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