1/* Malloc implementation for multiple threads without lock contention.
2 Copyright (C) 1996-2019 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4 Contributed by Wolfram Gloger <wg@malloc.de>
5 and Doug Lea <dl@cs.oswego.edu>, 2001.
6
7 The GNU C Library is free software; you can redistribute it and/or
8 modify it under the terms of the GNU Lesser General Public License as
9 published by the Free Software Foundation; either version 2.1 of the
10 License, or (at your option) any later version.
11
12 The GNU C Library is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
16
17 You should have received a copy of the GNU Lesser General Public
18 License along with the GNU C Library; see the file COPYING.LIB. If
19 not, see <http://www.gnu.org/licenses/>. */
20
21/*
22 This is a version (aka ptmalloc2) of malloc/free/realloc written by
23 Doug Lea and adapted to multiple threads/arenas by Wolfram Gloger.
24
25 There have been substantial changes made after the integration into
26 glibc in all parts of the code. Do not look for much commonality
27 with the ptmalloc2 version.
28
29* Version ptmalloc2-20011215
30 based on:
31 VERSION 2.7.0 Sun Mar 11 14:14:06 2001 Doug Lea (dl at gee)
32
33* Quickstart
34
35 In order to compile this implementation, a Makefile is provided with
36 the ptmalloc2 distribution, which has pre-defined targets for some
37 popular systems (e.g. "make posix" for Posix threads). All that is
38 typically required with regard to compiler flags is the selection of
39 the thread package via defining one out of USE_PTHREADS, USE_THR or
40 USE_SPROC. Check the thread-m.h file for what effects this has.
41 Many/most systems will additionally require USE_TSD_DATA_HACK to be
42 defined, so this is the default for "make posix".
43
44* Why use this malloc?
45
46 This is not the fastest, most space-conserving, most portable, or
47 most tunable malloc ever written. However it is among the fastest
48 while also being among the most space-conserving, portable and tunable.
49 Consistent balance across these factors results in a good general-purpose
50 allocator for malloc-intensive programs.
51
52 The main properties of the algorithms are:
53 * For large (>= 512 bytes) requests, it is a pure best-fit allocator,
54 with ties normally decided via FIFO (i.e. least recently used).
55 * For small (<= 64 bytes by default) requests, it is a caching
56 allocator, that maintains pools of quickly recycled chunks.
57 * In between, and for combinations of large and small requests, it does
58 the best it can trying to meet both goals at once.
59 * For very large requests (>= 128KB by default), it relies on system
60 memory mapping facilities, if supported.
61
62 For a longer but slightly out of date high-level description, see
63 http://gee.cs.oswego.edu/dl/html/malloc.html
64
65 You may already by default be using a C library containing a malloc
66 that is based on some version of this malloc (for example in
67 linux). You might still want to use the one in this file in order to
68 customize settings or to avoid overheads associated with library
69 versions.
70
71* Contents, described in more detail in "description of public routines" below.
72
73 Standard (ANSI/SVID/...) functions:
74 malloc(size_t n);
75 calloc(size_t n_elements, size_t element_size);
76 free(void* p);
77 realloc(void* p, size_t n);
78 memalign(size_t alignment, size_t n);
79 valloc(size_t n);
80 mallinfo()
81 mallopt(int parameter_number, int parameter_value)
82
83 Additional functions:
84 independent_calloc(size_t n_elements, size_t size, void* chunks[]);
85 independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]);
86 pvalloc(size_t n);
87 malloc_trim(size_t pad);
88 malloc_usable_size(void* p);
89 malloc_stats();
90
91* Vital statistics:
92
93 Supported pointer representation: 4 or 8 bytes
94 Supported size_t representation: 4 or 8 bytes
95 Note that size_t is allowed to be 4 bytes even if pointers are 8.
96 You can adjust this by defining INTERNAL_SIZE_T
97
98 Alignment: 2 * sizeof(size_t) (default)
99 (i.e., 8 byte alignment with 4byte size_t). This suffices for
100 nearly all current machines and C compilers. However, you can
101 define MALLOC_ALIGNMENT to be wider than this if necessary.
102
103 Minimum overhead per allocated chunk: 4 or 8 bytes
104 Each malloced chunk has a hidden word of overhead holding size
105 and status information.
106
107 Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead)
108 8-byte ptrs: 24/32 bytes (including, 4/8 overhead)
109
110 When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte
111 ptrs but 4 byte size) or 24 (for 8/8) additional bytes are
112 needed; 4 (8) for a trailing size field and 8 (16) bytes for
113 free list pointers. Thus, the minimum allocatable size is
114 16/24/32 bytes.
115
116 Even a request for zero bytes (i.e., malloc(0)) returns a
117 pointer to something of the minimum allocatable size.
118
119 The maximum overhead wastage (i.e., number of extra bytes
120 allocated than were requested in malloc) is less than or equal
121 to the minimum size, except for requests >= mmap_threshold that
122 are serviced via mmap(), where the worst case wastage is 2 *
123 sizeof(size_t) bytes plus the remainder from a system page (the
124 minimal mmap unit); typically 4096 or 8192 bytes.
125
126 Maximum allocated size: 4-byte size_t: 2^32 minus about two pages
127 8-byte size_t: 2^64 minus about two pages
128
129 It is assumed that (possibly signed) size_t values suffice to
130 represent chunk sizes. `Possibly signed' is due to the fact
131 that `size_t' may be defined on a system as either a signed or
132 an unsigned type. The ISO C standard says that it must be
133 unsigned, but a few systems are known not to adhere to this.
134 Additionally, even when size_t is unsigned, sbrk (which is by
135 default used to obtain memory from system) accepts signed
136 arguments, and may not be able to handle size_t-wide arguments
137 with negative sign bit. Generally, values that would
138 appear as negative after accounting for overhead and alignment
139 are supported only via mmap(), which does not have this
140 limitation.
141
142 Requests for sizes outside the allowed range will perform an optional
143 failure action and then return null. (Requests may also
144 also fail because a system is out of memory.)
145
146 Thread-safety: thread-safe
147
148 Compliance: I believe it is compliant with the 1997 Single Unix Specification
149 Also SVID/XPG, ANSI C, and probably others as well.
150
151* Synopsis of compile-time options:
152
153 People have reported using previous versions of this malloc on all
154 versions of Unix, sometimes by tweaking some of the defines
155 below. It has been tested most extensively on Solaris and Linux.
156 People also report using it in stand-alone embedded systems.
157
158 The implementation is in straight, hand-tuned ANSI C. It is not
159 at all modular. (Sorry!) It uses a lot of macros. To be at all
160 usable, this code should be compiled using an optimizing compiler
161 (for example gcc -O3) that can simplify expressions and control
162 paths. (FAQ: some macros import variables as arguments rather than
163 declare locals because people reported that some debuggers
164 otherwise get confused.)
165
166 OPTION DEFAULT VALUE
167
168 Compilation Environment options:
169
170 HAVE_MREMAP 0
171
172 Changing default word sizes:
173
174 INTERNAL_SIZE_T size_t
175
176 Configuration and functionality options:
177
178 USE_PUBLIC_MALLOC_WRAPPERS NOT defined
179 USE_MALLOC_LOCK NOT defined
180 MALLOC_DEBUG NOT defined
181 REALLOC_ZERO_BYTES_FREES 1
182 TRIM_FASTBINS 0
183
184 Options for customizing MORECORE:
185
186 MORECORE sbrk
187 MORECORE_FAILURE -1
188 MORECORE_CONTIGUOUS 1
189 MORECORE_CANNOT_TRIM NOT defined
190 MORECORE_CLEARS 1
191 MMAP_AS_MORECORE_SIZE (1024 * 1024)
192
193 Tuning options that are also dynamically changeable via mallopt:
194
195 DEFAULT_MXFAST 64 (for 32bit), 128 (for 64bit)
196 DEFAULT_TRIM_THRESHOLD 128 * 1024
197 DEFAULT_TOP_PAD 0
198 DEFAULT_MMAP_THRESHOLD 128 * 1024
199 DEFAULT_MMAP_MAX 65536
200
201 There are several other #defined constants and macros that you
202 probably don't want to touch unless you are extending or adapting malloc. */
203
204/*
205 void* is the pointer type that malloc should say it returns
206*/
207
208#ifndef void
209#define void void
210#endif /*void*/
211
212#include <stddef.h> /* for size_t */
213#include <stdlib.h> /* for getenv(), abort() */
214#include <unistd.h> /* for __libc_enable_secure */
215
216#include <atomic.h>
217#include <_itoa.h>
218#include <bits/wordsize.h>
219#include <sys/sysinfo.h>
220
221#include <ldsodefs.h>
222
223#include <unistd.h>
224#include <stdio.h> /* needed for malloc_stats */
225#include <errno.h>
226#include <assert.h>
227
228#include <shlib-compat.h>
229
230/* For uintptr_t. */
231#include <stdint.h>
232
233/* For va_arg, va_start, va_end. */
234#include <stdarg.h>
235
236/* For MIN, MAX, powerof2. */
237#include <sys/param.h>
238
239/* For ALIGN_UP et. al. */
240#include <libc-pointer-arith.h>
241
242/* For DIAG_PUSH/POP_NEEDS_COMMENT et al. */
243#include <libc-diag.h>
244
245#include <malloc/malloc-internal.h>
246
247/* For SINGLE_THREAD_P. */
248#include <sysdep-cancel.h>
249
250/*
251 Debugging:
252
253 Because freed chunks may be overwritten with bookkeeping fields, this
254 malloc will often die when freed memory is overwritten by user
255 programs. This can be very effective (albeit in an annoying way)
256 in helping track down dangling pointers.
257
258 If you compile with -DMALLOC_DEBUG, a number of assertion checks are
259 enabled that will catch more memory errors. You probably won't be
260 able to make much sense of the actual assertion errors, but they
261 should help you locate incorrectly overwritten memory. The checking
262 is fairly extensive, and will slow down execution
263 noticeably. Calling malloc_stats or mallinfo with MALLOC_DEBUG set
264 will attempt to check every non-mmapped allocated and free chunk in
265 the course of computing the summmaries. (By nature, mmapped regions
266 cannot be checked very much automatically.)
267
268 Setting MALLOC_DEBUG may also be helpful if you are trying to modify
269 this code. The assertions in the check routines spell out in more
270 detail the assumptions and invariants underlying the algorithms.
271
272 Setting MALLOC_DEBUG does NOT provide an automated mechanism for
273 checking that all accesses to malloced memory stay within their
274 bounds. However, there are several add-ons and adaptations of this
275 or other mallocs available that do this.
276*/
277
278#ifndef MALLOC_DEBUG
279#define MALLOC_DEBUG 0
280#endif
281
282#ifndef NDEBUG
283# define __assert_fail(assertion, file, line, function) \
284 __malloc_assert(assertion, file, line, function)
285
286extern const char *__progname;
287
288static void
289__malloc_assert (const char *assertion, const char *file, unsigned int line,
290 const char *function)
291{
292 (void) __fxprintf (NULL, "%s%s%s:%u: %s%sAssertion `%s' failed.\n",
293 __progname, __progname[0] ? ": " : "",
294 file, line,
295 function ? function : "", function ? ": " : "",
296 assertion);
297 fflush (stderr);
298 abort ();
299}
300#endif
301
302#if USE_TCACHE
303/* We want 64 entries. This is an arbitrary limit, which tunables can reduce. */
304# define TCACHE_MAX_BINS 64
305# define MAX_TCACHE_SIZE tidx2usize (TCACHE_MAX_BINS-1)
306
307/* Only used to pre-fill the tunables. */
308# define tidx2usize(idx) (((size_t) idx) * MALLOC_ALIGNMENT + MINSIZE - SIZE_SZ)
309
310/* When "x" is from chunksize(). */
311# define csize2tidx(x) (((x) - MINSIZE + MALLOC_ALIGNMENT - 1) / MALLOC_ALIGNMENT)
312/* When "x" is a user-provided size. */
313# define usize2tidx(x) csize2tidx (request2size (x))
314
315/* With rounding and alignment, the bins are...
316 idx 0 bytes 0..24 (64-bit) or 0..12 (32-bit)
317 idx 1 bytes 25..40 or 13..20
318 idx 2 bytes 41..56 or 21..28
319 etc. */
320
321/* This is another arbitrary limit, which tunables can change. Each
322 tcache bin will hold at most this number of chunks. */
323# define TCACHE_FILL_COUNT 7
324#endif
325
326
327/*
328 REALLOC_ZERO_BYTES_FREES should be set if a call to
329 realloc with zero bytes should be the same as a call to free.
330 This is required by the C standard. Otherwise, since this malloc
331 returns a unique pointer for malloc(0), so does realloc(p, 0).
332*/
333
334#ifndef REALLOC_ZERO_BYTES_FREES
335#define REALLOC_ZERO_BYTES_FREES 1
336#endif
337
338/*
339 TRIM_FASTBINS controls whether free() of a very small chunk can
340 immediately lead to trimming. Setting to true (1) can reduce memory
341 footprint, but will almost always slow down programs that use a lot
342 of small chunks.
343
344 Define this only if you are willing to give up some speed to more
345 aggressively reduce system-level memory footprint when releasing
346 memory in programs that use many small chunks. You can get
347 essentially the same effect by setting MXFAST to 0, but this can
348 lead to even greater slowdowns in programs using many small chunks.
349 TRIM_FASTBINS is an in-between compile-time option, that disables
350 only those chunks bordering topmost memory from being placed in
351 fastbins.
352*/
353
354#ifndef TRIM_FASTBINS
355#define TRIM_FASTBINS 0
356#endif
357
358
359/* Definition for getting more memory from the OS. */
360#define MORECORE (*__morecore)
361#define MORECORE_FAILURE 0
362void * __default_morecore (ptrdiff_t);
363void *(*__morecore)(ptrdiff_t) = __default_morecore;
364
365
366#include <string.h>
367
368/*
369 MORECORE-related declarations. By default, rely on sbrk
370*/
371
372
373/*
374 MORECORE is the name of the routine to call to obtain more memory
375 from the system. See below for general guidance on writing
376 alternative MORECORE functions, as well as a version for WIN32 and a
377 sample version for pre-OSX macos.
378*/
379
380#ifndef MORECORE
381#define MORECORE sbrk
382#endif
383
384/*
385 MORECORE_FAILURE is the value returned upon failure of MORECORE
386 as well as mmap. Since it cannot be an otherwise valid memory address,
387 and must reflect values of standard sys calls, you probably ought not
388 try to redefine it.
389*/
390
391#ifndef MORECORE_FAILURE
392#define MORECORE_FAILURE (-1)
393#endif
394
395/*
396 If MORECORE_CONTIGUOUS is true, take advantage of fact that
397 consecutive calls to MORECORE with positive arguments always return
398 contiguous increasing addresses. This is true of unix sbrk. Even
399 if not defined, when regions happen to be contiguous, malloc will
400 permit allocations spanning regions obtained from different
401 calls. But defining this when applicable enables some stronger
402 consistency checks and space efficiencies.
403*/
404
405#ifndef MORECORE_CONTIGUOUS
406#define MORECORE_CONTIGUOUS 1
407#endif
408
409/*
410 Define MORECORE_CANNOT_TRIM if your version of MORECORE
411 cannot release space back to the system when given negative
412 arguments. This is generally necessary only if you are using
413 a hand-crafted MORECORE function that cannot handle negative arguments.
414*/
415
416/* #define MORECORE_CANNOT_TRIM */
417
418/* MORECORE_CLEARS (default 1)
419 The degree to which the routine mapped to MORECORE zeroes out
420 memory: never (0), only for newly allocated space (1) or always
421 (2). The distinction between (1) and (2) is necessary because on
422 some systems, if the application first decrements and then
423 increments the break value, the contents of the reallocated space
424 are unspecified.
425 */
426
427#ifndef MORECORE_CLEARS
428# define MORECORE_CLEARS 1
429#endif
430
431
432/*
433 MMAP_AS_MORECORE_SIZE is the minimum mmap size argument to use if
434 sbrk fails, and mmap is used as a backup. The value must be a
435 multiple of page size. This backup strategy generally applies only
436 when systems have "holes" in address space, so sbrk cannot perform
437 contiguous expansion, but there is still space available on system.
438 On systems for which this is known to be useful (i.e. most linux
439 kernels), this occurs only when programs allocate huge amounts of
440 memory. Between this, and the fact that mmap regions tend to be
441 limited, the size should be large, to avoid too many mmap calls and
442 thus avoid running out of kernel resources. */
443
444#ifndef MMAP_AS_MORECORE_SIZE
445#define MMAP_AS_MORECORE_SIZE (1024 * 1024)
446#endif
447
448/*
449 Define HAVE_MREMAP to make realloc() use mremap() to re-allocate
450 large blocks.
451*/
452
453#ifndef HAVE_MREMAP
454#define HAVE_MREMAP 0
455#endif
456
457/* We may need to support __malloc_initialize_hook for backwards
458 compatibility. */
459
460#if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_24)
461# define HAVE_MALLOC_INIT_HOOK 1
462#else
463# define HAVE_MALLOC_INIT_HOOK 0
464#endif
465
466
467/*
468 This version of malloc supports the standard SVID/XPG mallinfo
469 routine that returns a struct containing usage properties and
470 statistics. It should work on any SVID/XPG compliant system that has
471 a /usr/include/malloc.h defining struct mallinfo. (If you'd like to
472 install such a thing yourself, cut out the preliminary declarations
473 as described above and below and save them in a malloc.h file. But
474 there's no compelling reason to bother to do this.)
475
476 The main declaration needed is the mallinfo struct that is returned
477 (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a
478 bunch of fields that are not even meaningful in this version of
479 malloc. These fields are are instead filled by mallinfo() with
480 other numbers that might be of interest.
481*/
482
483
484/* ---------- description of public routines ------------ */
485
486/*
487 malloc(size_t n)
488 Returns a pointer to a newly allocated chunk of at least n bytes, or null
489 if no space is available. Additionally, on failure, errno is
490 set to ENOMEM on ANSI C systems.
491
492 If n is zero, malloc returns a minumum-sized chunk. (The minimum
493 size is 16 bytes on most 32bit systems, and 24 or 32 bytes on 64bit
494 systems.) On most systems, size_t is an unsigned type, so calls
495 with negative arguments are interpreted as requests for huge amounts
496 of space, which will often fail. The maximum supported value of n
497 differs across systems, but is in all cases less than the maximum
498 representable value of a size_t.
499*/
500void* __libc_malloc(size_t);
501libc_hidden_proto (__libc_malloc)
502
503/*
504 free(void* p)
505 Releases the chunk of memory pointed to by p, that had been previously
506 allocated using malloc or a related routine such as realloc.
507 It has no effect if p is null. It can have arbitrary (i.e., bad!)
508 effects if p has already been freed.
509
510 Unless disabled (using mallopt), freeing very large spaces will
511 when possible, automatically trigger operations that give
512 back unused memory to the system, thus reducing program footprint.
513*/
514void __libc_free(void*);
515libc_hidden_proto (__libc_free)
516
517/*
518 calloc(size_t n_elements, size_t element_size);
519 Returns a pointer to n_elements * element_size bytes, with all locations
520 set to zero.
521*/
522void* __libc_calloc(size_t, size_t);
523
524/*
525 realloc(void* p, size_t n)
526 Returns a pointer to a chunk of size n that contains the same data
527 as does chunk p up to the minimum of (n, p's size) bytes, or null
528 if no space is available.
529
530 The returned pointer may or may not be the same as p. The algorithm
531 prefers extending p when possible, otherwise it employs the
532 equivalent of a malloc-copy-free sequence.
533
534 If p is null, realloc is equivalent to malloc.
535
536 If space is not available, realloc returns null, errno is set (if on
537 ANSI) and p is NOT freed.
538
539 if n is for fewer bytes than already held by p, the newly unused
540 space is lopped off and freed if possible. Unless the #define
541 REALLOC_ZERO_BYTES_FREES is set, realloc with a size argument of
542 zero (re)allocates a minimum-sized chunk.
543
544 Large chunks that were internally obtained via mmap will always be
545 grown using malloc-copy-free sequences unless the system supports
546 MREMAP (currently only linux).
547
548 The old unix realloc convention of allowing the last-free'd chunk
549 to be used as an argument to realloc is not supported.
550*/
551void* __libc_realloc(void*, size_t);
552libc_hidden_proto (__libc_realloc)
553
554/*
555 memalign(size_t alignment, size_t n);
556 Returns a pointer to a newly allocated chunk of n bytes, aligned
557 in accord with the alignment argument.
558
559 The alignment argument should be a power of two. If the argument is
560 not a power of two, the nearest greater power is used.
561 8-byte alignment is guaranteed by normal malloc calls, so don't
562 bother calling memalign with an argument of 8 or less.
563
564 Overreliance on memalign is a sure way to fragment space.
565*/
566void* __libc_memalign(size_t, size_t);
567libc_hidden_proto (__libc_memalign)
568
569/*
570 valloc(size_t n);
571 Equivalent to memalign(pagesize, n), where pagesize is the page
572 size of the system. If the pagesize is unknown, 4096 is used.
573*/
574void* __libc_valloc(size_t);
575
576
577
578/*
579 mallopt(int parameter_number, int parameter_value)
580 Sets tunable parameters The format is to provide a
581 (parameter-number, parameter-value) pair. mallopt then sets the
582 corresponding parameter to the argument value if it can (i.e., so
583 long as the value is meaningful), and returns 1 if successful else
584 0. SVID/XPG/ANSI defines four standard param numbers for mallopt,
585 normally defined in malloc.h. Only one of these (M_MXFAST) is used
586 in this malloc. The others (M_NLBLKS, M_GRAIN, M_KEEP) don't apply,
587 so setting them has no effect. But this malloc also supports four
588 other options in mallopt. See below for details. Briefly, supported
589 parameters are as follows (listed defaults are for "typical"
590 configurations).
591
592 Symbol param # default allowed param values
593 M_MXFAST 1 64 0-80 (0 disables fastbins)
594 M_TRIM_THRESHOLD -1 128*1024 any (-1U disables trimming)
595 M_TOP_PAD -2 0 any
596 M_MMAP_THRESHOLD -3 128*1024 any (or 0 if no MMAP support)
597 M_MMAP_MAX -4 65536 any (0 disables use of mmap)
598*/
599int __libc_mallopt(int, int);
600libc_hidden_proto (__libc_mallopt)
601
602
603/*
604 mallinfo()
605 Returns (by copy) a struct containing various summary statistics:
606
607 arena: current total non-mmapped bytes allocated from system
608 ordblks: the number of free chunks
609 smblks: the number of fastbin blocks (i.e., small chunks that
610 have been freed but not use resused or consolidated)
611 hblks: current number of mmapped regions
612 hblkhd: total bytes held in mmapped regions
613 usmblks: always 0
614 fsmblks: total bytes held in fastbin blocks
615 uordblks: current total allocated space (normal or mmapped)
616 fordblks: total free space
617 keepcost: the maximum number of bytes that could ideally be released
618 back to system via malloc_trim. ("ideally" means that
619 it ignores page restrictions etc.)
620
621 Because these fields are ints, but internal bookkeeping may
622 be kept as longs, the reported values may wrap around zero and
623 thus be inaccurate.
624*/
625struct mallinfo __libc_mallinfo(void);
626
627
628/*
629 pvalloc(size_t n);
630 Equivalent to valloc(minimum-page-that-holds(n)), that is,
631 round up n to nearest pagesize.
632 */
633void* __libc_pvalloc(size_t);
634
635/*
636 malloc_trim(size_t pad);
637
638 If possible, gives memory back to the system (via negative
639 arguments to sbrk) if there is unused memory at the `high' end of
640 the malloc pool. You can call this after freeing large blocks of
641 memory to potentially reduce the system-level memory requirements
642 of a program. However, it cannot guarantee to reduce memory. Under
643 some allocation patterns, some large free blocks of memory will be
644 locked between two used chunks, so they cannot be given back to
645 the system.
646
647 The `pad' argument to malloc_trim represents the amount of free
648 trailing space to leave untrimmed. If this argument is zero,
649 only the minimum amount of memory to maintain internal data
650 structures will be left (one page or less). Non-zero arguments
651 can be supplied to maintain enough trailing space to service
652 future expected allocations without having to re-obtain memory
653 from the system.
654
655 Malloc_trim returns 1 if it actually released any memory, else 0.
656 On systems that do not support "negative sbrks", it will always
657 return 0.
658*/
659int __malloc_trim(size_t);
660
661/*
662 malloc_usable_size(void* p);
663
664 Returns the number of bytes you can actually use in
665 an allocated chunk, which may be more than you requested (although
666 often not) due to alignment and minimum size constraints.
667 You can use this many bytes without worrying about
668 overwriting other allocated objects. This is not a particularly great
669 programming practice. malloc_usable_size can be more useful in
670 debugging and assertions, for example:
671
672 p = malloc(n);
673 assert(malloc_usable_size(p) >= 256);
674
675*/
676size_t __malloc_usable_size(void*);
677
678/*
679 malloc_stats();
680 Prints on stderr the amount of space obtained from the system (both
681 via sbrk and mmap), the maximum amount (which may be more than
682 current if malloc_trim and/or munmap got called), and the current
683 number of bytes allocated via malloc (or realloc, etc) but not yet
684 freed. Note that this is the number of bytes allocated, not the
685 number requested. It will be larger than the number requested
686 because of alignment and bookkeeping overhead. Because it includes
687 alignment wastage as being in use, this figure may be greater than
688 zero even when no user-level chunks are allocated.
689
690 The reported current and maximum system memory can be inaccurate if
691 a program makes other calls to system memory allocation functions
692 (normally sbrk) outside of malloc.
693
694 malloc_stats prints only the most commonly interesting statistics.
695 More information can be obtained by calling mallinfo.
696
697*/
698void __malloc_stats(void);
699
700/*
701 posix_memalign(void **memptr, size_t alignment, size_t size);
702
703 POSIX wrapper like memalign(), checking for validity of size.
704*/
705int __posix_memalign(void **, size_t, size_t);
706
707/* mallopt tuning options */
708
709/*
710 M_MXFAST is the maximum request size used for "fastbins", special bins
711 that hold returned chunks without consolidating their spaces. This
712 enables future requests for chunks of the same size to be handled
713 very quickly, but can increase fragmentation, and thus increase the
714 overall memory footprint of a program.
715
716 This malloc manages fastbins very conservatively yet still
717 efficiently, so fragmentation is rarely a problem for values less
718 than or equal to the default. The maximum supported value of MXFAST
719 is 80. You wouldn't want it any higher than this anyway. Fastbins
720 are designed especially for use with many small structs, objects or
721 strings -- the default handles structs/objects/arrays with sizes up
722 to 8 4byte fields, or small strings representing words, tokens,
723 etc. Using fastbins for larger objects normally worsens
724 fragmentation without improving speed.
725
726 M_MXFAST is set in REQUEST size units. It is internally used in
727 chunksize units, which adds padding and alignment. You can reduce
728 M_MXFAST to 0 to disable all use of fastbins. This causes the malloc
729 algorithm to be a closer approximation of fifo-best-fit in all cases,
730 not just for larger requests, but will generally cause it to be
731 slower.
732*/
733
734
735/* M_MXFAST is a standard SVID/XPG tuning option, usually listed in malloc.h */
736#ifndef M_MXFAST
737#define M_MXFAST 1
738#endif
739
740#ifndef DEFAULT_MXFAST
741#define DEFAULT_MXFAST (64 * SIZE_SZ / 4)
742#endif
743
744
745/*
746 M_TRIM_THRESHOLD is the maximum amount of unused top-most memory
747 to keep before releasing via malloc_trim in free().
748
749 Automatic trimming is mainly useful in long-lived programs.
750 Because trimming via sbrk can be slow on some systems, and can
751 sometimes be wasteful (in cases where programs immediately
752 afterward allocate more large chunks) the value should be high
753 enough so that your overall system performance would improve by
754 releasing this much memory.
755
756 The trim threshold and the mmap control parameters (see below)
757 can be traded off with one another. Trimming and mmapping are
758 two different ways of releasing unused memory back to the
759 system. Between these two, it is often possible to keep
760 system-level demands of a long-lived program down to a bare
761 minimum. For example, in one test suite of sessions measuring
762 the XF86 X server on Linux, using a trim threshold of 128K and a
763 mmap threshold of 192K led to near-minimal long term resource
764 consumption.
765
766 If you are using this malloc in a long-lived program, it should
767 pay to experiment with these values. As a rough guide, you
768 might set to a value close to the average size of a process
769 (program) running on your system. Releasing this much memory
770 would allow such a process to run in memory. Generally, it's
771 worth it to tune for trimming rather tham memory mapping when a
772 program undergoes phases where several large chunks are
773 allocated and released in ways that can reuse each other's
774 storage, perhaps mixed with phases where there are no such
775 chunks at all. And in well-behaved long-lived programs,
776 controlling release of large blocks via trimming versus mapping
777 is usually faster.
778
779 However, in most programs, these parameters serve mainly as
780 protection against the system-level effects of carrying around
781 massive amounts of unneeded memory. Since frequent calls to
782 sbrk, mmap, and munmap otherwise degrade performance, the default
783 parameters are set to relatively high values that serve only as
784 safeguards.
785
786 The trim value It must be greater than page size to have any useful
787 effect. To disable trimming completely, you can set to
788 (unsigned long)(-1)
789
790 Trim settings interact with fastbin (MXFAST) settings: Unless
791 TRIM_FASTBINS is defined, automatic trimming never takes place upon
792 freeing a chunk with size less than or equal to MXFAST. Trimming is
793 instead delayed until subsequent freeing of larger chunks. However,
794 you can still force an attempted trim by calling malloc_trim.
795
796 Also, trimming is not generally possible in cases where
797 the main arena is obtained via mmap.
798
799 Note that the trick some people use of mallocing a huge space and
800 then freeing it at program startup, in an attempt to reserve system
801 memory, doesn't have the intended effect under automatic trimming,
802 since that memory will immediately be returned to the system.
803*/
804
805#define M_TRIM_THRESHOLD -1
806
807#ifndef DEFAULT_TRIM_THRESHOLD
808#define DEFAULT_TRIM_THRESHOLD (128 * 1024)
809#endif
810
811/*
812 M_TOP_PAD is the amount of extra `padding' space to allocate or
813 retain whenever sbrk is called. It is used in two ways internally:
814
815 * When sbrk is called to extend the top of the arena to satisfy
816 a new malloc request, this much padding is added to the sbrk
817 request.
818
819 * When malloc_trim is called automatically from free(),
820 it is used as the `pad' argument.
821
822 In both cases, the actual amount of padding is rounded
823 so that the end of the arena is always a system page boundary.
824
825 The main reason for using padding is to avoid calling sbrk so
826 often. Having even a small pad greatly reduces the likelihood
827 that nearly every malloc request during program start-up (or
828 after trimming) will invoke sbrk, which needlessly wastes
829 time.
830
831 Automatic rounding-up to page-size units is normally sufficient
832 to avoid measurable overhead, so the default is 0. However, in
833 systems where sbrk is relatively slow, it can pay to increase
834 this value, at the expense of carrying around more memory than
835 the program needs.
836*/
837
838#define M_TOP_PAD -2
839
840#ifndef DEFAULT_TOP_PAD
841#define DEFAULT_TOP_PAD (0)
842#endif
843
844/*
845 MMAP_THRESHOLD_MAX and _MIN are the bounds on the dynamically
846 adjusted MMAP_THRESHOLD.
847*/
848
849#ifndef DEFAULT_MMAP_THRESHOLD_MIN
850#define DEFAULT_MMAP_THRESHOLD_MIN (128 * 1024)
851#endif
852
853#ifndef DEFAULT_MMAP_THRESHOLD_MAX
854 /* For 32-bit platforms we cannot increase the maximum mmap
855 threshold much because it is also the minimum value for the
856 maximum heap size and its alignment. Going above 512k (i.e., 1M
857 for new heaps) wastes too much address space. */
858# if __WORDSIZE == 32
859# define DEFAULT_MMAP_THRESHOLD_MAX (512 * 1024)
860# else
861# define DEFAULT_MMAP_THRESHOLD_MAX (4 * 1024 * 1024 * sizeof(long))
862# endif
863#endif
864
865/*
866 M_MMAP_THRESHOLD is the request size threshold for using mmap()
867 to service a request. Requests of at least this size that cannot
868 be allocated using already-existing space will be serviced via mmap.
869 (If enough normal freed space already exists it is used instead.)
870
871 Using mmap segregates relatively large chunks of memory so that
872 they can be individually obtained and released from the host
873 system. A request serviced through mmap is never reused by any
874 other request (at least not directly; the system may just so
875 happen to remap successive requests to the same locations).
876
877 Segregating space in this way has the benefits that:
878
879 1. Mmapped space can ALWAYS be individually released back
880 to the system, which helps keep the system level memory
881 demands of a long-lived program low.
882 2. Mapped memory can never become `locked' between
883 other chunks, as can happen with normally allocated chunks, which
884 means that even trimming via malloc_trim would not release them.
885 3. On some systems with "holes" in address spaces, mmap can obtain
886 memory that sbrk cannot.
887
888 However, it has the disadvantages that:
889
890 1. The space cannot be reclaimed, consolidated, and then
891 used to service later requests, as happens with normal chunks.
892 2. It can lead to more wastage because of mmap page alignment
893 requirements
894 3. It causes malloc performance to be more dependent on host
895 system memory management support routines which may vary in
896 implementation quality and may impose arbitrary
897 limitations. Generally, servicing a request via normal
898 malloc steps is faster than going through a system's mmap.
899
900 The advantages of mmap nearly always outweigh disadvantages for
901 "large" chunks, but the value of "large" varies across systems. The
902 default is an empirically derived value that works well in most
903 systems.
904
905
906 Update in 2006:
907 The above was written in 2001. Since then the world has changed a lot.
908 Memory got bigger. Applications got bigger. The virtual address space
909 layout in 32 bit linux changed.
910
911 In the new situation, brk() and mmap space is shared and there are no
912 artificial limits on brk size imposed by the kernel. What is more,
913 applications have started using transient allocations larger than the
914 128Kb as was imagined in 2001.
915
916 The price for mmap is also high now; each time glibc mmaps from the
917 kernel, the kernel is forced to zero out the memory it gives to the
918 application. Zeroing memory is expensive and eats a lot of cache and
919 memory bandwidth. This has nothing to do with the efficiency of the
920 virtual memory system, by doing mmap the kernel just has no choice but
921 to zero.
922
923 In 2001, the kernel had a maximum size for brk() which was about 800
924 megabytes on 32 bit x86, at that point brk() would hit the first
925 mmaped shared libaries and couldn't expand anymore. With current 2.6
926 kernels, the VA space layout is different and brk() and mmap
927 both can span the entire heap at will.
928
929 Rather than using a static threshold for the brk/mmap tradeoff,
930 we are now using a simple dynamic one. The goal is still to avoid
931 fragmentation. The old goals we kept are
932 1) try to get the long lived large allocations to use mmap()
933 2) really large allocations should always use mmap()
934 and we're adding now:
935 3) transient allocations should use brk() to avoid forcing the kernel
936 having to zero memory over and over again
937
938 The implementation works with a sliding threshold, which is by default
939 limited to go between 128Kb and 32Mb (64Mb for 64 bitmachines) and starts
940 out at 128Kb as per the 2001 default.
941
942 This allows us to satisfy requirement 1) under the assumption that long
943 lived allocations are made early in the process' lifespan, before it has
944 started doing dynamic allocations of the same size (which will
945 increase the threshold).
946
947 The upperbound on the threshold satisfies requirement 2)
948
949 The threshold goes up in value when the application frees memory that was
950 allocated with the mmap allocator. The idea is that once the application
951 starts freeing memory of a certain size, it's highly probable that this is
952 a size the application uses for transient allocations. This estimator
953 is there to satisfy the new third requirement.
954
955*/
956
957#define M_MMAP_THRESHOLD -3
958
959#ifndef DEFAULT_MMAP_THRESHOLD
960#define DEFAULT_MMAP_THRESHOLD DEFAULT_MMAP_THRESHOLD_MIN
961#endif
962
963/*
964 M_MMAP_MAX is the maximum number of requests to simultaneously
965 service using mmap. This parameter exists because
966 some systems have a limited number of internal tables for
967 use by mmap, and using more than a few of them may degrade
968 performance.
969
970 The default is set to a value that serves only as a safeguard.
971 Setting to 0 disables use of mmap for servicing large requests.
972*/
973
974#define M_MMAP_MAX -4
975
976#ifndef DEFAULT_MMAP_MAX
977#define DEFAULT_MMAP_MAX (65536)
978#endif
979
980#include <malloc.h>
981
982#ifndef RETURN_ADDRESS
983#define RETURN_ADDRESS(X_) (NULL)
984#endif
985
986/* Forward declarations. */
987struct malloc_chunk;
988typedef struct malloc_chunk* mchunkptr;
989
990/* Internal routines. */
991
992static void* _int_malloc(mstate, size_t);
993static void _int_free(mstate, mchunkptr, int);
994static void* _int_realloc(mstate, mchunkptr, INTERNAL_SIZE_T,
995 INTERNAL_SIZE_T);
996static void* _int_memalign(mstate, size_t, size_t);
997static void* _mid_memalign(size_t, size_t, void *);
998
999static void malloc_printerr(const char *str) __attribute__ ((noreturn));
1000
1001static void* mem2mem_check(void *p, size_t sz);
1002static void top_check(void);
1003static void munmap_chunk(mchunkptr p);
1004#if HAVE_MREMAP
1005static mchunkptr mremap_chunk(mchunkptr p, size_t new_size);
1006#endif
1007
1008static void* malloc_check(size_t sz, const void *caller);
1009static void free_check(void* mem, const void *caller);
1010static void* realloc_check(void* oldmem, size_t bytes,
1011 const void *caller);
1012static void* memalign_check(size_t alignment, size_t bytes,
1013 const void *caller);
1014
1015/* ------------------ MMAP support ------------------ */
1016
1017
1018#include <fcntl.h>
1019#include <sys/mman.h>
1020
1021#if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
1022# define MAP_ANONYMOUS MAP_ANON
1023#endif
1024
1025#ifndef MAP_NORESERVE
1026# define MAP_NORESERVE 0
1027#endif
1028
1029#define MMAP(addr, size, prot, flags) \
1030 __mmap((addr), (size), (prot), (flags)|MAP_ANONYMOUS|MAP_PRIVATE, -1, 0)
1031
1032
1033/*
1034 ----------------------- Chunk representations -----------------------
1035*/
1036
1037
1038/*
1039 This struct declaration is misleading (but accurate and necessary).
1040 It declares a "view" into memory allowing access to necessary
1041 fields at known offsets from a given base. See explanation below.
1042*/
1043
1044struct malloc_chunk {
1045
1046 INTERNAL_SIZE_T mchunk_prev_size; /* Size of previous chunk (if free). */
1047 INTERNAL_SIZE_T mchunk_size; /* Size in bytes, including overhead. */
1048
1049 struct malloc_chunk* fd; /* double links -- used only if free. */
1050 struct malloc_chunk* bk;
1051
1052 /* Only used for large blocks: pointer to next larger size. */
1053 struct malloc_chunk* fd_nextsize; /* double links -- used only if free. */
1054 struct malloc_chunk* bk_nextsize;
1055};
1056
1057
1058/*
1059 malloc_chunk details:
1060
1061 (The following includes lightly edited explanations by Colin Plumb.)
1062
1063 Chunks of memory are maintained using a `boundary tag' method as
1064 described in e.g., Knuth or Standish. (See the paper by Paul
1065 Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a
1066 survey of such techniques.) Sizes of free chunks are stored both
1067 in the front of each chunk and at the end. This makes
1068 consolidating fragmented chunks into bigger chunks very fast. The
1069 size fields also hold bits representing whether chunks are free or
1070 in use.
1071
1072 An allocated chunk looks like this:
1073
1074
1075 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1076 | Size of previous chunk, if unallocated (P clear) |
1077 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1078 | Size of chunk, in bytes |A|M|P|
1079 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1080 | User data starts here... .
1081 . .
1082 . (malloc_usable_size() bytes) .
1083 . |
1084nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1085 | (size of chunk, but used for application data) |
1086 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1087 | Size of next chunk, in bytes |A|0|1|
1088 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1089
1090 Where "chunk" is the front of the chunk for the purpose of most of
1091 the malloc code, but "mem" is the pointer that is returned to the
1092 user. "Nextchunk" is the beginning of the next contiguous chunk.
1093
1094 Chunks always begin on even word boundaries, so the mem portion
1095 (which is returned to the user) is also on an even word boundary, and
1096 thus at least double-word aligned.
1097
1098 Free chunks are stored in circular doubly-linked lists, and look like this:
1099
1100 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1101 | Size of previous chunk, if unallocated (P clear) |
1102 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1103 `head:' | Size of chunk, in bytes |A|0|P|
1104 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1105 | Forward pointer to next chunk in list |
1106 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1107 | Back pointer to previous chunk in list |
1108 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1109 | Unused space (may be 0 bytes long) .
1110 . .
1111 . |
1112nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1113 `foot:' | Size of chunk, in bytes |
1114 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1115 | Size of next chunk, in bytes |A|0|0|
1116 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1117
1118 The P (PREV_INUSE) bit, stored in the unused low-order bit of the
1119 chunk size (which is always a multiple of two words), is an in-use
1120 bit for the *previous* chunk. If that bit is *clear*, then the
1121 word before the current chunk size contains the previous chunk
1122 size, and can be used to find the front of the previous chunk.
1123 The very first chunk allocated always has this bit set,
1124 preventing access to non-existent (or non-owned) memory. If
1125 prev_inuse is set for any given chunk, then you CANNOT determine
1126 the size of the previous chunk, and might even get a memory
1127 addressing fault when trying to do so.
1128
1129 The A (NON_MAIN_ARENA) bit is cleared for chunks on the initial,
1130 main arena, described by the main_arena variable. When additional
1131 threads are spawned, each thread receives its own arena (up to a
1132 configurable limit, after which arenas are reused for multiple
1133 threads), and the chunks in these arenas have the A bit set. To
1134 find the arena for a chunk on such a non-main arena, heap_for_ptr
1135 performs a bit mask operation and indirection through the ar_ptr
1136 member of the per-heap header heap_info (see arena.c).
1137
1138 Note that the `foot' of the current chunk is actually represented
1139 as the prev_size of the NEXT chunk. This makes it easier to
1140 deal with alignments etc but can be very confusing when trying
1141 to extend or adapt this code.
1142
1143 The three exceptions to all this are:
1144
1145 1. The special chunk `top' doesn't bother using the
1146 trailing size field since there is no next contiguous chunk
1147 that would have to index off it. After initialization, `top'
1148 is forced to always exist. If it would become less than
1149 MINSIZE bytes long, it is replenished.
1150
1151 2. Chunks allocated via mmap, which have the second-lowest-order
1152 bit M (IS_MMAPPED) set in their size fields. Because they are
1153 allocated one-by-one, each must contain its own trailing size
1154 field. If the M bit is set, the other bits are ignored
1155 (because mmapped chunks are neither in an arena, nor adjacent
1156 to a freed chunk). The M bit is also used for chunks which
1157 originally came from a dumped heap via malloc_set_state in
1158 hooks.c.
1159
1160 3. Chunks in fastbins are treated as allocated chunks from the
1161 point of view of the chunk allocator. They are consolidated
1162 with their neighbors only in bulk, in malloc_consolidate.
1163*/
1164
1165/*
1166 ---------- Size and alignment checks and conversions ----------
1167*/
1168
1169/* conversion from malloc headers to user pointers, and back */
1170
1171#define chunk2mem(p) ((void*)((char*)(p) + 2*SIZE_SZ))
1172#define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*SIZE_SZ))
1173
1174/* The smallest possible chunk */
1175#define MIN_CHUNK_SIZE (offsetof(struct malloc_chunk, fd_nextsize))
1176
1177/* The smallest size we can malloc is an aligned minimal chunk */
1178
1179#define MINSIZE \
1180 (unsigned long)(((MIN_CHUNK_SIZE+MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK))
1181
1182/* Check if m has acceptable alignment */
1183
1184#define aligned_OK(m) (((unsigned long)(m) & MALLOC_ALIGN_MASK) == 0)
1185
1186#define misaligned_chunk(p) \
1187 ((uintptr_t)(MALLOC_ALIGNMENT == 2 * SIZE_SZ ? (p) : chunk2mem (p)) \
1188 & MALLOC_ALIGN_MASK)
1189
1190
1191/*
1192 Check if a request is so large that it would wrap around zero when
1193 padded and aligned. To simplify some other code, the bound is made
1194 low enough so that adding MINSIZE will also not wrap around zero.
1195 */
1196
1197#define REQUEST_OUT_OF_RANGE(req) \
1198 ((unsigned long) (req) >= \
1199 (unsigned long) (INTERNAL_SIZE_T) (-2 * MINSIZE))
1200
1201/* pad request bytes into a usable size -- internal version */
1202
1203#define request2size(req) \
1204 (((req) + SIZE_SZ + MALLOC_ALIGN_MASK < MINSIZE) ? \
1205 MINSIZE : \
1206 ((req) + SIZE_SZ + MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK)
1207
1208/* Same, except also perform an argument and result check. First, we check
1209 that the padding done by request2size didn't result in an integer
1210 overflow. Then we check (using REQUEST_OUT_OF_RANGE) that the resulting
1211 size isn't so large that a later alignment would lead to another integer
1212 overflow. */
1213#define checked_request2size(req, sz) \
1214({ \
1215 (sz) = request2size (req); \
1216 if (((sz) < (req)) \
1217 || REQUEST_OUT_OF_RANGE (sz)) \
1218 { \
1219 __set_errno (ENOMEM); \
1220 return 0; \
1221 } \
1222})
1223
1224/*
1225 --------------- Physical chunk operations ---------------
1226 */
1227
1228
1229/* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */
1230#define PREV_INUSE 0x1
1231
1232/* extract inuse bit of previous chunk */
1233#define prev_inuse(p) ((p)->mchunk_size & PREV_INUSE)
1234
1235
1236/* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */
1237#define IS_MMAPPED 0x2
1238
1239/* check for mmap()'ed chunk */
1240#define chunk_is_mmapped(p) ((p)->mchunk_size & IS_MMAPPED)
1241
1242
1243/* size field is or'ed with NON_MAIN_ARENA if the chunk was obtained
1244 from a non-main arena. This is only set immediately before handing
1245 the chunk to the user, if necessary. */
1246#define NON_MAIN_ARENA 0x4
1247
1248/* Check for chunk from main arena. */
1249#define chunk_main_arena(p) (((p)->mchunk_size & NON_MAIN_ARENA) == 0)
1250
1251/* Mark a chunk as not being on the main arena. */
1252#define set_non_main_arena(p) ((p)->mchunk_size |= NON_MAIN_ARENA)
1253
1254
1255/*
1256 Bits to mask off when extracting size
1257
1258 Note: IS_MMAPPED is intentionally not masked off from size field in
1259 macros for which mmapped chunks should never be seen. This should
1260 cause helpful core dumps to occur if it is tried by accident by
1261 people extending or adapting this malloc.
1262 */
1263#define SIZE_BITS (PREV_INUSE | IS_MMAPPED | NON_MAIN_ARENA)
1264
1265/* Get size, ignoring use bits */
1266#define chunksize(p) (chunksize_nomask (p) & ~(SIZE_BITS))
1267
1268/* Like chunksize, but do not mask SIZE_BITS. */
1269#define chunksize_nomask(p) ((p)->mchunk_size)
1270
1271/* Ptr to next physical malloc_chunk. */
1272#define next_chunk(p) ((mchunkptr) (((char *) (p)) + chunksize (p)))
1273
1274/* Size of the chunk below P. Only valid if !prev_inuse (P). */
1275#define prev_size(p) ((p)->mchunk_prev_size)
1276
1277/* Set the size of the chunk below P. Only valid if !prev_inuse (P). */
1278#define set_prev_size(p, sz) ((p)->mchunk_prev_size = (sz))
1279
1280/* Ptr to previous physical malloc_chunk. Only valid if !prev_inuse (P). */
1281#define prev_chunk(p) ((mchunkptr) (((char *) (p)) - prev_size (p)))
1282
1283/* Treat space at ptr + offset as a chunk */
1284#define chunk_at_offset(p, s) ((mchunkptr) (((char *) (p)) + (s)))
1285
1286/* extract p's inuse bit */
1287#define inuse(p) \
1288 ((((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size) & PREV_INUSE)
1289
1290/* set/clear chunk as being inuse without otherwise disturbing */
1291#define set_inuse(p) \
1292 ((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size |= PREV_INUSE
1293
1294#define clear_inuse(p) \
1295 ((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size &= ~(PREV_INUSE)
1296
1297
1298/* check/set/clear inuse bits in known places */
1299#define inuse_bit_at_offset(p, s) \
1300 (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size & PREV_INUSE)
1301
1302#define set_inuse_bit_at_offset(p, s) \
1303 (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size |= PREV_INUSE)
1304
1305#define clear_inuse_bit_at_offset(p, s) \
1306 (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size &= ~(PREV_INUSE))
1307
1308
1309/* Set size at head, without disturbing its use bit */
1310#define set_head_size(p, s) ((p)->mchunk_size = (((p)->mchunk_size & SIZE_BITS) | (s)))
1311
1312/* Set size/use field */
1313#define set_head(p, s) ((p)->mchunk_size = (s))
1314
1315/* Set size at footer (only when chunk is not in use) */
1316#define set_foot(p, s) (((mchunkptr) ((char *) (p) + (s)))->mchunk_prev_size = (s))
1317
1318
1319#pragma GCC poison mchunk_size
1320#pragma GCC poison mchunk_prev_size
1321
1322/*
1323 -------------------- Internal data structures --------------------
1324
1325 All internal state is held in an instance of malloc_state defined
1326 below. There are no other static variables, except in two optional
1327 cases:
1328 * If USE_MALLOC_LOCK is defined, the mALLOC_MUTEx declared above.
1329 * If mmap doesn't support MAP_ANONYMOUS, a dummy file descriptor
1330 for mmap.
1331
1332 Beware of lots of tricks that minimize the total bookkeeping space
1333 requirements. The result is a little over 1K bytes (for 4byte
1334 pointers and size_t.)
1335 */
1336
1337/*
1338 Bins
1339
1340 An array of bin headers for free chunks. Each bin is doubly
1341 linked. The bins are approximately proportionally (log) spaced.
1342 There are a lot of these bins (128). This may look excessive, but
1343 works very well in practice. Most bins hold sizes that are
1344 unusual as malloc request sizes, but are more usual for fragments
1345 and consolidated sets of chunks, which is what these bins hold, so
1346 they can be found quickly. All procedures maintain the invariant
1347 that no consolidated chunk physically borders another one, so each
1348 chunk in a list is known to be preceeded and followed by either
1349 inuse chunks or the ends of memory.
1350
1351 Chunks in bins are kept in size order, with ties going to the
1352 approximately least recently used chunk. Ordering isn't needed
1353 for the small bins, which all contain the same-sized chunks, but
1354 facilitates best-fit allocation for larger chunks. These lists
1355 are just sequential. Keeping them in order almost never requires
1356 enough traversal to warrant using fancier ordered data
1357 structures.
1358
1359 Chunks of the same size are linked with the most
1360 recently freed at the front, and allocations are taken from the
1361 back. This results in LRU (FIFO) allocation order, which tends
1362 to give each chunk an equal opportunity to be consolidated with
1363 adjacent freed chunks, resulting in larger free chunks and less
1364 fragmentation.
1365
1366 To simplify use in double-linked lists, each bin header acts
1367 as a malloc_chunk. This avoids special-casing for headers.
1368 But to conserve space and improve locality, we allocate
1369 only the fd/bk pointers of bins, and then use repositioning tricks
1370 to treat these as the fields of a malloc_chunk*.
1371 */
1372
1373typedef struct malloc_chunk *mbinptr;
1374
1375/* addressing -- note that bin_at(0) does not exist */
1376#define bin_at(m, i) \
1377 (mbinptr) (((char *) &((m)->bins[((i) - 1) * 2])) \
1378 - offsetof (struct malloc_chunk, fd))
1379
1380/* analog of ++bin */
1381#define next_bin(b) ((mbinptr) ((char *) (b) + (sizeof (mchunkptr) << 1)))
1382
1383/* Reminders about list directionality within bins */
1384#define first(b) ((b)->fd)
1385#define last(b) ((b)->bk)
1386
1387/*
1388 Indexing
1389
1390 Bins for sizes < 512 bytes contain chunks of all the same size, spaced
1391 8 bytes apart. Larger bins are approximately logarithmically spaced:
1392
1393 64 bins of size 8
1394 32 bins of size 64
1395 16 bins of size 512
1396 8 bins of size 4096
1397 4 bins of size 32768
1398 2 bins of size 262144
1399 1 bin of size what's left
1400
1401 There is actually a little bit of slop in the numbers in bin_index
1402 for the sake of speed. This makes no difference elsewhere.
1403
1404 The bins top out around 1MB because we expect to service large
1405 requests via mmap.
1406
1407 Bin 0 does not exist. Bin 1 is the unordered list; if that would be
1408 a valid chunk size the small bins are bumped up one.
1409 */
1410
1411#define NBINS 128
1412#define NSMALLBINS 64
1413#define SMALLBIN_WIDTH MALLOC_ALIGNMENT
1414#define SMALLBIN_CORRECTION (MALLOC_ALIGNMENT > 2 * SIZE_SZ)
1415#define MIN_LARGE_SIZE ((NSMALLBINS - SMALLBIN_CORRECTION) * SMALLBIN_WIDTH)
1416
1417#define in_smallbin_range(sz) \
1418 ((unsigned long) (sz) < (unsigned long) MIN_LARGE_SIZE)
1419
1420#define smallbin_index(sz) \
1421 ((SMALLBIN_WIDTH == 16 ? (((unsigned) (sz)) >> 4) : (((unsigned) (sz)) >> 3))\
1422 + SMALLBIN_CORRECTION)
1423
1424#define largebin_index_32(sz) \
1425 (((((unsigned long) (sz)) >> 6) <= 38) ? 56 + (((unsigned long) (sz)) >> 6) :\
1426 ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1427 ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1428 ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1429 ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1430 126)
1431
1432#define largebin_index_32_big(sz) \
1433 (((((unsigned long) (sz)) >> 6) <= 45) ? 49 + (((unsigned long) (sz)) >> 6) :\
1434 ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1435 ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1436 ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1437 ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1438 126)
1439
1440// XXX It remains to be seen whether it is good to keep the widths of
1441// XXX the buckets the same or whether it should be scaled by a factor
1442// XXX of two as well.
1443#define largebin_index_64(sz) \
1444 (((((unsigned long) (sz)) >> 6) <= 48) ? 48 + (((unsigned long) (sz)) >> 6) :\
1445 ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1446 ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1447 ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1448 ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1449 126)
1450
1451#define largebin_index(sz) \
1452 (SIZE_SZ == 8 ? largebin_index_64 (sz) \
1453 : MALLOC_ALIGNMENT == 16 ? largebin_index_32_big (sz) \
1454 : largebin_index_32 (sz))
1455
1456#define bin_index(sz) \
1457 ((in_smallbin_range (sz)) ? smallbin_index (sz) : largebin_index (sz))
1458
1459/* Take a chunk off a bin list. */
1460static void
1461unlink_chunk (mstate av, mchunkptr p)
1462{
1463 if (chunksize (p) != prev_size (next_chunk (p)))
1464 malloc_printerr ("corrupted size vs. prev_size");
1465
1466 mchunkptr fd = p->fd;
1467 mchunkptr bk = p->bk;
1468
1469 if (__builtin_expect (fd->bk != p || bk->fd != p, 0))
1470 malloc_printerr ("corrupted double-linked list");
1471
1472 fd->bk = bk;
1473 bk->fd = fd;
1474 if (!in_smallbin_range (chunksize_nomask (p)) && p->fd_nextsize != NULL)
1475 {
1476 if (p->fd_nextsize->bk_nextsize != p
1477 || p->bk_nextsize->fd_nextsize != p)
1478 malloc_printerr ("corrupted double-linked list (not small)");
1479
1480 if (fd->fd_nextsize == NULL)
1481 {
1482 if (p->fd_nextsize == p)
1483 fd->fd_nextsize = fd->bk_nextsize = fd;
1484 else
1485 {
1486 fd->fd_nextsize = p->fd_nextsize;
1487 fd->bk_nextsize = p->bk_nextsize;
1488 p->fd_nextsize->bk_nextsize = fd;
1489 p->bk_nextsize->fd_nextsize = fd;
1490 }
1491 }
1492 else
1493 {
1494 p->fd_nextsize->bk_nextsize = p->bk_nextsize;
1495 p->bk_nextsize->fd_nextsize = p->fd_nextsize;
1496 }
1497 }
1498}
1499
1500/*
1501 Unsorted chunks
1502
1503 All remainders from chunk splits, as well as all returned chunks,
1504 are first placed in the "unsorted" bin. They are then placed
1505 in regular bins after malloc gives them ONE chance to be used before
1506 binning. So, basically, the unsorted_chunks list acts as a queue,
1507 with chunks being placed on it in free (and malloc_consolidate),
1508 and taken off (to be either used or placed in bins) in malloc.
1509
1510 The NON_MAIN_ARENA flag is never set for unsorted chunks, so it
1511 does not have to be taken into account in size comparisons.
1512 */
1513
1514/* The otherwise unindexable 1-bin is used to hold unsorted chunks. */
1515#define unsorted_chunks(M) (bin_at (M, 1))
1516
1517/*
1518 Top
1519
1520 The top-most available chunk (i.e., the one bordering the end of
1521 available memory) is treated specially. It is never included in
1522 any bin, is used only if no other chunk is available, and is
1523 released back to the system if it is very large (see
1524 M_TRIM_THRESHOLD). Because top initially
1525 points to its own bin with initial zero size, thus forcing
1526 extension on the first malloc request, we avoid having any special
1527 code in malloc to check whether it even exists yet. But we still
1528 need to do so when getting memory from system, so we make
1529 initial_top treat the bin as a legal but unusable chunk during the
1530 interval between initialization and the first call to
1531 sysmalloc. (This is somewhat delicate, since it relies on
1532 the 2 preceding words to be zero during this interval as well.)
1533 */
1534
1535/* Conveniently, the unsorted bin can be used as dummy top on first call */
1536#define initial_top(M) (unsorted_chunks (M))
1537
1538/*
1539 Binmap
1540
1541 To help compensate for the large number of bins, a one-level index
1542 structure is used for bin-by-bin searching. `binmap' is a
1543 bitvector recording whether bins are definitely empty so they can
1544 be skipped over during during traversals. The bits are NOT always
1545 cleared as soon as bins are empty, but instead only
1546 when they are noticed to be empty during traversal in malloc.
1547 */
1548
1549/* Conservatively use 32 bits per map word, even if on 64bit system */
1550#define BINMAPSHIFT 5
1551#define BITSPERMAP (1U << BINMAPSHIFT)
1552#define BINMAPSIZE (NBINS / BITSPERMAP)
1553
1554#define idx2block(i) ((i) >> BINMAPSHIFT)
1555#define idx2bit(i) ((1U << ((i) & ((1U << BINMAPSHIFT) - 1))))
1556
1557#define mark_bin(m, i) ((m)->binmap[idx2block (i)] |= idx2bit (i))
1558#define unmark_bin(m, i) ((m)->binmap[idx2block (i)] &= ~(idx2bit (i)))
1559#define get_binmap(m, i) ((m)->binmap[idx2block (i)] & idx2bit (i))
1560
1561/*
1562 Fastbins
1563
1564 An array of lists holding recently freed small chunks. Fastbins
1565 are not doubly linked. It is faster to single-link them, and
1566 since chunks are never removed from the middles of these lists,
1567 double linking is not necessary. Also, unlike regular bins, they
1568 are not even processed in FIFO order (they use faster LIFO) since
1569 ordering doesn't much matter in the transient contexts in which
1570 fastbins are normally used.
1571
1572 Chunks in fastbins keep their inuse bit set, so they cannot
1573 be consolidated with other free chunks. malloc_consolidate
1574 releases all chunks in fastbins and consolidates them with
1575 other free chunks.
1576 */
1577
1578typedef struct malloc_chunk *mfastbinptr;
1579#define fastbin(ar_ptr, idx) ((ar_ptr)->fastbinsY[idx])
1580
1581/* offset 2 to use otherwise unindexable first 2 bins */
1582#define fastbin_index(sz) \
1583 ((((unsigned int) (sz)) >> (SIZE_SZ == 8 ? 4 : 3)) - 2)
1584
1585
1586/* The maximum fastbin request size we support */
1587#define MAX_FAST_SIZE (80 * SIZE_SZ / 4)
1588
1589#define NFASTBINS (fastbin_index (request2size (MAX_FAST_SIZE)) + 1)
1590
1591/*
1592 FASTBIN_CONSOLIDATION_THRESHOLD is the size of a chunk in free()
1593 that triggers automatic consolidation of possibly-surrounding
1594 fastbin chunks. This is a heuristic, so the exact value should not
1595 matter too much. It is defined at half the default trim threshold as a
1596 compromise heuristic to only attempt consolidation if it is likely
1597 to lead to trimming. However, it is not dynamically tunable, since
1598 consolidation reduces fragmentation surrounding large chunks even
1599 if trimming is not used.
1600 */
1601
1602#define FASTBIN_CONSOLIDATION_THRESHOLD (65536UL)
1603
1604/*
1605 NONCONTIGUOUS_BIT indicates that MORECORE does not return contiguous
1606 regions. Otherwise, contiguity is exploited in merging together,
1607 when possible, results from consecutive MORECORE calls.
1608
1609 The initial value comes from MORECORE_CONTIGUOUS, but is
1610 changed dynamically if mmap is ever used as an sbrk substitute.
1611 */
1612
1613#define NONCONTIGUOUS_BIT (2U)
1614
1615#define contiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) == 0)
1616#define noncontiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) != 0)
1617#define set_noncontiguous(M) ((M)->flags |= NONCONTIGUOUS_BIT)
1618#define set_contiguous(M) ((M)->flags &= ~NONCONTIGUOUS_BIT)
1619
1620/* Maximum size of memory handled in fastbins. */
1621static INTERNAL_SIZE_T global_max_fast;
1622
1623/*
1624 Set value of max_fast.
1625 Use impossibly small value if 0.
1626 Precondition: there are no existing fastbin chunks in the main arena.
1627 Since do_check_malloc_state () checks this, we call malloc_consolidate ()
1628 before changing max_fast. Note other arenas will leak their fast bin
1629 entries if max_fast is reduced.
1630 */
1631
1632#define set_max_fast(s) \
1633 global_max_fast = (((s) == 0) \
1634 ? SMALLBIN_WIDTH : ((s + SIZE_SZ) & ~MALLOC_ALIGN_MASK))
1635
1636static inline INTERNAL_SIZE_T
1637get_max_fast (void)
1638{
1639 /* Tell the GCC optimizers that global_max_fast is never larger
1640 than MAX_FAST_SIZE. This avoids out-of-bounds array accesses in
1641 _int_malloc after constant propagation of the size parameter.
1642 (The code never executes because malloc preserves the
1643 global_max_fast invariant, but the optimizers may not recognize
1644 this.) */
1645 if (global_max_fast > MAX_FAST_SIZE)
1646 __builtin_unreachable ();
1647 return global_max_fast;
1648}
1649
1650/*
1651 ----------- Internal state representation and initialization -----------
1652 */
1653
1654/*
1655 have_fastchunks indicates that there are probably some fastbin chunks.
1656 It is set true on entering a chunk into any fastbin, and cleared early in
1657 malloc_consolidate. The value is approximate since it may be set when there
1658 are no fastbin chunks, or it may be clear even if there are fastbin chunks
1659 available. Given it's sole purpose is to reduce number of redundant calls to
1660 malloc_consolidate, it does not affect correctness. As a result we can safely
1661 use relaxed atomic accesses.
1662 */
1663
1664
1665struct malloc_state
1666{
1667 /* Serialize access. */
1668 __libc_lock_define (, mutex);
1669
1670 /* Flags (formerly in max_fast). */
1671 int flags;
1672
1673 /* Set if the fastbin chunks contain recently inserted free blocks. */
1674 /* Note this is a bool but not all targets support atomics on booleans. */
1675 int have_fastchunks;
1676
1677 /* Fastbins */
1678 mfastbinptr fastbinsY[NFASTBINS];
1679
1680 /* Base of the topmost chunk -- not otherwise kept in a bin */
1681 mchunkptr top;
1682
1683 /* The remainder from the most recent split of a small request */
1684 mchunkptr last_remainder;
1685
1686 /* Normal bins packed as described above */
1687 mchunkptr bins[NBINS * 2 - 2];
1688
1689 /* Bitmap of bins */
1690 unsigned int binmap[BINMAPSIZE];
1691
1692 /* Linked list */
1693 struct malloc_state *next;
1694
1695 /* Linked list for free arenas. Access to this field is serialized
1696 by free_list_lock in arena.c. */
1697 struct malloc_state *next_free;
1698
1699 /* Number of threads attached to this arena. 0 if the arena is on
1700 the free list. Access to this field is serialized by
1701 free_list_lock in arena.c. */
1702 INTERNAL_SIZE_T attached_threads;
1703
1704 /* Memory allocated from the system in this arena. */
1705 INTERNAL_SIZE_T system_mem;
1706 INTERNAL_SIZE_T max_system_mem;
1707};
1708
1709struct malloc_par
1710{
1711 /* Tunable parameters */
1712 unsigned long trim_threshold;
1713 INTERNAL_SIZE_T top_pad;
1714 INTERNAL_SIZE_T mmap_threshold;
1715 INTERNAL_SIZE_T arena_test;
1716 INTERNAL_SIZE_T arena_max;
1717
1718 /* Memory map support */
1719 int n_mmaps;
1720 int n_mmaps_max;
1721 int max_n_mmaps;
1722 /* the mmap_threshold is dynamic, until the user sets
1723 it manually, at which point we need to disable any
1724 dynamic behavior. */
1725 int no_dyn_threshold;
1726
1727 /* Statistics */
1728 INTERNAL_SIZE_T mmapped_mem;
1729 INTERNAL_SIZE_T max_mmapped_mem;
1730
1731 /* First address handed out by MORECORE/sbrk. */
1732 char *sbrk_base;
1733
1734#if USE_TCACHE
1735 /* Maximum number of buckets to use. */
1736 size_t tcache_bins;
1737 size_t tcache_max_bytes;
1738 /* Maximum number of chunks in each bucket. */
1739 size_t tcache_count;
1740 /* Maximum number of chunks to remove from the unsorted list, which
1741 aren't used to prefill the cache. */
1742 size_t tcache_unsorted_limit;
1743#endif
1744};
1745
1746/* There are several instances of this struct ("arenas") in this
1747 malloc. If you are adapting this malloc in a way that does NOT use
1748 a static or mmapped malloc_state, you MUST explicitly zero-fill it
1749 before using. This malloc relies on the property that malloc_state
1750 is initialized to all zeroes (as is true of C statics). */
1751
1752static struct malloc_state main_arena =
1753{
1754 .mutex = _LIBC_LOCK_INITIALIZER,
1755 .next = &main_arena,
1756 .attached_threads = 1
1757};
1758
1759/* These variables are used for undumping support. Chunked are marked
1760 as using mmap, but we leave them alone if they fall into this
1761 range. NB: The chunk size for these chunks only includes the
1762 initial size field (of SIZE_SZ bytes), there is no trailing size
1763 field (unlike with regular mmapped chunks). */
1764static mchunkptr dumped_main_arena_start; /* Inclusive. */
1765static mchunkptr dumped_main_arena_end; /* Exclusive. */
1766
1767/* True if the pointer falls into the dumped arena. Use this after
1768 chunk_is_mmapped indicates a chunk is mmapped. */
1769#define DUMPED_MAIN_ARENA_CHUNK(p) \
1770 ((p) >= dumped_main_arena_start && (p) < dumped_main_arena_end)
1771
1772/* There is only one instance of the malloc parameters. */
1773
1774static struct malloc_par mp_ =
1775{
1776 .top_pad = DEFAULT_TOP_PAD,
1777 .n_mmaps_max = DEFAULT_MMAP_MAX,
1778 .mmap_threshold = DEFAULT_MMAP_THRESHOLD,
1779 .trim_threshold = DEFAULT_TRIM_THRESHOLD,
1780#define NARENAS_FROM_NCORES(n) ((n) * (sizeof (long) == 4 ? 2 : 8))
1781 .arena_test = NARENAS_FROM_NCORES (1)
1782#if USE_TCACHE
1783 ,
1784 .tcache_count = TCACHE_FILL_COUNT,
1785 .tcache_bins = TCACHE_MAX_BINS,
1786 .tcache_max_bytes = tidx2usize (TCACHE_MAX_BINS-1),
1787 .tcache_unsorted_limit = 0 /* No limit. */
1788#endif
1789};
1790
1791/*
1792 Initialize a malloc_state struct.
1793
1794 This is called from ptmalloc_init () or from _int_new_arena ()
1795 when creating a new arena.
1796 */
1797
1798static void
1799malloc_init_state (mstate av)
1800{
1801 int i;
1802 mbinptr bin;
1803
1804 /* Establish circular links for normal bins */
1805 for (i = 1; i < NBINS; ++i)
1806 {
1807 bin = bin_at (av, i);
1808 bin->fd = bin->bk = bin;
1809 }
1810
1811#if MORECORE_CONTIGUOUS
1812 if (av != &main_arena)
1813#endif
1814 set_noncontiguous (av);
1815 if (av == &main_arena)
1816 set_max_fast (DEFAULT_MXFAST);
1817 atomic_store_relaxed (&av->have_fastchunks, false);
1818
1819 av->top = initial_top (av);
1820}
1821
1822/*
1823 Other internal utilities operating on mstates
1824 */
1825
1826static void *sysmalloc (INTERNAL_SIZE_T, mstate);
1827static int systrim (size_t, mstate);
1828static void malloc_consolidate (mstate);
1829
1830
1831/* -------------- Early definitions for debugging hooks ---------------- */
1832
1833/* Define and initialize the hook variables. These weak definitions must
1834 appear before any use of the variables in a function (arena.c uses one). */
1835#ifndef weak_variable
1836/* In GNU libc we want the hook variables to be weak definitions to
1837 avoid a problem with Emacs. */
1838# define weak_variable weak_function
1839#endif
1840
1841/* Forward declarations. */
1842static void *malloc_hook_ini (size_t sz,
1843 const void *caller) __THROW;
1844static void *realloc_hook_ini (void *ptr, size_t sz,
1845 const void *caller) __THROW;
1846static void *memalign_hook_ini (size_t alignment, size_t sz,
1847 const void *caller) __THROW;
1848
1849#if HAVE_MALLOC_INIT_HOOK
1850void weak_variable (*__malloc_initialize_hook) (void) = NULL;
1851compat_symbol (libc, __malloc_initialize_hook,
1852 __malloc_initialize_hook, GLIBC_2_0);
1853#endif
1854
1855void weak_variable (*__free_hook) (void *__ptr,
1856 const void *) = NULL;
1857void *weak_variable (*__malloc_hook)
1858 (size_t __size, const void *) = malloc_hook_ini;
1859void *weak_variable (*__realloc_hook)
1860 (void *__ptr, size_t __size, const void *)
1861 = realloc_hook_ini;
1862void *weak_variable (*__memalign_hook)
1863 (size_t __alignment, size_t __size, const void *)
1864 = memalign_hook_ini;
1865void weak_variable (*__after_morecore_hook) (void) = NULL;
1866
1867/* This function is called from the arena shutdown hook, to free the
1868 thread cache (if it exists). */
1869static void tcache_thread_shutdown (void);
1870
1871/* ------------------ Testing support ----------------------------------*/
1872
1873static int perturb_byte;
1874
1875static void
1876alloc_perturb (char *p, size_t n)
1877{
1878 if (__glibc_unlikely (perturb_byte))
1879 memset (p, perturb_byte ^ 0xff, n);
1880}
1881
1882static void
1883free_perturb (char *p, size_t n)
1884{
1885 if (__glibc_unlikely (perturb_byte))
1886 memset (p, perturb_byte, n);
1887}
1888
1889
1890
1891#include <stap-probe.h>
1892
1893/* ------------------- Support for multiple arenas -------------------- */
1894#include "arena.c"
1895
1896/*
1897 Debugging support
1898
1899 These routines make a number of assertions about the states
1900 of data structures that should be true at all times. If any
1901 are not true, it's very likely that a user program has somehow
1902 trashed memory. (It's also possible that there is a coding error
1903 in malloc. In which case, please report it!)
1904 */
1905
1906#if !MALLOC_DEBUG
1907
1908# define check_chunk(A, P)
1909# define check_free_chunk(A, P)
1910# define check_inuse_chunk(A, P)
1911# define check_remalloced_chunk(A, P, N)
1912# define check_malloced_chunk(A, P, N)
1913# define check_malloc_state(A)
1914
1915#else
1916
1917# define check_chunk(A, P) do_check_chunk (A, P)
1918# define check_free_chunk(A, P) do_check_free_chunk (A, P)
1919# define check_inuse_chunk(A, P) do_check_inuse_chunk (A, P)
1920# define check_remalloced_chunk(A, P, N) do_check_remalloced_chunk (A, P, N)
1921# define check_malloced_chunk(A, P, N) do_check_malloced_chunk (A, P, N)
1922# define check_malloc_state(A) do_check_malloc_state (A)
1923
1924/*
1925 Properties of all chunks
1926 */
1927
1928static void
1929do_check_chunk (mstate av, mchunkptr p)
1930{
1931 unsigned long sz = chunksize (p);
1932 /* min and max possible addresses assuming contiguous allocation */
1933 char *max_address = (char *) (av->top) + chunksize (av->top);
1934 char *min_address = max_address - av->system_mem;
1935
1936 if (!chunk_is_mmapped (p))
1937 {
1938 /* Has legal address ... */
1939 if (p != av->top)
1940 {
1941 if (contiguous (av))
1942 {
1943 assert (((char *) p) >= min_address);
1944 assert (((char *) p + sz) <= ((char *) (av->top)));
1945 }
1946 }
1947 else
1948 {
1949 /* top size is always at least MINSIZE */
1950 assert ((unsigned long) (sz) >= MINSIZE);
1951 /* top predecessor always marked inuse */
1952 assert (prev_inuse (p));
1953 }
1954 }
1955 else if (!DUMPED_MAIN_ARENA_CHUNK (p))
1956 {
1957 /* address is outside main heap */
1958 if (contiguous (av) && av->top != initial_top (av))
1959 {
1960 assert (((char *) p) < min_address || ((char *) p) >= max_address);
1961 }
1962 /* chunk is page-aligned */
1963 assert (((prev_size (p) + sz) & (GLRO (dl_pagesize) - 1)) == 0);
1964 /* mem is aligned */
1965 assert (aligned_OK (chunk2mem (p)));
1966 }
1967}
1968
1969/*
1970 Properties of free chunks
1971 */
1972
1973static void
1974do_check_free_chunk (mstate av, mchunkptr p)
1975{
1976 INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE | NON_MAIN_ARENA);
1977 mchunkptr next = chunk_at_offset (p, sz);
1978
1979 do_check_chunk (av, p);
1980
1981 /* Chunk must claim to be free ... */
1982 assert (!inuse (p));
1983 assert (!chunk_is_mmapped (p));
1984
1985 /* Unless a special marker, must have OK fields */
1986 if ((unsigned long) (sz) >= MINSIZE)
1987 {
1988 assert ((sz & MALLOC_ALIGN_MASK) == 0);
1989 assert (aligned_OK (chunk2mem (p)));
1990 /* ... matching footer field */
1991 assert (prev_size (next_chunk (p)) == sz);
1992 /* ... and is fully consolidated */
1993 assert (prev_inuse (p));
1994 assert (next == av->top || inuse (next));
1995
1996 /* ... and has minimally sane links */
1997 assert (p->fd->bk == p);
1998 assert (p->bk->fd == p);
1999 }
2000 else /* markers are always of size SIZE_SZ */
2001 assert (sz == SIZE_SZ);
2002}
2003
2004/*
2005 Properties of inuse chunks
2006 */
2007
2008static void
2009do_check_inuse_chunk (mstate av, mchunkptr p)
2010{
2011 mchunkptr next;
2012
2013 do_check_chunk (av, p);
2014
2015 if (chunk_is_mmapped (p))
2016 return; /* mmapped chunks have no next/prev */
2017
2018 /* Check whether it claims to be in use ... */
2019 assert (inuse (p));
2020
2021 next = next_chunk (p);
2022
2023 /* ... and is surrounded by OK chunks.
2024 Since more things can be checked with free chunks than inuse ones,
2025 if an inuse chunk borders them and debug is on, it's worth doing them.
2026 */
2027 if (!prev_inuse (p))
2028 {
2029 /* Note that we cannot even look at prev unless it is not inuse */
2030 mchunkptr prv = prev_chunk (p);
2031 assert (next_chunk (prv) == p);
2032 do_check_free_chunk (av, prv);
2033 }
2034
2035 if (next == av->top)
2036 {
2037 assert (prev_inuse (next));
2038 assert (chunksize (next) >= MINSIZE);
2039 }
2040 else if (!inuse (next))
2041 do_check_free_chunk (av, next);
2042}
2043
2044/*
2045 Properties of chunks recycled from fastbins
2046 */
2047
2048static void
2049do_check_remalloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s)
2050{
2051 INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE | NON_MAIN_ARENA);
2052
2053 if (!chunk_is_mmapped (p))
2054 {
2055 assert (av == arena_for_chunk (p));
2056 if (chunk_main_arena (p))
2057 assert (av == &main_arena);
2058 else
2059 assert (av != &main_arena);
2060 }
2061
2062 do_check_inuse_chunk (av, p);
2063
2064 /* Legal size ... */
2065 assert ((sz & MALLOC_ALIGN_MASK) == 0);
2066 assert ((unsigned long) (sz) >= MINSIZE);
2067 /* ... and alignment */
2068 assert (aligned_OK (chunk2mem (p)));
2069 /* chunk is less than MINSIZE more than request */
2070 assert ((long) (sz) - (long) (s) >= 0);
2071 assert ((long) (sz) - (long) (s + MINSIZE) < 0);
2072}
2073
2074/*
2075 Properties of nonrecycled chunks at the point they are malloced
2076 */
2077
2078static void
2079do_check_malloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s)
2080{
2081 /* same as recycled case ... */
2082 do_check_remalloced_chunk (av, p, s);
2083
2084 /*
2085 ... plus, must obey implementation invariant that prev_inuse is
2086 always true of any allocated chunk; i.e., that each allocated
2087 chunk borders either a previously allocated and still in-use
2088 chunk, or the base of its memory arena. This is ensured
2089 by making all allocations from the `lowest' part of any found
2090 chunk. This does not necessarily hold however for chunks
2091 recycled via fastbins.
2092 */
2093
2094 assert (prev_inuse (p));
2095}
2096
2097
2098/*
2099 Properties of malloc_state.
2100
2101 This may be useful for debugging malloc, as well as detecting user
2102 programmer errors that somehow write into malloc_state.
2103
2104 If you are extending or experimenting with this malloc, you can
2105 probably figure out how to hack this routine to print out or
2106 display chunk addresses, sizes, bins, and other instrumentation.
2107 */
2108
2109static void
2110do_check_malloc_state (mstate av)
2111{
2112 int i;
2113 mchunkptr p;
2114 mchunkptr q;
2115 mbinptr b;
2116 unsigned int idx;
2117 INTERNAL_SIZE_T size;
2118 unsigned long total = 0;
2119 int max_fast_bin;
2120
2121 /* internal size_t must be no wider than pointer type */
2122 assert (sizeof (INTERNAL_SIZE_T) <= sizeof (char *));
2123
2124 /* alignment is a power of 2 */
2125 assert ((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT - 1)) == 0);
2126
2127 /* Check the arena is initialized. */
2128 assert (av->top != 0);
2129
2130 /* No memory has been allocated yet, so doing more tests is not possible. */
2131 if (av->top == initial_top (av))
2132 return;
2133
2134 /* pagesize is a power of 2 */
2135 assert (powerof2(GLRO (dl_pagesize)));
2136
2137 /* A contiguous main_arena is consistent with sbrk_base. */
2138 if (av == &main_arena && contiguous (av))
2139 assert ((char *) mp_.sbrk_base + av->system_mem ==
2140 (char *) av->top + chunksize (av->top));
2141
2142 /* properties of fastbins */
2143
2144 /* max_fast is in allowed range */
2145 assert ((get_max_fast () & ~1) <= request2size (MAX_FAST_SIZE));
2146
2147 max_fast_bin = fastbin_index (get_max_fast ());
2148
2149 for (i = 0; i < NFASTBINS; ++i)
2150 {
2151 p = fastbin (av, i);
2152
2153 /* The following test can only be performed for the main arena.
2154 While mallopt calls malloc_consolidate to get rid of all fast
2155 bins (especially those larger than the new maximum) this does
2156 only happen for the main arena. Trying to do this for any
2157 other arena would mean those arenas have to be locked and
2158 malloc_consolidate be called for them. This is excessive. And
2159 even if this is acceptable to somebody it still cannot solve
2160 the problem completely since if the arena is locked a
2161 concurrent malloc call might create a new arena which then
2162 could use the newly invalid fast bins. */
2163
2164 /* all bins past max_fast are empty */
2165 if (av == &main_arena && i > max_fast_bin)
2166 assert (p == 0);
2167
2168 while (p != 0)
2169 {
2170 /* each chunk claims to be inuse */
2171 do_check_inuse_chunk (av, p);
2172 total += chunksize (p);
2173 /* chunk belongs in this bin */
2174 assert (fastbin_index (chunksize (p)) == i);
2175 p = p->fd;
2176 }
2177 }
2178
2179 /* check normal bins */
2180 for (i = 1; i < NBINS; ++i)
2181 {
2182 b = bin_at (av, i);
2183
2184 /* binmap is accurate (except for bin 1 == unsorted_chunks) */
2185 if (i >= 2)
2186 {
2187 unsigned int binbit = get_binmap (av, i);
2188 int empty = last (b) == b;
2189 if (!binbit)
2190 assert (empty);
2191 else if (!empty)
2192 assert (binbit);
2193 }
2194
2195 for (p = last (b); p != b; p = p->bk)
2196 {
2197 /* each chunk claims to be free */
2198 do_check_free_chunk (av, p);
2199 size = chunksize (p);
2200 total += size;
2201 if (i >= 2)
2202 {
2203 /* chunk belongs in bin */
2204 idx = bin_index (size);
2205 assert (idx == i);
2206 /* lists are sorted */
2207 assert (p->bk == b ||
2208 (unsigned long) chunksize (p->bk) >= (unsigned long) chunksize (p));
2209
2210 if (!in_smallbin_range (size))
2211 {
2212 if (p->fd_nextsize != NULL)
2213 {
2214 if (p->fd_nextsize == p)
2215 assert (p->bk_nextsize == p);
2216 else
2217 {
2218 if (p->fd_nextsize == first (b))
2219 assert (chunksize (p) < chunksize (p->fd_nextsize));
2220 else
2221 assert (chunksize (p) > chunksize (p->fd_nextsize));
2222
2223 if (p == first (b))
2224 assert (chunksize (p) > chunksize (p->bk_nextsize));
2225 else
2226 assert (chunksize (p) < chunksize (p->bk_nextsize));
2227 }
2228 }
2229 else
2230 assert (p->bk_nextsize == NULL);
2231 }
2232 }
2233 else if (!in_smallbin_range (size))
2234 assert (p->fd_nextsize == NULL && p->bk_nextsize == NULL);
2235 /* chunk is followed by a legal chain of inuse chunks */
2236 for (q = next_chunk (p);
2237 (q != av->top && inuse (q) &&
2238 (unsigned long) (chunksize (q)) >= MINSIZE);
2239 q = next_chunk (q))
2240 do_check_inuse_chunk (av, q);
2241 }
2242 }
2243
2244 /* top chunk is OK */
2245 check_chunk (av, av->top);
2246}
2247#endif
2248
2249
2250/* ----------------- Support for debugging hooks -------------------- */
2251#include "hooks.c"
2252
2253
2254/* ----------- Routines dealing with system allocation -------------- */
2255
2256/*
2257 sysmalloc handles malloc cases requiring more memory from the system.
2258 On entry, it is assumed that av->top does not have enough
2259 space to service request for nb bytes, thus requiring that av->top
2260 be extended or replaced.
2261 */
2262
2263static void *
2264sysmalloc (INTERNAL_SIZE_T nb, mstate av)
2265{
2266 mchunkptr old_top; /* incoming value of av->top */
2267 INTERNAL_SIZE_T old_size; /* its size */
2268 char *old_end; /* its end address */
2269
2270 long size; /* arg to first MORECORE or mmap call */
2271 char *brk; /* return value from MORECORE */
2272
2273 long correction; /* arg to 2nd MORECORE call */
2274 char *snd_brk; /* 2nd return val */
2275
2276 INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of new space */
2277 INTERNAL_SIZE_T end_misalign; /* partial page left at end of new space */
2278 char *aligned_brk; /* aligned offset into brk */
2279
2280 mchunkptr p; /* the allocated/returned chunk */
2281 mchunkptr remainder; /* remainder from allocation */
2282 unsigned long remainder_size; /* its size */
2283
2284
2285 size_t pagesize = GLRO (dl_pagesize);
2286 bool tried_mmap = false;
2287
2288
2289 /*
2290 If have mmap, and the request size meets the mmap threshold, and
2291 the system supports mmap, and there are few enough currently
2292 allocated mmapped regions, try to directly map this request
2293 rather than expanding top.
2294 */
2295
2296 if (av == NULL
2297 || ((unsigned long) (nb) >= (unsigned long) (mp_.mmap_threshold)
2298 && (mp_.n_mmaps < mp_.n_mmaps_max)))
2299 {
2300 char *mm; /* return value from mmap call*/
2301
2302 try_mmap:
2303 /*
2304 Round up size to nearest page. For mmapped chunks, the overhead
2305 is one SIZE_SZ unit larger than for normal chunks, because there
2306 is no following chunk whose prev_size field could be used.
2307
2308 See the front_misalign handling below, for glibc there is no
2309 need for further alignments unless we have have high alignment.
2310 */
2311 if (MALLOC_ALIGNMENT == 2 * SIZE_SZ)
2312 size = ALIGN_UP (nb + SIZE_SZ, pagesize);
2313 else
2314 size = ALIGN_UP (nb + SIZE_SZ + MALLOC_ALIGN_MASK, pagesize);
2315 tried_mmap = true;
2316
2317 /* Don't try if size wraps around 0 */
2318 if ((unsigned long) (size) > (unsigned long) (nb))
2319 {
2320 mm = (char *) (MMAP (0, size, PROT_READ | PROT_WRITE, 0));
2321
2322 if (mm != MAP_FAILED)
2323 {
2324 /*
2325 The offset to the start of the mmapped region is stored
2326 in the prev_size field of the chunk. This allows us to adjust
2327 returned start address to meet alignment requirements here
2328 and in memalign(), and still be able to compute proper
2329 address argument for later munmap in free() and realloc().
2330 */
2331
2332 if (MALLOC_ALIGNMENT == 2 * SIZE_SZ)
2333 {
2334 /* For glibc, chunk2mem increases the address by 2*SIZE_SZ and
2335 MALLOC_ALIGN_MASK is 2*SIZE_SZ-1. Each mmap'ed area is page
2336 aligned and therefore definitely MALLOC_ALIGN_MASK-aligned. */
2337 assert (((INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK) == 0);
2338 front_misalign = 0;
2339 }
2340 else
2341 front_misalign = (INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK;
2342 if (front_misalign > 0)
2343 {
2344 correction = MALLOC_ALIGNMENT - front_misalign;
2345 p = (mchunkptr) (mm + correction);
2346 set_prev_size (p, correction);
2347 set_head (p, (size - correction) | IS_MMAPPED);
2348 }
2349 else
2350 {
2351 p = (mchunkptr) mm;
2352 set_prev_size (p, 0);
2353 set_head (p, size | IS_MMAPPED);
2354 }
2355
2356 /* update statistics */
2357
2358 int new = atomic_exchange_and_add (&mp_.n_mmaps, 1) + 1;
2359 atomic_max (&mp_.max_n_mmaps, new);
2360
2361 unsigned long sum;
2362 sum = atomic_exchange_and_add (&mp_.mmapped_mem, size) + size;
2363 atomic_max (&mp_.max_mmapped_mem, sum);
2364
2365 check_chunk (av, p);
2366
2367 return chunk2mem (p);
2368 }
2369 }
2370 }
2371
2372 /* There are no usable arenas and mmap also failed. */
2373 if (av == NULL)
2374 return 0;
2375
2376 /* Record incoming configuration of top */
2377
2378 old_top = av->top;
2379 old_size = chunksize (old_top);
2380 old_end = (char *) (chunk_at_offset (old_top, old_size));
2381
2382 brk = snd_brk = (char *) (MORECORE_FAILURE);
2383
2384 /*
2385 If not the first time through, we require old_size to be
2386 at least MINSIZE and to have prev_inuse set.
2387 */
2388
2389 assert ((old_top == initial_top (av) && old_size == 0) ||
2390 ((unsigned long) (old_size) >= MINSIZE &&
2391 prev_inuse (old_top) &&
2392 ((unsigned long) old_end & (pagesize - 1)) == 0));
2393
2394 /* Precondition: not enough current space to satisfy nb request */
2395 assert ((unsigned long) (old_size) < (unsigned long) (nb + MINSIZE));
2396
2397
2398 if (av != &main_arena)
2399 {
2400 heap_info *old_heap, *heap;
2401 size_t old_heap_size;
2402
2403 /* First try to extend the current heap. */
2404 old_heap = heap_for_ptr (old_top);
2405 old_heap_size = old_heap->size;
2406 if ((long) (MINSIZE + nb - old_size) > 0
2407 && grow_heap (old_heap, MINSIZE + nb - old_size) == 0)
2408 {
2409 av->system_mem += old_heap->size - old_heap_size;
2410 set_head (old_top, (((char *) old_heap + old_heap->size) - (char *) old_top)
2411 | PREV_INUSE);
2412 }
2413 else if ((heap = new_heap (nb + (MINSIZE + sizeof (*heap)), mp_.top_pad)))
2414 {
2415 /* Use a newly allocated heap. */
2416 heap->ar_ptr = av;
2417 heap->prev = old_heap;
2418 av->system_mem += heap->size;
2419 /* Set up the new top. */
2420 top (av) = chunk_at_offset (heap, sizeof (*heap));
2421 set_head (top (av), (heap->size - sizeof (*heap)) | PREV_INUSE);
2422
2423 /* Setup fencepost and free the old top chunk with a multiple of
2424 MALLOC_ALIGNMENT in size. */
2425 /* The fencepost takes at least MINSIZE bytes, because it might
2426 become the top chunk again later. Note that a footer is set
2427 up, too, although the chunk is marked in use. */
2428 old_size = (old_size - MINSIZE) & ~MALLOC_ALIGN_MASK;
2429 set_head (chunk_at_offset (old_top, old_size + 2 * SIZE_SZ), 0 | PREV_INUSE);
2430 if (old_size >= MINSIZE)
2431 {
2432 set_head (chunk_at_offset (old_top, old_size), (2 * SIZE_SZ) | PREV_INUSE);
2433 set_foot (chunk_at_offset (old_top, old_size), (2 * SIZE_SZ));
2434 set_head (old_top, old_size | PREV_INUSE | NON_MAIN_ARENA);
2435 _int_free (av, old_top, 1);
2436 }
2437 else
2438 {
2439 set_head (old_top, (old_size + 2 * SIZE_SZ) | PREV_INUSE);
2440 set_foot (old_top, (old_size + 2 * SIZE_SZ));
2441 }
2442 }
2443 else if (!tried_mmap)
2444 /* We can at least try to use to mmap memory. */
2445 goto try_mmap;
2446 }
2447 else /* av == main_arena */
2448
2449
2450 { /* Request enough space for nb + pad + overhead */
2451 size = nb + mp_.top_pad + MINSIZE;
2452
2453 /*
2454 If contiguous, we can subtract out existing space that we hope to
2455 combine with new space. We add it back later only if
2456 we don't actually get contiguous space.
2457 */
2458
2459 if (contiguous (av))
2460 size -= old_size;
2461
2462 /*
2463 Round to a multiple of page size.
2464 If MORECORE is not contiguous, this ensures that we only call it
2465 with whole-page arguments. And if MORECORE is contiguous and
2466 this is not first time through, this preserves page-alignment of
2467 previous calls. Otherwise, we correct to page-align below.
2468 */
2469
2470 size = ALIGN_UP (size, pagesize);
2471
2472 /*
2473 Don't try to call MORECORE if argument is so big as to appear
2474 negative. Note that since mmap takes size_t arg, it may succeed
2475 below even if we cannot call MORECORE.
2476 */
2477
2478 if (size > 0)
2479 {
2480 brk = (char *) (MORECORE (size));
2481 LIBC_PROBE (memory_sbrk_more, 2, brk, size);
2482 }
2483
2484 if (brk != (char *) (MORECORE_FAILURE))
2485 {
2486 /* Call the `morecore' hook if necessary. */
2487 void (*hook) (void) = atomic_forced_read (__after_morecore_hook);
2488 if (__builtin_expect (hook != NULL, 0))
2489 (*hook)();
2490 }
2491 else
2492 {
2493 /*
2494 If have mmap, try using it as a backup when MORECORE fails or
2495 cannot be used. This is worth doing on systems that have "holes" in
2496 address space, so sbrk cannot extend to give contiguous space, but
2497 space is available elsewhere. Note that we ignore mmap max count
2498 and threshold limits, since the space will not be used as a
2499 segregated mmap region.
2500 */
2501
2502 /* Cannot merge with old top, so add its size back in */
2503 if (contiguous (av))
2504 size = ALIGN_UP (size + old_size, pagesize);
2505
2506 /* If we are relying on mmap as backup, then use larger units */
2507 if ((unsigned long) (size) < (unsigned long) (MMAP_AS_MORECORE_SIZE))
2508 size = MMAP_AS_MORECORE_SIZE;
2509
2510 /* Don't try if size wraps around 0 */
2511 if ((unsigned long) (size) > (unsigned long) (nb))
2512 {
2513 char *mbrk = (char *) (MMAP (0, size, PROT_READ | PROT_WRITE, 0));
2514
2515 if (mbrk != MAP_FAILED)
2516 {
2517 /* We do not need, and cannot use, another sbrk call to find end */
2518 brk = mbrk;
2519 snd_brk = brk + size;
2520
2521 /*
2522 Record that we no longer have a contiguous sbrk region.
2523 After the first time mmap is used as backup, we do not
2524 ever rely on contiguous space since this could incorrectly
2525 bridge regions.
2526 */
2527 set_noncontiguous (av);
2528 }
2529 }
2530 }
2531
2532 if (brk != (char *) (MORECORE_FAILURE))
2533 {
2534 if (mp_.sbrk_base == 0)
2535 mp_.sbrk_base = brk;
2536 av->system_mem += size;
2537
2538 /*
2539 If MORECORE extends previous space, we can likewise extend top size.
2540 */
2541
2542 if (brk == old_end && snd_brk == (char *) (MORECORE_FAILURE))
2543 set_head (old_top, (size + old_size) | PREV_INUSE);
2544
2545 else if (contiguous (av) && old_size && brk < old_end)
2546 /* Oops! Someone else killed our space.. Can't touch anything. */
2547 malloc_printerr ("break adjusted to free malloc space");
2548
2549 /*
2550 Otherwise, make adjustments:
2551
2552 * If the first time through or noncontiguous, we need to call sbrk
2553 just to find out where the end of memory lies.
2554
2555 * We need to ensure that all returned chunks from malloc will meet
2556 MALLOC_ALIGNMENT
2557
2558 * If there was an intervening foreign sbrk, we need to adjust sbrk
2559 request size to account for fact that we will not be able to
2560 combine new space with existing space in old_top.
2561
2562 * Almost all systems internally allocate whole pages at a time, in
2563 which case we might as well use the whole last page of request.
2564 So we allocate enough more memory to hit a page boundary now,
2565 which in turn causes future contiguous calls to page-align.
2566 */
2567
2568 else
2569 {
2570 front_misalign = 0;
2571 end_misalign = 0;
2572 correction = 0;
2573 aligned_brk = brk;
2574
2575 /* handle contiguous cases */
2576 if (contiguous (av))
2577 {
2578 /* Count foreign sbrk as system_mem. */
2579 if (old_size)
2580 av->system_mem += brk - old_end;
2581
2582 /* Guarantee alignment of first new chunk made from this space */
2583
2584 front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK;
2585 if (front_misalign > 0)
2586 {
2587 /*
2588 Skip over some bytes to arrive at an aligned position.
2589 We don't need to specially mark these wasted front bytes.
2590 They will never be accessed anyway because
2591 prev_inuse of av->top (and any chunk created from its start)
2592 is always true after initialization.
2593 */
2594
2595 correction = MALLOC_ALIGNMENT - front_misalign;
2596 aligned_brk += correction;
2597 }
2598
2599 /*
2600 If this isn't adjacent to existing space, then we will not
2601 be able to merge with old_top space, so must add to 2nd request.
2602 */
2603
2604 correction += old_size;
2605
2606 /* Extend the end address to hit a page boundary */
2607 end_misalign = (INTERNAL_SIZE_T) (brk + size + correction);
2608 correction += (ALIGN_UP (end_misalign, pagesize)) - end_misalign;
2609
2610 assert (correction >= 0);
2611 snd_brk = (char *) (MORECORE (correction));
2612
2613 /*
2614 If can't allocate correction, try to at least find out current
2615 brk. It might be enough to proceed without failing.
2616
2617 Note that if second sbrk did NOT fail, we assume that space
2618 is contiguous with first sbrk. This is a safe assumption unless
2619 program is multithreaded but doesn't use locks and a foreign sbrk
2620 occurred between our first and second calls.
2621 */
2622
2623 if (snd_brk == (char *) (MORECORE_FAILURE))
2624 {
2625 correction = 0;
2626 snd_brk = (char *) (MORECORE (0));
2627 }
2628 else
2629 {
2630 /* Call the `morecore' hook if necessary. */
2631 void (*hook) (void) = atomic_forced_read (__after_morecore_hook);
2632 if (__builtin_expect (hook != NULL, 0))
2633 (*hook)();
2634 }
2635 }
2636
2637 /* handle non-contiguous cases */
2638 else
2639 {
2640 if (MALLOC_ALIGNMENT == 2 * SIZE_SZ)
2641 /* MORECORE/mmap must correctly align */
2642 assert (((unsigned long) chunk2mem (brk) & MALLOC_ALIGN_MASK) == 0);
2643 else
2644 {
2645 front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK;
2646 if (front_misalign > 0)
2647 {
2648 /*
2649 Skip over some bytes to arrive at an aligned position.
2650 We don't need to specially mark these wasted front bytes.
2651 They will never be accessed anyway because
2652 prev_inuse of av->top (and any chunk created from its start)
2653 is always true after initialization.
2654 */
2655
2656 aligned_brk += MALLOC_ALIGNMENT - front_misalign;
2657 }
2658 }
2659
2660 /* Find out current end of memory */
2661 if (snd_brk == (char *) (MORECORE_FAILURE))
2662 {
2663 snd_brk = (char *) (MORECORE (0));
2664 }
2665 }
2666
2667 /* Adjust top based on results of second sbrk */
2668 if (snd_brk != (char *) (MORECORE_FAILURE))
2669 {
2670 av->top = (mchunkptr) aligned_brk;
2671 set_head (av->top, (snd_brk - aligned_brk + correction) | PREV_INUSE);
2672 av->system_mem += correction;
2673
2674 /*
2675 If not the first time through, we either have a
2676 gap due to foreign sbrk or a non-contiguous region. Insert a
2677 double fencepost at old_top to prevent consolidation with space
2678 we don't own. These fenceposts are artificial chunks that are
2679 marked as inuse and are in any case too small to use. We need
2680 two to make sizes and alignments work out.
2681 */
2682
2683 if (old_size != 0)
2684 {
2685 /*
2686 Shrink old_top to insert fenceposts, keeping size a
2687 multiple of MALLOC_ALIGNMENT. We know there is at least
2688 enough space in old_top to do this.
2689 */
2690 old_size = (old_size - 4 * SIZE_SZ) & ~MALLOC_ALIGN_MASK;
2691 set_head (old_top, old_size | PREV_INUSE);
2692
2693 /*
2694 Note that the following assignments completely overwrite
2695 old_top when old_size was previously MINSIZE. This is
2696 intentional. We need the fencepost, even if old_top otherwise gets
2697 lost.
2698 */
2699 set_head (chunk_at_offset (old_top, old_size),
2700 (2 * SIZE_SZ) | PREV_INUSE);
2701 set_head (chunk_at_offset (old_top, old_size + 2 * SIZE_SZ),
2702 (2 * SIZE_SZ) | PREV_INUSE);
2703
2704 /* If possible, release the rest. */
2705 if (old_size >= MINSIZE)
2706 {
2707 _int_free (av, old_top, 1);
2708 }
2709 }
2710 }
2711 }
2712 }
2713 } /* if (av != &main_arena) */
2714
2715 if ((unsigned long) av->system_mem > (unsigned long) (av->max_system_mem))
2716 av->max_system_mem = av->system_mem;
2717 check_malloc_state (av);
2718
2719 /* finally, do the allocation */
2720 p = av->top;
2721 size = chunksize (p);
2722
2723 /* check that one of the above allocation paths succeeded */
2724 if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE))
2725 {
2726 remainder_size = size - nb;
2727 remainder = chunk_at_offset (p, nb);
2728 av->top = remainder;
2729 set_head (p, nb | PREV_INUSE | (av != &main_arena ? NON_MAIN_ARENA : 0));
2730 set_head (remainder, remainder_size | PREV_INUSE);
2731 check_malloced_chunk (av, p, nb);
2732 return chunk2mem (p);
2733 }
2734
2735 /* catch all failure paths */
2736 __set_errno (ENOMEM);
2737 return 0;
2738}
2739
2740
2741/*
2742 systrim is an inverse of sorts to sysmalloc. It gives memory back
2743 to the system (via negative arguments to sbrk) if there is unused
2744 memory at the `high' end of the malloc pool. It is called
2745 automatically by free() when top space exceeds the trim
2746 threshold. It is also called by the public malloc_trim routine. It
2747 returns 1 if it actually released any memory, else 0.
2748 */
2749
2750static int
2751systrim (size_t pad, mstate av)
2752{
2753 long top_size; /* Amount of top-most memory */
2754 long extra; /* Amount to release */
2755 long released; /* Amount actually released */
2756 char *current_brk; /* address returned by pre-check sbrk call */
2757 char *new_brk; /* address returned by post-check sbrk call */
2758 size_t pagesize;
2759 long top_area;
2760
2761 pagesize = GLRO (dl_pagesize);
2762 top_size = chunksize (av->top);
2763
2764 top_area = top_size - MINSIZE - 1;
2765 if (top_area <= pad)
2766 return 0;
2767
2768 /* Release in pagesize units and round down to the nearest page. */
2769 extra = ALIGN_DOWN(top_area - pad, pagesize);
2770
2771 if (extra == 0)
2772 return 0;
2773
2774 /*
2775 Only proceed if end of memory is where we last set it.
2776 This avoids problems if there were foreign sbrk calls.
2777 */
2778 current_brk = (char *) (MORECORE (0));
2779 if (current_brk == (char *) (av->top) + top_size)
2780 {
2781 /*
2782 Attempt to release memory. We ignore MORECORE return value,
2783 and instead call again to find out where new end of memory is.
2784 This avoids problems if first call releases less than we asked,
2785 of if failure somehow altered brk value. (We could still
2786 encounter problems if it altered brk in some very bad way,
2787 but the only thing we can do is adjust anyway, which will cause
2788 some downstream failure.)
2789 */
2790
2791 MORECORE (-extra);
2792 /* Call the `morecore' hook if necessary. */
2793 void (*hook) (void) = atomic_forced_read (__after_morecore_hook);
2794 if (__builtin_expect (hook != NULL, 0))
2795 (*hook)();
2796 new_brk = (char *) (MORECORE (0));
2797
2798 LIBC_PROBE (memory_sbrk_less, 2, new_brk, extra);
2799
2800 if (new_brk != (char *) MORECORE_FAILURE)
2801 {
2802 released = (long) (current_brk - new_brk);
2803
2804 if (released != 0)
2805 {
2806 /* Success. Adjust top. */
2807 av->system_mem -= released;
2808 set_head (av->top, (top_size - released) | PREV_INUSE);
2809 check_malloc_state (av);
2810 return 1;
2811 }
2812 }
2813 }
2814 return 0;
2815}
2816
2817static void
2818munmap_chunk (mchunkptr p)
2819{
2820 size_t pagesize = GLRO (dl_pagesize);
2821 INTERNAL_SIZE_T size = chunksize (p);
2822
2823 assert (chunk_is_mmapped (p));
2824
2825 /* Do nothing if the chunk is a faked mmapped chunk in the dumped
2826 main arena. We never free this memory. */
2827 if (DUMPED_MAIN_ARENA_CHUNK (p))
2828 return;
2829
2830 uintptr_t mem = (uintptr_t) chunk2mem (p);
2831 uintptr_t block = (uintptr_t) p - prev_size (p);
2832 size_t total_size = prev_size (p) + size;
2833 /* Unfortunately we have to do the compilers job by hand here. Normally
2834 we would test BLOCK and TOTAL-SIZE separately for compliance with the
2835 page size. But gcc does not recognize the optimization possibility
2836 (in the moment at least) so we combine the two values into one before
2837 the bit test. */
2838 if (__glibc_unlikely ((block | total_size) & (pagesize - 1)) != 0
2839 || __glibc_unlikely (!powerof2 (mem & (pagesize - 1))))
2840 malloc_printerr ("munmap_chunk(): invalid pointer");
2841
2842 atomic_decrement (&mp_.n_mmaps);
2843 atomic_add (&mp_.mmapped_mem, -total_size);
2844
2845 /* If munmap failed the process virtual memory address space is in a
2846 bad shape. Just leave the block hanging around, the process will
2847 terminate shortly anyway since not much can be done. */
2848 __munmap ((char *) block, total_size);
2849}
2850
2851#if HAVE_MREMAP
2852
2853static mchunkptr
2854mremap_chunk (mchunkptr p, size_t new_size)
2855{
2856 size_t pagesize = GLRO (dl_pagesize);
2857 INTERNAL_SIZE_T offset = prev_size (p);
2858 INTERNAL_SIZE_T size = chunksize (p);
2859 char *cp;
2860
2861 assert (chunk_is_mmapped (p));
2862
2863 uintptr_t block = (uintptr_t) p - offset;
2864 uintptr_t mem = (uintptr_t) chunk2mem(p);
2865 size_t total_size = offset + size;
2866 if (__glibc_unlikely ((block | total_size) & (pagesize - 1)) != 0
2867 || __glibc_unlikely (!powerof2 (mem & (pagesize - 1))))
2868 malloc_printerr("mremap_chunk(): invalid pointer");
2869
2870 /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */
2871 new_size = ALIGN_UP (new_size + offset + SIZE_SZ, pagesize);
2872
2873 /* No need to remap if the number of pages does not change. */
2874 if (total_size == new_size)
2875 return p;
2876
2877 cp = (char *) __mremap ((char *) block, total_size, new_size,
2878 MREMAP_MAYMOVE);
2879
2880 if (cp == MAP_FAILED)
2881 return 0;
2882
2883 p = (mchunkptr) (cp + offset);
2884
2885 assert (aligned_OK (chunk2mem (p)));
2886
2887 assert (prev_size (p) == offset);
2888 set_head (p, (new_size - offset) | IS_MMAPPED);
2889
2890 INTERNAL_SIZE_T new;
2891 new = atomic_exchange_and_add (&mp_.mmapped_mem, new_size - size - offset)
2892 + new_size - size - offset;
2893 atomic_max (&mp_.max_mmapped_mem, new);
2894 return p;
2895}
2896#endif /* HAVE_MREMAP */
2897
2898/*------------------------ Public wrappers. --------------------------------*/
2899
2900#if USE_TCACHE
2901
2902/* We overlay this structure on the user-data portion of a chunk when
2903 the chunk is stored in the per-thread cache. */
2904typedef struct tcache_entry
2905{
2906 struct tcache_entry *next;
2907 /* This field exists to detect double frees. */
2908 struct tcache_perthread_struct *key;
2909} tcache_entry;
2910
2911/* There is one of these for each thread, which contains the
2912 per-thread cache (hence "tcache_perthread_struct"). Keeping
2913 overall size low is mildly important. Note that COUNTS and ENTRIES
2914 are redundant (we could have just counted the linked list each
2915 time), this is for performance reasons. */
2916typedef struct tcache_perthread_struct
2917{
2918 char counts[TCACHE_MAX_BINS];
2919 tcache_entry *entries[TCACHE_MAX_BINS];
2920} tcache_perthread_struct;
2921
2922static __thread bool tcache_shutting_down = false;
2923static __thread tcache_perthread_struct *tcache = NULL;
2924
2925/* Caller must ensure that we know tc_idx is valid and there's room
2926 for more chunks. */
2927static __always_inline void
2928tcache_put (mchunkptr chunk, size_t tc_idx)
2929{
2930 tcache_entry *e = (tcache_entry *) chunk2mem (chunk);
2931 assert (tc_idx < TCACHE_MAX_BINS);
2932
2933 /* Mark this chunk as "in the tcache" so the test in _int_free will
2934 detect a double free. */
2935 e->key = tcache;
2936
2937 e->next = tcache->entries[tc_idx];
2938 tcache->entries[tc_idx] = e;
2939 ++(tcache->counts[tc_idx]);
2940}
2941
2942/* Caller must ensure that we know tc_idx is valid and there's
2943 available chunks to remove. */
2944static __always_inline void *
2945tcache_get (size_t tc_idx)
2946{
2947 tcache_entry *e = tcache->entries[tc_idx];
2948 assert (tc_idx < TCACHE_MAX_BINS);
2949 assert (tcache->counts[tc_idx] > 0);
2950 tcache->entries[tc_idx] = e->next;
2951 --(tcache->counts[tc_idx]);
2952 e->key = NULL;
2953 return (void *) e;
2954}
2955
2956static void
2957tcache_thread_shutdown (void)
2958{
2959 int i;
2960 tcache_perthread_struct *tcache_tmp = tcache;
2961
2962 if (!tcache)
2963 return;
2964
2965 /* Disable the tcache and prevent it from being reinitialized. */
2966 tcache = NULL;
2967 tcache_shutting_down = true;
2968
2969 /* Free all of the entries and the tcache itself back to the arena
2970 heap for coalescing. */
2971 for (i = 0; i < TCACHE_MAX_BINS; ++i)
2972 {
2973 while (tcache_tmp->entries[i])
2974 {
2975 tcache_entry *e = tcache_tmp->entries[i];
2976 tcache_tmp->entries[i] = e->next;
2977 __libc_free (e);
2978 }
2979 }
2980
2981 __libc_free (tcache_tmp);
2982}
2983
2984static void
2985tcache_init(void)
2986{
2987 mstate ar_ptr;
2988 void *victim = 0;
2989 const size_t bytes = sizeof (tcache_perthread_struct);
2990
2991 if (tcache_shutting_down)
2992 return;
2993
2994 arena_get (ar_ptr, bytes);
2995 victim = _int_malloc (ar_ptr, bytes);
2996 if (!victim && ar_ptr != NULL)
2997 {
2998 ar_ptr = arena_get_retry (ar_ptr, bytes);
2999 victim = _int_malloc (ar_ptr, bytes);
3000 }
3001
3002
3003 if (ar_ptr != NULL)
3004 __libc_lock_unlock (ar_ptr->mutex);
3005
3006 /* In a low memory situation, we may not be able to allocate memory
3007 - in which case, we just keep trying later. However, we
3008 typically do this very early, so either there is sufficient
3009 memory, or there isn't enough memory to do non-trivial
3010 allocations anyway. */
3011 if (victim)
3012 {
3013 tcache = (tcache_perthread_struct *) victim;
3014 memset (tcache, 0, sizeof (tcache_perthread_struct));
3015 }
3016
3017}
3018
3019# define MAYBE_INIT_TCACHE() \
3020 if (__glibc_unlikely (tcache == NULL)) \
3021 tcache_init();
3022
3023#else /* !USE_TCACHE */
3024# define MAYBE_INIT_TCACHE()
3025
3026static void
3027tcache_thread_shutdown (void)
3028{
3029 /* Nothing to do if there is no thread cache. */
3030}
3031
3032#endif /* !USE_TCACHE */
3033
3034void *
3035__libc_malloc (size_t bytes)
3036{
3037 mstate ar_ptr;
3038 void *victim;
3039
3040 void *(*hook) (size_t, const void *)
3041 = atomic_forced_read (__malloc_hook);
3042 if (__builtin_expect (hook != NULL, 0))
3043 return (*hook)(bytes, RETURN_ADDRESS (0));
3044#if USE_TCACHE
3045 /* int_free also calls request2size, be careful to not pad twice. */
3046 size_t tbytes;
3047 checked_request2size (bytes, tbytes);
3048 size_t tc_idx = csize2tidx (tbytes);
3049
3050 MAYBE_INIT_TCACHE ();
3051
3052 DIAG_PUSH_NEEDS_COMMENT;
3053 if (tc_idx < mp_.tcache_bins
3054 /*&& tc_idx < TCACHE_MAX_BINS*/ /* to appease gcc */
3055 && tcache
3056 && tcache->entries[tc_idx] != NULL)
3057 {
3058 return tcache_get (tc_idx);
3059 }
3060 DIAG_POP_NEEDS_COMMENT;
3061#endif
3062
3063 if (SINGLE_THREAD_P)
3064 {
3065 victim = _int_malloc (&main_arena, bytes);
3066 assert (!victim || chunk_is_mmapped (mem2chunk (victim)) ||
3067 &main_arena == arena_for_chunk (mem2chunk (victim)));
3068 return victim;
3069 }
3070
3071 arena_get (ar_ptr, bytes);
3072
3073 victim = _int_malloc (ar_ptr, bytes);
3074 /* Retry with another arena only if we were able to find a usable arena
3075 before. */
3076 if (!victim && ar_ptr != NULL)
3077 {
3078 LIBC_PROBE (memory_malloc_retry, 1, bytes);
3079 ar_ptr = arena_get_retry (ar_ptr, bytes);
3080 victim = _int_malloc (ar_ptr, bytes);
3081 }
3082
3083 if (ar_ptr != NULL)
3084 __libc_lock_unlock (ar_ptr->mutex);
3085
3086 assert (!victim || chunk_is_mmapped (mem2chunk (victim)) ||
3087 ar_ptr == arena_for_chunk (mem2chunk (victim)));
3088 return victim;
3089}
3090libc_hidden_def (__libc_malloc)
3091
3092void
3093__libc_free (void *mem)
3094{
3095 mstate ar_ptr;
3096 mchunkptr p; /* chunk corresponding to mem */
3097
3098 void (*hook) (void *, const void *)
3099 = atomic_forced_read (__free_hook);
3100 if (__builtin_expect (hook != NULL, 0))
3101 {
3102 (*hook)(mem, RETURN_ADDRESS (0));
3103 return;
3104 }
3105
3106 if (mem == 0) /* free(0) has no effect */
3107 return;
3108
3109 p = mem2chunk (mem);
3110
3111 if (chunk_is_mmapped (p)) /* release mmapped memory. */
3112 {
3113 /* See if the dynamic brk/mmap threshold needs adjusting.
3114 Dumped fake mmapped chunks do not affect the threshold. */
3115 if (!mp_.no_dyn_threshold
3116 && chunksize_nomask (p) > mp_.mmap_threshold
3117 && chunksize_nomask (p) <= DEFAULT_MMAP_THRESHOLD_MAX
3118 && !DUMPED_MAIN_ARENA_CHUNK (p))
3119 {
3120 mp_.mmap_threshold = chunksize (p);
3121 mp_.trim_threshold = 2 * mp_.mmap_threshold;
3122 LIBC_PROBE (memory_mallopt_free_dyn_thresholds, 2,
3123 mp_.mmap_threshold, mp_.trim_threshold);
3124 }
3125 munmap_chunk (p);
3126 return;
3127 }
3128
3129 MAYBE_INIT_TCACHE ();
3130
3131 ar_ptr = arena_for_chunk (p);
3132 _int_free (ar_ptr, p, 0);
3133}
3134libc_hidden_def (__libc_free)
3135
3136void *
3137__libc_realloc (void *oldmem, size_t bytes)
3138{
3139 mstate ar_ptr;
3140 INTERNAL_SIZE_T nb; /* padded request size */
3141
3142 void *newp; /* chunk to return */
3143
3144 void *(*hook) (void *, size_t, const void *) =
3145 atomic_forced_read (__realloc_hook);
3146 if (__builtin_expect (hook != NULL, 0))
3147 return (*hook)(oldmem, bytes, RETURN_ADDRESS (0));
3148
3149#if REALLOC_ZERO_BYTES_FREES
3150 if (bytes == 0 && oldmem != NULL)
3151 {
3152 __libc_free (oldmem); return 0;
3153 }
3154#endif
3155
3156 /* realloc of null is supposed to be same as malloc */
3157 if (oldmem == 0)
3158 return __libc_malloc (bytes);
3159
3160 /* chunk corresponding to oldmem */
3161 const mchunkptr oldp = mem2chunk (oldmem);
3162 /* its size */
3163 const INTERNAL_SIZE_T oldsize = chunksize (oldp);
3164
3165 if (chunk_is_mmapped (oldp))
3166 ar_ptr = NULL;
3167 else
3168 {
3169 MAYBE_INIT_TCACHE ();
3170 ar_ptr = arena_for_chunk (oldp);
3171 }
3172
3173 /* Little security check which won't hurt performance: the allocator
3174 never wrapps around at the end of the address space. Therefore
3175 we can exclude some size values which might appear here by
3176 accident or by "design" from some intruder. We need to bypass
3177 this check for dumped fake mmap chunks from the old main arena
3178 because the new malloc may provide additional alignment. */
3179 if ((__builtin_expect ((uintptr_t) oldp > (uintptr_t) -oldsize, 0)
3180 || __builtin_expect (misaligned_chunk (oldp), 0))
3181 && !DUMPED_MAIN_ARENA_CHUNK (oldp))
3182 malloc_printerr ("realloc(): invalid pointer");
3183
3184 checked_request2size (bytes, nb);
3185
3186 if (chunk_is_mmapped (oldp))
3187 {
3188 /* If this is a faked mmapped chunk from the dumped main arena,
3189 always make a copy (and do not free the old chunk). */
3190 if (DUMPED_MAIN_ARENA_CHUNK (oldp))
3191 {
3192 /* Must alloc, copy, free. */
3193 void *newmem = __libc_malloc (bytes);
3194 if (newmem == 0)
3195 return NULL;
3196 /* Copy as many bytes as are available from the old chunk
3197 and fit into the new size. NB: The overhead for faked
3198 mmapped chunks is only SIZE_SZ, not 2 * SIZE_SZ as for
3199 regular mmapped chunks. */
3200 if (bytes > oldsize - SIZE_SZ)
3201 bytes = oldsize - SIZE_SZ;
3202 memcpy (newmem, oldmem, bytes);
3203 return newmem;
3204 }
3205
3206 void *newmem;
3207
3208#if HAVE_MREMAP
3209 newp = mremap_chunk (oldp, nb);
3210 if (newp)
3211 return chunk2mem (newp);
3212#endif
3213 /* Note the extra SIZE_SZ overhead. */
3214 if (oldsize - SIZE_SZ >= nb)
3215 return oldmem; /* do nothing */
3216
3217 /* Must alloc, copy, free. */
3218 newmem = __libc_malloc (bytes);
3219 if (newmem == 0)
3220 return 0; /* propagate failure */
3221
3222 memcpy (newmem, oldmem, oldsize - 2 * SIZE_SZ);
3223 munmap_chunk (oldp);
3224 return newmem;
3225 }
3226
3227 if (SINGLE_THREAD_P)
3228 {
3229 newp = _int_realloc (ar_ptr, oldp, oldsize, nb);
3230 assert (!newp || chunk_is_mmapped (mem2chunk (newp)) ||
3231 ar_ptr == arena_for_chunk (mem2chunk (newp)));
3232
3233 return newp;
3234 }
3235
3236 __libc_lock_lock (ar_ptr->mutex);
3237
3238 newp = _int_realloc (ar_ptr, oldp, oldsize, nb);
3239
3240 __libc_lock_unlock (ar_ptr->mutex);
3241 assert (!newp || chunk_is_mmapped (mem2chunk (newp)) ||
3242 ar_ptr == arena_for_chunk (mem2chunk (newp)));
3243
3244 if (newp == NULL)
3245 {
3246 /* Try harder to allocate memory in other arenas. */
3247 LIBC_PROBE (memory_realloc_retry, 2, bytes, oldmem);
3248 newp = __libc_malloc (bytes);
3249 if (newp != NULL)
3250 {
3251 memcpy (newp, oldmem, oldsize - SIZE_SZ);
3252 _int_free (ar_ptr, oldp, 0);
3253 }
3254 }
3255
3256 return newp;
3257}
3258libc_hidden_def (__libc_realloc)
3259
3260void *
3261__libc_memalign (size_t alignment, size_t bytes)
3262{
3263 void *address = RETURN_ADDRESS (0);
3264 return _mid_memalign (alignment, bytes, address);
3265}
3266
3267static void *
3268_mid_memalign (size_t alignment, size_t bytes, void *address)
3269{
3270 mstate ar_ptr;
3271 void *p;
3272
3273 void *(*hook) (size_t, size_t, const void *) =
3274 atomic_forced_read (__memalign_hook);
3275 if (__builtin_expect (hook != NULL, 0))
3276 return (*hook)(alignment, bytes, address);
3277
3278 /* If we need less alignment than we give anyway, just relay to malloc. */
3279 if (alignment <= MALLOC_ALIGNMENT)
3280 return __libc_malloc (bytes);
3281
3282 /* Otherwise, ensure that it is at least a minimum chunk size */
3283 if (alignment < MINSIZE)
3284 alignment = MINSIZE;
3285
3286 /* If the alignment is greater than SIZE_MAX / 2 + 1 it cannot be a
3287 power of 2 and will cause overflow in the check below. */
3288 if (alignment > SIZE_MAX / 2 + 1)
3289 {
3290 __set_errno (EINVAL);
3291 return 0;
3292 }
3293
3294 /* Check for overflow. */
3295 if (bytes > SIZE_MAX - alignment - MINSIZE)
3296 {
3297 __set_errno (ENOMEM);
3298 return 0;
3299 }
3300
3301
3302 /* Make sure alignment is power of 2. */
3303 if (!powerof2 (alignment))
3304 {
3305 size_t a = MALLOC_ALIGNMENT * 2;
3306 while (a < alignment)
3307 a <<= 1;
3308 alignment = a;
3309 }
3310
3311 if (SINGLE_THREAD_P)
3312 {
3313 p = _int_memalign (&main_arena, alignment, bytes);
3314 assert (!p || chunk_is_mmapped (mem2chunk (p)) ||
3315 &main_arena == arena_for_chunk (mem2chunk (p)));
3316
3317 return p;
3318 }
3319
3320 arena_get (ar_ptr, bytes + alignment + MINSIZE);
3321
3322 p = _int_memalign (ar_ptr, alignment, bytes);
3323 if (!p && ar_ptr != NULL)
3324 {
3325 LIBC_PROBE (memory_memalign_retry, 2, bytes, alignment);
3326 ar_ptr = arena_get_retry (ar_ptr, bytes);
3327 p = _int_memalign (ar_ptr, alignment, bytes);
3328 }
3329
3330 if (ar_ptr != NULL)
3331 __libc_lock_unlock (ar_ptr->mutex);
3332
3333 assert (!p || chunk_is_mmapped (mem2chunk (p)) ||
3334 ar_ptr == arena_for_chunk (mem2chunk (p)));
3335 return p;
3336}
3337/* For ISO C11. */
3338weak_alias (__libc_memalign, aligned_alloc)
3339libc_hidden_def (__libc_memalign)
3340
3341void *
3342__libc_valloc (size_t bytes)
3343{
3344 if (__malloc_initialized < 0)
3345 ptmalloc_init ();
3346
3347 void *address = RETURN_ADDRESS (0);
3348 size_t pagesize = GLRO (dl_pagesize);
3349 return _mid_memalign (pagesize, bytes, address);
3350}
3351
3352void *
3353__libc_pvalloc (size_t bytes)
3354{
3355 if (__malloc_initialized < 0)
3356 ptmalloc_init ();
3357
3358 void *address = RETURN_ADDRESS (0);
3359 size_t pagesize = GLRO (dl_pagesize);
3360 size_t rounded_bytes = ALIGN_UP (bytes, pagesize);
3361
3362 /* Check for overflow. */
3363 if (bytes > SIZE_MAX - 2 * pagesize - MINSIZE)
3364 {
3365 __set_errno (ENOMEM);
3366 return 0;
3367 }
3368
3369 return _mid_memalign (pagesize, rounded_bytes, address);
3370}
3371
3372void *
3373__libc_calloc (size_t n, size_t elem_size)
3374{
3375 mstate av;
3376 mchunkptr oldtop, p;
3377 INTERNAL_SIZE_T bytes, sz, csz, oldtopsize;
3378 void *mem;
3379 unsigned long clearsize;
3380 unsigned long nclears;
3381 INTERNAL_SIZE_T *d;
3382
3383 /* size_t is unsigned so the behavior on overflow is defined. */
3384 bytes = n * elem_size;
3385#define HALF_INTERNAL_SIZE_T \
3386 (((INTERNAL_SIZE_T) 1) << (8 * sizeof (INTERNAL_SIZE_T) / 2))
3387 if (__builtin_expect ((n | elem_size) >= HALF_INTERNAL_SIZE_T, 0))
3388 {
3389 if (elem_size != 0 && bytes / elem_size != n)
3390 {
3391 __set_errno (ENOMEM);
3392 return 0;
3393 }
3394 }
3395
3396 void *(*hook) (size_t, const void *) =
3397 atomic_forced_read (__malloc_hook);
3398 if (__builtin_expect (hook != NULL, 0))
3399 {
3400 sz = bytes;
3401 mem = (*hook)(sz, RETURN_ADDRESS (0));
3402 if (mem == 0)
3403 return 0;
3404
3405 return memset (mem, 0, sz);
3406 }
3407
3408 sz = bytes;
3409
3410 MAYBE_INIT_TCACHE ();
3411
3412 if (SINGLE_THREAD_P)
3413 av = &main_arena;
3414 else
3415 arena_get (av, sz);
3416
3417 if (av)
3418 {
3419 /* Check if we hand out the top chunk, in which case there may be no
3420 need to clear. */
3421#if MORECORE_CLEARS
3422 oldtop = top (av);
3423 oldtopsize = chunksize (top (av));
3424# if MORECORE_CLEARS < 2
3425 /* Only newly allocated memory is guaranteed to be cleared. */
3426 if (av == &main_arena &&
3427 oldtopsize < mp_.sbrk_base + av->max_system_mem - (char *) oldtop)
3428 oldtopsize = (mp_.sbrk_base + av->max_system_mem - (char *) oldtop);
3429# endif
3430 if (av != &main_arena)
3431 {
3432 heap_info *heap = heap_for_ptr (oldtop);
3433 if (oldtopsize < (char *) heap + heap->mprotect_size - (char *) oldtop)
3434 oldtopsize = (char *) heap + heap->mprotect_size - (char *) oldtop;
3435 }
3436#endif
3437 }
3438 else
3439 {
3440 /* No usable arenas. */
3441 oldtop = 0;
3442 oldtopsize = 0;
3443 }
3444 mem = _int_malloc (av, sz);
3445
3446 assert (!mem || chunk_is_mmapped (mem2chunk (mem)) ||
3447 av == arena_for_chunk (mem2chunk (mem)));
3448
3449 if (!SINGLE_THREAD_P)
3450 {
3451 if (mem == 0 && av != NULL)
3452 {
3453 LIBC_PROBE (memory_calloc_retry, 1, sz);
3454 av = arena_get_retry (av, sz);
3455 mem = _int_malloc (av, sz);
3456 }
3457
3458 if (av != NULL)
3459 __libc_lock_unlock (av->mutex);
3460 }
3461
3462 /* Allocation failed even after a retry. */
3463 if (mem == 0)
3464 return 0;
3465
3466 p = mem2chunk (mem);
3467
3468 /* Two optional cases in which clearing not necessary */
3469 if (chunk_is_mmapped (p))
3470 {
3471 if (__builtin_expect (perturb_byte, 0))
3472 return memset (mem, 0, sz);
3473
3474 return mem;
3475 }
3476
3477 csz = chunksize (p);
3478
3479#if MORECORE_CLEARS
3480 if (perturb_byte == 0 && (p == oldtop && csz > oldtopsize))
3481 {
3482 /* clear only the bytes from non-freshly-sbrked memory */
3483 csz = oldtopsize;
3484 }
3485#endif
3486
3487 /* Unroll clear of <= 36 bytes (72 if 8byte sizes). We know that
3488 contents have an odd number of INTERNAL_SIZE_T-sized words;
3489 minimally 3. */
3490 d = (INTERNAL_SIZE_T *) mem;
3491 clearsize = csz - SIZE_SZ;
3492 nclears = clearsize / sizeof (INTERNAL_SIZE_T);
3493 assert (nclears >= 3);
3494
3495 if (nclears > 9)
3496 return memset (d, 0, clearsize);
3497
3498 else
3499 {
3500 *(d + 0) = 0;
3501 *(d + 1) = 0;
3502 *(d + 2) = 0;
3503 if (nclears > 4)
3504 {
3505 *(d + 3) = 0;
3506 *(d + 4) = 0;
3507 if (nclears > 6)
3508 {
3509 *(d + 5) = 0;
3510 *(d + 6) = 0;
3511 if (nclears > 8)
3512 {
3513 *(d + 7) = 0;
3514 *(d + 8) = 0;
3515 }
3516 }
3517 }
3518 }
3519
3520 return mem;
3521}
3522
3523/*
3524 ------------------------------ malloc ------------------------------
3525 */
3526
3527static void *
3528_int_malloc (mstate av, size_t bytes)
3529{
3530 INTERNAL_SIZE_T nb; /* normalized request size */
3531 unsigned int idx; /* associated bin index */
3532 mbinptr bin; /* associated bin */
3533
3534 mchunkptr victim; /* inspected/selected chunk */
3535 INTERNAL_SIZE_T size; /* its size */
3536 int victim_index; /* its bin index */
3537
3538 mchunkptr remainder; /* remainder from a split */
3539 unsigned long remainder_size; /* its size */
3540
3541 unsigned int block; /* bit map traverser */
3542 unsigned int bit; /* bit map traverser */
3543 unsigned int map; /* current word of binmap */
3544
3545 mchunkptr fwd; /* misc temp for linking */
3546 mchunkptr bck; /* misc temp for linking */
3547
3548#if USE_TCACHE
3549 size_t tcache_unsorted_count; /* count of unsorted chunks processed */
3550#endif
3551
3552 /*
3553 Convert request size to internal form by adding SIZE_SZ bytes
3554 overhead plus possibly more to obtain necessary alignment and/or
3555 to obtain a size of at least MINSIZE, the smallest allocatable
3556 size. Also, checked_request2size traps (returning 0) request sizes
3557 that are so large that they wrap around zero when padded and
3558 aligned.
3559 */
3560
3561 checked_request2size (bytes, nb);
3562
3563 /* There are no usable arenas. Fall back to sysmalloc to get a chunk from
3564 mmap. */
3565 if (__glibc_unlikely (av == NULL))
3566 {
3567 void *p = sysmalloc (nb, av);
3568 if (p != NULL)
3569 alloc_perturb (p, bytes);
3570 return p;
3571 }
3572
3573 /*
3574 If the size qualifies as a fastbin, first check corresponding bin.
3575 This code is safe to execute even if av is not yet initialized, so we
3576 can try it without checking, which saves some time on this fast path.
3577 */
3578
3579#define REMOVE_FB(fb, victim, pp) \
3580 do \
3581 { \
3582 victim = pp; \
3583 if (victim == NULL) \
3584 break; \
3585 } \
3586 while ((pp = catomic_compare_and_exchange_val_acq (fb, victim->fd, victim)) \
3587 != victim); \
3588
3589 if ((unsigned long) (nb) <= (unsigned long) (get_max_fast ()))
3590 {
3591 idx = fastbin_index (nb);
3592 mfastbinptr *fb = &fastbin (av, idx);
3593 mchunkptr pp;
3594 victim = *fb;
3595
3596 if (victim != NULL)
3597 {
3598 if (SINGLE_THREAD_P)
3599 *fb = victim->fd;
3600 else
3601 REMOVE_FB (fb, pp, victim);
3602 if (__glibc_likely (victim != NULL))
3603 {
3604 size_t victim_idx = fastbin_index (chunksize (victim));
3605 if (__builtin_expect (victim_idx != idx, 0))
3606 malloc_printerr ("malloc(): memory corruption (fast)");
3607 check_remalloced_chunk (av, victim, nb);
3608#if USE_TCACHE
3609 /* While we're here, if we see other chunks of the same size,
3610 stash them in the tcache. */
3611 size_t tc_idx = csize2tidx (nb);
3612 if (tcache && tc_idx < mp_.tcache_bins)
3613 {
3614 mchunkptr tc_victim;
3615
3616 /* While bin not empty and tcache not full, copy chunks. */
3617 while (tcache->counts[tc_idx] < mp_.tcache_count
3618 && (tc_victim = *fb) != NULL)
3619 {
3620 if (SINGLE_THREAD_P)
3621 *fb = tc_victim->fd;
3622 else
3623 {
3624 REMOVE_FB (fb, pp, tc_victim);
3625 if (__glibc_unlikely (tc_victim == NULL))
3626 break;
3627 }
3628 tcache_put (tc_victim, tc_idx);
3629 }
3630 }
3631#endif
3632 void *p = chunk2mem (victim);
3633 alloc_perturb (p, bytes);
3634 return p;
3635 }
3636 }
3637 }
3638
3639 /*
3640 If a small request, check regular bin. Since these "smallbins"
3641 hold one size each, no searching within bins is necessary.
3642 (For a large request, we need to wait until unsorted chunks are
3643 processed to find best fit. But for small ones, fits are exact
3644 anyway, so we can check now, which is faster.)
3645 */
3646
3647 if (in_smallbin_range (nb))
3648 {
3649 idx = smallbin_index (nb);
3650 bin = bin_at (av, idx);
3651
3652 if ((victim = last (bin)) != bin)
3653 {
3654 bck = victim->bk;
3655 if (__glibc_unlikely (bck->fd != victim))
3656 malloc_printerr ("malloc(): smallbin double linked list corrupted");
3657 set_inuse_bit_at_offset (victim, nb);
3658 bin->bk = bck;
3659 bck->fd = bin;
3660
3661 if (av != &main_arena)
3662 set_non_main_arena (victim);
3663 check_malloced_chunk (av, victim, nb);
3664#if USE_TCACHE
3665 /* While we're here, if we see other chunks of the same size,
3666 stash them in the tcache. */
3667 size_t tc_idx = csize2tidx (nb);
3668 if (tcache && tc_idx < mp_.tcache_bins)
3669 {
3670 mchunkptr tc_victim;
3671
3672 /* While bin not empty and tcache not full, copy chunks over. */
3673 while (tcache->counts[tc_idx] < mp_.tcache_count
3674 && (tc_victim = last (bin)) != bin)
3675 {
3676 if (tc_victim != 0)
3677 {
3678 bck = tc_victim->bk;
3679 set_inuse_bit_at_offset (tc_victim, nb);
3680 if (av != &main_arena)
3681 set_non_main_arena (tc_victim);
3682 bin->bk = bck;
3683 bck->fd = bin;
3684
3685 tcache_put (tc_victim, tc_idx);
3686 }
3687 }
3688 }
3689#endif
3690 void *p = chunk2mem (victim);
3691 alloc_perturb (p, bytes);
3692 return p;
3693 }
3694 }
3695
3696 /*
3697 If this is a large request, consolidate fastbins before continuing.
3698 While it might look excessive to kill all fastbins before
3699 even seeing if there is space available, this avoids
3700 fragmentation problems normally associated with fastbins.
3701 Also, in practice, programs tend to have runs of either small or
3702 large requests, but less often mixtures, so consolidation is not
3703 invoked all that often in most programs. And the programs that
3704 it is called frequently in otherwise tend to fragment.
3705 */
3706
3707 else
3708 {
3709 idx = largebin_index (nb);
3710 if (atomic_load_relaxed (&av->have_fastchunks))
3711 malloc_consolidate (av);
3712 }
3713
3714 /*
3715 Process recently freed or remaindered chunks, taking one only if
3716 it is exact fit, or, if this a small request, the chunk is remainder from
3717 the most recent non-exact fit. Place other traversed chunks in
3718 bins. Note that this step is the only place in any routine where
3719 chunks are placed in bins.
3720
3721 The outer loop here is needed because we might not realize until
3722 near the end of malloc that we should have consolidated, so must
3723 do so and retry. This happens at most once, and only when we would
3724 otherwise need to expand memory to service a "small" request.
3725 */
3726
3727#if USE_TCACHE
3728 INTERNAL_SIZE_T tcache_nb = 0;
3729 size_t tc_idx = csize2tidx (nb);
3730 if (tcache && tc_idx < mp_.tcache_bins)
3731 tcache_nb = nb;
3732 int return_cached = 0;
3733
3734 tcache_unsorted_count = 0;
3735#endif
3736
3737 for (;; )
3738 {
3739 int iters = 0;
3740 while ((victim = unsorted_chunks (av)->bk) != unsorted_chunks (av))
3741 {
3742 bck = victim->bk;
3743 size = chunksize (victim);
3744 mchunkptr next = chunk_at_offset (victim, size);
3745
3746 if (__glibc_unlikely (size <= 2 * SIZE_SZ)
3747 || __glibc_unlikely (size > av->system_mem))
3748 malloc_printerr ("malloc(): invalid size (unsorted)");
3749 if (__glibc_unlikely (chunksize_nomask (next) < 2 * SIZE_SZ)
3750 || __glibc_unlikely (chunksize_nomask (next) > av->system_mem))
3751 malloc_printerr ("malloc(): invalid next size (unsorted)");
3752 if (__glibc_unlikely ((prev_size (next) & ~(SIZE_BITS)) != size))
3753 malloc_printerr ("malloc(): mismatching next->prev_size (unsorted)");
3754 if (__glibc_unlikely (bck->fd != victim)
3755 || __glibc_unlikely (victim->fd != unsorted_chunks (av)))
3756 malloc_printerr ("malloc(): unsorted double linked list corrupted");
3757 if (__glibc_unlikely (prev_inuse (next)))
3758 malloc_printerr ("malloc(): invalid next->prev_inuse (unsorted)");
3759
3760 /*
3761 If a small request, try to use last remainder if it is the
3762 only chunk in unsorted bin. This helps promote locality for
3763 runs of consecutive small requests. This is the only
3764 exception to best-fit, and applies only when there is
3765 no exact fit for a small chunk.
3766 */
3767
3768 if (in_smallbin_range (nb) &&
3769 bck == unsorted_chunks (av) &&
3770 victim == av->last_remainder &&
3771 (unsigned long) (size) > (unsigned long) (nb + MINSIZE))
3772 {
3773 /* split and reattach remainder */
3774 remainder_size = size - nb;
3775 remainder = chunk_at_offset (victim, nb);
3776 unsorted_chunks (av)->bk = unsorted_chunks (av)->fd = remainder;
3777 av->last_remainder = remainder;
3778 remainder->bk = remainder->fd = unsorted_chunks (av);
3779 if (!in_smallbin_range (remainder_size))
3780 {
3781 remainder->fd_nextsize = NULL;
3782 remainder->bk_nextsize = NULL;
3783 }
3784
3785 set_head (victim, nb | PREV_INUSE |
3786 (av != &main_arena ? NON_MAIN_ARENA : 0));
3787 set_head (remainder, remainder_size | PREV_INUSE);
3788 set_foot (remainder, remainder_size);
3789
3790 check_malloced_chunk (av, victim, nb);
3791 void *p = chunk2mem (victim);
3792 alloc_perturb (p, bytes);
3793 return p;
3794 }
3795
3796 /* remove from unsorted list */
3797 if (__glibc_unlikely (bck->fd != victim))
3798 malloc_printerr ("malloc(): corrupted unsorted chunks 3");
3799 unsorted_chunks (av)->bk = bck;
3800 bck->fd = unsorted_chunks (av);
3801
3802 /* Take now instead of binning if exact fit */
3803
3804 if (size == nb)
3805 {
3806 set_inuse_bit_at_offset (victim, size);
3807 if (av != &main_arena)
3808 set_non_main_arena (victim);
3809#if USE_TCACHE
3810 /* Fill cache first, return to user only if cache fills.
3811 We may return one of these chunks later. */
3812 if (tcache_nb
3813 && tcache->counts[tc_idx] < mp_.tcache_count)
3814 {
3815 tcache_put (victim, tc_idx);
3816 return_cached = 1;
3817 continue;
3818 }
3819 else
3820 {
3821#endif
3822 check_malloced_chunk (av, victim, nb);
3823 void *p = chunk2mem (victim);
3824 alloc_perturb (p, bytes);
3825 return p;
3826#if USE_TCACHE
3827 }
3828#endif
3829 }
3830
3831 /* place chunk in bin */
3832
3833 if (in_smallbin_range (size))
3834 {
3835 victim_index = smallbin_index (size);
3836 bck = bin_at (av, victim_index);
3837 fwd = bck->fd;
3838 }
3839 else
3840 {
3841 victim_index = largebin_index (size);
3842 bck = bin_at (av, victim_index);
3843 fwd = bck->fd;
3844
3845 /* maintain large bins in sorted order */
3846 if (fwd != bck)
3847 {
3848 /* Or with inuse bit to speed comparisons */
3849 size |= PREV_INUSE;
3850 /* if smaller than smallest, bypass loop below */
3851 assert (chunk_main_arena (bck->bk));
3852 if ((unsigned long) (size)
3853 < (unsigned long) chunksize_nomask (bck->bk))
3854 {
3855 fwd = bck;
3856 bck = bck->bk;
3857
3858 victim->fd_nextsize = fwd->fd;
3859 victim->bk_nextsize = fwd->fd->bk_nextsize;
3860 fwd->fd->bk_nextsize = victim->bk_nextsize->fd_nextsize = victim;
3861 }
3862 else
3863 {
3864 assert (chunk_main_arena (fwd));
3865 while ((unsigned long) size < chunksize_nomask (fwd))
3866 {
3867 fwd = fwd->fd_nextsize;
3868 assert (chunk_main_arena (fwd));
3869 }
3870
3871 if ((unsigned long) size
3872 == (unsigned long) chunksize_nomask (fwd))
3873 /* Always insert in the second position. */
3874 fwd = fwd->fd;
3875 else
3876 {
3877 victim->fd_nextsize = fwd;
3878 victim->bk_nextsize = fwd->bk_nextsize;
3879 if (__glibc_unlikely (fwd->bk_nextsize->fd_nextsize != fwd))
3880 malloc_printerr ("malloc(): largebin double linked list corrupted (nextsize)");
3881 fwd->bk_nextsize = victim;
3882 victim->bk_nextsize->fd_nextsize = victim;
3883 }
3884 bck = fwd->bk;
3885 if (bck->fd != fwd)
3886 malloc_printerr ("malloc(): largebin double linked list corrupted (bk)");
3887 }
3888 }
3889 else
3890 victim->fd_nextsize = victim->bk_nextsize = victim;
3891 }
3892
3893 mark_bin (av, victim_index);
3894 victim->bk = bck;
3895 victim->fd = fwd;
3896 fwd->bk = victim;
3897 bck->fd = victim;
3898
3899#if USE_TCACHE
3900 /* If we've processed as many chunks as we're allowed while
3901 filling the cache, return one of the cached ones. */
3902 ++tcache_unsorted_count;
3903 if (return_cached
3904 && mp_.tcache_unsorted_limit > 0
3905 && tcache_unsorted_count > mp_.tcache_unsorted_limit)
3906 {
3907 return tcache_get (tc_idx);
3908 }
3909#endif
3910
3911#define MAX_ITERS 10000
3912 if (++iters >= MAX_ITERS)
3913 break;
3914 }
3915
3916#if USE_TCACHE
3917 /* If all the small chunks we found ended up cached, return one now. */
3918 if (return_cached)
3919 {
3920 return tcache_get (tc_idx);
3921 }
3922#endif
3923
3924 /*
3925 If a large request, scan through the chunks of current bin in
3926 sorted order to find smallest that fits. Use the skip list for this.
3927 */
3928
3929 if (!in_smallbin_range (nb))
3930 {
3931 bin = bin_at (av, idx);
3932
3933 /* skip scan if empty or largest chunk is too small */
3934 if ((victim = first (bin)) != bin
3935 && (unsigned long) chunksize_nomask (victim)
3936 >= (unsigned long) (nb))
3937 {
3938 victim = victim->bk_nextsize;
3939 while (((unsigned long) (size = chunksize (victim)) <
3940 (unsigned long) (nb)))
3941 victim = victim->bk_nextsize;
3942
3943 /* Avoid removing the first entry for a size so that the skip
3944 list does not have to be rerouted. */
3945 if (victim != last (bin)
3946 && chunksize_nomask (victim)
3947 == chunksize_nomask (victim->fd))
3948 victim = victim->fd;
3949
3950 remainder_size = size - nb;
3951 unlink_chunk (av, victim);
3952
3953 /* Exhaust */
3954 if (remainder_size < MINSIZE)
3955 {
3956 set_inuse_bit_at_offset (victim, size);
3957 if (av != &main_arena)
3958 set_non_main_arena (victim);
3959 }
3960 /* Split */
3961 else
3962 {
3963 remainder = chunk_at_offset (victim, nb);
3964 /* We cannot assume the unsorted list is empty and therefore
3965 have to perform a complete insert here. */
3966 bck = unsorted_chunks (av);
3967 fwd = bck->fd;
3968 if (__glibc_unlikely (fwd->bk != bck))
3969 malloc_printerr ("malloc(): corrupted unsorted chunks");
3970 remainder->bk = bck;
3971 remainder->fd = fwd;
3972 bck->fd = remainder;
3973 fwd->bk = remainder;
3974 if (!in_smallbin_range (remainder_size))
3975 {
3976 remainder->fd_nextsize = NULL;
3977 remainder->bk_nextsize = NULL;
3978 }
3979 set_head (victim, nb | PREV_INUSE |
3980 (av != &main_arena ? NON_MAIN_ARENA : 0));
3981 set_head (remainder, remainder_size | PREV_INUSE);
3982 set_foot (remainder, remainder_size);
3983 }
3984 check_malloced_chunk (av, victim, nb);
3985 void *p = chunk2mem (victim);
3986 alloc_perturb (p, bytes);
3987 return p;
3988 }
3989 }
3990
3991 /*
3992 Search for a chunk by scanning bins, starting with next largest
3993 bin. This search is strictly by best-fit; i.e., the smallest
3994 (with ties going to approximately the least recently used) chunk
3995 that fits is selected.
3996
3997 The bitmap avoids needing to check that most blocks are nonempty.
3998 The particular case of skipping all bins during warm-up phases
3999 when no chunks have been returned yet is faster than it might look.
4000 */
4001
4002 ++idx;
4003 bin = bin_at (av, idx);
4004 block = idx2block (idx);
4005 map = av->binmap[block];
4006 bit = idx2bit (idx);
4007
4008 for (;; )
4009 {
4010 /* Skip rest of block if there are no more set bits in this block. */
4011 if (bit > map || bit == 0)
4012 {
4013 do
4014 {
4015 if (++block >= BINMAPSIZE) /* out of bins */
4016 goto use_top;
4017 }
4018 while ((map = av->binmap[block]) == 0);
4019
4020 bin = bin_at (av, (block << BINMAPSHIFT));
4021 bit = 1;
4022 }
4023
4024 /* Advance to bin with set bit. There must be one. */
4025 while ((bit & map) == 0)
4026 {
4027 bin = next_bin (bin);
4028 bit <<= 1;
4029 assert (bit != 0);
4030 }
4031
4032 /* Inspect the bin. It is likely to be non-empty */
4033 victim = last (bin);
4034
4035 /* If a false alarm (empty bin), clear the bit. */
4036 if (victim == bin)
4037 {
4038 av->binmap[block] = map &= ~bit; /* Write through */
4039 bin = next_bin (bin);
4040 bit <<= 1;
4041 }
4042
4043 else
4044 {
4045 size = chunksize (victim);
4046
4047 /* We know the first chunk in this bin is big enough to use. */
4048 assert ((unsigned long) (size) >= (unsigned long) (nb));
4049
4050 remainder_size = size - nb;
4051
4052 /* unlink */
4053 unlink_chunk (av, victim);
4054
4055 /* Exhaust */
4056 if (remainder_size < MINSIZE)
4057 {
4058 set_inuse_bit_at_offset (victim, size);
4059 if (av != &main_arena)
4060 set_non_main_arena (victim);
4061 }
4062
4063 /* Split */
4064 else
4065 {
4066 remainder = chunk_at_offset (victim, nb);
4067
4068 /* We cannot assume the unsorted list is empty and therefore
4069 have to perform a complete insert here. */
4070 bck = unsorted_chunks (av);
4071 fwd = bck->fd;
4072 if (__glibc_unlikely (fwd->bk != bck))
4073 malloc_printerr ("malloc(): corrupted unsorted chunks 2");
4074 remainder->bk = bck;
4075 remainder->fd = fwd;
4076 bck->fd = remainder;
4077 fwd->bk = remainder;
4078
4079 /* advertise as last remainder */
4080 if (in_smallbin_range (nb))
4081 av->last_remainder = remainder;
4082 if (!in_smallbin_range (remainder_size))
4083 {
4084 remainder->fd_nextsize = NULL;
4085 remainder->bk_nextsize = NULL;
4086 }
4087 set_head (victim, nb | PREV_INUSE |
4088 (av != &main_arena ? NON_MAIN_ARENA : 0));
4089 set_head (remainder, remainder_size | PREV_INUSE);
4090 set_foot (remainder, remainder_size);
4091 }
4092 check_malloced_chunk (av, victim, nb);
4093 void *p = chunk2mem (victim);
4094 alloc_perturb (p, bytes);
4095 return p;
4096 }
4097 }
4098
4099 use_top:
4100 /*
4101 If large enough, split off the chunk bordering the end of memory
4102 (held in av->top). Note that this is in accord with the best-fit
4103 search rule. In effect, av->top is treated as larger (and thus
4104 less well fitting) than any other available chunk since it can
4105 be extended to be as large as necessary (up to system
4106 limitations).
4107
4108 We require that av->top always exists (i.e., has size >=
4109 MINSIZE) after initialization, so if it would otherwise be
4110 exhausted by current request, it is replenished. (The main
4111 reason for ensuring it exists is that we may need MINSIZE space
4112 to put in fenceposts in sysmalloc.)
4113 */
4114
4115 victim = av->top;
4116 size = chunksize (victim);
4117
4118 if (__glibc_unlikely (size > av->system_mem))
4119 malloc_printerr ("malloc(): corrupted top size");
4120
4121 if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE))
4122 {
4123 remainder_size = size - nb;
4124 remainder = chunk_at_offset (victim, nb);
4125 av->top = remainder;
4126 set_head (victim, nb | PREV_INUSE |
4127 (av != &main_arena ? NON_MAIN_ARENA : 0));
4128 set_head (remainder, remainder_size | PREV_INUSE);
4129
4130 check_malloced_chunk (av, victim, nb);
4131 void *p = chunk2mem (victim);
4132 alloc_perturb (p, bytes);
4133 return p;
4134 }
4135
4136 /* When we are using atomic ops to free fast chunks we can get
4137 here for all block sizes. */
4138 else if (atomic_load_relaxed (&av->have_fastchunks))
4139 {
4140 malloc_consolidate (av);
4141 /* restore original bin index */
4142 if (in_smallbin_range (nb))
4143 idx = smallbin_index (nb);
4144 else
4145 idx = largebin_index (nb);
4146 }
4147
4148 /*
4149 Otherwise, relay to handle system-dependent cases
4150 */
4151 else
4152 {
4153 void *p = sysmalloc (nb, av);
4154 if (p != NULL)
4155 alloc_perturb (p, bytes);
4156 return p;
4157 }
4158 }
4159}
4160
4161/*
4162 ------------------------------ free ------------------------------
4163 */
4164
4165static void
4166_int_free (mstate av, mchunkptr p, int have_lock)
4167{
4168 INTERNAL_SIZE_T size; /* its size */
4169 mfastbinptr *fb; /* associated fastbin */
4170 mchunkptr nextchunk; /* next contiguous chunk */
4171 INTERNAL_SIZE_T nextsize; /* its size */
4172 int nextinuse; /* true if nextchunk is used */
4173 INTERNAL_SIZE_T prevsize; /* size of previous contiguous chunk */
4174 mchunkptr bck; /* misc temp for linking */
4175 mchunkptr fwd; /* misc temp for linking */
4176
4177 size = chunksize (p);
4178
4179 /* Little security check which won't hurt performance: the
4180 allocator never wrapps around at the end of the address space.
4181 Therefore we can exclude some size values which might appear
4182 here by accident or by "design" from some intruder. */
4183 if (__builtin_expect ((uintptr_t) p > (uintptr_t) -size, 0)
4184 || __builtin_expect (misaligned_chunk (p), 0))
4185 malloc_printerr ("free(): invalid pointer");
4186 /* We know that each chunk is at least MINSIZE bytes in size or a
4187 multiple of MALLOC_ALIGNMENT. */
4188 if (__glibc_unlikely (size < MINSIZE || !aligned_OK (size)))
4189 malloc_printerr ("free(): invalid size");
4190
4191 check_inuse_chunk(av, p);
4192
4193#if USE_TCACHE
4194 {
4195 size_t tc_idx = csize2tidx (size);
4196 if (tcache != NULL && tc_idx < mp_.tcache_bins)
4197 {
4198 /* Check to see if it's already in the tcache. */
4199 tcache_entry *e = (tcache_entry *) chunk2mem (p);
4200
4201 /* This test succeeds on double free. However, we don't 100%
4202 trust it (it also matches random payload data at a 1 in
4203 2^<size_t> chance), so verify it's not an unlikely
4204 coincidence before aborting. */
4205 if (__glibc_unlikely (e->key == tcache))
4206 {
4207 tcache_entry *tmp;
4208 LIBC_PROBE (memory_tcache_double_free, 2, e, tc_idx);
4209 for (tmp = tcache->entries[tc_idx];
4210 tmp;
4211 tmp = tmp->next)
4212 if (tmp == e)
4213 malloc_printerr ("free(): double free detected in tcache 2");
4214 /* If we get here, it was a coincidence. We've wasted a
4215 few cycles, but don't abort. */
4216 }
4217
4218 if (tcache->counts[tc_idx] < mp_.tcache_count)
4219 {
4220 tcache_put (p, tc_idx);
4221 return;
4222 }
4223 }
4224 }
4225#endif
4226
4227 /*
4228 If eligible, place chunk on a fastbin so it can be found
4229 and used quickly in malloc.
4230 */
4231
4232 if ((unsigned long)(size) <= (unsigned long)(get_max_fast ())
4233
4234#if TRIM_FASTBINS
4235 /*
4236 If TRIM_FASTBINS set, don't place chunks
4237 bordering top into fastbins
4238 */
4239 && (chunk_at_offset(p, size) != av->top)
4240#endif
4241 ) {
4242
4243 if (__builtin_expect (chunksize_nomask (chunk_at_offset (p, size))
4244 <= 2 * SIZE_SZ, 0)
4245 || __builtin_expect (chunksize (chunk_at_offset (p, size))
4246 >= av->system_mem, 0))
4247 {
4248 bool fail = true;
4249 /* We might not have a lock at this point and concurrent modifications
4250 of system_mem might result in a false positive. Redo the test after
4251 getting the lock. */
4252 if (!have_lock)
4253 {
4254 __libc_lock_lock (av->mutex);
4255 fail = (chunksize_nomask (chunk_at_offset (p, size)) <= 2 * SIZE_SZ
4256 || chunksize (chunk_at_offset (p, size)) >= av->system_mem);
4257 __libc_lock_unlock (av->mutex);
4258 }
4259
4260 if (fail)
4261 malloc_printerr ("free(): invalid next size (fast)");
4262 }
4263
4264 free_perturb (chunk2mem(p), size - 2 * SIZE_SZ);
4265
4266 atomic_store_relaxed (&av->have_fastchunks, true);
4267 unsigned int idx = fastbin_index(size);
4268 fb = &fastbin (av, idx);
4269
4270 /* Atomically link P to its fastbin: P->FD = *FB; *FB = P; */
4271 mchunkptr old = *fb, old2;
4272
4273 if (SINGLE_THREAD_P)
4274 {
4275 /* Check that the top of the bin is not the record we are going to
4276 add (i.e., double free). */
4277 if (__builtin_expect (old == p, 0))
4278 malloc_printerr ("double free or corruption (fasttop)");
4279 p->fd = old;
4280 *fb = p;
4281 }
4282 else
4283 do
4284 {
4285 /* Check that the top of the bin is not the record we are going to
4286 add (i.e., double free). */
4287 if (__builtin_expect (old == p, 0))
4288 malloc_printerr ("double free or corruption (fasttop)");
4289 p->fd = old2 = old;
4290 }
4291 while ((old = catomic_compare_and_exchange_val_rel (fb, p, old2))
4292 != old2);
4293
4294 /* Check that size of fastbin chunk at the top is the same as
4295 size of the chunk that we are adding. We can dereference OLD
4296 only if we have the lock, otherwise it might have already been
4297 allocated again. */
4298 if (have_lock && old != NULL
4299 && __builtin_expect (fastbin_index (chunksize (old)) != idx, 0))
4300 malloc_printerr ("invalid fastbin entry (free)");
4301 }
4302
4303 /*
4304 Consolidate other non-mmapped chunks as they arrive.
4305 */
4306
4307 else if (!chunk_is_mmapped(p)) {
4308
4309 /* If we're single-threaded, don't lock the arena. */
4310 if (SINGLE_THREAD_P)
4311 have_lock = true;
4312
4313 if (!have_lock)
4314 __libc_lock_lock (av->mutex);
4315
4316 nextchunk = chunk_at_offset(p, size);
4317
4318 /* Lightweight tests: check whether the block is already the
4319 top block. */
4320 if (__glibc_unlikely (p == av->top))
4321 malloc_printerr ("double free or corruption (top)");
4322 /* Or whether the next chunk is beyond the boundaries of the arena. */
4323 if (__builtin_expect (contiguous (av)
4324 && (char *) nextchunk
4325 >= ((char *) av->top + chunksize(av->top)), 0))
4326 malloc_printerr ("double free or corruption (out)");
4327 /* Or whether the block is actually not marked used. */
4328 if (__glibc_unlikely (!prev_inuse(nextchunk)))
4329 malloc_printerr ("double free or corruption (!prev)");
4330
4331 nextsize = chunksize(nextchunk);
4332 if (__builtin_expect (chunksize_nomask (nextchunk) <= 2 * SIZE_SZ, 0)
4333 || __builtin_expect (nextsize >= av->system_mem, 0))
4334 malloc_printerr ("free(): invalid next size (normal)");
4335
4336 free_perturb (chunk2mem(p), size - 2 * SIZE_SZ);
4337
4338 /* consolidate backward */
4339 if (!prev_inuse(p)) {
4340 prevsize = prev_size (p);
4341 size += prevsize;
4342 p = chunk_at_offset(p, -((long) prevsize));
4343 if (__glibc_unlikely (chunksize(p) != prevsize))
4344 malloc_printerr ("corrupted size vs. prev_size while consolidating");
4345 unlink_chunk (av, p);
4346 }
4347
4348 if (nextchunk != av->top) {
4349 /* get and clear inuse bit */
4350 nextinuse = inuse_bit_at_offset(nextchunk, nextsize);
4351
4352 /* consolidate forward */
4353 if (!nextinuse) {
4354 unlink_chunk (av, nextchunk);
4355 size += nextsize;
4356 } else
4357 clear_inuse_bit_at_offset(nextchunk, 0);
4358
4359 /*
4360 Place the chunk in unsorted chunk list. Chunks are
4361 not placed into regular bins until after they have
4362 been given one chance to be used in malloc.
4363 */
4364
4365 bck = unsorted_chunks(av);
4366 fwd = bck->fd;
4367 if (__glibc_unlikely (fwd->bk != bck))
4368 malloc_printerr ("free(): corrupted unsorted chunks");
4369 p->fd = fwd;
4370 p->bk = bck;
4371 if (!in_smallbin_range(size))
4372 {
4373 p->fd_nextsize = NULL;
4374 p->bk_nextsize = NULL;
4375 }
4376 bck->fd = p;
4377 fwd->bk = p;
4378
4379 set_head(p, size | PREV_INUSE);
4380 set_foot(p, size);
4381
4382 check_free_chunk(av, p);
4383 }
4384
4385 /*
4386 If the chunk borders the current high end of memory,
4387 consolidate into top
4388 */
4389
4390 else {
4391 size += nextsize;
4392 set_head(p, size | PREV_INUSE);
4393 av->top = p;
4394 check_chunk(av, p);
4395 }
4396
4397 /*
4398 If freeing a large space, consolidate possibly-surrounding
4399 chunks. Then, if the total unused topmost memory exceeds trim
4400 threshold, ask malloc_trim to reduce top.
4401
4402 Unless max_fast is 0, we don't know if there are fastbins
4403 bordering top, so we cannot tell for sure whether threshold
4404 has been reached unless fastbins are consolidated. But we
4405 don't want to consolidate on each free. As a compromise,
4406 consolidation is performed if FASTBIN_CONSOLIDATION_THRESHOLD
4407 is reached.
4408 */
4409
4410 if ((unsigned long)(size) >= FASTBIN_CONSOLIDATION_THRESHOLD) {
4411 if (atomic_load_relaxed (&av->have_fastchunks))
4412 malloc_consolidate(av);
4413
4414 if (av == &main_arena) {
4415#ifndef MORECORE_CANNOT_TRIM
4416 if ((unsigned long)(chunksize(av->top)) >=
4417 (unsigned long)(mp_.trim_threshold))
4418 systrim(mp_.top_pad, av);
4419#endif
4420 } else {
4421 /* Always try heap_trim(), even if the top chunk is not
4422 large, because the corresponding heap might go away. */
4423 heap_info *heap = heap_for_ptr(top(av));
4424
4425 assert(heap->ar_ptr == av);
4426 heap_trim(heap, mp_.top_pad);
4427 }
4428 }
4429
4430 if (!have_lock)
4431 __libc_lock_unlock (av->mutex);
4432 }
4433 /*
4434 If the chunk was allocated via mmap, release via munmap().
4435 */
4436
4437 else {
4438 munmap_chunk (p);
4439 }
4440}
4441
4442/*
4443 ------------------------- malloc_consolidate -------------------------
4444
4445 malloc_consolidate is a specialized version of free() that tears
4446 down chunks held in fastbins. Free itself cannot be used for this
4447 purpose since, among other things, it might place chunks back onto
4448 fastbins. So, instead, we need to use a minor variant of the same
4449 code.
4450*/
4451
4452static void malloc_consolidate(mstate av)
4453{
4454 mfastbinptr* fb; /* current fastbin being consolidated */
4455 mfastbinptr* maxfb; /* last fastbin (for loop control) */
4456 mchunkptr p; /* current chunk being consolidated */
4457 mchunkptr nextp; /* next chunk to consolidate */
4458 mchunkptr unsorted_bin; /* bin header */
4459 mchunkptr first_unsorted; /* chunk to link to */
4460
4461 /* These have same use as in free() */
4462 mchunkptr nextchunk;
4463 INTERNAL_SIZE_T size;
4464 INTERNAL_SIZE_T nextsize;
4465 INTERNAL_SIZE_T prevsize;
4466 int nextinuse;
4467
4468 atomic_store_relaxed (&av->have_fastchunks, false);
4469
4470 unsorted_bin = unsorted_chunks(av);
4471
4472 /*
4473 Remove each chunk from fast bin and consolidate it, placing it
4474 then in unsorted bin. Among other reasons for doing this,
4475 placing in unsorted bin avoids needing to calculate actual bins
4476 until malloc is sure that chunks aren't immediately going to be
4477 reused anyway.
4478 */
4479
4480 maxfb = &fastbin (av, NFASTBINS - 1);
4481 fb = &fastbin (av, 0);
4482 do {
4483 p = atomic_exchange_acq (fb, NULL);
4484 if (p != 0) {
4485 do {
4486 {
4487 unsigned int idx = fastbin_index (chunksize (p));
4488 if ((&fastbin (av, idx)) != fb)
4489 malloc_printerr ("malloc_consolidate(): invalid chunk size");
4490 }
4491
4492 check_inuse_chunk(av, p);
4493 nextp = p->fd;
4494
4495 /* Slightly streamlined version of consolidation code in free() */
4496 size = chunksize (p);
4497 nextchunk = chunk_at_offset(p, size);
4498 nextsize = chunksize(nextchunk);
4499
4500 if (!prev_inuse(p)) {
4501 prevsize = prev_size (p);
4502 size += prevsize;
4503 p = chunk_at_offset(p, -((long) prevsize));
4504 if (__glibc_unlikely (chunksize(p) != prevsize))
4505 malloc_printerr ("corrupted size vs. prev_size in fastbins");
4506 unlink_chunk (av, p);
4507 }
4508
4509 if (nextchunk != av->top) {
4510 nextinuse = inuse_bit_at_offset(nextchunk, nextsize);
4511
4512 if (!nextinuse) {
4513 size += nextsize;
4514 unlink_chunk (av, nextchunk);
4515 } else
4516 clear_inuse_bit_at_offset(nextchunk, 0);
4517
4518 first_unsorted = unsorted_bin->fd;
4519 unsorted_bin->fd = p;
4520 first_unsorted->bk = p;
4521
4522 if (!in_smallbin_range (size)) {
4523 p->fd_nextsize = NULL;
4524 p->bk_nextsize = NULL;
4525 }
4526
4527 set_head(p, size | PREV_INUSE);
4528 p->bk = unsorted_bin;
4529 p->fd = first_unsorted;
4530 set_foot(p, size);
4531 }
4532
4533 else {
4534 size += nextsize;
4535 set_head(p, size | PREV_INUSE);
4536 av->top = p;
4537 }
4538
4539 } while ( (p = nextp) != 0);
4540
4541 }
4542 } while (fb++ != maxfb);
4543}
4544
4545/*
4546 ------------------------------ realloc ------------------------------
4547*/
4548
4549void*
4550_int_realloc(mstate av, mchunkptr oldp, INTERNAL_SIZE_T oldsize,
4551 INTERNAL_SIZE_T nb)
4552{
4553 mchunkptr newp; /* chunk to return */
4554 INTERNAL_SIZE_T newsize; /* its size */
4555 void* newmem; /* corresponding user mem */
4556
4557 mchunkptr next; /* next contiguous chunk after oldp */
4558
4559 mchunkptr remainder; /* extra space at end of newp */
4560 unsigned long remainder_size; /* its size */
4561
4562 /* oldmem size */
4563 if (__builtin_expect (chunksize_nomask (oldp) <= 2 * SIZE_SZ, 0)
4564 || __builtin_expect (oldsize >= av->system_mem, 0))
4565 malloc_printerr ("realloc(): invalid old size");
4566
4567 check_inuse_chunk (av, oldp);
4568
4569 /* All callers already filter out mmap'ed chunks. */
4570 assert (!chunk_is_mmapped (oldp));
4571
4572 next = chunk_at_offset (oldp, oldsize);
4573 INTERNAL_SIZE_T nextsize = chunksize (next);
4574 if (__builtin_expect (chunksize_nomask (next) <= 2 * SIZE_SZ, 0)
4575 || __builtin_expect (nextsize >= av->system_mem, 0))
4576 malloc_printerr ("realloc(): invalid next size");
4577
4578 if ((unsigned long) (oldsize) >= (unsigned long) (nb))
4579 {
4580 /* already big enough; split below */
4581 newp = oldp;
4582 newsize = oldsize;
4583 }
4584
4585 else
4586 {
4587 /* Try to expand forward into top */
4588 if (next == av->top &&
4589 (unsigned long) (newsize = oldsize + nextsize) >=
4590 (unsigned long) (nb + MINSIZE))
4591 {
4592 set_head_size (oldp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0));
4593 av->top = chunk_at_offset (oldp, nb);
4594 set_head (av->top, (newsize - nb) | PREV_INUSE);
4595 check_inuse_chunk (av, oldp);
4596 return chunk2mem (oldp);
4597 }
4598
4599 /* Try to expand forward into next chunk; split off remainder below */
4600 else if (next != av->top &&
4601 !inuse (next) &&
4602 (unsigned long) (newsize = oldsize + nextsize) >=
4603 (unsigned long) (nb))
4604 {
4605 newp = oldp;
4606 unlink_chunk (av, next);
4607 }
4608
4609 /* allocate, copy, free */
4610 else
4611 {
4612 newmem = _int_malloc (av, nb - MALLOC_ALIGN_MASK);
4613 if (newmem == 0)
4614 return 0; /* propagate failure */
4615
4616 newp = mem2chunk (newmem);
4617 newsize = chunksize (newp);
4618
4619 /*
4620 Avoid copy if newp is next chunk after oldp.
4621 */
4622 if (newp == next)
4623 {
4624 newsize += oldsize;
4625 newp = oldp;
4626 }
4627 else
4628 {
4629 memcpy (newmem, chunk2mem (oldp), oldsize - SIZE_SZ);
4630 _int_free (av, oldp, 1);
4631 check_inuse_chunk (av, newp);
4632 return chunk2mem (newp);
4633 }
4634 }
4635 }
4636
4637 /* If possible, free extra space in old or extended chunk */
4638
4639 assert ((unsigned long) (newsize) >= (unsigned long) (nb));
4640
4641 remainder_size = newsize - nb;
4642
4643 if (remainder_size < MINSIZE) /* not enough extra to split off */
4644 {
4645 set_head_size (newp, newsize | (av != &main_arena ? NON_MAIN_ARENA : 0));
4646 set_inuse_bit_at_offset (newp, newsize);
4647 }
4648 else /* split remainder */
4649 {
4650 remainder = chunk_at_offset (newp, nb);
4651 set_head_size (newp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0));
4652 set_head (remainder, remainder_size | PREV_INUSE |
4653 (av != &main_arena ? NON_MAIN_ARENA : 0));
4654 /* Mark remainder as inuse so free() won't complain */
4655 set_inuse_bit_at_offset (remainder, remainder_size);
4656 _int_free (av, remainder, 1);
4657 }
4658
4659 check_inuse_chunk (av, newp);
4660 return chunk2mem (newp);
4661}
4662
4663/*
4664 ------------------------------ memalign ------------------------------
4665 */
4666
4667static void *
4668_int_memalign (mstate av, size_t alignment, size_t bytes)
4669{
4670 INTERNAL_SIZE_T nb; /* padded request size */
4671 char *m; /* memory returned by malloc call */
4672 mchunkptr p; /* corresponding chunk */
4673 char *brk; /* alignment point within p */
4674 mchunkptr newp; /* chunk to return */
4675 INTERNAL_SIZE_T newsize; /* its size */
4676 INTERNAL_SIZE_T leadsize; /* leading space before alignment point */
4677 mchunkptr remainder; /* spare room at end to split off */
4678 unsigned long remainder_size; /* its size */
4679 INTERNAL_SIZE_T size;
4680
4681
4682
4683 checked_request2size (bytes, nb);
4684
4685 /*
4686 Strategy: find a spot within that chunk that meets the alignment
4687 request, and then possibly free the leading and trailing space.
4688 */
4689
4690
4691 /* Check for overflow. */
4692 if (nb > SIZE_MAX - alignment - MINSIZE)
4693 {
4694 __set_errno (ENOMEM);
4695 return 0;
4696 }
4697
4698 /* Call malloc with worst case padding to hit alignment. */
4699
4700 m = (char *) (_int_malloc (av, nb + alignment + MINSIZE));
4701
4702 if (m == 0)
4703 return 0; /* propagate failure */
4704
4705 p = mem2chunk (m);
4706
4707 if ((((unsigned long) (m)) % alignment) != 0) /* misaligned */
4708
4709 { /*
4710 Find an aligned spot inside chunk. Since we need to give back
4711 leading space in a chunk of at least MINSIZE, if the first
4712 calculation places us at a spot with less than MINSIZE leader,
4713 we can move to the next aligned spot -- we've allocated enough
4714 total room so that this is always possible.
4715 */
4716 brk = (char *) mem2chunk (((unsigned long) (m + alignment - 1)) &
4717 - ((signed long) alignment));
4718 if ((unsigned long) (brk - (char *) (p)) < MINSIZE)
4719 brk += alignment;
4720
4721 newp = (mchunkptr) brk;
4722 leadsize = brk - (char *) (p);
4723 newsize = chunksize (p) - leadsize;
4724
4725 /* For mmapped chunks, just adjust offset */
4726 if (chunk_is_mmapped (p))
4727 {
4728 set_prev_size (newp, prev_size (p) + leadsize);
4729 set_head (newp, newsize | IS_MMAPPED);
4730 return chunk2mem (newp);
4731 }
4732
4733 /* Otherwise, give back leader, use the rest */
4734 set_head (newp, newsize | PREV_INUSE |
4735 (av != &main_arena ? NON_MAIN_ARENA : 0));
4736 set_inuse_bit_at_offset (newp, newsize);
4737 set_head_size (p, leadsize | (av != &main_arena ? NON_MAIN_ARENA : 0));
4738 _int_free (av, p, 1);
4739 p = newp;
4740
4741 assert (newsize >= nb &&
4742 (((unsigned long) (chunk2mem (p))) % alignment) == 0);
4743 }
4744
4745 /* Also give back spare room at the end */
4746 if (!chunk_is_mmapped (p))
4747 {
4748 size = chunksize (p);
4749 if ((unsigned long) (size) > (unsigned long) (nb + MINSIZE))
4750 {
4751 remainder_size = size - nb;
4752 remainder = chunk_at_offset (p, nb);
4753 set_head (remainder, remainder_size | PREV_INUSE |
4754 (av != &main_arena ? NON_MAIN_ARENA : 0));
4755 set_head_size (p, nb);
4756 _int_free (av, remainder, 1);
4757 }
4758 }
4759
4760 check_inuse_chunk (av, p);
4761 return chunk2mem (p);
4762}
4763
4764
4765/*
4766 ------------------------------ malloc_trim ------------------------------
4767 */
4768
4769static int
4770mtrim (mstate av, size_t pad)
4771{
4772 /* Ensure all blocks are consolidated. */
4773 malloc_consolidate (av);
4774
4775 const size_t ps = GLRO (dl_pagesize);
4776 int psindex = bin_index (ps);
4777 const size_t psm1 = ps - 1;
4778
4779 int result = 0;
4780 for (int i = 1; i < NBINS; ++i)
4781 if (i == 1 || i >= psindex)
4782 {
4783 mbinptr bin = bin_at (av, i);
4784
4785 for (mchunkptr p = last (bin); p != bin; p = p->bk)
4786 {
4787 INTERNAL_SIZE_T size = chunksize (p);
4788
4789 if (size > psm1 + sizeof (struct malloc_chunk))
4790 {
4791 /* See whether the chunk contains at least one unused page. */
4792 char *paligned_mem = (char *) (((uintptr_t) p
4793 + sizeof (struct malloc_chunk)
4794 + psm1) & ~psm1);
4795
4796 assert ((char *) chunk2mem (p) + 4 * SIZE_SZ <= paligned_mem);
4797 assert ((char *) p + size > paligned_mem);
4798
4799 /* This is the size we could potentially free. */
4800 size -= paligned_mem - (char *) p;
4801
4802 if (size > psm1)
4803 {
4804#if MALLOC_DEBUG
4805 /* When debugging we simulate destroying the memory
4806 content. */
4807 memset (paligned_mem, 0x89, size & ~psm1);
4808#endif
4809 __madvise (paligned_mem, size & ~psm1, MADV_DONTNEED);
4810
4811 result = 1;
4812 }
4813 }
4814 }
4815 }
4816
4817#ifndef MORECORE_CANNOT_TRIM
4818 return result | (av == &main_arena ? systrim (pad, av) : 0);
4819
4820#else
4821 return result;
4822#endif
4823}
4824
4825
4826int
4827__malloc_trim (size_t s)
4828{
4829 int result = 0;
4830
4831 if (__malloc_initialized < 0)
4832 ptmalloc_init ();
4833
4834 mstate ar_ptr = &main_arena;
4835 do
4836 {
4837 __libc_lock_lock (ar_ptr->mutex);
4838 result |= mtrim (ar_ptr, s);
4839 __libc_lock_unlock (ar_ptr->mutex);
4840
4841 ar_ptr = ar_ptr->next;
4842 }
4843 while (ar_ptr != &main_arena);
4844
4845 return result;
4846}
4847
4848
4849/*
4850 ------------------------- malloc_usable_size -------------------------
4851 */
4852
4853static size_t
4854musable (void *mem)
4855{
4856 mchunkptr p;
4857 if (mem != 0)
4858 {
4859 p = mem2chunk (mem);
4860
4861 if (__builtin_expect (using_malloc_checking == 1, 0))
4862 return malloc_check_get_size (p);
4863
4864 if (chunk_is_mmapped (p))
4865 {
4866 if (DUMPED_MAIN_ARENA_CHUNK (p))
4867 return chunksize (p) - SIZE_SZ;
4868 else
4869 return chunksize (p) - 2 * SIZE_SZ;
4870 }
4871 else if (inuse (p))
4872 return chunksize (p) - SIZE_SZ;
4873 }
4874 return 0;
4875}
4876
4877
4878size_t
4879__malloc_usable_size (void *m)
4880{
4881 size_t result;
4882
4883 result = musable (m);
4884 return result;
4885}
4886
4887/*
4888 ------------------------------ mallinfo ------------------------------
4889 Accumulate malloc statistics for arena AV into M.
4890 */
4891
4892static void
4893int_mallinfo (mstate av, struct mallinfo *m)
4894{
4895 size_t i;
4896 mbinptr b;
4897 mchunkptr p;
4898 INTERNAL_SIZE_T avail;
4899 INTERNAL_SIZE_T fastavail;
4900 int nblocks;
4901 int nfastblocks;
4902
4903 check_malloc_state (av);
4904
4905 /* Account for top */
4906 avail = chunksize (av->top);
4907 nblocks = 1; /* top always exists */
4908
4909 /* traverse fastbins */
4910 nfastblocks = 0;
4911 fastavail = 0;
4912
4913 for (i = 0; i < NFASTBINS; ++i)
4914 {
4915 for (p = fastbin (av, i); p != 0; p = p->fd)
4916 {
4917 ++nfastblocks;
4918 fastavail += chunksize (p);
4919 }
4920 }
4921
4922 avail += fastavail;
4923
4924 /* traverse regular bins */
4925 for (i = 1; i < NBINS; ++i)
4926 {
4927 b = bin_at (av, i);
4928 for (p = last (b); p != b; p = p->bk)
4929 {
4930 ++nblocks;
4931 avail += chunksize (p);
4932 }
4933 }
4934
4935 m->smblks += nfastblocks;
4936 m->ordblks += nblocks;
4937 m->fordblks += avail;
4938 m->uordblks += av->system_mem - avail;
4939 m->arena += av->system_mem;
4940 m->fsmblks += fastavail;
4941 if (av == &main_arena)
4942 {
4943 m->hblks = mp_.n_mmaps;
4944 m->hblkhd = mp_.mmapped_mem;
4945 m->usmblks = 0;
4946 m->keepcost = chunksize (av->top);
4947 }
4948}
4949
4950
4951struct mallinfo
4952__libc_mallinfo (void)
4953{
4954 struct mallinfo m;
4955 mstate ar_ptr;
4956
4957 if (__malloc_initialized < 0)
4958 ptmalloc_init ();
4959
4960 memset (&m, 0, sizeof (m));
4961 ar_ptr = &main_arena;
4962 do
4963 {
4964 __libc_lock_lock (ar_ptr->mutex);
4965 int_mallinfo (ar_ptr, &m);
4966 __libc_lock_unlock (ar_ptr->mutex);
4967
4968 ar_ptr = ar_ptr->next;
4969 }
4970 while (ar_ptr != &main_arena);
4971
4972 return m;
4973}
4974
4975/*
4976 ------------------------------ malloc_stats ------------------------------
4977 */
4978
4979void
4980__malloc_stats (void)
4981{
4982 int i;
4983 mstate ar_ptr;
4984 unsigned int in_use_b = mp_.mmapped_mem, system_b = in_use_b;
4985
4986 if (__malloc_initialized < 0)
4987 ptmalloc_init ();
4988 _IO_flockfile (stderr);
4989 int old_flags2 = stderr->_flags2;
4990 stderr->_flags2 |= _IO_FLAGS2_NOTCANCEL;
4991 for (i = 0, ar_ptr = &main_arena;; i++)
4992 {
4993 struct mallinfo mi;
4994
4995 memset (&mi, 0, sizeof (mi));
4996 __libc_lock_lock (ar_ptr->mutex);
4997 int_mallinfo (ar_ptr, &mi);
4998 fprintf (stderr, "Arena %d:\n", i);
4999 fprintf (stderr, "system bytes = %10u\n", (unsigned int) mi.arena);
5000 fprintf (stderr, "in use bytes = %10u\n", (unsigned int) mi.uordblks);
5001#if MALLOC_DEBUG > 1
5002 if (i > 0)
5003 dump_heap (heap_for_ptr (top (ar_ptr)));
5004#endif
5005 system_b += mi.arena;
5006 in_use_b += mi.uordblks;
5007 __libc_lock_unlock (ar_ptr->mutex);
5008 ar_ptr = ar_ptr->next;
5009 if (ar_ptr == &main_arena)
5010 break;
5011 }
5012 fprintf (stderr, "Total (incl. mmap):\n");
5013 fprintf (stderr, "system bytes = %10u\n", system_b);
5014 fprintf (stderr, "in use bytes = %10u\n", in_use_b);
5015 fprintf (stderr, "max mmap regions = %10u\n", (unsigned int) mp_.max_n_mmaps);
5016 fprintf (stderr, "max mmap bytes = %10lu\n",
5017 (unsigned long) mp_.max_mmapped_mem);
5018 stderr->_flags2 = old_flags2;
5019 _IO_funlockfile (stderr);
5020}
5021
5022
5023/*
5024 ------------------------------ mallopt ------------------------------
5025 */
5026static __always_inline int
5027do_set_trim_threshold (size_t value)
5028{
5029 LIBC_PROBE (memory_mallopt_trim_threshold, 3, value, mp_.trim_threshold,
5030 mp_.no_dyn_threshold);
5031 mp_.trim_threshold = value;
5032 mp_.no_dyn_threshold = 1;
5033 return 1;
5034}
5035
5036static __always_inline int
5037do_set_top_pad (size_t value)
5038{
5039 LIBC_PROBE (memory_mallopt_top_pad, 3, value, mp_.top_pad,
5040 mp_.no_dyn_threshold);
5041 mp_.top_pad = value;
5042 mp_.no_dyn_threshold = 1;
5043 return 1;
5044}
5045
5046static __always_inline int
5047do_set_mmap_threshold (size_t value)
5048{
5049 /* Forbid setting the threshold too high. */
5050 if (value <= HEAP_MAX_SIZE / 2)
5051 {
5052 LIBC_PROBE (memory_mallopt_mmap_threshold, 3, value, mp_.mmap_threshold,
5053 mp_.no_dyn_threshold);
5054 mp_.mmap_threshold = value;
5055 mp_.no_dyn_threshold = 1;
5056 return 1;
5057 }
5058 return 0;
5059}
5060
5061static __always_inline int
5062do_set_mmaps_max (int32_t value)
5063{
5064 LIBC_PROBE (memory_mallopt_mmap_max, 3, value, mp_.n_mmaps_max,
5065 mp_.no_dyn_threshold);
5066 mp_.n_mmaps_max = value;
5067 mp_.no_dyn_threshold = 1;
5068 return 1;
5069}
5070
5071static __always_inline int
5072do_set_mallopt_check (int32_t value)
5073{
5074 return 1;
5075}
5076
5077static __always_inline int
5078do_set_perturb_byte (int32_t value)
5079{
5080 LIBC_PROBE (memory_mallopt_perturb, 2, value, perturb_byte);
5081 perturb_byte = value;
5082 return 1;
5083}
5084
5085static __always_inline int
5086do_set_arena_test (size_t value)
5087{
5088 LIBC_PROBE (memory_mallopt_arena_test, 2, value, mp_.arena_test);
5089 mp_.arena_test = value;
5090 return 1;
5091}
5092
5093static __always_inline int
5094do_set_arena_max (size_t value)
5095{
5096 LIBC_PROBE (memory_mallopt_arena_max, 2, value, mp_.arena_max);
5097 mp_.arena_max = value;
5098 return 1;
5099}
5100
5101#if USE_TCACHE
5102static __always_inline int
5103do_set_tcache_max (size_t value)
5104{
5105 if (value >= 0 && value <= MAX_TCACHE_SIZE)
5106 {
5107 LIBC_PROBE (memory_tunable_tcache_max_bytes, 2, value, mp_.tcache_max_bytes);
5108 mp_.tcache_max_bytes = value;
5109 mp_.tcache_bins = csize2tidx (request2size(value)) + 1;
5110 }
5111 return 1;
5112}
5113
5114static __always_inline int
5115do_set_tcache_count (size_t value)
5116{
5117 LIBC_PROBE (memory_tunable_tcache_count, 2, value, mp_.tcache_count);
5118 mp_.tcache_count = value;
5119 return 1;
5120}
5121
5122static __always_inline int
5123do_set_tcache_unsorted_limit (size_t value)
5124{
5125 LIBC_PROBE (memory_tunable_tcache_unsorted_limit, 2, value, mp_.tcache_unsorted_limit);
5126 mp_.tcache_unsorted_limit = value;
5127 return 1;
5128}
5129#endif
5130
5131int
5132__libc_mallopt (int param_number, int value)
5133{
5134 mstate av = &main_arena;
5135 int res = 1;
5136
5137 if (__malloc_initialized < 0)
5138 ptmalloc_init ();
5139 __libc_lock_lock (av->mutex);
5140
5141 LIBC_PROBE (memory_mallopt, 2, param_number, value);
5142
5143 /* We must consolidate main arena before changing max_fast
5144 (see definition of set_max_fast). */
5145 malloc_consolidate (av);
5146
5147 switch (param_number)
5148 {
5149 case M_MXFAST:
5150 if (value >= 0 && value <= MAX_FAST_SIZE)
5151 {
5152 LIBC_PROBE (memory_mallopt_mxfast, 2, value, get_max_fast ());
5153 set_max_fast (value);
5154 }
5155 else
5156 res = 0;
5157 break;
5158
5159 case M_TRIM_THRESHOLD:
5160 do_set_trim_threshold (value);
5161 break;
5162
5163 case M_TOP_PAD:
5164 do_set_top_pad (value);
5165 break;
5166
5167 case M_MMAP_THRESHOLD:
5168 res = do_set_mmap_threshold (value);
5169 break;
5170
5171 case M_MMAP_MAX:
5172 do_set_mmaps_max (value);
5173 break;
5174
5175 case M_CHECK_ACTION:
5176 do_set_mallopt_check (value);
5177 break;
5178
5179 case M_PERTURB:
5180 do_set_perturb_byte (value);
5181 break;
5182
5183 case M_ARENA_TEST:
5184 if (value > 0)
5185 do_set_arena_test (value);
5186 break;
5187
5188 case M_ARENA_MAX:
5189 if (value > 0)
5190 do_set_arena_max (value);
5191 break;
5192 }
5193 __libc_lock_unlock (av->mutex);
5194 return res;
5195}
5196libc_hidden_def (__libc_mallopt)
5197
5198
5199/*
5200 -------------------- Alternative MORECORE functions --------------------
5201 */
5202
5203
5204/*
5205 General Requirements for MORECORE.
5206
5207 The MORECORE function must have the following properties:
5208
5209 If MORECORE_CONTIGUOUS is false:
5210
5211 * MORECORE must allocate in multiples of pagesize. It will
5212 only be called with arguments that are multiples of pagesize.
5213
5214 * MORECORE(0) must return an address that is at least
5215 MALLOC_ALIGNMENT aligned. (Page-aligning always suffices.)
5216
5217 else (i.e. If MORECORE_CONTIGUOUS is true):
5218
5219 * Consecutive calls to MORECORE with positive arguments
5220 return increasing addresses, indicating that space has been
5221 contiguously extended.
5222
5223 * MORECORE need not allocate in multiples of pagesize.
5224 Calls to MORECORE need not have args of multiples of pagesize.
5225
5226 * MORECORE need not page-align.
5227
5228 In either case:
5229
5230 * MORECORE may allocate more memory than requested. (Or even less,
5231 but this will generally result in a malloc failure.)
5232
5233 * MORECORE must not allocate memory when given argument zero, but
5234 instead return one past the end address of memory from previous
5235 nonzero call. This malloc does NOT call MORECORE(0)
5236 until at least one call with positive arguments is made, so
5237 the initial value returned is not important.
5238
5239 * Even though consecutive calls to MORECORE need not return contiguous
5240 addresses, it must be OK for malloc'ed chunks to span multiple
5241 regions in those cases where they do happen to be contiguous.
5242
5243 * MORECORE need not handle negative arguments -- it may instead
5244 just return MORECORE_FAILURE when given negative arguments.
5245 Negative arguments are always multiples of pagesize. MORECORE
5246 must not misinterpret negative args as large positive unsigned
5247 args. You can suppress all such calls from even occurring by defining
5248 MORECORE_CANNOT_TRIM,
5249
5250 There is some variation across systems about the type of the
5251 argument to sbrk/MORECORE. If size_t is unsigned, then it cannot
5252 actually be size_t, because sbrk supports negative args, so it is
5253 normally the signed type of the same width as size_t (sometimes
5254 declared as "intptr_t", and sometimes "ptrdiff_t"). It doesn't much
5255 matter though. Internally, we use "long" as arguments, which should
5256 work across all reasonable possibilities.
5257
5258 Additionally, if MORECORE ever returns failure for a positive
5259 request, then mmap is used as a noncontiguous system allocator. This
5260 is a useful backup strategy for systems with holes in address spaces
5261 -- in this case sbrk cannot contiguously expand the heap, but mmap
5262 may be able to map noncontiguous space.
5263
5264 If you'd like mmap to ALWAYS be used, you can define MORECORE to be
5265 a function that always returns MORECORE_FAILURE.
5266
5267 If you are using this malloc with something other than sbrk (or its
5268 emulation) to supply memory regions, you probably want to set
5269 MORECORE_CONTIGUOUS as false. As an example, here is a custom
5270 allocator kindly contributed for pre-OSX macOS. It uses virtually
5271 but not necessarily physically contiguous non-paged memory (locked
5272 in, present and won't get swapped out). You can use it by
5273 uncommenting this section, adding some #includes, and setting up the
5274 appropriate defines above:
5275
5276 *#define MORECORE osMoreCore
5277 *#define MORECORE_CONTIGUOUS 0
5278
5279 There is also a shutdown routine that should somehow be called for
5280 cleanup upon program exit.
5281
5282 *#define MAX_POOL_ENTRIES 100
5283 *#define MINIMUM_MORECORE_SIZE (64 * 1024)
5284 static int next_os_pool;
5285 void *our_os_pools[MAX_POOL_ENTRIES];
5286
5287 void *osMoreCore(int size)
5288 {
5289 void *ptr = 0;
5290 static void *sbrk_top = 0;
5291
5292 if (size > 0)
5293 {
5294 if (size < MINIMUM_MORECORE_SIZE)
5295 size = MINIMUM_MORECORE_SIZE;
5296 if (CurrentExecutionLevel() == kTaskLevel)
5297 ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0);
5298 if (ptr == 0)
5299 {
5300 return (void *) MORECORE_FAILURE;
5301 }
5302 // save ptrs so they can be freed during cleanup
5303 our_os_pools[next_os_pool] = ptr;
5304 next_os_pool++;
5305 ptr = (void *) ((((unsigned long) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK);
5306 sbrk_top = (char *) ptr + size;
5307 return ptr;
5308 }
5309 else if (size < 0)
5310 {
5311 // we don't currently support shrink behavior
5312 return (void *) MORECORE_FAILURE;
5313 }
5314 else
5315 {
5316 return sbrk_top;
5317 }
5318 }
5319
5320 // cleanup any allocated memory pools
5321 // called as last thing before shutting down driver
5322
5323 void osCleanupMem(void)
5324 {
5325 void **ptr;
5326
5327 for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++)
5328 if (*ptr)
5329 {
5330 PoolDeallocate(*ptr);
5331 * ptr = 0;
5332 }
5333 }
5334
5335 */
5336
5337
5338/* Helper code. */
5339
5340extern char **__libc_argv attribute_hidden;
5341
5342static void
5343malloc_printerr (const char *str)
5344{
5345 __libc_message (do_abort, "%s\n", str);
5346 __builtin_unreachable ();
5347}
5348
5349/* We need a wrapper function for one of the additions of POSIX. */
5350int
5351__posix_memalign (void **memptr, size_t alignment, size_t size)
5352{
5353 void *mem;
5354
5355 /* Test whether the SIZE argument is valid. It must be a power of
5356 two multiple of sizeof (void *). */
5357 if (alignment % sizeof (void *) != 0
5358 || !powerof2 (alignment / sizeof (void *))
5359 || alignment == 0)
5360 return EINVAL;
5361
5362
5363 void *address = RETURN_ADDRESS (0);
5364 mem = _mid_memalign (alignment, size, address);
5365
5366 if (mem != NULL)
5367 {
5368 *memptr = mem;
5369 return 0;
5370 }
5371
5372 return ENOMEM;
5373}
5374weak_alias (__posix_memalign, posix_memalign)
5375
5376
5377int
5378__malloc_info (int options, FILE *fp)
5379{
5380 /* For now, at least. */
5381 if (options != 0)
5382 return EINVAL;
5383
5384 int n = 0;
5385 size_t total_nblocks = 0;
5386 size_t total_nfastblocks = 0;
5387 size_t total_avail = 0;
5388 size_t total_fastavail = 0;
5389 size_t total_system = 0;
5390 size_t total_max_system = 0;
5391 size_t total_aspace = 0;
5392 size_t total_aspace_mprotect = 0;
5393
5394
5395
5396 if (__malloc_initialized < 0)
5397 ptmalloc_init ();
5398
5399 fputs ("<malloc version=\"1\">\n", fp);
5400
5401 /* Iterate over all arenas currently in use. */
5402 mstate ar_ptr = &main_arena;
5403 do
5404 {
5405 fprintf (fp, "<heap nr=\"%d\">\n<sizes>\n", n++);
5406
5407 size_t nblocks = 0;
5408 size_t nfastblocks = 0;
5409 size_t avail = 0;
5410 size_t fastavail = 0;
5411 struct
5412 {
5413 size_t from;
5414 size_t to;
5415 size_t total;
5416 size_t count;
5417 } sizes[NFASTBINS + NBINS - 1];
5418#define nsizes (sizeof (sizes) / sizeof (sizes[0]))
5419
5420 __libc_lock_lock (ar_ptr->mutex);
5421
5422 for (size_t i = 0; i < NFASTBINS; ++i)
5423 {
5424 mchunkptr p = fastbin (ar_ptr, i);
5425 if (p != NULL)
5426 {
5427 size_t nthissize = 0;
5428 size_t thissize = chunksize (p);
5429
5430 while (p != NULL)
5431 {
5432 ++nthissize;
5433 p = p->fd;
5434 }
5435
5436 fastavail += nthissize * thissize;
5437 nfastblocks += nthissize;
5438 sizes[i].from = thissize - (MALLOC_ALIGNMENT - 1);
5439 sizes[i].to = thissize;
5440 sizes[i].count = nthissize;
5441 }
5442 else
5443 sizes[i].from = sizes[i].to = sizes[i].count = 0;
5444
5445 sizes[i].total = sizes[i].count * sizes[i].to;
5446 }
5447
5448
5449 mbinptr bin;
5450 struct malloc_chunk *r;
5451
5452 for (size_t i = 1; i < NBINS; ++i)
5453 {
5454 bin = bin_at (ar_ptr, i);
5455 r = bin->fd;
5456 sizes[NFASTBINS - 1 + i].from = ~((size_t) 0);
5457 sizes[NFASTBINS - 1 + i].to = sizes[NFASTBINS - 1 + i].total
5458 = sizes[NFASTBINS - 1 + i].count = 0;
5459
5460 if (r != NULL)
5461 while (r != bin)
5462 {
5463 size_t r_size = chunksize_nomask (r);
5464 ++sizes[NFASTBINS - 1 + i].count;
5465 sizes[NFASTBINS - 1 + i].total += r_size;
5466 sizes[NFASTBINS - 1 + i].from
5467 = MIN (sizes[NFASTBINS - 1 + i].from, r_size);
5468 sizes[NFASTBINS - 1 + i].to = MAX (sizes[NFASTBINS - 1 + i].to,
5469 r_size);
5470
5471 r = r->fd;
5472 }
5473
5474 if (sizes[NFASTBINS - 1 + i].count == 0)
5475 sizes[NFASTBINS - 1 + i].from = 0;
5476 nblocks += sizes[NFASTBINS - 1 + i].count;
5477 avail += sizes[NFASTBINS - 1 + i].total;
5478 }
5479
5480 size_t heap_size = 0;
5481 size_t heap_mprotect_size = 0;
5482 size_t heap_count = 0;
5483 if (ar_ptr != &main_arena)
5484 {
5485 /* Iterate over the arena heaps from back to front. */
5486 heap_info *heap = heap_for_ptr (top (ar_ptr));
5487 do
5488 {
5489 heap_size += heap->size;
5490 heap_mprotect_size += heap->mprotect_size;
5491 heap = heap->prev;
5492 ++heap_count;
5493 }
5494 while (heap != NULL);
5495 }
5496
5497 __libc_lock_unlock (ar_ptr->mutex);
5498
5499 total_nfastblocks += nfastblocks;
5500 total_fastavail += fastavail;
5501
5502 total_nblocks += nblocks;
5503 total_avail += avail;
5504
5505 for (size_t i = 0; i < nsizes; ++i)
5506 if (sizes[i].count != 0 && i != NFASTBINS)
5507 fprintf (fp, " \
5508 <size from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n",
5509 sizes[i].from, sizes[i].to, sizes[i].total, sizes[i].count);
5510
5511 if (sizes[NFASTBINS].count != 0)
5512 fprintf (fp, "\
5513 <unsorted from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n",
5514 sizes[NFASTBINS].from, sizes[NFASTBINS].to,
5515 sizes[NFASTBINS].total, sizes[NFASTBINS].count);
5516
5517 total_system += ar_ptr->system_mem;
5518 total_max_system += ar_ptr->max_system_mem;
5519
5520 fprintf (fp,
5521 "</sizes>\n<total type=\"fast\" count=\"%zu\" size=\"%zu\"/>\n"
5522 "<total type=\"rest\" count=\"%zu\" size=\"%zu\"/>\n"
5523 "<system type=\"current\" size=\"%zu\"/>\n"
5524 "<system type=\"max\" size=\"%zu\"/>\n",
5525 nfastblocks, fastavail, nblocks, avail,
5526 ar_ptr->system_mem, ar_ptr->max_system_mem);
5527
5528 if (ar_ptr != &main_arena)
5529 {
5530 fprintf (fp,
5531 "<aspace type=\"total\" size=\"%zu\"/>\n"
5532 "<aspace type=\"mprotect\" size=\"%zu\"/>\n"
5533 "<aspace type=\"subheaps\" size=\"%zu\"/>\n",
5534 heap_size, heap_mprotect_size, heap_count);
5535 total_aspace += heap_size;
5536 total_aspace_mprotect += heap_mprotect_size;
5537 }
5538 else
5539 {
5540 fprintf (fp,
5541 "<aspace type=\"total\" size=\"%zu\"/>\n"
5542 "<aspace type=\"mprotect\" size=\"%zu\"/>\n",
5543 ar_ptr->system_mem, ar_ptr->system_mem);
5544 total_aspace += ar_ptr->system_mem;
5545 total_aspace_mprotect += ar_ptr->system_mem;
5546 }
5547
5548 fputs ("</heap>\n", fp);
5549 ar_ptr = ar_ptr->next;
5550 }
5551 while (ar_ptr != &main_arena);
5552
5553 fprintf (fp,
5554 "<total type=\"fast\" count=\"%zu\" size=\"%zu\"/>\n"
5555 "<total type=\"rest\" count=\"%zu\" size=\"%zu\"/>\n"
5556 "<total type=\"mmap\" count=\"%d\" size=\"%zu\"/>\n"
5557 "<system type=\"current\" size=\"%zu\"/>\n"
5558 "<system type=\"max\" size=\"%zu\"/>\n"
5559 "<aspace type=\"total\" size=\"%zu\"/>\n"
5560 "<aspace type=\"mprotect\" size=\"%zu\"/>\n"
5561 "</malloc>\n",
5562 total_nfastblocks, total_fastavail, total_nblocks, total_avail,
5563 mp_.n_mmaps, mp_.mmapped_mem,
5564 total_system, total_max_system,
5565 total_aspace, total_aspace_mprotect);
5566
5567 return 0;
5568}
5569weak_alias (__malloc_info, malloc_info)
5570
5571
5572strong_alias (__libc_calloc, __calloc) weak_alias (__libc_calloc, calloc)
5573strong_alias (__libc_free, __free) strong_alias (__libc_free, free)
5574strong_alias (__libc_malloc, __malloc) strong_alias (__libc_malloc, malloc)
5575strong_alias (__libc_memalign, __memalign)
5576weak_alias (__libc_memalign, memalign)
5577strong_alias (__libc_realloc, __realloc) strong_alias (__libc_realloc, realloc)
5578strong_alias (__libc_valloc, __valloc) weak_alias (__libc_valloc, valloc)
5579strong_alias (__libc_pvalloc, __pvalloc) weak_alias (__libc_pvalloc, pvalloc)
5580strong_alias (__libc_mallinfo, __mallinfo)
5581weak_alias (__libc_mallinfo, mallinfo)
5582strong_alias (__libc_mallopt, __mallopt) weak_alias (__libc_mallopt, mallopt)
5583
5584weak_alias (__malloc_stats, malloc_stats)
5585weak_alias (__malloc_usable_size, malloc_usable_size)
5586weak_alias (__malloc_trim, malloc_trim)
5587
5588#if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_26)
5589compat_symbol (libc, __libc_free, cfree, GLIBC_2_0);
5590#endif
5591
5592/* ------------------------------------------------------------
5593 History:
5594
5595 [see ftp://g.oswego.edu/pub/misc/malloc.c for the history of dlmalloc]
5596
5597 */
5598/*
5599 * Local variables:
5600 * c-basic-offset: 2
5601 * End:
5602 */
5603