
1857
1858#if HAVE_MALLOC_INIT_HOOK
1859void weak_variable (*__malloc_initialize_hook) (void) = NULL;
1860compat_symbol (libc, __malloc_initialize_hook,
1861 __malloc_initialize_hook, GLIBC_2_0);
1862#endif
1863
1864void weak_variable (*__free_hook) (void *__ptr,
1865 const void *) = NULL;
1866void *weak_variable (*__malloc_hook)
1867 (size_t __size, const void *) = malloc_hook_ini;
1868void *weak_variable (*__realloc_hook)
1869 (void *__ptr, size_t __size, const void *)
1870 = realloc_hook_ini;
1871void *weak_variable (*__memalign_hook)
1872 (size_t __alignment, size_t __size, const void *)
1873 = memalign_hook_ini;
1874void weak_variable (*__after_morecore_hook) (void) = NULL;
1875
1876/* This function is called from the arena shutdown hook, to free the
1877 thread cache (if it exists). */
1878static void tcache_thread_shutdown (void);
1879
1880/* ------------------ Testing support ----------------------------------*/
1881
1882static int perturb_byte;
1883
1884static void
1885alloc_perturb (char *p, size_t n)
1886{
1887 if (__glibc_unlikely (perturb_byte))
1888 memset (p, perturb_byte ^ 0xff, n);
1889}
1890
1891static void
1892free_perturb (char *p, size_t n)
1893{
1894 if (__glibc_unlikely (perturb_byte))
1895 memset (p, perturb_byte, n);
1896}
1897
1898
1899
1900#include <stap-probe.h>
1901
1902/* ------------------- Support for multiple arenas -------------------- */
1903#include "arena.c"
1904
1905/*
1906 Debugging support
1907
1908 These routines make a number of assertions about the states
1909 of data structures that should be true at all times. If any
1910 are not true, it's very likely that a user program has somehow
1911 trashed memory. (It's also possible that there is a coding error
1912 in malloc. In which case, please report it!)
1913 */
1914
1915#if !MALLOC_DEBUG
1916
1917# define check_chunk(A, P)
1918# define check_free_chunk(A, P)
1919# define check_inuse_chunk(A, P)
1920# define check_remalloced_chunk(A, P, N)
1921# define check_malloced_chunk(A, P, N)
1922# define check_malloc_state(A)
1923
1924#else
1925
1926# define check_chunk(A, P) do_check_chunk (A, P)
1927# define check_free_chunk(A, P) do_check_free_chunk (A, P)
1928# define check_inuse_chunk(A, P) do_check_inuse_chunk (A, P)
1929# define check_remalloced_chunk(A, P, N) do_check_remalloced_chunk (A, P, N)
1930# define check_malloced_chunk(A, P, N) do_check_malloced_chunk (A, P, N)
1931# define check_malloc_state(A) do_check_malloc_state (A)
1932
1933/*
1934 Properties of all chunks
1935 */
1936
1937static void
1938do_check_chunk (mstate av, mchunkptr p)
1939{
1940 unsigned long sz = chunksize (p);
1941 /* min and max possible addresses assuming contiguous allocation */
1942 char *max_address = (char *) (av->top) + chunksize (av->top);
1943 char *min_address = max_address - av->system_mem;
1944
1945 if (!chunk_is_mmapped (p))
1946 {
1947 /* Has legal address ... */
1948 if (p != av->top)
1949 {
1950 if (contiguous (av))
1951 {
1952 assert (((char *) p) >= min_address);
1953 assert (((char *) p + sz) <= ((char *) (av->top)));
1954 }
1955 }
1956 else
1957 {
1958 /* top size is always at least MINSIZE */
1959 assert ((unsigned long) (sz) >= MINSIZE);
1960 /* top predecessor always marked inuse */
1961 assert (prev_inuse (p));
1962 }
1963 }
1964 else if (!DUMPED_MAIN_ARENA_CHUNK (p))
1965 {
1966 /* address is outside main heap */
1967 if (contiguous (av) && av->top != initial_top (av))
1968 {
1969 assert (((char *) p) < min_address || ((char *) p) >= max_address);
1970 }
1971 /* chunk is page-aligned */
1972 assert (((prev_size (p) + sz) & (GLRO (dl_pagesize) - 1)) == 0);
1973 /* mem is aligned */
1974 assert (aligned_OK (chunk2mem (p)));
1975 }
1976}
1977
1978/*
1979 Properties of free chunks
1980 */
1981
1982static void
1983do_check_free_chunk (mstate av, mchunkptr p)
1984{
1985 INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE | NON_MAIN_ARENA);
1986 mchunkptr next = chunk_at_offset (p, sz);
1987
1988 do_check_chunk (av, p);
1989
1990 /* Chunk must claim to be free ... */
1991 assert (!inuse (p));
1992 assert (!chunk_is_mmapped (p));
1993
1994 /* Unless a special marker, must have OK fields */
1995 if ((unsigned long) (sz) >= MINSIZE)
1996 {
1997 assert ((sz & MALLOC_ALIGN_MASK) == 0);
1998 assert (aligned_OK (chunk2mem (p)));
1999 /* ... matching footer field */
2000 assert (prev_size (next_chunk (p)) == sz);
2001 /* ... and is fully consolidated */
2002 assert (prev_inuse (p));
2003 assert (next == av->top || inuse (next));
2004
2005 /* ... and has minimally sane links */
2006 assert (p->fd->bk == p);
2007 assert (p->bk->fd == p);
2008 }
2009 else /* markers are always of size SIZE_SZ */
2010 assert (sz == SIZE_SZ);
2011}
2012
2013/*
2014 Properties of inuse chunks
2015 */
2016
2017static void
2018do_check_inuse_chunk (mstate av, mchunkptr p)
2019{
2020 mchunkptr next;
2021
2022 do_check_chunk (av, p);
2023
2024 if (chunk_is_mmapped (p))
2025 return; /* mmapped chunks have no next/prev */
2026
2027 /* Check whether it claims to be in use ... */
2028 assert (inuse (p));
2029
2030 next = next_chunk (p);
2031
2032 /* ... and is surrounded by OK chunks.
2033 Since more things can be checked with free chunks than inuse ones,
2034 if an inuse chunk borders them and debug is on, it's worth doing them.
2035 */
2036 if (!prev_inuse (p))
2037 {
2038 /* Note that we cannot even look at prev unless it is not inuse */
2039 mchunkptr prv = prev_chunk (p);
2040 assert (next_chunk (prv) == p);
2041 do_check_free_chunk (av, prv);
2042 }
2043
2044 if (next == av->top)
2045 {
2046 assert (prev_inuse (next));
2047 assert (chunksize (next) >= MINSIZE);
2048 }
2049 else if (!inuse (next))
2050 do_check_free_chunk (av, next);
2051}
2052
2053/*
2054 Properties of chunks recycled from fastbins
2055 */
2056
2057static void
2058do_check_remalloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s)
2059{
2060 INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE | NON_MAIN_ARENA);
2061
2062 if (!chunk_is_mmapped (p))
2063 {
2064 assert (av == arena_for_chunk (p));
2065 if (chunk_main_arena (p))
2066 assert (av == &main_arena);
2067 else
2068 assert (av != &main_arena);
2069 }
2070
2071 do_check_inuse_chunk (av, p);
2072
2073 /* Legal size ... */
2074 assert ((sz & MALLOC_ALIGN_MASK) == 0);
2075 assert ((unsigned long) (sz) >= MINSIZE);
2076 /* ... and alignment */
2077 assert (aligned_OK (chunk2mem (p)));
2078 /* chunk is less than MINSIZE more than request */
2079 assert ((long) (sz) - (long) (s) >= 0);
2080 assert ((long) (sz) - (long) (s + MINSIZE) < 0);
2081}
2082
2083/*
2084 Properties of nonrecycled chunks at the point they are malloced
2085 */
2086
2087static void
2088do_check_malloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s)
2089{
2090 /* same as recycled case ... */
2091 do_check_remalloced_chunk (av, p, s);
2092
2093 /*
2094 ... plus, must obey implementation invariant that prev_inuse is
2095 always true of any allocated chunk; i.e., that each allocated
2096 chunk borders either a previously allocated and still in-use
2097 chunk, or the base of its memory arena. This is ensured
2098 by making all allocations from the `lowest' part of any found
2099 chunk. This does not necessarily hold however for chunks
2100 recycled via fastbins.
2101 */
2102
2103 assert (prev_inuse (p));
2104}
2105
2106
2107/*
2108 Properties of malloc_state.
2109
2110 This may be useful for debugging malloc, as well as detecting user
2111 programmer errors that somehow write into malloc_state.
2112
2113 If you are extending or experimenting with this malloc, you can
2114 probably figure out how to hack this routine to print out or
2115 display chunk addresses, sizes, bins, and other instrumentation.
2116 */
2117
2118static void
2119do_check_malloc_state (mstate av)
2120{
2121 int i;
2122 mchunkptr p;
2123 mchunkptr q;
2124 mbinptr b;
2125 unsigned int idx;
2126 INTERNAL_SIZE_T size;
2127 unsigned long total = 0;
2128 int max_fast_bin;
2129
2130 /* internal size_t must be no wider than pointer type */
2131 assert (sizeof (INTERNAL_SIZE_T) <= sizeof (char *));
2132
2133 /* alignment is a power of 2 */
2134 assert ((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT - 1)) == 0);
2135
2136 /* Check the arena is initialized. */
2137 assert (av->top != 0);
2138
2139 /* No memory has been allocated yet, so doing more tests is not possible. */
2140 if (av->top == initial_top (av))
2141 return;
2142
2143 /* pagesize is a power of 2 */
2144 assert (powerof2(GLRO (dl_pagesize)));
2145
2146 /* A contiguous main_arena is consistent with sbrk_base. */
2147 if (av == &main_arena && contiguous (av))
2148 assert ((char *) mp_.sbrk_base + av->system_mem ==
2149 (char *) av->top + chunksize (av->top));
2150
2151 /* properties of fastbins */
2152
2153 /* max_fast is in allowed range */
2154 assert ((get_max_fast () & ~1) <= request2size (MAX_FAST_SIZE));
2155
2156 max_fast_bin = fastbin_index (get_max_fast ());
2157
2158 for (i = 0; i < NFASTBINS; ++i)
2159 {
2160 p = fastbin (av, i);
2161
2162 /* The following test can only be performed for the main arena.
2163 While mallopt calls malloc_consolidate to get rid of all fast
2164 bins (especially those larger than the new maximum) this does
2165 only happen for the main arena. Trying to do this for any
2166 other arena would mean those arenas have to be locked and
2167 malloc_consolidate be called for them. This is excessive. And
2168 even if this is acceptable to somebody it still cannot solve
2169 the problem completely since if the arena is locked a
2170 concurrent malloc call might create a new arena which then
2171 could use the newly invalid fast bins. */
2172
2173 /* all bins past max_fast are empty */
2174 if (av == &main_arena && i > max_fast_bin)
2175 assert (p == 0);
2176
2177 while (p != 0)
2178 {
2179 /* each chunk claims to be inuse */
2180 do_check_inuse_chunk (av, p);
2181 total += chunksize (p);
2182 /* chunk belongs in this bin */
2183 assert (fastbin_index (chunksize (p)) == i);
2184 p = p->fd;
2185 }
2186 }
2187
2188 /* check normal bins */
2189 for (i = 1; i < NBINS; ++i)
2190 {
2191 b = bin_at (av, i);
2192
2193 /* binmap is accurate (except for bin 1 == unsorted_chunks) */
2194 if (i >= 2)
2195 {
2196 unsigned int binbit = get_binmap (av, i);
2197 int empty = last (b) == b;
2198 if (!binbit)
2199 assert (empty);
2200 else if (!empty)
2201 assert (binbit);
2202 }
2203
2204 for (p = last (b); p != b; p = p->bk)
2205 {
2206 /* each chunk claims to be free */
2207 do_check_free_chunk (av, p);
2208 size = chunksize (p);
2209 total += size;
2210 if (i >= 2)
2211 {
2212 /* chunk belongs in bin */
2213 idx = bin_index (size);
2214 assert (idx == i);
2215 /* lists are sorted */
2216 assert (p->bk == b ||
2217 (unsigned long) chunksize (p->bk) >= (unsigned long) chunksize (p));
2218
2219 if (!in_smallbin_range (size))
2220 {
2221 if (p->fd_nextsize != NULL)
2222 {
2223 if (p->fd_nextsize == p)
2224 assert (p->bk_nextsize == p);
2225 else
2226 {
2227 if (p->fd_nextsize == first (b))
2228 assert (chunksize (p) < chunksize (p->fd_nextsize));
2229 else
2230 assert (chunksize (p) > chunksize (p->fd_nextsize));
2231
2232 if (p == first (b))
2233 assert (chunksize (p) > chunksize (p->bk_nextsize));
2234 else
2235 assert (chunksize (p) < chunksize (p->bk_nextsize));
2236 }
2237 }
2238 else
2239 assert (p->bk_nextsize == NULL);
2240 }
2241 }
2242 else if (!in_smallbin_range (size))
2243 assert (p->fd_nextsize == NULL && p->bk_nextsize == NULL);
2244 /* chunk is followed by a legal chain of inuse chunks */
2245 for (q = next_chunk (p);
2246 (q != av->top && inuse (q) &&
2247 (unsigned long) (chunksize (q)) >= MINSIZE);
2248 q = next_chunk (q))
2249 do_check_inuse_chunk (av, q);
2250 }
2251 }
2252
2253 /* top chunk is OK */
2254 check_chunk (av, av->top);
2255}
2256#endif
2257
2258
2259/* ----------------- Support for debugging hooks -------------------- */
2260#include "hooks.c"
2261
2262
2263/* ----------- Routines dealing with system allocation -------------- */
2264
2265/*
2266 sysmalloc handles malloc cases requiring more memory from the system.
2267 On entry, it is assumed that av->top does not have enough
2268 space to service request for nb bytes, thus requiring that av->top
2269 be extended or replaced.
2270 */
2271
2272static void *
2273sysmalloc (INTERNAL_SIZE_T nb, mstate av)
2274{
2275 mchunkptr old_top; /* incoming value of av->top */
2276 INTERNAL_SIZE_T old_size; /* its size */
2277 char *old_end; /* its end address */
2278
2279 long size; /* arg to first MORECORE or mmap call */
2280 char *brk; /* return value from MORECORE */
2281
2282 long correction; /* arg to 2nd MORECORE call */
2283 char *snd_brk; /* 2nd return val */
2284
2285 INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of new space */
2286 INTERNAL_SIZE_T end_misalign; /* partial page left at end of new space */
2287 char *aligned_brk; /* aligned offset into brk */
2288
2289 mchunkptr p; /* the allocated/returned chunk */
2290 mchunkptr remainder; /* remainder from allocation */
2291 unsigned long remainder_size; /* its size */
2292
2293
2294 size_t pagesize = GLRO (dl_pagesize);
2295 bool tried_mmap = false;
2296
2297
2298 /*
2299 If have mmap, and the request size meets the mmap threshold, and
2300 the system supports mmap, and there are few enough currently
2301 allocated mmapped regions, try to directly map this request
2302 rather than expanding top.
2303 */
2304
2305 if (av == NULL
2306 || ((unsigned long) (nb) >= (unsigned long) (mp_.mmap_threshold)
2307 && (mp_.n_mmaps < mp_.n_mmaps_max)))
2308 {
2309 char *mm; /* return value from mmap call*/
2310
2311 try_mmap:
2312 /*
2313 Round up size to nearest page. For mmapped chunks, the overhead
2314 is one SIZE_SZ unit larger than for normal chunks, because there
2315 is no following chunk whose prev_size field could be used.
2316
2317 See the front_misalign handling below, for glibc there is no
2318 need for further alignments unless we have have high alignment.
2319 */
2320 if (MALLOC_ALIGNMENT == 2 * SIZE_SZ)
2321 size = ALIGN_UP (nb + SIZE_SZ, pagesize);
2322 else
2323 size = ALIGN_UP (nb + SIZE_SZ + MALLOC_ALIGN_MASK, pagesize);
2324 tried_mmap = true;
2325
2326 /* Don't try if size wraps around 0 */
2327 if ((unsigned long) (size) > (unsigned long) (nb))
2328 {
2329 mm = (char *) (MMAP (0, size, PROT_READ | PROT_WRITE, 0));
2330
2331 if (mm != MAP_FAILED)
2332 {
2333 /*
2334 The offset to the start of the mmapped region is stored
2335 in the prev_size field of the chunk. This allows us to adjust
2336 returned start address to meet alignment requirements here
2337 and in memalign(), and still be able to compute proper
2338 address argument for later munmap in free() and realloc().
2339 */
2340
2341 if (MALLOC_ALIGNMENT == 2 * SIZE_SZ)
2342 {
2343 /* For glibc, chunk2mem increases the address by 2*SIZE_SZ and
2344 MALLOC_ALIGN_MASK is 2*SIZE_SZ-1. Each mmap'ed area is page
2345 aligned and therefore definitely MALLOC_ALIGN_MASK-aligned. */
2346 assert (((INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK) == 0);
2347 front_misalign = 0;
2348 }
2349 else
2350 front_misalign = (INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK;
2351 if (front_misalign > 0)
2352 {
2353 correction = MALLOC_ALIGNMENT - front_misalign;
2354 p = (mchunkptr) (mm + correction);
2355 set_prev_size (p, correction);
2356 set_head (p, (size - correction) | IS_MMAPPED);
2357 }
2358 else
2359 {
2360 p = (mchunkptr) mm;
2361 set_prev_size (p, 0);
2362 set_head (p, size | IS_MMAPPED);
2363 }
2364
2365 /* update statistics */
2366
2367 int new = atomic_exchange_and_add (&mp_.n_mmaps, 1) + 1;
2368 atomic_max (&mp_.max_n_mmaps, new);
2369
2370 unsigned long sum;
2371 sum = atomic_exchange_and_add (&mp_.mmapped_mem, size) + size;
2372 atomic_max (&mp_.max_mmapped_mem, sum);
2373
2374 check_chunk (av, p);
2375
2376 return chunk2mem (p);
2377 }
2378 }
2379 }
2380
2381 /* There are no usable arenas and mmap also failed. */
2382 if (av == NULL)
2383 return 0;
2384
2385 /* Record incoming configuration of top */
2386
2387 old_top = av->top;
2388 old_size = chunksize (old_top);
2389 old_end = (char *) (chunk_at_offset (old_top, old_size));
2390
2391 brk = snd_brk = (char *) (MORECORE_FAILURE);
2392
2393 /*
2394 If not the first time through, we require old_size to be
2395 at least MINSIZE and to have prev_inuse set.
2396 */
2397
2398 assert ((old_top == initial_top (av) && old_size == 0) ||
2399 ((unsigned long) (old_size) >= MINSIZE &&
2400 prev_inuse (old_top) &&
2401 ((unsigned long) old_end & (pagesize - 1)) == 0));
2402
2403 /* Precondition: not enough current space to satisfy nb request */
2404 assert ((unsigned long) (old_size) < (unsigned long) (nb + MINSIZE));
2405
2406
2407 if (av != &main_arena)
2408 {
2409 heap_info *old_heap, *heap;
2410 size_t old_heap_size;
2411
2412 /* First try to extend the current heap. */
2413 old_heap = heap_for_ptr (old_top);
2414 old_heap_size = old_heap->size;
2415 if ((long) (MINSIZE + nb - old_size) > 0
2416 && grow_heap (old_heap, MINSIZE + nb - old_size) == 0)
2417 {
2418 av->system_mem += old_heap->size - old_heap_size;
2419 set_head (old_top, (((char *) old_heap + old_heap->size) - (char *) old_top)
2420 | PREV_INUSE);
2421 }
2422 else if ((heap = new_heap (nb + (MINSIZE + sizeof (*heap)), mp_.top_pad)))
2423 {
2424 /* Use a newly allocated heap. */
2425 heap->ar_ptr = av;
2426 heap->prev = old_heap;
2427 av->system_mem += heap->size;
2428 /* Set up the new top. */
2429 top (av) = chunk_at_offset (heap, sizeof (*heap));
2430 set_head (top (av), (heap->size - sizeof (*heap)) | PREV_INUSE);
2431
2432 /* Setup fencepost and free the old top chunk with a multiple of
2433 MALLOC_ALIGNMENT in size. */
2434 /* The fencepost takes at least MINSIZE bytes, because it might
2435 become the top chunk again later. Note that a footer is set
2436 up, too, although the chunk is marked in use. */
2437 old_size = (old_size - MINSIZE) & ~MALLOC_ALIGN_MASK;
2438 set_head (chunk_at_offset (old_top, old_size + 2 * SIZE_SZ), 0 | PREV_INUSE);
2439 if (old_size >= MINSIZE)
2440 {
2441 set_head (chunk_at_offset (old_top, old_size), (2 * SIZE_SZ) | PREV_INUSE);
2442 set_foot (chunk_at_offset (old_top, old_size), (2 * SIZE_SZ));
2443 set_head (old_top, old_size | PREV_INUSE | NON_MAIN_ARENA);
2444 _int_free (av, old_top, 1);
2445 }
2446 else
2447 {
2448 set_head (old_top, (old_size + 2 * SIZE_SZ) | PREV_INUSE);
2449 set_foot (old_top, (old_size + 2 * SIZE_SZ));
2450 }
2451 }
2452 else if (!tried_mmap)
2453 /* We can at least try to use to mmap memory. */
2454 goto try_mmap;
2455 }
2456 else /* av == main_arena */
2457
2458
2459 { /* Request enough space for nb + pad + overhead */
2460 size = nb + mp_.top_pad + MINSIZE;
2461
2462 /*
2463 If contiguous, we can subtract out existing space that we hope to
2464 combine with new space. We add it back later only if
2465 we don't actually get contiguous space.
2466 */
2467
2468 if (contiguous (av))
2469 size -= old_size;
2470
2471 /*
2472 Round to a multiple of page size.
2473 If MORECORE is not contiguous, this ensures that we only call it
2474 with whole-page arguments. And if MORECORE is contiguous and
2475 this is not first time through, this preserves page-alignment of
2476 previous calls. Otherwise, we correct to page-align below.
2477 */
2478
2479 size = ALIGN_UP (size, pagesize);
2480
2481 /*
2482 Don't try to call MORECORE if argument is so big as to appear
2483 negative. Note that since mmap takes size_t arg, it may succeed
2484 below even if we cannot call MORECORE.
2485 */
2486
2487 if (size > 0)
2488 {
2489 brk = (char *) (MORECORE (size));
2490 LIBC_PROBE (memory_sbrk_more, 2, brk, size);
2491 }
2492
2493 if (brk != (char *) (MORECORE_FAILURE))
2494 {
2495 /* Call the `morecore' hook if necessary. */
2496 void (*hook) (void) = atomic_forced_read (__after_morecore_hook);
2497 if (__builtin_expect (hook != NULL, 0))
2498 (*hook)();
2499 }
2500 else
2501 {
2502 /*
2503 If have mmap, try using it as a backup when MORECORE fails or
2504 cannot be used. This is worth doing on systems that have "holes" in
2505 address space, so sbrk cannot extend to give contiguous space, but
2506 space is available elsewhere. Note that we ignore mmap max count
2507 and threshold limits, since the space will not be used as a
2508 segregated mmap region.
2509 */
2510
2511 /* Cannot merge with old top, so add its size back in */
2512 if (contiguous (av))
2513 size = ALIGN_UP (size + old_size, pagesize);
2514
2515 /* If we are relying on mmap as backup, then use larger units */
2516 if ((unsigned long) (size) < (unsigned long) (MMAP_AS_MORECORE_SIZE))
2517 size = MMAP_AS_MORECORE_SIZE;
2518
2519 /* Don't try if size wraps around 0 */
2520 if ((unsigned long) (size) > (unsigned long) (nb))
2521 {
2522 char *mbrk = (char *) (MMAP (0, size, PROT_READ | PROT_WRITE, 0));
2523
2524 if (mbrk != MAP_FAILED)
2525 {
2526 /* We do not need, and cannot use, another sbrk call to find end */
2527 brk = mbrk;
2528 snd_brk = brk + size;
2529
2530 /*
2531 Record that we no longer have a contiguous sbrk region.
2532 After the first time mmap is used as backup, we do not
2533 ever rely on contiguous space since this could incorrectly
2534 bridge regions.
2535 */
2536 set_noncontiguous (av);
2537 }
2538 }
2539 }
2540
2541 if (brk != (char *) (MORECORE_FAILURE))
2542 {
2543 if (mp_.sbrk_base == 0)
2544 mp_.sbrk_base = brk;
2545 av->system_mem += size;
2546
2547 /*
2548 If MORECORE extends previous space, we can likewise extend top size.
2549 */
2550
2551 if (brk == old_end && snd_brk == (char *) (MORECORE_FAILURE))
2552 set_head (old_top, (size + old_size) | PREV_INUSE);
2553
2554 else if (contiguous (av) && old_size && brk < old_end)
2555 /* Oops! Someone else killed our space.. Can't touch anything. */
2556 malloc_printerr ("break adjusted to free malloc space");
2557
2558 /*
2559 Otherwise, make adjustments:
2560
2561 * If the first time through or noncontiguous, we need to call sbrk
2562 just to find out where the end of memory lies.
2563
2564 * We need to ensure that all returned chunks from malloc will meet
2565 MALLOC_ALIGNMENT
2566
2567 * If there was an intervening foreign sbrk, we need to adjust sbrk
2568 request size to account for fact that we will not be able to
2569 combine new space with existing space in old_top.
2570
2571 * Almost all systems internally allocate whole pages at a time, in
2572 which case we might as well use the whole last page of request.
2573 So we allocate enough more memory to hit a page boundary now,
2574 which in turn causes future contiguous calls to page-align.
2575 */
2576
2577 else
2578 {
2579 front_misalign = 0;
2580 end_misalign = 0;
2581 correction = 0;
2582 aligned_brk = brk;
2583
2584 /* handle contiguous cases */
2585 if (contiguous (av))
2586 {
2587 /* Count foreign sbrk as system_mem. */
2588 if (old_size)
2589 av->system_mem += brk - old_end;
2590
2591 /* Guarantee alignment of first new chunk made from this space */
2592
2593 front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK;
2594 if (front_misalign > 0)
2595 {
2596 /*
2597 Skip over some bytes to arrive at an aligned position.
2598 We don't need to specially mark these wasted front bytes.
2599 They will never be accessed anyway because
2600 prev_inuse of av->top (and any chunk created from its start)
2601 is always true after initialization.
2602 */
2603
2604 correction = MALLOC_ALIGNMENT - front_misalign;
2605 aligned_brk += correction;
2606 }
2607
2608 /*
2609 If this isn't adjacent to existing space, then we will not
2610 be able to merge with old_top space, so must add to 2nd request.
2611 */
2612
2613 correction += old_size;
2614
2615 /* Extend the end address to hit a page boundary */
2616 end_misalign = (INTERNAL_SIZE_T) (brk + size + correction);
2617 correction += (ALIGN_UP (end_misalign, pagesize)) - end_misalign;
2618
2619 assert (correction >= 0);
2620 snd_brk = (char *) (MORECORE (correction));
2621
2622 /*
2623 If can't allocate correction, try to at least find out current
2624 brk. It might be enough to proceed without failing.
2625
2626 Note that if second sbrk did NOT fail, we assume that space
2627 is contiguous with first sbrk. This is a safe assumption unless
2628 program is multithreaded but doesn't use locks and a foreign sbrk
2629 occurred between our first and second calls.
2630 */
2631
2632 if (snd_brk == (char *) (MORECORE_FAILURE))
2633 {
2634 correction = 0;
2635 snd_brk = (char *) (MORECORE (0));
2636 }
2637 else
2638 {
2639 /* Call the `morecore' hook if necessary. */
2640 void (*hook) (void) = atomic_forced_read (__after_morecore_hook);
2641 if (__builtin_expect (hook != NULL, 0))
2642 (*hook)();
2643 }
2644 }
2645
2646 /* handle non-contiguous cases */
2647 else
2648 {
2649 if (MALLOC_ALIGNMENT == 2 * SIZE_SZ)
2650 /* MORECORE/mmap must correctly align */
2651 assert (((unsigned long) chunk2mem (brk) & MALLOC_ALIGN_MASK) == 0);
2652 else
2653 {
2654 front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK;
2655 if (front_misalign > 0)
2656 {
2657 /*
2658 Skip over some bytes to arrive at an aligned position.
2659 We don't need to specially mark these wasted front bytes.
2660 They will never be accessed anyway because
2661 prev_inuse of av->top (and any chunk created from its start)
2662 is always true after initialization.
2663 */
2664
2665 aligned_brk += MALLOC_ALIGNMENT - front_misalign;
2666 }
2667 }
2668
2669 /* Find out current end of memory */
2670 if (snd_brk == (char *) (MORECORE_FAILURE))
2671 {
2672 snd_brk = (char *) (MORECORE (0));
2673 }
2674 }
2675
2676 /* Adjust top based on results of second sbrk */
2677 if (snd_brk != (char *) (MORECORE_FAILURE))
2678 {
2679 av->top = (mchunkptr) aligned_brk;
2680 set_head (av->top, (snd_brk - aligned_brk + correction) | PREV_INUSE);
2681 av->system_mem += correction;
2682
2683 /*
2684 If not the first time through, we either have a
2685 gap due to foreign sbrk or a non-contiguous region. Insert a
2686 double fencepost at old_top to prevent consolidation with space
2687 we don't own. These fenceposts are artificial chunks that are
2688 marked as inuse and are in any case too small to use. We need
2689 two to make sizes and alignments work out.
2690 */
2691
2692 if (old_size != 0)
2693 {
2694 /*
2695 Shrink old_top to insert fenceposts, keeping size a
2696 multiple of MALLOC_ALIGNMENT. We know there is at least
2697 enough space in old_top to do this.
2698 */
2699 old_size = (old_size - 4 * SIZE_SZ) & ~MALLOC_ALIGN_MASK;
2700 set_head (old_top, old_size | PREV_INUSE);
2701
2702 /*
2703 Note that the following assignments completely overwrite
2704 old_top when old_size was previously MINSIZE. This is
2705 intentional. We need the fencepost, even if old_top otherwise gets
2706 lost.
2707 */
2708 set_head (chunk_at_offset (old_top, old_size),
2709 (2 * SIZE_SZ) | PREV_INUSE);
2710 set_head (chunk_at_offset (old_top, old_size + 2 * SIZE_SZ),
2711 (2 * SIZE_SZ) | PREV_INUSE);
2712
2713 /* If possible, release the rest. */
2714 if (old_size >= MINSIZE)
2715 {
2716 _int_free (av, old_top, 1);
2717 }
2718 }
2719 }
2720 }
2721 }
2722 } /* if (av != &main_arena) */
2723
2724 if ((unsigned long) av->system_mem > (unsigned long) (av->max_system_mem))
2725 av->max_system_mem = av->system_mem;
2726 check_malloc_state (av);
2727
2728 /* finally, do the allocation */
2729 p = av->top;
2730 size = chunksize (p);
2731
2732 /* check that one of the above allocation paths succeeded */
2733 if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE))
2734 {
2735 remainder_size = size - nb;
2736 remainder = chunk_at_offset (p, nb);
2737 av->top = remainder;
2738 set_head (p, nb | PREV_INUSE | (av != &main_arena ? NON_MAIN_ARENA : 0));
2739 set_head (remainder, remainder_size | PREV_INUSE);
2740 check_malloced_chunk (av, p, nb);
2741 return chunk2mem (p);
2742 }
2743
2744 /* catch all failure paths */
2745 __set_errno (ENOMEM);
2746 return 0;
2747}
2748
2749
2750/*
2751 systrim is an inverse of sorts to sysmalloc. It gives memory back
2752 to the system (via negative arguments to sbrk) if there is unused
2753 memory at the `high' end of the malloc pool. It is called
2754 automatically by free() when top space exceeds the trim
2755 threshold. It is also called by the public malloc_trim routine. It
2756 returns 1 if it actually released any memory, else 0.
2757 */
2758
2759static int
2760systrim (size_t pad, mstate av)
2761{
2762 long top_size; /* Amount of top-most memory */
2763 long extra; /* Amount to release */
2764 long released; /* Amount actually released */
2765 char *current_brk; /* address returned by pre-check sbrk call */
2766 char *new_brk; /* address returned by post-check sbrk call */
2767 size_t pagesize;
2768 long top_area;
2769
2770 pagesize = GLRO (dl_pagesize);
2771 top_size = chunksize (av->top);
2772
2773 top_area = top_size - MINSIZE - 1;
2774 if (top_area <= pad)
2775 return 0;
2776
2777 /* Release in pagesize units and round down to the nearest page. */
2778 extra = ALIGN_DOWN(top_area - pad, pagesize);
2779
2780 if (extra == 0)
2781 return 0;
2782
2783 /*
2784 Only proceed if end of memory is where we last set it.
2785 This avoids problems if there were foreign sbrk calls.
2786 */
2787 current_brk = (char *) (MORECORE (0));
2788 if (current_brk == (char *) (av->top) + top_size)
2789 {
2790 /*
2791 Attempt to release memory. We ignore MORECORE return value,
2792 and instead call again to find out where new end of memory is.
2793 This avoids problems if first call releases less than we asked,
2794 of if failure somehow altered brk value. (We could still
2795 encounter problems if it altered brk in some very bad way,
2796 but the only thing we can do is adjust anyway, which will cause
2797 some downstream failure.)
2798 */
2799
2800 MORECORE (-extra);
2801 /* Call the `morecore' hook if necessary. */
2802 void (*hook) (void) = atomic_forced_read (__after_morecore_hook);
2803 if (__builtin_expect (hook != NULL, 0))
2804 (*hook)();
2805 new_brk = (char *) (MORECORE (0));
2806
2807 LIBC_PROBE (memory_sbrk_less, 2, new_brk, extra);
2808
2809 if (new_brk != (char *) MORECORE_FAILURE)
2810 {
2811 released = (long) (current_brk - new_brk);
2812
2813 if (released != 0)
2814 {
2815 /* Success. Adjust top. */
2816 av->system_mem -= released;
2817 set_head (av->top, (top_size - released) | PREV_INUSE);
2818 check_malloc_state (av);
2819 return 1;
2820 }
2821 }
2822 }
2823 return 0;
2824}
2825
2826static void
2827munmap_chunk (mchunkptr p)
2828{
2829 INTERNAL_SIZE_T size = chunksize (p);
2830
2831 assert (chunk_is_mmapped (p));
2832
2833 /* Do nothing if the chunk is a faked mmapped chunk in the dumped
2834 main arena. We never free this memory. */
2835 if (DUMPED_MAIN_ARENA_CHUNK (p))
2836 return;
2837
2838 uintptr_t block = (uintptr_t) p - prev_size (p);
2839 size_t total_size = prev_size (p) + size;
2840 /* Unfortunately we have to do the compilers job by hand here. Normally
2841 we would test BLOCK and TOTAL-SIZE separately for compliance with the
2842 page size. But gcc does not recognize the optimization possibility
2843 (in the moment at least) so we combine the two values into one before
2844 the bit test. */
2845 if (__builtin_expect (((block | total_size) & (GLRO (dl_pagesize) - 1)) != 0, 0))
2846 malloc_printerr ("munmap_chunk(): invalid pointer");
2847
2848 atomic_decrement (&mp_.n_mmaps);
2849 atomic_add (&mp_.mmapped_mem, -total_size);
2850
2851 /* If munmap failed the process virtual memory address space is in a
2852 bad shape. Just leave the block hanging around, the process will
2853 terminate shortly anyway since not much can be done. */
2854 __munmap ((char *) block, total_size);
2855}
2856
2857#if HAVE_MREMAP
2858
2859static mchunkptr
2860mremap_chunk (mchunkptr p, size_t new_size)
2861{
2862 size_t pagesize = GLRO (dl_pagesize);
2863 INTERNAL_SIZE_T offset = prev_size (p);
2864 INTERNAL_SIZE_T size = chunksize (p);
2865 char *cp;
2866
2867 assert (chunk_is_mmapped (p));
2868 assert (((size + offset) & (GLRO (dl_pagesize) - 1)) == 0);
2869
2870 /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */
2871 new_size = ALIGN_UP (new_size + offset + SIZE_SZ, pagesize);
2872
2873 /* No need to remap if the number of pages does not change. */
2874 if (size + offset == new_size)
2875 return p;
2876
2877 cp = (char *) __mremap ((char *) p - offset, size + offset, new_size,
2878 MREMAP_MAYMOVE);
2879
2880 if (cp == MAP_FAILED)
2881 return 0;
2882
2883 p = (mchunkptr) (cp + offset);
2884
2885 assert (aligned_OK (chunk2mem (p)));
2886
2887 assert (prev_size (p) == offset);
2888 set_head (p, (new_size - offset) | IS_MMAPPED);
2889
2890 INTERNAL_SIZE_T new;
2891 new = atomic_exchange_and_add (&mp_.mmapped_mem, new_size - size - offset)
2892 + new_size - size - offset;
2893 atomic_max (&mp_.max_mmapped_mem, new);
2894 return p;
2895}
2896#endif /* HAVE_MREMAP */
2897
2898/*------------------------ Public wrappers. --------------------------------*/
2899
2900#if USE_TCACHE
2901
2902/* We overlay this structure on the user-data portion of a chunk when
2903 the chunk is stored in the per-thread cache. */
2904typedef struct tcache_entry
2905{
2906 struct tcache_entry *next;
2907} tcache_entry;
2908
2909/* There is one of these for each thread, which contains the
2910 per-thread cache (hence "tcache_perthread_struct"). Keeping
2911 overall size low is mildly important. Note that COUNTS and ENTRIES
2912 are redundant (we could have just counted the linked list each
2913 time), this is for performance reasons. */
2914typedef struct tcache_perthread_struct
2915{
2916 char counts[TCACHE_MAX_BINS];
2917 tcache_entry *entries[TCACHE_MAX_BINS];
2918} tcache_perthread_struct;
2919
2920static __thread bool tcache_shutting_down = false;
2921static __thread tcache_perthread_struct *tcache = NULL;
2922
2923/* Caller must ensure that we know tc_idx is valid and there's room
2924 for more chunks. */
2925static __always_inline void
2926tcache_put (mchunkptr chunk, size_t tc_idx)
2927{
2928 tcache_entry *e = (tcache_entry *) chunk2mem (chunk);
2929 assert (tc_idx < TCACHE_MAX_BINS);
2930 e->next = tcache->entries[tc_idx];
2931 tcache->entries[tc_idx] = e;
2932 ++(tcache->counts[tc_idx]);
2933}
2934
2935/* Caller must ensure that we know tc_idx is valid and there's
2936 available chunks to remove. */
2937static __always_inline void *
2938tcache_get (size_t tc_idx)
2939{
2940 tcache_entry *e = tcache->entries[tc_idx];
2941 assert (tc_idx < TCACHE_MAX_BINS);
2942 assert (tcache->entries[tc_idx] > 0);
2943 tcache->entries[tc_idx] = e->next;
2944 --(tcache->counts[tc_idx]);
2945 return (void *) e;
2946}
2947
2948static void
2949tcache_thread_shutdown (void)
2950{
2951 int i;
2952 tcache_perthread_struct *tcache_tmp = tcache;
2953
2954 if (!tcache)
2955 return;
2956
2957 /* Disable the tcache and prevent it from being reinitialized. */
2958 tcache = NULL;
2959 tcache_shutting_down = true;
2960
2961 /* Free all of the entries and the tcache itself back to the arena
2962 heap for coalescing. */
2963 for (i = 0; i < TCACHE_MAX_BINS; ++i)
2964 {
2965 while (tcache_tmp->entries[i])
2966 {
2967 tcache_entry *e = tcache_tmp->entries[i];
2968 tcache_tmp->entries[i] = e->next;
2969 __libc_free (e);
2970 }
2971 }
2972
2973 __libc_free (tcache_tmp);
2974}
2975
2976static void
2977tcache_init(void)
2978{
2979 mstate ar_ptr;
2980 void *victim = 0;
2981 const size_t bytes = sizeof (tcache_perthread_struct);
2982
2983 if (tcache_shutting_down)
2984 return;
2985
2986 arena_get (ar_ptr, bytes);
2987 victim = _int_malloc (ar_ptr, bytes);
2988 if (!victim && ar_ptr != NULL)
2989 {
2990 ar_ptr = arena_get_retry (ar_ptr, bytes);
2991 victim = _int_malloc (ar_ptr, bytes);
2992 }
2993
2994
2995 if (ar_ptr != NULL)
2996 __libc_lock_unlock (ar_ptr->mutex);
2997
2998 /* In a low memory situation, we may not be able to allocate memory
2999 - in which case, we just keep trying later. However, we
3000 typically do this very early, so either there is sufficient
3001 memory, or there isn't enough memory to do non-trivial
3002 allocations anyway. */
3003 if (victim)
3004 {
3005 tcache = (tcache_perthread_struct *) victim;
3006 memset (tcache, 0, sizeof (tcache_perthread_struct));
3007 }
3008
3009}
3010
3011# define MAYBE_INIT_TCACHE() \
3012 if (__glibc_unlikely (tcache == NULL)) \
3013 tcache_init();
3014
3015#else /* !USE_TCACHE */
3016# define MAYBE_INIT_TCACHE()
3017
3018static void
3019tcache_thread_shutdown (void)
3020{
3021 /* Nothing to do if there is no thread cache. */
3022}
3023
3024#endif /* !USE_TCACHE */
3025
3026void *
3027__libc_malloc (size_t bytes)
3028{
3029 mstate ar_ptr;
3030 void *victim;
3031
3032 void *(*hook) (size_t, const void *)
3033 = atomic_forced_read (__malloc_hook);
3034 if (__builtin_expect (hook != NULL, 0))
3035 return (*hook)(bytes, RETURN_ADDRESS (0));
3036#if USE_TCACHE
3037 /* int_free also calls request2size, be careful to not pad twice. */
3038 size_t tbytes;
3039 checked_request2size (bytes, tbytes);
3040 size_t tc_idx = csize2tidx (tbytes);
3041
3042 MAYBE_INIT_TCACHE ();
3043
3044 DIAG_PUSH_NEEDS_COMMENT;
3045 if (tc_idx < mp_.tcache_bins
3046 /*&& tc_idx < TCACHE_MAX_BINS*/ /* to appease gcc */
3047 && tcache
3048 && tcache->entries[tc_idx] != NULL)
3049 {
3050 return tcache_get (tc_idx);
3051 }
3052 DIAG_POP_NEEDS_COMMENT;
3053#endif
3054
3055 if (SINGLE_THREAD_P)
3056 {
3057 victim = _int_malloc (&main_arena, bytes);
3058 assert (!victim || chunk_is_mmapped (mem2chunk (victim)) ||
3059 &main_arena == arena_for_chunk (mem2chunk (victim)));
3060 return victim;
3061 }
3062
3063 arena_get (ar_ptr, bytes);
3064
3065 victim = _int_malloc (ar_ptr, bytes);
3066 /* Retry with another arena only if we were able to find a usable arena
3067 before. */
3068 if (!victim && ar_ptr != NULL)
3069 {
3070 LIBC_PROBE (memory_malloc_retry, 1, bytes);
3071 ar_ptr = arena_get_retry (ar_ptr, bytes);
3072 victim = _int_malloc (ar_ptr, bytes);
3073 }
3074
3075 if (ar_ptr != NULL)
3076 __libc_lock_unlock (ar_ptr->mutex);
3077
3078 assert (!victim || chunk_is_mmapped (mem2chunk (victim)) ||
3079 ar_ptr == arena_for_chunk (mem2chunk (victim)));
3080 return victim;
3081}
3082libc_hidden_def (__libc_malloc)
3083
3084void
3085__libc_free (void *mem)
3086{
3087 mstate ar_ptr;
3088 mchunkptr p; /* chunk corresponding to mem */
3089
3090 void (*hook) (void *, const void *)
3091 = atomic_forced_read (__free_hook);
3092 if (__builtin_expect (hook != NULL, 0))
3093 {
3094 (*hook)(mem, RETURN_ADDRESS (0));
3095 return;
3096 }
3097
3098 if (mem == 0) /* free(0) has no effect */
3099 return;
3100
3101 p = mem2chunk (mem);
3102
3103 if (chunk_is_mmapped (p)) /* release mmapped memory. */
3104 {
3105 /* See if the dynamic brk/mmap threshold needs adjusting.
3106 Dumped fake mmapped chunks do not affect the threshold. */
3107 if (!mp_.no_dyn_threshold
3108 && chunksize_nomask (p) > mp_.mmap_threshold
3109 && chunksize_nomask (p) <= DEFAULT_MMAP_THRESHOLD_MAX
3110 && !DUMPED_MAIN_ARENA_CHUNK (p))
3111 {
3112 mp_.mmap_threshold = chunksize (p);
3113 mp_.trim_threshold = 2 * mp_.mmap_threshold;
3114 LIBC_PROBE (memory_mallopt_free_dyn_thresholds, 2,
3115 mp_.mmap_threshold, mp_.trim_threshold);
3116 }
3117 munmap_chunk (p);
3118 return;
3119 }
3120
3121 MAYBE_INIT_TCACHE ();
3122
3123 ar_ptr = arena_for_chunk (p);
3124 _int_free (ar_ptr, p, 0);
3125}
3126libc_hidden_def (__libc_free)
3127
3128void *
3129__libc_realloc (void *oldmem, size_t bytes)
3130{
3131 mstate ar_ptr;
3132 INTERNAL_SIZE_T nb; /* padded request size */
3133
3134 void *newp; /* chunk to return */
3135
3136 void *(*hook) (void *, size_t, const void *) =
3137 atomic_forced_read (__realloc_hook);
3138 if (__builtin_expect (hook != NULL, 0))
3139 return (*hook)(oldmem, bytes, RETURN_ADDRESS (0));
3140
3141#if REALLOC_ZERO_BYTES_FREES
3142 if (bytes == 0 && oldmem != NULL)
3143 {
3144 __libc_free (oldmem); return 0;
3145 }
3146#endif
3147
3148 /* realloc of null is supposed to be same as malloc */
3149 if (oldmem == 0)
3150 return __libc_malloc (bytes);
3151
3152 /* chunk corresponding to oldmem */
3153 const mchunkptr oldp = mem2chunk (oldmem);
3154 /* its size */
3155 const INTERNAL_SIZE_T oldsize = chunksize (oldp);
3156
3157 if (chunk_is_mmapped (oldp))
3158 ar_ptr = NULL;
3159 else
3160 {
3161 MAYBE_INIT_TCACHE ();
3162 ar_ptr = arena_for_chunk (oldp);
3163 }
3164
3165 /* Little security check which won't hurt performance: the allocator
3166 never wrapps around at the end of the address space. Therefore
3167 we can exclude some size values which might appear here by
3168 accident or by "design" from some intruder. We need to bypass
3169 this check for dumped fake mmap chunks from the old main arena
3170 because the new malloc may provide additional alignment. */
3171 if ((__builtin_expect ((uintptr_t) oldp > (uintptr_t) -oldsize, 0)
3172 || __builtin_expect (misaligned_chunk (oldp), 0))
3173 && !DUMPED_MAIN_ARENA_CHUNK (oldp))
3174 malloc_printerr ("realloc(): invalid pointer");
3175
3176 checked_request2size (bytes, nb);
3177
3178 if (chunk_is_mmapped (oldp))
3179 {
3180 /* If this is a faked mmapped chunk from the dumped main arena,
3181 always make a copy (and do not free the old chunk). */
3182 if (DUMPED_MAIN_ARENA_CHUNK (oldp))
3183 {
3184 /* Must alloc, copy, free. */
3185 void *newmem = __libc_malloc (bytes);
3186 if (newmem == 0)
3187 return NULL;
3188 /* Copy as many bytes as are available from the old chunk
3189 and fit into the new size. NB: The overhead for faked
3190 mmapped chunks is only SIZE_SZ, not 2 * SIZE_SZ as for
3191 regular mmapped chunks. */
3192 if (bytes > oldsize - SIZE_SZ)
3193 bytes = oldsize - SIZE_SZ;
3194 memcpy (newmem, oldmem, bytes);
3195 return newmem;
3196 }
3197
3198 void *newmem;
3199
3200#if HAVE_MREMAP
3201 newp = mremap_chunk (oldp, nb);
3202 if (newp)
3203 return chunk2mem (newp);
3204#endif
3205 /* Note the extra SIZE_SZ overhead. */
3206 if (oldsize - SIZE_SZ >= nb)
3207 return oldmem; /* do nothing */
3208
3209 /* Must alloc, copy, free. */
3210 newmem = __libc_malloc (bytes);
3211 if (newmem == 0)
3212 return 0; /* propagate failure */
3213
3214 memcpy (newmem, oldmem, oldsize - 2 * SIZE_SZ);
3215 munmap_chunk (oldp);
3216 return newmem;
3217 }
3218
3219 if (SINGLE_THREAD_P)
3220 {
3221 newp = _int_realloc (ar_ptr, oldp, oldsize, nb);
3222 assert (!newp || chunk_is_mmapped (mem2chunk (newp)) ||
3223 ar_ptr == arena_for_chunk (mem2chunk (newp)));
3224
3225 return newp;
3226 }
3227
3228 __libc_lock_lock (ar_ptr->mutex);
3229
3230 newp = _int_realloc (ar_ptr, oldp, oldsize, nb);
3231
3232 __libc_lock_unlock (ar_ptr->mutex);
3233 assert (!newp || chunk_is_mmapped (mem2chunk (newp)) ||
3234 ar_ptr == arena_for_chunk (mem2chunk (newp)));
3235
3236 if (newp == NULL)
3237 {
3238 /* Try harder to allocate memory in other arenas. */
3239 LIBC_PROBE (memory_realloc_retry, 2, bytes, oldmem);
3240 newp = __libc_malloc (bytes);
3241 if (newp != NULL)
3242 {
3243 memcpy (newp, oldmem, oldsize - SIZE_SZ);
3244 _int_free (ar_ptr, oldp, 0);
3245 }
3246 }
3247
3248 return newp;
3249}
3250libc_hidden_def (__libc_realloc)
3251
3252void *
3253__libc_memalign (size_t alignment, size_t bytes)
3254{
3255 void *address = RETURN_ADDRESS (0);
3256 return _mid_memalign (alignment, bytes, address);
3257}
3258
3259static void *
3260_mid_memalign (size_t alignment, size_t bytes, void *address)
3261{
3262 mstate ar_ptr;
3263 void *p;
3264
3265 void *(*hook) (size_t, size_t, const void *) =
3266 atomic_forced_read (__memalign_hook);
3267 if (__builtin_expect (hook != NULL, 0))
3268 return (*hook)(alignment, bytes, address);
3269
3270 /* If we need less alignment than we give anyway, just relay to malloc. */
3271 if (alignment <= MALLOC_ALIGNMENT)
3272 return __libc_malloc (bytes);
3273
3274 /* Otherwise, ensure that it is at least a minimum chunk size */
3275 if (alignment < MINSIZE)
3276 alignment = MINSIZE;
3277
3278 /* If the alignment is greater than SIZE_MAX / 2 + 1 it cannot be a
3279 power of 2 and will cause overflow in the check below. */
3280 if (alignment > SIZE_MAX / 2 + 1)
3281 {
3282 __set_errno (EINVAL);
3283 return 0;
3284 }
3285
3286 /* Check for overflow. */
3287 if (bytes > SIZE_MAX - alignment - MINSIZE)
3288 {
3289 __set_errno (ENOMEM);
3290 return 0;
3291 }
3292
3293
3294 /* Make sure alignment is power of 2. */
3295 if (!powerof2 (alignment))
3296 {
3297 size_t a = MALLOC_ALIGNMENT * 2;
3298 while (a < alignment)
3299 a <<= 1;
3300 alignment = a;
3301 }
3302
3303 if (SINGLE_THREAD_P)
3304 {
3305 p = _int_memalign (&main_arena, alignment, bytes);
3306 assert (!p || chunk_is_mmapped (mem2chunk (p)) ||
3307 &main_arena == arena_for_chunk (mem2chunk (p)));
3308
3309 return p;
3310 }
3311
3312 arena_get (ar_ptr, bytes + alignment + MINSIZE);
3313
3314 p = _int_memalign (ar_ptr, alignment, bytes);
3315 if (!p && ar_ptr != NULL)
3316 {
3317 LIBC_PROBE (memory_memalign_retry, 2, bytes, alignment);
3318 ar_ptr = arena_get_retry (ar_ptr, bytes);
3319 p = _int_memalign (ar_ptr, alignment, bytes);
3320 }
3321
3322 if (ar_ptr != NULL)
3323 __libc_lock_unlock (ar_ptr->mutex);
3324
3325 assert (!p || chunk_is_mmapped (mem2chunk (p)) ||
3326 ar_ptr == arena_for_chunk (mem2chunk (p)));
3327 return p;
3328}
3329/* For ISO C11. */
3330weak_alias (__libc_memalign, aligned_alloc)
3331libc_hidden_def (__libc_memalign)
3332
3333void *
3334__libc_valloc (size_t bytes)
3335{
3336 if (__malloc_initialized < 0)
3337 ptmalloc_init ();
3338
3339 void *address = RETURN_ADDRESS (0);
3340 size_t pagesize = GLRO (dl_pagesize);
3341 return _mid_memalign (pagesize, bytes, address);
3342}
3343
3344void *
3345__libc_pvalloc (size_t bytes)
3346{
3347 if (__malloc_initialized < 0)
3348 ptmalloc_init ();
3349
3350 void *address = RETURN_ADDRESS (0);
3351 size_t pagesize = GLRO (dl_pagesize);
3352 size_t rounded_bytes = ALIGN_UP (bytes, pagesize);
3353
3354 /* Check for overflow. */
3355 if (bytes > SIZE_MAX - 2 * pagesize - MINSIZE)
3356 {
3357 __set_errno (ENOMEM);
3358 return 0;
3359 }
3360
3361 return _mid_memalign (pagesize, rounded_bytes, address);
3362}
3363
3364void *
3365__libc_calloc (size_t n, size_t elem_size)
3366{
3367 mstate av;
3368 mchunkptr oldtop, p;
3369 INTERNAL_SIZE_T bytes, sz, csz, oldtopsize;
3370 void *mem;
3371 unsigned long clearsize;
3372 unsigned long nclears;
3373 INTERNAL_SIZE_T *d;
3374
3375 /* size_t is unsigned so the behavior on overflow is defined. */
3376 bytes = n * elem_size;
3377#define HALF_INTERNAL_SIZE_T \
3378 (((INTERNAL_SIZE_T) 1) << (8 * sizeof (INTERNAL_SIZE_T) / 2))
3379 if (__builtin_expect ((n | elem_size) >= HALF_INTERNAL_SIZE_T, 0))
3380 {
3381 if (elem_size != 0 && bytes / elem_size != n)
3382 {
3383 __set_errno (ENOMEM);
3384 return 0;
3385 }
3386 }
3387
3388 void *(*hook) (size_t, const void *) =
3389 atomic_forced_read (__malloc_hook);
3390 if (__builtin_expect (hook != NULL, 0))
3391 {
3392 sz = bytes;
3393 mem = (*hook)(sz, RETURN_ADDRESS (0));
3394 if (mem == 0)
3395 return 0;
3396
3397 return memset (mem, 0, sz);
3398 }
3399
3400 sz = bytes;
3401
3402 MAYBE_INIT_TCACHE ();
3403
3404 if (SINGLE_THREAD_P)
3405 av = &main_arena;
3406 else
3407 arena_get (av, sz);
3408
3409 if (av)
3410 {
3411 /* Check if we hand out the top chunk, in which case there may be no
3412 need to clear. */
3413#if MORECORE_CLEARS
3414 oldtop = top (av);
3415 oldtopsize = chunksize (top (av));
3416# if MORECORE_CLEARS < 2
3417 /* Only newly allocated memory is guaranteed to be cleared. */
3418 if (av == &main_arena &&
3419 oldtopsize < mp_.sbrk_base + av->max_system_mem - (char *) oldtop)
3420 oldtopsize = (mp_.sbrk_base + av->max_system_mem - (char *) oldtop);
3421# endif
3422 if (av != &main_arena)
3423 {
3424 heap_info *heap = heap_for_ptr (oldtop);
3425 if (oldtopsize < (char *) heap + heap->mprotect_size - (char *) oldtop)
3426 oldtopsize = (char *) heap + heap->mprotect_size - (char *) oldtop;
3427 }
3428#endif
3429 }
3430 else
3431 {
3432 /* No usable arenas. */
3433 oldtop = 0;
3434 oldtopsize = 0;
3435 }
3436 mem = _int_malloc (av, sz);
3437
3438 assert (!mem || chunk_is_mmapped (mem2chunk (mem)) ||
3439 av == arena_for_chunk (mem2chunk (mem)));
3440
3441 if (!SINGLE_THREAD_P)
3442 {
3443 if (mem == 0 && av != NULL)
3444 {
3445 LIBC_PROBE (memory_calloc_retry, 1, sz);
3446 av = arena_get_retry (av, sz);
3447 mem = _int_malloc (av, sz);
3448 }
3449
3450 if (av != NULL)
3451 __libc_lock_unlock (av->mutex);
3452 }
3453
3454 /* Allocation failed even after a retry. */
3455 if (mem == 0)
3456 return 0;
3457
3458 p = mem2chunk (mem);
3459
3460 /* Two optional cases in which clearing not necessary */
3461 if (chunk_is_mmapped (p))
3462 {
3463 if (__builtin_expect (perturb_byte, 0))
3464 return memset (mem, 0, sz);
3465
3466 return mem;
3467 }
3468
3469 csz = chunksize (p);
3470
3471#if MORECORE_CLEARS
3472 if (perturb_byte == 0 && (p == oldtop && csz > oldtopsize))
3473 {
3474 /* clear only the bytes from non-freshly-sbrked memory */
3475 csz = oldtopsize;
3476 }
3477#endif
3478
3479 /* Unroll clear of <= 36 bytes (72 if 8byte sizes). We know that
3480 contents have an odd number of INTERNAL_SIZE_T-sized words;
3481 minimally 3. */
3482 d = (INTERNAL_SIZE_T *) mem;
3483 clearsize = csz - SIZE_SZ;
3484 nclears = clearsize / sizeof (INTERNAL_SIZE_T);
3485 assert (nclears >= 3);
3486
3487 if (nclears > 9)
3488 return memset (d, 0, clearsize);
3489
3490 else
3491 {
3492 *(d + 0) = 0;
3493 *(d + 1) = 0;
3494 *(d + 2) = 0;
3495 if (nclears > 4)
3496 {
3497 *(d + 3) = 0;
3498 *(d + 4) = 0;
3499 if (nclears > 6)
3500 {
3501 *(d + 5) = 0;
3502 *(d + 6) = 0;
3503 if (nclears > 8)
3504 {
3505 *(d + 7) = 0;
3506 *(d + 8) = 0;
3507 }
3508 }
3509 }
3510 }
3511
3512 return mem;
3513}
3514
3515/*
3516 ------------------------------ malloc ------------------------------
3517 */
3518
3519static void *
3520_int_malloc (mstate av, size_t bytes)
3521{
3522 INTERNAL_SIZE_T nb; /* normalized request size */
3523 unsigned int idx; /* associated bin index */
3524 mbinptr bin; /* associated bin */
3525
3526 mchunkptr victim; /* inspected/selected chunk */
3527 INTERNAL_SIZE_T size; /* its size */
3528 int victim_index; /* its bin index */
3529
3530 mchunkptr remainder; /* remainder from a split */
3531 unsigned long remainder_size; /* its size */
3532
3533 unsigned int block; /* bit map traverser */
3534 unsigned int bit; /* bit map traverser */
3535 unsigned int map; /* current word of binmap */
3536
3537 mchunkptr fwd; /* misc temp for linking */
3538 mchunkptr bck; /* misc temp for linking */
3539
3540#if USE_TCACHE
3541 size_t tcache_unsorted_count; /* count of unsorted chunks processed */
3542#endif
3543
3544 /*
3545 Convert request size to internal form by adding SIZE_SZ bytes
3546 overhead plus possibly more to obtain necessary alignment and/or
3547 to obtain a size of at least MINSIZE, the smallest allocatable
3548 size. Also, checked_request2size traps (returning 0) request sizes
3549 that are so large that they wrap around zero when padded and
3550 aligned.
3551 */
3552
3553 checked_request2size (bytes, nb);
3554
3555 /* There are no usable arenas. Fall back to sysmalloc to get a chunk from
3556 mmap. */
3557 if (__glibc_unlikely (av == NULL))
3558 {
3559 void *p = sysmalloc (nb, av);
3560 if (p != NULL)
3561 alloc_perturb (p, bytes);
3562 return p;
3563 }
3564
3565 /*
3566 If the size qualifies as a fastbin, first check corresponding bin.
3567 This code is safe to execute even if av is not yet initialized, so we
3568 can try it without checking, which saves some time on this fast path.
3569 */
3570
3571#define REMOVE_FB(fb, victim, pp) \
3572 do \
3573 { \
3574 victim = pp; \
3575 if (victim == NULL) \
3576 break; \
3577 } \
3578 while ((pp = catomic_compare_and_exchange_val_acq (fb, victim->fd, victim)) \
3579 != victim); \
3580
3581 if ((unsigned long) (nb) <= (unsigned long) (get_max_fast ()))
3582 {
3583 idx = fastbin_index (nb);
3584 mfastbinptr *fb = &fastbin (av, idx);
3585 mchunkptr pp;
3586 victim = *fb;
3587
3588 if (victim != NULL)
3589 {
3590 if (SINGLE_THREAD_P)
3591 *fb = victim->fd;
3592 else
3593 REMOVE_FB (fb, pp, victim);
3594 if (__glibc_likely (victim != NULL))
3595 {
3596 size_t victim_idx = fastbin_index (chunksize (victim));
3597 if (__builtin_expect (victim_idx != idx, 0))
3598 malloc_printerr ("malloc(): memory corruption (fast)");
3599 check_remalloced_chunk (av, victim, nb);
3600#if USE_TCACHE
3601 /* While we're here, if we see other chunks of the same size,
3602 stash them in the tcache. */
3603 size_t tc_idx = csize2tidx (nb);
3604 if (tcache && tc_idx < mp_.tcache_bins)
3605 {
3606 mchunkptr tc_victim;
3607
3608 /* While bin not empty and tcache not full, copy chunks. */
3609 while (tcache->counts[tc_idx] < mp_.tcache_count
3610 && (tc_victim = *fb) != NULL)
3611 {
3612 if (SINGLE_THREAD_P)
3613 *fb = tc_victim->fd;
3614 else
3615 {
3616 REMOVE_FB (fb, pp, tc_victim);
3617 if (__glibc_unlikely (tc_victim == NULL))
3618 break;
3619 }
3620 tcache_put (tc_victim, tc_idx);
3621 }
3622 }
3623#endif
3624 void *p = chunk2mem (victim);
3625 alloc_perturb (p, bytes);
3626 return p;
3627 }
3628 }
3629 }
3630
3631 /*
3632 If a small request, check regular bin. Since these "smallbins"
3633 hold one size each, no searching within bins is necessary.
3634 (For a large request, we need to wait until unsorted chunks are
3635 processed to find best fit. But for small ones, fits are exact
3636 anyway, so we can check now, which is faster.)
3637 */
3638
3639 if (in_smallbin_range (nb))
3640 {
3641 idx = smallbin_index (nb);
3642 bin = bin_at (av, idx);
3643
3644 if ((victim = last (bin)) != bin)
3645 {
3646 bck = victim->bk;
3647 if (__glibc_unlikely (bck->fd != victim))
3648 malloc_printerr ("malloc(): smallbin double linked list corrupted");
3649 set_inuse_bit_at_offset (victim, nb);
3650 bin->bk = bck;
3651 bck->fd = bin;
3652
3653 if (av != &main_arena)
3654 set_non_main_arena (victim);
3655 check_malloced_chunk (av, victim, nb);
3656#if USE_TCACHE
3657 /* While we're here, if we see other chunks of the same size,
3658 stash them in the tcache. */
3659 size_t tc_idx = csize2tidx (nb);
3660 if (tcache && tc_idx < mp_.tcache_bins)
3661 {
3662 mchunkptr tc_victim;
3663
3664 /* While bin not empty and tcache not full, copy chunks over. */
3665 while (tcache->counts[tc_idx] < mp_.tcache_count
3666 && (tc_victim = last (bin)) != bin)
3667 {
3668 if (tc_victim != 0)
3669 {
3670 bck = tc_victim->bk;
3671 set_inuse_bit_at_offset (tc_victim, nb);
3672 if (av != &main_arena)
3673 set_non_main_arena (tc_victim);
3674 bin->bk = bck;
3675 bck->fd = bin;
3676
3677 tcache_put (tc_victim, tc_idx);
3678 }
3679 }
3680 }
3681#endif
3682 void *p = chunk2mem (victim);
3683 alloc_perturb (p, bytes);
3684 return p;
3685 }
3686 }
3687
3688 /*
3689 If this is a large request, consolidate fastbins before continuing.
3690 While it might look excessive to kill all fastbins before
3691 even seeing if there is space available, this avoids
3692 fragmentation problems normally associated with fastbins.
3693 Also, in practice, programs tend to have runs of either small or
3694 large requests, but less often mixtures, so consolidation is not
3695 invoked all that often in most programs. And the programs that
3696 it is called frequently in otherwise tend to fragment.
3697 */
3698
3699 else
3700 {
3701 idx = largebin_index (nb);
3702 if (atomic_load_relaxed (&av->have_fastchunks))
3703 malloc_consolidate (av);
3704 }
3705
3706 /*
3707 Process recently freed or remaindered chunks, taking one only if
3708 it is exact fit, or, if this a small request, the chunk is remainder from
3709 the most recent non-exact fit. Place other traversed chunks in
3710 bins. Note that this step is the only place in any routine where
3711 chunks are placed in bins.
3712
3713 The outer loop here is needed because we might not realize until
3714 near the end of malloc that we should have consolidated, so must
3715 do so and retry. This happens at most once, and only when we would
3716 otherwise need to expand memory to service a "small" request.
3717 */
3718
3719#if USE_TCACHE
3720 INTERNAL_SIZE_T tcache_nb = 0;
3721 size_t tc_idx = csize2tidx (nb);
3722 if (tcache && tc_idx < mp_.tcache_bins)
3723 tcache_nb = nb;
3724 int return_cached = 0;
3725
3726 tcache_unsorted_count = 0;
3727#endif
3728
3729 for (;; )
3730 {
3731 int iters = 0;
3732 while ((victim = unsorted_chunks (av)->bk) != unsorted_chunks (av))
3733 {
3734 bck = victim->bk;
3735 if (__builtin_expect (chunksize_nomask (victim) <= 2 * SIZE_SZ, 0)
3736 || __builtin_expect (chunksize_nomask (victim)
3737 > av->system_mem, 0))
3738 malloc_printerr ("malloc(): memory corruption");
3739 size = chunksize (victim);
3740
3741 /*
3742 If a small request, try to use last remainder if it is the
3743 only chunk in unsorted bin. This helps promote locality for
3744 runs of consecutive small requests. This is the only
3745 exception to best-fit, and applies only when there is
3746 no exact fit for a small chunk.
3747 */
3748
3749 if (in_smallbin_range (nb) &&
3750 bck == unsorted_chunks (av) &&
3751 victim == av->last_remainder &&
3752 (unsigned long) (size) > (unsigned long) (nb + MINSIZE))
3753 {
3754 /* split and reattach remainder */
3755 remainder_size = size - nb;
3756 remainder = chunk_at_offset (victim, nb);
3757 unsorted_chunks (av)->bk = unsorted_chunks (av)->fd = remainder;
3758 av->last_remainder = remainder;
3759 remainder->bk = remainder->fd = unsorted_chunks (av);
3760 if (!in_smallbin_range (remainder_size))
3761 {
3762 remainder->fd_nextsize = NULL;
3763 remainder->bk_nextsize = NULL;
3764 }
3765
3766 set_head (victim, nb | PREV_INUSE |
3767 (av != &main_arena ? NON_MAIN_ARENA : 0));
3768 set_head (remainder, remainder_size | PREV_INUSE);
3769 set_foot (remainder, remainder_size);
3770
3771 check_malloced_chunk (av, victim, nb);
3772 void *p = chunk2mem (victim);
3773 alloc_perturb (p, bytes);
3774 return p;
3775 }
3776
3777 /* remove from unsorted list */
3778 if (__glibc_unlikely (bck->fd != victim))
3779 malloc_printerr ("malloc(): corrupted unsorted chunks 3");
3780 unsorted_chunks (av)->bk = bck;
3781 bck->fd = unsorted_chunks (av);
3782
3783 /* Take now instead of binning if exact fit */
3784
3785 if (size == nb)
3786 {
3787 set_inuse_bit_at_offset (victim, size);
3788 if (av != &main_arena)
3789 set_non_main_arena (victim);
3790#if USE_TCACHE
3791 /* Fill cache first, return to user only if cache fills.
3792 We may return one of these chunks later. */
3793 if (tcache_nb
3794 && tcache->counts[tc_idx] < mp_.tcache_count)
3795 {
3796 tcache_put (victim, tc_idx);
3797 return_cached = 1;
3798 continue;
3799 }
3800 else
3801 {
3802#endif
3803 check_malloced_chunk (av, victim, nb);
3804 void *p = chunk2mem (victim);
3805 alloc_perturb (p, bytes);
3806 return p;
3807#if USE_TCACHE
3808 }
3809#endif
3810 }
3811
3812 /* place chunk in bin */
3813
3814 if (in_smallbin_range (size))
3815 {
3816 victim_index = smallbin_index (size);
3817 bck = bin_at (av, victim_index);
3818 fwd = bck->fd;
3819 }
3820 else
3821 {
3822 victim_index = largebin_index (size);
3823 bck = bin_at (av, victim_index);
3824 fwd = bck->fd;
3825
3826 /* maintain large bins in sorted order */
3827 if (fwd != bck)
3828 {
3829 /* Or with inuse bit to speed comparisons */
3830 size |= PREV_INUSE;
3831 /* if smaller than smallest, bypass loop below */
3832 assert (chunk_main_arena (bck->bk));
3833 if ((unsigned long) (size)
3834 < (unsigned long) chunksize_nomask (bck->bk))
3835 {
3836 fwd = bck;
3837 bck = bck->bk;
3838
3839 victim->fd_nextsize = fwd->fd;
3840 victim->bk_nextsize = fwd->fd->bk_nextsize;
3841 fwd->fd->bk_nextsize = victim->bk_nextsize->fd_nextsize = victim;
3842 }
3843 else
3844 {
3845 assert (chunk_main_arena (fwd));
3846 while ((unsigned long) size < chunksize_nomask (fwd))
3847 {
3848 fwd = fwd->fd_nextsize;
3849 assert (chunk_main_arena (fwd));
3850 }
3851
3852 if ((unsigned long) size
3853 == (unsigned long) chunksize_nomask (fwd))
3854 /* Always insert in the second position. */
3855 fwd = fwd->fd;
3856 else
3857 {
3858 victim->fd_nextsize = fwd;
3859 victim->bk_nextsize = fwd->bk_nextsize;
3860 fwd->bk_nextsize = victim;
3861 victim->bk_nextsize->fd_nextsize = victim;
3862 }
3863 bck = fwd->bk;
3864 }
3865 }
3866 else
3867 victim->fd_nextsize = victim->bk_nextsize = victim;
3868 }
3869
3870 mark_bin (av, victim_index);
3871 victim->bk = bck;
3872 victim->fd = fwd;
3873 fwd->bk = victim;
3874 bck->fd = victim;
3875
3876#if USE_TCACHE
3877 /* If we've processed as many chunks as we're allowed while
3878 filling the cache, return one of the cached ones. */
3879 ++tcache_unsorted_count;
3880 if (return_cached
3881 && mp_.tcache_unsorted_limit > 0
3882 && tcache_unsorted_count > mp_.tcache_unsorted_limit)
3883 {
3884 return tcache_get (tc_idx);
3885 }
3886#endif
3887
3888#define MAX_ITERS 10000
3889 if (++iters >= MAX_ITERS)
3890 break;
3891 }
3892
3893#if USE_TCACHE
3894 /* If all the small chunks we found ended up cached, return one now. */
3895 if (return_cached)
3896 {
3897 return tcache_get (tc_idx);
3898 }
3899#endif
3900
3901 /*
3902 If a large request, scan through the chunks of current bin in
3903 sorted order to find smallest that fits. Use the skip list for this.
3904 */
3905
3906 if (!in_smallbin_range (nb))
3907 {
3908 bin = bin_at (av, idx);
3909
3910 /* skip scan if empty or largest chunk is too small */
3911 if ((victim = first (bin)) != bin
3912 && (unsigned long) chunksize_nomask (victim)
3913 >= (unsigned long) (nb))
3914 {
3915 victim = victim->bk_nextsize;
3916 while (((unsigned long) (size = chunksize (victim)) <
3917 (unsigned long) (nb)))
3918 victim = victim->bk_nextsize;
3919
3920 /* Avoid removing the first entry for a size so that the skip
3921 list does not have to be rerouted. */
3922 if (victim != last (bin)
3923 && chunksize_nomask (victim)
3924 == chunksize_nomask (victim->fd))
3925 victim = victim->fd;
3926
3927 remainder_size = size - nb;
3928 unlink (av, victim, bck, fwd);
3929
3930 /* Exhaust */
3931 if (remainder_size < MINSIZE)
3932 {
3933 set_inuse_bit_at_offset (victim, size);
3934 if (av != &main_arena)
3935 set_non_main_arena (victim);
3936 }
3937 /* Split */
3938 else
3939 {
3940 remainder = chunk_at_offset (victim, nb);
3941 /* We cannot assume the unsorted list is empty and therefore
3942 have to perform a complete insert here. */
3943 bck = unsorted_chunks (av);
3944 fwd = bck->fd;
3945 if (__glibc_unlikely (fwd->bk != bck))
3946 malloc_printerr ("malloc(): corrupted unsorted chunks");
3947 remainder->bk = bck;
3948 remainder->fd = fwd;
3949 bck->fd = remainder;
3950 fwd->bk = remainder;
3951 if (!in_smallbin_range (remainder_size))
3952 {
3953 remainder->fd_nextsize = NULL;
3954 remainder->bk_nextsize = NULL;
3955 }
3956 set_head (victim, nb | PREV_INUSE |
3957 (av != &main_arena ? NON_MAIN_ARENA : 0));
3958 set_head (remainder, remainder_size | PREV_INUSE);
3959 set_foot (remainder, remainder_size);
3960 }
3961 check_malloced_chunk (av, victim, nb);
3962 void *p = chunk2mem (victim);
3963 alloc_perturb (p, bytes);
3964 return p;
3965 }
3966 }
3967
3968 /*
3969 Search for a chunk by scanning bins, starting with next largest
3970 bin. This search is strictly by best-fit; i.e., the smallest
3971 (with ties going to approximately the least recently used) chunk
3972 that fits is selected.
3973
3974 The bitmap avoids needing to check that most blocks are nonempty.
3975 The particular case of skipping all bins during warm-up phases
3976 when no chunks have been returned yet is faster than it might look.
3977 */
3978
3979 ++idx;
3980 bin = bin_at (av, idx);
3981 block = idx2block (idx);
3982 map = av->binmap[block];
3983 bit = idx2bit (idx);
3984
3985 for (;; )
3986 {
3987 /* Skip rest of block if there are no more set bits in this block. */
3988 if (bit > map || bit == 0)
3989 {
3990 do
3991 {
3992 if (++block >= BINMAPSIZE) /* out of bins */
3993 goto use_top;
3994 }
3995 while ((map = av->binmap[block]) == 0);
3996
3997 bin = bin_at (av, (block << BINMAPSHIFT));
3998 bit = 1;
3999 }
4000
4001 /* Advance to bin with set bit. There must be one. */
4002 while ((bit & map) == 0)
4003 {
4004 bin = next_bin (bin);
4005 bit <<= 1;
4006 assert (bit != 0);
4007 }
4008
4009 /* Inspect the bin. It is likely to be non-empty */
4010 victim = last (bin);
4011
4012 /* If a false alarm (empty bin), clear the bit. */
4013 if (victim == bin)
4014 {
4015 av->binmap[block] = map &= ~bit; /* Write through */
4016 bin = next_bin (bin);
4017 bit <<= 1;
4018 }
4019
4020 else
4021 {
4022 size = chunksize (victim);
4023
4024 /* We know the first chunk in this bin is big enough to use. */
4025 assert ((unsigned long) (size) >= (unsigned long) (nb));
4026
4027 remainder_size = size - nb;
4028
4029 /* unlink */
4030 unlink (av, victim, bck, fwd);
4031
4032 /* Exhaust */
4033 if (remainder_size < MINSIZE)
4034 {
4035 set_inuse_bit_at_offset (victim, size);
4036 if (av != &main_arena)
4037 set_non_main_arena (victim);
4038 }
4039
4040 /* Split */
4041 else
4042 {
4043 remainder = chunk_at_offset (victim, nb);
4044
4045 /* We cannot assume the unsorted list is empty and therefore
4046 have to perform a complete insert here. */
4047 bck = unsorted_chunks (av);
4048 fwd = bck->fd;
4049 if (__glibc_unlikely (fwd->bk != bck))
4050 malloc_printerr ("malloc(): corrupted unsorted chunks 2");
4051 remainder->bk = bck;
4052 remainder->fd = fwd;
4053 bck->fd = remainder;
4054 fwd->bk = remainder;
4055
4056 /* advertise as last remainder */
4057 if (in_smallbin_range (nb))
4058 av->last_remainder = remainder;
4059 if (!in_smallbin_range (remainder_size))
4060 {
4061 remainder->fd_nextsize = NULL;
4062 remainder->bk_nextsize = NULL;
4063 }
4064 set_head (victim, nb | PREV_INUSE |
4065 (av != &main_arena ? NON_MAIN_ARENA : 0));
4066 set_head (remainder, remainder_size | PREV_INUSE);
4067 set_foot (remainder, remainder_size);
4068 }
4069 check_malloced_chunk (av, victim, nb);
4070 void *p = chunk2mem (victim);
4071 alloc_perturb (p, bytes);
4072 return p;
4073 }
4074 }
4075
4076 use_top:
4077 /*
4078 If large enough, split off the chunk bordering the end of memory
4079 (held in av->top). Note that this is in accord with the best-fit
4080 search rule. In effect, av->top is treated as larger (and thus
4081 less well fitting) than any other available chunk since it can
4082 be extended to be as large as necessary (up to system
4083 limitations).
4084
4085 We require that av->top always exists (i.e., has size >=
4086 MINSIZE) after initialization, so if it would otherwise be
4087 exhausted by current request, it is replenished. (The main
4088 reason for ensuring it exists is that we may need MINSIZE space
4089 to put in fenceposts in sysmalloc.)
4090 */
4091
4092 victim = av->top;
4093 size = chunksize (victim);
4094
4095 if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE))
4096 {
4097 remainder_size = size - nb;
4098 remainder = chunk_at_offset (victim, nb);
4099 av->top = remainder;
4100 set_head (victim, nb | PREV_INUSE |
4101 (av != &main_arena ? NON_MAIN_ARENA : 0));
4102 set_head (remainder, remainder_size | PREV_INUSE);
4103
4104 check_malloced_chunk (av, victim, nb);
4105 void *p = chunk2mem (victim);
4106 alloc_perturb (p, bytes);
4107 return p;
4108 }
4109
4110 /* When we are using atomic ops to free fast chunks we can get
4111 here for all block sizes. */
4112 else if (atomic_load_relaxed (&av->have_fastchunks))
4113 {
4114 malloc_consolidate (av);
4115 /* restore original bin index */
4116 if (in_smallbin_range (nb))
4117 idx = smallbin_index (nb);
4118 else
4119 idx = largebin_index (nb);
4120 }
4121
4122 /*
4123 Otherwise, relay to handle system-dependent cases
4124 */
4125 else
4126 {
4127 void *p = sysmalloc (nb, av);
4128 if (p != NULL)
4129 alloc_perturb (p, bytes);
4130 return p;
4131 }
4132 }
4133}
4134
4135/*
4136 ------------------------------ free ------------------------------
4137 */
4138
4139static void
4140_int_free (mstate av, mchunkptr p, int have_lock)
4141{
4142 INTERNAL_SIZE_T size; /* its size */
4143 mfastbinptr *fb; /* associated fastbin */
4144 mchunkptr nextchunk; /* next contiguous chunk */
4145 INTERNAL_SIZE_T nextsize; /* its size */
4146 int nextinuse; /* true if nextchunk is used */
4147 INTERNAL_SIZE_T prevsize; /* size of previous contiguous chunk */
4148 mchunkptr bck; /* misc temp for linking */
4149 mchunkptr fwd; /* misc temp for linking */
4150
4151 size = chunksize (p);
4152
4153 /* Little security check which won't hurt performance: the
4154 allocator never wrapps around at the end of the address space.
4155 Therefore we can exclude some size values which might appear
4156 here by accident or by "design" from some intruder. */
4157 if (__builtin_expect ((uintptr_t) p > (uintptr_t) -size, 0)
4158 || __builtin_expect (misaligned_chunk (p), 0))
4159 malloc_printerr ("free(): invalid pointer");
4160 /* We know that each chunk is at least MINSIZE bytes in size or a
4161 multiple of MALLOC_ALIGNMENT. */
4162 if (__glibc_unlikely (size < MINSIZE || !aligned_OK (size)))
4163 malloc_printerr ("free(): invalid size");
4164
4165 check_inuse_chunk(av, p);
4166
4167#if USE_TCACHE
4168 {
4169 size_t tc_idx = csize2tidx (size);
4170
4171 if (tcache
4172 && tc_idx < mp_.tcache_bins
4173 && tcache->counts[tc_idx] < mp_.tcache_count)
4174 {
4175 tcache_put (p, tc_idx);
4176 return;
4177 }
4178 }
4179#endif
4180
4181 /*
4182 If eligible, place chunk on a fastbin so it can be found
4183 and used quickly in malloc.
4184 */
4185
4186 if ((unsigned long)(size) <= (unsigned long)(get_max_fast ())
4187
4188#if TRIM_FASTBINS
4189 /*
4190 If TRIM_FASTBINS set, don't place chunks
4191 bordering top into fastbins
4192 */
4193 && (chunk_at_offset(p, size) != av->top)
4194#endif
4195 ) {
4196
4197 if (__builtin_expect (chunksize_nomask (chunk_at_offset (p, size))
4198 <= 2 * SIZE_SZ, 0)
4199 || __builtin_expect (chunksize (chunk_at_offset (p, size))
4200 >= av->system_mem, 0))
4201 {
4202 bool fail = true;
4203 /* We might not have a lock at this point and concurrent modifications
4204 of system_mem might result in a false positive. Redo the test after
4205 getting the lock. */
4206 if (!have_lock)
4207 {
4208 __libc_lock_lock (av->mutex);
4209 fail = (chunksize_nomask (chunk_at_offset (p, size)) <= 2 * SIZE_SZ
4210 || chunksize (chunk_at_offset (p, size)) >= av->system_mem);
4211 __libc_lock_unlock (av->mutex);
4212 }
4213
4214 if (fail)
4215 malloc_printerr ("free(): invalid next size (fast)");
4216 }
4217
4218 free_perturb (chunk2mem(p), size - 2 * SIZE_SZ);
4219
4220 atomic_store_relaxed (&av->have_fastchunks, true);
4221 unsigned int idx = fastbin_index(size);
4222 fb = &fastbin (av, idx);
4223
4224 /* Atomically link P to its fastbin: P->FD = *FB; *FB = P; */
4225 mchunkptr old = *fb, old2;
4226
4227 if (SINGLE_THREAD_P)
4228 {
4229 /* Check that the top of the bin is not the record we are going to
4230 add (i.e., double free). */
4231 if (__builtin_expect (old == p, 0))
4232 malloc_printerr ("double free or corruption (fasttop)");
4233 p->fd = old;
4234 *fb = p;
4235 }
4236 else
4237 do
4238 {
4239 /* Check that the top of the bin is not the record we are going to
4240 add (i.e., double free). */
4241 if (__builtin_expect (old == p, 0))
4242 malloc_printerr ("double free or corruption (fasttop)");
4243 p->fd = old2 = old;
4244 }
4245 while ((old = catomic_compare_and_exchange_val_rel (fb, p, old2))
4246 != old2);
4247
4248 /* Check that size of fastbin chunk at the top is the same as
4249 size of the chunk that we are adding. We can dereference OLD
4250 only if we have the lock, otherwise it might have already been
4251 allocated again. */
4252 if (have_lock && old != NULL
4253 && __builtin_expect (fastbin_index (chunksize (old)) != idx, 0))
4254 malloc_printerr ("invalid fastbin entry (free)");
4255 }
4256
4257 /*
4258 Consolidate other non-mmapped chunks as they arrive.
4259 */
4260
4261 else if (!chunk_is_mmapped(p)) {
4262
4263 /* If we're single-threaded, don't lock the arena. */
4264 if (SINGLE_THREAD_P)
4265 have_lock = true;
4266
4267 if (!have_lock)
4268 __libc_lock_lock (av->mutex);
4269
4270 nextchunk = chunk_at_offset(p, size);
4271
4272 /* Lightweight tests: check whether the block is already the
4273 top block. */
4274 if (__glibc_unlikely (p == av->top))
4275 malloc_printerr ("double free or corruption (top)");
4276 /* Or whether the next chunk is beyond the boundaries of the arena. */
4277 if (__builtin_expect (contiguous (av)
4278 && (char *) nextchunk
4279 >= ((char *) av->top + chunksize(av->top)), 0))
4280 malloc_printerr ("double free or corruption (out)");
4281 /* Or whether the block is actually not marked used. */
4282 if (__glibc_unlikely (!prev_inuse(nextchunk)))
4283 malloc_printerr ("double free or corruption (!prev)");
4284
4285 nextsize = chunksize(nextchunk);
4286 if (__builtin_expect (chunksize_nomask (nextchunk) <= 2 * SIZE_SZ, 0)
4287 || __builtin_expect (nextsize >= av->system_mem, 0))
4288 malloc_printerr ("free(): invalid next size (normal)");
4289
4290 free_perturb (chunk2mem(p), size - 2 * SIZE_SZ);
4291
4292 /* consolidate backward */
4293 if (!prev_inuse(p)) {
4294 prevsize = prev_size (p);
4295 size += prevsize;
4296 p = chunk_at_offset(p, -((long) prevsize));
4297 unlink(av, p, bck, fwd);
4298 }
4299
4300 if (nextchunk != av->top) {
4301 /* get and clear inuse bit */
4302 nextinuse = inuse_bit_at_offset(nextchunk, nextsize);
4303
4304 /* consolidate forward */
4305 if (!nextinuse) {
4306 unlink(av, nextchunk, bck, fwd);
4307 size += nextsize;
4308 } else
4309 clear_inuse_bit_at_offset(nextchunk, 0);
4310
4311 /*
4312 Place the chunk in unsorted chunk list. Chunks are
4313 not placed into regular bins until after they have
4314 been given one chance to be used in malloc.
4315 */
4316
4317 bck = unsorted_chunks(av);
4318 fwd = bck->fd;
4319 if (__glibc_unlikely (fwd->bk != bck))
4320 malloc_printerr ("free(): corrupted unsorted chunks");
4321 p->fd = fwd;
4322 p->bk = bck;
4323 if (!in_smallbin_range(size))
4324 {
4325 p->fd_nextsize = NULL;
4326 p->bk_nextsize = NULL;
4327 }
4328 bck->fd = p;
4329 fwd->bk = p;
4330
4331 set_head(p, size | PREV_INUSE);
4332 set_foot(p, size);
4333
4334 check_free_chunk(av, p);
4335 }
4336
4337 /*
4338 If the chunk borders the current high end of memory,
4339 consolidate into top
4340 */
4341
4342 else {
4343 size += nextsize;
4344 set_head(p, size | PREV_INUSE);
4345 av->top = p;
4346 check_chunk(av, p);
4347 }
4348
4349 /*
4350 If freeing a large space, consolidate possibly-surrounding
4351 chunks. Then, if the total unused topmost memory exceeds trim
4352 threshold, ask malloc_trim to reduce top.
4353
4354 Unless max_fast is 0, we don't know if there are fastbins
4355 bordering top, so we cannot tell for sure whether threshold
4356 has been reached unless fastbins are consolidated. But we
4357 don't want to consolidate on each free. As a compromise,
4358 consolidation is performed if FASTBIN_CONSOLIDATION_THRESHOLD
4359 is reached.
4360 */
4361
4362 if ((unsigned long)(size) >= FASTBIN_CONSOLIDATION_THRESHOLD) {
4363 if (atomic_load_relaxed (&av->have_fastchunks))
4364 malloc_consolidate(av);
4365
4366 if (av == &main_arena) {
4367#ifndef MORECORE_CANNOT_TRIM
4368 if ((unsigned long)(chunksize(av->top)) >=
4369 (unsigned long)(mp_.trim_threshold))
4370 systrim(mp_.top_pad, av);
4371#endif
4372 } else {
4373 /* Always try heap_trim(), even if the top chunk is not
4374 large, because the corresponding heap might go away. */
4375 heap_info *heap = heap_for_ptr(top(av));
4376
4377 assert(heap->ar_ptr == av);
4378 heap_trim(heap, mp_.top_pad);
4379 }
4380 }
4381
4382 if (!have_lock)
4383 __libc_lock_unlock (av->mutex);
4384 }
4385 /*
4386 If the chunk was allocated via mmap, release via munmap().
4387 */
4388
4389 else {
4390 munmap_chunk (p);
4391 }
4392}
4393
4394/*
4395 ------------------------- malloc_consolidate -------------------------
4396
4397 malloc_consolidate is a specialized version of free() that tears
4398 down chunks held in fastbins. Free itself cannot be used for this
4399 purpose since, among other things, it might place chunks back onto
4400 fastbins. So, instead, we need to use a minor variant of the same
4401 code.
4402*/
4403
4404static void malloc_consolidate(mstate av)
4405{
4406 mfastbinptr* fb; /* current fastbin being consolidated */
4407 mfastbinptr* maxfb; /* last fastbin (for loop control) */
4408 mchunkptr p; /* current chunk being consolidated */
4409 mchunkptr nextp; /* next chunk to consolidate */
4410 mchunkptr unsorted_bin; /* bin header */
4411 mchunkptr first_unsorted; /* chunk to link to */
4412
4413 /* These have same use as in free() */
4414 mchunkptr nextchunk;
4415 INTERNAL_SIZE_T size;
4416 INTERNAL_SIZE_T nextsize;
4417 INTERNAL_SIZE_T prevsize;
4418 int nextinuse;
4419 mchunkptr bck;
4420 mchunkptr fwd;
4421
4422 atomic_store_relaxed (&av->have_fastchunks, false);
4423
4424 unsorted_bin = unsorted_chunks(av);
4425
4426 /*
4427 Remove each chunk from fast bin and consolidate it, placing it
4428 then in unsorted bin. Among other reasons for doing this,
4429 placing in unsorted bin avoids needing to calculate actual bins
4430 until malloc is sure that chunks aren't immediately going to be
4431 reused anyway.
4432 */
4433
4434 maxfb = &fastbin (av, NFASTBINS - 1);
4435 fb = &fastbin (av, 0);
4436 do {
4437 p = atomic_exchange_acq (fb, NULL);
4438 if (p != 0) {
4439 do {
4440 {
4441 unsigned int idx = fastbin_index (chunksize (p));
4442 if ((&fastbin (av, idx)) != fb)
4443 malloc_printerr ("malloc_consolidate(): invalid chunk size");
4444 }
4445
4446 check_inuse_chunk(av, p);
4447 nextp = p->fd;
4448
4449 /* Slightly streamlined version of consolidation code in free() */
4450 size = chunksize (p);
4451 nextchunk = chunk_at_offset(p, size);
4452 nextsize = chunksize(nextchunk);
4453
4454 if (!prev_inuse(p)) {
4455 prevsize = prev_size (p);
4456 size += prevsize;
4457 p = chunk_at_offset(p, -((long) prevsize));
4458 unlink(av, p, bck, fwd);
4459 }
4460
4461 if (nextchunk != av->top) {
4462 nextinuse = inuse_bit_at_offset(nextchunk, nextsize);
4463
4464 if (!nextinuse) {
4465 size += nextsize;
4466 unlink(av, nextchunk, bck, fwd);
4467 } else
4468 clear_inuse_bit_at_offset(nextchunk, 0);
4469
4470 first_unsorted = unsorted_bin->fd;
4471 unsorted_bin->fd = p;
4472 first_unsorted->bk = p;
4473
4474 if (!in_smallbin_range (size)) {
4475 p->fd_nextsize = NULL;
4476 p->bk_nextsize = NULL;
4477 }
4478
4479 set_head(p, size | PREV_INUSE);
4480 p->bk = unsorted_bin;
4481 p->fd = first_unsorted;
4482 set_foot(p, size);
4483 }
4484
4485 else {
4486 size += nextsize;
4487 set_head(p, size | PREV_INUSE);
4488 av->top = p;
4489 }
4490
4491 } while ( (p = nextp) != 0);
4492
4493 }
4494 } while (fb++ != maxfb);
4495}
4496
4497/*
4498 ------------------------------ realloc ------------------------------
4499*/
4500
4501void*
4502_int_realloc(mstate av, mchunkptr oldp, INTERNAL_SIZE_T oldsize,
4503 INTERNAL_SIZE_T nb)
4504{
4505 mchunkptr newp; /* chunk to return */
4506 INTERNAL_SIZE_T newsize; /* its size */
4507 void* newmem; /* corresponding user mem */
4508
4509 mchunkptr next; /* next contiguous chunk after oldp */
4510
4511 mchunkptr remainder; /* extra space at end of newp */
4512 unsigned long remainder_size; /* its size */
4513
4514 mchunkptr bck; /* misc temp for linking */
4515 mchunkptr fwd; /* misc temp for linking */
4516
4517 unsigned long copysize; /* bytes to copy */
4518 unsigned int ncopies; /* INTERNAL_SIZE_T words to copy */
4519 INTERNAL_SIZE_T* s; /* copy source */
4520 INTERNAL_SIZE_T* d; /* copy destination */
4521
4522 /* oldmem size */
4523 if (__builtin_expect (chunksize_nomask (oldp) <= 2 * SIZE_SZ, 0)
4524 || __builtin_expect (oldsize >= av->system_mem, 0))
4525 malloc_printerr ("realloc(): invalid old size");
4526
4527 check_inuse_chunk (av, oldp);
4528
4529 /* All callers already filter out mmap'ed chunks. */
4530 assert (!chunk_is_mmapped (oldp));
4531
4532 next = chunk_at_offset (oldp, oldsize);
4533 INTERNAL_SIZE_T nextsize = chunksize (next);
4534 if (__builtin_expect (chunksize_nomask (next) <= 2 * SIZE_SZ, 0)
4535 || __builtin_expect (nextsize >= av->system_mem, 0))
4536 malloc_printerr ("realloc(): invalid next size");
4537
4538 if ((unsigned long) (oldsize) >= (unsigned long) (nb))
4539 {
4540 /* already big enough; split below */
4541 newp = oldp;
4542 newsize = oldsize;
4543 }
4544
4545 else
4546 {
4547 /* Try to expand forward into top */
4548 if (next == av->top &&
4549 (unsigned long) (newsize = oldsize + nextsize) >=
4550 (unsigned long) (nb + MINSIZE))
4551 {
4552 set_head_size (oldp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0));
4553 av->top = chunk_at_offset (oldp, nb);
4554 set_head (av->top, (newsize - nb) | PREV_INUSE);
4555 check_inuse_chunk (av, oldp);
4556 return chunk2mem (oldp);
4557 }
4558
4559 /* Try to expand forward into next chunk; split off remainder below */
4560 else if (next != av->top &&
4561 !inuse (next) &&
4562 (unsigned long) (newsize = oldsize + nextsize) >=
4563 (unsigned long) (nb))
4564 {
4565 newp = oldp;
4566 unlink (av, next, bck, fwd);
4567 }
4568
4569 /* allocate, copy, free */
4570 else
4571 {
4572 newmem = _int_malloc (av, nb - MALLOC_ALIGN_MASK);
4573 if (newmem == 0)
4574 return 0; /* propagate failure */
4575
4576 newp = mem2chunk (newmem);
4577 newsize = chunksize (newp);
4578
4579 /*
4580 Avoid copy if newp is next chunk after oldp.
4581 */
4582 if (newp == next)
4583 {
4584 newsize += oldsize;
4585 newp = oldp;
4586 }
4587 else
4588 {
4589 /*
4590 Unroll copy of <= 36 bytes (72 if 8byte sizes)
4591 We know that contents have an odd number of
4592 INTERNAL_SIZE_T-sized words; minimally 3.
4593 */
4594
4595 copysize = oldsize - SIZE_SZ;
4596 s = (INTERNAL_SIZE_T *) (chunk2mem (oldp));
4597 d = (INTERNAL_SIZE_T *) (newmem);
4598 ncopies = copysize / sizeof (INTERNAL_SIZE_T);
4599 assert (ncopies >= 3);
4600
4601 if (ncopies > 9)
4602 memcpy (d, s, copysize);
4603
4604 else
4605 {
4606 *(d + 0) = *(s + 0);
4607 *(d + 1) = *(s + 1);
4608 *(d + 2) = *(s + 2);
4609 if (ncopies > 4)
4610 {
4611 *(d + 3) = *(s + 3);
4612 *(d + 4) = *(s + 4);
4613 if (ncopies > 6)
4614 {
4615 *(d + 5) = *(s + 5);
4616 *(d + 6) = *(s + 6);
4617 if (ncopies > 8)
4618 {
4619 *(d + 7) = *(s + 7);
4620 *(d + 8) = *(s + 8);
4621 }
4622 }
4623 }
4624 }
4625
4626 _int_free (av, oldp, 1);
4627 check_inuse_chunk (av, newp);
4628 return chunk2mem (newp);
4629 }
4630 }
4631 }
4632
4633 /* If possible, free extra space in old or extended chunk */
4634
4635 assert ((unsigned long) (newsize) >= (unsigned long) (nb));
4636
4637 remainder_size = newsize - nb;
4638
4639 if (remainder_size < MINSIZE) /* not enough extra to split off */
4640 {
4641 set_head_size (newp, newsize | (av != &main_arena ? NON_MAIN_ARENA : 0));
4642 set_inuse_bit_at_offset (newp, newsize);
4643 }
4644 else /* split remainder */
4645 {
4646 remainder = chunk_at_offset (newp, nb);
4647 set_head_size (newp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0));
4648 set_head (remainder, remainder_size | PREV_INUSE |
4649 (av != &main_arena ? NON_MAIN_ARENA : 0));
4650 /* Mark remainder as inuse so free() won't complain */
4651 set_inuse_bit_at_offset (remainder, remainder_size);
4652 _int_free (av, remainder, 1);
4653 }
4654
4655 check_inuse_chunk (av, newp);
4656 return chunk2mem (newp);
4657}
4658
4659/*
4660 ------------------------------ memalign ------------------------------
4661 */
4662
4663static void *
4664_int_memalign (mstate av, size_t alignment, size_t bytes)
4665{
4666 INTERNAL_SIZE_T nb; /* padded request size */
4667 char *m; /* memory returned by malloc call */
4668 mchunkptr p; /* corresponding chunk */
4669 char *brk; /* alignment point within p */
4670 mchunkptr newp; /* chunk to return */
4671 INTERNAL_SIZE_T newsize; /* its size */
4672 INTERNAL_SIZE_T leadsize; /* leading space before alignment point */
4673 mchunkptr remainder; /* spare room at end to split off */
4674 unsigned long remainder_size; /* its size */
4675 INTERNAL_SIZE_T size;
4676
4677
4678
4679 checked_request2size (bytes, nb);
4680
4681 /*
4682 Strategy: find a spot within that chunk that meets the alignment
4683 request, and then possibly free the leading and trailing space.
4684 */
4685
4686
4687 /* Check for overflow. */
4688 if (nb > SIZE_MAX - alignment - MINSIZE)
4689 {
4690 __set_errno (ENOMEM);
4691 return 0;
4692 }
4693
4694 /* Call malloc with worst case padding to hit alignment. */
4695
4696 m = (char *) (_int_malloc (av, nb + alignment + MINSIZE));
4697
4698 if (m == 0)
4699 return 0; /* propagate failure */
4700
4701 p = mem2chunk (m);
4702
4703 if ((((unsigned long) (m)) % alignment) != 0) /* misaligned */
4704
4705 { /*
4706 Find an aligned spot inside chunk. Since we need to give back
4707 leading space in a chunk of at least MINSIZE, if the first
4708 calculation places us at a spot with less than MINSIZE leader,
4709 we can move to the next aligned spot -- we've allocated enough
4710 total room so that this is always possible.
4711 */
4712 brk = (char *) mem2chunk (((unsigned long) (m + alignment - 1)) &
4713 - ((signed long) alignment));
4714 if ((unsigned long) (brk - (char *) (p)) < MINSIZE)
4715 brk += alignment;
4716
4717 newp = (mchunkptr) brk;
4718 leadsize = brk - (char *) (p);
4719 newsize = chunksize (p) - leadsize;
4720
4721 /* For mmapped chunks, just adjust offset */
4722 if (chunk_is_mmapped (p))
4723 {
4724 set_prev_size (newp, prev_size (p) + leadsize);
4725 set_head (newp, newsize | IS_MMAPPED);
4726 return chunk2mem (newp);
4727 }
4728
4729 /* Otherwise, give back leader, use the rest */
4730 set_head (newp, newsize | PREV_INUSE |
4731 (av != &main_arena ? NON_MAIN_ARENA : 0));
4732 set_inuse_bit_at_offset (newp, newsize);
4733 set_head_size (p, leadsize | (av != &main_arena ? NON_MAIN_ARENA : 0));
4734 _int_free (av, p, 1);
4735 p = newp;
4736
4737 assert (newsize >= nb &&
4738 (((unsigned long) (chunk2mem (p))) % alignment) == 0);
4739 }
4740
4741 /* Also give back spare room at the end */
4742 if (!chunk_is_mmapped (p))
4743 {
4744 size = chunksize (p);
4745 if ((unsigned long) (size) > (unsigned long) (nb + MINSIZE))
4746 {
4747 remainder_size = size - nb;
4748 remainder = chunk_at_offset (p, nb);
4749 set_head (remainder, remainder_size | PREV_INUSE |
4750 (av != &main_arena ? NON_MAIN_ARENA : 0));
4751 set_head_size (p, nb);
4752 _int_free (av, remainder, 1);
4753 }
4754 }
4755
4756 check_inuse_chunk (av, p);
4757 return chunk2mem (p);
4758}
4759
4760
4761/*
4762 ------------------------------ malloc_trim ------------------------------
4763 */
4764
4765static int
4766mtrim (mstate av, size_t pad)
4767{
4768 /* Ensure all blocks are consolidated. */
4769 malloc_consolidate (av);
4770
4771 const size_t ps = GLRO (dl_pagesize);
4772 int psindex = bin_index (ps);
4773 const size_t psm1 = ps - 1;
4774
4775 int result = 0;
4776 for (int i = 1; i < NBINS; ++i)
4777 if (i == 1 || i >= psindex)
4778 {
4779 mbinptr bin = bin_at (av, i);
4780
4781 for (mchunkptr p = last (bin); p != bin; p = p->bk)
4782 {
4783 INTERNAL_SIZE_T size = chunksize (p);
4784
4785 if (size > psm1 + sizeof (struct malloc_chunk))
4786 {
4787 /* See whether the chunk contains at least one unused page. */
4788 char *paligned_mem = (char *) (((uintptr_t) p
4789 + sizeof (struct malloc_chunk)
4790 + psm1) & ~psm1);
4791
4792 assert ((char *) chunk2mem (p) + 4 * SIZE_SZ <= paligned_mem);
4793 assert ((char *) p + size > paligned_mem);
4794
4795 /* This is the size we could potentially free. */
4796 size -= paligned_mem - (char *) p;
4797
4798 if (size > psm1)
4799 {
4800#if MALLOC_DEBUG
4801 /* When debugging we simulate destroying the memory
4802 content. */
4803 memset (paligned_mem, 0x89, size & ~psm1);
4804#endif
4805 __madvise (paligned_mem, size & ~psm1, MADV_DONTNEED);
4806
4807 result = 1;
4808 }
4809 }
4810 }
4811 }
4812
4813#ifndef MORECORE_CANNOT_TRIM
4814 return result | (av == &main_arena ? systrim (pad, av) : 0);
4815
4816#else
4817 return result;
4818#endif
4819}
4820
4821
4822int
4823__malloc_trim (size_t s)
4824{
4825 int result = 0;
4826
4827 if (__malloc_initialized < 0)
4828 ptmalloc_init ();
4829
4830 mstate ar_ptr = &main_arena;
4831 do
4832 {
4833 __libc_lock_lock (ar_ptr->mutex);
4834 result |= mtrim (ar_ptr, s);
4835 __libc_lock_unlock (ar_ptr->mutex);
4836
4837 ar_ptr = ar_ptr->next;
4838 }
4839 while (ar_ptr != &main_arena);
4840
4841 return result;
4842}
4843
4844
4845/*
4846 ------------------------- malloc_usable_size -------------------------
4847 */
4848
4849static size_t
4850musable (void *mem)
4851{
4852 mchunkptr p;
4853 if (mem != 0)
4854 {
4855 p = mem2chunk (mem);
4856
4857 if (__builtin_expect (using_malloc_checking == 1, 0))
4858 return malloc_check_get_size (p);
4859
4860 if (chunk_is_mmapped (p))
4861 {
4862 if (DUMPED_MAIN_ARENA_CHUNK (p))
4863 return chunksize (p) - SIZE_SZ;
4864 else
4865 return chunksize (p) - 2 * SIZE_SZ;
4866 }
4867 else if (inuse (p))
4868 return chunksize (p) - SIZE_SZ;
4869 }
4870 return 0;
4871}
4872
4873
4874size_t
4875__malloc_usable_size (void *m)
4876{
4877 size_t result;
4878
4879 result = musable (m);
4880 return result;
4881}
4882
4883/*
4884 ------------------------------ mallinfo ------------------------------
4885 Accumulate malloc statistics for arena AV into M.
4886 */
4887
4888static void
4889int_mallinfo (mstate av, struct mallinfo *m)
4890{
4891 size_t i;
4892 mbinptr b;
4893 mchunkptr p;
4894 INTERNAL_SIZE_T avail;
4895 INTERNAL_SIZE_T fastavail;
4896 int nblocks;
4897 int nfastblocks;
4898
4899 check_malloc_state (av);
4900
4901 /* Account for top */
4902 avail = chunksize (av->top);
4903 nblocks = 1; /* top always exists */
4904
4905 /* traverse fastbins */
4906 nfastblocks = 0;
4907 fastavail = 0;
4908
4909 for (i = 0; i < NFASTBINS; ++i)
4910 {
4911 for (p = fastbin (av, i); p != 0; p = p->fd)
4912 {
4913 ++nfastblocks;
4914 fastavail += chunksize (p);
4915 }
4916 }
4917
4918 avail += fastavail;
4919
4920 /* traverse regular bins */
4921 for (i = 1; i < NBINS; ++i)
4922 {
4923 b = bin_at (av, i);
4924 for (p = last (b); p != b; p = p->bk)
4925 {
4926 ++nblocks;
4927 avail += chunksize (p);
4928 }
4929 }
4930
4931 m->smblks += nfastblocks;
4932 m->ordblks += nblocks;
4933 m->fordblks += avail;
4934 m->uordblks += av->system_mem - avail;
4935 m->arena += av->system_mem;
4936 m->fsmblks += fastavail;
4937 if (av == &main_arena)
4938 {
4939 m->hblks = mp_.n_mmaps;
4940 m->hblkhd = mp_.mmapped_mem;
4941 m->usmblks = 0;
4942 m->keepcost = chunksize (av->top);
4943 }
4944}
4945
4946
4947struct mallinfo
4948__libc_mallinfo (void)
4949{
4950 struct mallinfo m;
4951 mstate ar_ptr;
4952
4953 if (__malloc_initialized < 0)
4954 ptmalloc_init ();
4955
4956 memset (&m, 0, sizeof (m));
4957 ar_ptr = &main_arena;
4958 do
4959 {
4960 __libc_lock_lock (ar_ptr->mutex);
4961 int_mallinfo (ar_ptr, &m);
4962 __libc_lock_unlock (ar_ptr->mutex);
4963
4964 ar_ptr = ar_ptr->next;
4965 }
4966 while (ar_ptr != &main_arena);
4967
4968 return m;
4969}
4970
4971/*
4972 ------------------------------ malloc_stats ------------------------------
4973 */
4974
4975void
4976__malloc_stats (void)
4977{
4978 int i;
4979 mstate ar_ptr;
4980 unsigned int in_use_b = mp_.mmapped_mem, system_b = in_use_b;
4981
4982 if (__malloc_initialized < 0)
4983 ptmalloc_init ();
4984 _IO_flockfile (stderr);
4985 int old_flags2 = stderr->_flags2;
4986 stderr->_flags2 |= _IO_FLAGS2_NOTCANCEL;
4987 for (i = 0, ar_ptr = &main_arena;; i++)
4988 {
4989 struct mallinfo mi;
4990
4991 memset (&mi, 0, sizeof (mi));
4992 __libc_lock_lock (ar_ptr->mutex);
4993 int_mallinfo (ar_ptr, &mi);
4994 fprintf (stderr, "Arena %d:\n", i);
4995 fprintf (stderr, "system bytes = %10u\n", (unsigned int) mi.arena);
4996 fprintf (stderr, "in use bytes = %10u\n", (unsigned int) mi.uordblks);
4997#if MALLOC_DEBUG > 1
4998 if (i > 0)
4999 dump_heap (heap_for_ptr (top (ar_ptr)));
5000#endif
5001 system_b += mi.arena;
5002 in_use_b += mi.uordblks;
5003 __libc_lock_unlock (ar_ptr->mutex);
5004 ar_ptr = ar_ptr->next;
5005 if (ar_ptr == &main_arena)
5006 break;
5007 }
5008 fprintf (stderr, "Total (incl. mmap):\n");
5009 fprintf (stderr, "system bytes = %10u\n", system_b);
5010 fprintf (stderr, "in use bytes = %10u\n", in_use_b);
5011 fprintf (stderr, "max mmap regions = %10u\n", (unsigned int) mp_.max_n_mmaps);
5012 fprintf (stderr, "max mmap bytes = %10lu\n",
5013 (unsigned long) mp_.max_mmapped_mem);
5014 stderr->_flags2 = old_flags2;
5015 _IO_funlockfile (stderr);
5016}
5017
5018
5019/*
5020 ------------------------------ mallopt ------------------------------
5021 */
5022static inline int
5023__always_inline
5024do_set_trim_threshold (size_t value)
5025{
5026 LIBC_PROBE (memory_mallopt_trim_threshold, 3, value, mp_.trim_threshold,
5027 mp_.no_dyn_threshold);
5028 mp_.trim_threshold = value;
5029 mp_.no_dyn_threshold = 1;
5030 return 1;
5031}
5032
5033static inline int
5034__always_inline
5035do_set_top_pad (size_t value)
5036{
5037 LIBC_PROBE (memory_mallopt_top_pad, 3, value, mp_.top_pad,
5038 mp_.no_dyn_threshold);
5039 mp_.top_pad = value;
5040 mp_.no_dyn_threshold = 1;
5041 return 1;
5042}
5043
5044static inline int
5045__always_inline
5046do_set_mmap_threshold (size_t value)
5047{
5048 /* Forbid setting the threshold too high. */
5049 if (value <= HEAP_MAX_SIZE / 2)
5050 {
5051 LIBC_PROBE (memory_mallopt_mmap_threshold, 3, value, mp_.mmap_threshold,
5052 mp_.no_dyn_threshold);
5053 mp_.mmap_threshold = value;
5054 mp_.no_dyn_threshold = 1;
5055 return 1;
5056 }
5057 return 0;
5058}
5059
5060static inline int
5061__always_inline
5062do_set_mmaps_max (int32_t value)
5063{
5064 LIBC_PROBE (memory_mallopt_mmap_max, 3, value, mp_.n_mmaps_max,
5065 mp_.no_dyn_threshold);
5066 mp_.n_mmaps_max = value;
5067 mp_.no_dyn_threshold = 1;
5068 return 1;
5069}
5070
5071static inline int
5072__always_inline
5073do_set_mallopt_check (int32_t value)
5074{
5075 return 1;
5076}
5077
5078static inline int
5079__always_inline
5080do_set_perturb_byte (int32_t value)
5081{
5082 LIBC_PROBE (memory_mallopt_perturb, 2, value, perturb_byte);
5083 perturb_byte = value;
5084 return 1;
5085}
5086
5087static inline int
5088__always_inline
5089do_set_arena_test (size_t value)
5090{
5091 LIBC_PROBE (memory_mallopt_arena_test, 2, value, mp_.arena_test);
5092 mp_.arena_test = value;
5093 return 1;
5094}
5095
5096static inline int
5097__always_inline
5098do_set_arena_max (size_t value)
5099{
5100 LIBC_PROBE (memory_mallopt_arena_max, 2, value, mp_.arena_max);
5101 mp_.arena_max = value;
5102 return 1;
5103}
5104
5105#if USE_TCACHE
5106static inline int
5107__always_inline
5108do_set_tcache_max (size_t value)
5109{
5110 if (value >= 0 && value <= MAX_TCACHE_SIZE)
5111 {
5112 LIBC_PROBE (memory_tunable_tcache_max_bytes, 2, value, mp_.tcache_max_bytes);
5113 mp_.tcache_max_bytes = value;
5114 mp_.tcache_bins = csize2tidx (request2size(value)) + 1;
5115 }
5116 return 1;
5117}
5118
5119static inline int
5120__always_inline
5121do_set_tcache_count (size_t value)
5122{
5123 LIBC_PROBE (memory_tunable_tcache_count, 2, value, mp_.tcache_count);
5124 mp_.tcache_count = value;
5125 return 1;
5126}
5127
5128static inline int
5129__always_inline
5130do_set_tcache_unsorted_limit (size_t value)
5131{
5132 LIBC_PROBE (memory_tunable_tcache_unsorted_limit, 2, value, mp_.tcache_unsorted_limit);
5133 mp_.tcache_unsorted_limit = value;
5134 return 1;
5135}
5136#endif
5137
5138int
5139__libc_mallopt (int param_number, int value)
5140{
5141 mstate av = &main_arena;
5142 int res = 1;
5143
5144 if (__malloc_initialized < 0)
5145 ptmalloc_init ();
5146 __libc_lock_lock (av->mutex);
5147
5148 LIBC_PROBE (memory_mallopt, 2, param_number, value);
5149
5150 /* We must consolidate main arena before changing max_fast
5151 (see definition of set_max_fast). */
5152 malloc_consolidate (av);
5153
5154 switch (param_number)
5155 {
5156 case M_MXFAST:
5157 if (value >= 0 && value <= MAX_FAST_SIZE)
5158 {
5159 LIBC_PROBE (memory_mallopt_mxfast, 2, value, get_max_fast ());
5160 set_max_fast (value);
5161 }
5162 else
5163 res = 0;
5164 break;
5165
5166 case M_TRIM_THRESHOLD:
5167 do_set_trim_threshold (value);
5168 break;
5169
5170 case M_TOP_PAD:
5171 do_set_top_pad (value);
5172 break;
5173
5174 case M_MMAP_THRESHOLD:
5175 res = do_set_mmap_threshold (value);
5176 break;
5177
5178 case M_MMAP_MAX:
5179 do_set_mmaps_max (value);
5180 break;
5181
5182 case M_CHECK_ACTION:
5183 do_set_mallopt_check (value);
5184 break;
5185
5186 case M_PERTURB:
5187 do_set_perturb_byte (value);
5188 break;
5189
5190 case M_ARENA_TEST:
5191 if (value > 0)
5192 do_set_arena_test (value);
5193 break;
5194
5195 case M_ARENA_MAX:
5196 if (value > 0)
5197 do_set_arena_max (value);
5198 break;
5199 }
5200 __libc_lock_unlock (av->mutex);
5201 return res;
5202}
5203libc_hidden_def (__libc_mallopt)
5204
5205
5206/*
5207 -------------------- Alternative MORECORE functions --------------------
5208 */
5209
5210
5211/*
5212 General Requirements for MORECORE.
5213
5214 The MORECORE function must have the following properties:
5215
5216 If MORECORE_CONTIGUOUS is false:
5217
5218 * MORECORE must allocate in multiples of pagesize. It will
5219 only be called with arguments that are multiples of pagesize.
5220
5221 * MORECORE(0) must return an address that is at least
5222 MALLOC_ALIGNMENT aligned. (Page-aligning always suffices.)
5223
5224 else (i.e. If MORECORE_CONTIGUOUS is true):
5225
5226 * Consecutive calls to MORECORE with positive arguments
5227 return increasing addresses, indicating that space has been
5228 contiguously extended.
5229
5230 * MORECORE need not allocate in multiples of pagesize.
5231 Calls to MORECORE need not have args of multiples of pagesize.
5232
5233 * MORECORE need not page-align.
5234
5235 In either case:
5236
5237 * MORECORE may allocate more memory than requested. (Or even less,
5238 but this will generally result in a malloc failure.)
5239
5240 * MORECORE must not allocate memory when given argument zero, but
5241 instead return one past the end address of memory from previous
5242 nonzero call. This malloc does NOT call MORECORE(0)
5243 until at least one call with positive arguments is made, so
5244 the initial value returned is not important.
5245
5246 * Even though consecutive calls to MORECORE need not return contiguous
5247 addresses, it must be OK for malloc'ed chunks to span multiple
5248 regions in those cases where they do happen to be contiguous.
5249
5250 * MORECORE need not handle negative arguments -- it may instead
5251 just return MORECORE_FAILURE when given negative arguments.
5252 Negative arguments are always multiples of pagesize. MORECORE
5253 must not misinterpret negative args as large positive unsigned
5254 args. You can suppress all such calls from even occurring by defining
5255 MORECORE_CANNOT_TRIM,
5256
5257 There is some variation across systems about the type of the
5258 argument to sbrk/MORECORE. If size_t is unsigned, then it cannot
5259 actually be size_t, because sbrk supports negative args, so it is
5260 normally the signed type of the same width as size_t (sometimes
5261 declared as "intptr_t", and sometimes "ptrdiff_t"). It doesn't much
5262 matter though. Internally, we use "long" as arguments, which should
5263 work across all reasonable possibilities.
5264
5265 Additionally, if MORECORE ever returns failure for a positive
5266 request, then mmap is used as a noncontiguous system allocator. This
5267 is a useful backup strategy for systems with holes in address spaces
5268 -- in this case sbrk cannot contiguously expand the heap, but mmap
5269 may be able to map noncontiguous space.
5270
5271 If you'd like mmap to ALWAYS be used, you can define MORECORE to be
5272 a function that always returns MORECORE_FAILURE.
5273
5274 If you are using this malloc with something other than sbrk (or its
5275 emulation) to supply memory regions, you probably want to set
5276 MORECORE_CONTIGUOUS as false. As an example, here is a custom
5277 allocator kindly contributed for pre-OSX macOS. It uses virtually
5278 but not necessarily physically contiguous non-paged memory (locked
5279 in, present and won't get swapped out). You can use it by
5280 uncommenting this section, adding some #includes, and setting up the
5281 appropriate defines above:
5282
5283 *#define MORECORE osMoreCore
5284 *#define MORECORE_CONTIGUOUS 0
5285
5286 There is also a shutdown routine that should somehow be called for
5287 cleanup upon program exit.
5288
5289 *#define MAX_POOL_ENTRIES 100
5290 *#define MINIMUM_MORECORE_SIZE (64 * 1024)
5291 static int next_os_pool;
5292 void *our_os_pools[MAX_POOL_ENTRIES];
5293
5294 void *osMoreCore(int size)
5295 {
5296 void *ptr = 0;
5297 static void *sbrk_top = 0;
5298
5299 if (size > 0)
5300 {
5301 if (size < MINIMUM_MORECORE_SIZE)
5302 size = MINIMUM_MORECORE_SIZE;
5303 if (CurrentExecutionLevel() == kTaskLevel)
5304 ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0);
5305 if (ptr == 0)
5306 {
5307 return (void *) MORECORE_FAILURE;
5308 }
5309 // save ptrs so they can be freed during cleanup
5310 our_os_pools[next_os_pool] = ptr;
5311 next_os_pool++;
5312 ptr = (void *) ((((unsigned long) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK);
5313 sbrk_top = (char *) ptr + size;
5314 return ptr;
5315 }
5316 else if (size < 0)
5317 {
5318 // we don't currently support shrink behavior
5319 return (void *) MORECORE_FAILURE;
5320 }
5321 else
5322 {
5323 return sbrk_top;
5324 }
5325 }
5326
5327 // cleanup any allocated memory pools
5328 // called as last thing before shutting down driver
5329
5330 void osCleanupMem(void)
5331 {
5332 void **ptr;
5333
5334 for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++)
5335 if (*ptr)
5336 {
5337 PoolDeallocate(*ptr);
5338 * ptr = 0;
5339 }
5340 }
5341
5342 */
5343
5344
5345/* Helper code. */
5346
5347extern char **__libc_argv attribute_hidden;
5348
5349static void
5350malloc_printerr (const char *str)
5351{
5352 __libc_message (do_abort, "%s\n", str);
5353 __builtin_unreachable ();
5354}
5355
5356/* We need a wrapper function for one of the additions of POSIX. */
5357int
5358__posix_memalign (void **memptr, size_t alignment, size_t size)
5359{
5360 void *mem;
5361
5362 /* Test whether the SIZE argument is valid. It must be a power of
5363 two multiple of sizeof (void *). */
5364 if (alignment % sizeof (void *) != 0
5365 || !powerof2 (alignment / sizeof (void *))
5366 || alignment == 0)
5367 return EINVAL;
5368
5369
5370 void *address = RETURN_ADDRESS (0);
5371 mem = _mid_memalign (alignment, size, address);
5372
5373 if (mem != NULL)
5374 {
5375 *memptr = mem;
5376 return 0;
5377 }
5378
5379 return ENOMEM;
5380}
5381weak_alias (__posix_memalign, posix_memalign)
5382
5383
5384int
5385__malloc_info (int options, FILE *fp)
5386{
5387 /* For now, at least. */
5388 if (options != 0)
5389 return EINVAL;
5390
5391 int n = 0;
5392 size_t total_nblocks = 0;
5393 size_t total_nfastblocks = 0;
5394 size_t total_avail = 0;
5395 size_t total_fastavail = 0;
5396 size_t total_system = 0;
5397 size_t total_max_system = 0;
5398 size_t total_aspace = 0;
5399 size_t total_aspace_mprotect = 0;
5400
5401
5402
5403 if (__malloc_initialized < 0)
5404 ptmalloc_init ();
5405
5406 fputs ("<malloc version=\"1\">\n", fp);
5407
5408 /* Iterate over all arenas currently in use. */
5409 mstate ar_ptr = &main_arena;
5410 do
5411 {
5412 fprintf (fp, "<heap nr=\"%d\">\n<sizes>\n", n++);
5413
5414 size_t nblocks = 0;
5415 size_t nfastblocks = 0;
5416 size_t avail = 0;
5417 size_t fastavail = 0;
5418 struct
5419 {
5420 size_t from;
5421 size_t to;
5422 size_t total;
5423 size_t count;
5424 } sizes[NFASTBINS + NBINS - 1];
5425#define nsizes (sizeof (sizes) / sizeof (sizes[0]))
5426
5427 __libc_lock_lock (ar_ptr->mutex);
5428
5429 for (size_t i = 0; i < NFASTBINS; ++i)
5430 {
5431 mchunkptr p = fastbin (ar_ptr, i);
5432 if (p != NULL)
5433 {
5434 size_t nthissize = 0;
5435 size_t thissize = chunksize (p);
5436
5437 while (p != NULL)
5438 {
5439 ++nthissize;
5440 p = p->fd;
5441 }
5442
5443 fastavail += nthissize * thissize;
5444 nfastblocks += nthissize;
5445 sizes[i].from = thissize - (MALLOC_ALIGNMENT - 1);
5446 sizes[i].to = thissize;
5447 sizes[i].count = nthissize;
5448 }
5449 else
5450 sizes[i].from = sizes[i].to = sizes[i].count = 0;
5451
5452 sizes[i].total = sizes[i].count * sizes[i].to;
5453 }
5454
5455
5456 mbinptr bin;
5457 struct malloc_chunk *r;
5458
5459 for (size_t i = 1; i < NBINS; ++i)
5460 {
5461 bin = bin_at (ar_ptr, i);
5462 r = bin->fd;
5463 sizes[NFASTBINS - 1 + i].from = ~((size_t) 0);
5464 sizes[NFASTBINS - 1 + i].to = sizes[NFASTBINS - 1 + i].total
5465 = sizes[NFASTBINS - 1 + i].count = 0;
5466
5467 if (r != NULL)
5468 while (r != bin)
5469 {
5470 size_t r_size = chunksize_nomask (r);
5471 ++sizes[NFASTBINS - 1 + i].count;
5472 sizes[NFASTBINS - 1 + i].total += r_size;
5473 sizes[NFASTBINS - 1 + i].from
5474 = MIN (sizes[NFASTBINS - 1 + i].from, r_size);
5475 sizes[NFASTBINS - 1 + i].to = MAX (sizes[NFASTBINS - 1 + i].to,
5476 r_size);
5477
5478 r = r->fd;
5479 }
5480
5481 if (sizes[NFASTBINS - 1 + i].count == 0)
5482 sizes[NFASTBINS - 1 + i].from = 0;
5483 nblocks += sizes[NFASTBINS - 1 + i].count;
5484 avail += sizes[NFASTBINS - 1 + i].total;
5485 }
5486
5487 size_t heap_size = 0;
5488 size_t heap_mprotect_size = 0;
5489 size_t heap_count = 0;
5490 if (ar_ptr != &main_arena)
5491 {
5492 /* Iterate over the arena heaps from back to front. */
5493 heap_info *heap = heap_for_ptr (top (ar_ptr));
5494 do
5495 {
5496 heap_size += heap->size;
5497 heap_mprotect_size += heap->mprotect_size;
5498 heap = heap->prev;
5499 ++heap_count;
5500 }
5501 while (heap != NULL);
5502 }
5503
5504 __libc_lock_unlock (ar_ptr->mutex);
5505
5506 total_nfastblocks += nfastblocks;
5507 total_fastavail += fastavail;
5508
5509 total_nblocks += nblocks;
5510 total_avail += avail;
5511
5512 for (size_t i = 0; i < nsizes; ++i)
5513 if (sizes[i].count != 0 && i != NFASTBINS)
5514 fprintf (fp, " \
5515 <size from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n",
5516 sizes[i].from, sizes[i].to, sizes[i].total, sizes[i].count);
5517
5518 if (sizes[NFASTBINS].count != 0)
5519 fprintf (fp, "\
5520 <unsorted from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n",
5521 sizes[NFASTBINS].from, sizes[NFASTBINS].to,
5522 sizes[NFASTBINS].total, sizes[NFASTBINS].count);
5523
5524 total_system += ar_ptr->system_mem;
5525 total_max_system += ar_ptr->max_system_mem;
5526
5527 fprintf (fp,
5528 "</sizes>\n<total type=\"fast\" count=\"%zu\" size=\"%zu\"/>\n"
5529 "<total type=\"rest\" count=\"%zu\" size=\"%zu\"/>\n"
5530 "<system type=\"current\" size=\"%zu\"/>\n"
5531 "<system type=\"max\" size=\"%zu\"/>\n",
5532 nfastblocks, fastavail, nblocks, avail,
5533 ar_ptr->system_mem, ar_ptr->max_system_mem);
5534
5535 if (ar_ptr != &main_arena)
5536 {
5537 fprintf (fp,
5538 "<aspace type=\"total\" size=\"%zu\"/>\n"
5539 "<aspace type=\"mprotect\" size=\"%zu\"/>\n"
5540 "<aspace type=\"subheaps\" size=\"%zu\"/>\n",
5541 heap_size, heap_mprotect_size, heap_count);
5542 total_aspace += heap_size;
5543 total_aspace_mprotect += heap_mprotect_size;
5544 }
5545 else
5546 {
5547 fprintf (fp,
5548 "<aspace type=\"total\" size=\"%zu\"/>\n"
5549 "<aspace type=\"mprotect\" size=\"%zu\"/>\n",
5550 ar_ptr->system_mem, ar_ptr->system_mem);
5551 total_aspace += ar_ptr->system_mem;
5552 total_aspace_mprotect +=