malloc.c source code [glibc/malloc/malloc.c]

1	/ Malloc implementation for multiple threads without lock contention.*
2	Copyright (C) 1996-2021 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 <https://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	/ For memory tagging. /
246	#include <libc-mtag.h>
247
248	#include <malloc/malloc-internal.h>
249
250	/ For SINGLE_THREAD_P. /
251	#include <sysdep-cancel.h>
252
253	#include <libc-internal.h>
254
255	/ For tcache double-free check. /
256	#include <random-bits.h>
257	#include <sys/random.h>
258
259	/*
260	Debugging:
261
262	Because freed chunks may be overwritten with bookkeeping fields, this
263	malloc will often die when freed memory is overwritten by user
264	programs. This can be very effective (albeit in an annoying way)
265	in helping track down dangling pointers.
266
267	If you compile with -DMALLOC_DEBUG, a number of assertion checks are
268	enabled that will catch more memory errors. You probably won't be
269	able to make much sense of the actual assertion errors, but they
270	should help you locate incorrectly overwritten memory. The checking
271	is fairly extensive, and will slow down execution
272	noticeably. Calling malloc_stats or mallinfo with MALLOC_DEBUG set
273	will attempt to check every non-mmapped allocated and free chunk in
274	the course of computing the summmaries. (By nature, mmapped regions
275	cannot be checked very much automatically.)
276
277	Setting MALLOC_DEBUG may also be helpful if you are trying to modify
278	this code. The assertions in the check routines spell out in more
279	detail the assumptions and invariants underlying the algorithms.
280
281	Setting MALLOC_DEBUG does NOT provide an automated mechanism for
282	checking that all accesses to malloced memory stay within their
283	bounds. However, there are several add-ons and adaptations of this
284	or other mallocs available that do this.
285	*/
286
287	#ifndef MALLOC_DEBUG
288	#define MALLOC_DEBUG 0
289	#endif
290
291	#if IS_IN (libc)
292	#ifndef NDEBUG
293	# define __assert_fail(assertion, file, line, function) \
294	__malloc_assert(assertion, file, line, function)
295
296	extern const char *__progname;
297
298	static void
299	__malloc_assert (const char assertion, const* char file, unsigned* int line,
300	const char *function)
301	{
302	(void) __fxprintf (NULL, "%s%s%s:%u: %s%sAssertion `%s' failed.\n",
303	__progname, __progname[`0`] ? ": " : "",
304	file, line,
305	function ? function : "", function ? ": " : "",
306	assertion);
307	fflush (stderr);
308	abort ();
309	}
310	#endif
311	#endif
312
313	#if USE_TCACHE
314	/ We want 64 entries. This is an arbitrary limit, which tunables can reduce. /
315	# define TCACHE_MAX_BINS 64
316	# define MAX_TCACHE_SIZE tidx2usize (TCACHE_MAX_BINS-1)
317
318	/ Only used to pre-fill the tunables. /
319	# define tidx2usize(idx) (((size_t) idx) * MALLOC_ALIGNMENT + MINSIZE - SIZE_SZ)
320
321	/ When "x" is from chunksize(). /
322	# define csize2tidx(x) (((x) - MINSIZE + MALLOC_ALIGNMENT - 1) / MALLOC_ALIGNMENT)
323	/ When "x" is a user-provided size. /
324	# define usize2tidx(x) csize2tidx (request2size (x))
325
326	/ With rounding and alignment, the bins are...*
327	idx 0 bytes 0..24 (64-bit) or 0..12 (32-bit)
328	idx 1 bytes 25..40 or 13..20
329	idx 2 bytes 41..56 or 21..28
330	etc. /*
331
332	/ This is another arbitrary limit, which tunables can change. Each*
333	tcache bin will hold at most this number of chunks. /*
334	# define TCACHE_FILL_COUNT 7
335
336	/ Maximum chunks in tcache bins for tunables. This value must fit the range*
337	of tcache->counts[] entries, else they may overflow. /*
338	# define MAX_TCACHE_COUNT UINT16_MAX
339	#endif
340
341	/ Safe-Linking:*
342	Use randomness from ASLR (mmap_base) to protect single-linked lists
343	of Fast-Bins and TCache. That is, mask the "next" pointers of the
344	lists' chunks, and also perform allocation alignment checks on them.
345	This mechanism reduces the risk of pointer hijacking, as was done with
346	Safe-Unlinking in the double-linked lists of Small-Bins.
347	It assumes a minimum page size of 4096 bytes (12 bits). Systems with
348	larger pages provide less entropy, although the pointer mangling
349	still works. /*
350	#define PROTECT_PTR(pos, ptr) \
351	((__typeof (ptr)) ((((size_t) pos) >> 12) ^ ((size_t) ptr)))
352	#define REVEAL_PTR(ptr) PROTECT_PTR (&ptr, ptr)
353
354	/*
355	The REALLOC_ZERO_BYTES_FREES macro controls the behavior of realloc (p, 0)
356	when p is nonnull. If the macro is nonzero, the realloc call returns NULL;
357	otherwise, the call returns what malloc (0) would. In either case,
358	p is freed. Glibc uses a nonzero REALLOC_ZERO_BYTES_FREES, which
359	implements common historical practice.
360
361	ISO C17 says the realloc call has implementation-defined behavior,
362	and it might not even free p.
363	*/
364
365	#ifndef REALLOC_ZERO_BYTES_FREES
366	#define REALLOC_ZERO_BYTES_FREES 1
367	#endif
368
369	/*
370	TRIM_FASTBINS controls whether free() of a very small chunk can
371	immediately lead to trimming. Setting to true (1) can reduce memory
372	footprint, but will almost always slow down programs that use a lot
373	of small chunks.
374
375	Define this only if you are willing to give up some speed to more
376	aggressively reduce system-level memory footprint when releasing
377	memory in programs that use many small chunks. You can get
378	essentially the same effect by setting MXFAST to 0, but this can
379	lead to even greater slowdowns in programs using many small chunks.
380	TRIM_FASTBINS is an in-between compile-time option, that disables
381	only those chunks bordering topmost memory from being placed in
382	fastbins.
383	*/
384
385	#ifndef TRIM_FASTBINS
386	#define TRIM_FASTBINS 0
387	#endif
388
389	/ Definition for getting more memory from the OS. /
390	#include "morecore.c"
391
392	#define MORECORE (*__glibc_morecore)
393	#define MORECORE_FAILURE 0
394
395	/ Memory tagging. /
396
397	/ Some systems support the concept of tagging (sometimes known as*
398	coloring) memory locations on a fine grained basis. Each memory
399	location is given a color (normally allocated randomly) and
400	pointers are also colored. When the pointer is dereferenced, the
401	pointer's color is checked against the memory's color and if they
402	differ the access is faulted (sometimes lazily).
403
404	We use this in glibc by maintaining a single color for the malloc
405	data structures that are interleaved with the user data and then
406	assigning separate colors for each block allocation handed out. In
407	this way simple buffer overruns will be rapidly detected. When
408	memory is freed, the memory is recolored back to the glibc default
409	so that simple use-after-free errors can also be detected.
410
411	If memory is reallocated the buffer is recolored even if the
412	address remains the same. This has a performance impact, but
413	guarantees that the old pointer cannot mistakenly be reused (code
414	that compares old against new will see a mismatch and will then
415	need to behave as though realloc moved the data to a new location).
416
417	Internal API for memory tagging support.
418
419	The aim is to keep the code for memory tagging support as close to
420	the normal APIs in glibc as possible, so that if tagging is not
421	enabled in the library, or is disabled at runtime then standard
422	operations can continue to be used. Support macros are used to do
423	this:
424
425	void tag_new_zero_region (void ptr, size_t size)
426
427	Allocates a new tag, colors the memory with that tag, zeros the
428	memory and returns a pointer that is correctly colored for that
429	location. The non-tagging version will simply call memset with 0.
430
431	void tag_region (void ptr, size_t size)
432
433	Color the region of memory pointed to by PTR and size SIZE with
434	the color of PTR. Returns the original pointer.
435
436	void tag_new_usable (void ptr)
437
438	Allocate a new random color and use it to color the user region of
439	a chunk; this may include data from the subsequent chunk's header
440	if tagging is sufficiently fine grained. Returns PTR suitably
441	recolored for accessing the memory there.
442
443	void tag_at (void ptr)
444
445	Read the current color of the memory at the address pointed to by
446	PTR (ignoring it's current color) and return PTR recolored to that
447	color. PTR must be valid address in all other respects. When
448	tagging is not enabled, it simply returns the original pointer.
449	*/
450
451	#ifdef USE_MTAG
452	static bool mtag_enabled = false;
453	static int mtag_mmap_flags = `0`;
454	#else
455	# define mtag_enabled false
456	# define mtag_mmap_flags 0
457	#endif
458
459	static __always_inline void *
460	tag_region (void *ptr, size_t size)
461	{
462	if (__glibc_unlikely (mtag_enabled))
463	return __libc_mtag_tag_region (ptr, size);
464	return ptr;
465	}
466
467	static __always_inline void *
468	tag_new_zero_region (void *ptr, size_t size)
469	{
470	if (__glibc_unlikely (mtag_enabled))
471	return __libc_mtag_tag_zero_region (__libc_mtag_new_tag (ptr), size);
472	return memset (ptr, `0`, size);
473	}
474
475	/ Defined later. /
476	static void *
477	tag_new_usable (void *ptr);
478
479	static __always_inline void *
480	tag_at (void *ptr)
481	{
482	if (__glibc_unlikely (mtag_enabled))
483	return __libc_mtag_address_get_tag (ptr);
484	return ptr;
485	}
486
487	#include <string.h>
488
489	/*
490	MORECORE-related declarations. By default, rely on sbrk
491	*/
492
493
494	/*
495	MORECORE is the name of the routine to call to obtain more memory
496	from the system. See below for general guidance on writing
497	alternative MORECORE functions, as well as a version for WIN32 and a
498	sample version for pre-OSX macos.
499	*/
500
501	#ifndef MORECORE
502	#define MORECORE sbrk
503	#endif
504
505	/*
506	MORECORE_FAILURE is the value returned upon failure of MORECORE
507	as well as mmap. Since it cannot be an otherwise valid memory address,
508	and must reflect values of standard sys calls, you probably ought not
509	try to redefine it.
510	*/
511
512	#ifndef MORECORE_FAILURE
513	#define MORECORE_FAILURE (-1)
514	#endif
515
516	/*
517	If MORECORE_CONTIGUOUS is true, take advantage of fact that
518	consecutive calls to MORECORE with positive arguments always return
519	contiguous increasing addresses. This is true of unix sbrk. Even
520	if not defined, when regions happen to be contiguous, malloc will
521	permit allocations spanning regions obtained from different
522	calls. But defining this when applicable enables some stronger
523	consistency checks and space efficiencies.
524	*/
525
526	#ifndef MORECORE_CONTIGUOUS
527	#define MORECORE_CONTIGUOUS 1
528	#endif
529
530	/*
531	Define MORECORE_CANNOT_TRIM if your version of MORECORE
532	cannot release space back to the system when given negative
533	arguments. This is generally necessary only if you are using
534	a hand-crafted MORECORE function that cannot handle negative arguments.
535	*/
536
537	/ #define MORECORE_CANNOT_TRIM /
538
539	/ MORECORE_CLEARS (default 1)*
540	The degree to which the routine mapped to MORECORE zeroes out
541	memory: never (0), only for newly allocated space (1) or always
542	(2). The distinction between (1) and (2) is necessary because on
543	some systems, if the application first decrements and then
544	increments the break value, the contents of the reallocated space
545	are unspecified.
546	*/
547
548	#ifndef MORECORE_CLEARS
549	# define MORECORE_CLEARS 1
550	#endif
551
552
553	/*
554	MMAP_AS_MORECORE_SIZE is the minimum mmap size argument to use if
555	sbrk fails, and mmap is used as a backup. The value must be a
556	multiple of page size. This backup strategy generally applies only
557	when systems have "holes" in address space, so sbrk cannot perform
558	contiguous expansion, but there is still space available on system.
559	On systems for which this is known to be useful (i.e. most linux
560	kernels), this occurs only when programs allocate huge amounts of
561	memory. Between this, and the fact that mmap regions tend to be
562	limited, the size should be large, to avoid too many mmap calls and
563	thus avoid running out of kernel resources. /*
564
565	#ifndef MMAP_AS_MORECORE_SIZE
566	#define MMAP_AS_MORECORE_SIZE (1024 * 1024)
567	#endif
568
569	/*
570	Define HAVE_MREMAP to make realloc() use mremap() to re-allocate
571	large blocks.
572	*/
573
574	#ifndef HAVE_MREMAP
575	#define HAVE_MREMAP 0
576	#endif
577
578	/*
579	This version of malloc supports the standard SVID/XPG mallinfo
580	routine that returns a struct containing usage properties and
581	statistics. It should work on any SVID/XPG compliant system that has
582	a /usr/include/malloc.h defining struct mallinfo. (If you'd like to
583	install such a thing yourself, cut out the preliminary declarations
584	as described above and below and save them in a malloc.h file. But
585	there's no compelling reason to bother to do this.)
586
587	The main declaration needed is the mallinfo struct that is returned
588	(by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a
589	bunch of fields that are not even meaningful in this version of
590	malloc. These fields are are instead filled by mallinfo() with
591	other numbers that might be of interest.
592	*/
593
594
595	/ ---------- description of public routines ------------ /
596
597	#if IS_IN (libc)
598	/*
599	malloc(size_t n)
600	Returns a pointer to a newly allocated chunk of at least n bytes, or null
601	if no space is available. Additionally, on failure, errno is
602	set to ENOMEM on ANSI C systems.
603
604	If n is zero, malloc returns a minimum-sized chunk. (The minimum
605	size is 16 bytes on most 32bit systems, and 24 or 32 bytes on 64bit
606	systems.) On most systems, size_t is an unsigned type, so calls
607	with negative arguments are interpreted as requests for huge amounts
608	of space, which will often fail. The maximum supported value of n
609	differs across systems, but is in all cases less than the maximum
610	representable value of a size_t.
611	*/
612	void* __libc_malloc(size_t);
613	libc_hidden_proto (__libc_malloc)
614
615	/*
616	free(void p)*
617	Releases the chunk of memory pointed to by p, that had been previously
618	allocated using malloc or a related routine such as realloc.
619	It has no effect if p is null. It can have arbitrary (i.e., bad!)
620	effects if p has already been freed.
621
622	Unless disabled (using mallopt), freeing very large spaces will
623	when possible, automatically trigger operations that give
624	back unused memory to the system, thus reducing program footprint.
625	*/
626	void __libc_free(void*);
627	libc_hidden_proto (__libc_free)
628
629	/*
630	calloc(size_t n_elements, size_t element_size);
631	Returns a pointer to n_elements element_size bytes, with all locations*
632	set to zero.
633	*/
634	void* __libc_calloc(size_t, size_t);
635
636	/*
637	realloc(void p, size_t n)*
638	Returns a pointer to a chunk of size n that contains the same data
639	as does chunk p up to the minimum of (n, p's size) bytes, or null
640	if no space is available.
641
642	The returned pointer may or may not be the same as p. The algorithm
643	prefers extending p when possible, otherwise it employs the
644	equivalent of a malloc-copy-free sequence.
645
646	If p is null, realloc is equivalent to malloc.
647
648	If space is not available, realloc returns null, errno is set (if on
649	ANSI) and p is NOT freed.
650
651	if n is for fewer bytes than already held by p, the newly unused
652	space is lopped off and freed if possible. Unless the #define
653	REALLOC_ZERO_BYTES_FREES is set, realloc with a size argument of
654	zero (re)allocates a minimum-sized chunk.
655
656	Large chunks that were internally obtained via mmap will always be
657	grown using malloc-copy-free sequences unless the system supports
658	MREMAP (currently only linux).
659
660	The old unix realloc convention of allowing the last-free'd chunk
661	to be used as an argument to realloc is not supported.
662	*/
663	void* __libc_realloc(void*, size_t);
664	libc_hidden_proto (__libc_realloc)
665
666	/*
667	memalign(size_t alignment, size_t n);
668	Returns a pointer to a newly allocated chunk of n bytes, aligned
669	in accord with the alignment argument.
670
671	The alignment argument should be a power of two. If the argument is
672	not a power of two, the nearest greater power is used.
673	8-byte alignment is guaranteed by normal malloc calls, so don't
674	bother calling memalign with an argument of 8 or less.
675
676	Overreliance on memalign is a sure way to fragment space.
677	*/
678	void* __libc_memalign(size_t, size_t);
679	libc_hidden_proto (__libc_memalign)
680
681	/*
682	valloc(size_t n);
683	Equivalent to memalign(pagesize, n), where pagesize is the page
684	size of the system. If the pagesize is unknown, 4096 is used.
685	*/
686	void* __libc_valloc(size_t);
687
688
689
690	/*
691	mallinfo()
692	Returns (by copy) a struct containing various summary statistics:
693
694	arena: current total non-mmapped bytes allocated from system
695	ordblks: the number of free chunks
696	smblks: the number of fastbin blocks (i.e., small chunks that
697	have been freed but not use resused or consolidated)
698	hblks: current number of mmapped regions
699	hblkhd: total bytes held in mmapped regions
700	usmblks: always 0
701	fsmblks: total bytes held in fastbin blocks
702	uordblks: current total allocated space (normal or mmapped)
703	fordblks: total free space
704	keepcost: the maximum number of bytes that could ideally be released
705	back to system via malloc_trim. ("ideally" means that
706	it ignores page restrictions etc.)
707
708	Because these fields are ints, but internal bookkeeping may
709	be kept as longs, the reported values may wrap around zero and
710	thus be inaccurate.
711	*/
712	struct mallinfo2 __libc_mallinfo2(void);
713	libc_hidden_proto (__libc_mallinfo2)
714
715	struct mallinfo __libc_mallinfo(void);
716
717
718	/*
719	pvalloc(size_t n);
720	Equivalent to valloc(minimum-page-that-holds(n)), that is,
721	round up n to nearest pagesize.
722	*/
723	void* __libc_pvalloc(size_t);
724
725	/*
726	malloc_trim(size_t pad);
727
728	If possible, gives memory back to the system (via negative
729	arguments to sbrk) if there is unused memory at the `high' end of
730	the malloc pool. You can call this after freeing large blocks of
731	memory to potentially reduce the system-level memory requirements
732	of a program. However, it cannot guarantee to reduce memory. Under
733	some allocation patterns, some large free blocks of memory will be
734	locked between two used chunks, so they cannot be given back to
735	the system.
736
737	The `pad' argument to malloc_trim represents the amount of free
738	trailing space to leave untrimmed. If this argument is zero,
739	only the minimum amount of memory to maintain internal data
740	structures will be left (one page or less). Non-zero arguments
741	can be supplied to maintain enough trailing space to service
742	future expected allocations without having to re-obtain memory
743	from the system.
744
745	Malloc_trim returns 1 if it actually released any memory, else 0.
746	On systems that do not support "negative sbrks", it will always
747	return 0.
748	*/
749	int __malloc_trim(size_t);
750
751	/*
752	malloc_usable_size(void p);*
753
754	Returns the number of bytes you can actually use in
755	an allocated chunk, which may be more than you requested (although
756	often not) due to alignment and minimum size constraints.
757	You can use this many bytes without worrying about
758	overwriting other allocated objects. This is not a particularly great
759	programming practice. malloc_usable_size can be more useful in
760	debugging and assertions, for example:
761
762	p = malloc(n);
763	assert(malloc_usable_size(p) >= 256);
764
765	*/
766	size_t __malloc_usable_size(void*);
767
768	/*
769	malloc_stats();
770	Prints on stderr the amount of space obtained from the system (both
771	via sbrk and mmap), the maximum amount (which may be more than
772	current if malloc_trim and/or munmap got called), and the current
773	number of bytes allocated via malloc (or realloc, etc) but not yet
774	freed. Note that this is the number of bytes allocated, not the
775	number requested. It will be larger than the number requested
776	because of alignment and bookkeeping overhead. Because it includes
777	alignment wastage as being in use, this figure may be greater than
778	zero even when no user-level chunks are allocated.
779
780	The reported current and maximum system memory can be inaccurate if
781	a program makes other calls to system memory allocation functions
782	(normally sbrk) outside of malloc.
783
784	malloc_stats prints only the most commonly interesting statistics.
785	More information can be obtained by calling mallinfo.
786
787	*/
788	void __malloc_stats(void);
789
790	/*
791	posix_memalign(void memptr, size_t alignment, size_t size);
792
793	POSIX wrapper like memalign(), checking for validity of size.
794	*/
795	int __posix_memalign(void **, size_t, size_t);
796	#endif /* IS_IN (libc) */
797
798	/*
799	mallopt(int parameter_number, int parameter_value)
800	Sets tunable parameters The format is to provide a
801	(parameter-number, parameter-value) pair. mallopt then sets the
802	corresponding parameter to the argument value if it can (i.e., so
803	long as the value is meaningful), and returns 1 if successful else
804	0. SVID/XPG/ANSI defines four standard param numbers for mallopt,
805	normally defined in malloc.h. Only one of these (M_MXFAST) is used
806	in this malloc. The others (M_NLBLKS, M_GRAIN, M_KEEP) don't apply,
807	so setting them has no effect. But this malloc also supports four
808	other options in mallopt. See below for details. Briefly, supported
809	parameters are as follows (listed defaults are for "typical"
810	configurations).
811
812	Symbol param # default allowed param values
813	M_MXFAST 1 64 0-80 (0 disables fastbins)
814	M_TRIM_THRESHOLD -1 1281024 any (-1U disables trimming)*
815	M_TOP_PAD -2 0 any
816	M_MMAP_THRESHOLD -3 1281024 any (or 0 if no MMAP support)*
817	M_MMAP_MAX -4 65536 any (0 disables use of mmap)
818	*/
819	int __libc_mallopt(int, int);
820	#if IS_IN (libc)
821	libc_hidden_proto (__libc_mallopt)
822	#endif
823
824	/ mallopt tuning options /
825
826	/*
827	M_MXFAST is the maximum request size used for "fastbins", special bins
828	that hold returned chunks without consolidating their spaces. This
829	enables future requests for chunks of the same size to be handled
830	very quickly, but can increase fragmentation, and thus increase the
831	overall memory footprint of a program.
832
833	This malloc manages fastbins very conservatively yet still
834	efficiently, so fragmentation is rarely a problem for values less
835	than or equal to the default. The maximum supported value of MXFAST
836	is 80. You wouldn't want it any higher than this anyway. Fastbins
837	are designed especially for use with many small structs, objects or
838	strings -- the default handles structs/objects/arrays with sizes up
839	to 8 4byte fields, or small strings representing words, tokens,
840	etc. Using fastbins for larger objects normally worsens
841	fragmentation without improving speed.
842
843	M_MXFAST is set in REQUEST size units. It is internally used in
844	chunksize units, which adds padding and alignment. You can reduce
845	M_MXFAST to 0 to disable all use of fastbins. This causes the malloc
846	algorithm to be a closer approximation of fifo-best-fit in all cases,
847	not just for larger requests, but will generally cause it to be
848	slower.
849	*/
850
851
852	/ M_MXFAST is a standard SVID/XPG tuning option, usually listed in malloc.h /
853	#ifndef M_MXFAST
854	#define M_MXFAST 1
855	#endif
856
857	#ifndef DEFAULT_MXFAST
858	#define DEFAULT_MXFAST (64 * SIZE_SZ / 4)
859	#endif
860
861
862	/*
863	M_TRIM_THRESHOLD is the maximum amount of unused top-most memory
864	to keep before releasing via malloc_trim in free().
865
866	Automatic trimming is mainly useful in long-lived programs.
867	Because trimming via sbrk can be slow on some systems, and can
868	sometimes be wasteful (in cases where programs immediately
869	afterward allocate more large chunks) the value should be high
870	enough so that your overall system performance would improve by
871	releasing this much memory.
872
873	The trim threshold and the mmap control parameters (see below)
874	can be traded off with one another. Trimming and mmapping are
875	two different ways of releasing unused memory back to the
876	system. Between these two, it is often possible to keep
877	system-level demands of a long-lived program down to a bare
878	minimum. For example, in one test suite of sessions measuring
879	the XF86 X server on Linux, using a trim threshold of 128K and a
880	mmap threshold of 192K led to near-minimal long term resource
881	consumption.
882
883	If you are using this malloc in a long-lived program, it should
884	pay to experiment with these values. As a rough guide, you
885	might set to a value close to the average size of a process
886	(program) running on your system. Releasing this much memory
887	would allow such a process to run in memory. Generally, it's
888	worth it to tune for trimming rather tham memory mapping when a
889	program undergoes phases where several large chunks are
890	allocated and released in ways that can reuse each other's
891	storage, perhaps mixed with phases where there are no such
892	chunks at all. And in well-behaved long-lived programs,
893	controlling release of large blocks via trimming versus mapping
894	is usually faster.
895
896	However, in most programs, these parameters serve mainly as
897	protection against the system-level effects of carrying around
898	massive amounts of unneeded memory. Since frequent calls to
899	sbrk, mmap, and munmap otherwise degrade performance, the default
900	parameters are set to relatively high values that serve only as
901	safeguards.
902
903	The trim value It must be greater than page size to have any useful
904	effect. To disable trimming completely, you can set to
905	(unsigned long)(-1)
906
907	Trim settings interact with fastbin (MXFAST) settings: Unless
908	TRIM_FASTBINS is defined, automatic trimming never takes place upon
909	freeing a chunk with size less than or equal to MXFAST. Trimming is
910	instead delayed until subsequent freeing of larger chunks. However,
911	you can still force an attempted trim by calling malloc_trim.
912
913	Also, trimming is not generally possible in cases where
914	the main arena is obtained via mmap.
915
916	Note that the trick some people use of mallocing a huge space and
917	then freeing it at program startup, in an attempt to reserve system
918	memory, doesn't have the intended effect under automatic trimming,
919	since that memory will immediately be returned to the system.
920	*/
921
922	#define M_TRIM_THRESHOLD -1
923
924	#ifndef DEFAULT_TRIM_THRESHOLD
925	#define DEFAULT_TRIM_THRESHOLD (128 * 1024)
926	#endif
927
928	/*
929	M_TOP_PAD is the amount of extra `padding' space to allocate or
930	retain whenever sbrk is called. It is used in two ways internally:
931
932	* When sbrk is called to extend the top of the arena to satisfy
933	a new malloc request, this much padding is added to the sbrk
934	request.
935
936	* When malloc_trim is called automatically from free(),
937	it is used as the `pad' argument.
938
939	In both cases, the actual amount of padding is rounded
940	so that the end of the arena is always a system page boundary.
941
942	The main reason for using padding is to avoid calling sbrk so
943	often. Having even a small pad greatly reduces the likelihood
944	that nearly every malloc request during program start-up (or
945	after trimming) will invoke sbrk, which needlessly wastes
946	time.
947
948	Automatic rounding-up to page-size units is normally sufficient
949	to avoid measurable overhead, so the default is 0. However, in
950	systems where sbrk is relatively slow, it can pay to increase
951	this value, at the expense of carrying around more memory than
952	the program needs.
953	*/
954
955	#define M_TOP_PAD -2
956
957	#ifndef DEFAULT_TOP_PAD
958	#define DEFAULT_TOP_PAD (0)
959	#endif
960
961	/*
962	MMAP_THRESHOLD_MAX and _MIN are the bounds on the dynamically
963	adjusted MMAP_THRESHOLD.
964	*/
965
966	#ifndef DEFAULT_MMAP_THRESHOLD_MIN
967	#define DEFAULT_MMAP_THRESHOLD_MIN (128 * 1024)
968	#endif
969
970	#ifndef DEFAULT_MMAP_THRESHOLD_MAX
971	/ For 32-bit platforms we cannot increase the maximum mmap*
972	threshold much because it is also the minimum value for the
973	maximum heap size and its alignment. Going above 512k (i.e., 1M
974	for new heaps) wastes too much address space. /*
975	# if __WORDSIZE == 32
976	# define DEFAULT_MMAP_THRESHOLD_MAX (512 * 1024)
977	# else
978	# define DEFAULT_MMAP_THRESHOLD_MAX (4 * 1024 * 1024 * sizeof(long))
979	# endif
980	#endif
981
982	/*
983	M_MMAP_THRESHOLD is the request size threshold for using mmap()
984	to service a request. Requests of at least this size that cannot
985	be allocated using already-existing space will be serviced via mmap.
986	(If enough normal freed space already exists it is used instead.)
987
988	Using mmap segregates relatively large chunks of memory so that
989	they can be individually obtained and released from the host
990	system. A request serviced through mmap is never reused by any
991	other request (at least not directly; the system may just so
992	happen to remap successive requests to the same locations).
993
994	Segregating space in this way has the benefits that:
995
996	1. Mmapped space can ALWAYS be individually released back
997	to the system, which helps keep the system level memory
998	demands of a long-lived program low.
999	2. Mapped memory can never become `locked' between
1000	other chunks, as can happen with normally allocated chunks, which
1001	means that even trimming via malloc_trim would not release them.
1002	3. On some systems with "holes" in address spaces, mmap can obtain
1003	memory that sbrk cannot.
1004
1005	However, it has the disadvantages that:
1006
1007	1. The space cannot be reclaimed, consolidated, and then
1008	used to service later requests, as happens with normal chunks.
1009	2. It can lead to more wastage because of mmap page alignment
1010	requirements
1011	3. It causes malloc performance to be more dependent on host
1012	system memory management support routines which may vary in
1013	implementation quality and may impose arbitrary
1014	limitations. Generally, servicing a request via normal
1015	malloc steps is faster than going through a system's mmap.
1016
1017	The advantages of mmap nearly always outweigh disadvantages for
1018	"large" chunks, but the value of "large" varies across systems. The
1019	default is an empirically derived value that works well in most
1020	systems.
1021
1022
1023	Update in 2006:
1024	The above was written in 2001. Since then the world has changed a lot.
1025	Memory got bigger. Applications got bigger. The virtual address space
1026	layout in 32 bit linux changed.
1027
1028	In the new situation, brk() and mmap space is shared and there are no
1029	artificial limits on brk size imposed by the kernel. What is more,
1030	applications have started using transient allocations larger than the
1031	128Kb as was imagined in 2001.
1032
1033	The price for mmap is also high now; each time glibc mmaps from the
1034	kernel, the kernel is forced to zero out the memory it gives to the
1035	application. Zeroing memory is expensive and eats a lot of cache and
1036	memory bandwidth. This has nothing to do with the efficiency of the
1037	virtual memory system, by doing mmap the kernel just has no choice but
1038	to zero.
1039
1040	In 2001, the kernel had a maximum size for brk() which was about 800
1041	megabytes on 32 bit x86, at that point brk() would hit the first
1042	mmaped shared libaries and couldn't expand anymore. With current 2.6
1043	kernels, the VA space layout is different and brk() and mmap
1044	both can span the entire heap at will.
1045
1046	Rather than using a static threshold for the brk/mmap tradeoff,
1047	we are now using a simple dynamic one. The goal is still to avoid
1048	fragmentation. The old goals we kept are
1049	1) try to get the long lived large allocations to use mmap()
1050	2) really large allocations should always use mmap()
1051	and we're adding now:
1052	3) transient allocations should use brk() to avoid forcing the kernel
1053	having to zero memory over and over again
1054
1055	The implementation works with a sliding threshold, which is by default
1056	limited to go between 128Kb and 32Mb (64Mb for 64 bitmachines) and starts
1057	out at 128Kb as per the 2001 default.
1058
1059	This allows us to satisfy requirement 1) under the assumption that long
1060	lived allocations are made early in the process' lifespan, before it has
1061	started doing dynamic allocations of the same size (which will
1062	increase the threshold).
1063
1064	The upperbound on the threshold satisfies requirement 2)
1065
1066	The threshold goes up in value when the application frees memory that was
1067	allocated with the mmap allocator. The idea is that once the application
1068	starts freeing memory of a certain size, it's highly probable that this is
1069	a size the application uses for transient allocations. This estimator
1070	is there to satisfy the new third requirement.
1071
1072	*/
1073
1074	#define M_MMAP_THRESHOLD -3
1075
1076	#ifndef DEFAULT_MMAP_THRESHOLD
1077	#define DEFAULT_MMAP_THRESHOLD DEFAULT_MMAP_THRESHOLD_MIN
1078	#endif
1079
1080	/*
1081	M_MMAP_MAX is the maximum number of requests to simultaneously
1082	service using mmap. This parameter exists because
1083	some systems have a limited number of internal tables for
1084	use by mmap, and using more than a few of them may degrade
1085	performance.
1086
1087	The default is set to a value that serves only as a safeguard.
1088	Setting to 0 disables use of mmap for servicing large requests.
1089	*/
1090
1091	#define M_MMAP_MAX -4
1092
1093	#ifndef DEFAULT_MMAP_MAX
1094	#define DEFAULT_MMAP_MAX (65536)
1095	#endif
1096
1097	#include <malloc.h>
1098
1099	#ifndef RETURN_ADDRESS
1100	#define RETURN_ADDRESS(X_) (NULL)
1101	#endif
1102
1103	/ Forward declarations. /
1104	struct malloc_chunk;
1105	typedef struct malloc_chunk* mchunkptr;
1106
1107	/ Internal routines. /
1108
1109	static void* _int_malloc(mstate, size_t);
1110	static void _int_free(mstate, mchunkptr, int);
1111	static void* _int_realloc(mstate, mchunkptr, INTERNAL_SIZE_T,
1112	INTERNAL_SIZE_T);
1113	static void* _int_memalign(mstate, size_t, size_t);
1114	#if IS_IN (libc)
1115	static void* _mid_memalign(size_t, size_t, void *);
1116	#endif
1117
1118	static void malloc_printerr(const char str) __attribute__* ((noreturn));
1119
1120	static void munmap_chunk(mchunkptr p);
1121	#if HAVE_MREMAP
1122	static mchunkptr mremap_chunk(mchunkptr p, size_t new_size);
1123	#endif
1124
1125	/ ------------------ MMAP support ------------------ /
1126
1127
1128	#include <fcntl.h>
1129	#include <sys/mman.h>
1130
1131	#if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
1132	# define MAP_ANONYMOUS MAP_ANON
1133	#endif
1134
1135	#ifndef MAP_NORESERVE
1136	# define MAP_NORESERVE 0
1137	#endif
1138
1139	#define MMAP(addr, size, prot, flags) \
1140	__mmap((addr), (size), (prot), (flags)\|MAP_ANONYMOUS\|MAP_PRIVATE, -1, 0)
1141
1142
1143	/*
1144	----------------------- Chunk representations -----------------------
1145	*/
1146
1147
1148	/*
1149	This struct declaration is misleading (but accurate and necessary).
1150	It declares a "view" into memory allowing access to necessary
1151	fields at known offsets from a given base. See explanation below.
1152	*/
1153
1154	struct malloc_chunk {
1155
1156	INTERNAL_SIZE_T mchunk_prev_size; / Size of previous chunk (if free). /
1157	INTERNAL_SIZE_T mchunk_size; / Size in bytes, including overhead. /
1158
1159	struct malloc_chunk* fd; / double links -- used only if free. /
1160	struct malloc_chunk* bk;
1161
1162	/ Only used for large blocks: pointer to next larger size. /
1163	struct malloc_chunk* fd_nextsize; / double links -- used only if free. /
1164	struct malloc_chunk* bk_nextsize;
1165	};
1166
1167
1168	/*
1169	malloc_chunk details:
1170
1171	(The following includes lightly edited explanations by Colin Plumb.)
1172
1173	Chunks of memory are maintained using a `boundary tag' method as
1174	described in e.g., Knuth or Standish. (See the paper by Paul
1175	Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a
1176	survey of such techniques.) Sizes of free chunks are stored both
1177	in the front of each chunk and at the end. This makes
1178	consolidating fragmented chunks into bigger chunks very fast. The
1179	size fields also hold bits representing whether chunks are free or
1180	in use.
1181
1182	An allocated chunk looks like this:
1183
1184
1185	chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1186	\| Size of previous chunk, if unallocated (P clear) \|
1187	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1188	\| Size of chunk, in bytes \|A\|M\|P\|
1189	mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1190	\| User data starts here... .
1191	. .
1192	. (malloc_usable_size() bytes) .
1193	. \|
1194	nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1195	\| (size of chunk, but used for application data) \|
1196	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1197	\| Size of next chunk, in bytes \|A\|0\|1\|
1198	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1199
1200	Where "chunk" is the front of the chunk for the purpose of most of
1201	the malloc code, but "mem" is the pointer that is returned to the
1202	user. "Nextchunk" is the beginning of the next contiguous chunk.
1203
1204	Chunks always begin on even word boundaries, so the mem portion
1205	(which is returned to the user) is also on an even word boundary, and
1206	thus at least double-word aligned.
1207
1208	Free chunks are stored in circular doubly-linked lists, and look like this:
1209
1210	chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1211	\| Size of previous chunk, if unallocated (P clear) \|
1212	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1213	`head:' \| Size of chunk, in bytes \|A\|0\|P\|
1214	mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1215	\| Forward pointer to next chunk in list \|
1216	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1217	\| Back pointer to previous chunk in list \|
1218	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1219	\| Unused space (may be 0 bytes long) .
1220	. .
1221	. \|
1222	nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1223	`foot:' \| Size of chunk, in bytes \|
1224	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1225	\| Size of next chunk, in bytes \|A\|0\|0\|
1226	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1227
1228	The P (PREV_INUSE) bit, stored in the unused low-order bit of the
1229	chunk size (which is always a multiple of two words), is an in-use
1230	bit for the previous* chunk. If that bit is clear, then the*
1231	word before the current chunk size contains the previous chunk
1232	size, and can be used to find the front of the previous chunk.
1233	The very first chunk allocated always has this bit set,
1234	preventing access to non-existent (or non-owned) memory. If
1235	prev_inuse is set for any given chunk, then you CANNOT determine
1236	the size of the previous chunk, and might even get a memory
1237	addressing fault when trying to do so.
1238
1239	The A (NON_MAIN_ARENA) bit is cleared for chunks on the initial,
1240	main arena, described by the main_arena variable. When additional
1241	threads are spawned, each thread receives its own arena (up to a
1242	configurable limit, after which arenas are reused for multiple
1243	threads), and the chunks in these arenas have the A bit set. To
1244	find the arena for a chunk on such a non-main arena, heap_for_ptr
1245	performs a bit mask operation and indirection through the ar_ptr
1246	member of the per-heap header heap_info (see arena.c).
1247
1248	Note that the `foot' of the current chunk is actually represented
1249	as the prev_size of the NEXT chunk. This makes it easier to
1250	deal with alignments etc but can be very confusing when trying
1251	to extend or adapt this code.
1252
1253	The three exceptions to all this are:
1254
1255	1. The special chunk `top' doesn't bother using the
1256	trailing size field since there is no next contiguous chunk
1257	that would have to index off it. After initialization, `top'
1258	is forced to always exist. If it would become less than
1259	MINSIZE bytes long, it is replenished.
1260
1261	2. Chunks allocated via mmap, which have the second-lowest-order
1262	bit M (IS_MMAPPED) set in their size fields. Because they are
1263	allocated one-by-one, each must contain its own trailing size
1264	field. If the M bit is set, the other bits are ignored
1265	(because mmapped chunks are neither in an arena, nor adjacent
1266	to a freed chunk). The M bit is also used for chunks which
1267	originally came from a dumped heap via malloc_set_state in
1268	hooks.c.
1269
1270	3. Chunks in fastbins are treated as allocated chunks from the
1271	point of view of the chunk allocator. They are consolidated
1272	with their neighbors only in bulk, in malloc_consolidate.
1273	*/
1274
1275	/*
1276	---------- Size and alignment checks and conversions ----------
1277	*/
1278
1279	/ Conversion from malloc headers to user pointers, and back. When*
1280	using memory tagging the user data and the malloc data structure
1281	headers have distinct tags. Converting fully from one to the other
1282	involves extracting the tag at the other address and creating a
1283	suitable pointer using it. That can be quite expensive. There are
1284	cases when the pointers are not dereferenced (for example only used
1285	for alignment check) so the tags are not relevant, and there are
1286	cases when user data is not tagged distinctly from malloc headers
1287	(user data is untagged because tagging is done late in malloc and
1288	early in free). User memory tagging across internal interfaces:
1289
1290	sysmalloc: Returns untagged memory.
1291	_int_malloc: Returns untagged memory.
1292	_int_free: Takes untagged memory.
1293	_int_memalign: Returns untagged memory.
1294	_int_memalign: Returns untagged memory.
1295	_mid_memalign: Returns tagged memory.
1296	_int_realloc: Takes and returns tagged memory.
1297	*/
1298
1299	/ The chunk header is two SIZE_SZ elements, but this is used widely, so*
1300	we define it here for clarity later. /*
1301	#define CHUNK_HDR_SZ (2 * SIZE_SZ)
1302
1303	/ Convert a chunk address to a user mem pointer without correcting*
1304	the tag. /*
1305	#define chunk2mem(p) ((void)((char)(p) + CHUNK_HDR_SZ))
1306
1307	/ Convert a chunk address to a user mem pointer and extract the right tag. /
1308	#define chunk2mem_tag(p) ((void)tag_at ((char)(p) + CHUNK_HDR_SZ))
1309
1310	/ Convert a user mem pointer to a chunk address and extract the right tag. /
1311	#define mem2chunk(mem) ((mchunkptr)tag_at (((char*)(mem) - CHUNK_HDR_SZ)))
1312
1313	/ The smallest possible chunk /
1314	#define MIN_CHUNK_SIZE (offsetof(struct malloc_chunk, fd_nextsize))
1315
1316	/ The smallest size we can malloc is an aligned minimal chunk /
1317
1318	#define MINSIZE \
1319	(unsigned long)(((MIN_CHUNK_SIZE+MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK))
1320
1321	/ Check if m has acceptable alignment /
1322
1323	#define aligned_OK(m) (((unsigned long)(m) & MALLOC_ALIGN_MASK) == 0)
1324
1325	#define misaligned_chunk(p) \
1326	((uintptr_t)(MALLOC_ALIGNMENT == CHUNK_HDR_SZ ? (p) : chunk2mem (p)) \
1327	& MALLOC_ALIGN_MASK)
1328
1329	/ pad request bytes into a usable size -- internal version /
1330	/ Note: This must be a macro that evaluates to a compile time constant*
1331	if passed a literal constant. /*
1332	#define request2size(req) \
1333	(((req) + SIZE_SZ + MALLOC_ALIGN_MASK < MINSIZE) ? \
1334	MINSIZE : \
1335	((req) + SIZE_SZ + MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK)
1336
1337	/ Check if REQ overflows when padded and aligned and if the resulting value*
1338	is less than PTRDIFF_T. Returns TRUE and the requested size or MINSIZE in
1339	case the value is less than MINSIZE on SZ or false if any of the previous
1340	check fail. /*
1341	static inline bool
1342	checked_request2size (size_t req, size_t *sz) __nonnull (`1`)
1343	{
1344	if (__glibc_unlikely (req > PTRDIFF_MAX))
1345	return false;
1346
1347	/ When using tagged memory, we cannot share the end of the user*
1348	block with the header for the next chunk, so ensure that we
1349	allocate blocks that are rounded up to the granule size. Take
1350	care not to overflow from close to MAX_SIZE_T to a small
1351	number. Ideally, this would be part of request2size(), but that
1352	must be a macro that produces a compile time constant if passed
1353	a constant literal. /*
1354	if (__glibc_unlikely (mtag_enabled))
1355	{
1356	/ Ensure this is not evaluated if !mtag_enabled, see gcc PR 99551. /
1357	asm ("");
1358
1359	req = (req + (__MTAG_GRANULE_SIZE - `1`)) &
1360	~(size_t)(__MTAG_GRANULE_SIZE - `1`);
1361	}
1362
1363	*sz = request2size (req);
1364	return true;
1365	}
1366
1367	/*
1368	--------------- Physical chunk operations ---------------
1369	*/
1370
1371
1372	/ size field is or'ed with PREV_INUSE when previous adjacent chunk in use /
1373	#define PREV_INUSE 0x1
1374
1375	/ extract inuse bit of previous chunk /
1376	#define prev_inuse(p) ((p)->mchunk_size & PREV_INUSE)
1377
1378
1379	/ size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() /
1380	#define IS_MMAPPED 0x2
1381
1382	/ check for mmap()'ed chunk /
1383	#define chunk_is_mmapped(p) ((p)->mchunk_size & IS_MMAPPED)
1384
1385
1386	/ size field is or'ed with NON_MAIN_ARENA if the chunk was obtained*
1387	from a non-main arena. This is only set immediately before handing
1388	the chunk to the user, if necessary. /*
1389	#define NON_MAIN_ARENA 0x4
1390
1391	/ Check for chunk from main arena. /
1392	#define chunk_main_arena(p) (((p)->mchunk_size & NON_MAIN_ARENA) == 0)
1393
1394	/ Mark a chunk as not being on the main arena. /
1395	#define set_non_main_arena(p) ((p)->mchunk_size \|= NON_MAIN_ARENA)
1396
1397
1398	/*
1399	Bits to mask off when extracting size
1400
1401	Note: IS_MMAPPED is intentionally not masked off from size field in
1402	macros for which mmapped chunks should never be seen. This should
1403	cause helpful core dumps to occur if it is tried by accident by
1404	people extending or adapting this malloc.
1405	*/
1406	#define SIZE_BITS (PREV_INUSE \| IS_MMAPPED \| NON_MAIN_ARENA)
1407
1408	/ Get size, ignoring use bits /
1409	#define chunksize(p) (chunksize_nomask (p) & ~(SIZE_BITS))
1410
1411	/ Like chunksize, but do not mask SIZE_BITS. /
1412	#define chunksize_nomask(p) ((p)->mchunk_size)
1413
1414	/ Ptr to next physical malloc_chunk. /
1415	#define next_chunk(p) ((mchunkptr) (((char *) (p)) + chunksize (p)))
1416
1417	/ Size of the chunk below P. Only valid if !prev_inuse (P). /
1418	#define prev_size(p) ((p)->mchunk_prev_size)
1419
1420	/ Set the size of the chunk below P. Only valid if !prev_inuse (P). /
1421	#define set_prev_size(p, sz) ((p)->mchunk_prev_size = (sz))
1422
1423	/ Ptr to previous physical malloc_chunk. Only valid if !prev_inuse (P). /
1424	#define prev_chunk(p) ((mchunkptr) (((char *) (p)) - prev_size (p)))
1425
1426	/ Treat space at ptr + offset as a chunk /
1427	#define chunk_at_offset(p, s) ((mchunkptr) (((char *) (p)) + (s)))
1428
1429	/ extract p's inuse bit /
1430	#define inuse(p) \
1431	((((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size) & PREV_INUSE)
1432
1433	/ set/clear chunk as being inuse without otherwise disturbing /
1434	#define set_inuse(p) \
1435	((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size \|= PREV_INUSE
1436
1437	#define clear_inuse(p) \
1438	((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size &= ~(PREV_INUSE)
1439
1440
1441	/ check/set/clear inuse bits in known places /
1442	#define inuse_bit_at_offset(p, s) \
1443	(((mchunkptr) (((char *) (p)) + (s)))->mchunk_size & PREV_INUSE)
1444
1445	#define set_inuse_bit_at_offset(p, s) \
1446	(((mchunkptr) (((char *) (p)) + (s)))->mchunk_size \|= PREV_INUSE)
1447
1448	#define clear_inuse_bit_at_offset(p, s) \
1449	(((mchunkptr) (((char *) (p)) + (s)))->mchunk_size &= ~(PREV_INUSE))
1450
1451
1452	/ Set size at head, without disturbing its use bit /
1453	#define set_head_size(p, s) ((p)->mchunk_size = (((p)->mchunk_size & SIZE_BITS) \| (s)))
1454
1455	/ Set size/use field /
1456	#define set_head(p, s) ((p)->mchunk_size = (s))
1457
1458	/ Set size at footer (only when chunk is not in use) /
1459	#define set_foot(p, s) (((mchunkptr) ((char *) (p) + (s)))->mchunk_prev_size = (s))
1460
1461	#pragma GCC poison mchunk_size
1462	#pragma GCC poison mchunk_prev_size
1463
1464	/ This is the size of the real usable data in the chunk. Not valid for*
1465	dumped heap chunks. /*
1466	#define memsize(p) \
1467	(__MTAG_GRANULE_SIZE > SIZE_SZ && __glibc_unlikely (mtag_enabled) ? \
1468	chunksize (p) - CHUNK_HDR_SZ : \
1469	chunksize (p) - CHUNK_HDR_SZ + (chunk_is_mmapped (p) ? 0 : SIZE_SZ))
1470
1471	/ If memory tagging is enabled the layout changes to accommodate the granule*
1472	size, this is wasteful for small allocations so not done by default.
1473	Both the chunk header and user data has to be granule aligned. /*
1474	_Static_assert (__MTAG_GRANULE_SIZE <= CHUNK_HDR_SZ,
1475	"memory tagging is not supported with large granule.");
1476
1477	static __always_inline void *
1478	tag_new_usable (void *ptr)
1479	{
1480	if (__glibc_unlikely (mtag_enabled) && ptr)
1481	{
1482	mchunkptr cp = mem2chunk(ptr);
1483	ptr = __libc_mtag_tag_region (__libc_mtag_new_tag (ptr), memsize (cp));
1484	}
1485	return ptr;
1486	}
1487
1488	/*
1489	-------------------- Internal data structures --------------------
1490
1491	All internal state is held in an instance of malloc_state defined
1492	below. There are no other static variables, except in two optional
1493	cases:
1494	* If USE_MALLOC_LOCK is defined, the mALLOC_MUTEx declared above.
1495	* If mmap doesn't support MAP_ANONYMOUS, a dummy file descriptor
1496	for mmap.
1497
1498	Beware of lots of tricks that minimize the total bookkeeping space
1499	requirements. The result is a little over 1K bytes (for 4byte
1500	pointers and size_t.)
1501	*/
1502
1503	/*
1504	Bins
1505
1506	An array of bin headers for free chunks. Each bin is doubly
1507	linked. The bins are approximately proportionally (log) spaced.
1508	There are a lot of these bins (128). This may look excessive, but
1509	works very well in practice. Most bins hold sizes that are
1510	unusual as malloc request sizes, but are more usual for fragments
1511	and consolidated sets of chunks, which is what these bins hold, so
1512	they can be found quickly. All procedures maintain the invariant
1513	that no consolidated chunk physically borders another one, so each
1514	chunk in a list is known to be preceeded and followed by either
1515	inuse chunks or the ends of memory.
1516
1517	Chunks in bins are kept in size order, with ties going to the
1518	approximately least recently used chunk. Ordering isn't needed
1519	for the small bins, which all contain the same-sized chunks, but
1520	facilitates best-fit allocation for larger chunks. These lists
1521	are just sequential. Keeping them in order almost never requires
1522	enough traversal to warrant using fancier ordered data
1523	structures.
1524
1525	Chunks of the same size are linked with the most
1526	recently freed at the front, and allocations are taken from the
1527	back. This results in LRU (FIFO) allocation order, which tends
1528	to give each chunk an equal opportunity to be consolidated with
1529	adjacent freed chunks, resulting in larger free chunks and less
1530	fragmentation.
1531
1532	To simplify use in double-linked lists, each bin header acts
1533	as a malloc_chunk. This avoids special-casing for headers.
1534	But to conserve space and improve locality, we allocate
1535	only the fd/bk pointers of bins, and then use repositioning tricks
1536	to treat these as the fields of a malloc_chunk.*
1537	*/
1538
1539	typedef struct malloc_chunk *mbinptr;
1540
1541	/ addressing -- note that bin_at(0) does not exist /
1542	#define bin_at(m, i) \
1543	(mbinptr) (((char ) &((m)->bins[((i) - 1) 2])) \
1544	- offsetof (struct malloc_chunk, fd))
1545
1546	/ analog of ++bin /
1547	#define next_bin(b) ((mbinptr) ((char *) (b) + (sizeof (mchunkptr) << 1)))
1548
1549	/ Reminders about list directionality within bins /
1550	#define first(b) ((b)->fd)
1551	#define last(b) ((b)->bk)
1552
1553	/*
1554	Indexing
1555
1556	Bins for sizes < 512 bytes contain chunks of all the same size, spaced
1557	8 bytes apart. Larger bins are approximately logarithmically spaced:
1558
1559	64 bins of size 8
1560	32 bins of size 64
1561	16 bins of size 512
1562	8 bins of size 4096
1563	4 bins of size 32768
1564	2 bins of size 262144
1565	1 bin of size what's left
1566
1567	There is actually a little bit of slop in the numbers in bin_index
1568	for the sake of speed. This makes no difference elsewhere.
1569
1570	The bins top out around 1MB because we expect to service large
1571	requests via mmap.
1572
1573	Bin 0 does not exist. Bin 1 is the unordered list; if that would be
1574	a valid chunk size the small bins are bumped up one.
1575	*/
1576
1577	#define NBINS 128
1578	#define NSMALLBINS 64
1579	#define SMALLBIN_WIDTH MALLOC_ALIGNMENT
1580	#define SMALLBIN_CORRECTION (MALLOC_ALIGNMENT > CHUNK_HDR_SZ)
1581	#define MIN_LARGE_SIZE ((NSMALLBINS - SMALLBIN_CORRECTION) * SMALLBIN_WIDTH)
1582
1583	#define in_smallbin_range(sz) \
1584	((unsigned long) (sz) < (unsigned long) MIN_LARGE_SIZE)
1585
1586	#define smallbin_index(sz) \
1587	((SMALLBIN_WIDTH == 16 ? (((unsigned) (sz)) >> 4) : (((unsigned) (sz)) >> 3))\
1588	+ SMALLBIN_CORRECTION)
1589
1590	#define largebin_index_32(sz) \
1591	(((((unsigned long) (sz)) >> 6) <= 38) ? 56 + (((unsigned long) (sz)) >> 6) :\
1592	((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1593	((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1594	((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1595	((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1596	126)
1597
1598	#define largebin_index_32_big(sz) \
1599	(((((unsigned long) (sz)) >> 6) <= 45) ? 49 + (((unsigned long) (sz)) >> 6) :\
1600	((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1601	((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1602	((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1603	((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1604	126)
1605
1606	// XXX It remains to be seen whether it is good to keep the widths of
1607	// XXX the buckets the same or whether it should be scaled by a factor
1608	// XXX of two as well.
1609	#define largebin_index_64(sz) \
1610	(((((unsigned long) (sz)) >> 6) <= 48) ? 48 + (((unsigned long) (sz)) >> 6) :\
1611	((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1612	((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1613	((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1614	((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1615	126)
1616
1617	#define largebin_index(sz) \
1618	(SIZE_SZ == 8 ? largebin_index_64 (sz) \
1619	: MALLOC_ALIGNMENT == 16 ? largebin_index_32_big (sz) \
1620	: largebin_index_32 (sz))
1621
1622	#define bin_index(sz) \
1623	((in_smallbin_range (sz)) ? smallbin_index (sz) : largebin_index (sz))
1624
1625	/ Take a chunk off a bin list. /
1626	static void
1627	unlink_chunk (mstate av, mchunkptr p)
1628	{
1629	if (chunksize (p) != prev_size (next_chunk (p)))
1630	malloc_printerr ("corrupted size vs. prev_size");
1631
1632	mchunkptr fd = p->fd;
1633	mchunkptr bk = p->bk;
1634
1635	if (__builtin_expect (fd->bk != p \|\| bk->fd != p, `0`))
1636	malloc_printerr ("corrupted double-linked list");
1637
1638	fd->bk = bk;
1639	bk->fd = fd;
1640	if (!in_smallbin_range (chunksize_nomask (p)) && p->fd_nextsize != NULL)
1641	{
1642	if (p->fd_nextsize->bk_nextsize != p
1643	\|\| p->bk_nextsize->fd_nextsize != p)
1644	malloc_printerr ("corrupted double-linked list (not small)");
1645
1646	if (fd->fd_nextsize == NULL)
1647	{
1648	if (p->fd_nextsize == p)
1649	fd->fd_nextsize = fd->bk_nextsize = fd;
1650	else
1651	{
1652	fd->fd_nextsize = p->fd_nextsize;
1653	fd->bk_nextsize = p->bk_nextsize;
1654	p->fd_nextsize->bk_nextsize = fd;
1655	p->bk_nextsize->fd_nextsize = fd;
1656	}
1657	}
1658	else
1659	{
1660	p->fd_nextsize->bk_nextsize = p->bk_nextsize;
1661	p->bk_nextsize->fd_nextsize = p->fd_nextsize;
1662	}
1663	}
1664	}
1665
1666	/*
1667	Unsorted chunks
1668
1669	All remainders from chunk splits, as well as all returned chunks,
1670	are first placed in the "unsorted" bin. They are then placed
1671	in regular bins after malloc gives them ONE chance to be used before
1672	binning. So, basically, the unsorted_chunks list acts as a queue,
1673	with chunks being placed on it in free (and malloc_consolidate),
1674	and taken off (to be either used or placed in bins) in malloc.
1675
1676	The NON_MAIN_ARENA flag is never set for unsorted chunks, so it
1677	does not have to be taken into account in size comparisons.
1678	*/
1679
1680	/ The otherwise unindexable 1-bin is used to hold unsorted chunks. /
1681	#define unsorted_chunks(M) (bin_at (M, 1))
1682
1683	/*
1684	Top
1685
1686	The top-most available chunk (i.e., the one bordering the end of
1687	available memory) is treated specially. It is never included in
1688	any bin, is used only if no other chunk is available, and is
1689	released back to the system if it is very large (see
1690	M_TRIM_THRESHOLD). Because top initially
1691	points to its own bin with initial zero size, thus forcing
1692	extension on the first malloc request, we avoid having any special
1693	code in malloc to check whether it even exists yet. But we still
1694	need to do so when getting memory from system, so we make
1695	initial_top treat the bin as a legal but unusable chunk during the
1696	interval between initialization and the first call to
1697	sysmalloc. (This is somewhat delicate, since it relies on
1698	the 2 preceding words to be zero during this interval as well.)
1699	*/
1700
1701	/ Conveniently, the unsorted bin can be used as dummy top on first call /
1702	#define initial_top(M) (unsorted_chunks (M))
1703
1704	/*
1705	Binmap
1706
1707	To help compensate for the large number of bins, a one-level index
1708	structure is used for bin-by-bin searching. `binmap' is a
1709	bitvector recording whether bins are definitely empty so they can
1710	be skipped over during during traversals. The bits are NOT always
1711	cleared as soon as bins are empty, but instead only
1712	when they are noticed to be empty during traversal in malloc.
1713	*/
1714
1715	/ Conservatively use 32 bits per map word, even if on 64bit system /
1716	#define BINMAPSHIFT 5
1717	#define BITSPERMAP (1U << BINMAPSHIFT)
1718	#define BINMAPSIZE (NBINS / BITSPERMAP)
1719
1720	#define idx2block(i) ((i) >> BINMAPSHIFT)
1721	#define idx2bit(i) ((1U << ((i) & ((1U << BINMAPSHIFT) - 1))))
1722
1723	#define mark_bin(m, i) ((m)->binmap[idx2block (i)] \|= idx2bit (i))
1724	#define unmark_bin(m, i) ((m)->binmap[idx2block (i)] &= ~(idx2bit (i)))
1725	#define get_binmap(m, i) ((m)->binmap[idx2block (i)] & idx2bit (i))
1726
1727	/*
1728	Fastbins
1729
1730	An array of lists holding recently freed small chunks. Fastbins
1731	are not doubly linked. It is faster to single-link them, and
1732	since chunks are never removed from the middles of these lists,
1733	double linking is not necessary. Also, unlike regular bins, they
1734	are not even processed in FIFO order (they use faster LIFO) since
1735	ordering doesn't much matter in the transient contexts in which
1736	fastbins are normally used.
1737
1738	Chunks in fastbins keep their inuse bit set, so they cannot
1739	be consolidated with other free chunks. malloc_consolidate
1740	releases all chunks in fastbins and consolidates them with
1741	other free chunks.
1742	*/
1743
1744	typedef struct malloc_chunk *mfastbinptr;
1745	#define fastbin(ar_ptr, idx) ((ar_ptr)->fastbinsY[idx])
1746
1747	/ offset 2 to use otherwise unindexable first 2 bins /
1748	#define fastbin_index(sz) \
1749	((((unsigned int) (sz)) >> (SIZE_SZ == 8 ? 4 : 3)) - 2)
1750
1751
1752	/ The maximum fastbin request size we support /
1753	#define MAX_FAST_SIZE (80 * SIZE_SZ / 4)
1754
1755	#define NFASTBINS (fastbin_index (request2size (MAX_FAST_SIZE)) + 1)
1756
1757	/*
1758	FASTBIN_CONSOLIDATION_THRESHOLD is the size of a chunk in free()
1759	that triggers automatic consolidation of possibly-surrounding
1760	fastbin chunks. This is a heuristic, so the exact value should not
1761	matter too much. It is defined at half the default trim threshold as a
1762	compromise heuristic to only attempt consolidation if it is likely
1763	to lead to trimming. However, it is not dynamically tunable, since
1764	consolidation reduces fragmentation surrounding large chunks even
1765	if trimming is not used.
1766	*/
1767
1768	#define FASTBIN_CONSOLIDATION_THRESHOLD (65536UL)
1769
1770	/*
1771	NONCONTIGUOUS_BIT indicates that MORECORE does not return contiguous
1772	regions. Otherwise, contiguity is exploited in merging together,
1773	when possible, results from consecutive MORECORE calls.
1774
1775	The initial value comes from MORECORE_CONTIGUOUS, but is
1776	changed dynamically if mmap is ever used as an sbrk substitute.
1777	*/
1778
1779	#define NONCONTIGUOUS_BIT (2U)
1780
1781	#define contiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) == 0)
1782	#define noncontiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) != 0)
1783	#define set_noncontiguous(M) ((M)->flags \|= NONCONTIGUOUS_BIT)
1784	#define set_contiguous(M) ((M)->flags &= ~NONCONTIGUOUS_BIT)
1785
1786	/ Maximum size of memory handled in fastbins. /
1787	static INTERNAL_SIZE_T global_max_fast;
1788
1789	/*
1790	Set value of max_fast.
1791	Use impossibly small value if 0.
1792	Precondition: there are no existing fastbin chunks in the main arena.
1793	Since do_check_malloc_state () checks this, we call malloc_consolidate ()
1794	before changing max_fast. Note other arenas will leak their fast bin
1795	entries if max_fast is reduced.
1796	*/
1797
1798	#define set_max_fast(s) \
1799	global_max_fast = (((size_t) (s) <= MALLOC_ALIGN_MASK - SIZE_SZ) \
1800	? MIN_CHUNK_SIZE / 2 : ((s + SIZE_SZ) & ~MALLOC_ALIGN_MASK))
1801
1802	static inline INTERNAL_SIZE_T
1803	get_max_fast (void)
1804	{
1805	/ Tell the GCC optimizers that global_max_fast is never larger*
1806	than MAX_FAST_SIZE. This avoids out-of-bounds array accesses in
1807	_int_malloc after constant propagation of the size parameter.
1808	(The code never executes because malloc preserves the
1809	global_max_fast invariant, but the optimizers may not recognize
1810	this.) /*
1811	if (global_max_fast > MAX_FAST_SIZE)
1812	__builtin_unreachable ();
1813	return global_max_fast;
1814	}
1815
1816	/*
1817	----------- Internal state representation and initialization -----------
1818	*/
1819
1820	/*
1821	have_fastchunks indicates that there are probably some fastbin chunks.
1822	It is set true on entering a chunk into any fastbin, and cleared early in
1823	malloc_consolidate. The value is approximate since it may be set when there
1824	are no fastbin chunks, or it may be clear even if there are fastbin chunks
1825	available. Given it's sole purpose is to reduce number of redundant calls to
1826	malloc_consolidate, it does not affect correctness. As a result we can safely
1827	use relaxed atomic accesses.
1828	*/
1829
1830
1831	struct malloc_state
1832	{
1833	/ Serialize access. /
1834	__libc_lock_define (, mutex);
1835
1836	/ Flags (formerly in max_fast). /
1837	int flags;
1838
1839	/ Set if the fastbin chunks contain recently inserted free blocks. /
1840	/ Note this is a bool but not all targets support atomics on booleans. /
1841	int have_fastchunks;
1842
1843	/ Fastbins /
1844	mfastbinptr fastbinsY[NFASTBINS];
1845
1846	/ Base of the topmost chunk -- not otherwise kept in a bin /
1847	mchunkptr top;
1848
1849	/ The remainder from the most recent split of a small request /
1850	mchunkptr last_remainder;
1851
1852	/ Normal bins packed as described above /
1853	mchunkptr bins[NBINS * `2` - `2`];
1854
1855	/ Bitmap of bins /
1856	unsigned int binmap[BINMAPSIZE];
1857
1858	/ Linked list /
1859	struct malloc_state *next;
1860
1861	/ Linked list for free arenas. Access to this field is serialized*
1862	by free_list_lock in arena.c. /*
1863	struct malloc_state *next_free;
1864
1865	/ Number of threads attached to this arena. 0 if the arena is on*
1866	the free list. Access to this field is serialized by
1867	free_list_lock in arena.c. /*
1868	INTERNAL_SIZE_T attached_threads;
1869
1870	/ Memory allocated from the system in this arena. /
1871	INTERNAL_SIZE_T system_mem;
1872	INTERNAL_SIZE_T max_system_mem;
1873	};
1874
1875	struct malloc_par
1876	{
1877	/ Tunable parameters /
1878	unsigned long trim_threshold;
1879	INTERNAL_SIZE_T top_pad;
1880	INTERNAL_SIZE_T mmap_threshold;
1881	INTERNAL_SIZE_T arena_test;
1882	INTERNAL_SIZE_T arena_max;
1883
1884	/ Memory map support /
1885	int n_mmaps;
1886	int n_mmaps_max;
1887	int max_n_mmaps;
1888	/ the mmap_threshold is dynamic, until the user sets*
1889	it manually, at which point we need to disable any
1890	dynamic behavior. /*
1891	int no_dyn_threshold;
1892
1893	/ Statistics /
1894	INTERNAL_SIZE_T mmapped_mem;
1895	INTERNAL_SIZE_T max_mmapped_mem;
1896
1897	/ First address handed out by MORECORE/sbrk. /
1898	char *sbrk_base;
1899
1900	#if USE_TCACHE
1901	/ Maximum number of buckets to use. /
1902	size_t tcache_bins;
1903	size_t tcache_max_bytes;
1904	/ Maximum number of chunks in each bucket. /
1905	size_t tcache_count;
1906	/ Maximum number of chunks to remove from the unsorted list, which*
1907	aren't used to prefill the cache. /*
1908	size_t tcache_unsorted_limit;
1909	#endif
1910	};
1911
1912	/ There are several instances of this struct ("arenas") in this*
1913	malloc. If you are adapting this malloc in a way that does NOT use
1914	a static or mmapped malloc_state, you MUST explicitly zero-fill it
1915	before using. This malloc relies on the property that malloc_state
1916	is initialized to all zeroes (as is true of C statics). /*
1917
1918	static struct malloc_state main_arena =
1919	{
1920	.mutex = _LIBC_LOCK_INITIALIZER,
1921	.next = &main_arena,
1922	.attached_threads = `1`
1923	};
1924
1925	/ There is only one instance of the malloc parameters. /
1926
1927	static struct malloc_par mp_ =
1928	{
1929	.top_pad = DEFAULT_TOP_PAD,
1930	.n_mmaps_max = DEFAULT_MMAP_MAX,
1931	.mmap_threshold = DEFAULT_MMAP_THRESHOLD,
1932	.trim_threshold = DEFAULT_TRIM_THRESHOLD,
1933	#define NARENAS_FROM_NCORES(n) ((n) * (sizeof (long) == 4 ? 2 : 8))
1934	.arena_test = NARENAS_FROM_NCORES (`1`)
1935	#if USE_TCACHE
1936	,
1937	.tcache_count = TCACHE_FILL_COUNT,
1938	.tcache_bins = TCACHE_MAX_BINS,
1939	.tcache_max_bytes = tidx2usize (TCACHE_MAX_BINS-`1`),
1940	.tcache_unsorted_limit = `0` / No limit. /
1941	#endif
1942	};
1943
1944	/*
1945	Initialize a malloc_state struct.
1946
1947	This is called from ptmalloc_init () or from _int_new_arena ()
1948	when creating a new arena.
1949	*/
1950
1951	static void
1952	malloc_init_state (mstate av)
1953	{
1954	int i;
1955	mbinptr bin;
1956
1957	/ Establish circular links for normal bins /
1958	for (i = `1`; i < NBINS; ++i)
1959	{
1960	bin = bin_at (av, i);
1961	bin->fd = bin->bk = bin;
1962	}
1963
1964	#if MORECORE_CONTIGUOUS
1965	if (av != &main_arena)
1966	#endif
1967	set_noncontiguous (av);
1968	if (av == &main_arena)
1969	set_max_fast (DEFAULT_MXFAST);
1970	atomic_store_relaxed (&av->have_fastchunks, false);
1971
1972	av->top = initial_top (av);
1973	}
1974
1975	/*
1976	Other internal utilities operating on mstates
1977	*/
1978
1979	static void *sysmalloc (INTERNAL_SIZE_T, mstate);
1980	static int systrim (size_t, mstate);
1981	static void malloc_consolidate (mstate);
1982
1983
1984	/ -------------- Early definitions for debugging hooks ---------------- /
1985
1986	/ This function is called from the arena shutdown hook, to free the*
1987	thread cache (if it exists). /*
1988	static void tcache_thread_shutdown (void);
1989
1990	/ ------------------ Testing support ----------------------------------/
1991
1992	static int perturb_byte;
1993
1994	static void
1995	alloc_perturb (char *p, size_t n)
1996	{
1997	if (__glibc_unlikely (perturb_byte))
1998	memset (p, perturb_byte ^ `0xff`, n);
1999	}
2000
2001	static void
2002	free_perturb (char *p, size_t n)
2003	{
2004	if (__glibc_unlikely (perturb_byte))
2005	memset (p, perturb_byte, n);
2006	}
2007
2008
2009
2010	#include <stap-probe.h>
2011
2012	/ ------------------- Support for multiple arenas -------------------- /
2013	#include "arena.c"
2014
2015	/*
2016	Debugging support
2017
2018	These routines make a number of assertions about the states
2019	of data structures that should be true at all times. If any
2020	are not true, it's very likely that a user program has somehow
2021	trashed memory. (It's also possible that there is a coding error
2022	in malloc. In which case, please report it!)
2023	*/
2024
2025	#if !MALLOC_DEBUG
2026
2027	# define check_chunk(A, P)
2028	# define check_free_chunk(A, P)
2029	# define check_inuse_chunk(A, P)
2030	# define check_remalloced_chunk(A, P, N)
2031	# define check_malloced_chunk(A, P, N)
2032	# define check_malloc_state(A)
2033
2034	#else
2035
2036	# define check_chunk(A, P) do_check_chunk (A, P)
2037	# define check_free_chunk(A, P) do_check_free_chunk (A, P)
2038	# define check_inuse_chunk(A, P) do_check_inuse_chunk (A, P)
2039	# define check_remalloced_chunk(A, P, N) do_check_remalloced_chunk (A, P, N)
2040	# define check_malloced_chunk(A, P, N) do_check_malloced_chunk (A, P, N)
2041	# define check_malloc_state(A) do_check_malloc_state (A)
2042
2043	/*
2044	Properties of all chunks
2045	*/
2046
2047	static void
2048	do_check_chunk (mstate av, mchunkptr p)
2049	{
2050	unsigned long sz = chunksize (p);
2051	/ min and max possible addresses assuming contiguous allocation /
2052	char max_address = (char* *) (av->top) + chunksize (av->top);
2053	char *min_address = max_address - av->system_mem;
2054
2055	if (!chunk_is_mmapped (p))
2056	{
2057	/ Has legal address ... /
2058	if (p != av->top)
2059	{
2060	if (contiguous (av))
2061	{
2062	assert (((char *) p) >= min_address);
2063	assert (((char ) p + sz) <= ((char* *) (av->top)));
2064	}
2065	}
2066	else
2067	{
2068	/ top size is always at least MINSIZE /
2069	assert ((unsigned long) (sz) >= MINSIZE);
2070	/ top predecessor always marked inuse /
2071	assert (prev_inuse (p));
2072	}
2073	}
2074	else
2075	{
2076	/ address is outside main heap /
2077	if (contiguous (av) && av->top != initial_top (av))
2078	{
2079	assert (((char ) p) < min_address \|\| ((char* *) p) >= max_address);
2080	}
2081	/ chunk is page-aligned /
2082	assert (((prev_size (p) + sz) & (GLRO (dl_pagesize) - `1`)) == `0`);
2083	/ mem is aligned /
2084	assert (aligned_OK (chunk2mem (p)));
2085	}
2086	}
2087
2088	/*
2089	Properties of free chunks
2090	*/
2091
2092	static void
2093	do_check_free_chunk (mstate av, mchunkptr p)
2094	{
2095	INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE \| NON_MAIN_ARENA);
2096	mchunkptr next = chunk_at_offset (p, sz);
2097
2098	do_check_chunk (av, p);
2099
2100	/ Chunk must claim to be free ... /
2101	assert (!inuse (p));
2102	assert (!chunk_is_mmapped (p));
2103
2104	/ Unless a special marker, must have OK fields /
2105	if ((unsigned long) (sz) >= MINSIZE)
2106	{
2107	assert ((sz & MALLOC_ALIGN_MASK) == `0`);
2108	assert (aligned_OK (chunk2mem (p)));
2109	/ ... matching footer field /
2110	assert (prev_size (next_chunk (p)) == sz);
2111	/ ... and is fully consolidated /
2112	assert (prev_inuse (p));
2113	assert (next == av->top \|\| inuse (next));
2114
2115	/ ... and has minimally sane links /
2116	assert (p->fd->bk == p);
2117	assert (p->bk->fd == p);
2118	}
2119	else / markers are always of size SIZE_SZ /
2120	assert (sz == SIZE_SZ);
2121	}
2122
2123	/*
2124	Properties of inuse chunks
2125	*/
2126
2127	static void
2128	do_check_inuse_chunk (mstate av, mchunkptr p)
2129	{
2130	mchunkptr next;
2131
2132	do_check_chunk (av, p);
2133
2134	if (chunk_is_mmapped (p))
2135	return; / mmapped chunks have no next/prev /
2136
2137	/ Check whether it claims to be in use ... /
2138	assert (inuse (p));
2139
2140	next = next_chunk (p);
2141
2142	/ ... and is surrounded by OK chunks.*
2143	Since more things can be checked with free chunks than inuse ones,
2144	if an inuse chunk borders them and debug is on, it's worth doing them.
2145	*/
2146	if (!prev_inuse (p))
2147	{
2148	/ Note that we cannot even look at prev unless it is not inuse /
2149	mchunkptr prv = prev_chunk (p);
2150	assert (next_chunk (prv) == p);
2151	do_check_free_chunk (av, prv);
2152	}
2153
2154	if (next == av->top)
2155	{
2156	assert (prev_inuse (next));
2157	assert (chunksize (next) >= MINSIZE);
2158	}
2159	else if (!inuse (next))
2160	do_check_free_chunk (av, next);
2161	}
2162
2163	/*
2164	Properties of chunks recycled from fastbins
2165	*/
2166
2167	static void
2168	do_check_remalloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s)
2169	{
2170	INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE \| NON_MAIN_ARENA);
2171
2172	if (!chunk_is_mmapped (p))
2173	{
2174	assert (av == arena_for_chunk (p));
2175	if (chunk_main_arena (p))
2176	assert (av == &main_arena);
2177	else
2178	assert (av != &main_arena);
2179	}
2180
2181	do_check_inuse_chunk (av, p);
2182
2183	/ Legal size ... /
2184	assert ((sz & MALLOC_ALIGN_MASK) == `0`);
2185	assert ((unsigned long) (sz) >= MINSIZE);
2186	/ ... and alignment /
2187	assert (aligned_OK (chunk2mem (p)));
2188	/ chunk is less than MINSIZE more than request /
2189	assert ((long) (sz) - (long) (s) >= `0`);
2190	assert ((long) (sz) - (long) (s + MINSIZE) < `0`);
2191	}
2192
2193	/*
2194	Properties of nonrecycled chunks at the point they are malloced
2195	*/
2196
2197	static void
2198	do_check_malloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s)
2199	{
2200	/ same as recycled case ... /
2201	do_check_remalloced_chunk (av, p, s);
2202
2203	/*
2204	... plus, must obey implementation invariant that prev_inuse is
2205	always true of any allocated chunk; i.e., that each allocated
2206	chunk borders either a previously allocated and still in-use
2207	chunk, or the base of its memory arena. This is ensured
2208	by making all allocations from the `lowest' part of any found
2209	chunk. This does not necessarily hold however for chunks
2210	recycled via fastbins.
2211	*/
2212
2213	assert (prev_inuse (p));
2214	}
2215
2216
2217	/*
2218	Properties of malloc_state.
2219
2220	This may be useful for debugging malloc, as well as detecting user
2221	programmer errors that somehow write into malloc_state.
2222
2223	If you are extending or experimenting with this malloc, you can
2224	probably figure out how to hack this routine to print out or
2225	display chunk addresses, sizes, bins, and other instrumentation.
2226	*/
2227
2228	static void
2229	do_check_malloc_state (mstate av)
2230	{
2231	int i;
2232	mchunkptr p;
2233	mchunkptr q;
2234	mbinptr b;
2235	unsigned int idx;
2236	INTERNAL_SIZE_T size;
2237	unsigned long total = `0`;
2238	int max_fast_bin;
2239
2240	/ internal size_t must be no wider than pointer type /
2241	assert (sizeof (INTERNAL_SIZE_T) <= sizeof (char *));
2242
2243	/ alignment is a power of 2 /
2244	assert ((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT - `1`)) == `0`);
2245
2246	/ Check the arena is initialized. /
2247	assert (av->top != `0`);
2248
2249	/ No memory has been allocated yet, so doing more tests is not possible. /
2250	if (av->top == initial_top (av))
2251	return;
2252
2253	/ pagesize is a power of 2 /
2254	assert (powerof2(GLRO (dl_pagesize)));
2255
2256	/ A contiguous main_arena is consistent with sbrk_base. /
2257	if (av == &main_arena && contiguous (av))
2258	assert ((char *) mp_.sbrk_base + av->system_mem ==
2259	(char *) av->top + chunksize (av->top));
2260
2261	/ properties of fastbins /
2262
2263	/ max_fast is in allowed range /
2264	assert ((get_max_fast () & ~`1`) <= request2size (MAX_FAST_SIZE));
2265
2266	max_fast_bin = fastbin_index (get_max_fast ());
2267
2268	for (i = `0`; i < NFASTBINS; ++i)
2269	{
2270	p = fastbin (av, i);
2271
2272	/ The following test can only be performed for the main arena.*
2273	While mallopt calls malloc_consolidate to get rid of all fast
2274	bins (especially those larger than the new maximum) this does
2275	only happen for the main arena. Trying to do this for any
2276	other arena would mean those arenas have to be locked and
2277	malloc_consolidate be called for them. This is excessive. And
2278	even if this is acceptable to somebody it still cannot solve
2279	the problem completely since if the arena is locked a
2280	concurrent malloc call might create a new arena which then
2281	could use the newly invalid fast bins. /*
2282
2283	/ all bins past max_fast are empty /
2284	if (av == &main_arena && i > max_fast_bin)
2285	assert (p == `0`);
2286
2287	while (p != `0`)
2288	{
2289	if (__glibc_unlikely (misaligned_chunk (p)))
2290	malloc_printerr ("do_check_malloc_state(): "
2291	"unaligned fastbin chunk detected");
2292	/ each chunk claims to be inuse /
2293	do_check_inuse_chunk (av, p);
2294	total += chunksize (p);
2295	/ chunk belongs in this bin /
2296	assert (fastbin_index (chunksize (p)) == i);
2297	p = REVEAL_PTR (p->fd);
2298	}
2299	}
2300
2301	/ check normal bins /
2302	for (i = `1`; i < NBINS; ++i)
2303	{
2304	b = bin_at (av, i);
2305
2306	/ binmap is accurate (except for bin 1 == unsorted_chunks) /
2307	if (i >= `2`)
2308	{
2309	unsigned int binbit = get_binmap (av, i);
2310	int empty = last (b) == b;
2311	if (!binbit)
2312	assert (empty);
2313	else if (!empty)
2314	assert (binbit);
2315	}
2316
2317	for (p = last (b); p != b; p = p->bk)
2318	{
2319	/ each chunk claims to be free /
2320	do_check_free_chunk (av, p);
2321	size = chunksize (p);
2322	total += size;
2323	if (i >= `2`)
2324	{
2325	/ chunk belongs in bin /
2326	idx = bin_index (size);
2327	assert (idx == i);
2328	/ lists are sorted /
2329	assert (p->bk == b \|\|
2330	(unsigned long) chunksize (p->bk) >= (unsigned long) chunksize (p));
2331
2332	if (!in_smallbin_range (size))
2333	{
2334	if (p->fd_nextsize != NULL)
2335	{
2336	if (p->fd_nextsize == p)
2337	assert (p->bk_nextsize == p);
2338	else
2339	{
2340	if (p->fd_nextsize == first (b))
2341	assert (chunksize (p) < chunksize (p->fd_nextsize));
2342	else
2343	assert (chunksize (p) > chunksize (p->fd_nextsize));
2344
2345	if (p == first (b))
2346	assert (chunksize (p) > chunksize (p->bk_nextsize));
2347	else
2348	assert (chunksize (p) < chunksize (p->bk_nextsize));
2349	}
2350	}
2351	else
2352	assert (p->bk_nextsize == NULL);
2353	}
2354	}
2355	else if (!in_smallbin_range (size))
2356	assert (p->fd_nextsize == NULL && p->bk_nextsize == NULL);
2357	/ chunk is followed by a legal chain of inuse chunks /
2358	for (q = next_chunk (p);
2359	(q != av->top && inuse (q) &&
2360	(unsigned long) (chunksize (q)) >= MINSIZE);
2361	q = next_chunk (q))
2362	do_check_inuse_chunk (av, q);
2363	}
2364	}
2365
2366	/ top chunk is OK /
2367	check_chunk (av, av->top);
2368	}
2369	#endif
2370
2371
2372	/ ----------------- Support for debugging hooks -------------------- /
2373	#if IS_IN (libc)
2374	#include "hooks.c"
2375	#endif
2376
2377
2378	/ ----------- Routines dealing with system allocation -------------- /
2379
2380	/*
2381	sysmalloc handles malloc cases requiring more memory from the system.
2382	On entry, it is assumed that av->top does not have enough
2383	space to service request for nb bytes, thus requiring that av->top
2384	be extended or replaced.
2385	*/
2386
2387	static void *
2388	sysmalloc (INTERNAL_SIZE_T nb, mstate av)
2389	{
2390	mchunkptr old_top; / incoming value of av->top /
2391	INTERNAL_SIZE_T old_size; / its size /
2392	char old_end; /* its end address /
2393
2394	long size; / arg to first MORECORE or mmap call /
2395	char brk; /* return value from MORECORE /
2396
2397	long correction; / arg to 2nd MORECORE call /
2398	char snd_brk; /* 2nd return val /
2399
2400	INTERNAL_SIZE_T front_misalign; / unusable bytes at front of new space /
2401	INTERNAL_SIZE_T end_misalign; / partial page left at end of new space /
2402	char aligned_brk; /* aligned offset into brk /
2403
2404	mchunkptr p; / the allocated/returned chunk /
2405	mchunkptr remainder; / remainder from allocation /
2406	unsigned long remainder_size; / its size /
2407
2408
2409	size_t pagesize = GLRO (dl_pagesize);
2410	bool tried_mmap = false;
2411
2412
2413	/*
2414	If have mmap, and the request size meets the mmap threshold, and
2415	the system supports mmap, and there are few enough currently
2416	allocated mmapped regions, try to directly map this request
2417	rather than expanding top.
2418	*/
2419
2420	if (av == NULL
2421	\|\| ((unsigned long) (nb) >= (unsigned long) (mp_.mmap_threshold)
2422	&& (mp_.n_mmaps < mp_.n_mmaps_max)))
2423	{
2424	char mm; /* return value from mmap call/
2425
2426	try_mmap:
2427	/*
2428	Round up size to nearest page. For mmapped chunks, the overhead
2429	is one SIZE_SZ unit larger than for normal chunks, because there
2430	is no following chunk whose prev_size field could be used.
2431
2432	See the front_misalign handling below, for glibc there is no
2433	need for further alignments unless we have have high alignment.
2434	*/
2435	if (MALLOC_ALIGNMENT == CHUNK_HDR_SZ)
2436	size = ALIGN_UP (nb + SIZE_SZ, pagesize);
2437	else
2438	size = ALIGN_UP (nb + SIZE_SZ + MALLOC_ALIGN_MASK, pagesize);
2439	tried_mmap = true;
2440
2441	/ Don't try if size wraps around 0 /
2442	if ((unsigned long) (size) > (unsigned long) (nb))
2443	{
2444	mm = (char *) (MMAP (`0`, size,
2445	mtag_mmap_flags \| PROT_READ \| PROT_WRITE, `0`));
2446
2447	if (mm != MAP_FAILED)
2448	{
2449	/*
2450	The offset to the start of the mmapped region is stored
2451	in the prev_size field of the chunk. This allows us to adjust
2452	returned start address to meet alignment requirements here
2453	and in memalign(), and still be able to compute proper
2454	address argument for later munmap in free() and realloc().
2455	*/
2456
2457	if (MALLOC_ALIGNMENT == CHUNK_HDR_SZ)
2458	{
2459	/ For glibc, chunk2mem increases the address by*
2460	CHUNK_HDR_SZ and MALLOC_ALIGN_MASK is
2461	CHUNK_HDR_SZ-1. Each mmap'ed area is page
2462	aligned and therefore definitely
2463	MALLOC_ALIGN_MASK-aligned. /*
2464	assert (((INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK) == `0`);
2465	front_misalign = `0`;
2466	}
2467	else
2468	front_misalign = (INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK;
2469	if (front_misalign > `0`)
2470	{
2471	correction = MALLOC_ALIGNMENT - front_misalign;
2472	p = (mchunkptr) (mm + correction);
2473	set_prev_size (p, correction);
2474	set_head (p, (size - correction) \| IS_MMAPPED);
2475	}
2476	else
2477	{
2478	p = (mchunkptr) mm;
2479	set_prev_size (p, `0`);
2480	set_head (p, size \| IS_MMAPPED);
2481	}
2482
2483	/ update statistics /
2484
2485	int new = atomic_exchange_and_add (&mp_.n_mmaps, `1`) + `1`;
2486	atomic_max (&mp_.max_n_mmaps, new);
2487
2488	unsigned long sum;
2489	sum = atomic_exchange_and_add (&mp_.mmapped_mem, size) + size;
2490	atomic_max (&mp_.max_mmapped_mem, sum);
2491
2492	check_chunk (av, p);
2493
2494	return chunk2mem (p);
2495	}
2496	}
2497	}
2498
2499	/ There are no usable arenas and mmap also failed. /
2500	if (av == NULL)
2501	return `0`;
2502
2503	/ Record incoming configuration of top /
2504
2505	old_top = av->top;
2506	old_size = chunksize (old_top);
2507	old_end = (char *) (chunk_at_offset (old_top, old_size));
2508
2509	brk = snd_brk = (char *) (MORECORE_FAILURE);
2510
2511	/*
2512	If not the first time through, we require old_size to be
2513	at least MINSIZE and to have prev_inuse set.
2514	*/
2515
2516	assert ((old_top == initial_top (av) && old_size == `0`) \|\|
2517	((unsigned long) (old_size) >= MINSIZE &&
2518	prev_inuse (old_top) &&
2519	((unsigned long) old_end & (pagesize - `1`)) == `0`));
2520
2521	/ Precondition: not enough current space to satisfy nb request /
2522	assert ((unsigned long) (old_size) < (unsigned long) (nb + MINSIZE));
2523
2524
2525	if (av != &main_arena)
2526	{
2527	heap_info old_heap, heap;
2528	size_t old_heap_size;
2529
2530	/ First try to extend the current heap. /
2531	old_heap = heap_for_ptr (old_top);
2532	old_heap_size = old_heap->size;
2533	if ((long) (MINSIZE + nb - old_size) > `0`
2534	&& grow_heap (old_heap, MINSIZE + nb - old_size) == `0`)
2535	{
2536	av->system_mem += old_heap->size - old_heap_size;
2537	set_head (old_top, (((char ) old_heap + old_heap->size) - (char* *) old_top)
2538	\| PREV_INUSE);
2539	}
2540	else if ((heap = new_heap (nb + (MINSIZE + sizeof (*heap)), mp_.top_pad)))
2541	{
2542	/ Use a newly allocated heap. /
2543	heap->ar_ptr = av;
2544	heap->prev = old_heap;
2545	av->system_mem += heap->size;
2546	/ Set up the new top. /
2547	top (av) = chunk_at_offset (heap, sizeof (*heap));
2548	set_head (top (av), (heap->size - sizeof (*heap)) \| PREV_INUSE);
2549
2550	/ Setup fencepost and free the old top chunk with a multiple of*
2551	MALLOC_ALIGNMENT in size. /*
2552	/ The fencepost takes at least MINSIZE bytes, because it might*
2553	become the top chunk again later. Note that a footer is set
2554	up, too, although the chunk is marked in use. /*
2555	old_size = (old_size - MINSIZE) & ~MALLOC_ALIGN_MASK;
2556	set_head (chunk_at_offset (old_top, old_size + CHUNK_HDR_SZ),
2557	`0` \| PREV_INUSE);
2558	if (old_size >= MINSIZE)
2559	{
2560	set_head (chunk_at_offset (old_top, old_size),
2561	CHUNK_HDR_SZ \| PREV_INUSE);
2562	set_foot (chunk_at_offset (old_top, old_size), CHUNK_HDR_SZ);
2563	set_head (old_top, old_size \| PREV_INUSE \| NON_MAIN_ARENA);
2564	_int_free (av, old_top, `1`);
2565	}
2566	else
2567	{
2568	set_head (old_top, (old_size + CHUNK_HDR_SZ) \| PREV_INUSE);
2569	set_foot (old_top, (old_size + CHUNK_HDR_SZ));
2570	}
2571	}
2572	else if (!tried_mmap)
2573	/ We can at least try to use to mmap memory. /
2574	goto try_mmap;
2575	}
2576	else / av == main_arena /
2577
2578
2579	{ / Request enough space for nb + pad + overhead /
2580	size = nb + mp_.top_pad + MINSIZE;
2581
2582	/*
2583	If contiguous, we can subtract out existing space that we hope to
2584	combine with new space. We add it back later only if
2585	we don't actually get contiguous space.
2586	*/
2587
2588	if (contiguous (av))
2589	size -= old_size;
2590
2591	/*
2592	Round to a multiple of page size.
2593	If MORECORE is not contiguous, this ensures that we only call it
2594	with whole-page arguments. And if MORECORE is contiguous and
2595	this is not first time through, this preserves page-alignment of
2596	previous calls. Otherwise, we correct to page-align below.
2597	*/
2598
2599	size = ALIGN_UP (size, pagesize);
2600
2601	/*
2602	Don't try to call MORECORE if argument is so big as to appear
2603	negative. Note that since mmap takes size_t arg, it may succeed
2604	below even if we cannot call MORECORE.
2605	*/
2606
2607	if (size > `0`)
2608	{
2609	brk = (char *) (MORECORE (size));
2610	LIBC_PROBE (memory_sbrk_more, `2`, brk, size);
2611	}
2612
2613	if (brk == (char *) (MORECORE_FAILURE))
2614	{
2615	/*
2616	If have mmap, try using it as a backup when MORECORE fails or
2617	cannot be used. This is worth doing on systems that have "holes" in
2618	address space, so sbrk cannot extend to give contiguous space, but
2619	space is available elsewhere. Note that we ignore mmap max count
2620	and threshold limits, since the space will not be used as a
2621	segregated mmap region.
2622	*/
2623
2624	/ Cannot merge with old top, so add its size back in /
2625	if (contiguous (av))
2626	size = ALIGN_UP (size + old_size, pagesize);
2627
2628	/ If we are relying on mmap as backup, then use larger units /
2629	if ((unsigned long) (size) < (unsigned long) (MMAP_AS_MORECORE_SIZE))
2630	size = MMAP_AS_MORECORE_SIZE;
2631
2632	/ Don't try if size wraps around 0 /
2633	if ((unsigned long) (size) > (unsigned long) (nb))
2634	{
2635	char mbrk = (char* *) (MMAP (`0`, size,
2636	mtag_mmap_flags \| PROT_READ \| PROT_WRITE,
2637	`0`));
2638
2639	if (mbrk != MAP_FAILED)
2640	{
2641	/ We do not need, and cannot use, another sbrk call to find end /
2642	brk = mbrk;
2643	snd_brk = brk + size;
2644
2645	/*
2646	Record that we no longer have a contiguous sbrk region.
2647	After the first time mmap is used as backup, we do not
2648	ever rely on contiguous space since this could incorrectly
2649	bridge regions.
2650	*/
2651	set_noncontiguous (av);
2652	}
2653	}
2654	}
2655
2656	if (brk != (char *) (MORECORE_FAILURE))
2657	{
2658	if (mp_.sbrk_base == `0`)
2659	mp_.sbrk_base = brk;
2660	av->system_mem += size;
2661
2662	/*
2663	If MORECORE extends previous space, we can likewise extend top size.
2664	*/
2665
2666	if (brk == old_end && snd_brk == (char *) (MORECORE_FAILURE))
2667	set_head (old_top, (size + old_size) \| PREV_INUSE);
2668
2669	else if (contiguous (av) && old_size && brk < old_end)
2670	/ Oops! Someone else killed our space.. Can't touch anything. /
2671	malloc_printerr ("break adjusted to free malloc space");
2672
2673	/*
2674	Otherwise, make adjustments:
2675
2676	* If the first time through or noncontiguous, we need to call sbrk
2677	just to find out where the end of memory lies.
2678
2679	* We need to ensure that all returned chunks from malloc will meet
2680	MALLOC_ALIGNMENT
2681
2682	* If there was an intervening foreign sbrk, we need to adjust sbrk
2683	request size to account for fact that we will not be able to
2684	combine new space with existing space in old_top.
2685
2686	* Almost all systems internally allocate whole pages at a time, in
2687	which case we might as well use the whole last page of request.
2688	So we allocate enough more memory to hit a page boundary now,
2689	which in turn causes future contiguous calls to page-align.
2690	*/
2691
2692	else
2693	{
2694	front_misalign = `0`;
2695	end_misalign = `0`;
2696	correction = `0`;
2697	aligned_brk = brk;
2698
2699	/ handle contiguous cases /
2700	if (contiguous (av))
2701	{
2702	/ Count foreign sbrk as system_mem. /
2703	if (old_size)
2704	av->system_mem += brk - old_end;
2705
2706	/ Guarantee alignment of first new chunk made from this space /
2707
2708	front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK;
2709	if (front_misalign > `0`)
2710	{
2711	/*
2712	Skip over some bytes to arrive at an aligned position.
2713	We don't need to specially mark these wasted front bytes.
2714	They will never be accessed anyway because
2715	prev_inuse of av->top (and any chunk created from its start)
2716	is always true after initialization.
2717	*/
2718
2719	correction = MALLOC_ALIGNMENT - front_misalign;
2720	aligned_brk += correction;
2721	}
2722
2723	/*
2724	If this isn't adjacent to existing space, then we will not
2725	be able to merge with old_top space, so must add to 2nd request.
2726	*/
2727
2728	correction += old_size;
2729
2730	/ Extend the end address to hit a page boundary /
2731	end_misalign = (INTERNAL_SIZE_T) (brk + size + correction);
2732	correction += (ALIGN_UP (end_misalign, pagesize)) - end_misalign;
2733
2734	assert (correction >= `0`);
2735	snd_brk = (char *) (MORECORE (correction));
2736
2737	/*
2738	If can't allocate correction, try to at least find out current
2739	brk. It might be enough to proceed without failing.
2740
2741	Note that if second sbrk did NOT fail, we assume that space
2742	is contiguous with first sbrk. This is a safe assumption unless
2743	program is multithreaded but doesn't use locks and a foreign sbrk
2744	occurred between our first and second calls.
2745	*/
2746
2747	if (snd_brk == (char *) (MORECORE_FAILURE))
2748	{
2749	correction = `0`;
2750	snd_brk = (char *) (MORECORE (`0`));
2751	}
2752	}
2753
2754	/ handle non-contiguous cases /
2755	else
2756	{
2757	if (MALLOC_ALIGNMENT == CHUNK_HDR_SZ)
2758	/ MORECORE/mmap must correctly align /
2759	assert (((unsigned long) chunk2mem (brk) & MALLOC_ALIGN_MASK) == `0`);
2760	else
2761	{
2762	front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK;
2763	if (front_misalign > `0`)
2764	{
2765	/*
2766	Skip over some bytes to arrive at an aligned position.
2767	We don't need to specially mark these wasted front bytes.
2768	They will never be accessed anyway because
2769	prev_inuse of av->top (and any chunk created from its start)
2770	is always true after initialization.
2771	*/
2772
2773	aligned_brk += MALLOC_ALIGNMENT - front_misalign;
2774	}
2775	}
2776
2777	/ Find out current end of memory /
2778	if (snd_brk == (char *) (MORECORE_FAILURE))
2779	{
2780	snd_brk = (char *) (MORECORE (`0`));
2781	}
2782	}
2783
2784	/ Adjust top based on results of second sbrk /
2785	if (snd_brk != (char *) (MORECORE_FAILURE))
2786	{
2787	av->top = (mchunkptr) aligned_brk;
2788	set_head (av->top, (snd_brk - aligned_brk + correction) \| PREV_INUSE);
2789	av->system_mem += correction;
2790
2791	/*
2792	If not the first time through, we either have a
2793	gap due to foreign sbrk or a non-contiguous region. Insert a
2794	double fencepost at old_top to prevent consolidation with space
2795	we don't own. These fenceposts are artificial chunks that are
2796	marked as inuse and are in any case too small to use. We need
2797	two to make sizes and alignments work out.
2798	*/
2799
2800	if (old_size != `0`)
2801	{
2802	/*
2803	Shrink old_top to insert fenceposts, keeping size a
2804	multiple of MALLOC_ALIGNMENT. We know there is at least
2805	enough space in old_top to do this.
2806	*/
2807	old_size = (old_size - `2` * CHUNK_HDR_SZ) & ~MALLOC_ALIGN_MASK;
2808	set_head (old_top, old_size \| PREV_INUSE);
2809
2810	/*
2811	Note that the following assignments completely overwrite
2812	old_top when old_size was previously MINSIZE. This is
2813	intentional. We need the fencepost, even if old_top otherwise gets
2814	lost.
2815	*/
2816	set_head (chunk_at_offset (old_top, old_size),
2817	CHUNK_HDR_SZ \| PREV_INUSE);
2818	set_head (chunk_at_offset (old_top,
2819	old_size + CHUNK_HDR_SZ),
2820	CHUNK_HDR_SZ \| PREV_INUSE);
2821
2822	/ If possible, release the rest. /
2823	if (old_size >= MINSIZE)
2824	{
2825	_int_free (av, old_top, `1`);
2826	}
2827	}
2828	}
2829	}
2830	}
2831	} / if (av != &main_arena) /
2832
2833	if ((unsigned long) av->system_mem > (unsigned long) (av->max_system_mem))
2834	av->max_system_mem = av->system_mem;
2835	check_malloc_state (av);
2836
2837	/ finally, do the allocation /
2838	p = av->top;
2839	size = chunksize (p);
2840
2841	/ check that one of the above allocation paths succeeded /
2842	if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE))
2843	{
2844	remainder_size = size - nb;
2845	remainder = chunk_at_offset (p, nb);
2846	av->top = remainder;
2847	set_head (p, nb \| PREV_INUSE \| (av != &main_arena ? NON_MAIN_ARENA : `0`));
2848	set_head (remainder, remainder_size \| PREV_INUSE);
2849	check_malloced_chunk (av, p, nb);
2850	return chunk2mem (p);
2851	}
2852
2853	/ catch all failure paths /
2854	__set_errno (ENOMEM);
2855	return `0`;
2856	}
2857
2858
2859	/*
2860	systrim is an inverse of sorts to sysmalloc. It gives memory back
2861	to the system (via negative arguments to sbrk) if there is unused
2862	memory at the `high' end of the malloc pool. It is called
2863	automatically by free() when top space exceeds the trim
2864	threshold. It is also called by the public malloc_trim routine. It
2865	returns 1 if it actually released any memory, else 0.
2866	*/
2867
2868	static int
2869	systrim (size_t pad, mstate av)
2870	{
2871	long top_size; / Amount of top-most memory /
2872	long extra; / Amount to release /
2873	long released; / Amount actually released /
2874	char current_brk; /* address returned by pre-check sbrk call /
2875	char new_brk; /* address returned by post-check sbrk call /
2876	size_t pagesize;
2877	long top_area;
2878
2879	pagesize = GLRO (dl_pagesize);
2880	top_size = chunksize (av->top);
2881
2882	top_area = top_size - MINSIZE - `1`;
2883	if (top_area <= pad)
2884	return `0`;
2885
2886	/ Release in pagesize units and round down to the nearest page. /
2887	extra = ALIGN_DOWN(top_area - pad, pagesize);
2888
2889	if (extra == `0`)
2890	return `0`;
2891
2892	/*
2893	Only proceed if end of memory is where we last set it.
2894	This avoids problems if there were foreign sbrk calls.
2895	*/
2896	current_brk = (char *) (MORECORE (`0`));
2897	if (current_brk == (char *) (av->top) + top_size)
2898	{
2899	/*
2900	Attempt to release memory. We ignore MORECORE return value,
2901	and instead call again to find out where new end of memory is.
2902	This avoids problems if first call releases less than we asked,
2903	of if failure somehow altered brk value. (We could still
2904	encounter problems if it altered brk in some very bad way,
2905	but the only thing we can do is adjust anyway, which will cause
2906	some downstream failure.)
2907	*/
2908
2909	MORECORE (-extra);
2910	new_brk = (char *) (MORECORE (`0`));
2911
2912	LIBC_PROBE (memory_sbrk_less, `2`, new_brk, extra);
2913
2914	if (new_brk != (char *) MORECORE_FAILURE)
2915	{
2916	released = (long) (current_brk - new_brk);
2917
2918	if (released != `0`)
2919	{
2920	/ Success. Adjust top. /
2921	av->system_mem -= released;
2922	set_head (av->top, (top_size - released) \| PREV_INUSE);
2923	check_malloc_state (av);
2924	return `1`;
2925	}
2926	}
2927	}
2928	return `0`;
2929	}
2930
2931	static void
2932	munmap_chunk (mchunkptr p)
2933	{
2934	size_t pagesize = GLRO (dl_pagesize);
2935	INTERNAL_SIZE_T size = chunksize (p);
2936
2937	assert (chunk_is_mmapped (p));
2938
2939	uintptr_t mem = (uintptr_t) chunk2mem (p);
2940	uintptr_t block = (uintptr_t) p - prev_size (p);
2941	size_t total_size = prev_size (p) + size;
2942	/ Unfortunately we have to do the compilers job by hand here. Normally*
2943	we would test BLOCK and TOTAL-SIZE separately for compliance with the
2944	page size. But gcc does not recognize the optimization possibility
2945	(in the moment at least) so we combine the two values into one before
2946	the bit test. /*
2947	if (__glibc_unlikely ((block \| total_size) & (pagesize - `1`)) != `0`
2948	\|\| __glibc_unlikely (!powerof2 (mem & (pagesize - `1`))))
2949	malloc_printerr ("munmap_chunk(): invalid pointer");
2950
2951	atomic_decrement (&mp_.n_mmaps);
2952	atomic_add (&mp_.mmapped_mem, -total_size);
2953
2954	/ If munmap failed the process virtual memory address space is in a*
2955	bad shape. Just leave the block hanging around, the process will
2956	terminate shortly anyway since not much can be done. /*
2957	__munmap ((char *) block, total_size);
2958	}
2959
2960	#if HAVE_MREMAP
2961
2962	static mchunkptr
2963	mremap_chunk (mchunkptr p, size_t new_size)
2964	{
2965	size_t pagesize = GLRO (dl_pagesize);
2966	INTERNAL_SIZE_T offset = prev_size (p);
2967	INTERNAL_SIZE_T size = chunksize (p);
2968	char *cp;
2969
2970	assert (chunk_is_mmapped (p));
2971
2972	uintptr_t block = (uintptr_t) p - offset;
2973	uintptr_t mem = (uintptr_t) chunk2mem(p);
2974	size_t total_size = offset + size;
2975	if (__glibc_unlikely ((block \| total_size) & (pagesize - `1`)) != `0`
2976	\|\| __glibc_unlikely (!powerof2 (mem & (pagesize - `1`))))
2977	malloc_printerr("mremap_chunk(): invalid pointer");
2978
2979	/ Note the extra SIZE_SZ overhead as in mmap_chunk(). /
2980	new_size = ALIGN_UP (new_size + offset + SIZE_SZ, pagesize);
2981
2982	/ No need to remap if the number of pages does not change. /
2983	if (total_size == new_size)
2984	return p;
2985
2986	cp = (char ) __mremap ((char* *) block, total_size, new_size,
2987	MREMAP_MAYMOVE);
2988
2989	if (cp == MAP_FAILED)
2990	return `0`;
2991
2992	p = (mchunkptr) (cp + offset);
2993
2994	assert (aligned_OK (chunk2mem (p)));
2995
2996	assert (prev_size (p) == offset);
2997	set_head (p, (new_size - offset) \| IS_MMAPPED);
2998
2999	INTERNAL_SIZE_T new;
3000	new = atomic_exchange_and_add (&mp_.mmapped_mem, new_size - size - offset)
3001	+ new_size - size - offset;
3002	atomic_max (&mp_.max_mmapped_mem, new);
3003	return p;
3004	}
3005	#endif /* HAVE_MREMAP */
3006
3007	/------------------------ Public wrappers. --------------------------------/
3008
3009	#if USE_TCACHE
3010
3011	/ We overlay this structure on the user-data portion of a chunk when*
3012	the chunk is stored in the per-thread cache. /*
3013	typedef struct tcache_entry
3014	{
3015	struct tcache_entry *next;
3016	/ This field exists to detect double frees. /
3017	uintptr_t key;
3018	} tcache_entry;
3019
3020	/ There is one of these for each thread, which contains the*
3021	per-thread cache (hence "tcache_perthread_struct"). Keeping
3022	overall size low is mildly important. Note that COUNTS and ENTRIES
3023	are redundant (we could have just counted the linked list each
3024	time), this is for performance reasons. /*
3025	typedef struct tcache_perthread_struct
3026	{
3027	uint16_t counts[TCACHE_MAX_BINS];
3028	tcache_entry *entries[TCACHE_MAX_BINS];
3029	} tcache_perthread_struct;
3030
3031	static __thread bool tcache_shutting_down = false;
3032	static __thread tcache_perthread_struct *tcache = NULL;
3033
3034	/ Process-wide key to try and catch a double-free in the same thread. /
3035	static uintptr_t tcache_key;
3036
3037	/ The value of tcache_key does not really have to be a cryptographically*
3038	secure random number. It only needs to be arbitrary enough so that it does
3039	not collide with values present in applications. If a collision does happen
3040	consistently enough, it could cause a degradation in performance since the
3041	entire list is checked to check if the block indeed has been freed the
3042	second time. The odds of this happening are exceedingly low though, about 1
3043	in 2^wordsize. There is probably a higher chance of the performance
3044	degradation being due to a double free where the first free happened in a
3045	different thread; that's a case this check does not cover. /*
3046	static void
3047	tcache_key_initialize (void)
3048	{
3049	if (__getrandom (&tcache_key, sizeof(tcache_key), GRND_NONBLOCK)
3050	!= sizeof (tcache_key))
3051	{
3052	tcache_key = random_bits ();
3053	#if __WORDSIZE == 64
3054	tcache_key = (tcache_key << `32`) \| random_bits ();
3055	#endif
3056	}
3057	}
3058
3059	/ Caller must ensure that we know tc_idx is valid and there's room*
3060	for more chunks. /*
3061	static __always_inline void
3062	tcache_put (mchunkptr chunk, size_t tc_idx)
3063	{
3064	tcache_entry e = (tcache_entry ) chunk2mem (chunk);
3065
3066	/ Mark this chunk as "in the tcache" so the test in _int_free will*
3067	detect a double free. /*
3068	e->key = tcache_key;
3069
3070	e->next = PROTECT_PTR (&e->next, tcache->entries[tc_idx]);
3071	tcache->entries[tc_idx] = e;
3072	++(tcache->counts[tc_idx]);
3073	}
3074
3075	/ Caller must ensure that we know tc_idx is valid and there's*
3076	available chunks to remove. /*
3077	static __always_inline void *
3078	tcache_get (size_t tc_idx)
3079	{
3080	tcache_entry *e = tcache->entries[tc_idx];
3081	if (__glibc_unlikely (!aligned_OK (e)))
3082	malloc_printerr ("malloc(): unaligned tcache chunk detected");
3083	tcache->entries[tc_idx] = REVEAL_PTR (e->next);
3084	--(tcache->counts[tc_idx]);
3085	e->key = `0`;
3086	return (void *) e;
3087	}
3088
3089	static void
3090	tcache_thread_shutdown (void)
3091	{
3092	int i;
3093	tcache_perthread_struct *tcache_tmp = tcache;
3094
3095	tcache_shutting_down = true;
3096
3097	if (!tcache)
3098	return;
3099
3100	/ Disable the tcache and prevent it from being reinitialized. /
3101	tcache = NULL;
3102
3103	/ Free all of the entries and the tcache itself back to the arena*
3104	heap for coalescing. /*
3105	for (i = `0`; i < TCACHE_MAX_BINS; ++i)
3106	{
3107	while (tcache_tmp->entries[i])
3108	{
3109	tcache_entry *e = tcache_tmp->entries[i];
3110	if (__glibc_unlikely (!aligned_OK (e)))
3111	malloc_printerr ("tcache_thread_shutdown(): "
3112	"unaligned tcache chunk detected");
3113	tcache_tmp->entries[i] = REVEAL_PTR (e->next);
3114	__libc_free (e);
3115	}
3116	}
3117
3118	__libc_free (tcache_tmp);
3119	}
3120
3121	static void
3122	tcache_init(void)
3123	{
3124	mstate ar_ptr;
3125	void *victim = `0`;
3126	const size_t bytes = sizeof (tcache_perthread_struct);
3127
3128	if (tcache_shutting_down)
3129	return;
3130
3131	arena_get (ar_ptr, bytes);
3132	victim = _int_malloc (ar_ptr, bytes);
3133	if (!victim && ar_ptr != NULL)
3134	{
3135	ar_ptr = arena_get_retry (ar_ptr, bytes);
3136	victim = _int_malloc (ar_ptr, bytes);
3137	}
3138
3139
3140	if (ar_ptr != NULL)
3141	__libc_lock_unlock (ar_ptr->mutex);
3142
3143	/ In a low memory situation, we may not be able to allocate memory*
3144	- in which case, we just keep trying later. However, we
3145	typically do this very early, so either there is sufficient
3146	memory, or there isn't enough memory to do non-trivial
3147	allocations anyway. /*
3148	if (victim)
3149	{
3150	tcache = (tcache_perthread_struct *) victim;
3151	memset (tcache, `0`, sizeof (tcache_perthread_struct));
3152	}
3153
3154	}
3155
3156	# define MAYBE_INIT_TCACHE() \
3157	if (__glibc_unlikely (tcache == NULL)) \
3158	tcache_init();
3159
3160	#else /* !USE_TCACHE */
3161	# define MAYBE_INIT_TCACHE()
3162
3163	static void
3164	tcache_thread_shutdown (void)
3165	{
3166	/ Nothing to do if there is no thread cache. /
3167	}
3168
3169	#endif /* !USE_TCACHE */
3170
3171	#if IS_IN (libc)
3172	void *
3173	__libc_malloc (size_t bytes)
3174	{
3175	mstate ar_ptr;
3176	void *victim;
3177
3178	_Static_assert (PTRDIFF_MAX <= SIZE_MAX / `2`,
3179	"PTRDIFF_MAX is not more than half of SIZE_MAX");
3180
3181	if (!__malloc_initialized)
3182	ptmalloc_init ();
3183	#if USE_TCACHE
3184	/ int_free also calls request2size, be careful to not pad twice. /
3185	size_t tbytes;
3186	if (!checked_request2size (bytes, &tbytes))
3187	{
3188	__set_errno (ENOMEM);
3189	return NULL;
3190	}
3191	size_t tc_idx = csize2tidx (tbytes);
3192
3193	MAYBE_INIT_TCACHE ();
3194
3195	DIAG_PUSH_NEEDS_COMMENT;
3196	if (tc_idx < mp_.tcache_bins
3197	&& tcache
3198	&& tcache->counts[tc_idx] > `0`)
3199	{
3200	victim = tcache_get (tc_idx);
3201	return tag_new_usable (victim);
3202	}
3203	DIAG_POP_NEEDS_COMMENT;
3204	#endif
3205
3206	if (SINGLE_THREAD_P)
3207	{
3208	victim = tag_new_usable (_int_malloc (&main_arena, bytes));
3209	assert (!victim \|\| chunk_is_mmapped (mem2chunk (victim)) \|\|
3210	&main_arena == arena_for_chunk (mem2chunk (victim)));
3211	return victim;
3212	}
3213
3214	arena_get (ar_ptr, bytes);
3215
3216	victim = _int_malloc (ar_ptr, bytes);
3217	/ Retry with another arena only if we were able to find a usable arena*
3218	before. /*
3219	if (!victim && ar_ptr != NULL)
3220	{
3221	LIBC_PROBE (memory_malloc_retry, `1`, bytes);
3222	ar_ptr = arena_get_retry (ar_ptr, bytes);
3223	victim = _int_malloc (ar_ptr, bytes);
3224	}
3225
3226	if (ar_ptr != NULL)
3227	__libc_lock_unlock (ar_ptr->mutex);
3228
3229	victim = tag_new_usable (victim);
3230
3231	assert (!victim \|\| chunk_is_mmapped (mem2chunk (victim)) \|\|
3232	ar_ptr == arena_for_chunk (mem2chunk (victim)));
3233	return victim;
3234	}
3235	libc_hidden_def (__libc_malloc)
3236
3237	void
3238	__libc_free (void *mem)
3239	{
3240	mstate ar_ptr;
3241	mchunkptr p; / chunk corresponding to mem /
3242
3243	if (mem == `0`) / free(0) has no effect /
3244	return;
3245
3246	/ Quickly check that the freed pointer matches the tag for the memory.*
3247	This gives a useful double-free detection. /*
3248	if (__glibc_unlikely (mtag_enabled))
3249	(volatile* char *)mem;
3250
3251	int err = errno;
3252
3253	p = mem2chunk (mem);
3254
3255	if (chunk_is_mmapped (p)) / release mmapped memory. /
3256	{
3257	/ See if the dynamic brk/mmap threshold needs adjusting.*
3258	Dumped fake mmapped chunks do not affect the threshold. /*
3259	if (!mp_.no_dyn_threshold
3260	&& chunksize_nomask (p) > mp_.mmap_threshold
3261	&& chunksize_nomask (p) <= DEFAULT_MMAP_THRESHOLD_MAX)
3262	{
3263	mp_.mmap_threshold = chunksize (p);
3264	mp_.trim_threshold = `2` * mp_.mmap_threshold;
3265	LIBC_PROBE (memory_mallopt_free_dyn_thresholds, `2`,
3266	mp_.mmap_threshold, mp_.trim_threshold);
3267	}
3268	munmap_chunk (p);
3269	}
3270	else
3271	{
3272	MAYBE_INIT_TCACHE ();
3273
3274	/ Mark the chunk as belonging to the library again. /
3275	(void)tag_region (chunk2mem (p), memsize (p));
3276
3277	ar_ptr = arena_for_chunk (p);
3278	_int_free (ar_ptr, p, `0`);
3279	}
3280
3281	__set_errno (err);
3282	}
3283	libc_hidden_def (__libc_free)
3284
3285	void *
3286	__libc_realloc (void *oldmem, size_t bytes)
3287	{
3288	mstate ar_ptr;
3289	INTERNAL_SIZE_T nb; / padded request size /
3290
3291	void newp; /* chunk to return /
3292
3293	if (!__malloc_initialized)
3294	ptmalloc_init ();
3295
3296	#if REALLOC_ZERO_BYTES_FREES
3297	if (bytes == `0` && oldmem != NULL)
3298	{
3299	__libc_free (oldmem); return `0`;
3300	}
3301	#endif
3302
3303	/ realloc of null is supposed to be same as malloc /
3304	if (oldmem == `0`)
3305	return __libc_malloc (bytes);
3306
3307	/ Perform a quick check to ensure that the pointer's tag matches the*
3308	memory's tag. /*
3309	if (__glibc_unlikely (mtag_enabled))
3310	(volatile* char*) oldmem;
3311
3312	/ chunk corresponding to oldmem /
3313	const mchunkptr oldp = mem2chunk (oldmem);
3314	/ its size /
3315	const INTERNAL_SIZE_T oldsize = chunksize (oldp);
3316
3317	if (chunk_is_mmapped (oldp))
3318	ar_ptr = NULL;
3319	else
3320	{
3321	MAYBE_INIT_TCACHE ();
3322	ar_ptr = arena_for_chunk (oldp);
3323	}
3324
3325	/ Little security check which won't hurt performance: the allocator*
3326	never wrapps around at the end of the address space. Therefore
3327	we can exclude some size values which might appear here by
3328	accident or by "design" from some intruder. /*
3329	if ((__builtin_expect ((uintptr_t) oldp > (uintptr_t) -oldsize, `0`)
3330	\|\| __builtin_expect (misaligned_chunk (oldp), `0`)))
3331	malloc_printerr ("realloc(): invalid pointer");
3332
3333	if (!checked_request2size (bytes, &nb))
3334	{
3335	__set_errno (ENOMEM);
3336	return NULL;
3337	}
3338
3339	if (chunk_is_mmapped (oldp))
3340	{
3341	void *newmem;
3342
3343	#if HAVE_MREMAP
3344	newp = mremap_chunk (oldp, nb);
3345	if (newp)
3346	{
3347	void *newmem = chunk2mem_tag (newp);
3348	/ Give the new block a different tag. This helps to ensure*
3349	that stale handles to the previous mapping are not
3350	reused. There's a performance hit for both us and the
3351	caller for doing this, so we might want to
3352	reconsider. /*
3353	return tag_new_usable (newmem);
3354	}
3355	#endif
3356	/ Note the extra SIZE_SZ overhead. /
3357	if (oldsize - SIZE_SZ >= nb)
3358	return oldmem; / do nothing /
3359
3360	/ Must alloc, copy, free. /
3361	newmem = __libc_malloc (bytes);
3362	if (newmem == `0`)
3363	return `0`; / propagate failure /
3364
3365	memcpy (newmem, oldmem, oldsize - CHUNK_HDR_SZ);
3366	munmap_chunk (oldp);
3367	return newmem;
3368	}
3369
3370	if (SINGLE_THREAD_P)
3371	{
3372	newp = _int_realloc (ar_ptr, oldp, oldsize, nb);
3373	assert (!newp \|\| chunk_is_mmapped (mem2chunk (newp)) \|\|
3374	ar_ptr == arena_for_chunk (mem2chunk (newp)));
3375
3376	return newp;
3377	}
3378
3379	__libc_lock_lock (ar_ptr->mutex);
3380
3381	newp = _int_realloc (ar_ptr, oldp, oldsize, nb);
3382
3383	__libc_lock_unlock (ar_ptr->mutex);
3384	assert (!newp \|\| chunk_is_mmapped (mem2chunk (newp)) \|\|
3385	ar_ptr == arena_for_chunk (mem2chunk (newp)));
3386
3387	if (newp == NULL)
3388	{
3389	/ Try harder to allocate memory in other arenas. /
3390	LIBC_PROBE (memory_realloc_retry, `2`, bytes, oldmem);
3391	newp = __libc_malloc (bytes);
3392	if (newp != NULL)
3393	{
3394	size_t sz = memsize (oldp);
3395	memcpy (newp, oldmem, sz);
3396	(void) tag_region (chunk2mem (oldp), sz);
3397	_int_free (ar_ptr, oldp, `0`);
3398	}
3399	}
3400
3401	return newp;
3402	}
3403	libc_hidden_def (__libc_realloc)
3404
3405	void *
3406	__libc_memalign (size_t alignment, size_t bytes)
3407	{
3408	if (!__malloc_initialized)
3409	ptmalloc_init ();
3410
3411	void *address = RETURN_ADDRESS (`0`);
3412	return _mid_memalign (alignment, bytes, address);
3413	}
3414
3415	static void *
3416	_mid_memalign (size_t alignment, size_t bytes, void *address)
3417	{
3418	mstate ar_ptr;
3419	void *p;
3420
3421	/ If we need less alignment than we give anyway, just relay to malloc. /
3422	if (alignment <= MALLOC_ALIGNMENT)
3423	return __libc_malloc (bytes);
3424
3425	/ Otherwise, ensure that it is at least a minimum chunk size /
3426	if (alignment < MINSIZE)
3427	alignment = MINSIZE;
3428
3429	/ If the alignment is greater than SIZE_MAX / 2 + 1 it cannot be a*
3430	power of 2 and will cause overflow in the check below. /*
3431	if (alignment > SIZE_MAX / `2` + `1`)
3432	{
3433	__set_errno (EINVAL);
3434	return `0`;
3435	}
3436
3437
3438	/ Make sure alignment is power of 2. /
3439	if (!powerof2 (alignment))
3440	{
3441	size_t a = MALLOC_ALIGNMENT * `2`;
3442	while (a < alignment)
3443	a <<= `1`;
3444	alignment = a;
3445	}
3446
3447	if (SINGLE_THREAD_P)
3448	{
3449	p = _int_memalign (&main_arena, alignment, bytes);
3450	assert (!p \|\| chunk_is_mmapped (mem2chunk (p)) \|\|
3451	&main_arena == arena_for_chunk (mem2chunk (p)));
3452	return tag_new_usable (p);
3453	}
3454
3455	arena_get (ar_ptr, bytes + alignment + MINSIZE);
3456
3457	p = _int_memalign (ar_ptr, alignment, bytes);
3458	if (!p && ar_ptr != NULL)
3459	{
3460	LIBC_PROBE (memory_memalign_retry, `2`, bytes, alignment);
3461	ar_ptr = arena_get_retry (ar_ptr, bytes);
3462	p = _int_memalign (ar_ptr, alignment, bytes);
3463	}
3464
3465	if (ar_ptr != NULL)
3466	__libc_lock_unlock (ar_ptr->mutex);
3467
3468	assert (!p \|\| chunk_is_mmapped (mem2chunk (p)) \|\|
3469	ar_ptr == arena_for_chunk (mem2chunk (p)));
3470	return tag_new_usable (p);
3471	}
3472	/ For ISO C11. /
3473	weak_alias (__libc_memalign, aligned_alloc)
3474	libc_hidden_def (__libc_memalign)
3475
3476	void *
3477	__libc_valloc (size_t bytes)
3478	{
3479	if (!__malloc_initialized)
3480	ptmalloc_init ();
3481
3482	void *address = RETURN_ADDRESS (`0`);
3483	size_t pagesize = GLRO (dl_pagesize);
3484	return _mid_memalign (pagesize, bytes, address);
3485	}
3486
3487	void *
3488	__libc_pvalloc (size_t bytes)
3489	{
3490	if (!__malloc_initialized)
3491	ptmalloc_init ();
3492
3493	void *address = RETURN_ADDRESS (`0`);
3494	size_t pagesize = GLRO (dl_pagesize);
3495	size_t rounded_bytes;
3496	/ ALIGN_UP with overflow check. /
3497	if (__glibc_unlikely (__builtin_add_overflow (bytes,
3498	pagesize - `1`,
3499	&rounded_bytes)))
3500	{
3501	__set_errno (ENOMEM);
3502	return `0`;
3503	}
3504	rounded_bytes = rounded_bytes & -(pagesize - `1`);
3505
3506	return _mid_memalign (pagesize, rounded_bytes, address);
3507	}
3508
3509	void *
3510	__libc_calloc (size_t n, size_t elem_size)
3511	{
3512	mstate av;
3513	mchunkptr oldtop;
3514	INTERNAL_SIZE_T sz, oldtopsize;
3515	void *mem;
3516	unsigned long clearsize;
3517	unsigned long nclears;
3518	INTERNAL_SIZE_T *d;
3519	ptrdiff_t bytes;
3520
3521	if (__glibc_unlikely (__builtin_mul_overflow (n, elem_size, &bytes)))
3522	{
3523	__set_errno (ENOMEM);
3524	return NULL;
3525	}
3526
3527	sz = bytes;
3528
3529	if (!__malloc_initialized)
3530	ptmalloc_init ();
3531
3532	MAYBE_INIT_TCACHE ();
3533
3534	if (SINGLE_THREAD_P)
3535	av = &main_arena;
3536	else
3537	arena_get (av, sz);
3538
3539	if (av)
3540	{
3541	/ Check if we hand out the top chunk, in which case there may be no*
3542	need to clear. /*
3543	#if MORECORE_CLEARS
3544	oldtop = top (av);
3545	oldtopsize = chunksize (top (av));
3546	# if MORECORE_CLEARS < 2
3547	/ Only newly allocated memory is guaranteed to be cleared. /
3548	if (av == &main_arena &&
3549	oldtopsize < mp_.sbrk_base + av->max_system_mem - (char *) oldtop)
3550	oldtopsize = (mp_.sbrk_base + av->max_system_mem - (char *) oldtop);
3551	# endif
3552	if (av != &main_arena)
3553	{
3554	heap_info *heap = heap_for_ptr (oldtop);
3555	if (oldtopsize < (char ) heap + heap->mprotect_size - (char* *) oldtop)
3556	oldtopsize = (char ) heap + heap->mprotect_size - (char* *) oldtop;
3557	}
3558	#endif
3559	}
3560	else
3561	{
3562	/ No usable arenas. /
3563	oldtop = `0`;
3564	oldtopsize = `0`;
3565	}
3566	mem = _int_malloc (av, sz);
3567
3568	assert (!mem \|\| chunk_is_mmapped (mem2chunk (mem)) \|\|
3569	av == arena_for_chunk (mem2chunk (mem)));
3570
3571	if (!SINGLE_THREAD_P)
3572	{
3573	if (mem == `0` && av != NULL)
3574	{
3575	LIBC_PROBE (memory_calloc_retry, `1`, sz);
3576	av = arena_get_retry (av, sz);
3577	mem = _int_malloc (av, sz);
3578	}
3579
3580	if (av != NULL)
3581	__libc_lock_unlock (av->mutex);
3582	}
3583
3584	/ Allocation failed even after a retry. /
3585	if (mem == `0`)
3586	return `0`;
3587
3588	mchunkptr p = mem2chunk (mem);
3589
3590	/ If we are using memory tagging, then we need to set the tags*
3591	regardless of MORECORE_CLEARS, so we zero the whole block while
3592	doing so. /*
3593	if (__glibc_unlikely (mtag_enabled))
3594	return tag_new_zero_region (mem, memsize (p));
3595
3596	INTERNAL_SIZE_T csz = chunksize (p);
3597
3598	/ Two optional cases in which clearing not necessary /
3599	if (chunk_is_mmapped (p))
3600	{
3601	if (__builtin_expect (perturb_byte, `0`))
3602	return memset (mem, `0`, sz);
3603
3604	return mem;
3605	}
3606
3607	#if MORECORE_CLEARS
3608	if (perturb_byte == `0` && (p == oldtop && csz > oldtopsize))
3609	{
3610	/ clear only the bytes from non-freshly-sbrked memory /
3611	csz = oldtopsize;
3612	}
3613	#endif
3614
3615	/ Unroll clear of <= 36 bytes (72 if 8byte sizes). We know that*
3616	contents have an odd number of INTERNAL_SIZE_T-sized words;
3617	minimally 3. /*
3618	d = (INTERNAL_SIZE_T *) mem;
3619	clearsize = csz - SIZE_SZ;
3620	nclears = clearsize / sizeof (INTERNAL_SIZE_T);
3621	assert (nclears >= `3`);
3622
3623	if (nclears > `9`)
3624	return memset (d, `0`, clearsize);
3625
3626	else
3627	{
3628	*(d + `0`) = `0`;
3629	*(d + `1`) = `0`;
3630	*(d + `2`) = `0`;
3631	if (nclears > `4`)
3632	{
3633	*(d + `3`) = `0`;
3634	*(d + `4`) = `0`;
3635	if (nclears > `6`)
3636	{
3637	*(d + `5`) = `0`;
3638	*(d + `6`) = `0`;
3639	if (nclears > `8`)
3640	{
3641	*(d + `7`) = `0`;
3642	*(d + `8`) = `0`;
3643	}
3644	}
3645	}
3646	}
3647
3648	return mem;
3649	}
3650	#endif /* IS_IN (libc) */
3651
3652	/*
3653	------------------------------ malloc ------------------------------
3654	*/
3655
3656	static void *
3657	_int_malloc (mstate av, size_t bytes)
3658	{
3659	INTERNAL_SIZE_T nb; / normalized request size /
3660	unsigned int idx; / associated bin index /
3661	mbinptr bin; / associated bin /
3662
3663	mchunkptr victim; / inspected/selected chunk /
3664	INTERNAL_SIZE_T size; / its size /
3665	int victim_index; / its bin index /
3666
3667	mchunkptr remainder; / remainder from a split /
3668	unsigned long remainder_size; / its size /
3669
3670	unsigned int block; / bit map traverser /
3671	unsigned int bit; / bit map traverser /
3672	unsigned int map; / current word of binmap /
3673
3674	mchunkptr fwd; / misc temp for linking /
3675	mchunkptr bck; / misc temp for linking /
3676
3677	#if USE_TCACHE
3678	size_t tcache_unsorted_count; / count of unsorted chunks processed /
3679	#endif
3680
3681	/*
3682	Convert request size to internal form by adding SIZE_SZ bytes
3683	overhead plus possibly more to obtain necessary alignment and/or
3684	to obtain a size of at least MINSIZE, the smallest allocatable
3685	size. Also, checked_request2size returns false for request sizes
3686	that are so large that they wrap around zero when padded and
3687	aligned.
3688	*/
3689
3690	if (!checked_request2size (bytes, &nb))
3691	{
3692	__set_errno (ENOMEM);
3693	return NULL;
3694	}
3695
3696	/ There are no usable arenas. Fall back to sysmalloc to get a chunk from*
3697	mmap. /*
3698	if (__glibc_unlikely (av == NULL))
3699	{
3700	void *p = sysmalloc (nb, av);
3701	if (p != NULL)
3702	alloc_perturb (p, bytes);
3703	return p;
3704	}
3705
3706	/*
3707	If the size qualifies as a fastbin, first check corresponding bin.
3708	This code is safe to execute even if av is not yet initialized, so we
3709	can try it without checking, which saves some time on this fast path.
3710	*/
3711
3712	#define REMOVE_FB(fb, victim, pp) \
3713	do \
3714	{ \
3715	victim = pp; \
3716	if (victim == NULL) \
3717	break; \
3718	pp = REVEAL_PTR (victim->fd); \
3719	if (__glibc_unlikely (pp != NULL && misaligned_chunk (pp))) \
3720	malloc_printerr ("malloc(): unaligned fastbin chunk detected"); \
3721	} \
3722	while ((pp = catomic_compare_and_exchange_val_acq (fb, pp, victim)) \
3723	!= victim); \
3724
3725	if ((unsigned long) (nb) <= (unsigned long) (get_max_fast ()))
3726	{
3727	idx = fastbin_index (nb);
3728	mfastbinptr *fb = &fastbin (av, idx);
3729	mchunkptr pp;
3730	victim = *fb;
3731
3732	if (victim != NULL)
3733	{
3734	if (__glibc_unlikely (misaligned_chunk (victim)))
3735	malloc_printerr ("malloc(): unaligned fastbin chunk detected 2");
3736
3737	if (SINGLE_THREAD_P)
3738	*fb = REVEAL_PTR (victim->fd);
3739	else
3740	REMOVE_FB (fb, pp, victim);
3741	if (__glibc_likely (victim != NULL))
3742	{
3743	size_t victim_idx = fastbin_index (chunksize (victim));
3744	if (__builtin_expect (victim_idx != idx, `0`))
3745	malloc_printerr ("malloc(): memory corruption (fast)");
3746	check_remalloced_chunk (av, victim, nb);
3747	#if USE_TCACHE
3748	/ While we're here, if we see other chunks of the same size,*
3749	stash them in the tcache. /*
3750	size_t tc_idx = csize2tidx (nb);
3751	if (tcache && tc_idx < mp_.tcache_bins)
3752	{
3753	mchunkptr tc_victim;
3754
3755	/ While bin not empty and tcache not full, copy chunks. /
3756	while (tcache->counts[tc_idx] < mp_.tcache_count
3757	&& (tc_victim = *fb) != NULL)
3758	{
3759	if (__glibc_unlikely (misaligned_chunk (tc_victim)))
3760	malloc_printerr ("malloc(): unaligned fastbin chunk detected 3");
3761	if (SINGLE_THREAD_P)
3762	*fb = REVEAL_PTR (tc_victim->fd);
3763	else
3764	{
3765	REMOVE_FB (fb, pp, tc_victim);
3766	if (__glibc_unlikely (tc_victim == NULL))
3767	break;
3768	}
3769	tcache_put (tc_victim, tc_idx);
3770	}
3771	}
3772	#endif
3773	void *p = chunk2mem (victim);
3774	alloc_perturb (p, bytes);
3775	return p;
3776	}
3777	}
3778	}
3779
3780	/*
3781	If a small request, check regular bin. Since these "smallbins"
3782	hold one size each, no searching within bins is necessary.
3783	(For a large request, we need to wait until unsorted chunks are
3784	processed to find best fit. But for small ones, fits are exact
3785	anyway, so we can check now, which is faster.)
3786	*/
3787
3788	if (in_smallbin_range (nb))
3789	{
3790	idx = smallbin_index (nb);
3791	bin = bin_at (av, idx);
3792
3793	if ((victim = last (bin)) != bin)
3794	{
3795	bck = victim->bk;
3796	if (__glibc_unlikely (bck->fd != victim))
3797	malloc_printerr ("malloc(): smallbin double linked list corrupted");
3798	set_inuse_bit_at_offset (victim, nb);
3799	bin->bk = bck;
3800	bck->fd = bin;
3801
3802	if (av != &main_arena)
3803	set_non_main_arena (victim);
3804	check_malloced_chunk (av, victim, nb);
3805	#if USE_TCACHE
3806	/ While we're here, if we see other chunks of the same size,*
3807	stash them in the tcache. /*
3808	size_t tc_idx = csize2tidx (nb);
3809	if (tcache && tc_idx < mp_.tcache_bins)
3810	{
3811	mchunkptr tc_victim;
3812
3813	/ While bin not empty and tcache not full, copy chunks over. /
3814	while (tcache->counts[tc_idx] < mp_.tcache_count
3815	&& (tc_victim = last (bin)) != bin)
3816	{
3817	if (tc_victim != `0`)
3818	{
3819	bck = tc_victim->bk;
3820	set_inuse_bit_at_offset (tc_victim, nb);
3821	if (av != &main_arena)
3822	set_non_main_arena (tc_victim);
3823	bin->bk = bck;
3824	bck->fd = bin;
3825
3826	tcache_put (tc_victim, tc_idx);
3827	}
3828	}
3829	}
3830	#endif
3831	void *p = chunk2mem (victim);
3832	alloc_perturb (p, bytes);
3833	return p;
3834	}
3835	}
3836
3837	/*
3838	If this is a large request, consolidate fastbins before continuing.
3839	While it might look excessive to kill all fastbins before
3840	even seeing if there is space available, this avoids
3841	fragmentation problems normally associated with fastbins.
3842	Also, in practice, programs tend to have runs of either small or
3843	large requests, but less often mixtures, so consolidation is not
3844	invoked all that often in most programs. And the programs that
3845	it is called frequently in otherwise tend to fragment.
3846	*/
3847
3848	else
3849	{
3850	idx = largebin_index (nb);
3851	if (atomic_load_relaxed (&av->have_fastchunks))
3852	malloc_consolidate (av);
3853	}
3854
3855	/*
3856	Process recently freed or remaindered chunks, taking one only if
3857	it is exact fit, or, if this a small request, the chunk is remainder from
3858	the most recent non-exact fit. Place other traversed chunks in
3859	bins. Note that this step is the only place in any routine where
3860	chunks are placed in bins.
3861
3862	The outer loop here is needed because we might not realize until
3863	near the end of malloc that we should have consolidated, so must
3864	do so and retry. This happens at most once, and only when we would
3865	otherwise need to expand memory to service a "small" request.
3866	*/
3867
3868	#if USE_TCACHE
3869	INTERNAL_SIZE_T tcache_nb = `0`;
3870	size_t tc_idx = csize2tidx (nb);
3871	if (tcache && tc_idx < mp_.tcache_bins)
3872	tcache_nb = nb;
3873	int return_cached = `0`;
3874
3875	tcache_unsorted_count = `0`;
3876	#endif
3877
3878	for (;; )
3879	{
3880	int iters = `0`;
3881	while ((victim = unsorted_chunks (av)->bk) != unsorted_chunks (av))
3882	{
3883	bck = victim->bk;
3884	size = chunksize (victim);
3885	mchunkptr next = chunk_at_offset (victim, size);
3886
3887	if (__glibc_unlikely (size <= CHUNK_HDR_SZ)
3888	\|\| __glibc_unlikely (size > av->system_mem))
3889	malloc_printerr ("malloc(): invalid size (unsorted)");
3890	if (__glibc_unlikely (chunksize_nomask (next) < CHUNK_HDR_SZ)
3891	\|\| __glibc_unlikely (chunksize_nomask (next) > av->system_mem))
3892	malloc_printerr ("malloc(): invalid next size (unsorted)");
3893	if (__glibc_unlikely ((prev_size (next) & ~(SIZE_BITS)) != size))
3894	malloc_printerr ("malloc(): mismatching next->prev_size (unsorted)");
3895	if (__glibc_unlikely (bck->fd != victim)
3896	\|\| __glibc_unlikely (victim->fd != unsorted_chunks (av)))
3897	malloc_printerr ("malloc(): unsorted double linked list corrupted");
3898	if (__glibc_unlikely (prev_inuse (next)))
3899	malloc_printerr ("malloc(): invalid next->prev_inuse (unsorted)");
3900
3901	/*
3902	If a small request, try to use last remainder if it is the
3903	only chunk in unsorted bin. This helps promote locality for
3904	runs of consecutive small requests. This is the only
3905	exception to best-fit, and applies only when there is
3906	no exact fit for a small chunk.
3907	*/
3908
3909	if (in_smallbin_range (nb) &&
3910	bck == unsorted_chunks (av) &&
3911	victim == av->last_remainder &&
3912	(unsigned long) (size) > (unsigned long) (nb + MINSIZE))
3913	{
3914	/ split and reattach remainder /
3915	remainder_size = size - nb;
3916	remainder = chunk_at_offset (victim, nb);
3917	unsorted_chunks (av)->bk = unsorted_chunks (av)->fd = remainder;
3918	av->last_remainder = remainder;
3919	remainder->bk = remainder->fd = unsorted_chunks (av);
3920	if (!in_smallbin_range (remainder_size))
3921	{
3922	remainder->fd_nextsize = NULL;
3923	remainder->bk_nextsize = NULL;
3924	}
3925
3926	set_head (victim, nb \| PREV_INUSE \|
3927	(av != &main_arena ? NON_MAIN_ARENA : `0`));
3928	set_head (remainder, remainder_size \| PREV_INUSE);
3929	set_foot (remainder, remainder_size);
3930
3931	check_malloced_chunk (av, victim, nb);
3932	void *p = chunk2mem (victim);
3933	alloc_perturb (p, bytes);
3934	return p;
3935	}
3936
3937	/ remove from unsorted list /
3938	if (__glibc_unlikely (bck->fd != victim))
3939	malloc_printerr ("malloc(): corrupted unsorted chunks 3");
3940	unsorted_chunks (av)->bk = bck;
3941	bck->fd = unsorted_chunks (av);
3942
3943	/ Take now instead of binning if exact fit /
3944
3945	if (size == nb)
3946	{
3947	set_inuse_bit_at_offset (victim, size);
3948	if (av != &main_arena)
3949	set_non_main_arena (victim);
3950	#if USE_TCACHE
3951	/ Fill cache first, return to user only if cache fills.*
3952	We may return one of these chunks later. /*
3953	if (tcache_nb
3954	&& tcache->counts[tc_idx] < mp_.tcache_count)
3955	{
3956	tcache_put (victim, tc_idx);
3957	return_cached = `1`;
3958	continue;
3959	}
3960	else
3961	{
3962	#endif
3963	check_malloced_chunk (av, victim, nb);
3964	void *p = chunk2mem (victim);
3965	alloc_perturb (p, bytes);
3966	return p;
3967	#if USE_TCACHE
3968	}
3969	#endif
3970	}
3971
3972	/ place chunk in bin /
3973
3974	if (in_smallbin_range (size))
3975	{
3976	victim_index = smallbin_index (size);
3977	bck = bin_at (av, victim_index);
3978	fwd = bck->fd;
3979	}
3980	else
3981	{
3982	victim_index = largebin_index (size);
3983	bck = bin_at (av, victim_index);
3984	fwd = bck->fd;
3985
3986	/ maintain large bins in sorted order /
3987	if (fwd != bck)
3988	{
3989	/ Or with inuse bit to speed comparisons /
3990	size \|= PREV_INUSE;
3991	/ if smaller than smallest, bypass loop below /
3992	assert (chunk_main_arena (bck->bk));
3993	if ((unsigned long) (size)
3994	< (unsigned long) chunksize_nomask (bck->bk))
3995	{
3996	fwd = bck;
3997	bck = bck->bk;
3998
3999	victim->fd_nextsize = fwd->fd;
4000	victim->bk_nextsize = fwd->fd->bk_nextsize;
4001	fwd->fd->bk_nextsize = victim->bk_nextsize->fd_nextsize = victim;
4002	}
4003	else
4004	{
4005	assert (chunk_main_arena (fwd));
4006	while ((unsigned long) size < chunksize_nomask (fwd))
4007	{
4008	fwd = fwd->fd_nextsize;
4009	assert (chunk_main_arena (fwd));
4010	}
4011
4012	if ((unsigned long) size
4013	== (unsigned long) chunksize_nomask (fwd))
4014	/ Always insert in the second position. /
4015	fwd = fwd->fd;
4016	else
4017	{
4018	victim->fd_nextsize = fwd;
4019	victim->bk_nextsize = fwd->bk_nextsize;
4020	if (__glibc_unlikely (fwd->bk_nextsize->fd_nextsize != fwd))
4021	malloc_printerr ("malloc(): largebin double linked list corrupted (nextsize)");
4022	fwd->bk_nextsize = victim;
4023	victim->bk_nextsize->fd_nextsize = victim;
4024	}
4025	bck = fwd->bk;
4026	if (bck->fd != fwd)
4027	malloc_printerr ("malloc(): largebin double linked list corrupted (bk)");
4028	}
4029	}
4030	else
4031	victim->fd_nextsize = victim->bk_nextsize = victim;
4032	}
4033
4034	mark_bin (av, victim_index);
4035	victim->bk = bck;
4036	victim->fd = fwd;
4037	fwd->bk = victim;
4038	bck->fd = victim;
4039
4040	#if USE_TCACHE
4041	/ If we've processed as many chunks as we're allowed while*
4042	filling the cache, return one of the cached ones. /*
4043	++tcache_unsorted_count;
4044	if (return_cached
4045	&& mp_.tcache_unsorted_limit > `0`
4046	&& tcache_unsorted_count > mp_.tcache_unsorted_limit)
4047	{
4048	return tcache_get (tc_idx);
4049	}
4050	#endif
4051
4052	#define MAX_ITERS 10000
4053	if (++iters >= MAX_ITERS)
4054	break;
4055	}
4056
4057	#if USE_TCACHE
4058	/ If all the small chunks we found ended up cached, return one now. /
4059	if (return_cached)
4060	{
4061	return tcache_get (tc_idx);
4062	}
4063	#endif
4064
4065	/*
4066	If a large request, scan through the chunks of current bin in
4067	sorted order to find smallest that fits. Use the skip list for this.
4068	*/
4069
4070	if (!in_smallbin_range (nb))
4071	{
4072	bin = bin_at (av, idx);
4073
4074	/ skip scan if empty or largest chunk is too small /
4075	if ((victim = first (bin)) != bin
4076	&& (unsigned long) chunksize_nomask (victim)
4077	>= (unsigned long) (nb))
4078	{
4079	victim = victim->bk_nextsize;
4080	while (((unsigned long) (size = chunksize (victim)) <
4081	(unsigned long) (nb)))
4082	victim = victim->bk_nextsize;
4083
4084	/ Avoid removing the first entry for a size so that the skip*
4085	list does not have to be rerouted. /*
4086	if (victim != last (bin)
4087	&& chunksize_nomask (victim)
4088	== chunksize_nomask (victim->fd))
4089	victim = victim->fd;
4090
4091	remainder_size = size - nb;
4092	unlink_chunk (av, victim);
4093
4094	/ Exhaust /
4095	if (remainder_size < MINSIZE)
4096	{
4097	set_inuse_bit_at_offset (victim, size);
4098	if (av != &main_arena)
4099	set_non_main_arena (victim);
4100	}
4101	/ Split /
4102	else
4103	{
4104	remainder = chunk_at_offset (victim, nb);
4105	/ We cannot assume the unsorted list is empty and therefore*
4106	have to perform a complete insert here. /*
4107	bck = unsorted_chunks (av);
4108	fwd = bck->fd;
4109	if (__glibc_unlikely (fwd->bk != bck))
4110	malloc_printerr ("malloc(): corrupted unsorted chunks");
4111	remainder->bk = bck;
4112	remainder->fd = fwd;
4113	bck->fd = remainder;
4114	fwd->bk = remainder;
4115	if (!in_smallbin_range (remainder_size))
4116	{
4117	remainder->fd_nextsize = NULL;
4118	remainder->bk_nextsize = NULL;
4119	}
4120	set_head (victim, nb \| PREV_INUSE \|
4121	(av != &main_arena ? NON_MAIN_ARENA : `0`));
4122	set_head (remainder, remainder_size \| PREV_INUSE);
4123	set_foot (remainder, remainder_size);
4124	}
4125	check_malloced_chunk (av, victim, nb);
4126	void *p = chunk2mem (victim);
4127	alloc_perturb (p, bytes);
4128	return p;
4129	}
4130	}
4131
4132	/*
4133	Search for a chunk by scanning bins, starting with next largest
4134	bin. This search is strictly by best-fit; i.e., the smallest
4135	(with ties going to approximately the least recently used) chunk
4136	that fits is selected.
4137
4138	The bitmap avoids needing to check that most blocks are nonempty.
4139	The particular case of skipping all bins during warm-up phases
4140	when no chunks have been returned yet is faster than it might look.
4141	*/
4142
4143	++idx;
4144	bin = bin_at (av, idx);
4145	block = idx2block (idx);
4146	map = av->binmap[block];
4147	bit = idx2bit (idx);
4148
4149	for (;; )
4150	{
4151	/ Skip rest of block if there are no more set bits in this block. /
4152	if (bit > map \|\| bit == `0`)
4153	{
4154	do
4155	{
4156	if (++block >= BINMAPSIZE) / out of bins /
4157	goto use_top;
4158	}
4159	while ((map = av->binmap[block]) == `0`);
4160
4161	bin = bin_at (av, (block << BINMAPSHIFT));
4162	bit = `1`;
4163	}
4164
4165	/ Advance to bin with set bit. There must be one. /
4166	while ((bit & map) == `0`)
4167	{
4168	bin = next_bin (bin);
4169	bit <<= `1`;
4170	assert (bit != `0`);
4171	}
4172
4173	/ Inspect the bin. It is likely to be non-empty /
4174	victim = last (bin);
4175
4176	/ If a false alarm (empty bin), clear the bit. /
4177	if (victim == bin)
4178	{
4179	av->binmap[block] = map &= ~bit; / Write through /
4180	bin = next_bin (bin);
4181	bit <<= `1`;
4182	}
4183
4184	else
4185	{
4186	size = chunksize (victim);
4187
4188	/ We know the first chunk in this bin is big enough to use. /
4189	assert ((unsigned long) (size) >= (unsigned long) (nb));
4190
4191	remainder_size = size - nb;
4192
4193	/ unlink /
4194	unlink_chunk (av, victim);
4195
4196	/ Exhaust /
4197	if (remainder_size < MINSIZE)
4198	{
4199	set_inuse_bit_at_offset (victim, size);
4200	if (av != &main_arena)
4201	set_non_main_arena (victim);
4202	}
4203
4204	/ Split /
4205	else
4206	{
4207	remainder = chunk_at_offset (victim, nb);
4208
4209	/ We cannot assume the unsorted list is empty and therefore*
4210	have to perform a complete insert here. /*
4211	bck = unsorted_chunks (av);
4212	fwd = bck->fd;
4213	if (__glibc_unlikely (fwd->bk != bck))
4214	malloc_printerr ("malloc(): corrupted unsorted chunks 2");
4215	remainder->bk = bck;
4216	remainder->fd = fwd;
4217	bck->fd = remainder;
4218	fwd->bk = remainder;
4219
4220	/ advertise as last remainder /
4221	if (in_smallbin_range (nb))
4222	av->last_remainder = remainder;
4223	if (!in_smallbin_range (remainder_size))
4224	{
4225	remainder->fd_nextsize = NULL;
4226	remainder->bk_nextsize = NULL;
4227	}
4228	set_head (victim, nb \| PREV_INUSE \|
4229	(av != &main_arena ? NON_MAIN_ARENA : `0`));
4230	set_head (remainder, remainder_size \| PREV_INUSE);
4231	set_foot (remainder, remainder_size);
4232	}
4233	check_malloced_chunk (av, victim, nb);
4234	void *p = chunk2mem (victim);
4235	alloc_perturb (p, bytes);
4236	return p;
4237	}
4238	}
4239
4240	use_top:
4241	/*
4242	If large enough, split off the chunk bordering the end of memory
4243	(held in av->top). Note that this is in accord with the best-fit
4244	search rule. In effect, av->top is treated as larger (and thus
4245	less well fitting) than any other available chunk since it can
4246	be extended to be as large as necessary (up to system
4247	limitations).
4248
4249	We require that av->top always exists (i.e., has size >=
4250	MINSIZE) after initialization, so if it would otherwise be
4251	exhausted by current request, it is replenished. (The main
4252	reason for ensuring it exists is that we may need MINSIZE space
4253	to put in fenceposts in sysmalloc.)
4254	*/
4255
4256	victim = av->top;
4257	size = chunksize (victim);
4258
4259	if (__glibc_unlikely (size > av->system_mem))
4260	malloc_printerr ("malloc(): corrupted top size");
4261
4262	if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE))
4263	{
4264	remainder_size = size - nb;
4265	remainder = chunk_at_offset (victim, nb);
4266	av->top = remainder;
4267	set_head (victim, nb \| PREV_INUSE \|
4268	(av != &main_arena ? NON_MAIN_ARENA : `0`));
4269	set_head (remainder, remainder_size \| PREV_INUSE);
4270
4271	check_malloced_chunk (av, victim, nb);
4272	void *p = chunk2mem (victim);
4273	alloc_perturb (p, bytes);
4274	return p;
4275	}
4276
4277	/ When we are using atomic ops to free fast chunks we can get*
4278	here for all block sizes. /*
4279	else if (atomic_load_relaxed (&av->have_fastchunks))
4280	{
4281	malloc_consolidate (av);
4282	/ restore original bin index /
4283	if (in_smallbin_range (nb))
4284	idx = smallbin_index (nb);
4285	else
4286	idx = largebin_index (nb);
4287	}
4288
4289	/*
4290	Otherwise, relay to handle system-dependent cases
4291	*/
4292	else
4293	{
4294	void *p = sysmalloc (nb, av);
4295	if (p != NULL)
4296	alloc_perturb (p, bytes);
4297	return p;
4298	}
4299	}
4300	}
4301
4302	/*
4303	------------------------------ free ------------------------------
4304	*/
4305
4306	static void
4307	_int_free (mstate av, mchunkptr p, int have_lock)
4308	{
4309	INTERNAL_SIZE_T size; / its size /
4310	mfastbinptr fb; /* associated fastbin /
4311	mchunkptr nextchunk; / next contiguous chunk /
4312	INTERNAL_SIZE_T nextsize; / its size /
4313	int nextinuse; / true if nextchunk is used /
4314	INTERNAL_SIZE_T prevsize; / size of previous contiguous chunk /
4315	mchunkptr bck; / misc temp for linking /
4316	mchunkptr fwd; / misc temp for linking /
4317
4318	size = chunksize (p);
4319
4320	/ Little security check which won't hurt performance: the*
4321	allocator never wrapps around at the end of the address space.
4322	Therefore we can exclude some size values which might appear
4323	here by accident or by "design" from some intruder. /*
4324	if (__builtin_expect ((uintptr_t) p > (uintptr_t) -size, `0`)
4325	\|\| __builtin_expect (misaligned_chunk (p), `0`))
4326	malloc_printerr ("free(): invalid pointer");
4327	/ We know that each chunk is at least MINSIZE bytes in size or a*
4328	multiple of MALLOC_ALIGNMENT. /*
4329	if (__glibc_unlikely (size < MINSIZE \|\| !aligned_OK (size)))
4330	malloc_printerr ("free(): invalid size");
4331
4332	check_inuse_chunk(av, p);
4333
4334	#if USE_TCACHE
4335	{
4336	size_t tc_idx = csize2tidx (size);
4337	if (tcache != NULL && tc_idx < mp_.tcache_bins)
4338	{
4339	/ Check to see if it's already in the tcache. /
4340	tcache_entry e = (tcache_entry ) chunk2mem (p);
4341
4342	/ This test succeeds on double free. However, we don't 100%*
4343	trust it (it also matches random payload data at a 1 in
4344	2^<size_t> chance), so verify it's not an unlikely
4345	coincidence before aborting. /*
4346	if (__glibc_unlikely (e->key == tcache_key))
4347	{
4348	tcache_entry *tmp;
4349	size_t cnt = `0`;
4350	LIBC_PROBE (memory_tcache_double_free, `2`, e, tc_idx);
4351	for (tmp = tcache->entries[tc_idx];
4352	tmp;
4353	tmp = REVEAL_PTR (tmp->next), ++cnt)
4354	{
4355	if (cnt >= mp_.tcache_count)
4356	malloc_printerr ("free(): too many chunks detected in tcache");
4357	if (__glibc_unlikely (!aligned_OK (tmp)))
4358	malloc_printerr ("free(): unaligned chunk detected in tcache 2");
4359	if (tmp == e)
4360	malloc_printerr ("free(): double free detected in tcache 2");
4361	/ If we get here, it was a coincidence. We've wasted a*
4362	few cycles, but don't abort. /*
4363	}
4364	}
4365
4366	if (tcache->counts[tc_idx] < mp_.tcache_count)
4367	{
4368	tcache_put (p, tc_idx);
4369	return;
4370	}
4371	}
4372	}
4373	#endif
4374
4375	/*
4376	If eligible, place chunk on a fastbin so it can be found
4377	and used quickly in malloc.
4378	*/
4379
4380	if ((unsigned long)(size) <= (unsigned long)(get_max_fast ())
4381
4382	#if TRIM_FASTBINS
4383	/*
4384	If TRIM_FASTBINS set, don't place chunks
4385	bordering top into fastbins
4386	*/
4387	&& (chunk_at_offset(p, size) != av->top)
4388	#endif
4389	) {
4390
4391	if (__builtin_expect (chunksize_nomask (chunk_at_offset (p, size))
4392	<= CHUNK_HDR_SZ, `0`)
4393	\|\| __builtin_expect (chunksize (chunk_at_offset (p, size))
4394	>= av->system_mem, `0`))
4395	{
4396	bool fail = true;
4397	/ We might not have a lock at this point and concurrent modifications*
4398	of system_mem might result in a false positive. Redo the test after
4399	getting the lock. /*
4400	if (!have_lock)
4401	{
4402	__libc_lock_lock (av->mutex);
4403	fail = (chunksize_nomask (chunk_at_offset (p, size)) <= CHUNK_HDR_SZ
4404	\|\| chunksize (chunk_at_offset (p, size)) >= av->system_mem);
4405	__libc_lock_unlock (av->mutex);
4406	}
4407
4408	if (fail)
4409	malloc_printerr ("free(): invalid next size (fast)");
4410	}
4411
4412	free_perturb (chunk2mem(p), size - CHUNK_HDR_SZ);
4413
4414	atomic_store_relaxed (&av->have_fastchunks, true);
4415	unsigned int idx = fastbin_index(size);
4416	fb = &fastbin (av, idx);
4417
4418	/ Atomically link P to its fastbin: P->FD = FB; FB = P; /
4419	mchunkptr old = *fb, old2;
4420
4421	if (SINGLE_THREAD_P)
4422	{
4423	/ Check that the top of the bin is not the record we are going to*
4424	add (i.e., double free). /*
4425	if (__builtin_expect (old == p, `0`))
4426	malloc_printerr ("double free or corruption (fasttop)");
4427	p->fd = PROTECT_PTR (&p->fd, old);
4428	*fb = p;
4429	}
4430	else
4431	do
4432	{
4433	/ Check that the top of the bin is not the record we are going to*
4434	add (i.e., double free). /*
4435	if (__builtin_expect (old == p, `0`))
4436	malloc_printerr ("double free or corruption (fasttop)");
4437	old2 = old;
4438	p->fd = PROTECT_PTR (&p->fd, old);
4439	}
4440	while ((old = catomic_compare_and_exchange_val_rel (fb, p, old2))
4441	!= old2);
4442
4443	/ Check that size of fastbin chunk at the top is the same as*
4444	size of the chunk that we are adding. We can dereference OLD
4445	only if we have the lock, otherwise it might have already been
4446	allocated again. /*
4447	if (have_lock && old != NULL
4448	&& __builtin_expect (fastbin_index (chunksize (old)) != idx, `0`))
4449	malloc_printerr ("invalid fastbin entry (free)");
4450	}
4451
4452	/*
4453	Consolidate other non-mmapped chunks as they arrive.
4454	*/
4455
4456	else if (!chunk_is_mmapped(p)) {
4457
4458	/ If we're single-threaded, don't lock the arena. /
4459	if (SINGLE_THREAD_P)
4460	have_lock = true;
4461
4462	if (!have_lock)
4463	__libc_lock_lock (av->mutex);
4464
4465	nextchunk = chunk_at_offset(p, size);
4466
4467	/ Lightweight tests: check whether the block is already the*
4468	top block. /*
4469	if (__glibc_unlikely (p == av->top))
4470	malloc_printerr ("double free or corruption (top)");
4471	/ Or whether the next chunk is beyond the boundaries of the arena. /
4472	if (__builtin_expect (contiguous (av)
4473	&& (char *) nextchunk
4474	>= ((char *) av->top + chunksize(av->top)), `0`))
4475	malloc_printerr ("double free or corruption (out)");
4476	/ Or whether the block is actually not marked used. /
4477	if (__glibc_unlikely (!prev_inuse(nextchunk)))
4478	malloc_printerr ("double free or corruption (!prev)");
4479
4480	nextsize = chunksize(nextchunk);
4481	if (__builtin_expect (chunksize_nomask (nextchunk) <= CHUNK_HDR_SZ, `0`)
4482	\|\| __builtin_expect (nextsize >= av->system_mem, `0`))
4483	malloc_printerr ("free(): invalid next size (normal)");
4484
4485	free_perturb (chunk2mem(p), size - CHUNK_HDR_SZ);
4486
4487	/ consolidate backward /
4488	if (!prev_inuse(p)) {
4489	prevsize = prev_size (p);
4490	size += prevsize;
4491	p = chunk_at_offset(p, -((long) prevsize));
4492	if (__glibc_unlikely (chunksize(p) != prevsize))
4493	malloc_printerr ("corrupted size vs. prev_size while consolidating");
4494	unlink_chunk (av, p);
4495	}
4496
4497	if (nextchunk != av->top) {
4498	/ get and clear inuse bit /
4499	nextinuse = inuse_bit_at_offset(nextchunk, nextsize);
4500
4501	/ consolidate forward /
4502	if (!nextinuse) {
4503	unlink_chunk (av, nextchunk);
4504	size += nextsize;
4505	} else
4506	clear_inuse_bit_at_offset(nextchunk, `0`);
4507
4508	/*
4509	Place the chunk in unsorted chunk list. Chunks are
4510	not placed into regular bins until after they have
4511	been given one chance to be used in malloc.
4512	*/
4513
4514	bck = unsorted_chunks(av);
4515	fwd = bck->fd;
4516	if (__glibc_unlikely (fwd->bk != bck))
4517	malloc_printerr ("free(): corrupted unsorted chunks");
4518	p->fd = fwd;
4519	p->bk = bck;
4520	if (!in_smallbin_range(size))
4521	{
4522	p->fd_nextsize = NULL;
4523	p->bk_nextsize = NULL;
4524	}
4525	bck->fd = p;
4526	fwd->bk = p;
4527
4528	set_head(p, size \| PREV_INUSE);
4529	set_foot(p, size);
4530
4531	check_free_chunk(av, p);
4532	}
4533
4534	/*
4535	If the chunk borders the current high end of memory,
4536	consolidate into top
4537	*/
4538
4539	else {
4540	size += nextsize;
4541	set_head(p, size \| PREV_INUSE);
4542	av->top = p;
4543	check_chunk(av, p);
4544	}
4545
4546	/*
4547	If freeing a large space, consolidate possibly-surrounding
4548	chunks. Then, if the total unused topmost memory exceeds trim
4549	threshold, ask malloc_trim to reduce top.
4550
4551	Unless max_fast is 0, we don't know if there are fastbins
4552	bordering top, so we cannot tell for sure whether threshold
4553	has been reached unless fastbins are consolidated. But we
4554	don't want to consolidate on each free. As a compromise,
4555	consolidation is performed if FASTBIN_CONSOLIDATION_THRESHOLD
4556	is reached.
4557	*/
4558
4559	if ((unsigned long)(size) >= FASTBIN_CONSOLIDATION_THRESHOLD) {
4560	if (atomic_load_relaxed (&av->have_fastchunks))
4561	malloc_consolidate(av);
4562
4563	if (av == &main_arena) {
4564	#ifndef MORECORE_CANNOT_TRIM
4565	if ((unsigned long)(chunksize(av->top)) >=
4566	(unsigned long)(mp_.trim_threshold))
4567	systrim(mp_.top_pad, av);
4568	#endif
4569	} else {
4570	/ Always try heap_trim(), even if the top chunk is not*
4571	large, because the corresponding heap might go away. /*
4572	heap_info *heap = heap_for_ptr(top(av));
4573
4574	assert(heap->ar_ptr == av);
4575	heap_trim(heap, mp_.top_pad);
4576	}
4577	}
4578
4579	if (!have_lock)
4580	__libc_lock_unlock (av->mutex);
4581	}
4582	/*
4583	If the chunk was allocated via mmap, release via munmap().
4584	*/
4585
4586	else {
4587	munmap_chunk (p);
4588	}
4589	}
4590
4591	/*
4592	------------------------- malloc_consolidate -------------------------
4593
4594	malloc_consolidate is a specialized version of free() that tears
4595	down chunks held in fastbins. Free itself cannot be used for this
4596	purpose since, among other things, it might place chunks back onto
4597	fastbins. So, instead, we need to use a minor variant of the same
4598	code.
4599	*/
4600
4601	static void malloc_consolidate(mstate av)
4602	{
4603	mfastbinptr* fb; / current fastbin being consolidated /
4604	mfastbinptr* maxfb; / last fastbin (for loop control) /
4605	mchunkptr p; / current chunk being consolidated /
4606	mchunkptr nextp; / next chunk to consolidate /
4607	mchunkptr unsorted_bin; / bin header /
4608	mchunkptr first_unsorted; / chunk to link to /
4609
4610	/ These have same use as in free() /
4611	mchunkptr nextchunk;
4612	INTERNAL_SIZE_T size;
4613	INTERNAL_SIZE_T nextsize;
4614	INTERNAL_SIZE_T prevsize;
4615	int nextinuse;
4616
4617	atomic_store_relaxed (&av->have_fastchunks, false);
4618
4619	unsorted_bin = unsorted_chunks(av);
4620
4621	/*
4622	Remove each chunk from fast bin and consolidate it, placing it
4623	then in unsorted bin. Among other reasons for doing this,
4624	placing in unsorted bin avoids needing to calculate actual bins
4625	until malloc is sure that chunks aren't immediately going to be
4626	reused anyway.
4627	*/
4628
4629	maxfb = &fastbin (av, NFASTBINS - `1`);
4630	fb = &fastbin (av, `0`);
4631	do {
4632	p = atomic_exchange_acq (fb, NULL);
4633	if (p != `0`) {
4634	do {
4635	{
4636	if (__glibc_unlikely (misaligned_chunk (p)))
4637	malloc_printerr ("malloc_consolidate(): "
4638	"unaligned fastbin chunk detected");
4639
4640	unsigned int idx = fastbin_index (chunksize (p));
4641	if ((&fastbin (av, idx)) != fb)
4642	malloc_printerr ("malloc_consolidate(): invalid chunk size");
4643	}
4644
4645	check_inuse_chunk(av, p);
4646	nextp = REVEAL_PTR (p->fd);
4647
4648	/ Slightly streamlined version of consolidation code in free() /
4649	size = chunksize (p);
4650	nextchunk = chunk_at_offset(p, size);
4651	nextsize = chunksize(nextchunk);
4652
4653	if (!prev_inuse(p)) {
4654	prevsize = prev_size (p);
4655	size += prevsize;
4656	p = chunk_at_offset(p, -((long) prevsize));
4657	if (__glibc_unlikely (chunksize(p) != prevsize))
4658	malloc_printerr ("corrupted size vs. prev_size in fastbins");
4659	unlink_chunk (av, p);
4660	}
4661
4662	if (nextchunk != av->top) {
4663	nextinuse = inuse_bit_at_offset(nextchunk, nextsize);
4664
4665	if (!nextinuse) {
4666	size += nextsize;
4667	unlink_chunk (av, nextchunk);
4668	} else
4669	clear_inuse_bit_at_offset(nextchunk, `0`);
4670
4671	first_unsorted = unsorted_bin->fd;
4672	unsorted_bin->fd = p;
4673	first_unsorted->bk = p;
4674
4675	if (!in_smallbin_range (size)) {
4676	p->fd_nextsize = NULL;
4677	p->bk_nextsize = NULL;
4678	}
4679
4680	set_head(p, size \| PREV_INUSE);
4681	p->bk = unsorted_bin;
4682	p->fd = first_unsorted;
4683	set_foot(p, size);
4684	}
4685
4686	else {
4687	size += nextsize;
4688	set_head(p, size \| PREV_INUSE);
4689	av->top = p;
4690	}
4691
4692	} while ( (p = nextp) != `0`);
4693
4694	}
4695	} while (fb++ != maxfb);
4696	}
4697
4698	/*
4699	------------------------------ realloc ------------------------------
4700	*/
4701
4702	static void *
4703	_int_realloc (mstate av, mchunkptr oldp, INTERNAL_SIZE_T oldsize,
4704	INTERNAL_SIZE_T nb)
4705	{
4706	mchunkptr newp; / chunk to return /
4707	INTERNAL_SIZE_T newsize; / its size /
4708	void* newmem; / corresponding user mem /
4709
4710	mchunkptr next; / next contiguous chunk after oldp /
4711
4712	mchunkptr remainder; / extra space at end of newp /
4713	unsigned long remainder_size; / its size /
4714
4715	/ oldmem size /
4716	if (__builtin_expect (chunksize_nomask (oldp) <= CHUNK_HDR_SZ, `0`)
4717	\|\| __builtin_expect (oldsize >= av->system_mem, `0`))
4718	malloc_printerr ("realloc(): invalid old size");
4719
4720	check_inuse_chunk (av, oldp);
4721
4722	/ All callers already filter out mmap'ed chunks. /
4723	assert (!chunk_is_mmapped (oldp));
4724
4725	next = chunk_at_offset (oldp, oldsize);
4726	INTERNAL_SIZE_T nextsize = chunksize (next);
4727	if (__builtin_expect (chunksize_nomask (next) <= CHUNK_HDR_SZ, `0`)
4728	\|\| __builtin_expect (nextsize >= av->system_mem, `0`))
4729	malloc_printerr ("realloc(): invalid next size");
4730
4731	if ((unsigned long) (oldsize) >= (unsigned long) (nb))
4732	{
4733	/ already big enough; split below /
4734	newp = oldp;
4735	newsize = oldsize;
4736	}
4737
4738	else
4739	{
4740	/ Try to expand forward into top /
4741	if (next == av->top &&
4742	(unsigned long) (newsize = oldsize + nextsize) >=
4743	(unsigned long) (nb + MINSIZE))
4744	{
4745	set_head_size (oldp, nb \| (av != &main_arena ? NON_MAIN_ARENA : `0`));
4746	av->top = chunk_at_offset (oldp, nb);
4747	set_head (av->top, (newsize - nb) \| PREV_INUSE);
4748	check_inuse_chunk (av, oldp);
4749	return tag_new_usable (chunk2mem (oldp));
4750	}
4751
4752	/ Try to expand forward into next chunk; split off remainder below /
4753	else if (next != av->top &&
4754	!inuse (next) &&
4755	(unsigned long) (newsize = oldsize + nextsize) >=
4756	(unsigned long) (nb))
4757	{
4758	newp = oldp;
4759	unlink_chunk (av, next);
4760	}
4761
4762	/ allocate, copy, free /
4763	else
4764	{
4765	newmem = _int_malloc (av, nb - MALLOC_ALIGN_MASK);
4766	if (newmem == `0`)
4767	return `0`; / propagate failure /
4768
4769	newp = mem2chunk (newmem);
4770	newsize = chunksize (newp);
4771
4772	/*
4773	Avoid copy if newp is next chunk after oldp.
4774	*/
4775	if (newp == next)
4776	{
4777	newsize += oldsize;
4778	newp = oldp;
4779	}
4780	else
4781	{
4782	void *oldmem = chunk2mem (oldp);
4783	size_t sz = memsize (oldp);
4784	(void) tag_region (oldmem, sz);
4785	newmem = tag_new_usable (newmem);
4786	memcpy (newmem, oldmem, sz);
4787	_int_free (av, oldp, `1`);
4788	check_inuse_chunk (av, newp);
4789	return newmem;
4790	}
4791	}
4792	}
4793
4794	/ If possible, free extra space in old or extended chunk /
4795
4796	assert ((unsigned long) (newsize) >= (unsigned long) (nb));
4797
4798	remainder_size = newsize - nb;
4799
4800	if (remainder_size < MINSIZE) / not enough extra to split off /
4801	{
4802	set_head_size (newp, newsize \| (av != &main_arena ? NON_MAIN_ARENA : `0`));
4803	set_inuse_bit_at_offset (newp, newsize);
4804	}
4805	else / split remainder /
4806	{
4807	remainder = chunk_at_offset (newp, nb);
4808	/ Clear any user-space tags before writing the header. /
4809	remainder = tag_region (remainder, remainder_size);
4810	set_head_size (newp, nb \| (av != &main_arena ? NON_MAIN_ARENA : `0`));
4811	set_head (remainder, remainder_size \| PREV_INUSE \|
4812	(av != &main_arena ? NON_MAIN_ARENA : `0`));
4813	/ Mark remainder as inuse so free() won't complain /
4814	set_inuse_bit_at_offset (remainder, remainder_size);
4815	_int_free (av, remainder, `1`);
4816	}
4817
4818	check_inuse_chunk (av, newp);
4819	return tag_new_usable (chunk2mem (newp));
4820	}
4821
4822	/*
4823	------------------------------ memalign ------------------------------
4824	*/
4825
4826	static void *
4827	_int_memalign (mstate av, size_t alignment, size_t bytes)
4828	{
4829	INTERNAL_SIZE_T nb; / padded request size /
4830	char m; /* memory returned by malloc call /
4831	mchunkptr p; / corresponding chunk /
4832	char brk; /* alignment point within p /
4833	mchunkptr newp; / chunk to return /
4834	INTERNAL_SIZE_T newsize; / its size /
4835	INTERNAL_SIZE_T leadsize; / leading space before alignment point /
4836	mchunkptr remainder; / spare room at end to split off /
4837	unsigned long remainder_size; / its size /
4838	INTERNAL_SIZE_T size;
4839
4840
4841
4842	if (!checked_request2size (bytes, &nb))
4843	{
4844	__set_errno (ENOMEM);
4845	return NULL;
4846	}
4847
4848	/*
4849	Strategy: find a spot within that chunk that meets the alignment
4850	request, and then possibly free the leading and trailing space.
4851	*/
4852
4853	/ Call malloc with worst case padding to hit alignment. /
4854
4855	m = (char *) (_int_malloc (av, nb + alignment + MINSIZE));
4856
4857	if (m == `0`)
4858	return `0`; / propagate failure /
4859
4860	p = mem2chunk (m);
4861
4862	if ((((unsigned long) (m)) % alignment) != `0`) / misaligned /
4863
4864	{ /*
4865	Find an aligned spot inside chunk. Since we need to give back
4866	leading space in a chunk of at least MINSIZE, if the first
4867	calculation places us at a spot with less than MINSIZE leader,
4868	we can move to the next aligned spot -- we've allocated enough
4869	total room so that this is always possible.
4870	*/
4871	brk = (char ) mem2chunk (((unsigned* long) (m + alignment - `1`)) &
4872	- ((signed long) alignment));
4873	if ((unsigned long) (brk - (char *) (p)) < MINSIZE)
4874	brk += alignment;
4875
4876	newp = (mchunkptr) brk;
4877	leadsize = brk - (char *) (p);
4878	newsize = chunksize (p) - leadsize;
4879
4880	/ For mmapped chunks, just adjust offset /
4881	if (chunk_is_mmapped (p))
4882	{
4883	set_prev_size (newp, prev_size (p) + leadsize);
4884	set_head (newp, newsize \| IS_MMAPPED);
4885	return chunk2mem (newp);
4886	}
4887
4888	/ Otherwise, give back leader, use the rest /
4889	set_head (newp, newsize \| PREV_INUSE \|
4890	(av != &main_arena ? NON_MAIN_ARENA : `0`));
4891	set_inuse_bit_at_offset (newp, newsize);
4892	set_head_size (p, leadsize \| (av != &main_arena ? NON_MAIN_ARENA : `0`));
4893	_int_free (av, p, `1`);
4894	p = newp;
4895
4896	assert (newsize >= nb &&
4897	(((unsigned long) (chunk2mem (p))) % alignment) == `0`);
4898	}
4899
4900	/ Also give back spare room at the end /
4901	if (!chunk_is_mmapped (p))
4902	{
4903	size = chunksize (p);
4904	if ((unsigned long) (size) > (unsigned long) (nb + MINSIZE))
4905	{
4906	remainder_size = size - nb;
4907	remainder = chunk_at_offset (p, nb);
4908	set_head (remainder, remainder_size \| PREV_INUSE \|
4909	(av != &main_arena ? NON_MAIN_ARENA : `0`));
4910	set_head_size (p, nb);
4911	_int_free (av, remainder, `1`);
4912	}
4913	}
4914
4915	check_inuse_chunk (av, p);
4916	return chunk2mem (p);
4917	}
4918
4919
4920	/*
4921	------------------------------ malloc_trim ------------------------------
4922	*/
4923
4924	static int
4925	mtrim (mstate av, size_t pad)
4926	{
4927	/ Ensure all blocks are consolidated. /
4928	malloc_consolidate (av);
4929
4930	const size_t ps = GLRO (dl_pagesize);
4931	int psindex = bin_index (ps);
4932	const size_t psm1 = ps - `1`;
4933
4934	int result = `0`;
4935	for (int i = `1`; i < NBINS; ++i)
4936	if (i == `1` \|\| i >= psindex)
4937	{
4938	mbinptr bin = bin_at (av, i);
4939
4940	for (mchunkptr p = last (bin); p != bin; p = p->bk)
4941	{
4942	INTERNAL_SIZE_T size = chunksize (p);
4943
4944	if (size > psm1 + sizeof (struct malloc_chunk))
4945	{
4946	/ See whether the chunk contains at least one unused page. /
4947	char paligned_mem = (char* *) (((uintptr_t) p
4948	+ sizeof (struct malloc_chunk)
4949	+ psm1) & ~psm1);
4950
4951	assert ((char ) chunk2mem (p) + `2` CHUNK_HDR_SZ
4952	<= paligned_mem);
4953	assert ((char *) p + size > paligned_mem);
4954
4955	/ This is the size we could potentially free. /
4956	size -= paligned_mem - (char *) p;
4957
4958	if (size > psm1)
4959	{
4960	#if MALLOC_DEBUG
4961	/ When debugging we simulate destroying the memory*
4962	content. /*
4963	memset (paligned_mem, `0x89`, size & ~psm1);
4964	#endif
4965	__madvise (paligned_mem, size & ~psm1, MADV_DONTNEED);
4966
4967	result = `1`;
4968	}
4969	}
4970	}
4971	}
4972
4973	#ifndef MORECORE_CANNOT_TRIM
4974	return result \| (av == &main_arena ? systrim (pad, av) : `0`);
4975
4976	#else
4977	return result;
4978	#endif
4979	}
4980
4981
4982	int
4983	__malloc_trim (size_t s)
4984	{
4985	int result = `0`;
4986
4987	if (!__malloc_initialized)
4988	ptmalloc_init ();
4989
4990	mstate ar_ptr = &main_arena;
4991	do
4992	{
4993	__libc_lock_lock (ar_ptr->mutex);
4994	result \|= mtrim (ar_ptr, s);
4995	__libc_lock_unlock (ar_ptr->mutex);
4996
4997	ar_ptr = ar_ptr->next;
4998	}
4999	while (ar_ptr != &main_arena);
5000
5001	return result;
5002	}
5003
5004
5005	/*
5006	------------------------- malloc_usable_size -------------------------
5007	*/
5008
5009	static size_t
5010	musable (void *mem)
5011	{
5012	mchunkptr p;
5013	if (mem != `0`)
5014	{
5015	size_t result = `0`;
5016
5017	p = mem2chunk (mem);
5018
5019	if (chunk_is_mmapped (p))
5020	result = chunksize (p) - CHUNK_HDR_SZ;
5021	else if (inuse (p))
5022	result = memsize (p);
5023
5024	return result;
5025	}
5026	return `0`;
5027	}
5028
5029	#if IS_IN (libc)
5030	size_t
5031	__malloc_usable_size (void *m)
5032	{
5033	size_t result;
5034
5035	result = musable (m);
5036	return result;
5037	}
5038	#endif
5039
5040	/*
5041	------------------------------ mallinfo ------------------------------
5042	Accumulate malloc statistics for arena AV into M.
5043	*/
5044	static void
5045	int_mallinfo (mstate av, struct mallinfo2 *m)
5046	{
5047	size_t i;
5048	mbinptr b;
5049	mchunkptr p;
5050	INTERNAL_SIZE_T avail;
5051	INTERNAL_SIZE_T fastavail;
5052	int nblocks;
5053	int nfastblocks;
5054
5055	check_malloc_state (av);
5056
5057	/ Account for top /
5058	avail = chunksize (av->top);
5059	nblocks = `1`; / top always exists /
5060
5061	/ traverse fastbins /
5062	nfastblocks = `0`;
5063	fastavail = `0`;
5064
5065	for (i = `0`; i < NFASTBINS; ++i)
5066	{
5067	for (p = fastbin (av, i);
5068	p != `0`;
5069	p = REVEAL_PTR (p->fd))
5070	{
5071	if (__glibc_unlikely (misaligned_chunk (p)))
5072	malloc_printerr ("int_mallinfo(): "
5073	"unaligned fastbin chunk detected");
5074	++nfastblocks;
5075	fastavail += chunksize (p);
5076	}
5077	}
5078
5079	avail += fastavail;
5080
5081	/ traverse regular bins /
5082	for (i = `1`; i < NBINS; ++i)
5083	{
5084	b = bin_at (av, i);
5085	for (p = last (b); p != b; p = p->bk)
5086	{
5087	++nblocks;
5088	avail += chunksize (p);
5089	}
5090	}
5091
5092	m->smblks += nfastblocks;
5093	m->ordblks += nblocks;
5094	m->fordblks += avail;
5095	m->uordblks += av->system_mem - avail;
5096	m->arena += av->system_mem;
5097	m->fsmblks += fastavail;
5098	if (av == &main_arena)
5099	{
5100	m->hblks = mp_.n_mmaps;
5101	m->hblkhd = mp_.mmapped_mem;
5102	m->usmblks = `0`;
5103	m->keepcost = chunksize (av->top);
5104	}
5105	}
5106
5107
5108	struct mallinfo2
5109	__libc_mallinfo2 (void)
5110	{
5111	struct mallinfo2 m;
5112	mstate ar_ptr;
5113
5114	if (!__malloc_initialized)
5115	ptmalloc_init ();
5116
5117	memset (&m, `0`, sizeof (m));
5118	ar_ptr = &main_arena;
5119	do
5120	{
5121	__libc_lock_lock (ar_ptr->mutex);
5122	int_mallinfo (ar_ptr, &m);
5123	__libc_lock_unlock (ar_ptr->mutex);
5124
5125	ar_ptr = ar_ptr->next;
5126	}
5127	while (ar_ptr != &main_arena);
5128
5129	return m;
5130	}
5131	libc_hidden_def (__libc_mallinfo2)
5132
5133	struct mallinfo
5134	__libc_mallinfo (void)
5135	{
5136	struct mallinfo m;
5137	struct mallinfo2 m2 = __libc_mallinfo2 ();
5138
5139	m.arena = m2.arena;
5140	m.ordblks = m2.ordblks;
5141	m.smblks = m2.smblks;
5142	m.hblks = m2.hblks;
5143	m.hblkhd = m2.hblkhd;
5144	m.usmblks = m2.usmblks;
5145	m.fsmblks = m2.fsmblks;
5146	m.uordblks = m2.uordblks;
5147	m.fordblks = m2.fordblks;
5148	m.keepcost = m2.keepcost;
5149
5150	return m;
5151	}
5152
5153
5154	/*
5155	------------------------------ malloc_stats ------------------------------
5156	*/
5157
5158	void
5159	__malloc_stats (void)
5160	{
5161	int i;
5162	mstate ar_ptr;
5163	unsigned int in_use_b = mp_.mmapped_mem, system_b = in_use_b;
5164
5165	if (!__malloc_initialized)
5166	ptmalloc_init ();
5167	_IO_flockfile (stderr);
5168	int old_flags2 = stderr->_flags2;
5169	stderr->_flags2 \|= _IO_FLAGS2_NOTCANCEL;
5170	for (i = `0`, ar_ptr = &main_arena;; i++)
5171	{
5172	struct mallinfo2 mi;
5173
5174	memset (&mi, `0`, sizeof (mi));
5175	__libc_lock_lock (ar_ptr->mutex);
5176	int_mallinfo (ar_ptr, &mi);
5177	fprintf (stderr, "Arena %d:\n", i);
5178	fprintf (stderr, "system bytes = %10u\n", (unsigned int) mi.arena);
5179	fprintf (stderr, "in use bytes = %10u\n", (unsigned int) mi.uordblks);
5180	#if MALLOC_DEBUG > 1
5181	if (i > `0`)
5182	dump_heap (heap_for_ptr (top (ar_ptr)));
5183	#endif
5184	system_b += mi.arena;
5185	in_use_b += mi.uordblks;
5186	__libc_lock_unlock (ar_ptr->mutex);
5187	ar_ptr = ar_ptr->next;
5188	if (ar_ptr == &main_arena)
5189	break;
5190	}
5191	fprintf (stderr, "Total (incl. mmap):\n");
5192	fprintf (stderr, "system bytes = %10u\n", system_b);
5193	fprintf (stderr, "in use bytes = %10u\n", in_use_b);
5194	fprintf (stderr, "max mmap regions = %10u\n", (unsigned int) mp_.max_n_mmaps);
5195	fprintf (stderr, "max mmap bytes = %10lu\n",
5196	(unsigned long) mp_.max_mmapped_mem);
5197	stderr->_flags2 = old_flags2;
5198	_IO_funlockfile (stderr);
5199	}
5200
5201
5202	/*
5203	------------------------------ mallopt ------------------------------
5204	*/
5205	static __always_inline int
5206	do_set_trim_threshold (size_t value)
5207	{
5208	LIBC_PROBE (memory_mallopt_trim_threshold, `3`, value, mp_.trim_threshold,
5209	mp_.no_dyn_threshold);
5210	mp_.trim_threshold = value;
5211	mp_.no_dyn_threshold = `1`;
5212	return `1`;
5213	}
5214
5215	static __always_inline int
5216	do_set_top_pad (size_t value)
5217	{
5218	LIBC_PROBE (memory_mallopt_top_pad, `3`, value, mp_.top_pad,
5219	mp_.no_dyn_threshold);
5220	mp_.top_pad = value;
5221	mp_.no_dyn_threshold = `1`;
5222	return `1`;
5223	}
5224
5225	static __always_inline int
5226	do_set_mmap_threshold (size_t value)
5227	{
5228	/ Forbid setting the threshold too high. /
5229	if (value <= HEAP_MAX_SIZE / `2`)
5230	{
5231	LIBC_PROBE (memory_mallopt_mmap_threshold, `3`, value, mp_.mmap_threshold,
5232	mp_.no_dyn_threshold);
5233	mp_.mmap_threshold = value;
5234	mp_.no_dyn_threshold = `1`;
5235	return `1`;
5236	}
5237	return `0`;
5238	}
5239
5240	static __always_inline int
5241	do_set_mmaps_max (int32_t value)
5242	{
5243	LIBC_PROBE (memory_mallopt_mmap_max, `3`, value, mp_.n_mmaps_max,
5244	mp_.no_dyn_threshold);
5245	mp_.n_mmaps_max = value;
5246	mp_.no_dyn_threshold = `1`;
5247	return `1`;
5248	}
5249
5250	static __always_inline int
5251	do_set_mallopt_check (int32_t value)
5252	{
5253	return `1`;
5254	}
5255
5256	static __always_inline int
5257	do_set_perturb_byte (int32_t value)
5258	{
5259	LIBC_PROBE (memory_mallopt_perturb, `2`, value, perturb_byte);
5260	perturb_byte = value;
5261	return `1`;
5262	}
5263
5264	static __always_inline int
5265	do_set_arena_test (size_t value)
5266	{
5267	LIBC_PROBE (memory_mallopt_arena_test, `2`, value, mp_.arena_test);
5268	mp_.arena_test = value;
5269	return `1`;
5270	}
5271
5272	static __always_inline int
5273	do_set_arena_max (size_t value)
5274	{
5275	LIBC_PROBE (memory_mallopt_arena_max, `2`, value, mp_.arena_max);
5276	mp_.arena_max = value;
5277	return `1`;
5278	}
5279
5280	#if USE_TCACHE
5281	static __always_inline int
5282	do_set_tcache_max (size_t value)
5283	{
5284	if (value <= MAX_TCACHE_SIZE)
5285	{
5286	LIBC_PROBE (memory_tunable_tcache_max_bytes, `2`, value, mp_.tcache_max_bytes);
5287	mp_.tcache_max_bytes = value;
5288	mp_.tcache_bins = csize2tidx (request2size(value)) + `1`;
5289	return `1`;
5290	}
5291	return `0`;
5292	}
5293
5294	static __always_inline int
5295	do_set_tcache_count (size_t value)
5296	{
5297	if (value <= MAX_TCACHE_COUNT)
5298	{
5299	LIBC_PROBE (memory_tunable_tcache_count, `2`, value, mp_.tcache_count);
5300	mp_.tcache_count = value;
5301	return `1`;
5302	}
5303	return `0`;
5304	}
5305
5306	static __always_inline int
5307	do_set_tcache_unsorted_limit (size_t value)
5308	{
5309	LIBC_PROBE (memory_tunable_tcache_unsorted_limit, `2`, value, mp_.tcache_unsorted_limit);
5310	mp_.tcache_unsorted_limit = value;
5311	return `1`;
5312	}
5313	#endif
5314
5315	static inline int
5316	__always_inline
5317	do_set_mxfast (size_t value)
5318	{
5319	if (value <= MAX_FAST_SIZE)
5320	{
5321	LIBC_PROBE (memory_mallopt_mxfast, `2`, value, get_max_fast ());
5322	set_max_fast (value);
5323	return `1`;
5324	}
5325	return `0`;
5326	}
5327
5328	int
5329	__libc_mallopt (int param_number, int value)
5330	{
5331	mstate av = &main_arena;
5332	int res = `1`;
5333
5334	if (!__malloc_initialized)
5335	ptmalloc_init ();
5336	__libc_lock_lock (av->mutex);
5337
5338	LIBC_PROBE (memory_mallopt, `2`, param_number, value);
5339
5340	/ We must consolidate main arena before changing max_fast*
5341	(see definition of set_max_fast). /*
5342	malloc_consolidate (av);
5343
5344	/ Many of these helper functions take a size_t. We do not worry*
5345	about overflow here, because negative int values will wrap to
5346	very large size_t values and the helpers have sufficient range
5347	checking for such conversions. Many of these helpers are also
5348	used by the tunables macros in arena.c. /*
5349
5350	switch (param_number)
5351	{
5352	case M_MXFAST:
5353	res = do_set_mxfast (value);
5354	break;
5355
5356	case M_TRIM_THRESHOLD:
5357	res = do_set_trim_threshold (value);
5358	break;
5359
5360	case M_TOP_PAD:
5361	res = do_set_top_pad (value);
5362	break;
5363
5364	case M_MMAP_THRESHOLD:
5365	res = do_set_mmap_threshold (value);
5366	break;
5367
5368	case M_MMAP_MAX:
5369	res = do_set_mmaps_max (value);
5370	break;
5371
5372	case M_CHECK_ACTION:
5373	res = do_set_mallopt_check (value);
5374	break;
5375
5376	case M_PERTURB:
5377	res = do_set_perturb_byte (value);
5378	break;
5379
5380	case M_ARENA_TEST:
5381	if (value > `0`)
5382	res = do_set_arena_test (value);
5383	break;
5384
5385	case M_ARENA_MAX:
5386	if (value > `0`)
5387	res = do_set_arena_max (value);
5388	break;
5389	}
5390	__libc_lock_unlock (av->mutex);
5391	return res;
5392	}
5393	libc_hidden_def (__libc_mallopt)
5394
5395
5396	/*
5397	-------------------- Alternative MORECORE functions --------------------
5398	*/
5399
5400
5401	/*
5402	General Requirements for MORECORE.
5403
5404	The MORECORE function must have the following properties:
5405
5406	If MORECORE_CONTIGUOUS is false:
5407
5408	* MORECORE must allocate in multiples of pagesize. It will
5409	only be called with arguments that are multiples of pagesize.
5410
5411	* MORECORE(0) must return an address that is at least
5412	MALLOC_ALIGNMENT aligned. (Page-aligning always suffices.)
5413
5414	else (i.e. If MORECORE_CONTIGUOUS is true):
5415
5416	* Consecutive calls to MORECORE with positive arguments
5417	return increasing addresses, indicating that space has been
5418	contiguously extended.
5419
5420	* MORECORE need not allocate in multiples of pagesize.
5421	Calls to MORECORE need not have args of multiples of pagesize.
5422
5423	* MORECORE need not page-align.
5424
5425	In either case:
5426
5427	* MORECORE may allocate more memory than requested. (Or even less,
5428	but this will generally result in a malloc failure.)
5429
5430	* MORECORE must not allocate memory when given argument zero, but
5431	instead return one past the end address of memory from previous
5432	nonzero call. This malloc does NOT call MORECORE(0)
5433	until at least one call with positive arguments is made, so
5434	the initial value returned is not important.
5435
5436	* Even though consecutive calls to MORECORE need not return contiguous
5437	addresses, it must be OK for malloc'ed chunks to span multiple
5438	regions in those cases where they do happen to be contiguous.
5439
5440	* MORECORE need not handle negative arguments -- it may instead
5441	just return MORECORE_FAILURE when given negative arguments.
5442	Negative arguments are always multiples of pagesize. MORECORE
5443	must not misinterpret negative args as large positive unsigned
5444	args. You can suppress all such calls from even occurring by defining
5445	MORECORE_CANNOT_TRIM,
5446
5447	There is some variation across systems about the type of the
5448	argument to sbrk/MORECORE. If size_t is unsigned, then it cannot
5449	actually be size_t, because sbrk supports negative args, so it is
5450	normally the signed type of the same width as size_t (sometimes
5451	declared as "intptr_t", and sometimes "ptrdiff_t"). It doesn't much
5452	matter though. Internally, we use "long" as arguments, which should
5453	work across all reasonable possibilities.
5454
5455	Additionally, if MORECORE ever returns failure for a positive
5456	request, then mmap is used as a noncontiguous system allocator. This
5457	is a useful backup strategy for systems with holes in address spaces
5458	-- in this case sbrk cannot contiguously expand the heap, but mmap
5459	may be able to map noncontiguous space.
5460
5461	If you'd like mmap to ALWAYS be used, you can define MORECORE to be
5462	a function that always returns MORECORE_FAILURE.
5463
5464	If you are using this malloc with something other than sbrk (or its
5465	emulation) to supply memory regions, you probably want to set
5466	MORECORE_CONTIGUOUS as false. As an example, here is a custom
5467	allocator kindly contributed for pre-OSX macOS. It uses virtually
5468	but not necessarily physically contiguous non-paged memory (locked
5469	in, present and won't get swapped out). You can use it by
5470	uncommenting this section, adding some #includes, and setting up the
5471	appropriate defines above:
5472
5473	*#define MORECORE osMoreCore
5474	*#define MORECORE_CONTIGUOUS 0
5475
5476	There is also a shutdown routine that should somehow be called for
5477	cleanup upon program exit.
5478
5479	*#define MAX_POOL_ENTRIES 100
5480	#define MINIMUM_MORECORE_SIZE (64 1024)
5481	static int next_os_pool;
5482	void our_os_pools[MAX_POOL_ENTRIES];*
5483
5484	void osMoreCore(int size)*
5485	{
5486	void ptr = 0;*
5487	static void sbrk_top = 0;*
5488
5489	if (size > 0)
5490	{
5491	if (size < MINIMUM_MORECORE_SIZE)
5492	size = MINIMUM_MORECORE_SIZE;
5493	if (CurrentExecutionLevel() == kTaskLevel)
5494	ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0);
5495	if (ptr == 0)
5496	{
5497	return (void ) MORECORE_FAILURE;*
5498	}
5499	// save ptrs so they can be freed during cleanup
5500	our_os_pools[next_os_pool] = ptr;
5501	next_os_pool++;
5502	ptr = (void ) ((((unsigned long) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK);*
5503	sbrk_top = (char ) ptr + size;*
5504	return ptr;
5505	}
5506	else if (size < 0)
5507	{
5508	// we don't currently support shrink behavior
5509	return (void ) MORECORE_FAILURE;*
5510	}
5511	else
5512	{
5513	return sbrk_top;
5514	}
5515	}
5516
5517	// cleanup any allocated memory pools
5518	// called as last thing before shutting down driver
5519
5520	void osCleanupMem(void)
5521	{
5522	void ptr;
5523
5524	for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++)
5525	if (ptr)*
5526	{
5527	PoolDeallocate(ptr);*
5528	* ptr = 0;
5529	}
5530	}
5531
5532	*/
5533
5534
5535	/ Helper code. /
5536
5537	extern char **__libc_argv attribute_hidden;
5538
5539	static void
5540	malloc_printerr (const char *str)
5541	{
5542	#if IS_IN (libc)
5543	__libc_message (do_abort, "%s\n", str);
5544	#else
5545	__libc_fatal (str);
5546	#endif
5547	__builtin_unreachable ();
5548	}
5549
5550	#if IS_IN (libc)
5551	/ We need a wrapper function for one of the additions of POSIX. /
5552	int
5553	__posix_memalign (void **memptr, size_t alignment, size_t size)
5554	{
5555	void *mem;
5556
5557	if (!__malloc_initialized)
5558	ptmalloc_init ();
5559
5560	/ Test whether the SIZE argument is valid. It must be a power of*
5561	two multiple of sizeof (void ). /
5562	if (alignment % sizeof (void *) != `0`
5563	\|\| !powerof2 (alignment / sizeof (void *))
5564	\|\| alignment == `0`)
5565	return EINVAL;
5566
5567
5568	void *address = RETURN_ADDRESS (`0`);
5569	mem = _mid_memalign (alignment, size, address);
5570
5571	if (mem != NULL)
5572	{
5573	*memptr = mem;
5574	return `0`;
5575	}
5576
5577	return ENOMEM;
5578	}
5579	weak_alias (__posix_memalign, posix_memalign)
5580	#endif
5581
5582
5583	int
5584	__malloc_info (int options, FILE *fp)
5585	{
5586	/ For now, at least. /
5587	if (options != `0`)
5588	return EINVAL;
5589
5590	int n = `0`;
5591	size_t total_nblocks = `0`;
5592	size_t total_nfastblocks = `0`;
5593	size_t total_avail = `0`;
5594	size_t total_fastavail = `0`;
5595	size_t total_system = `0`;
5596	size_t total_max_system = `0`;
5597	size_t total_aspace = `0`;
5598	size_t total_aspace_mprotect = `0`;
5599
5600
5601
5602	if (!__malloc_initialized)
5603	ptmalloc_init ();
5604
5605	fputs ("<malloc version=\"1\">\n", fp);
5606
5607	/ Iterate over all arenas currently in use. /
5608	mstate ar_ptr = &main_arena;
5609	do
5610	{
5611	fprintf (fp, "<heap nr=\"%d\">\n<sizes>\n", n++);
5612
5613	size_t nblocks = `0`;
5614	size_t nfastblocks = `0`;
5615	size_t avail = `0`;
5616	size_t fastavail = `0`;
5617	struct
5618	{
5619	size_t from;
5620	size_t to;
5621	size_t total;
5622	size_t count;
5623	} sizes[NFASTBINS + NBINS - `1`];
5624	#define nsizes (sizeof (sizes) / sizeof (sizes[0]))
5625
5626	__libc_lock_lock (ar_ptr->mutex);
5627
5628	/ Account for top chunk. The top-most available chunk is*
5629	treated specially and is never in any bin. See "initial_top"
5630	comments. /*
5631	avail = chunksize (ar_ptr->top);
5632	nblocks = `1`; / Top always exists. /
5633
5634	for (size_t i = `0`; i < NFASTBINS; ++i)
5635	{
5636	mchunkptr p = fastbin (ar_ptr, i);
5637	if (p != NULL)
5638	{
5639	size_t nthissize = `0`;
5640	size_t thissize = chunksize (p);
5641
5642	while (p != NULL)
5643	{
5644	if (__glibc_unlikely (misaligned_chunk (p)))
5645	malloc_printerr ("__malloc_info(): "
5646	"unaligned fastbin chunk detected");
5647	++nthissize;
5648	p = REVEAL_PTR (p->fd);
5649	}
5650
5651	fastavail += nthissize * thissize;
5652	nfastblocks += nthissize;
5653	sizes[i].from = thissize - (MALLOC_ALIGNMENT - `1`);
5654	sizes[i].to = thissize;
5655	sizes[i].count = nthissize;
5656	}
5657	else
5658	sizes[i].from = sizes[i].to = sizes[i].count = `0`;
5659
5660	sizes[i].total = sizes[i].count * sizes[i].to;
5661	}
5662
5663
5664	mbinptr bin;
5665	struct malloc_chunk *r;
5666
5667	for (size_t i = `1`; i < NBINS; ++i)
5668	{
5669	bin = bin_at (ar_ptr, i);
5670	r = bin->fd;
5671	sizes[NFASTBINS - `1` + i].from = ~((size_t) `0`);
5672	sizes[NFASTBINS - `1` + i].to = sizes[NFASTBINS - `1` + i].total
5673	= sizes[NFASTBINS - `1` + i].count = `0`;
5674
5675	if (r != NULL)
5676	while (r != bin)
5677	{
5678	size_t r_size = chunksize_nomask (r);
5679	++sizes[NFASTBINS - `1` + i].count;
5680	sizes[NFASTBINS - `1` + i].total += r_size;
5681	sizes[NFASTBINS - `1` + i].from
5682	= MIN (sizes[NFASTBINS - `1` + i].from, r_size);
5683	sizes[NFASTBINS - `1` + i].to = MAX (sizes[NFASTBINS - `1` + i].to,
5684	r_size);
5685
5686	r = r->fd;
5687	}
5688
5689	if (sizes[NFASTBINS - `1` + i].count == `0`)
5690	sizes[NFASTBINS - `1` + i].from = `0`;
5691	nblocks += sizes[NFASTBINS - `1` + i].count;
5692	avail += sizes[NFASTBINS - `1` + i].total;
5693	}
5694
5695	size_t heap_size = `0`;
5696	size_t heap_mprotect_size = `0`;
5697	size_t heap_count = `0`;
5698	if (ar_ptr != &main_arena)
5699	{
5700	/ Iterate over the arena heaps from back to front. /
5701	heap_info *heap = heap_for_ptr (top (ar_ptr));
5702	do
5703	{
5704	heap_size += heap->size;
5705	heap_mprotect_size += heap->mprotect_size;
5706	heap = heap->prev;
5707	++heap_count;
5708	}
5709	while (heap != NULL);
5710	}
5711
5712	__libc_lock_unlock (ar_ptr->mutex);
5713
5714	total_nfastblocks += nfastblocks;
5715	total_fastavail += fastavail;
5716
5717	total_nblocks += nblocks;
5718	total_avail += avail;
5719
5720	for (size_t i = `0`; i < nsizes; ++i)
5721	if (sizes[i].count != `0` && i != NFASTBINS)
5722	fprintf (fp, "\
5723	<size from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n",
5724	sizes[i].from, sizes[i].to, sizes[i].total, sizes[i].count);
5725
5726	if (sizes[NFASTBINS].count != `0`)
5727	fprintf (fp, "\
5728	<unsorted from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n",
5729	sizes[NFASTBINS].from, sizes[NFASTBINS].to,
5730	sizes[NFASTBINS].total, sizes[NFASTBINS].count);
5731
5732	total_system += ar_ptr->system_mem;
5733	total_max_system += ar_ptr->max_system_mem;
5734
5735	fprintf (fp,
5736	"</sizes>\n<total type=\"fast\" count=\"%zu\" size=\"%zu\"/>\n"
5737	"<total type=\"rest\" count=\"%zu\" size=\"%zu\"/>\n"
5738	"<system type=\"current\" size=\"%zu\"/>\n"
5739	"<system type=\"max\" size=\"%zu\"/>\n",
5740	nfastblocks, fastavail, nblocks, avail,
5741	ar_ptr->system_mem, ar_ptr->max_system_mem);
5742
5743	if (ar_ptr != &main_arena)
5744	{
5745	fprintf (fp,
5746	"<aspace type=\"total\" size=\"%zu\"/>\n"
5747	"<aspace type=\"mprotect\" size=\"%zu\"/>\n"
5748	"<aspace type=\"subheaps\" size=\"%zu\"/>\n",
5749	heap_size, heap_mprotect_size, heap_count);
5750	total_aspace += heap_size;
5751	total_aspace_mprotect += heap_mprotect_size;
5752	}
5753	else
5754	{
5755	fprintf (fp,
5756	"<aspace type=\"total\" size=\"%zu\"/>\n"
5757	"<aspace type=\"mprotect\" size=\"%zu\"/>\n",
5758	ar_ptr->system_mem, ar_ptr->system_mem);
5759	total_aspace += ar_ptr->system_mem;
5760	total_aspace_mprotect += ar_ptr->system_mem;
5761	}
5762
5763	fputs ("</heap>\n", fp);
5764	ar_ptr = ar_ptr->next;
5765	}
5766	while (ar_ptr != &main_arena);
5767
5768	fprintf (fp,
5769	"<total type=\"fast\" count=\"%zu\" size=\"%zu\"/>\n"
5770	"<total type=\"rest\" count=\"%zu\" size=\"%zu\"/>\n"
5771	"<total type=\"mmap\" count=\"%d\" size=\"%zu\"/>\n"
5772	"<system type=\"current\" size=\"%zu\"/>\n"
5773	"<system type=\"max\" size=\"%zu\"/>\n"
5774	"<aspace type=\"total\" size=\"%zu\"/>\n"
5775	"<aspace type=\"mprotect\" size=\"%zu\"/>\n"
5776	"</malloc>\n",
5777	total_nfastblocks, total_fastavail, total_nblocks, total_avail,
5778	mp_.n_mmaps, mp_.mmapped_mem,
5779	total_system, total_max_system,
5780	total_aspace, total_aspace_mprotect);
5781
5782	return `0`;
5783	}
5784	#if IS_IN (libc)
5785	weak_alias (__malloc_info, malloc_info)
5786
5787	strong_alias (__libc_calloc, __calloc) weak_alias (__libc_calloc, calloc)
5788	strong_alias (__libc_free, __free) strong_alias (__libc_free, free)
5789	strong_alias (__libc_malloc, __malloc) strong_alias (__libc_malloc, malloc)
5790	strong_alias (__libc_memalign, __memalign)
5791	weak_alias (__libc_memalign, memalign)
5792	strong_alias (__libc_realloc, __realloc) strong_alias (__libc_realloc, realloc)
5793	strong_alias (__libc_valloc, __valloc) weak_alias (__libc_valloc, valloc)
5794	strong_alias (__libc_pvalloc, __pvalloc) weak_alias (__libc_pvalloc, pvalloc)
5795	strong_alias (__libc_mallinfo, __mallinfo)
5796	weak_alias (__libc_mallinfo, mallinfo)
5797	strong_alias (__libc_mallinfo2, __mallinfo2)
5798	weak_alias (__libc_mallinfo2, mallinfo2)
5799	strong_alias (__libc_mallopt, __mallopt) weak_alias (__libc_mallopt, mallopt)
5800
5801	weak_alias (__malloc_stats, malloc_stats)
5802	weak_alias (__malloc_usable_size, malloc_usable_size)
5803	weak_alias (__malloc_trim, malloc_trim)
5804	#endif
5805
5806	#if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_26)
5807	compat_symbol (libc, __libc_free, cfree, GLIBC_2_0);
5808	#endif
5809
5810	/ ------------------------------------------------------------*
5811	History:
5812
5813	[see ftp://g.oswego.edu/pub/misc/malloc.c for the history of dlmalloc]
5814
5815	*/
5816	/*
5817	* Local variables:
5818	* c-basic-offset: 2
5819	* End:
5820	*/
5821

Browse the source code of glibc/malloc/malloc.c